Brookings Trade2005Forum Offshoring White-Collar Work Susan M. Collins and Lael Brainard editors
brookings institution press Washington, D.C.
Copyright © 2006 THE BROOKINGS INSTITUTION 1775 Massachusetts Avenue, N.W., Washington, D.C. 20036 ISSN 1520-5479 ISBN-13: 978-0-8157-1284-8 ISBN-10: 0-8157-1284-7
For information on subscriptions, standing orders, and individual copies, contact Brookings Institution Press, P.O. Box 465, Hanover, PA 17331-0465. Or call 866-698-0010. E-mail
[email protected]. Visit Brookings online at http://bookstore.brookings.edu. Brookings periodicals are available online through both Online Computer Library Center (contact OCLC subscriptions department at 800-848-5878, ext. 6251) and Project Muse (http://muse.jhu.edu). Authorization to photocopy items for internal or personal use or the internal or personal use of specific clients is granted by the Brookings Institution for libraries and other users registered with the Copyright Clearance Center Transactional Reporting Service, provided that the basic fee is paid to the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. For more information, please contact CCC at 508/750-8400. This authorization does not extend to other kinds of copying, such as copying for general distribution, or for creating new collective works, or for sale. Specific written permission for such copying must be obtained from the Permissions Department, Brookings Institution Press, 1775 Massachusetts Avenue, N.W., Washington, DC 20036; fax, 202/536-3623; e-mail,
[email protected].
Brookings Trade2005Forum Offshoring White-Collar Work Editors’ Summary
ix
What Do We Learn from Trade Theory?
james r. markusen Modeling the Offshoring of White-Collar Services: From Comparative Advantage to the New Theories of Trade and Foreign Direct Investment
1
Comments by Alan V. Deardorff and Douglas A. Irwin 24 Discussion 30
daniel trefler Service Offshoring: Threats and Opportunities
35
Comments by Dani Rodrik and Pol Antràs 61 Discussion 66
Exploring the Empirics
j. bradford jensen and lori g. kletzer Tradable Services: Understanding the Scope and Impact of Services Offshoring
75
Comments by Jared Bernstein and Robert C. Feenstra 117 Discussion 127
maria borga Trends in Employment at U.S. Multinational Companies: Evidence from Firm-Level Data
135
desirée van welsum and xavier reif Potential Offshoring: Evidence from Selected OECD Countries
165
Comments by Robert Z. Lawrence and Catherine L. Mann 195 Discussion 201
Offshoring—India’s Role
t. n. srinivasan Information-Technology-Enabled Services and India’s Growth Prospects
203
Comment by Anne Krueger 232 Discussion 236
rafiq dossani Globalization and the Offshoring of Services: The Case of India
241
Comment by Arvind Panagariya 268 Discussion 274
Lessons from Industry Studies
clair brown and greg linden Offshoring in the Semiconductor Industry: A Historical Perspective
279
Comment by Jeffrey T. Macher 323 Discussion 329
rosemary batt, virginia doellgast, and hyunji kwon Service Management and Employment Systems in U.S. and Indian Call Centers
335
Comment by Vivek Agrawal 361 Discussion 366
ravi aron and ying liu Determinants of Operational Risk in Global Sourcing of Financial Services: Evidence from Field Research
373
ashish arora The Emerging Offshore Software Industries and the U.S. Economy
399
frank levy and ari goelman Offshoring and Radiology
411
Discussion 424
What Role for Policy?
lael brainard, robert e. litan, and nicholas warren A Fairer Deal for America’s Workers in a New Era of Offshoring 427 Comment by Lawrence Mishel 448 Discussion 452
kimberly a. clausing The Role of U.S. Tax Policy in Offshoring Comment by Kevin A. Hassett 483 Discussion 486
457
the brookings institution The Brookings Institution is a private nonprofit organization devoted to research, education, and publication on important issues of domestic and foreign policy. Its principal purpose is to bring the highest quality independent research and analysis to bear on current and emerging policy problems. The Institution was founded on December 8, 1927, to merge the activities of the Institute for Government Research, founded in 1916, the Institute of Economics, founded in 1922, and the Robert Brookings Graduate School of Economics and Government, founded in 1924. Interpretations or conclusions in Brookings publications should be understood to be solely those of the authors. Board of Trustees John L. Thornton Chair Strobe Talbott President Zoë Baird Alan R. Batkin Richard C. Blum Geoffrey T. Boisi James W. Cicconi Arthur B. Culvahouse Jr. Kenneth W. Dam Vishakha N. Desai Thomas E. Donilon Mario Draghi Kenneth M. Duberstein
Lawrence K. Fish Cyrus F. Freidheim Jr. Bart Friedman David Friend Ann M. Fudge Jeffrey W. Greenberg Brian L. Greenspun William A. Haseltine Teresa Heinz Samuel Hellman Glenn H. Hutchins Joel Z. Hyatt Shirley Ann Jackson Kenneth Jacobs Suzanne Nora Johnson
Michael H. Jordan Harold Hongju Koh William A. Owens Frank H. Pearl John Edward Porter Steven Rattner Haim Saban Leonard D. Schaeffer Lawrence H. Summers David F. Swensen Larry D. Thompson Laura D’Andrea Tyson Antoine W. van Agtmael Beatrice W. Welters Daniel Yergin
Honorary Trustees Leonard Abramson Elizabeth E. Bailey Rex J. Bates Louis W. Cabot A. W. Clausen William T. Coleman Jr. Alan M. Dachs D. Ronald Daniel Robert A. Day Bruce B. Dayton Charles W. Duncan Jr. Walter Y. Elisha Robert F. Erburu Henry Louis Gates Jr. Robert D. Haas Lee H. Hamilton F. Warren Hellman
Robert A. Helman Roy M. Huffington James A. Johnson Ann Dibble Jordan Vernon E. Jordan Jr. Breene M. Kerr Marie L. Knowles James T. Lynn Jessica Tuchman Mathews David O. Maxwell Donald F. McHenry Robert S. McNamara Mary Patterson McPherson Arjay Miller Mario M. Morino Maconda Brown O’Connor Samuel Pisar
J. Woodward Redmond Charles W. Robinson James D. Robinson III Judith Rodin Warren B. Rudman B. Francis Saul II Ralph S. Saul Henry B. Schacht Michael P. Schulhof Joan E. Spero Vincent J. Trosino John C. Whitehead Stephen M. Wolf James D. Wolfensohn Ezra K. Zilkha
Brookings Trade Forum is a series of annual volumes that provide authoritative and indepth analysis on current and emerging issues in international economics. The series aims to explore questions on international trade and macroeconomics in an interdisciplinary fashion with both practitioners and academics and seeks to gather in one place papers that provide a thorough look at a particular topic affecting international economic policy. Leading experts in the field will contribute to each volume. This eighth issue contains edited versions of the papers and comments presented at a conference held at the Brookings Institution, May 12–13, 2005. This year’s forum focused on the offshoring of high-skilled work. Contributors use trade theory, empirics, industry studies, and policy analyses to illuminate the issues and implications of this complex phenomenon. The conference and journal were made possible by a generous grant from the Alfred P. Sloan Foundation. Additional support of the 2005 Trade Forum came from the William and Flora Hewlett Foundation and the Tokyo Club Foundation for Global Studies.
Coeditors
Staff
Advisers
Contributors
Susan M. Collins, Brookings Institution and Georgetown University Lael Brainard, Brookings Institution Lindsey Wilson, production assistant Janet Mowery, editorial associate Gabriel Chodorow-Reich, research assistant and verifier Eric Haven, research verifier Stephen Robblee, research verifier Nicholas Warren, senior research analyst Jagdish Bhagwati, Columbia University Richard N. Cooper, Harvard University Avinash Dixit, Princeton University Geza Feketekuty, Monterey Institute of International Studies Jeffrey A. Frankel, Harvard University Gene Grossman, Princeton University Gary Horlick, Wilmer, Cutler & Pickering Gary Hufbauer, Institute for International Economics John H. Jackson, Georgetown Law School Paul R. Krugman, Princeton University Maurice Obstfeld, University of California–Berkeley Sylvia Ostry, University of Toronto Kenneth Rogoff, International Monetary Fund Laura Tyson, London Business School Paula Stern, The Stern Group Alan Wolff, Dewey Balantine Vivek Agrawal, McKinsey and Company Pol Antràs, Harvard University Ravi Aron, University of Pennsylvania Ashish Arora, Carnegie Mellon University Rosemary Batt, Cornell University Jared Bernstein, Economic Policy Institute Maria Borga, U.S. Bureau of Economic Analysis
Lael Brainard, Brookings Institution Clair Brown, University of California–Berkeley Kimberly A. Clausing, Reed College Alan V. Deardorff, University of Michigan Virginia Doellgast, Cornell University Rafiq Dossani, Stanford University Robert C. Feenstra, University of California–Davis Ari Goelman, Massachusetts Institute of Technology Kevin A. Hassett, American Enterprise Institute Douglas A. Irwin, Dartmouth College J. Bradford Jensen, Institute for International Economics Lori G. Kletzer, University of California–Santa Cruz Anne Krueger, International Monetary Fund Hyunji Kwon, Cornell University Robert Z. Lawrence, Harvard University Frank Levy, Massachusetts Institute of Technology Greg Linden, University of California–Berkeley Robert E. Litan, Brookings Institution and Kauffman Foundation Ying Liu, University of Pennsylvania Jeffrey T. Macher, Georgetown University Catherine L. Mann, Institute of International Economics James R. Markusen, University of Colorado–Boulder Lawrence Mishel, Economic Policy Institute Arvind Panagariya, Columbia University Xavier Reif, Organization for Economic Cooperation and Development Dani Rodrik, Harvard University T. N. Srinivasan, Yale University Daniel Trefler, University of Toronto Desirée van Welsum, Organization for Economic Cooperation and Development Nicholas Warren, Brookings Institution Conference participants
Susan Aaronson, University of North Carolina Starynee Adams, Brookings Institution Mary Amiti, International Monetary Fund Claude Barfield, American Enterprise Institute Joshua Bivens, Economic Policy Institute Ronald Blackwell, AFL-CIO Mary Jane Bolle, Library of Congress Barry Bosworth, Brookings Institution Chad P. Bown, Brookings Institution and Brandeis University Sydney F. Collins, University of Miami Rishi Daga, Reliance Infocomm Annie Davis, Brookings Institution I. M. Destler, University of Maryland and Institute for International Economics Rebecca Dillender, Bureau of Internal Labor Affairs Howard Dobson, United States Department of Labor Kristin Forbes, Council of Economic Advisers Barbara Fraumeni, Bureau of Economic Analysis Richard Freeman, Harvard University
Conference participants (continued)
Carol Graham, Brookings Institution Gene Grossman, Princeton University Jane T. Haltmaier, Federal Reserve Board Ronil Hira, Rochester Institute of Technology Ned Howenstine, Bureau of Economic Analysis Kent Hughes, Woodrow Wilson International Center for Scholars Graham Ingham, International Monetary Fund Jane Irhig, Federal Reserve Board Karen Johnson, Federal Reserve Board Robert Johnson, Cargill Douglas Kaden, Oak Hill Capital Aslihan Kes, International Center for Research on Women Judy Knepper, United States Government Accountability Office Ralph Kozlow, Bureau of Economic Analysis Martha Laboissiere, McKinsey and Company Frank Langfitt, National Public Radio Thea Lee, AFL-CIO Philip I. Levy, Council of Economic Advisers Paul Magnusson, Business Week Magazine Dalia Marin, University of Munich Richard Mataloni, Bureau of Economic Analysis Lawrence McNeil, Bureau of Economic Analysis Emily McWithey, Brookings Institution Theodore Moran, Georgetown University Janet Norwood, New York Conference Board Gail Pesyna, The Sloan Foundation Anbihn Phan, Princeton University David Ratner, Economic Policy Institute J. David Richardson, Syracuse University and Institute for International Economics Fernando Robles, George Washington University Howard Rosen, TAA Coalition Jeffrey Russell, Duke University Kenneth Ryder, National Academy of Public Administration Isabelle Sawhill, Brookings Institution Gary Saxonhouse, University of Michigan Phillip Swagel, American Enterprise Institute Strobe Talbott, Brookings Institution Daniel Tartullo, Georgetown University Law Center Michael Teitelbaum, The Sloan Foundation Edwin M. Truman, Institute for International Economics Robert Vilhauer, Boeing Company David Walters, Office of the United States Trade Representative Timothy Wedding, United States Government Accountability Office Shang-Jin Wei, International Monetary Fund Obie Whichard, Bureau of Economic Advisers Beth Anne Wilson, Federal Reserve Board Karen Wilson, Boeing Company Loren Yager, United States Government Accountability Office William J. Zeile, Bureau of Economic Analysis
LAEL BRAINARD SUSAN M. COLLINS
Offshoring White-Collar Work: Editors’ Summary
O
ffshoring of services burst into America’s public consciousness in 2003, raising concerns about the nation’s future competitiveness. For the first time, highly skilled white-collar workers in the United States perceived themselves to be in direct competition with lower-paid foreign workers. Viewed from across the Indian Ocean, services offshoring appeared to promise accelerated development— where poor countries could compete successfully with much richer ones by focusing their talent pool on a select group of high-value activities. For the first time in many decades, mainstream economists launched a serious debate over the possibility that trade could be zero sum (albeit only in special circumstances). With the media stoking the popular imagination through dramatic, though often anecdotal or inconsistent “statistics,” the Brookings Institution embarked on an effort to analyze the offshoring phenomenon. First, seeking to separate fact from speculation, we convened a full-day conference in 2004 that focused narrowly on the available data, their implications, and their limitations. We then launched a broader research project aimed at illuminating offshoring from a variety of complementary views: those of theory and empirics, industries and labor markets, and developed and developing countries. Participants with diverse perspectives presented their analyses at the Brookings Trade Forum Conference in May 2005. This volume contains the fourteen revised papers, as well as invited commentary by fourteen additional experts and summaries of the conference discussions. Here we highlight some of the key themes and conclusions that emerged from the project, and then briefly summarize the papers. The term offshoring has itself been the source of some confusion. In this volume, we distinguish between questions of location (whether an activity is undertaken in the home market or offshore) and questions of ownership (whether an activity is undertaken within an enterprise or is outsourced to an arm’s-length ix
x
Brookings Trade Forum: 2005
provider. Thus, by offshoring we mean the assignment of part or all of the value chain to an offshore (foreign) location, where an activity could be done either within the firm (in-house) or by a third party (at arm’s length). In contrast, we use the term outsourcing to mean the assignment of services to a third party, whether domestic or foreign. The research focused on the new dimensions of offshoring, which reflect both sector and occupation: offshoring of services, as well as offshoring of white-collar work. The contributors to this volume demonstrate that existing economic theory can go a long way toward capturing key dimensions of the services offshoring phenomenon. In particular, trade theory can explain why production processes that rely on skilled workers may migrate to locations in developing countries where such workers are relatively scarce; lack of complementary factors such as know-how may make them relatively inexpensive in these markets. In many respects, the issues raised by the new wave of offshoring parallel those that arose from the globalization of manufacturing that began some decades ago. But recent developments affect workers who tend to be more educated and to earn higher salaries. They often involve only a fragment of the value chain. And because of the nature of the relevant processes, recent theory suggests that levels of institutional development might constrain the extent to which developing countries can participate. Under quite sensible assumptions, standard trade models do not necessarily predict that high-wage economies such as the United States will gain from greater offshoring of services. One contribution of the project is to clarify the channels through which gains and losses might occur. Since existing evidence consistently shows that relatively few service and white-collar jobs have been offshored to date, concerns focus on what might happen in the future. How much of the labor force in the United States and elsewhere is really vulnerable? As the 2004 workshop made clear, the answer is difficult to discern from existing data. Thus a contribution of this project is to present and discuss alternative approaches to constructing such indicators—one based on quantitative analysis across U.S. industries and occupations and the other based on a more subjective assessment for a large number of OECD countries. The results suggest that more of the industrial country workers exposed to international trade are employed in services than in manufacturing, and show that those workers suffer substantial losses when they are displaced. They also highlight that trade goes in both directions and that foreign firms locating domestically provide job opportunities for workers in tradable service sectors. The spread of offshoring to services raises questions that are at least as important and consequential for developing countries as for the advanced ones. In this context, India emerges as a focus of both public and academic attention. Its econ-
Lael Brainard and Susan M. Collins
xi
omy is a study in contrasts—home to both a world-class information technology (IT) services sector and to one-quarter of the world’s poorest. Contributors to the project highlight both lessons and debates from recent research. There is broad agreement that India’s IT success is founded on a pool of highly educated English speakers and that IT services have thrived by working around poor infrastructure and taking advantage of episodic openings in the restrictive policy regime. Interestingly, among the potential constraints to future growth is a limited supply of suitably skilled workers, not concerns about India’s institutional weaknesses (such as corruption and patchy intellectual property enforcement). Significant disagreement emerged over the importance of India’s diaspora community to the IT success; whether the experience in IT will drive broader economic liberalization; and whether India could have reaped even greater dividends by pursuing a broader growth strategy premised on access to high-quality primary education, public investment in infrastructure, and sweeping policy reform. Analysis of individual service sectors yields rich insights into the offshoring phenomenon. Software services have a long history of offshoring, particularly in India, followed by call centers, semiconductor design, and business process outsourcing (BPO). Radiology is an example of a relatively new focus of public attention. Services differ profoundly from manufacturing in that quality inheres in the underlying processes rather than in a final physical output. As a result, the decision to offshore depends on a firm’s ability to specify and monitor processes and on its choice of internal or arm’s-length organizational forms. In some sectors—notably call centers—a cookbook approach to offshoring appears to prevent foreign providers from fully utilizing their skilled workers or achieving maximum efficiency. Unexploited opportunities for arbitrage also are present in the BPO sector, where field surveys suggest the striking finding that advanced economy buyers and developing economy suppliers perceive complexity very differently. For instance, while processes requiring intensive algorithmic computation tend to be rated as complex by managers in the United States and the United Kingdom, managers in India and Singapore give high complexity ratings to processes requiring judgmentdriven communication. Concerns about the loss of high-value services appear to be overblown in software services, call centers, and semiconductor design: the vertical decomposition of these services preserves the highest-value activities in the home market, while shedding lower-value processes to overseas providers. And professional credential requirements interact with other factors to make extensive offshoring of radiology services unlikely to come to pass. What is the policy agenda that emerges from the spread of offshoring into services? The papers in this volume focus on two areas: social insurance and corporate taxation. The United States lags behind other wealthy economies on
xii
Brookings Trade Forum: 2005
social insurance. With offshoring and technological change accelerating the rate at which workers’ job specific-skills depreciate, there is a strong case to be made for wage loss insurance, which would encourage workers to broaden their employment search and go back to work more quickly, while defraying the cost to employers of hiring and providing on-the-job training to new employees from different sectors. The policy agenda should also address economy-wide cost disadvantages. In this context, a case can also be made for simplifying the U.S. system of taxing multinationals’ worldwide income and for lowering the corporate tax rate while broadening the base.
Part I: What Do We Learn from Trade Theory? In the first paper of the volume, James Markusen examines offshoring of white-collar work from the perspective of a trade theorist. His approach is to construct a set of alternative conceptual frameworks, instead of focusing on a single detailed model based on one particular framework. The paper provides new and useful benchmarks for understanding the effects of offshoring on welfare and factor incomes among industrialized economies (which he calls the North) and emerging economies (the South). Markusen begins by identifying the real-world features of offshoring that he believes models should consider. First, he argues that offshoring of white-collar services is largely about technical and institutional innovations that allow new things to be traded, and not about marginal liberalizations that allow more and less costly trade in existing traded goods and services. Further, the newly traded services tend not to be final goods, but instead parts of a production chain that is decomposed or “vertically fragmented” geographically. These offshored services can be either upstream (such as software design) or downstream (such as call centers that offer after-sales services). And the offshored service activities in his model can be fragmented from the production of a final manufactured good as easily as from the production of a final service. Second, borrowing from mainstream trade theory, Markusen argues that it is important to incorporate differences in both technology across productive activities and factor supplies across countries. Fragments of the production process often differ in their factor intensities (that is, their relative use of unskilled labor, skilled labor, physical capital), while economies often differ in their factor endowments (the relative abundance of skilled and unskilled labor and physical capital). This may suggest, for example, that if production of a certain service
Lael Brainard and Susan M. Collins
xiii
requires mid-level skills, its production may be best suited to locations where such labor is cheap. However, the simplest off-the-shelf trade model in which factor intensities interact with factor endowments (the two-factor Heckscher-Ohlin model) implies that factors are always expensive where they are scarce. This is in stark contrast to the empirical fact that skilled labor is relatively cheap in some countries where it is also relatively scarce, such as India or China. Thus Markusen suggests expanding the basic model to incorporate an additional factor, which he labels “know-how.” The know-how could be physical capital, such as computers, networks, and satellite links, or knowledge capital, such as managerial, organizational, or marketing expertise. He makes the key assumption that know-how is complementary to skilled labor, thereby creating a potential missing input for countries otherwise well suited to the production of skill-intensive fragments. This feature provides an explanation for the “puzzle” that scarce factors can be relatively cheap. Markusen presents a series of five simple models. The first is a standard twogood, two-factor, two-region Heckscher-Ohlin model with perfect competition. His innovation is to permit fragmentation of the initially more skill-intensive good, leading to a shift of the service activity from North to South—the offshoring of services. The second model incorporates “know-how,” the third factor assumed to be located primarily in the North and a necessary complement to the skilled labor used in services. Markusen’s third model omits know-how but introduces multinational firms that produce with increasing returns to scale under conditions of imperfect competition. The fourth model combines the missing input model with the multinational production structure. Markusen also briefly discusses a fifth, somewhat more speculative, framework that considers some issues related to outsourcing—or whether offshoring is conducted within or outside the ownership boundaries of the multinational firm. Markusen’s preferred framework is the missing input model, particularly with its extension to multinationals. His models illustrate how the ability to offshore white-collar services can exploit this gap between skills and know-how, bringing modern corporate knowledge to these skilled workers. Although there is clearly a benefit to the world economy overall, the model also highlights that there are winners and losers, and in particular, that some northern groups may be vulnerable. The implications of offshoring that these models suggest to be unambiguous are welfare gains for the South overall and welfare gains for skilled labor in the South. The implications for unskilled labor in both regions, for skilled labor in the North, and for overall welfare in the North are all mixed.
xiv
Brookings Trade Forum: 2005
The second paper, by Daniel Trefler, elaborates on some of the issues Markusen considers in his fifth model. A central point of his analysis is that institutional development in potential host economies is likely to constrain the extent to which activities can be offshored, which he defines as the use of workers located abroad to provide sophisticated services to U.S. customers. Trefler sees two key features of offshoring that distinguish it from other types of international trade. He identifies each feature with a set of concerns from the North. First, service offshoring uses some of the most dynamic information and communication technologies (ICT). It thus has implications for the corporate innovation strategies that lie at the heart of U.S. competitiveness policies. He argues that this feature raises the concern that U.S. firms may be crowded out of the most innovative lines of business, and that there is already some evidence of this among Indian multinationals such as Satyam, HCL, and Tata. Second, Trefler notes that the employment of highly skilled white-collar workers in low-cost countries such as India raises the concern that offshoring may displace “good” U.S. jobs and depress salaries of high-paid workers, both of which would reduce the incentives of Americans to invest in their own human capital. A related concern is that the disruption from increased service offshoring may make it less worthwhile for firms to make long-term investments in their best workers. Although existing evidence suggests that service offshoring is currently small, Trefler stresses that many worry that, over the next ten to twenty years or more, imports from China and India will devastate the United States. He believes this concern is misplaced for two reasons. First, it ignores the law of comparative advantage, which states that no country can export all goods. He notes that similar concerns were expressed about Japan, where wages in 1959 were 10 percent of U.S. levels. However, the law of comparative advantage does not rule out the possibility that China and India will export high-tech goods and services, leaving the relatively low-skilled activities for Americans. This raises Trefler’s third and perhaps central theme: the crucial role of institutions. Current thinking about innovation-based long-term growth focuses on (1) institutions that protect property rights from preying politicians and bureaucrats, (2) institutions that provide a fully functioning legal framework for arm’slength transactions, and (3) institutions that balance the needs of innovators inside the corporation against the needs of investors outside the corporation. Trefler observes that these institutions are only beginning to take shape in China, India, and many of the other emerging markets that are potential offshoring hosts, and argues that they are unlikely to evolve quickly, even over a quartercentury horizon. Thus, Trefler concludes that China and India are a long way from being the world’s innovation giants.
Lael Brainard and Susan M. Collins
xv
Finally, Trefler believes that most of the sensible policies aimed at fostering American competitiveness in the service offshoring market are investmentpromoting framework policies. These encourage American workers, firms, and governments to invest in building productive assets such as human capital and new technologies. While acknowledging that such framework policies address a whole host of domestic competitiveness issues and are not unique to offshoring, Trefler argues that they should still be seen as appropriate ways to address concerns raised by services offshoring. He advocates providing more assistance to workers who are displaced, both skilled white-collar workers and unskilled workers displaced by low-end manufacturing imports. Many of these issues are addressed in the paper on wage insurance by Lael Brainard, Robert Litan, and Nicholas Warren.
Part II: Exploring the Empirics The next three papers in the volume address empirical issues. Traditionally, economists have treated services as nontraded activities. However, as globalized production expands beyond manufacturing to include a variety of services, this perspective is obsolescent, and the need for better measures of tradability is clear. Existing data on international flows of services provide at best an incomplete picture of the magnitude and dimensions of this growing phenomenon. J. Bradford Jensen and Lori Kletzer present a new method for identifying which service activities are vulnerable to international trade. Their approach distinguishes occupations as well as industries and enables them to examine the implications of globalizing the production of services for domestic employment and job loss. Specifically, Jensen and Kletzer use the geographic concentration of service activities in the United States to identify which service activities are traded domestically. They then classify activities that are traded domestically as potentially tradable internationally. Using the identified industries and occupations, they develop estimates of the number of workers who are in tradable activities in all sectors of the economy. This enables them to compare the demographic characteristics of workers in tradable and nontradable activities and the employment growth in traded and nontraded service activities. Their approach to measuring tradability of services relies on the economic intuition that production of services will be geographically concentrated only if the services are traded. This builds from the observation that production of traded manufactured goods tends to be geographically concentrated (to capitalize on
xvi
Brookings Trade Forum: 2005
increasing returns to scale, access to inputs like natural resources, etc.), while production of those that are not traded tends to be more widely distributed. Indeed, they show that the geographic concentration of some service activities within the United States is nearly as great as in manufacturing. Their paper highlights three main findings. First, they conclude that the number of workers potentially exposed to international trade in services is actually larger than the number of exposed workers in manufacturing. Second, workers in tradable sectors have higher skills and significantly higher earnings. The higher earnings do not appear to be solely a result of higher skill levels—in regressions controlling for observable characteristics, workers in selected tradable service activities have 16–17 percent higher earnings than similar workers in nontradable activities in the same sector. Third, employment in tradable activities grew more slowly in the period 1998 to 2002, but Jensen and Kletzer show that this was due primarily to employment losses in manufacturing. Within services, employment in tradable and nontradable activities grew at similar rates except at the lowest end of the skill distribution. Average employment growth was negative in low-skill tradable industries and occupations but positive (though low) in nontraded low-skill services. Jensen and Kletzer use the 2004 Displaced Worker Survey to examine the scope and cost of involuntary job loss. Briefly, they find some evidence that displacement rates are higher from tradable than from nontradable service industries. The difference is most notable in the Information sector; however, the authors note that this could simply represent displacements associated with the tech/telecom bubble and may not reflect offshoring. They also find somewhat higher displacement rates from tradable than from nontradable white-collar occupations. The same gap is not evident for blue-collar workers. Consistent with the employment characteristics, they find that workers displaced from tradable service activities are more educated, and have higher earnings, than workers displaced from nontradable activities. They are also more likely to be reemployed. Finally, they document that job loss from tradable and nontradable service activities is costly to workers (they typically endure a period of unemployment and are unlikely to earn as much in new jobs as they did in their former jobs). Generalizing from what is known about manufacturing worker job loss, the authors speculate that lower levels of job tenure and higher levels of educational attainment may be advantages in seeking reemployment. They would favor a less porous safety net, provided, for example, by extending Trade Adjustment Assistance (TAA) to services workers and extending wage insurance beyond TAA.
Lael Brainard and Susan M. Collins
xvii
The next two papers, by Maria Borga and by Desirée van Welsum and Xavier Reif, address evidence of existing and potential offshoring by U.S. and selected OECD multinational companies. Borga uses firm-level data from the Bureau of Economic Analysis (BEA) to examine the evidence on the extent of offshoring by U.S. parent companies to their foreign affiliates, and then to determine if, and how, offshoring is associated with changes in U.S. parent employment. Using survey data from 1,117 U.S. parent companies, she studied their offshoring behavior by examining the shares of imports of goods from affiliated and unaffiliated parties and of imports of services from affiliated parties in their total purchased inputs. Between 1994 and 2002, average employment growth was 39 percent for U.S. parent companies, a result of both mergers and acquisitions and the expansion of existing operations. Examining parents’ offshoring behavior, she finds distinctions between goods and services as purchased inputs, as well as between purchases from affiliates versus nonaffiliates. Overall, she concludes that the vast majority of U.S. parents’ purchased inputs are acquired from domestic sources, not imports in both 1994 and 2002. The average share of foreign affiliates’ sales to the local market increased, reaching 78 percent in 2002, while shares of sales to both third countries and the United States declined. The high share of sales to local markets demonstrates the importance of serving local customers in the parent’s decision to invest abroad. This suggests that serving local markets was an increasingly important motive for U.S. MNCs’ overseas expansion. Borga also compares the data for parent companies with employment gains and those with employment losses. The average increase in employment (82 percent) is roughly the same as the average increase in sales (84 percent) for parents with employment gains. But for parents that lost employment, the average decrease in employment was 30 percent, well above the 10 percent average decrease in sales. Both sets of parents increased their reliance on purchased inputs. However, those parents whose employment declined strengthened their ties to their affiliates by increasing their reliance on imported goods and, to a lesser extent, on imported services from them. On the other hand, parents that gained employees had a smaller increase in reliance on affiliated imports of goods and reduced their reliance on affiliated imports of services. The paper then turns to a decomposition of changes in parent employment among three factors: the change in output, the change in labor productivity, and the change in the use of purchased inputs in production. This decomposition shows that the increase in labor productivity was the most important factor for those parents that lost employment, followed by the loss of output and by the
xviii
Brookings Trade Forum: 2005
increased use of purchased inputs. For those parents that gained employment, increased output more than accounted for the change in employment. Because most of the inputs purchased by parents are supplied by domestic firms, rather than imported, the loss of jobs due to greater imports accounted for less than one-third of the total change in employment attributed to the greater use of purchased inputs. In the final section of the paper, Borga considers a variety of correlations, including those between employment and output, employment and labor productivity, employment and reliance on purchased inputs, and imports by parents and parents’ employment. In their paper, Desirée van Welsum and Xavier Reif also focus on the growing tradability of services and the implications for white-collar jobs previously shielded from international competition. Like Jensen and Kletzer, their work is motivated by the absence of any official data measuring the extent of the offshoring of services activities. Instead of a statistical approach, they consider a checklist of attributes and use their own judgment to identify tradable occupations. The authors use data on trade in services and occupational employment to estimate the current extent of globalization of services in OECD economies and its potential growth. They find that in many countries often mentioned in the offshoring debate both exports and imports of business and computer and information services have grown rapidly. OECD countries still account for over 75 percent of exports of these services, but their share is declining. Drawing on their earlier detailed analysis of occupational data for selected OECD countries, Van Welsum and Reif seek to determine the share of total employment that could potentially be affected by the international sourcing of IT- and ICT-enabled services. Their measure suggests that close to 20 percent of total employment could potentially be affected by offshoring, particularly in business services (for example, accounting, consulting), financial services, and research and development. The authors provide a simple descriptive regression analysis of the relationship between the share of employment potentially affected by offshoring and other economic and structural developments for OECD economies between 1996 and 2003. They do not find any systematic or significant evidence that either net outward investment or imports of business services are associated with declines in the share of employment potentially affected by offshoring—at least at the aggregate level. However, they do find that exports of business services are positively associated with the share of employment potentially affected by offshoring, suggesting that increases in demand and production have also raised demand for these types of ICT-using occupations. Other key factors found to be
Lael Brainard and Susan M. Collins
xix
positively associated with the share of employment potentially affected by offshoring are the comparative size of the service sector, the growing share of ICT investment in total fixed investment, and human capital.
Part III: Offshoring—India’s Role The next two papers, by T. N. Srinivasan and Rafiq Dossani, consider the economic implications of services offshoring for India. T. N. Srinivasan finds the contribution of information technology services to Indian growth to be significant. He traces the spectacular development of India’s IT sector to several sources, especially public investment in higher education and the creation of elite engineering schools. The policy regime was also influential: first as a restrictive force until the mid-1980s, subsequently as an enabling factor before the reforms of 1991, and thereafter as a proactive supportive force. Telecommunications reforms, government incentives for software technology parks (STPs), foreign investment, and venture capital all played prominent roles. Srinivasan argues that the Indian IT diaspora concentrated in Silicon Valley was a significant factor in the rise of India’s IT sector, although Rafiq Dossani finds evidence for the opposite view. Srinivasan notes that a significant number of Indian engineers who held senior positions in U.S. companies in the late 1990s subsequently helped persuade their senior management to establish operations in India. Moreover, the diaspora’s influence is spreading beyond the narrow confines of the Indian IT industry to the broader contours of India’s economic development and growth. Srinivasan includes a brief analytical discussion of the influence of IT in the growth process and as a source of dynamic comparative advantage. IT services are in effect universal intermediates, which are essential to any production activity and possibly most, if not all, consumption activities. Any technical progress in the IT sector is reflected first in productivity gains (or cost reductions) in the IT sector. Other sectors begin to experience productivity gains as they invest in equipment and processes to take advantage of the new lower-cost information technology, which diffuses the total factor productivity (TFP) gains gradually over time. Srinivasan next examines the prospects for and constraints on India’s fulfilling the high expectations for IT-led growth. The spectacular growth of software exports and information-technology-enabled services and business process outsourcing (ITES-BPO) has been underpinned by the relative abundance of relatively low-wage English-speaking skilled workers. Srinivasan notes that current
xx
Brookings Trade Forum: 2005
trends could lead to excess demand for ITES/IT workers, which would put upward pressure on compensation. If the share of Indian graduates employed in the industry does not decline, the implied annual growth rate of 11.3 percent is far below the 35 percent growth in the value of India’s ITES/IT output projected by India’s National Association of Software and Service Companies (NASSCOM) for 2003–12. Any increase in the compensation of ITES/IT workers in India with no change in their productivity would cut into India’s cost competitiveness. While India is likely to remain competitive for the foreseeable future at the low end of the skill spectrum, India’s competitive edge is likely to erode at the higher end of the spectrum, as Indian wages rise relative to U.S. wages. Already, worker turnover is high, particularly in smaller lower-value-added firms, which should put upward pressure on compensation. Potential future competitors include Bangladesh, Ireland, Pakistan, Sri Lanka, and the Philippines, all of which have pools of English-speaking workers suitable for ITES/IT employment, although only Bangladesh currently has wages below India’s. Competition from China also cannot be ruled out, since English-speaking ability can be acquired. Srinivasan believes the IT sector can make a significant contribution to India’s future growth, but cautions that accelerating the rate of GDP growth to 8 percent or more per year and sustaining it for several decades is a necessary though not sufficient condition for achieving the overarching objective of eradicating poverty. He also believes that the success of the IT sector can help to broaden and strengthen political support for reforms in other sectors—if three things happen. First, politicians must recognize the link between reforms and the success of the IT sector. Second, a much greater proportion of the population must experience the benefits from efficient and inexpensive IT services. Third, political differences over the pace and sequencing of further reforms will have to be resolved. In the next paper, Rafiq Dossani examines India’s experience with providing offshored services to developed nations, where certain components of services exporting have been well established for decades. India has catapulted itself in recent years into a leadership position among developing economies on services exports related to information technology. Dossani’s analysis offers central insights into how the services component of international trade has expanded and how a developing country like India could succeed in exporting services to developed countries. Dossani begins with a discussion of how technology has helped to overcome the intrinsically greater obstacles to trade in services relative to manufactured goods. First, digitization allowed the conversion of service flows into stocks of
Lael Brainard and Susan M. Collins
xxi
information, making it feasible to separate and store a service. The expanded ability to subdivide tasks reduced costs by offering the possibility of preparing the standardized components with lower-cost labor and, possibly, at another location. Second, digitization permitted noninformation service flows to be converted into information service flows—for instance, replacing the need for inperson sampling in a showroom with virtual sampling of goods delivered over the Internet. Third, costs associated with transmitting digitized material dropped significantly. Services such as writing software programs, which were offshored to India in the early 1970s, were enabled by digitized storage and, in the 1980s, by the standardization of programming languages. Still later, as digital transmission costs fell in the 1990s, even nonstorable services, such as customer care, could be offshored. Advances in information technology made the recent growth in offshoring possible by parsing the provision of certain services into components requiring different levels of skill and interactivity. Dossani finds that the initial impact in India of services offshoring was to generate high and growing levels of employment. However, the low-value-added and low-skill work that was being produced also provided few barriers to entry and was subject to automation. The resulting competition and price deflation mimicked the situation in manufacturing exports in developing countries and raised the likelihood that asymmetries of globalization could be repeated in services exports from developing countries. Dossani presents a multifaceted explanation of India’s success with offshoring services. Local entrepreneurship and a high level of infant industry protection allowed the Indian IT industry to reach a high growth path and allowed local skills to keep pace with global changes. A key advantage appears to be widespread education in the English language. Other institutional advantages are India’s mature judicial system, its conformance with WTO obligations, and a history of successful private enterprise that provided the talent for initiating and managing complex service projects. In contrast, China has better infrastructure but lacks a history of private entrepreneurship, a large population with knowledge of English, and a mature judicial system. Dossani disagrees with the widely held view that spending by the government on education was a key contributor to the success of offshoring, and also argues that the global Indian diaspora has been largely noninfluential, except during the past few years. He highlights the tendency for higher-stage work to remain in the developed countries. This is due to the lack of domain skills in India—a consequence of protectionist policies and the fact that India has no domestic demand for highend services to promote the development of such skilled workforces. Dossani
xxii
Brookings Trade Forum: 2005
does say that recent reforms intended to attract transnational corporations and transnationally trained diaspora could change the environment by introducing domain skills. Additional obstacles are technology, the need to protect proprietary knowledge, economies of scale and time, and mission criticality. Even when offshoring is possible, the in-house multinational is a favored organizational form in most cases. This enables much of the rents from sophisticated work to be captured in the developed country even though the work may be done in India.
Part IV: Lessons from Industry Studies The next five papers provide a fascinating set of industry-level studies of offshoring in diverse service sectors. Clair Brown and Greg Linden examine semiconductor design, which is a frequently cited component of the new wave of services offshoring. Semiconductor (or chip) companies were among the first to invest in offshore facilities to manufacture goods for imports back to the United States. Because meaningful data about the impact of the offshoring of chip design (and even of manufacturing) are limited, Brown and Linden rely on a more qualitative analysis for their key points. They conducted dozens of interviews with engineers and managers at numerous semiconductor and related companies in the United States, Asia, and Europe over the past twelve years. Their research also incorporates the rich store of publicly available information in trade journals and company reports. After briefly describing the stages of semiconductor production and their analytical framework, they examine the offshoring of assembly jobs, manufacturing, and design jobs. They also discuss what their conclusions mean for the United States. Before addressing semiconductor design directly, Brown and Linden analyze the impact on the U.S. semiconductor industry of the offshoring of semiconductor assembly and fabrication. They argue that the initial concern about losing domestic jobs in both stages turned out to be unfounded, as the industry used the situation to its competitive advantage by becoming cost competitive (through assembly offshoring) and by developing the “fabless” (design-only) sector (through foreign outsourcing of chip fabrication or manufacturing). Brown and Linden then analyze the ongoing offshoring of design jobs and compare this stage with the two stages that preceded it in order to explore the possible impact on domestic jobs and the U.S. semiconductor industry.
Lael Brainard and Susan M. Collins
xxiii
In the second industry study, Rosemary Batt, Virginia Doellgast, and Hyunji Kwon assess call centers through surveys of Indian and U.S. companies. To date, the depiction of this emerging sector has been based largely on anecdotal evidence rather than systematic empirical investigation of business practices— either in the United States or in India. Batt and her coauthors seek to map the extent of variation in service management and employment strategies among U.S. in-house, U.S. outsourced, and foreign offshore call centers that provide similar services to American customers. In addition, they test the impact of ownership status and firm strategies on worker turnover. The authors begin with a literature review, which shows that service management strategies and employment systems vary substantially—from professional approaches to service to highly transactional or cost-driven ones. In addition, work and employment systems typically are differentiated according to the level of education and training required; the level of discretion and collaborative problem-solving embedded in the design of work; and the level and type of compensation system designed to motivate effort. There are several straightforward implications from this literature on work design and employment systems. Companies are more likely to retain in-house those services that are more complex, involve customer transactions that are more nuanced or uncertain, and involve highly valued customers. In order to meet the demands of these types of products and customers, they are likely to focus on service quality and customization and to adopt a more professional approach to service than a subcontractor would. Centers operated by subcontractors, both in the United States and offshore, by contrast, are more likely to compete on costs by offering lower wages and benefits, using more standardized work processes, and closely monitoring performance. From the literature review, Batt, Doellgast, and Kwon develop several testable predictions differentiating U.S. in-house, U.S. outsourced, and foreign offshore establishments. In their sample of establishments, in-house centers tend to adopt a more coherent quasi-professional approach to service interactions than outsourced and offshore sites; in-house jobs are characterized by relatively higher levels of initial investments in training and pay, discretion, and problemsolving opportunities. In offshore centers, by contrast, workers have somewhat higher levels of formal education and receive more initial training than in-house centers but have fewer opportunities to make choices or solve problems. Further multivariate analyses show that U.S. outsourced and offshore centers have significantly higher quit rates. Ownership status is an important driver in the choice of management and employment practices, with U.S. outsourced and offshore
xxiv
Brookings Trade Forum: 2005
centers more constrained to follow standardized operating procedures and performance monitoring. At the level of managerial policy, the authors conclude that the extensive use of routinized work processes in call centers leads to high turnover, which limits options for customization and is associated with lower service quality and productivity. Moreover, to the extent that call centers hire college-educated workers, the highly constrained and monitored work system creates an inefficient use of human capital. The underutilization of human capital represents a substantial loss for Indian subcontractors, who are paying for skills that they are not using. To the extent that companies have complex service offerings or want to compete on the basis of service differentiation, quality, or customer loyalty, they are likely to retain customer interactions in-house, consistent with the transaction costs perspective and core competency argument. To date, this appears to be what most U.S. corporations are doing: after two decades of rapid growth of U.S. call centers, most industry estimates are consistent with the authors’ finding that less than 15 percent of U.S. call centers are run by third-party subcontractors; and only a tiny fraction have moved offshore. However, for those transactions that are simple and codifiable, Batt and her coauthors predict that companies are likely to continue expanding their operations offshore. Their data suggest that the strategy of outsourcing operations to U.S. subcontractors is likely to be a transitory one, as the modest reductions in labor costs may be offset by the high costs of turnover and low levels of employee skill. According to this scenario, the U.S. subcontracting sector, which grew dramatically in the 1990s, will be the hardest hit by Indian competition. The scenarios also depend on human resource development. In India, there is evidence that demand is outstripping the short-run supply of skilled labor in call center cities such as Bangalore and Chennai. Thus, there is a need for the Indian government to invest in the human resource infrastructure. The next industry study, by Ravi Aron and Ying Liu, investigates the offshore outsourcing of business processes in financial services. Aron and Liu’s findings are based on four years of field research and data collection from firms that provide offshore outsourcing services in India, Mauritius, Singapore, and Thailand, and from clients that buy these services in the United States and the United Kingdom. Operational risk is central to considerations of offshoring. Although the business media often claim that process complexity—the converse of codifiability—is the primary reason for operational risk, Aron and Liu’s field research finds that process complexity actually means different things to different managers. Managers in the West perceive complexity very differently from man-
Lael Brainard and Susan M. Collins
xxv
agers in the countries that provide offshoring services (India and Singapore). Managers in the United States and the United Kingdom tend to rate as complex those processes that call for analytical skills, algorithmic computational intensity, and many subtasks involving quantitative analysis. In contrast, managers in India and Singapore rate as highly complex processes where the work is not easily codifiable and requires judgment-driven communication and contextsensitive inference, and as less complex those that require quantitative analysis and algorithmic computational work. This astonishing finding leads Aron and Liu to ask if there might be a market for complexity arbitrage. Could offshoring release not just gains from wage arbitrage but also from classic specialization in relative comparative advantage? How much a firm gains from offshore outsourcing depends on how well it is able to manage operational risk. To measure this, Aron and Liu analyzs the factors that contribute to operational risk (from survey data). They use a “knowledge continuum” to describe the nature of different stages of information work and how that determines operational risk. They then regress the magnitude of observed operational errors against several of these attributes of the process, the outsourcing contract, and the workforce. Two factors have the greatest influence on operational risk. First, as the work involved in executing offshore processes becomes more codifiable, operational risk declines. Processes that are not easily codifiable or for which the agents need deep context-sensitive understanding of how the process is to be executed are more prone to operational errors. Second, when the buyer and provider of services can agree on a precise and unambiguous set of metrics of process quality, the resulting operational risk is low. However, when process quality itself is open to subjective interpretation, the operational risk is higher. These findings help to shed light on the optimal governance structure for sourcing different kinds of processes. Aron and Liu propose a governance structure that they call the extended organizational form (EOF), where (1) the buyer contracts the production process to the provider; (2) the buyer can inspect the provider’s output quality after production; and (3) the buyer’s managers can also exercise partial managerial control over the provider’s agents by monitoring the quality of process execution during the production. This hybrid mechanism allows buyers to exercise some managerial control across the boundaries of the firm without waiting for a process cycle to be completed. The paper concludes by comparing the efficiency in sourcing offshored services of the traditional in-house hierarchy, market-based outsourcing, and the EOF hybrid. The analysis shows that for relatively complex processes (as rated by the providers), the EOF is indeed the optimal governance choice and holds
xxvi
Brookings Trade Forum: 2005
the potential to unlock gains from wage and complexity arbitrage associated with offshore outsourcing. This organizational form contrasts with the more traditional hierarchical form used for the offshoring of call center work but may also discourage innovation that would improve quality and efficiency. In his paper, Ashish Arora examines the growth and evolution of services offshoring in the global software industry. This experience is particularly interesting because software was one of the first skill-intensive industries for which production moved to relatively low-wage countries. This short paper draws from Arora’s recently coauthored book, which provides extensive additional detail. Broadly speaking, software activities can be divided into design, coding, and maintenance. Arora argues that the latter two of these are analogous to production and entail relatively low-end tasks. These, not design, account for most of the offshoring to date. He distinguishes between those who work in the software industry and the much larger number who work in software occupations outside of the “core” computer equipment and software services industries. After providing an overview of the global software industry in terms of employment, sales, and exports, Arora describes how three countries, India, Ireland, and Israel, have emerged as centers of offshored software service through a combination of excess skilled labor, key innovations at the level of the firm, and good timing. In contrast, Brazil and China, the other two newcomers, have pursued a very different strategy, relying considerably less on exports, at least to date. Brazil has relied on a sophisticated domestic banking industry to generate demand, hoping that it will lead to the creation of an internationally competitive software industry. China appears to be following a more traditional importsubstitution model. Arora does not believe that the fast growth of export-oriented sectors in lower-wage countries yet threatens the supremacy of the United States as a producer of technology services. The continued importance of a close relationship between the producer and end-user of a software service, as well as the U.S. advantage in innovation-spurring institutions such as venture capital, suggests to him that offshored software services will remain concentrated in relatively lowvalue-added activities. But he recognizes that the sensitivity of this conclusion to factors specific to software production may leave workers in other potentially offshorable occupations feeling less secure. In “Offshoring and Radiology,” Frank Levy and Ari Goelman conclude that the much discussed reading of radiology images offshore by “cheap foreign doctors” is, to date, no more than an urban legend. Unlike software professionals, production workers, and call center operators, U.S. doctors (including radiologists) determine who qualifies as a doctor. Many radiology images are out-
Lael Brainard and Susan M. Collins
xxvii
sourced, some to offshore locations. But because of radiologists’ power, virtually all of these images are read by “nighthawk” firms that employ U.S.-boardcertified radiologists who are also accredited in the state and hospital in which the image was created. The typical nighthawk customer is a small U.S. hospital whose emergency room generates only a few images in a night. For these hospitals, hiring full-time or on-call radiologists would be prohibitively expensive, and it is often more efficient and cost-effective to contract with a nighthawk firm offshore. The authors also point out that radiologists’ power to restrict foreign competition is reinforced by other factors, including the cognitive nature of the work. In 2004 Levy and his coauthor Richard J. Murnane argued that tasks are easiest to offshore when they can be performed using deductive or inductive rules, a condition similar to that required to program a task for a computer. But the reading of most radiology images cannot be expressed in rules—for example, few images can be scanned by a computer. Thus, a radiologist’s output is hard to monitor, placing extra emphasis on the radiologist’s credentials. This emphasis on credentials interacts with the threat of malpractice litigation: few doctors would want to explain to a jury why an image was interpreted by an unlicensed radiologist. Correspondingly, a nighthawk firm cannot purchase malpractice insurance unless it can prove it uses board-certified radiologists. Levy and Goelman point out that 75 percent of the fee reimbursed by most health plans is a “technical fee” paid to the entity that owns the scanning equipment and that only 25 percent goes to the radiologist who reads the image. An insurer seeking to limit aggregate costs might in the future focus on limiting the number of scans through benefits management rather than trying to certify foreign radiologists. By contrast, the doctors themselves—radiologists and nonradiologists who own their own scanning equipment—may be a future source of foreign demand. The authors close by explaining why mammography might be a candidate for this kind of offshoring.
Part V: What Role for Policy? The final two papers focus on the implications of offshoring for U.S. policy. Lael Brainard, Robert E. Litan, and Nicholas Warren argue that there is a strong case for helping to insure the livelihoods of the widening pool of American workers who face insecurity associated with structural shifts in order to preserve the benefits of an open and innovative economy. They propose a new wage loss insurance program to provide incentives for more rapid reemployment and on-
xxviii
Brookings Trade Forum: 2005
the-job-training—a program that insures against earnings losses for permanently displaced workers who secure reemployment. Brainard and her coauthors estimate that it would cost roughly $3.5 billion a year to insure permanently displaced full-time workers (who secure reemployment) for 50 percent of their earnings loss up to a cap of $10,000 a year for two years. The authors argue that although the U.S. labor market ranks second to none in job turnover, America’s safety net for easing job transitions is one of the weakest among the wealthy economies. The main federally mandated unemployment insurance (UI) program contains so many restrictions that today only about 40 percent of all jobless workers receive benefits. Meanwhile, workers have long found it difficult, time-consuming, and expensive to prove that they are entitled to extended unemployment benefits under the nation’s Trade Adjustment Assistance (TAA) program. Despite important reforms in 2002, TAA has helped fewer than 75,000 new workers per year, while denying more than 40 percent of all employees’ petitions. And remarkably, the Department of Labor has interpreted the TAA statute as excluding service workers displaced by trade. Arguing that workers’ firm-specific skills are losing value at an accelerating pace in the face of offshoring and technological change, Brainard and her coauthors advocate supplementing the existing safety net with a new program that insures against wage loss, not just unemployment, for permanently displaced workers. Wage insurance would encourage workers to broaden their employment search and go back to work more quickly, while defraying the cost to employers of hiring and providing on-the-job training to new employees from different sectors. With wage insurance, the economy as a whole would benefit from shorter spells of joblessness and more efficient reskilling for workers. A chief goal of wage insurance is to speed the reemployment of workers who have been permanently displaced. Wage insurance is most likely to have overall positive economic benefits if it targets workers whose earnings would otherwise fall dramatically as forces outside their control devalue their firm-specific skills. A Canadian pilot wage insurance program reduced unemployment durations by 4.4 percent, on average, according to research by the Social Development and Research Corporation. This could amount to hundreds of millions of dollars in annual savings on unemployment insurance payments in the U.S. context. The authors emphasize that wage insurance also serves as a training subsidy for the worker’s new employer. The retraining and new skills that a displaced worker receives on a new job benefit both the worker and the new employer. Finally, evidence suggests that wage insurance encourages workers to broaden their job search to new types of jobs in new sectors.
Lael Brainard and Susan M. Collins
xxix
The authors estimate that the net cost of a fairly generous program that provides 50 percent replacement of lost earnings with a $10,000 annual payment cap for two years would be $3.5 billion a year, on average, which amounts to an insurance premium of roughly $25 per worker per year. This is substantially less than the $42.4 billion paid by the state and federal governments in unemployment insurance benefits in 2003. The authors project the costs of a wage insurance program for several different scenarios. Brainard and her coauthors argue that a comprehensive, incentive-based safety net for displaced workers that encourages rapid reemployment and onthe-job training is a benefit for workers and businesses alike. In the volume’s final paper, Kimberly Clausing examines the role of U.S. corporate tax policy on offshoring behavior. Under the current system, U.S. multinational firms are taxed on their worldwide income, although tax credits are granted for taxes paid to foreign governments. Since profits are only taxed upon repatriation to the United States, this system provides an incentive to locate real economic activity as well as profits in low-tax countries. In addition, there is an incentive to avoid locating (and earning profits) in high-tax countries, because U.S. tax credits are limited to the U.S. tax liability. Recent changes in tax law under the American Jobs Creation Act of 2004 strengthen these incentives by further lightening the taxation of foreign income and by granting a temporary tax holiday, lowering taxes on repatriations of dividends from low-tax countries. Clausing examines the incentives that are created by this system with respect to offshoring. Ceteris paribus, the U.S. tax system provides an incentive to offshore activities in low-tax countries and to offshore in-house rather than at arm’s length. Substantial empirical evidence documents that U.S. multinational firms are sensitive to tax rate differentials among countries in their decisions regarding where to invest; this responsiveness is increasing, in part because of the increasing globalization of U.S. business. In addition, the previous empirical evidence indicates that multinational firms are sensitive to tax differences among countries when they decide where to book profits. Such sensitivity has implications for government revenues in the United States and elsewhere. Clausing sees four potential goals for an international corporate tax system: enhancing efficiency, improving U.S. macroeconomic indicators, improving the competitiveness of U.S. multinationals, and generating government revenue. Most international tax systems reflect a balance of these goals, and these goals sometimes compete. For example, efforts to enhance the competitiveness of U.S.-based multinational firms may lead to an artificial tax preference
xxx
Brookings Trade Forum: 2005
favoring economic activity and profits in low-tax countries. Also, some goals are easier to achieve with tax policy than others; attempts to use tax policy to enhance U.S. macroeconomic indicators or the position of U.S. financial balances may be misguided. Finally, Clausing discusses the merits of several policy alternatives in the context of the current U.S. system and the policy goals. These alternatives include the adoption of a territorial system of taxing international income, under which foreign dividends would be exempt from taxation; the elimination of deferral of U.S. taxation on income earned in low-tax countries; and greater international coordination. All of these major changes would require important tradeoffs and should be made with caution. In the meantime, pragmatic smaller changes are likely to improve the functioning of the current U.S. tax system. Proposals to lower the corporate tax rate, broaden the tax base, strengthen enforcement, and simplify the tax system deserve close attention, because these changes would likely improve the performance of the U.S. tax system. Such changes could also help ensure that offshoring activities occur in a manner that enhances efficiency and is consistent with the national interest.
JAMES R. MARKUSEN University of Colorado–Boulder
Modeling the Offshoring of White-Collar Services: From Comparative Advantage to the New Theories of Trade and Foreign Direct Investment
I
have always viewed trade theory as consisting of a portfolio of models.1 There are many underlying causes of or motives for trade, and it is probably more productive to have a series of models analyzing just a few of these at a time than to attempt one grand model that includes all possible bases for trade. At the other extreme, we could envision a model for every industry and every country pair and perhaps every multinational firm. But at this point, theory coincides with case-study analysis and we learn nothing of any generality. So a parsimonious set of models, the number of elements greater than one but less than say one thousand, is probably a good scientific objective. My first question in approaching my assignment for the Brookings Trade Forum is whether we can make good progress from off-the-shelf elements of our portfolio of models, or do we need an entirely new approach? The methodology I use to answer this question is to first ask another question: what are the important characteristics of the offshoring of white-collar services that we wish to capture in a theory model? Having identified a number of these characteristics, I am led toward the conclusion that we can indeed go a long way by drawing from our portfolio of models, mixing and matching elements to create a useful, empirically relevant, and productive subtheory for offshoring white-collar services. 1. I got this idea from Tjalling Koopmans’s (1957) Three Essays on the State of Economic Science, which is still of great value today. Koopmans used the term “sequence of models.”
1
2
Brookings Trade Forum: 2005
I sketch the outlines of a number of candidate “template” models, each of which captures some aspect of the problem. From analytical insights and numerical simulations of these models, I am then able to answer questions about the effects and consequences of technological or institutional innovations that permit offshoring to arise. These include effects on the national income of each country, effects on the relative and real prices of skilled and unskilled labor in each country, and effects on the volume of trade in goods (for example, are trade in goods and trade in services complements or substitutes?). Before proceeding, I wish to emphasize that my goal here is to suggest ways of thinking about the issues in formal models. I was distressed following my presentation at the Brookings conference to find many people focusing on the results of some simulations, particularly with respect to “northern” welfare (read as U.S. welfare). All readers should understand that no theory says that a move from partial liberalization to full liberalization makes everyone better off. To push the point further, I am confident that I can concoct a model to generate any result desired by a reader with a deep pocketbook. I have tried hard to stick to reasonable and relevant structures and assumptions, but even so, qualitative results sometimes depend on specific parameter values, as we shall see. In the following section, I provide a brief overview of some of our theory portfolio and then identify some of the crucial aspects of offshoring we wish to capture. Finally I present a series of template models.
Our Theory Portfolio We can usefully draw from existing theories and models of trade in order to make progress on offshoring. I do not claim that the list is exhaustive or that alternative taxonomies might not be more useful; I just believe that these particular elements will prove useful. —Comparative advantage theories of trade in goods. Our traditional trade theory tends to focus on differences among countries as the primary motive for trade. The Ricardian model of trade, in which countries possess different technologies, is usually listed first. Second, the workhorse model of trade is factorproportions or Heckscher-Ohlin theory, in which differences in factor intensities among goods intersect with differences in factor endowments among countries to determine a pattern of comparative advantage and trade. This ever-popular approach not only gives an intuitive explanation for the direction of trade, but permits a detailed analysis of the distributional consequences of trade within countries and of aggregate gains from trade. Other country characteristics that fit
James R. Markusen
3
here include differences in market distortions among countries and country (internal market) size. —Non-comparative-advantage theories of trade. This category is largely the domain of the “new trade theory,” a term I dislike: “industrial organization approach” to trade is more apt (and avoids the awkward problem of what to call the theory that comes after the new theory). The principal motives for trade are scale economies, imperfect competition, and product differentiation. A subcategory of this branch of theory involves the existence of firm-specific assets, an especially useful approach to the theory of the multinational firm. These range from managerial and technology assets to brand names and trademarks. This approach has resurfaced more recently in heterogeneous firm models, in which (potential) firms get productivity “draws” from some distribution that make some firms more productive than others (Melitz 2003; Helpman, Melitz, and Yeaple 2004). Productivity in turn determines whether firms enter foreign markets and if so whether by exports or foreign production. —Trade in factors. While trade in goods has drawn the most attention in both theory and empirical analysis, the topic of trade in factors has always lurked in the background. Generalization of theoretical findings is difficult, but the loose consensus among trade economists is that trade in goods and trade in factors tend to be substitutes in comparative-advantage models. Indeed, Mundell’s (1957) early demonstration of this might explain the lack of interest in trade in factors. An elegant treatment of this substitutability is found in Jones, Coelho, and Easton (1986). Later it was shown that trade in goods and trade in factors tend to be complements for virtually any other causes of trade other than factor proportions (Markusen 1983) and even in some versions of factor-proportions models (Neary 1995). —Theories of foreign direct investment and arm’s-length trade in firm-specific assets. I think it is fair to say that, until the mid-1980s, FDI was just viewed as part of the theory of capital movements in a factor-proportions world. Eventually a huge amount of empirical evidence, most notably that most foreign direct investment (FDI) not only comes from but goes to other high-income capital-rich countries, led to new approaches to what we are now calling offshoring. Theory split into two branches. One could be called the vertical or resourceseeking approach, an early example of which is described by Helpman (1984). This is in fact a natural extension of factor-proportions comparative-advantage models in which activities differ in factor intensities and countries differ in factor proportions. The alternative is the horizontal or market-seeking approach, in which firms exploit firm-specific assets in multiple markets, an early example of which is described by Markusen (1984). The latter is more a part of non-comparative-
4
Brookings Trade Forum: 2005
advantage theory and, while both approaches are important, does the job of explaining the large volume of intra-industry FDI among the high-income countries. I believe it is accurate to say that the overwhelming weight of empirical evidence, beginning with Brainard (1997), is more consistent with the horizontal approach. Intertwined with this literature on FDI is a long-standing literature on “internalization,” now being called by its inverse name, “outsourcing.” Both terms address whether firms keep certain activities internal to the firm or use arm’slength contractors to supply intermediates or to provide assembly, services, distribution, and so forth. Early analyses include that by Dunning (1977). Some more recent authors seem unaware of this large literature, but it is still a pertinent antecedent, and changing the name from internalization to outsourcing does not change that fact. This literature argues that the choice between internal and arm’s-length modes depends on issues such as moral hazard, adverse selection, hold-up, contract enforcement, and intellectual property protection. —Trade in business services (non-factor, non-trade-mediating services). There was an earlier wave of interest in trade in business services in the late 1980s, in Canada in particular. In my view (and I was a participant) the theory that came out of this was not very successful. Several authors got bogged down in trying to define services, an elusive goal, as Daniel Trefler (this volume) has so nicely indicated with a quote from Justice Potter Stewart. One traditional view of business services is that they are hard to trade, requiring the spatial and temporal proximity of supplier and customer. Herbert Grubel (1987) went so far as to argue that all trade in services is embodied in goods or persons. It is very clear from the topics we are considering today that this view is at best badly outdated. One area where progress has been made is in the theory of the multinational. The modern view is that parent firms are exporters of the services of knowledgebased assets to foreign subsidiaries (although goods and intermediates are often traded as well) (Markusen 1995, 2002). —Liberalization: trade expansion at the extensive margin. Much traditional trade theory involves liberalization expanding the volume of trade in existing traded goods. We could call this expansion at the intensive margin. But these models do not seem appropriate to the current discussion, in which we are looking at new things being traded. Some existing theory bears on this. In comparative-advantage models, liberalization expands trade at the intensive margin, but some “middle” goods can become traded as trade costs fall, as in the DornbuschFischer-Samuelson (1977) Ricardian continuum model. Yi (2003) has a neat Ricardian model in which goods are produced in distinct stages of production
James R. Markusen
5
that can be geographically fragmented. Other valuable empirical evidence is found in Hummels, Ishii, and Yi (2001). In the theory of the horizontal multinational, investment liberalization allows intrafirm trade in the services of knowledge-based assets, so more things are traded. For vertical multinationals and arm’s-length offshoring, innovations in technology, liberalization, or institutions (intellectual property protection) allow fragmentation of the production chain so that more things are traded: capitalintensive intermediates go out, labor-intensive assembly takes place abroad in labor-abundant countries, with much of the final output shipped back to the parent country. In all of these models, trade expands at the extensive margin.
Empirical Characteristics of Offshoring of White-Collar Services to Capture in Theory Models Here is a wish list of characteristics for theoretical models of offshoring of white-collar services. —Expansion of trade at the extensive margin: new things traded due to innovations in communications and technology. This poses a number of challenges to theory, especially the fact that we are talking about nonmarginal changes and discrete movements of something being nontraded to potentially lots of trade. Traditional comparative-statics analysis is of little use: it focuses on marginal changes in activities that are already in use in the benchmark. —Vertical fragmentation of production: the new traded services tend to be intermediates, and they may be upstream, downstream, or not part of a sequence. Traded white-collar services often have important characteristics that cannot be captured in the simplest off-the-shelf models, which assume a set of final goods. One is that they may be firm-specific rather than bought and sold in arm’s-length markets. Another is that they may form part of a particular production sequence, such as a well-defined upstream (design) or downstream (after-sales service) component of overall production. A third is that there may be crucial complementarities among different elements of the production chain, such as between skilled labor and telecommunications equipment and infrastructure. —Offshoring of medium-skilled or even highly skilled services to skilledlabor-scarce countries. Is this at odds with factor-proportions theories? The simplest off-the-shelf 2 x 2 Heckscher-Ohlin model is not going to offer insights into why relatively skilled-labor-intensive services are being offshored to very skilled-labor-scarce countries. One of the most important tasks of theory, in my
6
Brookings Trade Forum: 2005
opinion, is to develop richer but empirically plausible models of why this is taking place. Yet the factor-proportions approach to trade does not necessarily have to be abandoned; but it must be enriched to include multiple goods or factors, or both, so that fragmentation and the complementarities just discussed can be analyzed. —Reversal in the direction of trade from existing multinationals models. New offshoring is exporting services back to high-income-country firms (intrafirm back to parents or via arm’s-length contracting). Trade in white-collar services is not new. The modern theory of the multinational has emphasized that parents are exporters of white-collar services, including management and engineering consulting, marketing, finance, and others, to their subsidiaries. One thing that is relatively new and that has generated much of the current interest is the reversal in the direction of trade that we are seeing. In some ways this is closely related to the previous point. —Firms, or specifically owners of knowledge-based assets, may offshore skilled-labor-intensive activities that are complements to these assets. A plausible worry is that skilled workers in the high-income countries are being hurt while their companies profit from offshoring. This cannot be dismissed and requires investigation. To me, it calls for at least a three-factor model, in which firms possess specific factors or other assets that are complements to skilled labor. One example is software engineers as complements to telecommunications equipment and network infrastructure, in which the third factor is physical capital. Or it could be that software engineers are complementary to managerial sophistication, organization infrastructure, and marketing channels. The complementary input is knowledge-based assets. Without services trade, you can train an engineer in India, but there will be no demand for his or her skills if there is nothing useful to do. The implication is that, in the absence of offshoring, these skilled workers are cheap even though they are relatively scarce in comparison with the availability of skilled workers in the country with the complementary factors. Offshoring that allows trade in the third factor causes that factor (or its services) to move to the skilled-laborscarce country to combine with cheap skilled labor there. This setup obviously has the elements of a story in which skilled labor is harmed in the high-income country, while owners of the complementary physical or knowledge-based assets benefit. This phenomenon is relatively easily modeled either in a competitive multifactor model or using Markusen’s knowledge-capital approach to the multinational. The latter approach has also proved a useful starting point for looking at
James R. Markusen
7
the internalization versus outsourcing decision in relation to the offshoring decision. The idea is that transacting in knowledge-based assets creates special problems for the owner (the multinational firm). For example, there are several laborturnover models in which workers in the host country absorb or learn the substance of the knowledge-based assets and can defect to start rival firms. Other issues that have been considered in the theory literature involve asymmetric information, reputations, and hold-up. The next few sections of the paper construct and analyze some simple template models that incorporate these features. All of the models presented have been coded into numerical simulation models using GAMS. Code for these models is available from the author. An appendix to the paper lays out the structure of model 1.
Models Model 1: A 3 x 2 x 2 Heckscher-Ohlin Model with Fragmentation Suppose we begin with a simple two-final-good, two-factor, two-country Heckscher-Ohlin model and then allow one good to fragment into two separate production activities, giving three in total. If we assume free trade, just considering free versus prohibitive fragmentation costs, we do not need to specify which is the upstream and which is the downstream activity. For a much more comprehensive treatment of this case, see Markusen and Venables (2005). For a more general approach, see Deardorff (2001, 2005a, 2005b). Here are the principal features of the model. (A) Two factors of production: skilled (H) and unskilled (U) labor (B) Two final goods, three production activities Y: unskilled-labor-intensive X: skilled-labor-intensive X: can fragment into high-tech manufacturing (M) and services (S) M: more skilled-labor-intensive than X S: middle skill intensity: less than X, more than Y (C) Two competitive, constant returns economies North: high-skilled-abundant South: low-skilled-abundant The service component of good X is thus chosen to have a middle factor intensity between integrated X and good Y; specifically, the complete ranking from most to least skill intensity is: M > X > S > Y. This choice definitely matters for
8
Brookings Trade Forum: 2005
the results. We are thinking here of things like business process outsourcing or call centers, which are less skill intensive than the overall industry, but more than a developing economy’s traditional sector of comparative advantage.2 I will report and analyze the qualitative results of numerical simulations. Begin with fragmentation banned; that is, M and S used for X must be produced in the same location. For this case, I calibrated the model so that the two countries are initially specialized in X (North) and Y (South) in free trade: factor endowments have a bigger spread than factor intensities. Now allow for the geographic fragmentation of X production. This results in some or all of the middle-skill-intensive service activity switching from North to South, with services exported back to North or M exported to South to be combined with S. This does not really matter with free trade, except that measured changes in trade volume will depend on which is which. For our purposes, it is perhaps better to think of the services as exported back to North, where they are combined with M to produce the completed good.3 There is a fundamental tension that arises in general equilibrium when the ability to fragment manufacturing and services is introduced. 1. Services, which are middle-skill intensive, shift from North to South, increasing the relative demand for skilled labor in both countries. This is an idea familiar from Feenstra and Hanson (1996a, 1996b, 1997) and also arises in multinationals models (Zhang and Markusen 1999). North sheds an activity that is unskilled-labor-intensive from its point of view, but South gains an activity that is skilled-labor-intensive from its point of view. Thus we expect the real and relative price of skilled labor to rise in both countries. 2. However, general equilibrium is bedeviled by terms-of-trade (TOT) effects: North moves from integrated X production to exporting M and importing S. A fall in the relative price of M harms North, possibly outweighing efficiency gains for North. The ability to fragment X production has an effect loosely related to a technical improvement. South can produce S more cheaply than integrated
2. In his comments on this paper, Douglas Irwin quite properly wonders about the robustness of results based on one particular ranking of factor intensities, yet also wants to avoid sliding down the “slippery slope” into taxonomy. I agree with both thoughts. My decision is to concentrate on a central case that I find the most empirically plausible: the offshored service has a middle intensity between Y and integrated X production. 3. Furthermore, allowing the service to be provided by southern workers is close to the same thing as moving foreign workers to North. If allowing the service to move results in factor-price equalization, they are exactly the same (provided the welfare of each country is that of its original residents).
James R. Markusen
9
North, and North can produce M more cheaply. But as the countries begin to specialize, their relative size will do a lot to determine the relative price of M versus S. The equilibrium relative price of M to S is higher the smaller North is. When North is large, an adverse terms-of-trade change can make it worse off than before fragmentation despite the efficiency gain. When North is not large, efficiency gains outweigh the terms-of-trade shift and both countries benefit.4 Closely related to this are two results that emerge from the simulations. First, my results indicated that while skilled labor is the relative gainer in North, both factors could suffer an absolute loss of real income when North is large. This occurs with a low equilibrium relative price for M as just mentioned. Second, results indicate that skilled labor is an absolute gainer in South, but that it might be a relative loser when South is large. South shifts its output to a more skilledlabor-intensive sector, but that sector (services) suffers a price fall relative to the no-fragmentation case. The latter effect is large when South is large: skilled labor gains absolutely but loses relatively. Results from my simulations over a range of parameter values can be summarized as: MODEL 1 RESULTS 1. South gains, North loses if North is large; both countries gain if South is large.5 2. Skilled labor is the relative gainer in North (but a real-income loser if South is small: TOT effect dominates). 3. Skilled labor is a real-income gainer in South (but gains relatively less if South is large: TOT effect dominates). 4. Unskilled labor is a real-income gainer in South, a loser in North. 5. Volume of trade in goods increases: goods and services trade are complements (but can fall if South is small: South is self-sufficient in S, does not export). 4. An alternative intuition about the terms-of-trade effect is as follows. North has a factor endowment that is well suited to integrated X production. When fragmentation is allowed, the equilibrium price of X falls, harming North, which specializes in X. The question is whether shifting to specialization in M more than recoups this loss. The answer is yes if North is not large. Perhaps this intuition also shows why South always gains: this terms-of-trade effect against X must benefit South. 5. A finding that North can lose is not new. Gomory and Baumol (2004) note this in a model with increasing returns. Samuelson (2004) shows a case with constant returns and perfect competition. Many other such cases occur in models with multinationals (Markusen 2002). Note that some results “guaranteeing” gains from trade compare autarky to free trade with fragmentation (Deardorff 2005a). To me this is irrelevant: the relevant question is comparing free trade in goods to free trade in goods and services.
10
Brookings Trade Forum: 2005
Despite having done countless runs of this model, I cannot guarantee that there are no other possibilities, and of course, reordering the factor intensities will change the results. What I can say is that it is easy to find ranges of parameters that generate these results, but we should all regard them as suggestive and not definitive. Model 2: A 3 x 3 x 2 Missing Input Model with Fragmentation My second model is designed to capture the idea that skilled labor can be cheap where it is scarce. It again has three production activities and two countries, but three factors. (A) Three factors of production: skilled (H) and unskilled (U) labor and know-how (K). K could be high-tech physical capital, such as telecommunications equipment and networks, or knowledge capital (managerial techniques, organization infrastructure, marketing channels). (B) Two final goods, three production activities Y: unskilled-labor-intensive X: skilled-labor- and know-how-intensive X: can fragment into high-tech manufacturing (M) and services (S) M: more skilled-labor-intensive than X S: skilled-labor- and know-how-intensive (C) Skilled labor and know-how are complements in the production of S (D) Two competitive, constant returns economies: North: high-skill- and know-how-abundant: South: low-skill-abundant, very know-how-scarce The complementarity between skilled labor and know-how in producing S is crucial. Specifically, this is modeled as a very low elasticity of substitution between H and K in producing S. When a country is very know-how-scarce, there is little for its skilled workers to do. You can train engineers, but there are no jobs for them. Assume initially that K (or its services) and skilled workers cannot move between countries. K is used with skilled workers largely in the North, which exports integrated X. The fundamental tension caused by introducing trade in K and S is now going to occur between northern and southern skilled workers. 1. Skilled labor is initially cheap in South (even though scarce) owing to a lack of K to work with. 2. Skilled labor in North and in South compete directly, introduction of trade in K moves K to South, shifting relative demand for skilled labor to South.
James R. Markusen
11
The introduction of the third factor, complementary to skilled labor, makes it straightforward to create a situation where skilled labor is initially cheap where it is scarce. In addition, the model is less sensitive to country-size issues, at least with respect to factor prices. But as in the previous model, there is a terms-oftrade issue for North. The ability of the owners of K to move their factor to South to work with cheap skilled labor there erodes not only the return to skilled labor in North, but also North’s implicit monopoly power over good X. The result in all of my simulations was that welfare decreases in North when trade in K and S is permitted and North is large. As in the previous model, both countries gain when North is not large. Here are my results for permitting trade in K and S. MODEL 2 RESULTS 1. South gains, North loses when North is large; both gain when North is not large. 2. Skilled labor is real-income loser in North, absolute gainer in South. 3. Real return to know-how rises in North, falls in South. 3. Unskilled labor is real-income gainer in North, loser in South. 4. Volume of trade in goods increases (complements services). Losses to North and northern skilled labor in particular are two of the things that analysts have worried about with respect to the offshoring of white-collar services. This model potentially validates the worry that northern business owners or owners of particular physical capital and knowledge-based assets will benefit considerably at the expense of northern skilled workers. Of course, the model is in part deliberately designed to do that, so this is hardly a coincidence. On the one hand, I cannot say with confidence that a thorough search would not lead to alternative models with quite different results. On the other hand, I would not have put this model forward if I did not find it empirically plausible and relevant. Model 3: A 3 x 2 x 2 Knowledge-Capital Model of Multinationals Now I would like to return to something close to model 1, but add in multinational firms following the knowledge-capital model of the multinational that I developed some time ago. This version of the model is based on Zhang and Markusen (1999). Markusen (2002) is the best source for the complete development of the theory, and this section is based on Chapter 9. (A) Two factors of production: skilled (H) and unskilled (U) labor (B) Two final goods, three production activities Y: unskilled-labor-intensive, constant returns, perfect competition
12
Brookings Trade Forum: 2005
X: skilled-labor-intensive, increasing returns at the firm level: firm and establishment-level fixed costs, constant marginal cost X: can fragment into high-tech manufacturing (M) and services (S) M: more skilled-labor-intensive than X. By assumption, only North can produce M. S: middle skill intensity: less than X, more than Y (C) Cournot output competition by X firms, free entry and exit in two firm types National firms: produce M and S in North, export X to South Multinational firms: produce M in North, which is exported to South, where S is produced, or vice versa (D) Two economies North: high-skill-abundant South: low-skill-abundant The reader will see that this resembles model 1 insofar as X can fragment into a skilled-labor-intensive phase and a medium-skilled-labor intensive phase. I have modeled the S phase as largely unskilled-labor-intensive in marginal costs, but establishment fixed costs as having a sizable skilled-labor component. I do not think that this affects the results. A nice feature of this model, aside from its probable empirical relevance, is that it avoids the “curse of Stolper-Samuelson”6 and the terms-of-trade effects that are so important in the competitive, constant-returns models. Because of procompetitive effects leading to increased firm scale and lower markups, it is entirely possible that both countries and all factors gain following a liberalization.7 The way this works in the present model is straightforward. Again, begin in a situation where trade in disembodied S is not allowed: S and M must be produced together. This is equivalent here to not allowing multinationals to enter. Having to use North’s factor endowment for both M and S is a binding constraint on the world economy and limits the number of firms in free-entry equilibrium, which in turn leads to a high markup and a low output per firm (high average cost). When this constraint is relaxed by allowing firms to fragment X, much, perhaps all, of service production moves to South. This again tends to have the Feenstra-Hanson effect of raising the relative demand for skilled labor 6. I didn’t invent this phrase, though I wish I had. I think I heard it first from Ron Jones. 7. Alternatively, we could model the final goods or intermediate services as differentiated, using the now well-known large-group monopolistic-competition framework. As far as welfare is concerned, there is a benefit from increased variety analogous to the procompetitive effect of the oligopoly model that tends to generate large welfare gains. See Ethier (1982) and extensions by Markusen (1989).
James R. Markusen
13
in both countries. But now this also increases the profitability of the existing firms, which leads to entry, which in turn leads to lower markups and higher output per firm (lower average costs). There are, however, a lot of possibilities depending on relative endowments and intensities and again on country size. Chapters 8 and 9 of Markusen (2002) show that many outcomes are possible. I can say that it is easy to find parameter values for which allowing fragmentation leads to welfare increases for both countries and gains for skilled labor in both countries. I have to admit that I did not find a set of parameters for which the real prices of all four factors increase, however. I tended to find that the real return to unskilled labor in North fell following fragmentation and trade in services. Here are some typical, but not robust, results for the effects of allowing multinationals to enter, equivalent here to allowing trade in services. MODEL 3 RESULTS (for a range of parameterizations) 1. South gains, North gains. 2. Skilled labor is a relative and an absolute gainer in North and South. 3. Unskilled labor loses in North and gains in South. 4. Procompetitive effects lead to more firms, lower markups, higher output per firm. Model 4: A 3 x 3 x 2 Model That Combines the Knowledge-Capital Model with the Missing Factor Model Our fourth template combines the knowledge-capital model with the missing factor model. I take the skilled labor in North and assume that some portion of it is factor K, which is complementary to skilled labor in producing establishment fixed costs. In fact, I coded up this model first and then moved to model 3 by simply allowing the substitution between K and H in producing establishment fixed costs to move to infinity. Otherwise, the models are the same. In the initial equilibrium, trade in K (or the services of K) is not permitted; alternatively, multinational firms are not permitted to enter. These results are then compared with allowing trade in K, or equivalently allowing multinationals to enter. Again, a range of outcomes is possible. But for the same parameterization that model 3 just indicated, the liberalization here generates a stronger adverse terms-of-trade effect for North. North’s welfare declines if North is large. As with model 2, now the skilled labor in North competes directly with skilled labor in South. The introduction of multinationals moves K from North to work with initially cheaper skilled labor in South. This lowers the real return to skilled labor in North with the big beneficiary being owners of the factor K.
14
Brookings Trade Forum: 2005
MODEL 4 RESULTS 1. South gains, North loses if North is big; both gain if South is big. 2. Skilled labor is relative and a real-income loser in North, a gainer in South. 3. Large gain in North for the owners of know-how. 4. Unskilled labor gains in both countries. 5. Procompetitive effects lead to more firms, lower markups, higher output per firm.
The Offshoring-Outsourcing (Location–Mode Choice) Relationship Internalization or its inverse, outsourcing, is a decision about the boundaries of the firm and what activities to keep inside or internal to the firm’s ownership structure and which to contract to arm’s-length firms. Multinationals offshore but do not outsource, keeping their foreign activities within owned foreign affiliates. Firms that contract or license in some way to foreign firms are engaging in both offshoring and outsourcing. As I indicated earlier, the internalization decision, also known as mode choice, has a long history of analysis, particularly in the international business literature. Its rediscovery under the name outsourcing has coincided with many researchers overlooking this long tradition. In any case, the traditional focus of the internalization/outsourcing decision has been on the various transactions costs, particularly when offshoring, of doing business at arm’s length rather than internal to the firm. It is important to keep offshoring and outsourcing decisions distinct: they are location choice and mode choice decisions, respectively. A factor that encourages outsourcing might at the same time discourage offshoring in favor of exporting from the home country or choosing a third country. There are in fact a number of factors associated with producing abroad that do precisely this: they encourage outsourcing but discourage offshoring. Some of these are: —restrictions on foreign investment —restrictions on the right of establishment —restrictions on immigration (generally temporary) of foreign business personnel —lack of intellectual property protection —lack of contract enforcement The first three in this list generally follow from the fact that offshoring requires setting up a foreign subsidiary, which in turn requires foreign investment and the use of home-country personnel in the host country for some period of time.
James R. Markusen
15
Thus, problems in any of these three areas would encourage a firm to outsource to a local firm, but they also discourage offshoring relative to other outside alternatives. The last two points involve various aspects of moral hazard and hold-up when firms make investments, both sunk physical capital and investments in training local workers, in connection with establishing a subsidiary. Again, they tend to encourage outsourcing but discourage offshoring. The problem of transferring knowledge and skills probably exists for both modes of offshoring, about which I will say more shortly. Two formal approaches found in the theory literature may be useful. The first I will call the “labor-learning model.” Variations of this are set forth by Ethier and Markusen (1996), Fosfuri, Motta, and Rønde (2001), Markusen (2001), and Glass and Saggi (2002). All of these papers discuss multinationals that make a foreign investment that is profitable owing to knowledge-based assets of the firm. However, workers in the host country acquire this knowledge themselves and can later defect to start a rival firm. If binding contracts cannot be written, then the firm will have to pay these workers a premium in subsequent periods to hold them in the firm. Thus, the multinational must share rents with local employees if contracts cannot be written or enforced. A second promising line of research involves the Grossman and Hart (1986) hold-up model, which has been developed in a series of papers by Antràs (Antràs 2003, 2005; Antràs and Helpman 2004). Here the idea is that the multinational firm and a local agent must each make sunk, relationship-specific investments in a project. In the absence of complete contracts or contract enforcement, this creates a bilateral ex post hold-up problem. The optimal mode of entry is generally that ownership, defined as residual rights in assets if bargaining breaks down, should go to the party with the larger sunk investment. As in the labor-learning model, this approach requires the multinational to share rents with a local agent whether or not that agent is the manager of an owned subsidiary or the owner of an arm’s-length contractor. Alternative assumptions can produce alternative correlations between offshoring and outsourcing. Suppose that a firm wishes to supply a product X in South. If the fixed costs of setting up a foreign plant are not too large but large relative to the sunk investments of the local partner, then the firm will tend to choose both offshoring (in preference to exporting) and internalization: a negative correlation between offshoring and outsourcing. A difference between the labor-learning model and the hold-up model is that, in the latter, both the multinational and the local manager make ex ante sunk investments that generate bilateral hold-up. In the labor-learning model, workers acquire bargaining power ex post as they learn. I think that both approaches
16
Brookings Trade Forum: 2005
have something to contribute to the offshoring of white-collar services. There is no question that there are a lot of training costs for the foreign workers. Information technology (IT) and business process outsourcing (BPO) activities are often learned on the job by workers who already possess good general skills. It is my understanding that many call-center workers are trained by independent firms before landing a call-center job. Here is my suggestion for one approach that combines the labor-learning approach and the sunk-cost hold-up approach. Think of this as, perhaps, a model of business process outsourcing or call centers. Model 5: Template for an Integrated Outsourcing-Offshoring Approach (A) Begin with the “missing input model,” two time periods. (B) Interpret this as firm-specific knowledge capital à la Markusen’s knowledge-capital model: skilled workers in the host country are cheap because they lack crucial physical or knowledge-based inputs. (C) With appropriate technological and institutional conditions, this asset can be “exported” by a firm (used abroad) in combination with local skilled workers. (D) However, local workers “absorb” the relevant knowledge and are able to “defect” to start rival firms on their own in the second period. (E) Also assume a capital investment in land, structures, and telecommunications is needed. Whoever owns this defines whether the project is a subsidiary (internalized) or an arm’s-length relationship (outsourced). The cost must be borne by the multinational. I suppose that many researchers in the international business field would conjecture that given complete and enforceable contracting, the firm would prefer outsourcing on a simple cost basis, so let us make that assumption. MODEL 5 RESULTS (conjecture! this paper has not been written!) 1. Given complete and enforceable contracts, outsourcing is preferred (by assumption). 2. If contracts are not enforced, then the multinational will want to own the physical capital—that is, internalization is chosen by the Grossman-Hart-Antràs argument. 3. However, even if it is possible to contract for physical capital (local firm contracts to pay a mortgage), the firm may still want to own it if it is not possible to contract for the intellectual capital (skills) that is transferred to local workers, in order to reduce the ex post hold-up problem of skilled workers threatening to leave.
James R. Markusen
17
4. On the other hand, if the learned skills of the foreign manager are relationship-specific—that is, they are only useful to the contracting multinational—then there is limited hold-up from the workers and outsourcing would be preferred. Indeed, in this case it seems as though the firm would have ex post hold-up power, and so the manager would want to own the capital. Again, this is conjecture. I am working on this project, but not yet certain of the results. As a final point, recall again that the agency costs and rent-sharing costs to the firm, whether they be less in the internal or the outsourcing mode, also affect the firm’s offshoring choice. For a firm seeking to serve the local host market, these costs may lead the firm to choose exports rather than offshoring. For a firm seeking to serve its own home market, these costs may lead it to choose domestic outsourcing or search for a third supplier.
Summary and Conclusions I have argued in this paper that we can make good progress in understanding the offshoring of white-collar services at the theory level from our existing portfolio of models. Many important features of offshoring white-collar services can be modeled from a recipe that mixes and matches elements from the existing inventory. Useful elements from our portfolio include: —vertical fragmentation of production —expansion of trade at the extensive margin —fragments that differ in factor intensities, countries that differ in endowments —knowledge stocks of countries or firms that are complementary to skilled labor; these create missing inputs for countries otherwise well suited to skillintensive fragments —knowledge-based assets that create particular contractual and agency issues for firms engaging in international business. Existing models of laborlearning and hold-up are useful places to start in considering the outsourcing (mode) choice in relation to the offshoring (location) choice. These features allow us to construct relatively simple and tractable generalequilibrium models that predict changes induced by fragmentation on aggregate welfare, factor prices (income distribution), the location of production activities, and the direction and volume of trade. While I view this paper as listing a number of plausible and empirically relevant ways of modeling the offshoring of white-collar services, it was clear at the conference that many people were much more interested in specific results from these models. Unfortunately, it is hard to offer robust conclusions, especially
18
Brookings Trade Forum: 2005
about the aggregate welfare of countries. Trade theorists are well aware of the underlying problem: any move from partial liberalization to more liberalization (allowing more things to be traded in our case) often does not result in Pareto improvements for the trading partners. It would be intellectually dishonest for me to report only cases in which everyone is better off. Overall, my simulation models suggest a clear gain for world welfare and for South in particular, but North may lose if it is large. This result is very familiar to all trade theorists in different contexts; for example, a large country may well prefer a Nash equilibrium in tariff rates to free trade with a small country. Stephen Magee and Kwang-Yeol Yoo (2005) have argued persuasively that the United States is not a large country in the sense of my models, which might give us some comfort. Results on factor-price changes are interesting and consistent with a range of existing literature. In the two-factor models, skilled labor is the relative and (usually) absolute gainer in both regions, as activities that are not skill intensive from North’s point of view are transferred to South, where they are. Results for unskilled labor are more mixed. I have been asked to indicate which model may fit reality the best, and I have to say that I think that the three-factor “missing input” model is my favorite, preferably with multinationals. I have called the third factor “knowhow”: it could be knowledge capital, high-tech physical capital, or highly skilled knowledge workers, including management. I started working with this model in connection with Central and Eastern Europe, where the productivity of workers with excellent skills in math, science, and engineering was very low: they were missing the crucial organizational, managerial, quality control, and marketing skills needed to complement their other skills. Many case studies I have read about East Asia suggest that the same circumstances prevail. You can educate scientists or engineers, but if there is nothing for them to do they will not be productive. I capture this by making North rich in know-how as distinct from general skilled labor, and by making know-how a strong complement to general skilled labor. The result is that skilled labor is cheap in South where it is scarce, a principal stylized fact that has generated much of the white-collar offshoring. This model sets up a tension between the general northern skilled labor and the southern skilled labor; perhaps routine programmers and routine businessprocess workers are examples. Allowing fragmentation moves know-how, or rather the services of the know-how, to South, generating big gains from the owners of the know-how and losses for general skilled labor in North. We should keep in mind, however, that much of the know-how is surely embodied
James R. Markusen
19
in the human capital of highly skilled tech workers, managers, marketers, and so forth, and this may suggest that second-level white-collar workers are the ones who are most at risk. I will close with another caveat about theory. Strong and robust findings about welfare gains from fragmentation are not forthcoming from a general approach to theory. Alternatively, it is usually possible to find some strange model that generates whatever result a client wants. I have tried to construct models that I feel are plausible and relevant, but some residual ambiguity remains. I hope I have at least left us with some sense of why this ambiguity exists.
Appendix Example of the Simulation Models Used in the Paper The models in the paper seem simple enough. The first model begins with the classic 2 x 2 x 2 workhorse model of trade theory. All students of economics learn the analytics of this model, many of these as undergraduates. Any economist can reasonably conjecture that introducing the ability to geographically fragment some activities should still permit analytical solutions. Unfortunately, it is not nearly that simple. Let us take model 1 as an example, our simplest model. We go from two to four production activities: Y, M, S, and final X instead of just Y and X. The number of possible production specialization patterns for a country goes from three to fifteen (this assumes that you have to produce some of something). In addition, there are a great many more possible trade patterns, the number going from two to fourteen. In other words, the dimensionality of the model increases greatly, making the simple analytical methods we are used to much less useful. Second, the entire model must now be formulated in terms of inequalities, not equations. We do not know which of these will hold for a particular set of parameters (for example, which production activities and which trade activities are slack), and, indeed, the set that holds with equality will typically change a lot when parameters are changed. Allowing fragmentation can, for example, reverse the direction of trade in X or Y, or both. The models are termed nonlinear complementarity problems in math programming language: each weak inequality is associated with a non-negative complementary variable. If an inequality holds as a strict equality in the solution, the complementary variable is positive; if it holds as a strict inequality (for example, marginal cost exceeds price), the complementary variable (quantity in this case), is zero. Traditional comparative-statics techniques used on sets of equalities are of no
20
Brookings Trade Forum: 2005
use. If you read some of the existing literature on production fragmentation, you will then understand why almost none of it actually solves for a world general equilibrium. Thus, I have chosen to approach the template models using simulations. I use software from GAMS, which has the only powerful and robust nonlinear complementarity solver in the business. All models consist of three sets of inequalities and complementary variables. First, there are zero-profit inequalities for each production and trade activity: marginal cost is greater than or equal to price. The complementary variable is the output or “activity level” of that activity. A quantity variable is complementary to the price inequality. Second, there are market-clearing inequalities: the supply of a commodity (good, factor, import, or other) is greater than or equal to its demand. The complementary variable is the price of that commodity. A price variable is complementary to a quantity equation. Finally there is an income balance equation for each country. In this appendix, I give the set of inequalities for model 1. There are thirtyfour production and trade activities, twenty-four “commodities” (final goods, intermediate goods, imports and exports, and utility, which is treated as a good produced from inputs of X and Y in the code), and two income levels. Walras’s Law makes one equation redundant: I fix the world price of Y at 1 and drop the world market-clearing equation for Y. Thus the entire model consists of fiftynine weak inequalities, each with an associated non-negative variable. In the body of the paper, I introduced only the notation needed to describe the models in basic economic terms. Here are the definitions of additional notation needed for the formal model. pki pcki pk wji ck(...) EKi IKi Wi Ii
producer price (that is, marginal cost) of good k in country i (k = Y, X, M, S; i = North, South) consumer/import price of good k in country i world price of good k price of primary factor j in country i (j = U, H) unit cost of producing good k (includes “cost” of producing utility: the unit expenditure function) exports of good k by country i imports of good k by country i welfare of country i income of country i
Other:
James R. Markusen
21
(1) market-clearing inequalities make extensive use of Shepard’s lemma, in which the unit demand for a good or factor is the derivative of the unit cost function with respect to the price of that good or factor. (2) very small trade costs, 0.05 percent, are used to prevent “ties” or, more formally, model degeneracy. This prevents, for example, its being equally profitable to both import and export a good, which would lead to indeterminacy of gross trade flows and possibly a failure to solve (infinitely many solutions differing in gross trade flows). (3) solutions without fragmentation permitted are computed by constraining import and export activities for S and M to be zero. (4) all production activities using more than one input are Cobb-Douglas, with shares as follows: Y U: 0.70 H: 0.30 M U: 0.17 H: 0.83 S U: 0.60 H: 0.40 X M: 0.70 S: 0.30 implied shares of primary factors in X produced from domestic M and S are the inverse of share for Y. X U: 0.30 H: 0.70 (5) factor endowment ratios for North and South are symmetric: North H: 0.90 U: 0.10 South H: 0.10 U: 0.90 The model is thus symmetric between North and South and between X and Y without fragmentation. Endowment ratios have a bigger spread (9/1) than intensities (7/3). If countries are the same size, then the no-fragmentation equilibrium is symmetric with both countries specialized, a goods terms of trade of one, factor-price ratios that are inverses in the two countries, and equal welfare across countries.
22
Brookings Trade Forum: 2005 Inequalities zero profit inequalities
Complementary variable activity levels
description (no. of inequalities) production of Y in i (2) production of M in i (2) production of S in i (2) production of X i from Si , Mi (2) production of X i from Sj , Mi (2) production of X i from Si , Mj (2) production of X i from Sj , Mj (2) exports of M by i (2) imports of M by i (2) exports of S by i (2) imports of S by i (2) exports of X by i (2) imports of X by i (2) exports of Y by i (2) imports of Y by i (2) home supply of Yi to i (2) production of welfare in i (2) supply - demand of Y prod (2)
supply - demand of Y cons (2)
supply - demand for M i (2)
supply - demand for imported M (2)
James R. Markusen market clearing inequalities
23 prices
description (no. of inequalities) supply - demand for Si (2)
supply - demand for imported S (2)
supply - demand for X i (2) world market and price for Y (1) world market and price for M (1) world market and price for S (1) world market and price for X (1) supply - demand for utility (2) market for unskilled labor in i (2)
market for skilled labor in i (2)
Income balance inequalities
incomes
description (no. of inequalities) income balance in country i (2)
Comments and Discussion
Alan V. Deardorff: Jim Markusen has done an excellent job of putting the phenomenon of offshoring into perspective. He has done this by drawing upon models of international trade and foreign direct investment to tell a series of stories about why offshoring may occur and what effects it will have. These stories take the form of several explicit models, for each of which he explains the structure and reports results. The results reported are only qualitative, but they are based on what were apparently numerous quantitative simulations. As I read his paper, I found the explanations given very intuitive and helpful. I especially appreciated his emphasis on why scarce factors may be cheap in some countries despite their scarcity, a result of being complementary to another factor that is even scarcer. As he puts it, scarce factors have nothing to do. Offshoring to take advantage of these services may make the services of the complementary factor available, benefiting both them and the world. I like that story a lot. When it came to the results of the models, however, I found myself more confused. Each model had its own list of results, and I had to keep flipping pages to compare them. So the main contribution I will make in these comments is to put these results into one place. In table 1, I report both the assumptions of Markusen’s models 1–4 and the results, all taken directly from his paper. Below the table is the notation I use, supplementing Markusen’s factors H, U, and K and his goods/fragments Y, X, M, and S with factor prices s for skilled labor, u for unskilled labor, and r for know-how. Most of the assumptions, at the top of the table, are common to all four models, which are distinguished by whether they have two or three factors and by whether they assume perfect or Cournot competition. 24
James R. Markusen
25
Table 1. Markusen’s Models: Assumptions and Results for Countries and Factors Model H-O with fragmentation 1 Assumptions Final goods Fragments (of X) Skill intensity ranking North abundance South abundance Factors of production Competition in X
2 (H, U) Perfect
Results North (country) South (country) North (s/w) North (s) North (w) North (r) South (s/w) South (s) South (w) South (r)
–a + + –a – … –a + … …
Missing Input (MI) 2
Multinational (MNC) 3
2 (Y, X) 2 (M, S) M>X>S>Y H (and K) U 3 (H, U, K) 2 (H, U) Perfect Cournot
–a + … – + + … + – –
+ + + + – … + + + …
MI + MNC 4
3 (H, U, K) Cournot
–a + – – + + + + + …
a. If North is large and South is small. Factors H = skilled labor Y = unskilled labor K = know-how Factor prices (real) w = unskilled wage s = skilled wage r = return to know-how
Goods and fragments Y = unskilled-labor-intensive final good X = skilled-labor-intensive final good M = high-tech manufacturing fragment of X production S = service fragment of X production
The bottom portion of the table shows the results that Markusen reports for welfare of the two countries and the two or three factors (in the form of their real factor prices) in each country as a result of introducing offshoring/fragmentation. I present simple plus and minus signs, except for those results that depend on country size due to terms-of-trade effects, in which case a footnote indicates that these signs hold only if North is a large country relative to South. In all cases my understanding from Markusen’s paper is that these qualitative results may not be valid for all possible parameters, even within the assumptions stated here,
26
Brookings Trade Forum: 2005
but that these signs represent the solutions for what he takes to be the most plausible range of parameters. The message I take from this collection of results is that almost anything can happen. Even within the constraints of the four models’ common assumptions, these apparently minor variations of assumed structure of competition (perfect versus Cournot) and number of factors (two versus three, with—importantly— the third factor complementary to one of the others) permits a mixture of both plus and minus signs in all but two of the main rows in this table. (I ignore the rows for return to know-how, since this factor does not appear in two of the models.) Only for country South as a whole and for skilled labor within South do the models yield an unambiguous conclusion: these two constituencies gain in all four models. Unskilled labor has mixed results in both countries, as does skilled labor in North and country North as a whole. This is not intended as a criticism of the models or of Markusen for presenting them. On the contrary, I think we need to know that sensible economic models do not provide a consensus message on some issues, and clearly that is the case for offshoring. As Markusen himself emphasizes, this should not be a surprise, since that has been true if one looks objectively at the literatures on other issues of trade theory, including even the gains from trade liberalization. In particular, it is indeed possible that the developed world, represented here by the country North, may lose from offshoring, and so especially may skilled labor within the developed world. This is not a new observation, and Markusen acknowledges others who have said it before, but it is at least helpful to have more light shed on the mechanisms by which this may happen, and Markusen’s models provide that light. There is one thing that Markusen does not do with his model that I wish he had, which would be to calculate the effect on welfare of the world as a whole. Admittedly, actual residents of the real world may not care about that, since their own welfare will be better tracked by the separate countries and factors. But a trade economist like myself would like to know the answer to this. I presume that the answer must be that the world as a whole must gain from the introduction of offshoring, since it represents an improvement in the efficiency with which resources are used worldwide, and the models here, or at least those with perfect competition, do not seem to have the sorts of market distortions that could render an improvement in efficiency harmful. But admittedly, the welfare effect on the world as a whole, even if it were reported, would not help us to resolve these ambiguities in effects on countries and factors. And that raises the question of how we might best go about resolving these ambiguities.
James R. Markusen
27
On that, I am even more at sea than Markusen’s models. Should we use case studies of individual episodes of offshoring for which more detailed information might be available? That would certainly be useful, but it could hardly be complete, since such case studies would probably be confined to single industries and they would miss the important general-equilibrium and terms-of-trade effects that drive models like Markusen’s. Alternatively, one might imagine econometric studies of offshoring together with national incomes that could provide estimates of the effect of the former on the latter. But surely, if we cannot even reach a consensus on the effects of trade more broadly on economic growth, then this approach does not seem very promising. Finally, Markusen’s own modeling suggests that perhaps we can build more elaborate models, incorporating the features he has here, but more deliberately replicating the data and parameters of the real world. Such computable general equilibrium (CGE) modeling has become commonplace for analysis of trade policy changes, and perhaps it can be applied productively here. But the truth is that such models rest very critically on assumptions made about model and market structure, which the data themselves are seldom able to inform. Seeing how sensitive Markusen’s simple models are to such assumptions, I doubt that CGE models applied to this problem could tell us much more than we already know here: that almost anything can happen. If that is the state of our knowledge about offshoring, what should be our policy advice? Knowing that losses from offshoring are possible, should we recommend that protectionist policies be employed to prevent it? Surely not, since in our ignorance we might as easily be depriving ourselves of benefits as of costs. Should we therefore advocate that offshoring be permitted to proceed unabated, regardless of the cost that it may impose? Perhaps, but if we are honest about our confidence that it will be beneficial, we may not be listened to. Maybe the best approach is not to condemn or to praise offshoring across the board, but to consider each example of offshoring on a case-by-case basis. That makes sense, except that I am not sure what information we should even want to have in order to judge it case-by-case. In the end, although I very much appreciate the insights that Markusen has given us with his series of models, I find myself knowing even less about the likely effects of offshoring than I did before I read his paper. Douglas A. Irwin: Let me begin by saying that I am very sympathetic to the sentiments expressed by James Markusen at the beginning of his paper—that trade theory is a portfolio of models, and that the art of economics is matching
28
Brookings Trade Forum: 2005
the phenomena you wish to analyze with the structural assumptions that you build into a model. As John Maynard Keynes wrote in the early 1920s: “The theory of economics does not furnish a body of settled conclusions immediately applicable to policy. It is a method rather than a doctrine, an apparatus of the mind, a technique of thinking, which helps its possessor to draw correct conclusions” (see Moggeridge 1983). I think this is the spirit in which Jim Markusen presents his paper. Markusen is off to a very good start in moving the analytics of the debate over offshoring forward. The contending views of Samuelson (2004) and Bhagwati, Panagariya, and Srinivasan (2004) represent the first published intuitions of trade economists on the issue, but the phenomenon cries out for a more detailed theoretical analysis. This is what Markusen does, mixing elements of standard trade models to match what we think is happening in international trade in services, and in the offshoring of white-collar workers in particular. The simulations he conducts will give all trade economists a new benchmark for thinking about the possible effects of offshoring on factor incomes and welfare at home and abroad. The one danger in this approach, however, is the risk of sliding down the slippery slope toward taxonomy. The problem is that the underlying phenomenon of offshoring is not without ambiguity (is it simply newly tradable services?). This means that we do not necessarily know the best modeling strategy for approaching it. It is not even clear what the comparative static exercise should be (a reduction in trading costs for services?). Since we are not sure how to model offshoring correctly, perhaps the best strategy (adopted by Markusen) is to play with a variety of structural assumptions. But this leaves us with a multitude of possible outcomes, and the varying results may not be too informative. Even with this danger in mind, Markusen should be commended for trying to sort out what possible outcomes emerge from a standard calibrated trade model simulation. In this comment, I will not take issue with the particular simulations that he undertakes, but rather make three points that suggest that we should take the results with a healthy dose of skepticism. First, in line with the title of the conference, Markusen assumes throughout that it is moderately to highly skilled labor that gets offshored. He does not hide the fact that his results hinge on a specific factor-intensity ordering, and indeed suggests the ordering is crucial to his findings. Yet the types of personnel that are commonly discussed as being offshored include (a) call center workers and secretaries (presumably unskilled labor), (b) financial analysts and traders (presumably moderately skilled labor), and (c) radiologists and information technology personnel (presumably highly skilled labor). It seems that, across a
James R. Markusen
29
variety of service activities, many different types of labor are potentially ripe for offshoring. As a result, I am not sure that we can generalize strictly on the basis of factor intensity what sort of outcomes we might expect. It could be that some interaction with technology, rather than the skill-type of labor, is the defining characteristic of what activities can be offshored to other countries. (This can be considered a call for more empirical work on who and what gets offshored.) Second, Markusen finds that, under a range of modeling approaches, North tends to lose from the tradability of services because of adverse terms-of-trade effects. This consistent finding reminds me of the tendency for older-generation, single-country-computable general equilibrium models employing the Armington assumption to find that a unilateral tariff reduction would be welfare worsening owing to adverse terms-of-trade effects—even for seemingly small, pricetaking countries such as Israel or Malaysia. Later study showed that these results were an artifact of modeling strategy (the Armington assumption of product differentiation in trade). This assumption built into the model a degree of market power that the countries did not actually possess. I raise this point as a reminder that the dominant terms-of-trade effect could, to some extent, be model driven. Alternative modeling approaches would likely lead to a different set of outcomes. For example, much of what is offshored seems to fall into the category of intermediate business services. If these services were built into an Ethier type of production function, where product differentiation and variety are valued, then making these services tradable would allow firms to add to the array of intermediates that they could potentially consume. Like trade in intermediate goods, the possibility of trade in these intermediate services might have positive productivity effects that would redound to the benefit of consumers in both North and South. Third, the work of Alan Deardorff (2005a, 2005b) in this area is relevant. Deardorff finds that defining comparative advantage is difficult in a world of Ricardian trade with intermediate inputs. In a Ricardian framework, Deardorff finds that trade patterns exhibit a high degree of sensitivity to trade costs. As my first comment suggests, it could be that sensitivity to trade costs is the important defining characteristic of offshoring, not a specific type of factor intensity. By contrast, in the pure technology case considered by Deardorff, allowing trade in intermediate goods almost always results in an economic benefit to North. (Deardorff’s other work on lumpy countries might help to explain why there is offshoring to countries abundant in unskilled labor. Both coastal China and Bangalore, India, might be viewed as agglomerations of physical or human capital within larger polities abundant in unskilled labor.)
30
Brookings Trade Forum: 2005
As a final point, the economic historian in me cannot help but observe that David Landes, in his book The Wealth and Poverty of Nations, noted that outsourcing and offshoring go back to thirteenth-century Europe. The putting-out system, whereby textile firms located in urban areas would employ cheap labor (primarily women) in the countryside as part of the production process, was opposed by the city guild officials and generated tremendous controversy at the time. I doubt that there were thirteenth-century modeling efforts to analyze the issue, but it is clear that offshoring and outsourcing are not necessarily new phenomena. General Discussion: Robert Feenstra launched the general discussion. He first raised the issue of whether India has abundant skilled labor. Recent data he has seen show that India’s share of the world’s scientists exceeds its share of world GDP. This is the correct criterion for factor abundance according to the Heckscher-Ohlin model. However, he noted that India is, of course, relatively much more abundant in unskilled labor. Feenstra also discussed the implications of offshoring for real wages of both skilled and unskilled labor in the North. In a model he developed with Gordon Hanson, it is possible for both types of real wages to rise, so that the North gains from offshoring. This reflects a Wal-Mart effect, in which offshoring raises world efficiency and reduces consumer prices. Thus, he questioned whether the results were as clear-cut as the models presented in James Markusen’s paper suggest. Shang-Jin Wei argued that a class of models was potentially missing from Markusen’s paper. He pointed out that services trade goes in both directions, with rich countries both importing from and exporting to developing countries. This feature is not captured in any of Markusen’s models. In fact, while media reports have been focusing on service outsourcing from the United States to India and other developing countries, the United States actually runs a surplus in services. He suggested that allowing for skill-intensive work to be offshored in both directions could provide a channel through which skilled labor in the North would gain. Dalia Marin discussed lessons from research, by herself and by others, that examines implications of offshoring to eastern Europe by Germany and Austria. The effects are difficult to tease out for Germany, because eastern Europe accounts for a very small percentage of its outgoing foreign investments. In contrast, most of the foreign investment from Austria goes to the region. She stated that the empirical analysis does support the conclusion that skilled Austrian labor has been hurt. In more recent work with more recent data she finds that German human capital has been hurt by outsourcing worldwide (including to eastern Europe). Her finding suggests that the distinction between manufacturing outsourcing and service outsourcing does not seem to be the right one, since Germany outsources
James R. Markusen
31
mainly in the manufacturing sector. What appears to matter is that the outsourced activity of German firms is almost three times as skill-intensive as the firms’activity in Germany. Catherine Mann stressed that severe measurement issues limit how much we actually know about the effects of trade in services for workers. Instead of simply jumping from trade to workers, she argued that we need to understand the mechanisms better: how services trade influences prices, which would affect production and technology, and through those channels affect employment. But we know very little about prices and costs for services, particularly in areas such as software, where services are often bundled together with goods. Many participants agreed that supporting the statistical community in finding ways to improve available measures should be a high priority.
32
Brookings Trade Forum: 2005
References Antràs, Pol. 2003. “Firms, Contracts, and Trade Structure.” Quarterly Journal of Economics 118 (4): 1375–1418. ———. 2005. “Incomplete Contracts and the Product Cycle.” American Economic Review (September): 1054–73. Antràs, Pol, and Elhanan Helpman. 2004. “Global Sourcing.” Journal of Political Economy 112 (3): 552–580. Bhagwati, Jagdish, Arvind Panagariya, and T. N. Srinivasan. 2004. “The Muddles over Outsourcing.” Journal of Economic Perspectives 18 (Fall): 93–114. Brainard, S. Lael. 1997. “An Empirical Assessment of the Proximity-Concentration Tradeoff between Multinational Sales and Trade.” American Economic Review 87 (4): 520–44. Deardorff, Alan V. 2001. “Fragmentation in Simple Trade Models.” North American Journal of Economics and Finance 12 (2): 121–37. ———. 2005a. “Ricardian Comparative Advantage with Intermediate Inputs.” North American Journal of Economics and Finance 16 (1): 11–34. ———. 2005b. “A Trade Theorist’s Take on Skilled-Labor Outsourcing.” International Review of Economics and Finance 14 (3): 259–71. Dornbusch, Rudiger, Stanley Fischer, and Paul A. Samuelson. 1977. “Comparative Advantage, Trade, and Payments in a Ricardian Model with a Continuum of Goods.” American Economic Review 67 (5): 823–39. Dunning, John H. 1977. “Trade, Location of Economic Activity and the Multinational Enterprise: A Search for an Eclectic Approach.” In The International Allocation of Economic Activity, edited by B. Ohlin, P. O. Hesselborn, and P. M. Wijkman, pp. 395–418. New York: Macmillan. Ethier, Wilfred. 1982. “National and International Returns to Scale in the Modern Theory of International Trade.” American Economic Review 72 (3): 389–405. Ethier, Wilfred, and James R. Markusen. 1996. “Multinational Firms, Technology Diffusion and Trade.” Journal of International Economics 41 (1): 1–28. Feenstra, Robert C., and Gordon H. Hanson. 1996a. “Foreign Investment, Outsourcing, and Relative Wages.” In The Political Economy of Trade Policy: Papers in Honor of Jagdish Bhagwati, edited by R. C. Feenstra, G. M. Grossman, and D. A. Irwin, pp. 89–127. MIT Press. ———. 1996b. “Globalization, Outsourcing, and Wage Inequality.” American Economic Review 86 (2): 240–45. ———. 1997. “Foreign Direct Investment and Relative Wages: Evidence from Mexico’s Maquiladoras.” Journal of International Economics 42 (3): 371–93. Fosfuri, Andrea, Massimo Motta, and Thomas Rønde. 2001. “Foreign Direct Investment and Spillovers through Workers’ Mobility.” Journal of International Economics 53 (1): 205–22. Glass, Amy Joyce, and Kamal Saggi. 2002. “Multinational Firms and Technology Transfer.” Scandinavian Journal of Economics 104 (4): 495–513.
James R. Markusen
33
Gomory, Ralph E., And William J. Baumol. 2004. “Globalization: Prospects, Promise, and Problems.” Journal of Policy Modeling 26 (4): 425–38. Grossman, Stanford J., and Oliver D. Hart. 1986. “The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration.” Journal of Political Economy 94 (4): 691–719. Grubel, Herbert. 1987. “All Traded Services Are Embodied in Materials or People.” World Economy 10 (3): 319–30. Helpman, Elhanan.1984. “A Simple Theory of Trade with Multinational Corporations.” Journal of Political Economy 92 (3): 451–71. Helpman, Elhanan, Mark J. Melitz, and Stephen R. Yeaple. 2004. “Exports versus FDI with Heterogeneous Firms.” American Economic Review 94 (1): 300–16. Hummels, David, Jun Ishii, and Kie-Mu Yi. 2001. “The Nature and Growth of Vertical Specialization in World Trade.” Journal of International Economics 54 (1): 75–96. Jones, Ronald W., I. Coelho, and Stephen T. Easton.1986. “The Theory of International Factor Flows: The Basic Model.” Journal of International Economics 20 (3–4): 313–27. Koopmans, Tjalling. 1957. Three Essays on the State of Economic Science. New York: McGraw-Hill. Landes, David. 1998. The Wealth and Poverty of Nations: Why Some Are So Rich and Some Are So Poor. New York: W. W. Norton. Magee, Stephen P., and Kwang-Yeol Yoo. 2005. “The United States Is a Small Country in World Trade: Further Evidence and Implications for Globalization.” Working Paper. University of Texas. Markusen, James R. 1983. “Factor Movements and Commodity Trade as Complements.” Journal of International Economics 14 (3–4): 341–56. ———. 1984. “Multinationals, Multi-Plant Economies, and the Gains from Trade.” Journal of International Economics 16 (3–4): 205–26. ———. 1989. “Trade in Producer Services and in Other Specialized Intermediate Inputs.” American Economic Review 79 (1): 85–95. ———. 1995. “The Boundaries of Multinational Firms and the Theory of International Trade.” Journal of Economic Perspectives 9 (2): 169–89. ———. 2001. “Contracts, Intellectual Property Rights, and Multinational Investment in Developing Countries.” Journal of International Economics 53 (1): 189–204. ———. 2002. Multinational Firms and the Theory of International Trade. MIT Press. Markusen, James R., and Anthony J. Venables. 2005. “A Multi-country Approach to Factors Proportions Trade and Trade Costs.” Working Paper 11051. Cambridge, Mass.: National Bureau of Economic Research. Melitz, Mark J. 2003. “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity.” Econometrica 71 (6): 1695–1725. Moggeridge, Donald, ed. 1983. The Collected Writings of John Maynard Keynes. Cambridge, Eng.: Macmillan. Mundell, Robert A. 1957. “International Trade and Factor Mobility.” American Economic Review 47 (3): 321–35.
34
Brookings Trade Forum: 2005
Neary, J. Peter. 1995. “Factor Mobility and International Trade.” Canadian Journal of Economics 28: S4–S23. Samuelson, Paul A. 2004. “Where Ricardo and Mill Rebut and Confirm Arguments of Mainstream Economists Supporting Globalization.” Journal of Economic Perspectives 18 (3): 135–46. Yi, Kei-Mu. 2003. “Can Vertical Specialization Explain the Growth of World Trade.” Journal of Political Economy 111 (1): 52–102. Zhang, Kevin Honglin, and James R. Markusen. 1999. “Vertical Multinational and Host Country Characteristics.” Journal of Development Economics 59 (2): 233–52.
DANIEL TREFLER University of Toronto
Service Offshoring: Threats and Opportunities
W
hen asked to provide a framework piece on offshoring, I decided it would be much easier to have the work done by an Indian consulting firm. A quick bit of research turned up a perfect corporate partner. Not surprisingly, the company has a London-based front end—it is a fact of the industry that many customers prefer to work through a Western intermediary. The company quoted the job at $63,000, no taxes. That’s about one-tenth of what an American management consulting firm would charge, but still too rich for my academic salary. So you are stuck with me. The experience taught me two things. First, you can outsource abroad just about anything, from which I conclude that all of our jobs are threatened. Second, the big money in offshore outsourcing goes to the OECD business analysts who help customers communicate their needs to business process outsourcers in low-cost countries. I conclude from this that offshoring brings remarkable opportunities to us all. Therein lies the paradox of offshoring: it is both a threat and an opportunity. In considering international offshoring, two trends scream out for our attention. The first is the rise of China as the world’s manufacturer. Surprisingly, many American firms have yet to wake up to this sea change in their sourcing possibilities. Better information about the strategic offshoring options available to American firms is desperately needed. Aside from this, the rise of China’s
I am indebted to Someshwar Rao and Prakash Sharma of Industry Canada for initiating this study and to both Runjuan Liu of the University of Alberta and Nathan Nunn of the University of British Columbia for help with completing this paper. Lael Brainard and Dani Rodrik provided many insightful comments, as did Susan Collins, who also provided extensive editorial direction. Belinda Lobo cheerfully provided secretarial support. Industry Canada graciously funded most of this study. I am also grateful to the Brookings Institution for funding.
35
36
Brookings Trade Forum: 2005
manufacturing sector poses no new public policy issues. All the familiar arguments hold. On the one hand, international trade is disruptive for workers and firms engaged in import-competing industries. On the other hand, international trade provides the benefits of lower prices to consumers and offers new opportunities for producers (both workers and firms) to expand into foreign markets. In aggregate, the benefits outweigh the costs. What remains for policymakers is the crucial task of ensuring that we generously care for our most disadvantaged, since these unskilled workers are the ones who will bear the brunt of the Chinese offshoring onslaught. The second extraordinary development in international trade has been the rapid growth of traded services involving innovative, technology-intensive processes and employing high-paid white-collar workers. In the past it was unheard of for low-cost countries such as India to be exporting high-value-added services. Now it is common to find Indian software programmers customizing sophisticated software applications for businesses worldwide. This development fundamentally alters the way we must think about innovation-based corporate strategy and public policies that affect the flexibility of the white-collar labor market. The United States faces a choice. It can insulate itself from the global competitive pressures that come with offshoring to low-cost countries. Such policies will protect firms and workers in the short run. However, there is at least some weak evidence that protectionism retards growth.1 In addition, insulating policies will likely encourage foreign countries to deny us market access. Considering that the United States is a major supplier of traded services to the rest of the world, insular policies are about as useful as a blow-dryer in an igloo. Alternatively, the United States can pursue domestic framework policies that promote the competitiveness of U.S. firms and workers. These framework policies would encourage productivity-enhancing investments both by individuals (for example, in human capital) and by firms (for example, in R&D and advanced technologies). The building blocks for globally competitive American firms are domestic policies that encourage continual investments in upgrading and innovation by individuals and firms. When it comes to the U.S. public policy response to offshoring, my best advice is: think globally, invest locally. Finally, let’s not forget about compassion. The American government must be prepared to generously help its most disadvantaged, for they are at greatest risk from the downside of offshoring. 1. See Nunn and Trefler (2005) for support of this view and Rodriguez and Rodrik (2001) for a scathing rebuke of the openness-and-growth literature.
Daniel Trefler
37
What Is Offshoring? There is no universal definition of offshoring, and one task of the Brookings Trade Forum is to decide how broad a set of phenomena to examine. The approach taken by all commentators on offshoring is to attempt a careful definition. This is a natural but misguided approach. We must first start by identifying America’s broad public policy objectives and then identify which aspects of offshoring enhance or impinge on our ability to meet these objectives. In my view there are two complementary objectives: (1) promoting competitiveness and raising incomes; and (2) advancing core values of community and caring through redistributive policies. The most interesting policies are the few that promote both objectives. These objectives will help us delineate the boundaries of a discussion of offshoring by answering three definitional questions. Offshoring, Nearshoring, Inshoring, or All of the Above? As I have already stated, the most interesting aspects of new trends in the tradability of services is the offshoring of technology-intensive, high-end services to low-wage countries. There are two other phenomena of interest: (1) Nearshoring: Much of U.S. offshoring is nearshoring to Canada—for example, to a call center in Toronto that services customers in Chicago. Yet Canada is a country that is very close to the United States (whence nearshoring), and more important, a country that is hardly a low-wage producer. (2) Inshoring: The United States is a major supplier of traded services to the rest of the world. This exporting of services or inshoring cannot be ignored. (Slaughter [2004] calls this “insourcing.”) Offshoring, nearshoring, and inshoring must all be examined. Offshore Outsourcing or Foreign Direct Investment (FDI)? “Offshore outsourcing” describes an arm’s-length transaction between a U.S. firm and a foreign firm. In contrast, FDI describes a domestic firm with a controlling equity investment in a foreign establishment. Recent theories of international trade make it clear that the distinction between offshore outsourcing and FDI is intimately related to the question of whether the United States will retain the highest-paying jobs in the value chain or watch them migrate both to other OECD countries and to emerging low-cost countries such as China and India. One cannot understand this process without looking at what is called the make-
38
Brookings Trade Forum: 2005
or-buy decision—that is, the decision about whether to produce in-house using FDI or to offshore outsource using arm’s-length transactions. In a nutshell, the new theories state that when a project is sufficiently routinized that it can be fully scoped or described, then outsourcing is the appropriate relationship with a foreign service provider. When the project is difficult to describe from its outset, it should be done in-house via FDI. For example, see Antràs (2005). The difficult-to-describe projects are typically the innovative projects that generate the highest value added. Thus we need to understand how firms choose between offshore outsourcing and FDI if we are to understand how to keep high-paying jobs in the United States. My suggestion is thus to study both offshore outsourcing and FDI. Not all economists will agree. For example, Bhagwati, Panagariya, and Srinivasan (2004) argue that we should only be thinking about offshore outsourcing. On this one point, I think that Bhagwati, Panamanian, and Sinicising are wrong.2 It is fitting to develop this discussion of offshoring by providing examples of its pervasiveness and the difficulties of further definitional refinements. Example 1. Traditional “mode 3” FDI in the service sector: Citibank sets up an office in Hong Kong that provides limited services to Chinese customers. The office is staffed primarily by Chinese, and most of the key decisions are made in the United States. Example 2. Traditional “mode 4” FDI in the service sector: A U.S. architectural firm sets up an office in Shanghai to bid and work on local contracts. The firm sends its American architects to Shanghai on a long-term basis to do the design work. What distinguishes this from the previous example is that the control of decisions is largely in the hands of Americans who have temporarily migrated to Shanghai. Example 3. The service-trade revolution using an FDI mode of entry: Verizon sets up an information technology (IT) center in Bangalore that hires Indian programmers to write software for Verizon’s U.S. operations. Example 4. The service-trade revolution with an offshore outsourcing mode of entry: Satyam (India) sets up a contact center that makes Wells Fargo VISA marketing calls to potential customers in Seattle. The use of the term “mode” comes from the IMF (2005) Balance of Payments Manual and is used by all OECD countries in presenting their data.3 2. These misspelled names were introduced by a Chinese student who typed up my corrections to this paper. The typos typify the monitoring and agency problems associated with offshore outsourcing. 3. The manual distinguishes four modes based on the location of the supplier and consumer of the traded service. Mode 1: The supplier is in one country and the consumer is in another. Each
Daniel Trefler
39
Table 1. Definitions of Export-Oriented FDI Projects Related to Offshored Services Contact center services Help desk Technical support/advice After-sales support Employee inquiries Claims inquiries Customer support and advice Market research Answering services Prospecting Information services Customer relationship management
Back-office services Claims processing Accounts processing Transaction processing Query management processing Customer administration processing HR and payroll processing Data processing IT outsourcing Logistics processing Quality assurance Supplier invoices
IT services Software development Application testing Content development Engineering and design Product optimization Other high-end services Regional headquarters Architectural services Biotech and pharmaceuticals R&D Radiology, X-ray Distance education
Source: UNCTAD (2004) and author.
Table 1 provides many more examples of the types of activities that I believe we should focus on. These examples are classified into four areas: contact centers or what are commonly called call centers, back-office services, IT services, and other high-end services. It is worth noting a problem with refining the definition of offshoring. Most of us would be comfortable with the following statement: “Manulife is offshore outsourcing development of its new human resources software to India, while the plastic products industry is importing shopping bags from China.” Why is one “offshore outsourcing” and the other “importing”? In both cases, products currently made in Asia were previously made in-house in America, and in both cases there has been phenomenal growth over the past five years. There are no good answers to this question. Given this problem of definition (and other problems as well), finer definition of offshoring seems impossible. I therefore adopt the approach of U.S. Supreme Court Justice Potter Stewart in his attempt to define pornography: I can’t define it, but I know it when I see it.4 stays in his or her own country. Mode 2: The consumer moves to the supplier’s country to obtain the service. Mode 3: The supplier sets up a foreign affiliate in the consumer’s country. Mode 4: The supplier supplies the service by moving to the consumer’s country. For more, see International Monetary Fund, Balance of Payments Manual (www.imf.org/external/np/sta/ bop/BOPman.pdf [2005]). 4. Jacobellis v. Ohio, 378 U.S. 184 (1964).
40
Brookings Trade Forum: 2005
Occupations or Industries? What makes manufacturing interesting is the dramatic rise in manufacturing exports from low-cost countries, especially China. This export surge has already had a large impact on America’s least-skilled workers in industries such as garments. See Feenstra and Hanson (1996, 1999) and Trefler (1998). It is now poised to threaten America’s moderately skilled workers in such industries as auto parts. However, to my mind these developments pose no new public policy issues that have not already been discussed in the context of conventional import competition. The reason it is not new is that it affects occupations that have always been impacted by international trade. On the other hand, the revolution in the world’s ability to trade in services is something new. At least some of the new service trade involves highly skilled white-collar workers operating in low-cost countries such as India. Successful policy responses aimed at assisting skilled labor will likely be very different from policy responses aimed at assisting less-skilled labor. This distinction has had no play in the offshoring debate but is likely crucial for reasons to be explained below. Thus, service offshoring poses new policy challenges not raised by manufacturing offshoring because it involves white-collar workers. My view will find critics. Most researchers argue that the rise of China as the world’s manufacturer poses such important challenges that it must be included in every discussion of international trade policy. I look forward to a healthy debate of this point. Another problem with focusing on industries rather than occupations stems from recent changes in traditional manufacturing. With the offshore outsourcing of back-office jobs by manufacturing firms, we tend to think that the line between manufacturing and services is becoming cleaner. However, the opposite is also happening. When Microsoft introduced its Xbox game player, it hired Singaporebased Flextronics (the contract manufacturing giant) to build a factory in lowwage Guadalajara, Mexico, that was supplied with standardized parts from China. Design of the core proprietary technology was outsourced to Nvidia Corp. of the Bay Area and manufactured in Taiwan. Clearly, Xbox could not have been brought to market in this way without tremendous logistics support. As such, Xbox is a manufactured product that embodies a significant service component. This example is commonplace. Accenture (2004) reports that 43 percent of its customers outsource their supply chain management. This reflects the rise of contract manufacturers that both manufacture and provide manufacturing service support. Thus, in many respects traditional industry distinctions are blurring. Focusing on occupations is much cleaner and more useful for policy.
Daniel Trefler
41
White-Collar Workers and the New Trade Issues Raised What is most novel about the recent emergence of offshoring is that it affects white-collar workers employed in technology-intensive industries (be these services or manufacturing). We simply do not know what the net effects of this are because empirical trade economists have virtually no experience with this phenomenon. Three issues need to be researched: —Many (though not all) white-collar jobs are high-paying jobs, paying $70,000 or more a year. As a country, we are familiar with losing high-paying jobs (head-office service jobs and banking service jobs, for example) to other rich countries. What we are less familiar with is losing high-paying jobs to India. We certainly want to avoid losing these good jobs. However, these losses are somewhat offset by the jobs created for U.S. business analysts with IT expertise. These Americans work as highly paid intermediaries who interface between U.S. companies and Indian business service providers. —When a white-collar job is offshored, the value of an American worker’s industry-specific and firm-specific knowledge is destroyed. This stands in contrast to what happens when an unskilled worker is displaced. There is little valuable knowledge to be destroyed. It is unclear whether loss of such knowledge is an equity concern alone (because it hurts offshore-displaced workers) or whether it is also an efficiency concern (because it destroys valuable human capital). This needs to be investigated. —There is now a large literature showing that retraining programs are not effective for most displaced workers (see, for example, Baicker and Rehavi 2004). The argument is that unskilled workers are unskilled for a reason: they are missing the most fundamental of abilities, namely the ability to learn (see Heckman and Carneiro 2003 and Trefler 2004). This means that displaced unskilled workers need income transfers to handle trade shocks. In contrast, IT professionals are likely to be highly motivated individuals who would do well in retraining programs.
The 64,000 Job Question: Whither China and India? Behind the alarm about service offshoring is a sense that OECD countries are in danger of being overtaken by China, India, and a number of other developingcountry destinations for service offshoring. In the most alarming scenario, these countries have an infinite capacity to absorb OECD technologies and management strategies, to improve on them, and ultimately to compete head-to-head
42
Brookings Trade Forum: 2005
with the OECD. Finally, in this scenario, China and India with their newly acquired high-tech status will continue to pay low wages for skilled labor and will use this advantage to create an economic steamroller that crushes all OECD countries. There are two reasons why this argument is flawed. First, there is an ironclad economic law that prevents one country from ever dominating world trade. Second, there are political-economic reasons to doubt the speed at which this scenario can unfold. I review these reasons in detail. The Ironclad Law of Comparative Advantage I am a better researcher than my secretary. Surprisingly, I am also a better typist than he is. That is, I have an absolute advantage over my secretary in both research and typing. Nevertheless, I find my secretary to be indispensable. That is because I am relatively better at research than typing. Thus if I typed an hour less a day I could write one page of this paper, whereas if my secretary typed an hour less a day he could only write one sentence of this report. In economic jargon, I have a comparative advantage in research, and my secretary has a comparative advantage in typing. In the most alarmist scenarios about China and India, these countries will soon have an absolute advantage in producing all goods and services. However, the United States will continue to have a comparative advantage in the most knowledge-intensive goods and services. Thus, even in the most alarmist scenario, the United States will continue to export knowledge-intensive goods and services to China and India. With their low wages, what prevents these countries from exporting everything and importing nothing? If they import nothing they will be giving their goods away for free. I doubt they would agree to this. In addition, the Americans will need yuan to buy Chinese goods. As we demand more of their currency it will rise in value. Eventually, the yuan will rise so much in value that Chinese wages are no longer so dominantly competitive. (This is exactly the problem the United States faces when its currency is strengthening.) In real life there are things China can do to slow this process down, but China cannot forever keep the yuan undervalued. This is an ironclad law. Countries such as Germany in the 1960s and Japan in the 1970s ran afoul of the comparative advantage police. They ran huge trade surpluses that threatened to destroy U.S. manufacturing. Over time, however, their currencies strengthened to the point where these countries ceased being low-cost producers. In this context it is important to remember that in 1959 Japan had a highly skilled and disciplined
Daniel Trefler
43
labor force that was paid 10 percent of U.S. wages. Japan in 1959 was, from the limited perspective of offshoring, not that different from China today. Yet Japan never was able to dominate world manufacturing. Why? Because Japan succumbed to the comparative advantage police by steadily revaluing the yen. The same will eventually happen to China. It does not matter that they have hundreds of millions of citizens ready to work for next to nothing. If we buy too much from them, their currency will rise to the point where their low yuandenominated wages are wiped out by the currency conversion. It does not matter that Chinese workers are paid 4 yuan an hour unchanged over the next hundred years. If the yuan strengthens, Chinese dollar-denominated wages will rise. Like the Mounties, the comparative advantage police always get their man. Institutions and the Mystery of Modern Economic Growth The comparative advantage argument has one significant limitation. It is possible that China and India will develop a comparative advantage in knowledgeintensive goods and services, leaving the United States to produce T-shirts for the Shanghai market. In this scenario, the United States continues to export to China according to the law of comparative advantage. However, the United States becomes poor relative to China and possibly poor even in absolute terms. The argument for absolute impoverishment was first made by Graham (1923) and has been repeated by Hicks (1953), Johnson and Stafford (1993), Gomory and Baumol (2000), and most recently by Samuelson (2004). While the argument is logically correct, fortunately for the United States it is irrelevant. The problem with the argument is that it presumes that China and India will become the world’s technological leaders. Such a presumption is in flagrant contradiction to what we know about the role of domestic institutions for promoting innovation. Current thinking about the determinants of long-term economic growth focuses on the central role of domestic institutions. See Helpman (2004) for a review of the literature. In this view, there are limits to what China and India can produce under their current political-legal-economic regimes. As China and India expand the range of services they provide, they will eventually enter into services that depend on constant innovation. In the new institutions-and-growth view, innovation cannot occur without institutions that protect property rights, that provide a fully functioning legal framework for arm’s-length transactions, that support thick equity and debt markets, and that balance the needs of inside innovators against those of outside investors. Srinivasan (this volume) argues that these institutional constraints on growth were made irrelevant in India’s IT sector because of special regulatory provisions
44
Brookings Trade Forum: 2005
and protections afforded the sector. While I would certainly never want to disagree with Professor Srinivasan—wait, I think I already have once in this paper— my point is less about the development of a single sector and more about longterm, innovation-based, multisectoral, modern economic growth. In short, China and India will not be able to compete in innovation-intensive sectors without the “institutions of modern capitalism” (Rosenberg and Birdzell 1986) and its handmaiden, “the invention of invention”(Mokyr 1990). For China and India to compete over the very long haul, their institutions will have to look a lot more like OECD institutions. This is unlikely to occur even over a quartercentury horizon. Evidence on the Importance of Institutions for Long-Run Growth Figure 1 provides two examples of a now-burgeoning institutions-and-growth literature. The top panel plots GDP per capita in 1997 against the Kaufmann, Kraay, and Zoido-Lobaton (1999) rule-of-law index. This index ranks countries based on the degree of rent-seeking or opportunistic behavior that investors are exposed to. For example, when I make an equity investment in a U.S. company I have some confidence that I will see my money again—not always, but usually. In contrast, my equity investment in China is much more likely to be siphoned out of the company and forever lost to me. The figure 1 R 2 of 71 percent shows just how much rent-seeking behavior can retard growth. The bottom panel plots GDP per capita against the Gwartney and Lawson (2003) legal-quality index. This index captures the ability of firms to write enforceable contracts. The need for rule of law governing commercial transactions is obvious. Later in this paper I discuss how important it is for understanding offshoring. The bottom panel of figure 1 shows how important the quality of legal institutions is for growth. Of course, India and especially China have grown rapidly with weak institutions. But as Alwyn Young (1992, 1994) has pointed out, much of this growth is based on unsustainable factor supply growth rather than on productivity growth. It is the latter that is the basis for modern economic growth. For example, reforms in China and India have led to a movement of workers from farms to cities, thus providing manufacturing with an almost infinite supply of cheap labor. In contrast, long-run sustainable growth of the kind experienced in OECD countries is driven by innovation-induced productivity growth. And rock-solid institutions are the supporting architecture for innovation.
Daniel Trefler
45
Figure 1. Good Institutions Promote Growth Log real per capita GDP
10 9 Canada and USA
China
8
India R 2 = 0.71
7 6 5 –1.75
–0.75
0.25 Rule of law
1.25
2.25
Log real per capita GDP
10
9
Canada and USA
8
China India
R 2 = 0.60
7
6
3
5 Legal quality
7
Source: Data from Kaufmann, Kraay, and Zoido-Lobaton (1999) and Gwartny and Lawson (2003)
9
46
Brookings Trade Forum: 2005
Institutions in China and India Rich countries have good institutions. The quality of Chinese and Indian growth-enhancing institutions is at best moderate. Historically, very few countries experience rapid improvements in their domestic institutions. Rather, institutions develop at a glacial pace, over a century or more. The idea that China or India can rapidly develop these institutions is a complete misread of the sources of modern economic growth. How does this pan out in the specific contexts of China and India? In personal conversation, Wendy Dobson of the Rotman School of Management has identified five weaknesses in Chinese and Indian institutions: (1) the role of the government—particularly state-owned enterprises and corrupt officials—in preventing the efficient reallocation of resources such as capital; (2) a weak financial system that leaves firms under-resourced; (3) a social safety net that leads to labor market inflexibilities; (4) a lack of an endogenous capability to innovate, in part because entrepreneurs are hemmed in by the rent-seeking behavior of bureaucrats; (5) a one-party state in China and a corrupt alliance between bureaucrats and politicians in India that retards the development of a local entrepreneurial class. Although some of these institutional impediments are slowly evaporating, it will take decades before they all disappear. An Application to the Worldwide Software Industry To make the argument about institutions less abstract, consider how it plays out in the emerging centers of the worldwide software industry, that is, in China, India, Brazil, Ireland, and Israel (see table 2). The industry is very large in India, China, and Brazil. The combined employment of these three countries is 600,000, approaching the U.S. level of 1,024,000. On the other hand, sales per employee are very small in these countries. A U.S. software employee generates almost $200,000 of sales per employee, four times more than an Indian employee. This means that China, India, and Brazil are providing low-valueadded programming skills. Will the software industry in these low-cost countries grow and enter into higher-valued-added segments? Four significant institutional factors may prevent this. —Long-term software growth must be primarily driven by domestic developments. Apart from India, the software industries in these countries developed in response to local needs (Arora and Gambardella 2005; Arora, this volume). Banking and telecommunications drive software growth in Brazil and China,
Daniel Trefler
47
Table 2. The Software Industry Worldwide, 2002 or latest available year
Countries Brazil China India Ireland (multinational enterprises) Ireland (domestic) Israel United States Japan Germany
Sales ($U.S. billion)
Employment (1,000s)
Sales/ employment (1,000s)
Software sales/GDP (percent)
7.7 13.3 12.5
160 190 250
46 38 50
1.5 1.1 2.5
12.3 1.6 4.1 200.0 85.0 39.8
15 13 15 1,024 534 300
804 127 273 195 159 133
10.1 1.3 3.7 2.0 2.0 2.2
Source: Arora and Gambardella (2005).
software growth in Israel was driven by Israel’s high-tech sector, and software firms in Ireland developed by providing services to multinationals using Ireland to enter the European market. In each case, domestic factors drove the initial growth: exports came later. The message, then, is that the institutions that promote domestic-led growth must be in place. —Clusters. In order to have domestic-led growth, many pieces must fall into place simultaneously. For example, the weak financial systems in China, India, and Brazil leave firms under-resourced because insiders routinely steal from outside investors. Thus, firms in these countries are short not only on capital, but also on sophisticated financial advice provided by banks and venture capital firms. Further, downstream demanders of software such as banks are also underdeveloped because of poor national institutions. Therefore, software firms are missing sophisticated buyers who will push them to innovate and upgrade their products (Porter 1998). It is sometimes argued that R&D follows production. Thus, as the low end of the software industry migrates to India, product development will also migrate. Indeed, India’s National Association of Software and Service Companies (NASSCOM) boasts many new products. However, available evidence suggests strong limits to this process. Audretsch and Feldman (1996) show that as an industry matures and manufacturing moves to low-cost locations outside the cluster, R&D continues to occur inside the cluster. Jaffe, Trajtenberg, and Henderson (1993) explain why. Much of what is important for ongoing innovation involves the local exchange of tacit information, that is, information that cannot be codified and that can only be communicated face-to-
48
Brookings Trade Forum: 2005
face. All of this implies institutional limits to the development of an increasingly sophisticated software industry in China, India, and Brazil. —National innovation systems. A skilled labor force is critical for the growth of a domestic software industry. China, Brazil, and India all have large and growing university systems. Each country turns out about as many natural science and engineering degrees as the United States (see, for example, Bardhan and Kroll 2003; Arora and Gambardella 2005; and Arora 2005). It is often argued that this provides these countries with cheap skilled labor. I am more skeptical. If skilled labor is so abundant, why are IT sector wages rising by 15 percent a year in India? The answer is that there is often a significant gap between what the university provides and what the private sector needs. The most successful country in the world at bridging this gap has been the United States. As is well known, the U.S. university system co-evolved with private sector needs. The development of the state university system is a typical example. As a result, the U.S. university system has an unparalleled curriculum vitality. Further, Rothschild (2003) argues that the continued success of the U.S. university system has been driven by competition. On the one hand, U.S. universities compete fiercely among themselves for the best faculty and ideas. On the other hand, the system has diverse revenue sources, and the many funders of U.S. university research compete among themselves to fund the best projects. As a result, there are no misdirected top-down injunctions about how to run engineering schools, and good ideas are rarely suppressed. Universities in China, Brazil, and India are able to crank out large numbers of graduates, but they will be unable to train the world’s best graduates for many decades to come. —International technology transfer. There can be little doubt that OECD multinationals are teaching China and India how to compete. There is also an argument that we are selling ourselves short by underpricing these technology transfers. However, for better or worse, in an open society it is virtually impossible to act differently than we are currently doing. How far will the process of international technology transfer go? Countries with strong protection of intellectual property rights are the favored destination of multinational enterprises (MNEs): these companies go where institutions are strong. Thus weak institutions in China, India, and Brazil will place a limit on technology transfer in the software industry. This section has demonstrated in the context of the software industry that weak Chinese, Indian, and Brazilian domestic institutions will prevent these countries from migrating too far up the software value chain. The United States need not worry that in the next twenty years we will be reduced to mending the socks of Chinese businessmen.
Daniel Trefler
49
The Determinants of Offshore Outsourcing: The Contracting Environment The rise of service offshoring has two main drivers: —Technological improvements in the information and communications technology (ICT) sector. These improvements launched what UNCTAD (2004) calls the “service tradability revolution.” While the financial sector has been using ICTs for fifteen years, developments of the past five years have dramatically reduced costs to the point where ICTs are cheaply available to all. —The new “openness” consensus among political coalitions in developing countries. In the spring of 1992, Deng Xiaoping used a tour of southern China to call for a radical opening up of the Chinese economy to both domestic and foreign competition. Since then, southern China has been growing at 25 percent a year. Likewise, the 1991 financial crisis in India led to the dismantling of tariffs and restrictions on FDI. Across the developing world there has been a wave of reforms aimed at integrating these low-cost countries into the world economy. The rise of manufacturing offshoring has also been greatly facilitated by reductions in transportation costs and improvements in transportation logistics. Conventional wisdom has it that firms go offshore to reduce costs, usually to low-cost countries. This is a misleading view. For one thing, 85 percent of U.S. service offshoring is with other OECD countries. For another, many firms want access to foreign markets in order to tap into new sources of skilled workers, to position themselves in rapidly growing markets, and to be closer to foreign customers. Accenture (2004) reports that lower costs is only third on the list of the most important factors in choosing an offshore outsourcing provider (see figure 2). The first two are service providers’ expertise or capability and service providers’ flexibility. What is most interesting about the list in figure 2 is that most of the items cannot be easily codified or written down in a contract. Mirroring this fact, less than a third of the firms in the Accenture study feel that their offshore outsourcing contract is the key framework for managing the offshore outsourcing relationship. The former CEO of one huge corporation related to me the story of the lengthy contract negotiations he had for a greenfield investment in China. Years of negotiating with the Chinese culminated in a party to celebrate the conclusion of the contract talks. At the party, the Chinese host turned to the CEO and candidly told him that the contract meant nothing to the Chinese partners and that it was only signed to make the CEO comfortable! For the Chinese partners, the important thing was that they trusted the CEO.
50
Brookings Trade Forum: 2005
Figure 2. Importance to Firms of Factors in Choosing an Outsourcing Provider Percent Provider expertise/capability
86
Flexibility
81
Low cost/price
78
Industry knowledge
75
Ability to earn trust
74
Reputation
69
Culture fit
55
Creativity
51
Outsourcing team members
50
Provider's global reach
39
Prior relationship
36
Knowledge of company and key executives Offshore capabilities
34 30
Source: Data are from Accenture (2004).
The New Theories of Offshoring: Trade and Contracting The difficulty of writing and enforcing contracts has led to a new generation of theories about offshoring that focuses on contractual incompleteness. The core idea is that parties to a contract cannot specify all possible future contingencies, particularly when an American firm is operating in a foreign environment with which it is not entirely familiar. For concreteness, suppose that an Indian service provider is required to make an up-front investment in customizing software for a U.S. buyer’s human resource (HR) needs. Also suppose that there is only a single outcome of interest, namely the “quality” of the software. I make the extreme assumption that a court cannot judge quality or observe anything that might be informative of quality. The contract is incomplete in the sense that the court cannot properly enforce it. In addition, the contract may not be enforceable because the Indian court is corrupt or lacks the expert judges needed to properly adjudicate the dispute. As a result, after the customization investment is made, there is a bilateral hold-up problem. The buyer would like to offer a lower price for the software than initially agreed to, perhaps arguing that the customization is incomplete.
Daniel Trefler
51
Figure 3. Ex Post Renegotiation Leads to Inefficient Underinvestment in Customization Buyers (continuum) Service provider (finite number)
Customization (relationship-specific investment)
Of course, the Indian service provider is no fool. He fully anticipates that the buyer will renegotiate and so takes steps to protect himself. In particular, the Indian service provider will underinvest in customization. Figure 3 illustrates this point. There is a continuum of buyers spread out on a circle. Each point on the circle represents one buyer’s ideal HR software needs. The number of Indian service providers is finite, three in figure 3. A buyer wants to find a service provider who is a perfect match, but usually will not find one. Instead, the buyer will have to ask the service provider to make a relationship-specific investment in customization. There are several steps in the timeline of this analysis: —The U.S. buyer enters India in search of a local service provider. —The buyer and service provider match. —The buyer chooses an organizational form. That is, the buyer decides whether to offshore outsource or to vertically integrate using FDI to buy the service provider’s firm. —The service provider chooses a level of relationship-specific investment in customization. —The buyer renegotiates. The question is, should the buyer use offshore outsourcing or FDI as the mode of securing customized HR software? The answer depends on the outside options of the service provider. If the service provider can turn around and find
52
Brookings Trade Forum: 2005
another buyer whose HR software needs are similar to the original buyer’s needs, then the service provider can walk away from the old relationship and start up a new one at little cost. In this case of good outside options, the service provider is not overly concerned with hold-up problems and so makes most of the necessary customization investments. This means that the buyer does not have to provide incentives to the service provider to make up-front investments. Logic dictates that in this scenario the buyer should offshore outsource. In contrast, if the service provider’s outside options are poor, he will be concerned about hold-up, will not make the customization investments, and will provide low-quality service. The buyer will then have to use FDI if the buyer wants to control up-front investments in customization. Thus, the decision to offshore outsource or use FDI depends on the degree of hold-up, which in turn depends on (a) the outside options available to the service provider and (b) the quality of contract-enforcement institutions such as the legal system and government rentseeking behavior. A key issue is the question of precisely how FDI provides the right incentives for the service provider to invest in customization. The earliest forms of these models were based on what is called the transactions cost theory of the firm (see Coase 1937; Williamson 1975, 1985; Klein, Crawford, and Alchian 1978; and Grossman and Helpman 2002, 2003). A problem with this approach is that it assumes that vertical integration (FDI) magically eliminates hold-up problems within the firm. But how does this happen? After all, service providers within the firm still have incentives to underinvest by shirking. To address this concern, Grossman and Hart (1986) and others developed the property rights theory of the firm. In this theory, the focus is on how the service provider’s incentives are altered by allowing or not allowing the service provider control over the buyer’s core asset. In particular, control of the relationship-specific asset is given to the party whose effort most influences profits. If the buyer’s input into developing customized HR software is crucial, then it should be done using FDI. If the buyer can scope the project with precise specifications, then what is needed most is to provide high-powered incentives to the service provider. This is done by making the service provider the residual claimant on profits—that is, by offshore outsourcing. This insight has been built into models of offshoring by Antràs (2003, 2005), Grossman and Helpman (2005), and Antràs and Helpman (2004). Three related papers that are less about the inability to write complete contracts than about the unwillingness of courts in developing countries to enforce them appear in Ethier and Markusen (1996), Markusen (2001), and Nunn (2005).
Daniel Trefler
53
Empirical Evidence Supporting the New Trade Theories It is useful to review the two papers that combine theoretical insights with empirical support. These are Antràs (2003) and Nunn (2005). In Antràs (2003), both the buyer and the service provider make relationship-specific investments. The buyer invests capital and the provider invests labor. With offshore outsourcing, each party’s outside option in the renegotiation stage is 0, so there is underinvestment by both parties. With FDI, the buyer is allowed to take a fraction of the provider’s output. Thus the buyer’s outside option is and the provider’s outside option is 0. Thus, relative to offshore outsourcing, FDI induces more investment by the buyer and less investment by the provider. Restated, activities done via FDI will be relatively more capital-intensive than offshore outsourced activities. This yields an important empirical prediction. The larger capital’s share of an industry, the more sensitive profits are to the buyer’s capital underinvestment. Hence the property rights approach predicts that FDI will be the dominant organizational form. This is exactly what we see in figure 4 (see the notes to figure 4 for a complete explanation). Nunn (2005) changes the focus slightly. Instead of being interested in the inability to write complete contracts, he is interested in the extent to which a country’s legal system appropriately enforces contracts. In particular, in countries with poor contract enforcement institutions, buyers and service providers will be unwilling to make relationship-specific investments for fear that they will expose themselves in court to hold-up problems. Thus, goods requiring substantial relationship-specific investments will tend to be produced in countries with good contract-enforcement institutions. Figure 5 provides Nunn’s evidence on this mechanism (see the notes to figure 5 for a complete explanation). We tend to think that offshoring is driven almost exclusively by the search for low-cost labor. This is simply not true. Eighty-five percent of U.S. service offshoring is with other OECD countries. Many firms enter into service offshoring relationships in order to gain access to a skilled workforce, to be in a rapidly growing market, or simply to be closer to customers. For many firms, the problem of offshoring to low-cost countries is the contracting environment. These countries do not have the legal institutions that allow firms to write complete and enforceable contracts. As a result, opportunistic behavior by local entrepreneurs, bureaucrats, and politicians leads to hold-up problems, underinvestment in the relationship, and ultimately an unsatisfactory offshoring experience. China in the 1990s was massively subject to these hold-up problems. Clissold (2004) provides a vivid description of just how terrifying the weaknesses of China’s legal system were to foreign investors. Things are improving, but only slowly.
54
Brookings Trade Forum: 2005
Figure 4. The Share of U.S. Imports Controlled by MNEs Rises with the Capital Intensity a Share of U.S. imports controlled by MNEs 1
y = –6.86 + 1.17x (1.02) (0.24) R2 = 0.54 0 0 Capital-labor ratio by industry Source: Data are from Antràs (2003). a. Each point in the plot is an industry. The data plot an industry’s capital-labor ratio against the share of U.S. imports for that industry that are imported by MNEs from their affiliates. The more capital-intensive the industry, the larger the share of U.S. imports involving MNEs. Note that this is manufacturing trade rather than services trade and that offshore outsourcing is any arm’s-length transaction.
Offshoring and Dynamic Comparative Advantage The best way to understand the economic aspects of the spectacular ascendancy of East and South Asia is that it has not been driven by high-tech innovation. Rather, it has been driven by kaizen, which means “improvement” or “idiotproofness” in Japanese and is translated into English as “total quality control.” The reason that Asian economies have stormed on to the scene one by one is that quality or reliability competition is discontinuous. Once a firm meets or surpasses the quality of its lead competitors, it grabs huge market share. What has been happening in East and South Asia has been a steady process of incremental innovation. This is Rosenberg’s (1982) unsung hero of modern economic growth. As Rosenberg has argued persuasively, incremental innovations lead year in and year out to the modest but steady productivity gains underlying modern economic growth. Thus, to understand offshoring one must understand
Daniel Trefler
55
Figure 5. Contract Enforcement and Comparative Advantqage a Quality of institutions .8
(beta coef = .38, t statistic = 4.81)
.4
0 0
.5 Relationship-specific investments in exports
1.1
Exports by good-institution countries/exports by bad-institution countries 2.7 (beta coef = .40, t statistic = 6.37)
0
–2.7 –.1
.5 Importance of relationship-specific investments
1.1
Source: Data are fom Nunn (2005). a. In the top panel each point is an industry. Countries with strong institutions (as measured by the rule of law) tend to export goods that require large relationship-specific investments. In the bottom panel each point is a pair of countries. Relative to countries with a weak rule of law, countries with strong rule of law have exports that are skewed toward goods requiring large relationship-specific investments.
56
Brookings Trade Forum: 2005
Figure 6. Incentives: What Are Firms Trying to Achieve through Outsourcing? Incentive for continual performance improvement
53
Service provider to make greater investments up front
44
Create environment of pooled commitment
Groundbreaking endeavor
41
35
Source: Data are from Accenture (2004).
incremental growth, not pathbreaking innovations dominated by the Western countries that invented invention. For firms thinking about offshore outsourcing, the single most important incentive issue is how to encourage service providers to continually improve their performance (see figure 6). Improving performance is much more important to firms than “groundbreaking endeavor.” It is even more important than the up-front investments that we focused on in the previous section. The problem that most firms face is that what is being offshore outsourced is a small component of a larger system. This creates a tension. On the one hand, a buyer would like a service provider to contribute ever-improving component services. On the other hand, ironing out incompatibilities between interdependent components can be a drain on the buyer’s energies. The buyer can conserve its energies by tightly controlling the improvement process, but this may inadvertently stifle the service provider’s incentive to innovate. Puga and Trefler (2002) explore this tension using the novel concept of the imperfect substitutability of innovative effort. Imperfect substitutability is a measure of the costs imposed on one party (the buyer or service provider) by the innovative efforts of the other party (the service provider or buyer). To illustrate, consider the key component of a television, namely, the cathode ray tube (CRT). A CRT is basically an electron gun aimed at the phosphorcoated front screen of a vacuum tube. Rising consumer preference for flatter screens has created a tension between electron gun manufacturers such as
Daniel Trefler
57
Figure 7. The Cathode Ray Tube (CRT): A Complex System
Source: Puga and Trefler (2002).
Sony and vacuum tube manufacturers such as Asahi Glass. From the perspective of Asahi Glass, domes are better than flat surfaces at withstanding the implosion forces of the vacuum tube. Asahi would thus prefer the solution illustrated in figure 7. The CRT screen is flat from the viewer’s perspective but domed from the electron gun’s perspective. Sony would prefer a flat screen from the perspective of both the viewer and the gun because the variable thickness of the glass creates a prism effect that reduces the sharpness of the picture. This distortion can only be remedied by modifying the electron gun. Asahi’s solution imposes costs on Sony, while Sony’s solution imposes costs on Asahi. In our terminology, the innovative efforts of Asahi and Sony are imperfectly substitutable. Sony must decide in advance the conditions under which it will accept Asahi’s solution. The broader these conditions are, the more likely Asahi’s solution is to be adopted and the more resources Asahi will funnel into the project. That is, delegation of control over knowledge is an incentive device. What makes this view of how innovation is organized so useful is its implication for long-term growth. As multinationals from the developed world use more Chinese service providers, good matches, in the sense that relatively little customization is needed, are more likely. In these cases, the MNEs will delegate
58
Brookings Trade Forum: 2005
control over knowledge to their Chinese service providers. This will give the Chinese incentives to do more incremental innovation, which in turn will make them more knowledgeable. In the next period, therefore, these Chinese service providers will have a greater ability to serve MNEs. Compare two markets, one in which relatively few MNEs enter (such as Indonesia or market 1) and one in which many MNEs enter (such as China or market 2). On average, the service providers in China will require less customization in order to meet the needs of MNEs. This will encourage MNEs in China to delegate control over knowledge creation. This will create more knowledgeable service providers in the next period, which in turn will make them even more attractive to MNEs in the period after that. This will lead to even more MNEs arriving in the next period and thus to even less need for customization. In short, the market becomes more and more attractive as a place for offshore outsourcing. Such a process is exactly what took Taiwan from being a country of original equipment manufacturers (OEMs) to being a country of original design manufacturers (ODMs), and it is moving China from being an auto-parts supplier to producing passenger cars for Southeast Asia and engine blocks for the United States. This analysis explains what is currently happening in China and India and offers further insights into how these developments are embedded in an institutional and organizational context.
Policy Challenges By any international yardstick the United States is a rich and successful economy. However, it could do better, and if it does not actively work on doing better it will fall behind. The problem is that offshoring has raised the stakes in the global competition game. The primary effect of offshoring is that it makes it all the more important for the United States to adopt productivity-enhancing domestic policies. What follows is a list of the key policy issues. I start with what the United States should not do. It is perhaps worth focusing on three policies, two of which receive inappropriate attention and one of which may be receiving too little attention. Two Dumb Ideas It is tempting to approach the problem of how to benefit from offshoring as a problem of designing an industrial policy that successfully picks winners.
Daniel Trefler
59
This is a dumb idea. We should not be in the business of subsidizing contact centers, management consultants, financial institutions, or insurance companies. Sure, China does it and Japan did it. But we forget the dumb mistakes that Japan made (see, for example, Saxonhouse 1998). And do we want a Chinesestyle command economy that is great at catching up but unproven at leapfrogging and horrible at allowing individuals the personal freedoms to make economic choices? Another dumb idea is to adopt a protectionist stance. This will help in the short run, but it will provide the wrong long-run incentives for investing in productivity. Without the spur of international competition, U.S. productivity in protected industries will languish, leading to even deeper structural problems. The Destruction of Human Capital The new competition from offshoring will lead to lost jobs and bankruptcies. Each time a worker is separated from her firm, firm-specific human capital is lost. This reduces the incentives of both managers and workers alike to invest in developing firm-specific knowledge. For example, a highly paid IT consultant will typically know much more than just IT. She will know about the unique needs of her firm. Offshoring leads to more frequent separations between workers and firms, thus destroying important dimensions of American human capital. There is solid evidence to support concerns about the destruction of human capital. Wasmer (2002) demonstrates that the major differences between European and U.S. labor markets stem from differences in the specificity of human capital investments. Martin and Moldoveanu (2003) offer substantial evidence of the rising importance of human capital for firm value. For example, in 2000 Cisco Systems employees earned between $5 and $8 billion in option profits alone at a time when the company only made $4.6 billion. It is unclear whether the loss of knowledge that arises from worker-firm separations is an equity concern alone (because it hurts offshore-displaced workers) or whether it is also an efficiency concern (because it destroys valuable human capital). It becomes an efficiency issue if there are incomplete contracts governing worker-firm relationships. Specifically, after relationship-specific training has occurred, workers cannot credibly commit to staying with the firm. This contractual incompleteness leads to an underinvestment in training relative to some unattainable first-best contract. The main point is that offshoring may be exacerbating this inefficiency by leading to more frequent separations. Ensuring that firm-relevant human capital continues to be created in the United States is always a key issue. Whether offshoring creates an environment
60
Brookings Trade Forum: 2005
in which government intervention (new policies to promote human capital formation) is appropriate is an open question that demands to be researched. The policy issues that flow from this are simply not well understood. There is a tension between promoting long-term relationships and promoting flexibility. Flexibility describes how easy it is both for workers and for firms to terminate a relationship and find an alternative one. How do we design incentives for greater on-the-job training and formal job training programs in an environment where offshoring is likely to reduce the length and value of worker-firm relationships? How do we help workers carry accumulated skills across firms? Should corporate and personal taxes reflect our need to promote both greater specific investments as well as greater flexibility? Clearly, more research is needed in this area.
Policy Conclusions Most of the sensible policies aimed at fostering American competitiveness in the service offshoring market are investment-promoting framework policies. They encourage U.S. workers, firms, and governments to invest in building productive assets such as human capital and new technologies. Such framework policies address a whole host of domestic competitiveness issues and so are not unique to issues raised by service offshoring. Nevertheless, it would be a mistake to think that this makes framework policies less central to issues raised by service offshoring. Offshoring creates only a few new policy issues. First, it forces U.S. firms to be part of a global market and hence to compete globally. It thus makes framework policies that encourage investment and competitiveness all the more important. Second, it creates more churning among firms and workers, thus destroying human capital that is specific to worker-firm matches. We must think of policies that encourage these investments without at the same time creating the kinds of labor market inflexibilities that are the source of Euro-sclerosis. Third, it is central both politically and morally to find ways of helping workers displaced by service offshoring. In Trefler (2004), I offer one approach—investing in children at risk so they grow up with skills that allow them to escape the pressures of foreign competition. This and other redistributive policies are clearly affordable for the richest country on the planet.
Comments and Discussion
Dani Rodrik: Dan Trefler’s paper is full of valuable nuggets, but the core of the paper rests on two assertions: (1) by the very logic of comparative advantage, it is impossible for China, India, and other newcomers to take over everything that the United States and other advanced countries are currently producing; and (2) the weakness of institutions in these newcomers will necessarily retard the rate at which they converge with technology levels in the West. The first of these points is unassailable in its logic and is hardly controversial (at least in a roomful of economists). The second seems also on target and serves as a useful reminder that China and India remain by and large very poor countries with lots of work still ahead of them, despite their prowess in certain tradable activities. But putting the problem this way somehow minimizes the ability of these countries (and their emulators) to compete head-on with the United States in global markets. I would like to suggest a somewhat different angle on this question. I have been doing some work recently (with my colleague Ricardo Hausmann) that attempts to measure the “quality” of the export baskets of different countries. We basically quantify the income level that is associated with each country’s exports, which we call EXPY. Without going into too much detail, we do this first by identifying the average income level of countries that are the main exporters of any six-digit level product. This gives us the income level associated with each traded product. Then we calculate EXPY as the weighted average of these values for each individual country. Figure 8 shows how EXPY stacks up against per capita GDP. As expected (and almost by construction), rich countries tend to export goods that other rich countries export. But upon closer look, two things are important in this figure. First, the range of EXPY is a lot narrower than the range of per capita GDP. Country export baskets exhibit considerably greater similarity 61
62
Brookings Trade Forum: 2005
Figure 8. Relationship between Per Capita GDP and Income Level Associated with Exports Log EXPY
10 United States
China
9.5 India 9
8.5
8
6.5
7
7.5
8
8.5
9
9.5
10
10.5
Log GPD per capita Source: Author’s calculations.
than the underlying aggregate productivity of individual countries. Second, some of the key countries associated with outsourcing/offshoring have EXPY levels that are very high; in fact, several times higher than their GDP per capita. Consider, for example, the following three countries: the United States, China, and India. Their respective values (for 2003) are shown below. Overall productivity in the United States is about 7.5 times higher than that in China, and about 13 times higher than that in India. Trefler is right that these
United States China India a. Income associated with each country’s exports.
GDP per capita (US$)
EXPY (US$) a
35,484 4,726 2,732
15,977 13,575 10,701
Daniel Trefler
63
huge gaps will not close until institutional quality in China and India come to resemble that of the United States, which is unlikely to happen in our lifetime. But now look at the second column of numbers (EXPY), which shows the productivity level associated with each country’s exports. Here the differences are actually tiny in comparison with the previous gaps. The income level associated with U.S. exports exceeds that of China’s exports by only 18 percent, and that of India’s exports by about 50 percent. Furthermore, India’s software exports are not included in this comparison, since the EXPY are calculated for commodity exports only. The important lesson is that the nature of the competition that the United States faces (and will face in the future) is determined by the productivity not of the average foreign producer, but of the very best among them. What is special about international trade today is that the very best producers in these poor but huge economies are very good indeed. To the extent that one worries about such things (and it is not at all clear that one should), there is less reason to be complacent than Trefler would lead us to believe. My other disagreements with Trefler, to the extent that there are any, also relate to differences in emphasis. For example, I think he underestimates the role played by industrial policy in most of the success cases, and he downplays the need to think about intelligent industrial policies. As I have argued elsewhere, the trick in successful industrial policy is not to “pick the winners” (an impossible task if there ever were one), but to know how to “let the losers go” (a much less demanding standard).1 I also would have liked to see greater exploration of the circumstances under which the reduction in the value of job-specific human capital (due to outsourcing and offshoring) has an efficiency (as opposed to a purely distributional) consequence. After all, the key question in this debate is whether competition from China and India undermines technological dynamism in the United States. A plausible channel, to an economist at least, would be through the reduction in the incentive to invest in specific human capital. Trefler talks about this possibility but leaves us guessing as to how seriously we should take it.
1. Dani Rodrik, “Industrial Policy for the Twenty-First Century,” Harvard University, 2004 (http://ksghome.harvard.edu/~drodrik/UNIDOSep.pdf).
64
Brookings Trade Forum: 2005
Pol Antràs: In his insightful paper Daniel Trefler makes three basic points. First, the “offshoring debate” should focus on the offshoring of services because the enormous recent increase in the offshoring of manufactured goods poses no significant new challenges to conventional trade policy. Second, service offshoring is distinct because it affects high-skilled workers. This is important because unlike low-skilled workers, the skill of high-skilled workers is endogenous, and thus service offshoring may well affect U.S. workers’ incentives to acquire human capital. Trefler’s third main point is that although service offshoring poses threats to certain types of white-collar workers, the United States is not likely to lose comparative advantage in high-tech goods to China or India any time soon. This is due to the weak nature of institutions governing economic transactions in these countries. This last point can be paraphrased using international trade jargon: high-tech goods are “contract intensive,” China and India are “institutions scarce” relative to the United States, and the process of “institution accumulation” is a very slow one. Let me next discuss each of these three points in more detail. I do not disagree with the claim that offshoring of manufacturing poses no new challenges to conventional trade policy. Still, I believe that the remarkable growth in the offshoring of manufacturing calls for a change in focus in trade policy debates. As a result of the increased offshoring of manufacturing processes, the role of intermediate inputs in the volume of international trade flows has grown in importance. Although the theoretical literature on the determinants of trade protection has not ignored the fact that trade taxes and subsidies fall not only on final goods but also on intermediate inputs (see, for instance, Grossman and Helpman 1994), these considerations have not been properly implemented in the empirical literature and are often not sufficiently stressed in policy discussions. Beyond its effect on trade policy, the increasing international fragmentation of manufacturing production also has significant implications for the optimal design of fiscal policy and monetary policy. To understand this, it is important to remember that a significant fraction of international offshoring is actually conducted within the boundaries of multinational firms. These firms thus have the ability to use transfer pricing (that is, the pricing of goods and services transacted within the firm) to shift profits across locations in order to minimize their tax burden. The public finance literature has studied this phenomenon, but more work is needed in this important area. A much less emphasized point is that transfer pricing may also have important consequences for the optimal design of monetary policy. In particular, transfer pricing may allow multinational firms to cope better with exchange rate movements (see Lawrence and Rangan 1999 for
Daniel Trefler
65
some suggestive evidence). This asymmetric pass-through between internal and external vertical relationships may influence the effectiveness of exchange rate devaluations, and through this channel it may affect the way monetary policy ought to be conducted in countries that host a large number of affiliates of foreign-based multinational firms.1 Having stressed that much more work is needed in the study of the policy implications of manufacturing offshoring, let me now turn back to service offshoring. This brings me to Trefler’s second main point, namely that service offshoring poses new challenges because it affects relatively skilled white-collar workers. He emphasizes the dynamic costs associated with the offshoring of services jobs, which stem from the reduced incentives to acquire skills that are specific to certain occupations that may eventually be offshored to lower-wage countries. This is certainly a reason for concern, but I would also stress the importance of static costs related to the loss of occupation-specific human capital. Every time a U.S. white-collar worker is displaced and his job is offshored, a significant fraction of the knowledge he acquired on the job gets lost. This destruction of occupation-specific human capital may have important aggregate effects. A challenge for future research is trying to quantify these negative productivity effects of service offshoring. A source of inspiration should be the recent work of Kambourov and Manovskii (2004), who have found that the increase in occupational mobility in the United States during the 1980s generated productivity effects able to account for up to 90 percent of the increase in wage inequality recorded over that period. Service offshoring is different from manufacturing offshoring not only because it involves different types of agents. It is also distinct because it involves the offshoring of relatively knowledge-intensive tasks. Traditional theoretical frameworks for understanding offshoring do not explicitly consider the transmission of knowledge inherent in the international offshoring of services. In Antràs, Garicano, and Rossi-Hansberg (2006) we provide a model of offshoring that places knowledge flows and communication technologies at center stage. We show that characteristics of offshoring and its consequences for “northern” workers are critically affected by the state of information and communication technologies (ICTs). For instance, in the model, an improvement in ICTs is associated with a larger amount of offshoring in the world economy, but also with larger increases in “northern” within-worker inequality resulting from the delocation of certain stages of production. Furthermore, and given that in equilibrium 1. I am grateful to Marc Melitz for suggesting this link between transfer pricing and monetary policy.
66
Brookings Trade Forum: 2005
the industrialized North is a net exporter of knowledge services, we also show that if knowledge flows are not appropriately recorded in official statics, the trade balance of the North will tend to appear in deficit. We view this misrecording as a potential contributor to explaining recent U.S. trade imbalances. Finally, let me move to Trefler’s third main point. His argument rests on the premises that (a) innovation or high-tech sectors are “contract intensive” and that (b) China and India will not become relatively “institution abundant” for years to come. My knowledge of institutional economics is rather limited, so I do not have much to add to the latter point. This brings me to the question: what makes a production process contract-dependent? There is a burgeoning literature analyzing the effects of institutions on comparative advantage, from both a theoretical and an empirical perspective (see Levchenko 2004; Costinot 2005; Nunn 2005; and Antràs 2005). A caveat of these pioneering studies is that they fail to convincingly map the theoretical concept of contract dependence to some observable variables in the data. In Acemoglu, Antràs, and Helpman (2005) we model explicitly the frictions that emerge in multi-agent production in the absence of fully enforceable contracts and show that contract-dependence can naturally be related to task complementarities in production. A salient result of our analysis is that countries with good institutions will gain comparative advantage in production processes that feature high complementarities. In light of our results, Trefler’s first premise can be empirically validated by testing whether indeed innovation is a process that involves high levels of complementarity between the agents engaged in it. I conclude by noting another implication of service offshoring on which Trefler does not comment. In particular, service offshoring leads to a shrinkage of what in international macroeconomics is commonly referred to as the “nontradable sector.” Surely this effect is qualitatively identical to that created by manufacturing offshoring. But quantitatively, this further reduction of the nontradable sector may have important implications for several issues in international macroeconomics. In an influential paper, Obstfeld and Rogoff (2000) showed that six major puzzles in international macroeconomics could be jointly rationalized by appealing to the nontradable nature of certain economic transactions. If this were indeed the common explanation for all these puzzles, we would expect these puzzles to gradually vanish with the advent of service offshoring. Future research should establish whether this is indeed what we observe in the data. General Discussion: The formal presentations in this session launched a lively discussion. Alan Deardorff questioned the extent to which the types of services
Daniel Trefler
67
being offshored really do rely on firm-specific human capital that would be subject to the kinds of hold-up problems Daniel Trefler emphasized in his paper. Giving an example to the contrary, his impression was that workers who could speak English and had strong, up-to-date programming skills could move easily from job to job. He also noted that he did not believe that firm-specificity was likely to help explain what he views as one of the main puzzles raised by offshoring—that products that require skilled labor are being offshored to locations where that skilled labor is relatively scarce. Referring to results from his recent surveys of Indian firms, Rafiq Dossani pointed out that workers who undergo long and extensive training programs are not necessarily more valuable to the firm. For instance, a programmer with four years experience in C++ immediately earns more than someone who has gone through a year and a half of in-house training. Thus he argued that it was important to consider value to the firm when examining the role of firm-specific human capital. Thea Lee raised three sets of issues. First, she expressed the view that previous speakers had been overly dismissive of the potential role for trade policy responses to problems associated with offshoring. She stressed that many current trade rules were not designed to address trade in services. For example, the subsidies rules and countervailing duties remedy do not apply to services. And workers laid off from service sector jobs are not eligible for Trade Adjustment Assistance. If these tools are legitimate for goods, should they not also be legitimate for services? Second, she believed that Trefler’s analysis downplayed the threats posed by India and especially China. She saw his posing the scenario of China exporting everything and the United States exporting nothing as an unhelpful caricature. The question she sees as more relevant is whether China, through currency manipulation and industrial policies, might gain an edge in the products that the United States would wish to expand its exports of. Third, Lee focused on the distributional implications of Trefler’s analysis. She asked if there was evidence that the group of winners was narrowing to very highly skilled workers whose skills are nontransferable, as well as to owners of capital. She was particularly struck by the large share of the gains from trade that appeared to be going to intermediaries. She wondered whether middlemen were more prominent in services than in goods trade, and what this meant for the extent to which consumers gained through lower prices. Trefler responded that there have been large gains for consumers, and that this was reflected in lower import prices and improving U.S. terms of trade. As an example, he cited the PCs and other consumer electronics now commonplace in homes and offices. These have been standardized abroad and have become
68
Brookings Trade Forum: 2005
substantially cheaper. Catherine Mann cautioned, however, that a lack of data on prices for trade in services makes it very difficult to assess what has happened to the terms of trade. Dossani also saw industrial policy in a much more positive light than Trefler’s portrayal in his presentation. Dossani pointed out that less-developed economies such as China can find successful models to follow, making industrial policy more promising and less risky than it would be for countries on the leading edge. Trefler agreed that there may be the potential for successful industrial targeting in developing countries attempting to catch up but reiterated his view that this approach was not appropriate for countries at the technological forefront, such as the United States. Claire Brown cautioned against making the assumption that the relevant skilled labor markets clear. Instead, drawing from interview results of firms in the semiconductor industry, she argued that they are in fact highly rationed. For example, her work suggested there were two distinct groups of engineers: Asian engineers who have been educated and trained in the United States and who could work in either the United States or Asia, and American engineers who do not have the option of repatriating to Asia. Although we talk about the rising skill and educational levels in countries such as India and China, they have a very small cadre of people who can run companies and manage projects. Thus, she noted, the distributional issues for these two groups are very different; and these characteristics imply that the labor markets may be far from functioning efficiently. Theodore Moran emphasized that offshoring of services is, in many key respects, very similar to the offshoring of manufactured products that occurred in previous decades. In particular, both affect white- as well as blue-collar workers, and both involve internationalization of production within multinational corporations. In his view, the two are part of the same phenomenon. Susan Collins noted that the initial title of the project and conference had been services offshoring. This was changed to offshoring of white-collar work to reflect the fact that manufacturing has also been important in explaining white-collar employment trends. Robert Feenstra raised some issues of interpretation about Dani Rodrik’s graphs, which he found very interesting. First, he assumed that the actual income levels reflected averages for each country as a whole. In China, income in the coastal provinces, which do most of the exporting, have been considerably higher than the country average, and thus closer to the constructed trade income level. Second, he wondered whether the observed catch-up of actual income to constructed trade income was really due to continued innovation in China’s most dynamic sectors and provinces, or simply to the spread of these gains to the rest
Daniel Trefler
69
of the country, causing income in the hinterlands to catch up with income in the coastal areas. Finally, Collins asked whether the increased fractionalization of production and trade might have implications for the institutional constraints stressed in Trefler’s paper. She raised the possibility that the ability to siphon off particular pieces of a production process could provide new opportunities for countries with less well developed institutional structures to expand and that this dynamic could limit the informativeness of historical data relating institutional development and composition of production and trade.
70
Brookings Trade Forum: 2005
References Accenture. 2004. “Driving High-Performance Outsourcing: Best Practices from the Masters, Executive Survey Results.” Technical Report 2004. Acemoglu, Daron, Pol Antràs, and Elhanan Helpman. 2005. “Contracts and the Division of Labor.” Harvard University. Antràs, Pol. 2003. “Firms, Contracts, and Trade Structure.” Quarterly Journal of Economics 118 (November): 1375–1418. ———. 2005. “Incomplete Contracts and the Product Cycle.” American Economic Review (September): 1054–73. Antràs, Pol, and Elhanan Helpman. 2004. “Global Sourcing.” Journal of Political Economy 112 (3): 552–80. Antràs, Pol, Luis Garicano, and Esteban Rossi-Hansberg. 2006 (forthcoming). “Offshoring in a Knowledge Economy.” Quarterly Journal of Economics 121 (1). Arora, Ashish, and Alfonso Gambardella. 2005. “The Globalization of the Software Industry: Perspectives and Opportunities for Developed and Developing Countries.” Innovation Policy and the Economy 5: 1–32. Audretsch, David B., and Maryann P. Feldman. 1996. “R&D Spillovers and the Geography of Innovation and Production.” American Economic Review 86 (June): 630–40. Baicker, Katherine, and M. Marit Rehavi. 2004. “Policy Watch: Trade Adjustment Assistance.” Journal of Economic Perspectives 18 (Spring): 239–55. Bardhan, Ashok D., and Cynthia Kroll. 2003. “The New Wave of Outsourcing.” Working Paper 1103. University of California, Berkeley, Fisher Center for Real Estate and Urban Economics. Bhagwati, Jagdish, Arvind Panagariya, and T. N. Srinivasan. 2004. “The Muddles over Outsourcing.” Journal of Economic Perspectives 18 (Fall): 93–114. Clissold, Tim. 2004. Mr. China. London: Constable and Robinson. Coase, Ronald. 1937. “Nature of the Firm.” Economica 4 (November): 386–405. Costinot, Arnaud. 2005. “Contract Enforcement, Division of Labor, and the Pattern of Trade.” Princeton University. Ethier, Wilfred J., and James R. Markusen. 1996. “Multinational Firms, Technology Diffusion and Trade.” Journal of International Economics 41 (August): 1–28. Feenstra, Robert C., and Gordon H. Hanson. 1996. “Globalization, Outsourcing, and Wage Inequality.” American Economic Review Papers and Proceedings 86 (May): 240–45. ———. 1999. “The Impact of Outsourcing and High-Technology Capital on Wages: Estimates for the United States, 1979–1990.” Quarterly Journal of Economics 114 (August): 907–40. Gomory, Ralph E., and William J. Baumol. 2000. Global Trade and Conflicting National Interests. MIT Press.
Daniel Trefler
71
Graham, Frank D. 1923. “Some Aspects of Protection Further Considered.” Quarterly Journal of Economics 37 (February): 199–227. Grossman, Gene M., and Elhanan Helpman. 1994. “Protection for Sale.” American Economic Review 84 (4): 833–50. ———. 2002. “Integration versus Outsourcing in Industry Equilibrium.” Quarterly Journal of Economics 117 (1): 85–120. ———. 2003. “Outsourcing versus FDI in Industry Equilibrium.” Journal of the European Economic Association 1 (2–3): 317–27. ———. 2005. “Outsourcing in a Global Economy.” Review of Economic Studies 72 (1): 135–59. Grossman, Sanford J., and Oliver D. Hart. 1986. “Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration.” Journal of Political Economy 94 (4): 691–719. Gwartney, James, and Robert Lawson. 2003. Economic Freedom of the World: 2003 Annual Report. Vancouver: Fraser Institute. Heckman, James, and Pedro Carneiro. 2003. “Human Capital Policy.” Working Paper 9495. Cambridge, Mass.: National Bureau of Economic Research (February). Helpman, Elhanan. 2004. The Mystery of Economic Growth. Harvard University Press. Hicks, John R. 1953. “An Inaugural Lecture.” Oxford Economic Papers 5 (2): 117–35. Jaffe, Adam B., Manuel Trajtenberg, and Rebecca Henderson. 1993. “Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations.” Quarterly Journal of Economics 108 (3): 577–98. Johnson, George E., and Frank P. Stafford. 1993. “International Competition and Real Wages.” American Economic Review 83 (2): 127–30. Kambourov, Gueorgui, and Iourii Manovskii. 2004. “Occupational Mobility and Wage Inequality.” University of Pennsylvania. Kaufmann, Daniel, Aart Kraay, and Pablo Zoido-Lobaton. 1999. “Governance Matters.” Policy Research Working Paper 2196. Washington: World Bank. Klein, Benjamin, Robert G. Crawford, and Armen A. Alchian. 1978. “Vertical Integration, Appropriable Rents, and the Competitive Contracting Process.” Journal of Law and Economics 21 (2): 297–326. Lawrence, Robert Z., and Subramanian Rangan. 1999. A Prism on Globalization: Corporate Responses to the Dollar. Brookings. Levchenko, Andrei. 2004. “Institutional Quality and International Trade.” IMF Working Paper WP/04/231. Washington: International Monetary Fund. Markusen, James R. 2001. “Contracts, Intellectual Property Rights, and Multinational Investment in Developing Countries.” Journal of International Economics 53 (1): 189–204. Martin, Roger, and Mihnea C. Moldoveanu 2003. “Capital versus Talent: The Battle That’s Reshaping Business.” Harvard Business Review (July).
72
Brookings Trade Forum: 2005
Mokyr, Joel. 1990. The Lever of Riches: Technological Creativity and Economic Progress. New York: Oxford University Press. Nunn, Nathan. 2005. “Relationship-Specificity, Incomplete Contracts, and the Pattern of Trade.” University of Toronto. Nunn, Nathan, and Daniel Trefler. 2005. “The Political Economy of Tariffs and Growth.” University of Toronto. Obstfeld, Maurice, and Kenneth Rogoff. 2000. “The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?” In NBER Macroeconomics Annual 2000, edited by Ben Bernanke and Kenneth Rogoff. MIT Press. Porter, Michael E. 1998. “The Competitive Advantage of Nations.” In On Competition, edited by Michael E. Porter, pp. 155–95. Harvard Business School Press. Puga, Diego, and Daniel Trefler 2002. “Knowledge Creation and Control in Organizations.” Working Paper 9121. Cambridge, Mass.: National Bureau of Economic Research (August). Rodriguez, Francisco, and Dani Rodrik. 2001. “Trade Policy and Economic Growth: A Skeptic’s Guide to the Cross-National Evidence.” In NBER Macroeconomics Annual 2000, edited by Ben S. Bernanke and Kenneth Rogoff, pp. 261–325. MIT Press. Rosenberg, Nathan. 1982. “Technological Interdependence in the American Economy.” In Inside the Black Box: Technology and Economics, edited by Nathan Rosenberg, pp. 55–80. Cambridge University Press. Rosenberg, Nathan, and L. E. Birdzell. 1986. How the West Grew Rich: The Economic Transformation of the Industrial World. New York: Basic Books. Rothschild, Michael. 2003. “What Makes American Public Universities Great?” NBER Reporter (Winter). Samuelson, Paul A. 2004. “Where Ricardo and Mill Rebut and Confirm Arguments of Mainstream Economists Supporting Globalization.” Journal of Economic Perspectives 18 (3): 135–46. Saxonhouse, Gary R. 1998. “Structural Change and Japanese Economic History: Will the 21st Century Be Different?” American Economic Review Papers and Proceedings 88 (2): 408–11. Slaughter, Matthew J. 2004. “Insourcing Jobs: Making the Global Economy Work for America.” Working Paper. Tuck School of Business, Dartmouth College. Trefler, Daniel. 1998. “Immigrants and Natives in General Equilibrium Trade Models.” In The Immigration Debate: Studies on the Economic, Demographic, and Fiscal Effects of Immigration, edited by James P. Smith and Barry Edmonston, pp. 206–38.Washington: National Academy Press. ———. 2004. “Looking Backwards: How Childhood Experiences Impact a Nation’s Wealth.” University of Toronto. UNCTAD. 2004. World Investment Report. Geneva: United Nations.
Daniel Trefler
73
Wasmer, Etienne. 2002. “Interpreting Europe and U.S. Labor Market Differences: The Specificity of Human Capital Investments.” Discussion Paper 549. Bonn, Germany: IZA (August). Williamson, Oliver E. 1975. Markets and Hierarchies: Analysis and Antitrust Implications. New York: Free Press. ———. 1985. The Economic Institutions of Capitalism: Firms, Markets, Relational Contracting. New York: Free Press. Young, Alwyn 1992. “A Tale of Two Cities: Factor Accumulation and Technical Change in Hong Kong and Singapore.” In NBER Macroeconomics Annual, edited by Oliver Jean Blanchard and Stanley Fischer, pp. 13–54. MIT Press. ———. 1994. “Lessons from the East Asian NICs: A Contrarian View.” European Economic Review 38 (3–4): 964–73.
This page intentionally left blank
J. BRADFORD JENSEN Institute for International Economics LORI G. KLETZER University of California–Santa Cruz and Institute for International Economics
Tradable Services: Understanding the Scope and Impact of Services Offshoring
G
lobalization, particularly globalized production, is evolving and broadening from manufacturing into services. Services activities now account for a larger share of global trade than in the past. Services trade has almost doubled over the past decade: in the period 1992 to 2002, exports increased from $163 billion to $279 billion, and imports increased from $102 billion to $205 billion. These changes, and their implications for American firms and workers, have attracted widespread attention. Coincident with the broadening of global economic integration from manufacturing to services, the face of job displacement in the United States is changing. While manufacturing workers have historically accounted for more than half of displaced workers, over the period 2001–03, nonmanufacturing workers accounted for 70 percent of displaced workers.1 The share of job loss accounted for by workers displaced from information, financial services, and professional and business services nearly tripled, from 15 percent during the 1979–82 recession to 43 percent over the 2001–03 period. The industrial and occupational shift We appreciate the comments and suggestions of our Brookings Trade Forum discussants, Jared Bernstein and Robert Feenstra, as well as those of Andrew Bernard, Catherine Mann, Michael Mussa, Dave Richardson, Peter Schott, and seminar participants at the Institute for International Economics; the University of California, Santa Cruz; and the 2004 Empirical Investigations in International Trade conference. We gratefully acknowledge the support of the Alfred P. Sloan Foundation. 1. The shift in job loss from manufacturing and production workers toward service and whitecollar (nonproduction) workers has been in evidence since the recession of the early 1990s. At that time, concerns about downsizing and reengineering were coincident with a rise in the share of white-collar and service sector job loss. See Podgursky (1992); Farber (1993); Gardner (1993); and Kletzer (1995, 1998).
75
76
Brookings Trade Forum: 2005
in job loss has been associated with a rise in the probability of job loss for moreeducated workers.2 Bringing these two trends together, the changing mix of industries exposed to international trade in services may have deep implications for the structure of U.S. industry and labor markets in the future. Currently, there is little clear understanding of the role of services globalization in domestic employment change and job loss. More fundamentally, there is little clear understanding of the size and extent of services offshoring, how large it is likely to become in the near-term future, or what impact it is having on the U.S. economy. Fueled by the 2004 presidential race and continued slack in the labor market, the services offshoring debate became headline material. The literature on services offshoring is expanding rapidly. A nonexhaustive list of recent contributors includes: Amiti and Wei (2004); Arora and Gambardella (2004); Bardhan and Kroll (2003): Bhagwati, Panagariya, and Srinivasan (2004); Brainard and Litan (2004); Bronfenbrenner and Luce (2004); Dossani and Kenney (2003, 2004); Kirkegaard (2004); Mann (2003); Samuelson (2004); and Schultze (2004). Despite the attention, relatively little is known about how many jobs may be at risk of relocation or how much job loss is associated with the business decisions to offshore and outsource. There are a few prominent projections, advanced mostly by consulting firms. The dominant and most widely quoted projection of future job losses due to movement of jobs offshore is Forrester Research’s estimate of 3.3 million.3 Others include: Deloitte Research’s estimate that by 2008 the world’s largest financial service companies will have relocated up to 2 million jobs to low-cost countries offshore; Gartner Research’s prediction that by the end of 2004 10 percent of IT jobs at U.S. IT companies and 5 percent of IT jobs at non-IT companies will have moved offshore; and Goldman Sachs’s estimate that 300,000 to 400,000 services jobs have moved offshore in the past three years, and that 15,000 to 30,000 jobs a month, in manufacturing and services combined, will be subject to offshoring in the future.4 It is clear that changes in technology are enabling more activities to be traded internationally. What is unclear is how large these trends are likely to become, 2. It is still the case that less-educated workers have the highest rates of job loss overall. Over the 2001–03 period, the rate of job loss for workers with a high school diploma or less was .141; for workers with at least some college experience, the rate of job loss was .096 (estimates from the 2004 Displaced Worker Survey). See Farber (2005) for a more detailed examination of worker characteristics and the risk of job loss. 3. See McCarthy (2002). The Forrester projection was updated in 2004 to 3.4 million. 4. See, in order, Gentle (2003); Gartner Research (2004); and Tilton (2003).
J. Bradford Jensen and Lori G. Kletzer
77
the sectors and occupations affected to date and going forward, and the impact on workers of the resulting dislocations. Without understanding the nature and scope of the changes, it is difficult to formulate effective public policy to address emerging needs. This paper develops a new empirical approach to identifying, at a detailed level, service activities that are potentially exposed to international trade. We use the geographic concentration of service activities within the United States to identify which service activities are traded domestically. We classify activities that are traded domestically as potentially tradable internationally. Using the identified industries and occupations, we develop estimates of the number of workers who are in tradable activities for all sectors of the economy. We compare the demographic characteristics of workers in tradable and nontradable activities and employment growth in traded and nontraded service activities. We also examine the risk of job loss and other employment outcomes for workers in tradable activities. To preview the results, we find considerable employment shares in tradable service industries and occupations. Based on our estimates, there are more workers in tradable professional and business service industries than in tradable manufacturing industries. We also examine the characteristics of workers in tradable and nontradable activities and find that workers in tradable sectors have higher skills and significantly higher wages. Within specific sectors like professional services, the earnings differentials are even larger, approaching 20 percent. When we examine employment growth trends across traded and nontraded activities, tradable activities have lower growth rates, due primarily to employment losses in manufacturing. Within services, tradable and nontradable activities have similar growth rates except at the lowest end of the skill distribution. Low-skill tradable industries and occupations have negative average employment growth, whereas employment growth in nontraded, low-skill services is positive (though low). We also examine worker displacement rates in tradable and nontradable service activities. We see some evidence that displacement rates are higher from tradable service industries than from nontradable. We also find higher displacement rates from tradable white-collar occupations than from nontradable. Consistent with the characteristics of employed workers, we find that workers displaced from tradable service activities are more educated, with higher earnings, than workers displaced from nontradable activities. Job loss from tradable and nontradable service activities is costly to workers in terms of earnings losses (comparing new job earnings to old job earnings). Taken together, the results are consistent with the view that economic activity within the United
78
Brookings Trade Forum: 2005
States is moving toward a U.S. comparative advantage in services, similar to manufacturing. In the next section we describe our empirical approach to identifying tradable activities. The following sections describe the tradable and nontradable categories for both manufacturing and services activities; compare worker characteristics in tradable and nontradable services; explore the employment trends in tradable and nontradable services; and consider the most recent evidence on job displacement from tradable activities.
Empirical Approach Historically, services have been considered nontradable, with a paucity of empirical work examining trade in services relative to empirical work on manufacturing. To examine the potential impact of trade in services on the U.S. economy, we wanted to identify the size and scope of services trade at as detailed a level as possible. As many observers and researchers have noted, gathering detailed data on the extent of services offshoring is quite difficult. While the Bureau of Economic Analysis (BEA) provides data on international trade in services, the data on international trade in services that BEA publishes do not provide particularly detailed industry-level data. Table 1 shows the level of industry detail available from BEA. Our interest in examining trade in services in more detail than what is available through the BEA services trade data necessitated an alternative empirical approach to identifying tradable service activities. Our approach to identifying service activities that are potentially tradable is novel: we use the geographic concentration of service activities in the United States to identify industries and occupations that appear to be traded domestically. From this domestic information, we infer that service activities that can be traded within the United States are also potentially tradable internationally. Framework The economic intuition we rely on to develop our baseline measure of tradable services is that nontraded services will not exhibit geographic concentration in production. We observe that goods that are traded tend to be geographically concentrated (to capitalize on increasing returns to scale, access to inputs such as natural resources, etc.), while goods that are not traded tend to be more ubiquitously distributed. We apply this same intuition to service production.
J. Bradford Jensen and Lori G. Kletzer
79
Helpman and Krugman (1985) present a model that demonstrates this intuition. They model a world with two goods, two countries, and three industries, where the first industry is assumed to be a nontradable constant-returns sector, the second industry is an industry with differentiated varieties that are assumed to be costlessly traded, and the third industry is a tradable constant-returns sector. Helpman and Krugman derive the input vectors V(1), V(2), and V(3) for the integrated world equilibrium. With homothetic and identical tastes, if country j has a share s j of world income, it must allocate resources s j V(1) to the nontradable industry; that is, the production of the nontraded good must be allocated between countries in proportion to their shares of world income. Nontraded goods are distributed uniformly according to population and income. This intuition is revealed more descriptively by Paul Krugman, who notes, “In the late twentieth century the great bulk of our labor force makes services rather than goods. Many of these services are nontradable and simply follow the geographical distribution of the goods-producing population—fast-food outlets, daycare providers, divorce lawyers surely have locational Ginis pretty close to zero. Some services, however, especially in the financial sector, can be traded. Hartford is an insurance city; Chicago the center of futures trading; Los Angeles the entertainment capital; and so on. . . . The most spectacular examples of localization in today’s world are, in fact, services rather than manufacturing. . . . Transportation of goods has not gotten much cheaper in the past eighty years. . . . But the ability to transmit information has grown spectacularly, with telecommunications, computers, fiber optics, etc.”5 The idea is that when something is traded the production of the activity is concentrated in a particular region to take advantage of some economies in production. As a result, not all regions will support local production of the good, and some regions will devote a disproportionate share of productive activity to a good and then trade it.6 We use the geographic concentration of service activity within the United States as an indicator that the service is traded within the United States and thus potentially tradable internationally. The “locational Gini” referred to by Krugman is one of several ways to measure geographic concentration.7 The measures compare a region’s share of 5. Krugman (1991, p. 65). 6. The relationship between geographic concentration of production and trade, particularly exports, has a long tradition in both economic geography (where the measure used is the location quotient) and trade analysis (where the measure used is revealed comparative advantage). The measures of economic concentration used in this paper are different from the location quotient and revealed comparative advantage measures, but all the measures have a similar flavor in that they compare the share of production (or exports) in a particular region to an “expected” baseline. 7. Among the different empirical approaches to measuring geographic concentration and agglomeration are Duranton and Overman (2004).
80
Brookings Trade Forum: 2005
Table 1. Private Services Trade by Type, 2002 Millions of dollars Trade type
Exports, 2002
Imports, 2002
Travel Overseas Canada Mexico Passenger fares Other transportation Freight Port services Royalties and license fees Affiliated U.S. parents’ transactions U.S. affiliates’ transactions Unaffiliated Industrial processes Other Other private services Affiliated services U.S. parents’ transactions U.S. affiliates’ transactions Unaffiliated services Education Financial services Insurance services Telecommunications Business, professional, and technical services Accounting, auditing, and bookkeeping services Advertising
66,547 54,772 6,268 5,507 17,046 29,166 12,330 16,836 44,142 32,218 29,066 3,152 11,924 3,900 8,024 122,594 43,500 25,194 18,306 79,094 12,759 15,859 2,839 4,137 28,799 360 633
58,044 44,494 6,489 7,061 19,969 38,527 25,973 12,554 19,258 15,132 2,958 12,174 4,126 1,935 2,192 69,436 32,367 17,529 14,838 37,069 2,466 3,665 15,348 4,180 10,732 716 1,360 (continued)
employment in or output of an activity with the region’s share of overall economic activity. We make use of two common measures of geographic concentration; but before turning to those measures we address one more conceptual issue. Demand-Induced Agglomeration and Intermediate Services Measures of geographic concentration are a way to implement the intuition described above. Most measures of concentration use the region’s share of employment in an industry relative to the region’s share of total employment. The measures of concentration do not differentiate the reasons activity is concentrated. It does not matter whether production is concentrated because of the
J. Bradford Jensen and Lori G. Kletzer
81
Table 1. Private Services Trade by Type, 1992–2002 (Continued) Millions of dollars Trade type Agricultural, mining, and on-site processing services Agricultural and mining services Waste treatment and depollution services Architectural, engineering, and other technical services Computer and data processing services Construction, architectural, engineering, and mining services Construction Data base and other information services Industrial engineering Installation, maintenance, and repair of equipment Legal services Management, consulting, and public relations services Medical services Miscellaneous disbursements Operational leasing Research, development, and testing services Sports and performing arts Trade-related services Training services Other business, professional, and technical services Other unaffiliated services
Exports, 2002
Imports, 2002
366 346 20 1,916 3,004
273 259 14 312 1,057
n.a. 654 2,426 749 4,992 3,270 1,696 1,901 623 3,573 1,086 175 353 501 430 14,700
n.a. 226 236 185 812 768 1,188 n.a. 1,522 190 1,040 110 95 361 283 679
Source: Bureau of Economic Analysis. n.a. = not available.
location of natural resources, increasing returns in production, or spillovers due to the agglomeration of workers; the concentration of production indicates that the good or service is produced in a location different from where it is consumed. So, in general, the reason for the concentration does not matter to us, except in one instance. If a service is nontradable and demand for the service is concentrated (that is, if industries that use the nontraded service are geographically concentrated), the service industry will be geographically concentrated and we would incorrectly infer that the service is tradable. To incorporate this case into our approach, we extend the intuition from the framework. If a nontradable industry provides intermediate inputs to a downstream industry, we would expect the geographic distribution of the nontraded intermediate industry to follow the distribution of the downstream industry. Instead of being distributed with income, the nontraded good is distributed in proportion to the geographic distribution of demand for that industry.
82
Brookings Trade Forum: 2005
We construct region-specific measures of demand for each industry using the 1999 input-output use tables produced by the Bureau of Economic Analysis.8 This measure of industry demand share (IDSi,p ) represents how much geographic concentration there is in demand for a good or service i in a particular region p. We construct the demand for industry i in Place of Work Metro Area p by: IDSi,p = j (Yi,j /Yi * InEMPj,p /InEMPj ),
(1)
where Yi,j = the output of industry i used by industry j (including government and private households as “industries”); Yi = total output of industry i; InEMPj,p = industry j employment in region p; InEMPj = total employment in industry j. We include both direct use and investment in the “use” of industry i output by industry j. To construct the region-specific measures of demand for each occupation, we use the industry-region-specific demand measures described above and weight those by the share of occupation employment in an industry. ODSo,p = j (IDSj,p * OcEMPo, j /OcEMPo),
(2)
where IDSj,p = industry demand share for industry j in region p; OcEMPo,j = occupation o employment in industry j; and OcEMPo = total employment in occupation o. These adjustments take account of the concentration of downstream industry concentration and adjust the “denominator” in the geographic concentration measures that follow. Measuring Geographic Concentration The first measure of economic concentration, as described in Ellison and Glaeser (1997), is: ECi = p (si,p – xp)2.
(3)
8. For more information, see www.bea.doc.gov/bea/dn2/i-o.htm. We aggregate some BEA input-output (IO) industries to a level consistent with the industry classification used by the Census Bureau on the 2000 Decennial PUMS (Public Use Micro Sample).
J. Bradford Jensen and Lori G. Kletzer
83
The measure is an index for comparing a region’s share of industry employment (si,p) with the area’s share of aggregate activity/employment (xp). When an area’s employment share in an activity is significantly greater than the area’s share of aggregate employment, this is interpreted as indicating a concentration, or specialization, in the given activity. The index EC provides a national index for each industry, and measures of EC indicating geographic concentration are interpreted as indicative of trade in that activity, in the sense that “local” employment exceeds “local” demand in some areas and the difference is traded outside the area. We modify the EC measure to look at the difference between the region’s share of industry employment and the region’s share of industry demand, as noted above: ECi = p (si,p – IDSi,p)2.
(4)
The new measure of EC is an index for comparing a region’s share of an industry’s employment (si ) with the region’s share of demand for that industry (IDSi,p). We do not make the Herfindahl adjustment that Ellison and Glaeser (1999) use in their index of agglomeration because we are not interested in agglomeration (the co-location of different firms in the same industry), but are interested in pure geographic concentration (whether the concentration is due to one firm or a number of firms). If economic activity is concentrated because significant scale economies are captured within a firm, we do not want to discount this concentration. The second measure of geographic concentration we use is the Gini coefficient. The Gini coefficient (G) for the concentration of industry activity is given by: Gi = | 1 – p (Yi,p – 1 + Yp) * ( Xi,p – 1 – Xp ) | ,
(5)
where p’s index regions (sorted by the region’s share of industry employment), Yi,p is the cumulative share of industry i employment in region p, Yi,p – 1 is the cumulative share of industry i employment in the region (p – 1) with the next lowest share of industry employment, Xp is the cumulative share of total employment in region p, and Xp – 1 is the cumulative share of total employment in region p – 1. We modify the Gini measure to: Gi = | 1 – p (Yi,p – 1 + Yi,p) * (IDSi,p – 1 – IDSi,p) | , where IDSi,p is the region’s share of demand for industry i.
(6)
84
Brookings Trade Forum: 2005
Implementation We implement these measures using employment information from the 2000 Decennial Census of Population Public Use Micro Sample (PUMS) files. We use as our geographic entity the Consolidated Metropolitan Statistical Area or the Metropolitan Statistical Area where an individual reports working.9 We construct the measures of geographic concentration for each industry. Industries that are geographically concentrated are considered tradable. We recognize that the use of worker-level data to investigate economic concentration is somewhat unusual. We pursue this strategy because we are interested in both industrial concentration and occupational concentration. The ability to identify both industries and occupations that are tradable is an important feature of the empirical strategy because many of the service activities that are reportedly being globally sourced are tasks within the service “production” process (for example, a bank’s customer service/call center component may be moved offshore, but not the banking relationship); occupations correspond more closely to these types of activities than industries do. We construct the adjusted G and EC measures for both industries and occupations. The correlation between the EC measure and the G measure is quite high, .713 for industries and .732 for occupations. For the remainder of this paper, we focus on the G results.
Classifying Industries and Occupations as Tradable or Nontradable An important task in our empirical approach is to identify the level of geographic concentration that indicates that an industry or occupation is “tradable.”10 We started exploring where to impose the tradable/nontradable threshold with industries because we have a much better sense of which industries are
9. For regions, we use the Place of Work Consolidated Metropolitan Area (POWCMA5) field on the Decennial PUMS. When POWCMA is coded as a nonmetropolitan area or a mixed metro/nonmetro area, we concatenate the Place of Work state code with the POWCMA5 code. For more information on the 5 percent sample PUMS, see www.census.gov/Press-Release/www/ 2003/PUMS5.html. 10. While choosing the threshold for nontradable and tradable is inherently arbitrary, we ran a number of robustness checks on the results reported in the paper. With the exception of the share of employment in the tradable sector (which decreases as the threshold rises), the results are robust to the choice of threshold.
J. Bradford Jensen and Lori G. Kletzer
85
Figure 1. Geographic Concentration of Industries 0.9
0.8
0.7
Gini Coefficient
0.6
0.5
0.4
0.3
0.2
0.1
0 0
100
200
300
400
500
600
700
Retail
Transportation
800
900
1000
NAICS Industry Ag
Mining
Utilities
Construction
Manufacturing
Wholesale
Services
Public Admin
tradable, particularly goods-producing industries. We initially placed industries into three roughly equal groups: Gini class 1 (least geographically concentrated) when the industry Gini was less than .1; Gini class 2 when the industry Gini was between .1 and .3; Gini class 3 (most geographically concentrated) when the Gini coefficient was greater than or equal to .3. Approximately 36 percent of industries are in Gini class 1, about 37 percent are in Gini class 2, and 27 percent are in Gini class 3. Figure 1 plots the Gini coefficients for all industries by two-digit NAICS code. The pattern exhibited in figure 1 is generally consistent with our priors that tradable industries will be geographically concentrated. For example, industries in the goods-producing sectors of Agriculture, Mining, and Manufacturing are typically in the top two Gini classes. Only five of the ninety-two industries in these sectors are in Gini class 1: Cement and Concrete; Machine Shops; Miscellaneous Manufacturing n.e.c.; Structural Metals and Tanks; and Printing and Related Activities. All of these industries seem to be either nontraded because of a high weightto-value ratio (such as Cement and Concrete), or they are categories that include a range of potentially dissimilar activities (Miscellaneous Manufacturing n.e.c.) that make them appear to be broadly geographically distributed. Most agriculture,
86
Brookings Trade Forum: 2005
mining, and manufacturing products are considered tradable; so as a first-order approximation, classifying the lowest geographic concentration category (Gini class 1) as nontradable seems appropriate for these sectors.11 Using a Gini coefficient of .1 as the threshold for tradable seems to make sense in other sectors as well. Industries in the retail trade sector are primarily classified as nontradable. Industries in the Transportation sector are mostly classified as tradable. In Public Administration, most activities are nontradable; Public Finance and the military are exceptions. In the Service sector, industries are balanced between nontradable and tradable. Table 2 provides a complete list of service industries by 2-digit NAICS sector and the industry’s Gini class.12 Table 3 shows the share of employment classified in tradable industries by major NAICS group. Again, the employment shares across categories and industries conform to our priors. All employment in the Agriculture and Mining sectors is classified as tradable (in one of the top two Gini classes). In Manufacturing, most employment is in the tradable sector.13 Utilities are mostly nontradable and Construction is entirely nontraded. For the remainder of the paper, we categorize industries with a Gini coefficient below .1 as nontradable and industries with a Gini coefficient greater than or equal to .1 as tradable. Size and Scope of Tradable Service Industries We use the categorization of industries as tradable and nontradable to develop estimates of the employment potentially affected by trade in services. Table 4 shows the share of total employment in tradable and nontradable industries by major NAICS group. In contrast to traditional characterizations of services as predominantly nontradable, our categorization suggests that a significant share of 11. Another check on the industry classification is to examine the correlation of geographic concentration of manufacturing industries with the level of trade intensity in those industries. The mean industry trade share [(imports + exports)/domestic production] for Gini class 1 = .40, Gini class 2 = .57, Gini class 3 = .71. If Manufacturing Machinery n.e.c. is removed from Gini class 1 (by virtue of its not being a consistent industry), the mean trade share for that class falls to .35. The pattern revealed is one of a positive correlation between Gini class and mean trade share, with some notable variation within class. 12. Higher education may appear to stand out in table 2 as a nontradable service industry. U.S. colleges and universities, particularly research institutions, have an acknowledged global comparative advantage and attract many foreign students. The sector also includes community colleges that are, by design, geographically dispersed. The types of specialized scientific occupations associated with research institutions (the most likely to “export” educational services) are geographically concentrated and thus considered tradable. 13. Alternatively, if we modify the cutoff and use .2 as the break between tradable and nontradable, 28 percent of manufacturing employment would be in the nontradable sector.
J. Bradford Jensen and Lori G. Kletzer
87
Table 2. Service Industries, Gini Coefficient Class 2-digit NAICS
Industry description
Gini coefficient class
51 51 51 51 51 51 51 51 51 51 51
Information Newspaper publishers Radio and television broadcasting and cable Libraries and archives Wired telecommunications carriers Data processing services Other telecommunication services Publishing, except newspapers and software Other information services Motion pictures and video industries Sound recording industries Software publishing
1 1 1 2 2 2 2 3 3 3 3
52 52 52 52 52
Finance and insurance Savings institutions, including credit unions Banking and related activities Insurance carriers and related activities Nondepository credit and related activities Securities, commodities, funds, trusts, and other financial investment
1 1 2 2 3
53 53 53 53 53
Real estate and rental and leasing Video tape and disk rental Other consumer goods rental Commercial, industrial, and other intangible assets rental and leasing Real estate Automotive equipment rental and leasing
1 1 2 2 2
54 54 54 54 54 54 54 54 54 54
Professional, scientific, and technical services Veterinary services Accounting, tax preparation, bookkeeping and payroll services Architectural, engineering, and related services Other professional, scientific, and technical services Legal services Specialized design services Computer systems design and related services Advertising and related services Management, scientific, and technical consulting services Scientific research and development services
1 1 2 2 2 2 2 2 2 3
55
Management Management of companies and enterprises
2
56 56 56 56 56
Administrative support Waste management and remediation services Business support services Services to buildings and dwellings Landscaping services Employment services
1 1 1 1 2 (continued)
88
Brookings Trade Forum: 2005
Table 2. Service Industries, Gini Coefficient Class (Continued) 2-digit NAICS
Industry description
Gini coefficient class
56 56 56
Other administrative and other support services Investigation and security services Travel arrangement and reservation services
2 2 2
61 61 61 61
Education Elementary and secondary schools Colleges and universities, including junior colleges Other schools, instruction, and educational services Business, technical, and trade schools and training
1 1 1 2
62 62 62 62 62 62 62 62 62 62 62 62 62 62 62
Health care and social services Hospitals Nursing care facilities Vocational rehabilitation services Offices of physicians Outpatient care centers Offices of dentists Offices of optometrists Residential care facilities, without nursing Child day care services Home health care services Other health care services Office of chiropractors Individual and family services Community food and housing, and emergency services Offices of other health practitioners
1 1 1 1 1 1 1 1 1 1 1 1 1 2 2
71 71 71 71
Arts, entertainment, and recreation Bowling centers Other amusement, gambling, and recreation industries Museums, art galleries, historical sites, and similar institutions Independent artists, performing arts, spectator sports, and related industries
72 72 72 72
Accommodation Drinking places, alcoholic beverages 1 Restaurants and other food services 1 Recreational vehicle parks and camps, and rooming and boarding houses 1 Traveler accommodation 2 (continued)
1 1 2 2
total employment is in tradable service industries. For example, more workers are in tradable industries in the services sector than in the manufacturing sector. The sum of the share of total employment in industries that are tradable in professional services (NAICS 51–56) is 13.7 percent and larger than the share of employment in tradable manufacturing industries (12.4 percent). There are sizable service sec-
J. Bradford Jensen and Lori G. Kletzer
89
Table 2. Service Industries, Gini Coefficient Class (Continued) 2-digit NAICS
Industry description
81 81 81 81 81 81 81 81 81
Other services Beauty salons Funeral homes, cemeteries, and crematories Personal and household goods repair and maintenance Automotive repair and maintenance Barber shops Religious organizations Commercial and industrial machinery and equipment repair and maintenance Dry cleaning and laundry services Car washes Electronic and precision equipment repair and maintenance Civic, social, advocacy organizations, and grant-making and giving Nail salons and other personal care services Other personal services Business, professional, political, and similar organizations Labor unions Footwear and leather goods repair
92 92 92 92 92 92 92 92 92 92 92 92 92 92 92
Public administration Justice, public order, and safety activities Administration of human resource programs Other general government and support Executive offices and legislative bodies Military Reserves or National Guard Administration of economic programs and space research Administration of environmental quality and housing programs Public finance activities National security and international affairs U.S. Armed Forces, branch not specified U.S. Coast Guard U.S. Air Force U.S. Army U.S. Navy U.S. Marines
81 81 81 81 81 81 81
Gini coefficient class
1 1 1 1 1 1 1 1 1 1 1 2 2 2 3 3 1 1 1 1 1 1 1 2 3 3 3 3 3 3 3
tors correctly characterized as having low shares of employment in tradable industries (education, health care, personal services, and public administration). However, because the service sector is much larger than the manufacturing sector, the number of workers potentially exposed to international trade in services is actually larger than the number of exposed workers in manufacturing.
90
Brookings Trade Forum: 2005
Table 3. Share of Sector Employment by Gini Coefficient by NAICS Sector Percent NAICS sector 11 21 22 23 31 32 33 3M 42 44 45 4M 48 49 51 52 53 54 55 56 61 62 71 72 81 92
Description
Gini class 1
Gini class 2
Gini class 3
Agriculture Mining Utilities Construction Manufacturing Manufacturing Manufacturing Manufacturing Wholesale trade Retail trade Retail trade Retail trade Transportation and warehousing Transportation and warehousing Information Finance and insurance Real estate and rental and leasing Professional, scientific, technical services Management Administrative support Education Health care/social services Arts, entertainment, recreation Accommodation Other services Public administration
0 0 80.89 100.00 0 21.99 14.44 0 45.82 81.72 88.65 100.00 42.81 0 33.25 32.05 9.06 13.95 0 59.53 98.89 97.80 67.35 81.92 79.77 71.68
87.95 24.24 15.31 0 40.39 44.88 65.36 100.00 50.62 18.28 11.35 0 22.03 100.00 50.37 50.98 90.94 79.87 100.00 40.47 1.11 2.20 32.65 18.08 9.86 4.63
12.05 75.76 3.80 0 59.61 33.13 20.21 0 3.57 0 0 0 35.17 0 16.38 16.97 0 6.18 0 0 0 0 0 0 10.37 23.69
60.82
29.75
9.43
All Industries
Occupation Results We are also interested in categorizing occupations as tradable and nontradable. We are interested in identifying tradable occupations because, at least based on anecdotal reports in the press, some intermediate inputs into service production might be tradable even though the service industry is not (think computer programming for the banking industry). We use a similar methodology to classify occupations into tradable and nontradable categories. We construct a demand-weighted Gini coefficient for each occupation as described above and use the same Gini = .1 threshold for the nontradable/tradable categorization. Table 5 shows the share of employment by Major Standard Occu-
J. Bradford Jensen and Lori G. Kletzer
91
Table 4. Share of Total Employment in Tradable and Nontradable Industries by NAICS Sector Percent NAICS sector 11 21 22 23 31 32 33 3M 42 44 45 4M 48 49 51 52 53 54 55 56 61 62 71 72 81 92
Description Agriculture Mining Utilities Construction Manufacturing Manufacturing Manufacturing Manufacturing Wholesale trade Retail trade Retail trade Retail trade Transportation and warehousing Transportation and warehousing Information Finance and insurance Real estate and rental and leasing Professional, scientific, technical services Management Administrative support Education Health care/social services Arts, entertainment, recreation Accommodation Other services Public administration All industries
Nontradable
Tradable
0 0 0.76 6.86 0 0.81 1.16 0 1.66 5.90 2.91 0.62 1.32 0 1.04 1.64 0.16 0.82 0 1.99 8.75 10.90 1.12 4.52 3.76 4.14
1.36 0.39 0.18 0 2.17 2.86 6.86 0.53 1.96 1.32 0.37 0 1.76 1.27 2.08 3.47 1.63 5.08 0.06 1.35 0.10 0.25 0.54 1.00 0.95 1.63
60.82
39.18
pational Classification group by Gini class. The groupings are largely consistent with our priors. The occupational groups with large shares of employment classified as tradable include: Business and Financial Operations (68 percent); Computer and Mathematical Occupations (100 percent); Architecture and Engineering (63 percent), Legal (96 percent), and Life, Physical and Social Sciences (83 percent).14 The notable nontradable occupational groups include 14. Van Welsum and Reif (this volume) offer a list of U.S. occupations (at the 3-digit level) identified as “potentially affected by offshoring” in table A-2. As explained in the chapter, their method relies on occupations having “offshorability attributes” that rely on the use of information and communication technologies, highly codifiable knowledge, and no face-to-face contact. There is overlap
92
Brookings Trade Forum: 2005
Table 5. Share of Occupation Employment by Gini Coefficient by Major Occupation Category Percent SOC 2-digit code
Description
Gini class 1
Gini class 2
Gini class 3
11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55
Management Business/financial operations Computer/mathematical Architecture/engineering Life, physical, social sciences Community/social services Legal Education and library Arts, design, entertainment Health care practitioners/technicians Health care support Protective service Food preparation/serving Building maintenance Personal care service Sales and related Office/administrative support Farm, fish, forestry Construction/extraction Installation, maintenance, repair Production Transportation/material moving Military specific
34.48 31.73 0 36.04 16.32 100.00 3.78 99.54 17.13 86.56 96.73 59.83 95.68 98.54 82.64 75.41 93.14 0 61.37 90.00 80.30 89.20 0
61.15 65.96 73.07 58.31 58.61 0 96.22 0.46 75.02 13.10 3.27 40.17 4.32 1.46 7.22 21.82 6.66 81.01 36.18 8.89 17.15 5.86 0
4.37 2.32 26.93 5.65 25.08 0 0 0 7.85 0.34 0 0 0 0 10.13 2.77 0.20 18.99 2.45 1.11 2.55 4.95 100.00
71.66
24.86
3.47
All occupations
Education and Library (99 percent nontradable); Health Care Practitioners (86 percent); Health Care Support (97 percent), Food Preparation (96 percent). On the blue-collar side, 90 percent of employment in Installation, Maintenance, and Repair is classified as nontradable, as is 80 percent of Production and 89 percent of Transportation and Material Moving.15 between the two lists of occupations, although our method identifies a larger set of tradable occupations. Van Welsum and Vickery (2005) offer a list of U.S. industries potentially affected by offshoring, in table 6. Our detailed industry list shares similarities with theirs, but our list excludes a number of retail industries (dairy stores, liquor stores, and others) included in their list. 15. The geographic concentration results are at first counterintuitive for production occupations given the manufacturing industry results. Production occupations are typically not industryspecific but instead are functional activities and are thus distributed more broadly.
J. Bradford Jensen and Lori G. Kletzer
93
Table 6. Share of Employment in Tradable and Nontradable Occupations and Industries Percent Occupation category (SOC 2-digit code)
Nontradable occupations
Tradable occupations
Management occupations (11) Non-tradable industries Tradable industries
23.97 10.51
26.58 38.94
Business and financial operations occupations (13) Nontradable industries Tradable industries
14.11 17.61
27.72 40.56
Computer and mathematical occupations (15) Nontradable industries Tradable industries
0 0
24.22 75.78
Architecture and engineering occupations (17) Nontradable industries Tradable industries
8.46 27.59
13.30 50.66
Life, physical, and social science occupations (19) Nontradable industries Tradable industries
7.28 9.03
36.49 47.20
Legal occupations (23) Nontradable industries Tradable industries
3.54 0.24
18.89 77.33
50.03 21.64
10.79 17.54
All occupations Total nontradable industries Total tradable industries
The last two rows of table 6 show for all occupations how many workers are in occupations classified as tradable in industries classified as nontradable. In the aggregate, the share of workers in tradable occupations and nontradable industries is not large, about 10 percent. However, for business and professional occupations, the share of workers in tradable occupations in nontradable industries is much larger. The typical professional occupation has about 25 percent of employment in tradable occupations in nontradable industries. To the extent that firms can vertically “disintegrate” the provision of these intermediate service inputs, workers in these tradable occupations are potentially vulnerable to trade even though their industry is not tradable. This suggests that for service activities the share of workers potentially vulnerable to trade is probably understated. Outside of education and health care occupations, the typical white-collar occupation involves a potentially tradable activity.
94
Brookings Trade Forum: 2005
Table 7. Mean Earnings and Demographic Characteristics for Selected and All Industries Percent, unless otherwise noted Industry (NAICS code) Manufacturing (3x) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Information (51) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Finance and insurance (52) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age
Nontradable
Tradable
36,974 75.1 6.1 9.7 2.6 13.8 85.3 40.0
39,901 67.8 9.7 11.7 6.0 20.4 82.9 40.2
35,472 50.9 10.4 7.8 9.4 37.4 94.2 38.7
49,510 55.9 11.5 7.3 10.6 41.3 96.2 37.6
38,170 29.0 11.5 7.8 7.1 30.5 97.1 38.1
54,460 42.7 9.2 6.4 10.2 43.8 97.4 39.1 (continued)
Worker Characteristics Beyond mere employment counts, we also examine demographic characteristics such as education, age, gender, and earnings to identify whether there are differences between workers in tradable service activities and those in nontradable industries and occupations. These characteristics are available from the 2000 Decennial Census of Population Public Use Micro Sample (PUMS) 5 percent sample.16 Table 7 shows the demographic characteristics of workers in tradable industries and nontradable industries in aggregate. Workers in tradable industries have 16. For more information on the 5 percent sample PUMS see www.census.gov/PressRelease/www/2003/PUMS5.html.
J. Bradford Jensen and Lori G. Kletzer
95
Table 7. Mean Earnings and Demographic Characteristics for Selected and All Industries (Continued) Percent, unless otherwise noted Industry (NAICS code) Real estate and rental and leasing (53) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Professional, scientific, technical services (54) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Management (55) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Administrative support (56) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma All industries Employment income Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age
Nontradable
Tradable
23,056 58.1 9.1 10.8 1.9 13.3 84.7 31.1
42,915 51.1 8.6 9.7 6.7 29.7 90.6 42.4
42,246 35.3 5.1 5.0 16.6 52.5 97.1 39.5
57,959 57.1 5.5 5.6 25.7 59.5 97.8 39.3
… … … … … … … …
61,285 45.5 5.4 4.9 14.3 49.7 97.8 40.5
24,039 64.1 11.9 22.2 2.0 10.7 72.3 37.2
28,742 48.5 17.6 12.2 5.0 23.4 88.0 36.1
30,966 49.6 10.2 10.4 10.2 26.6 87.0 38.8
41,836 60.1 9.9 10.3 9.2 30.2 88.7 39.4
96
Brookings Trade Forum: 2005
higher incomes, are more likely to be male, and are more likely to have a college degree (though not an advanced degree). The table also breaks out these same characteristics for selected service industries classified as tradable and nontradable. We present the results for the manufacturing sector as a benchmark for demographic characteristics typically associated with trade-affected workers. Workers in tradable service industries are higher paid and more skilled than workers in tradable manufacturing. Within services, the most striking feature of the service industry results is the difference in annual earnings. Across all major service sector groups, the differential in earnings between tradable and nontradable industries is large, with tradable services having appreciably higher wages. Service workers in tradable industries also tend to have attained a higher level of education and are more likely to be male and white. Table 8 shows the results for all occupations divided into tradable and nontradable groups. Individuals in occupations identified as tradable tend to have higher earnings, are more likely to be male and have more years of schooling. The table also shows the same characteristics for selected occupations. Again, as in the industry results, workers in tradable occupations earn more and are more highly educated than workers in nontradable service occupations. In tables 9–12, we estimate a number of regressions to examine whether the earnings differentials in tradable industries and occupations are the result of higher educational attainment. Table 9 shows regression results for all industries and NAICS 51–56 industries. Across all industries, controlling for observable demographic characteristics and industry (2-digit NAICS) and regional (POWCMA) fixed effects, workers in tradable industries have 6 percent higher wages. For workers in professional and business service industries, the differential associated with being in a tradable industry is even larger. Again controlling for observable demographic characteristics, in the professional service sector, workers in tradable industries have almost 15 percent higher wages than workers in nontradable industries in the same sector. Table 10 shows a similar specification for occupations. The first column reports the results for all occupations, and the second column reports the results for “high-end” service occupations.17 Across all occupations, workers in tradable occupations receive 9 percent higher wages than workers in nontradable occupations. For high-end service occupations, workers in the tradable sector receive almost 13 percent higher wages, even after controlling for demographic characteristics and occupation group (2-digit SOC) and region. 17. High-end service occupations include SOC major groups 11, 13, 15, 17, 19, 23, and 29. See table 8 for the names of the SOC major groups.
J. Bradford Jensen and Lori G. Kletzer
97
Table 8. Mean Earnings and Demographic Characteristics for Occupations Percent, unless otherwise noted Industry (NAICS code) Management (11) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Business and financial operations (13) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Computer and mathematical occupations (15) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Architecture and engineering occupations (17) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age Life, Physical, and Social Science Occupations (19) Employment income (dollars) Male Percent African American Percent Hispanic With advanced degree With bachelor’s degree With high school diploma Age
Nontradable
Tradable
51,399 56.2 8.3 6.8 19.9 46.5 95.2 41.8
69,029 67.3 4.7 5.0 15.7 49.6 95.8 42.6
42,813 41.3 10.3 6.9 10.5 44.0 97.6 40.4
51,998 48.0 8.3 5.4 16.2 61.6 98.6 40.2
… … … … … … … …
54,297 70.3 6.8 4.5 17.8 59.9 99.1 37.3
40,505 82.5 5.7 6.4 5.3 26.2 96.2 39.4
62,115 89.0 3.9 4.1 25.5 76.2 99.9 40.6
29,339 57.4 7.0 7.2 11.6 40.0 96.4 36.0
50,000 59.2 4.6 4.0 54.4 85.3 99.2 40.3 (continued)
98
Brookings Trade Forum: 2005
Table 8. Mean Earnings and Demographic Characteristics for Occupations Percent, unless otherwise noted Industry (NAICS code) Legal Occupations (23) Employment income (dollars) Male Percent African American Percent Hispanic With advanced degree With bachelor’s degree With high school diploma Age Healthcare Practitioners and Technical Occupations (29) Employment income (dollars) Male Percent African American Percent Hispanic With advanced degree With bachelor’s degree With high school diploma Age Healthcare Support Occupations (31) Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age All Occupations Employment income (dollars) Male African American Hispanic With advanced degree With bachelor’s degree With high school diploma Age
Nontradable
Tradable
71,304 60.6 9.1 4.5 58.2 78.8 99.2 47.7
80,265 51.4 5.6 5.1 64.1 76.9 99.3 40.9
39,922 19.5 9.8 4.5 17.8 47.3 98.8 40.5
139,375 70.6 4.6 4.8 93.4 97.8 99.7 42.8
18,423 11.9 24.0 10.6 2.2 7.9 83.8 37.8
18,751 17.6 3.7 5.6 9.9 30.9 97.3 39.0
28,789 48.5 11.1 10.9 7.4 21.8 86.3 38.8
51,503 66.7 7.5 8.8 16.1 43.9 91.0 39.9
Table 11 examines whether the effects of being in a tradable industry and occupation are independent. Workers in tradable industries and tradable occupations are the omitted category. For all industries and occupations, workers in nontradable industries and nontradable occupations have 10 percent lower wages than workers in both tradable industries and occupations. Interestingly, the effect seems to be additive. Workers in either only a tradable industry or only
J. Bradford Jensen and Lori G. Kletzer
99
Table 9. OLS Regression Results, Tradable Industry Wage Differentialsa All industries Dependent variable: log (employment income) Tradable industry Male African American Hispanic Hours Weeks Advanced degree Bachelor’s degree Industry controls (2-digit NAICS) POWCMAb controls Summary statistics R2 N Weighted N
NAICS 50s
0.060 (0.0008) 0.214 (0.0006) –0.096 (0.0010) –0.215 (0.0010) 0.026 (0.0000) 0.040 (0.0000) 0.262 (0.0011) 0.380 (0.0008) Yes Yes
0.147 (0.0016) 0.225 (0.0014) –0.145 (0.0024) –0.218 (0.0026) 0.029 (0.0001) 0.039 (0.0001) 0.224 (0.0023) 0.325 (0.0017) Yes Yes
0.538 5,836,360 122,155,903
0.519 1,074,271 23,609,616
a. Standard error in parentheses. b. Place of Work Consolidated Metropolitan Area.
a tradable occupation receive wages about 5 percent lower than workers in both a tradable industry and a tradable occupation. In both professional service industries and “high-end” service occupations, the effect of being in a tradable industry and a tradable occupation is quite large. Workers in tradable industries and occupations in NAICS 50 sector receive wages 17 percent higher than workers in a nontradable industry and nontradable occupation within the same sector. For high-end service occupations, the differential is almost as large: workers in tradable industries and occupations make almost 16 percent more than workers in nontradable industries and occupations. These results demonstrate that tradable industries and occupations pay higher wages, even after controlling for observable characteristics. These effects appear to be independent: being in both a tradable industry and a tradable occupation is associated with a larger (almost double) income differential than being in either a tradable industry or occupation alone. The comparison of worker characteristics in tradable service activities suggests that tradable services are consistent with U.S. comparative advantage; they
100
Brookings Trade Forum: 2005
Table 10. OLS Regression Results, Tradable Occupation Wage Differentialsa All occupations Dependent variable: log (employment income) Tradable occupation
0.091 (0.0008) 0.215 (0.0006) –0.061 (0.0010) –0.187 (0.0010) 0.026 (0.0000) 0.039 (0.0000) 0.216 (0.0011) 0.303 (0.0008) Yes Yes
0.127 (0.0014) 0.245 (0.0013) –0.112 (0.0023) –0.168 (0.0027) 0.020 (0.0001) 0.038 (0.0001) 0.227 (0.0016) 0.297 (0.0013) Yes Yes
0.545 5,836,630 122,155,903
0.396 1,446,158 30,803,183
Male African American Hispanic Hours Weeks Advanced degree Bachelor’s degree Occupation controls (2-digit SOC) POWCMAc controls Summary statistics R2 N Weighted N
High-end service occupations b
a. Standard error in parentheses. b. High-end service occupations are occupations in SOC major groups 11, 13, 15, 17, 19, 23, and 29. c. Place of Work Consolidated Metropolitan Area.
are high-skill and high-wage activities (relative to both manufacturing and nontradable service activities).
Changes in Aggregate Employment Growth Much of the recent attention to services offshoring has emphasized job losses in specific occupational categories. We examine recent employment growth trends using both aggregate industry data from the Census Bureau’s County Business Patterns program and aggregate occupation data from the Bureau of Labor Statistics’ Occupational Employment Statistics program.18 We present the 18. The County Business Patterns program is an establishment-based data collection program that uses primarily administrative data and thus has nearly universal coverage of in-scope estab-
J. Bradford Jensen and Lori G. Kletzer
101
Table 11. OLS Regression Results, Tradable Industry and Occupation Wage Differentialsa All industries and occupations
NAICS 50s
High-end service occupations b
Dependent variable: Log (employment income) Nontradable industry and nontradable –0.098 –0.174 occupation (0.0011) (0.0026) Nontradable industry and tradable –0.055 –0.072 occupation (0.0012) (0.0026) Tradable industry and nontradable –0.055 –0.045 occupation (0.0010) (0.0022) Tradable industry and tradable occupation —Omitted category— Male 0.205 0.205 (0.0007) (0.0015) African American –0.064 –0.111 (0.0010) (0.0024) Hispanic –0.173 –0.169 (0.0010) (0.0026) Hours 0.025 0.027 (0.0000) (0.0001) Weeks 0.039 0.038 (0.0000) (0.0001) Advanced degree 0.223 0.197 (0.0011) (0.0024) Bachelor’s degree 0.279 0.245 (0.0008) (0.0017) Industry controls (2-digit NAICS) Yes Yes Occupation controls (2-digit SOC) Yes Yes POWCMAc controls Yes Yes Summary statistics R2 N Weighted N
0.545 5,836,630 122,155,903
0.540 1,074,271 23,609,616
–0.159 (0.0022) –0.050 (0.0019) –0.087 (0.0021) 0.244 (0.0013) –0.111 (0.0022) –0.158 (0.0026) 0.020 (0.0001) 0.036 (0.0001) 0.232 (0.0016) 0.276 (0.0013) Yes Yes Yes 0.419 1,446,158 30,803,183
a. Standard error in parentheses. b. High-end service occupations are occupations in SOC major groups 11, 13, 15, 17, 19, 23, and 29. c. Place of Work Consolidated Metropolitan Area.
data broken out as tradable/nontradable and by sector. The results in the previous section indicate that tradable activities in general and tradable services in particular require higher skills than other activities. High-skill activities are consistent with U.S. comparative advantage, and we would expect that as trade
lishments. For more information on County Business Patterns see www.census.gov/epcd/cbp/ view/cbpview.html. The Occupational Employment Statistics program is also an establishmentbased program, but it is collected through a survey instrument. For more information on the Occupational Employment Statistics see www.bls.gov/oes/home.htm.
102
Brookings Trade Forum: 2005
Figure 2. Industry Employment Growth, 1998–2002 0.8
Change in Log(Industry Employment)
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8 0
100
200
300
400
500
600
700
800
900
NAICS Industry Ag
Mining
Utilities
Construction
Manufacturing
Wholesale
Retail
Transportation
Services
increases, economic activity would shift to activities consistent with U.S. comparative advantage. Thus, we would expect higher-skill industries and occupations to have higher rates of employment growth. We also break out the employment growth rates by industry and occupation skill quartile.19 Figure 2 shows the change in industry employment (log) for the period 1998–2002 by NAICS code.20 Overall, employment in manufacturing industries shrank, and employment in service industries grew. Table 12 presents mean industry employment growth by tradable and nontradable sectors. In the aggregate, the mean tradable industry experienced an employment loss of almost 6 percent, while the mean nontradable industry experienced an employment gain of 5.6 percent. The lower panels of table 12 break out industries by sector, tradable category, and skill quartile. The lower panels of table 12 show that the 19. Industry and occupation skill quartiles are created by placing industries and occupations into skill quartiles based on the share of employees within the industry with a bachelor’s degree. 20. We are constrained to use 1998 as our starting point because it is the first year that County Business Patterns was produced on a NAICS basis; 2002 is the most recent year available. Public Administration is not in scope for the County Business Patterns program, so employment change figures are not available for this sector.
J. Bradford Jensen and Lori G. Kletzer
103
Table 12. Industry-Level Employment Change, by Industry Characteristics, 1998–2002 Industry classification Nontradable Tradable Ag, Min, Mfga Services Ag, Min, Mfg
Tradable v. nontradable
Nontradable Tradable Nontradable Tradable Nontradable Tradable
Services
Nontradable
Tradable
Skill quartile
Number of industries
Mean
Standard deviation
Skill Q1 Skill Q2 Skill Q1 Skill Q2 Skill Q3 Skill Q4 Skill Q1 Skill Q2 Skill Q3 Skill Q4 Skill Q1 Skill Q2 Skill Q3 Skill Q4
88 149 5 83 91 85 3 2 32 24 16 11 24 23 20 24 7 16 31 31
0.056 –0.059 –0.116 –0.173 0.067 0.076 –0.067 –0.190 –0.191 –0.203 –0.114 –0.147 0.016 0.084 0.015 0.156 –0.006 0.112 –0.007 0.139
0.114 0.198 0.099 0.161 0.107 0.145 0.102 0.015 0.169 0.148 0.103 0.216 0.080 0.098 0.106 0.088 0.233 0.104 0.095 0.148
a. Agriculture, Mining, Manufacturing.
employment losses are, on average, concentrated in the goods-producing sector (and in the lower portion of the skills distribution).21 In the service sector, the average nontradable industry experienced 6.7 percent growth, and the average tradable service industry experienced 7.6 percent growth. In general, industries in the lower-skill quartiles have a lower rate of employment growth. Tradable industries do not seem to have dramatically different employment outcomes than nontradable industries, though at the low end of the skill distribution tradable industries had, on average, employment losses.22
21. These results are consistent with Bernard, Jensen, and Schott (forthcoming 2006). Bernard, Jensen, and Schott use detailed, plant-level data to examine the impact of imports from low-wage countries on U.S. manufacturing. The results show that activity in U.S. manufacturing is shifting to industries consistent with U.S. comparative advantage. 22. Using a t test to compare the lowest-skill quartile with the highest-skill quartile in the tradable services industry group, we cannot reject the null hypothesis that the means are the same at the 10 percent level.
104
Brookings Trade Forum: 2005
Table 13. Occupation-Level Employment Change, by Occupation Characteristics, 1999–2003 Occupation classification Nontradable Tradable Ag, Prod, Ext, Cona Services Ag, Prod, Ext, Con
Tradable v. nontradable
Nontradable Tradable Nontradable Tradable Nontradable
Tradable
Services
Nontradable
Tradable
Skill quartile
Number of industries
Mean
Standard deviation
Skill Q1 Skill Q2 Skill Q3 Skill Q1 Skill Q2 Skill Q3 Skill Q1 Skill Q2 Skill Q3 Skill Q4 Skill Q1 Skill Q2 Skill Q3 Skill Q4
197 228 38 77 180 180 23 12 3 56 18 3 30 57 54 39 10 32 59 79
0.022 –0.004 –0.044 –0.141 0.036 0.059 –0.070 –0.026 0.056 –0.148 –0.150 0.014 0.005 0.037 0.021 0.078 –0.065 0.086 0.032 0.083
0.160 0.247 0.143 0.228 0.161 0.230 0.145 0.140 0.125 0.235 0.196 0.272 0.114 0.173 0.165 0.164 0.111 0.210 0.181 0.269
a. Agricultural, Production, Extractive, Construction b. Skill Q is Skill Quartile
Table 13 shows similar employment growth rates for 1999–2003 for occupation categories.23 Similar to industries, tradable occupations in aggregate have lower employment growth rates than nontradable industries on average. Also similar to industries, this is explained primarily by differences between production-related occupations and service activities. Tradable service occupations have, on average, higher employment growth rates than nontradable service occupations. It is interesting to note that, as in tradable industries, at the low end of the skill distribution tradable service occupations have negative employment growth. In comparison, the highest skill category has positive employment growth.24
23. We use 1999 as our starting year because it is the first year the Occupational Employment Survey was published on a Standard Occupational Classification basis. We use 2003 as the end point to have a four-year period consistent with the industry data. 24. Using a t test to compare the lowest-skill quartile with the highest-skill quartile in the tradable services occupation group, we can reject the null hypothesis that the means are the same.
J. Bradford Jensen and Lori G. Kletzer
105
The employment growth results are consistent with the comparative advantage framework. Employment is shifting toward activities that are consistent with U.S. comparative advantage. Industries and occupations that require higher skills are growing relative to low skill industries and occupations. In both tradable service industries and occupations, those in the lowest skill classes experience negative employment growth on average.
Evidence on the Risk of Job Loss and Characteristics of Displaced Workers The Displaced Worker Surveys (DWS) provide basic information on the scope and cost of involuntary job loss. The DWSs offer large sample sizes, are nationally representative, and allow several key elements to be investigated, including the incidence of job loss; the characteristics of workers affected; likelihood of reemployment; reemployment industry and occupation; and earnings changes.25 These surveys have been used extensively to study manufacturing job loss (see Kletzer 2001). The 2000 census provides the most up-to-date industry and occupational classifications of the services and white-collar jobs of primary interest. The need for updated detail on industry and occupation (currently) limits our use of the Displaced Worker Surveys to the most recent administration, in January 2004. Although we lose the ability to observe services and white-collar job loss over time, we gain the industry and occupational detail necessary for studying services offshoring. Job displacement from services Job loss rates by industry are reported in table 14, focusing on the 2001–03 period covered by the January 2004 Displaced Worker Survey. Remembering that this time period covered the dot.com bust and the most recent recession, the Information sector (NAICS 51) had a notably high rate of job loss (.232). Overall, the risk of job loss was lower in services than in manufacturing. As a reference point, table 14 includes job loss rates by industry for the period 1999–2001, from the 2002 Displaced Worker Survey. The industry classifications are different, reflecting the use of 1990 census codes for the 2002 survey. What is clear is that job loss rates increased from 1999–2001 to 2001–03, most 25. See the appendix for more information on the Displaced Worker Surveys.
Table 14. Job Loss Rates, by Industry Mean From the 2004 Displaced Worker Survey (2001–03) Industry Agriculture Mining Construction Manufacturing Wholesale and retail trade Transport and utilities Information Financial Professional and business services Education and health services Leisure and hospitality Other services Public administration Total Manufacturing, tradable Manufacturing, nontradable Nonmanufacturing, tradable Nonmanufacturing, nontradable
Total 2001–03 0.049 0.127 0.131 0.209 0.113 0.089 0.232 0.081 0.144 0.040 0.105 0.051 0.020 0.103
From the 2002 and 2004 Displaced Worker Surveys
Tradable
Nontradable
… … … … 0.077
… … … … 0.091
0.317 0.08 0.158 0.071 0.083 0.03
0.075 0.081 0.113 0.039 0.113 0.057
0.153
0.076
0.213 0.192 0.128 0.073
Dropping agriculture, mining, and construction Manufacturing, tradable 0.213 Manufacturing, nontradable 0.192 Nonmanufacturing, tradable 0.106 Nonmanufacturing, nontradable 0.054 Total
Industry
1999– 2001
2001– 03
Agriculture Mining Construction Manufacturing, durables Manufacturing, nondurables Transportation Communications Utilities and sanitary service Wholesale trade Retail trade Finance,insurance, and real estate Private household Business and repair services Personal services Entertainment and recreation Hospitals Other medical Educational services Social services Other professional services Forestry and fisheries Public administration
0.042 0.173 0.107 0.177 0.133 0.096 0.159 0.054 0.111 0.099 0.079 0.044 0.181 0.080 0.071 0.026 0.052 0.020 0.033 0.071 0.008 0.017
0.065 0.127 0.131 0.236 0.157 0.103 0.305 0.052 0.123 0.107 0.080 0.016 0.172 0.057 0.098 0.030 0.055 0.030 0.060 0.078 0.070 0.020
0.090
0.106
Total
0.126
0.058
Source: Authors’ calculations from the 2002 and 2004 Displaced Worker Surveys, using sampling weights.
J. Bradford Jensen and Lori G. Kletzer
107
notably in Communications (the former name of the sector for some of Information) and Manufacturing. When we apply our tradable/nontradable distinction to the overall economy, the rate of job loss is notably higher in tradable industries (.153) than in nontradable industries (.076). Within the broad sectors of manufacturing and nonmanufacturing, tradable industries also had higher rates of job loss. The tradable/nontradable distinction is small in manufacturing, with tradable industries having a job loss rate of .213, and nontradable (of which there are few) a rate of .192. Outside of manufacturing, the tradable distinction is large. Tradable nonmanufacturing industries have a rate of job loss of .128, and nontradable industries, .073. This difference is most notable in the Information sector, where the rate of job loss from tradable (3-digit) industries was .317 and the nontradable job loss rate was .075. Job loss rates by occupation are reported in table 15. The blue-collar occupations faced a higher rate of job loss (about .12) than the white-collar occupations (about .09). Workers in all occupational categories faced a higher rate of job loss in 2001–03 than in 1999–2001. Production workers faced the highest rate of job loss, .206 (the cross-occupation average was .106). Some of the white-collar occupational categories forecast to be at risk of services offshoring had high job loss rates (but lower than Production workers), including Business Operations Specialists (.143), Computer and Math (.177), and Architecture and Engineering (.128). In the overall economy, tradable occupations had a higher rate of job loss than nontradable occupations, with the greatest difference in white-collar occupations. White-collar workers in tradable occupations faced a job loss rate of .094, and workers in nontradable occupations faced a rate of .065. For blue-collar workers, the tradable job loss rate was .128 and the nontradable rate was .122. There is no clear pattern of exposure to the risk of job loss by tradability within detailed occupations. Parallel to our discussion of worker characteristics from the 2000 PUMS, table 16 reports demographic and educational characteristics for workers displaced from tradable and nontradable nonmanufacturing industries, with (tradable) manufacturing industries offered as a reference group. As noted by Kletzer (2001), workers displaced from nonmanufacturing industries are slightly younger, less tenured, less likely to be male, and considerably more educated than workers displaced from manufacturing. In tradable nonmanufacturing, 75 percent of displaced workers had at least some college experience. In manufacturing, the share of displaced workers with some college was 46 percent.
Table 15. Job Loss Rates, by Occupationa Mean From the 2004 Displaced Worker Survey (2001–03) Industry Management business, financial (white collar) Business operations specialists Financial specialists Professional and related (white collar) Computer and math Architecture and engineering Life, physical, and social science Service (white collar) Sales (white collar) Office and administrative support (white collar) Farming, forestry, fishery (blue collar)
From the 2002 and 2004 Displaced Worker Surveys Industry
1999– 2001
2001– 03
0.091 0.171 0.044
Executive, administrative, managerial Professional specialty Technician and related
0.086 0.059 0.088
0.094 0.066 0.110
0.109 0.177 0.113 0.057 0.072 0.123
0.033 n.a. 0.158 0.066 0.056 0.079
Sales Administrative support Private household Protective services Food, health, cleaning, personal Precision production, craft, repair
0.094 0.097 0.047 0.045 0.069 0.111
0.109 0.106
0.109
0.067
0.092
0.181
0.219
0.110
…
…
Operators, assemblers, inspectors Transportation and material moving equipment
0.103
0.112
Total 2001–03
Tradable
Nontradable
0.089 0.143 0.054
0.077 0.121 0.057
0.070 0.177 0.128 0.059 0.073 0.106
0.059 0.075 0.151
Construction and extractive (blue collar) Installation, maintenance, repair (blue collar) Production (blue collar) Transportation and material moving (blue collar)
0.149
…
…
0.112 0.206
0.117 0.163
0.083 0.169
0.117
0.057
0.096
0.102
0.101
0.078
Blue collar, tradable Blue collar, nontradable White collar, tradable White collar, nontradable
0.128 0.122 0.094 0.065
… … … …
… … … …
Full sample Blue collar, tradable Blue collar, nontradable White collar, tradable White collar, nontradable Full sample total
0.175 0.150 0.104 0.078 …
… … … … 0.122
… … … … 0.087
Total
Source: Authors’ calculations from the 2002 and 2004 Displaced Worker Surveys. a. Agriculture, Mining, and Construction omitted. n.a. Not available.
Handlers, cleaners, helpers Farming, forestry, fishery Armed forces Total
0.139 0.044
0.151 0.067
0.090
0.103
110
Brookings Trade Forum: 2005
Table 16 also shows that in tradable nonmanufacturing industries, displaced workers were more educated, more likely to have health insurance, more likely to lose full-time jobs, and more likely to have higher predisplacement earnings than workers displaced from nontradable industries. The educational attainment differences are stark: 42 percent of workers displaced from nontradable nonmanufacturing industries, but 24 percent of workers displaced from tradable nonmanufacturing industries, had a high school diploma or less. The educational differences show up in predisplacement weekly earnings. Postdisplacement, reemployment rates (also reported in table 16) are higher for displaced nonmanufacturing workers than for manufacturing workers. Reemployment rates are .75 and .77 for nontradable and tradable nonmanufacturing workers, respectively, .64 for manufacturing workers. The earnings cost of job displacement, well established for manufacturing workers, also affected nonmanufacturing workers. For the 2001–03 period, with the weak job recovery from the recession, we see large earnings losses. Median earnings losses are smaller for nonmanufacturing than for manufacturing, and a larger share of nonmanufacturing workers experience no earnings loss. Consistent with lower predisplacement earnings, workers displaced from nontradable nonmanufacturing industries experienced smaller earnings losses than workers displaced from tradable nonmanufacturing industries. Table 17 reports worker characteristics and reemployment outcomes for three services sectors: Information; Financial, Insurance and Real Estate; and Professional and Business Services. For the most part, workers in tradable industries in these sectors have higher levels of educational attainment. In Information and Professional and Business Services, predisplacement weekly earnings were higher in tradable industries than in nontradable industries. Consistent with higher earnings, more workers displaced from tradable industries reported that they had health insurance coverage than workers displaced from nontradable industries. Reemployment outcomes (reemployment rates or average earnings losses) are similar within sector, across the tradability of the detailed industries. Table 18 reports a similar breakdown, by occupation, for sectors: Management, Business and Financial; Professional and Related; Office and Administrative Support. Workers from tradable occupations have higher levels of education, within occupational group, than workers from nontradable occupations. Their predisplacement earnings were higher, as was the availability of health insurance coverage. Men are more highly represented in the tradable occupations. Again, there is no clear pattern of reemployment outcomes by tradability. Earnings losses range from 3 percent to 16 percent, with 40 to 50 percent of reemployed workers reporting no earnings loss.
J. Bradford Jensen and Lori G. Kletzer
111
Table 16. Characteristics of Displaced Workers, by Industrial Sector and Tradability Worker characteristics Age (mean in years) Standard deviation Job tenure (mean in years) Standard deviation Job tenure > ten years
Manufacturing, tradable
Nonmanufacturing, Nonmanufacturing, tradable nontradable
41.60 11.20 7.11 8.43 0.23
39.60 11.10 4.40 5.60 0.12
38.10 11.70 4.26 5.61 0.14
0.14 0.40 0.24 0.22 0.61
0.05 0.19 0.30 0.45 0.54
0.11 0.31 0.33 0.25 0.45
0.75 0.96
0.66 0.90
0.47 0.82
342.70 300.54
443.18 383.08
294.91 271.21
Share reemployed Of reemployed, share full-time
0.64 0.80
0.77 0.78
0.75 0.72
All reemployed Change in ln earnings (mean) Standard deviation Median change Share with no loss in earnings
–0.32 0.89 –0.15 0.42
–0.30 0.98 –0.11 0.45
–0.14 1.02 –0.03 0.51
Full-time to full-time Change in ln earnings (mean) Standard deviation Median change Share with no loss in earnings
–0.21 0.76 –0.10 0.42
–0.21 0.69 –0.07 0.46
–0.12 0.97 –0.03 0.52
Educational attainment (share) High school dropout High school graduate Some college College + Male In predisplacement job Share with health insurance Full-time If full-time, real weekly earnings (dollars) Standard deviation (dollars)
Source: Authors’ calculations from the 2004 Displaced Worker Survey, using sampling weights. Agriculture, Mining, and Construction omitted.
Conclusions This paper develops a new empirical approach to identifying, at a detailed level for the entire economy, industries and occupations that are tradable. Using the methodology, we find substantial employment in tradable service industries and occupations. Workers in these industries and occupations are more highly skilled and have higher earnings than workers in the manufacturing sector and nontradable service activities. The higher earnings are not solely a result of higher skill levels: in regressions controlling for observable characteristics,
Table 17. Characteristics of Selected Service Sector Displaced Workers, by Industry and Tradability Information Tradable Job tenure (mean in years) Standard deviation Job tenure > ten years Educational attainment (share) High school dropout High school graduate Some college College + Male In predisplacement job Share with health insurance Full-time If full-time, real weekly earnings (dollars) Standard deviation (dollars) Share reemployed Of reemployed, share full-time All reemployed Change in ln earnings (mean) Standard deviation Median change Share with no loss in earnings Full-time to full-time Change in ln earnings (mean) Standard deviation Median change Share with no loss in earnings
Nontradable
Financial, insurance, real estate Tradable
Nontradable
Professional and business services Tradable
Nontradable
5.80 7.37 0.192
4.51 7.25 0.16
5.82 7.00 0.167
8.28 9.14 0.259
3.55 3.98 0.066
3.24 4.68 0.109
0.032 0.207 0.262 0.499 0.559
0.00 0.038 0.45 0.512 0.668
0.04 0.179 0.389 0.392 0.47
0.046 0.243 0.354 0.357 0.479
0.047 0.157 0.261 0.535 0.527
0.173 0.446 0.196 0.186 0.527
0.82 0.93 530.82 409.45 0.72 0.76
0.62 0.87 387.98 350.69 0.81 0.87
0.62 0.91 409.88 380.43 0.61 0.80
0.73 0.94 542.51 454.14 0.68 0.82
0.66 0.91 504.61 415.82 0.71 0.80
0.36 0.83 273.95 251.57 0.62 0.73
–0.57 1.07 –0.34 0.346
–0.72 2.97 –0.024 0.469
–0.16 1.09 –0.08 0.456
0.013 0.499 0.03 0.531
–0.34 0.96 –0.08 0.457
–0.18 0.93 –0.03 0.468
–0.40 0.82 –0.25 0.36
–1.003 3.328 –0.07 0.344
–0.15 0.51 –0.047 0.457
0.018 0.36 –0.007 0.508
–0.185 0.737 –0.034 0.49
–0.162 0.999 –0.029 0.489
Source: Authors’ calculations from the 2004 Displaced Worker Survey, using sampling weights.
Table 18. Characteristics of Displaced Workers in Selected Service Occupations, by Occupation and Tradability Management, business, and financial Worker characteristics Job tenure (mean in years) Standard deviation Job tenure > ten years Educational attainment (share) High school dropout High school graduate Some college College + Male In pre-displacement job Share with health insurance Full-time If full-time, real weekly earnings (dollars) Standard deviation (dollars) Share reemployed Of reemployed, share full-time All reemployed Change in ln earnings (mean) Standard deviation Median change Share with no loss in earnings Full-time to full-time Change in 1n earnings (mean) Standard deviation Median change Share with no loss in earnings
Tradable
Nontradable
Professional and related Tradable
Nontradable
Office and administrative support Tradable
Nontradable
6.72 8.04 0.204
5.03 4.99 0.143
4.82 6.09 0.111
4.30 5.25 0.109
5.31 6.69 0.176
4.57 5.74 0.136
0.008 0.132 0.269 0.591 0.466
0.012 0.272 0.28 0.436 0.633
0.003 0.092 0.198 0.708 0.717
0.026 0.115 0.328 0.53 0.248
0.051 0.331 0.438 0.18 0.306
0.05 0.339 0.406 0.204 0.241
0.775 0.965 554.78 434.23 0.786 0.791
0.588 0.927 426.02 336.05 0.72 0.726
0.794 0.93 523.24 369.44 0.80 0.805
0.632 0.791 323.60 226.58 0.801 0.707
0.616 0.896 299.45 254.48 0.691 0.758
0.577 0.865 261.96 198.07 0.755 0.763
–0.374 1.08 –0.127 0.492
–0.364 1.144 –0.165 0.389
–0.34 1.155 –0.084 0.455
–0.14 0.811 –0.037 0.507
–0.227 0.677 –0.15 0.443
–0.093 1.063 –0.045 0.512
–0.205 0.852 –0.045 0.528
–0.357 1.165 –0.109 0.351
–0.318 1.176 –0.068 0.462
–0.128 0.343 –0.029 0.515
–0.113 0.455 –0.068 0.471
0.012 0.704 –0.025 0.542
Source: Authors’ calculations from the 2004 Displaced Worker Survey, using sampling weights.
114
Brookings Trade Forum: 2005
workers in selected tradable service activities earn 16–17 percent higher incomes than similar workers in nontradable activities in the same sector. Examining employment growth across industries and occupations, there is little evidence that tradable service industries or occupations grow more slowly than nontradable industries or occupations overall, though at the low end of the skill distribution employment growth is negative for tradable services. Highskill service activities have the highest employment growth rates. There is job insecurity associated with employment in tradable activities, including service activities. We find a higher rate of job loss from tradable industries than from nontradable industries, with the greatest difference outside of manufacturing. In comparison with an overall rate of job loss of .103 for 2001–03, tradable nonmanufacturing industries have a rate of job loss of .128 and nontradable industries .073 (though we note the possibility that these differences are driven by the tech bubble). Also within occupations, workers in tradable jobs faced a higher rate of job loss than workers in nontradable jobs, with the greatest difference within white-collar occupations. These results have several implications. First, it seems inappropriate to consider all service activities as inherently nontradable. The geographic concentration of some service activities within the United States is as great as in manufacturing and is consistent with the view that a number of service industries and occupations are tradable. The share of employment in tradable services is large enough that a better understanding of the forces shaping trade in services warrants our attention. At a minimum, more resources should be devoted to collecting and publishing considerably more detail on international service flows. Continuing to increase the amount of information collected on the use of intermediate service inputs within the United States would also increase our ability to track and understand developments in this large and growing sector. Second, the results presented in this paper suggest that tradable services are consistent with U.S. comparative advantage. While professional and business services jobs require higher skills and pay higher wages than manufacturing jobs in general, tradable services jobs in these sectors require even higher skills and are more highly paid than nontradable service activities. We would expect that as technological and organizational change increases the potential for trade in services, economic activity in the United States will shift to activities consistent with U.S. comparative advantage.26 It is therefore possible that further liberalization in international services trade would directly benefit workers and firms in
26. The United States maintains a positive trade balance in service activities; see table 1.
J. Bradford Jensen and Lori G. Kletzer
115
the United States. The policy community should devote more attention to understanding the impediments to services trade. Third, although tradable services have relatively high employment growth rates overall, at the low end of the skill distribution tradable service activities have negative employment growth. The potential for reallocation across activities in response to shifting trade patterns in services is real. Policymakers should prepare for additional reallocation among this group of workers. The process of adjustment to job displacement might be eased by service worker characteristics. For the most part, workers displaced from tradable services are different, in terms of job tenure and educational attainment, from workers displaced from (tradable) manufacturing industries. Generalizing from what we know from studies of manufacturing worker job loss, lower levels of job tenure and higher levels of educational attainment may be advantages in seeking reemployment. Given the current availability of data, it is too early to tell. We need data beyond the time period of the “jobless recovery.” We also need more information to discern whether workers in tradable activities face different reemployment outcomes than workers in nontradable activities. The evidence we do have tells us that job loss for services workers is costly. These costs underscore the need to have a less porous safety net (for example, by extending Trade Adjustment Assistance [TAA] to services workers and extending wage insurance beyond TAA). Lower rates of employment growth at the lower end of the skill distribution in tradable service activities may have implications for the retraining strategies and opportunities for displaced low-skill workers in both manufacturing and services.
Appendix: Displaced Worker Survey The Displaced Worker Survey is administered biennially as a supplement to the Current Population Survey (CPS). The first survey was administered in January 1984 and the most recent in January 2004. In each survey, adults (aged 20 years and older) in the regular monthly CPS were asked if they had lost a job in the preceding three- or five-year period due to “a plant closing, an employer going out of business, a layoff from which he/she was not recalled, or other similar reasons.”27 If the answer was yes, a series of questions followed concerning 27. For the 1984–92 surveys, the recall period was five years. Starting in 1994, the recall period was shortened to three years.
116
Brookings Trade Forum: 2005
the lost job and the period of joblessness. Other causes of job loss, such as quitting and firing, are not considered displacements.28 This categorization is consistent with our common understanding of job displacement: it occurs without personal prejudice in that terminations are related to the operating decisions of the employer and are independent of individual job performance. This operational definition is not without ambiguity: the displacements are “job” displacements, in the sense that an individual displaced from a job and rehired into a different job with the same employer is considered displaced. A key advantage of the DWS is its large-scale representative nature. As part of the CPS, it draws on a random sample of 60,000 households, which is weighted to be representative of the U.S. workforce. As a result, the surveys yield responses from large numbers of displaced workers in a wide set of industries. In exchange for breadth of coverage, the DWSs have two weaknesses relevant to any study of the costs of job loss. The first is the relatively short-term horizon. Individuals are surveyed just once, providing information about one postdisplacement point in time, rather than about their experiences over time. The second weakness is the lack of a readily available comparison group of nondisplaced workers. Without such a comparison group, we cannot investigate what would have happened to these workers if they had not been displaced. The lack of a comparison group leads to some unavoidable errors in measuring outcomes such as postdisplacement reemployment and earnings losses. The rate of job loss reported in the tables is calculated as in Farber (1993, 2003, 2005): it is the ratio of the (weighted) number of reported displacements divided by the (weighted) number of workers who were either employed at the survey date or reported a job loss but were not employed at the survey date. See Kletzer (2001) for more discussion of the issues that arise when using the DWSs to measure the incidence of job loss.
28. Individuals who respond that their job loss was due to the end of a seasonal job or the failure of a self-employed business are also not included.
Comments and Discussion
Jared Bernstein: Jensen and Kletzer have written a refreshingly clear and insightful paper that readers will find to be one of more useful contributions to the often fuzzy literature on offshoring. Much of this work has tried to identify the service or white-collar jobs at risk to offshore competition, but we have been stymied by the difficulty of using trade data on service flows for this purpose. These authors derive a clever method using geographical clustering for doing so, and while they may need to work a bit harder to convince skeptics, many will find their approach convincing, as I do. This innovative classification scheme sets the stage for the paper’s other main contribution: a description of the characteristics and earnings of those in tradable services relative to those in nontradable services. One criticism of the paper is that the title promises more than the authors, or anyone else for that matter, can yet deliver. That is, while they go further than others toward identifying the industries and occupations directly affected by offshoring, to truly capture the “scope and impact” of this growing competitive challenge, researchers need to go beyond the direct effects. The authors do point out that displaced workers in tradable services suffer large wage losses relative to other displaced workers, but (a) it is not clear that this is because they are in tradable services, and (b) surely the impact of offshoring goes beyond this subgroup. This latter point is critical. The implicit supply shock from adding millions of skilled workers to a relatively concentrated set of occupations and industries may have a significant negative impact on the wage structure of white-collar workers, much as the increase of trade in manufacturing goods with low-laborcost competitors has structurally altered the wage distribution of blue-collar 117
118
Brookings Trade Forum: 2005
workers. In short, a white-collar worker needn’t get displaced to feel the impact of this growing phenomenon.
Using Geographic Clustering to Identify Tradable Services The credibility of their paper rests on the authors’ novel method for identifying tradable services. They point out that BEA data on international trade flows in services are not disaggregated enough by industry to serve this purpose. But the problem goes deeper than this. As my EPI colleague Josh Bivens points out, these data, especially the highly relevant parts relating to information technology, are getting a bit hard to believe, given what so many firms are telling us about their service imports and what some other countries’ service export data suggest. Take, for example, data on the value of imports of computer-related services, which includes software writing, from India. Even with recent large upward revisions, the tiny magnitudes of the BEA numbers—for example, $330 million in 2003— are hard to believe. The Indian tech trade group NASSCOM puts this value at $4.7 billion. This is not to suggest that NASSCOM’s data capacity is superior to BEA’s. Rather, if you’re out to identify service jobs affected by offshoring, most analysts are suspicious of the quality of our data on the import of some key services associated with offshoring. At any rate, Jensen and Kletzer use the assumption that tradable firms exhibit geographic concentration. This assumption comes from research on the goods sector, where returns to scale, access to transportation nodes, and proximity to natural resources lead goods producers to congregate near each other. Is it reasonable to extend this to service production? Empirically, we can, without much effort, observe this concentration, or lack thereof. Silicon valleys and “research triangles” have appeared in numerous places over the past decades. Meanwhile, bowling alleys and child-care centers are scattered pretty much all over the place. In this regard, their transporting of this method of identifying tradable industries from goods to services does not seem a big stretch. There are, however, some differences between goods and services that will lead some readers to wonder if scale economies and access issues loom large enough in services to motivate geographic clustering. For example, to transport cars or steel, manufactures have historically needed to locate near waterways. But it is hard to see why this constraint would hold for, to take a very relevant case, transmitting information across the Internet. In fact, it is the sharp decline
J. Bradford Jensen and Lori G. Kletzer
119
in such costs that has allegedly motivated service firms to offshore data to extremely distant places. So they may need to work a little harder to convince skeptics. What are the specific benefits they have in mind that motivate tradable services to locate near each other? Are there some case studies they could cite? As mentioned, it is not hard to point to areas where high-tech firms are concentrated, but there could be lots of reasons for that, including niche education and labor markets: California’s Silicon Valley and North Carolina’s Research Triangle, for example, are both near universities with specialties in computer science. And where I live, in northern Virginia, our silicon alley, out Route 66 in the Dulles corridor, likely grew out of the desire to be close to federal government contractors and purchasers. What is the connection to international trade? And why shouldn’t nongeographically clustered service industries offshore some of their jobs? Hospitals, for example, score in the authors’ least geographically concentrated category, presumably because they are pretty pervasive across localities in our economy. But anecdotes suggest that hospitals are beginning to offshore some of their accounting services, certainly a plausible scenario (anecdotes also suggest hospitals are offshoring high-tech functions, like radiology services, but as the conference paper by Frank Levy and Ari Goelman (this volume) finds, this does not appear to be occurring).1 While I encourage them to work a little harder to convince the reader that their classification scheme is up to the task, a close look at their tables and figures reveals strong face validity. There are a few industries, such as hospitals, that seem questionably classified as nontradable (accounting, tax preparation, bookkeeping, and payroll services is another), but no such system will be perfect. In the case of the two examples I just mentioned, they are services that by their nature tend to be demanded in most localities and thus fly under the radar of their test. So perhaps Jensen and Kletzer can think of an added filter that would help address such industries.2 They presumably pick up some of these jobs in their occupational analysis. Their table 8, for example, shows that 11 percent of total employment is in tradable occupations in nontradable industries. Still, the apparent misclassification of a few industries may unsettle some readers.
1. Levy and Goelman show that both gatekeeper actions by U.S. radiologists and malpractice regulations explain why hospitals are hard-pressed to offshore such services. 2. I doubt anyone would squawk if they just added a few industries like hospitals and tax preparation services that are widely reported to be tradable services, even though they are not geographically concentrated.
120
Brookings Trade Forum: 2005
They are careful to avoid the following mistake: suppose a nontradable upstream service provides an intermediate service to a tradable downstream service industry. If the upstream firms need to locate near the downstream firms, they will be misclassified as tradable services. For example, if a computer firm both offshores programming tasks to India and outsources payroll services to a nearby firm, the authors could end up mistakenly labeling the upstream industry as a tradable service. To avoid this, they use input/output tables to parse the upstream services from the downstream ones. A final concern is in regard to the role of productivity growth in their method of using workers to identify where firms are clustered. If demand is constant, falling, or not growing too quickly, as was arguably the case over their period of study, firms with fast-growing productivity might be shedding workers. The impact of this on their analysis is not necessarily problematic, as long as the firms in such industries remain clustered (and it is hard to see why they would not). But this may be one reason why this type of analysis is usually based on more direct measures of industry output (one reason they are sticking with workers is because they want to examine occupations as well as industries).
Comparing the Characteristics of Workers in Tradable and Nontradable Jobs As one might have expected, given the anecdotes in the newspapers, jobs in tradable services pay more than those in nontradable services: a 35 percent annual earnings differential in tradable services, unadjusted for worker differences, and a large adjusted differential, discussed next. Such workers are also more likely to be male and have higher educational attainment. With a set of earnings regressions, the authors find a statistically and economically large premium associated with being in a tradable industry, a tradable occupation, and a combination of the two (in their later analysis on displacement, we see the downside of this—workers displaced from such jobs experience large relative losses). Relative to those in nontradable industries and occupations, the premium amounts to between 10 and 17 percent, depending on the sample. What is interesting here is that the impact of being in such industries and occupations is modeled as a sort of interaction, as the regressions already control for industries and occupations. The coefficient of interest thus tests whether an earnings premium exists above that already accounted for by the underlying industries or occupations that are also included in the tradable services indicator.
J. Bradford Jensen and Lori G. Kletzer
121
Such interactions are difficult to interpret. The descriptive statistics reveal that workers in tradable services have characteristics that by themselves are all associated with positive and significant coefficients in such regressions: they are disproportionately male, nonminority, and have higher educational attainment. Combine these characteristics and you get a fairly hefty wage boost beyond that accounted for by any one characteristic alone. Are such workers truly more productive, or are there other factors, such as bargaining power and discrimination, that might explain their premium relative to those who lack this set of characteristics? The result is also curious in relation to the tradable service categorization. One might expect that the wages of such workers face downward pressure from international competition relative to the wages of other workers with similar skill sets in nontradable industries and occupations. At least in these static regressions, that is not the case. It will be interesting to track the premium over time to see if this pressure develops. At any rate, the important point is that service workers exposed to trade competition have a lot to lose. The last section helps to quantify that point. This part of the paper includes two tables on changes in employment levels by industry and occupation. The goal here is to determine the extent to which job losses have occurred in recent years in tradable services, a question that is a bit of a holy grail, given the nervousness regarding the impact of offshoring services. As such, I thought the section got short shrift. This part of the analysis would have benefited from more discussion of the data and trying a little harder to separate out cyclical effects. On the first point, their sources for employment data are the Census Bureau’s County Business Patterns and the BLS Occupational Employment Statistics (OES). Neither of these sources is typically used to track aggregate employment changes, and readers will legitimately wonder whether they reflect the stylized facts of employment trends over the years in question (1998–2003). In fact, given the difference in employment trends between the two surveys that are universally used for such analysis—the BLS Establishment and Household surveys—some will question whether the facts are “stylized” at all. I took a cursory look at the total OES employment counts from 2000 through 2003, which seem to show a large growth of jobs over these years, which is hard to square with data from more reliable sources of aggregate employment growth (such as the Establishment survey). Also, one of the biggest challenges regarding the question of the impact of offshoring on job loss over recent years is separating an offshoring effect from that of the cycle. This is particularly tough given the burst of the IT bubble in late
122
Brookings Trade Forum: 2005
2000 and the resulting spike in layoffs in this sector. In table 12, the authors examine changes from 1998–2002, a period including a strong run-up in employment growth (1998–2000) and a recession (2001) and jobless recovery (2002). At the least, the authors might consider breaking out these two periods to add some accounting for these cyclical effects. Better yet, given the caveats regarding these data sets for this purpose and the difficulty untangling cycle from offshoring effects, they might want to be more cautious about their claims here. For example, claims comparing the employment growth of tradable and nontradable services made it into their abstract and could be widely cited. There is also a claim here regarding employment losses at the lower end of the skill distribution in tradable services, but this change is essentially zero in table 12 and (if I calculated the standard error correctly) statistically insignificant (at the 5 percent level) in table 13.3
Displaced Workers in Tradable Services The final section of the paper uses the Displaced Workers Survey (DWS) to examine the extent to which being in a tradable job raises a worker’s chance of displacement. Because of coding changes on industries and occupations, the authors cannot do comparisons across this biennial survey. But using the most recent survey, covering the years 2001–03, they find that those in tradable services face significantly higher displacement rates than those in nontradable services. For example, 31.7 percent of those in the tradable sectors of information services were laid off (not for cause) over these years, but only 7.5 percent of those in the nontradable sectors. Here again, the concern is that we are catching the cycle and the bursting of the tech bubble in the analysis, and thus not really isolating an offshoring effect. Information services includes both newspaper publishing (a nontradable service) and Internet publishing (a tradable service), and it is surely the case that a postbubble, large negative spike in domestic demand affected the former more than the latter. A simple difference-in-difference estimator might help to difference out the cycle, say using the changes in displacement in services that were nontradable. The problem is the introduction of new industry and occupation codes in the most recent DWS. However, the BLS has a version of the monthly CPS with 3. I divided the standard deviation by the square root of the number of industries, both given in the table (0.111/3.16) for a standard error of 0.035, which returns a t statistic of –1.85.
J. Bradford Jensen and Lori G. Kletzer
123
new sectoral codes starting in 2000, and although they could not track displacements, the authors should see if these files might enable them to compare wage and employment changes in tradable and nontradable services controlling for the cycle. The DWS has long showed that among displaced workers who are reemployed at the time of the survey, blue-collar production workers take the biggest hit in wages (the pay gap between their old and new jobs is above the average loss). But Jensen and Kletzer find negative effects of a similar magnitude for displaced workers in tradable services. The difference between the old and new wage was, on average, about –30 percent for workers displaced from tradable jobs in both manufacturing and services, and about –14 percent for those displaced from nontradable services. So workers in tradable services were more likely to be displaced during the recent downturn/jobless recovery, and for those who found new jobs at the time of the survey, these displacements were quite costly relative to nontradable services.
Summary Faced with the question of how we identify service workers directly affected by offshoring, Jensen and Kletzer come up with an elegant solution: borrow the observation from the goods-producing literature that firms engaged in trade exhibit geographic concentration. While some might question how well this assumption travels across these different sectors, their results are, for the most part, intuitively satisfying and believable. This aspect of the paper makes a useful contribution to what has been a major stumbling block in this fledgling literature, namely, identifying affected workers in tradable services. The paper’s other major contribution is its documentation of the characteristics of these workers, including their relative earnings. The paper has two shortcomings, both of which are evident in much work on offshoring. First, barring some attempt to control for cyclical effects, it is hard to know whether the job and wage loss effects they identify for workers in tradable services are due to their exposure to offshoring competition or to the protracted labor-market downturn over this period. While they get some traction in this argument by comparing tradable and nontradable services, the problem is that the negative cyclical demand shock was particularly acute in some of the same industries and occupations that have heavy weights in their tradable service category (like IT).
124
Brookings Trade Forum: 2005
Second, from a policy perspective, economists need to look far beyond those directly affected by offshoring to grasp the magnitude of the challenge it poses. Compared to the number who are and will be affected in some way by the competitive pressures from this form of trade, the number of workers who lose their jobs is surely very small. This by no means should lead us to give up on those who take the “direct hit”—workers displaced by service trade. Their needs are often the most acute, and in this regard, ideas like wage insurance and expanding Trade Adjustment Assistance are meritorious. But as Richard Freeman has discussed (this volume), the implicit supply shock from the introduction of millions of skilled workers into a relatively concentrated set of occupations and industries may have a significant impact on the wage structure of white-collar workers, just as the increase in trade in manufacturing goods has structurally altered the wage distribution of blue-collar workers, partially contributing to the post-1979 increase in wage inequality and real wage losses, particularly for men. In this sense, Jensen and Kletzer may be overstating the breadth of their work by giving their piece the subtitle: “Understanding the Scope and Impact of Services Offshoring.” They get us a long way, further than any previous forays, toward identifying the most visible victims of offshoring: those who lose their jobs. But if Samuelson and others are right about the impact of competitive pressures on the United States from trade with low-cost countries in sectors where we have held a comparative advantage, the scope and impact of offshoring could spill over far beyond those directly affected. Robert C. Feenstra: This is a good paper that introduces a new technique for classifying service industries as tradable and nontradable and then pursues a number of applications. The technique involves looking at the geographic concentration of service industries, using the idea that a more concentrated industry is most likely tradable. Geographic concentration is measured using population census data from the PUMS files, which also allow us to track individuals’ occupations as well as their industries of employment. So the paper not only introduces a new technique for measure of the tradability of industries or occupations, it also shows how it can be implemented on a dataset that is novel for trade economists. I actually thought of using the geographic concentration of industries to measure something about trade some years ago, when reading a Scientific American article (Landy 1999) dealing with the distribution of stars in the universe. The “cosmological principle” states that the universe overall is homogeneous, so galaxies have no particular pattern. That is true on a very large scale, but on
J. Bradford Jensen and Lori G. Kletzer
125
smaller scales, galaxies form into clusters that are fractal: even as the scale of observation is reduced, the basic pattern of galaxies is the same. The extent to which galaxies cluster together can be measured by their spatial correlation. When reading that article I thought that the same should be true of the location of economic activity: we could use spatial correlation or some other technique to measure the clustering of industries. That is exactly what the authors do here, using the Gini coefficient and a second measure of concentration. They find that the clustering or concentration of many service industries is just as strong as for manufacturing industries, implying that these service activities must be traded. While my reference to astronomy is just for fun, economists also use the concentration of industries to make conclusions about trade. Jean Imbs and Romain Wacziarg (2003) have shown, for example, that for developing countries the concentration of industries first falls and later increases as the countries mature, so the Gini coefficient follows a U-shaped pattern. For China, Alwyn Young (2000) found that after trade was opened the concentration of industries across provinces fell, which seemed to be contrary to comparative advantage, where we would expect regions to specialize. But later research found that industries in China later became more specialized across provinces, so the Gini coefficient also follows a U-shaped pattern in that country (see Naughton 2003; Poncent 2003). From these examples I conclude that using the concentration of industries to measure their trade orientation is well motivated and that the application to service industries is entirely new. Let us now consider the results of the paper. Using the Gini coefficients of geographic concentration, the authors divide industries into three groups: those with a Gini of less than 0.1 being the least concentrated, and therefore nontradable; those with a Gini above or equal to 0.3 being the most concentrated, and therefore tradable, and those with a Gini between 0.1 and 0.3 in an intermediate category, but also treated as tradable. The classification of industries into these three groups is appealing: there are only a handful of nontraded manufacturing industries, including cement and concrete, whereas service industries are evenly divided between nontraded and traded activities. There are some anomalies, however: the education sector is very diversified geographically, so it is classified as nontradable, despite the fact that it is a principal service export of the United States. The geographic diversification of education holds for elementary and high schools, as well as colleges and universities (see Jensen and Kletzer’s table 2), perhaps because of the land grant system in higher education. Because the authors use census data on individuals from the PUMS files, they can also distinguish tradable occupations as opposed to tradable industries.
126
Brookings Trade Forum: 2005
That is, they can measure the geographic concentration of job titles rather than just industries. These job titles are unfamiliar to trade economists, so some further explanation would be desirable. For example, occupational titles within the life, physical, and social sciences are mostly tradable; that is, these persons are geographically concentrated in their employment (see table 5). About half of these persons work in nontraded industries (such as education, which is not concentrated in space), and another half work in traded industries (see table 6). So at this point I could use some examples to understand the classifications: how can most of the employment in the life, physical, and social sciences be concentrated, when a significant number of these individuals work in education, which is not concentrated? In the next part of the paper, the authors investigate the characteristics of workers as classified by the tradability of their industry and occupation. Workers in traded industries are more highly skilled and are paid more than in nontraded industries, and this is especially true in traded service industries. The same is true for occupations: workers in tradable occupations earn more and have more education than those in nontradable occupations. Even if we strip out the effect of higher education, a wage premium persists for the traded industries, especially for traded service industries: these workers command a premium over and above their education level and demographic characteristics. The premium is about 6 percent for traded manufacturing and 15 percent for traded professional service industries. These results reminded me of two other related studies. First, Jeffrey Sachs and Howard Shatz (1998) made the point that services really are more skillintensive than manufacturing. The characterization of service jobs as flipping hamburgers is not true on average, where the jobs are more likely to be professional. Second, I was reminded of the earlier studies on the wage premiums in manufacturing by Larry Katz and Larry Summers (1989a, 1989b). They found that capital-intensive industries in manufacturing pay higher wages, and since these industries have higher exports, there is a wage premium in exporting. Trade economists were always squeamish about this finding, since it runs the risk of implying that being an exporter leads to paying higher wages, therefore suggesting that a subsidy to exports might help. On the contrary, most of us would believe that being more productive at the plant level leads to being an exporter and paying higher wages, with little or no role for export subsidies (see Fernandez 1989). The authors then investigate the growth across industries and occupations. In this I did not agree with the their expectations regarding which sectors would grow the most. For example, they state: “High-skill activities are consistent with
J. Bradford Jensen and Lori G. Kletzer
127
U.S. comparative advantage, and we would expect that as trade increases, economic activity would shift to activities consistent with U.S. comparative advantage. Thus we would expect higher-skill industries and occupations to have higher rates of employment growth.” My difficulty with this logic is that it all depends on whether the United States is benefiting from increased export opportunities in the sectors where it has comparative advantage, or, on the contrary, whether it is facing new competition in those sectors. Paul Samuelson (2004) suggests that outsourcing could cause the United States to face competition in sectors where it formerly had comparative advantage. That is different from what Jensen and Kletzer have in mind. What they actually find is that service employment expanded during the period 1998–2003 and manufacturing employment contracted, and this shift holds regardless of whether one looks at traded or nontraded industries. So on the issue of employment growth, the methods developed in this paper to measure tradability just do not give us any extra explanatory power. We are back to the hypothesis advanced by James Harrigan and Rita Balaban (1999) and also by Bernardo Blum (2004): namely, that it is the rise in the service sector in the United States, combined with the skill-intensity of that sector (Sachs and Shatz 1998), that explains the rising relative wages of skilled workers. We still do not know whether this shift toward services comes from demand pressure, trade, productivity, or some other cause. It would have been nice if the tradability of service industries gave us extra insight on this issue, but that is not what the empirical results here show. In the final section of the paper the authors examine job loss and the characteristics of displaced workers. This is an issue that Lori Kletzer has written on extensively, and the results here complement her earlier findings. Workers in tradable industries face a notably higher rate of job loss than those in nontradables. That is particularly true in service industries and in white-collar occupations. Nevertheless, it is still true that production workers in the United States have a higher rate of job loss than those in nonproduction and white-collar occupations, including those occupations that we believe are being affected by services outsourcing. General Discussion: Many participants commended the authors for their extremely creative and useful paper. The discussion also raised a variety of issues of interpretation and suggestions for further work, with some questioning how well domestic geographic concentration could capture international tradability. Perhaps not surprisingly, a number of speakers found the results surprising for particular industries or occupations.
128
Brookings Trade Forum: 2005
Lael Brainard highlighted two reasons why the authors’ concentration index approach to identifying tradability was particularly valuable. First, it can be applied across occupations as well as across industries. Second, it gets around the problem that direct measurement is more difficult for services than for goods production. Internationalization of services essentially entails linking domestic and foreign factors of production so that work moves between product or project teams. It is extremely difficult to quantify the value added from each step of the process. Brainard also wondered why the authors focused on a bivariate indicator (whether something was tradable or nontradable) in their empirical analyses instead of exploiting the continuous variable that they constructed. She and others saw their use of an essentially arbitrary threshold as throwing away potentially useful information. The revised version of the paper does provide the actual indicators for major sectors and occupations. Some participants suggested that it would be helpful to compare the results of the tradability measure constructed here with other available alternatives. This would be one way to explore how well it captures what we mean by tradability. Brainard noted that we have direct tradability indicators for merchandise. She expected to find that some highly tradable goods, such as sugar, are not particularly highly concentrated. Catherine Mann wondered whether the approach by Brad Jensen and Lori Kletzer had implications similar to the work by Frank Levy and Richard Murnane, which classifies tasks in terms of routinization. Susan Collins asked how similar it was to the classification by Desirée van Welsum and Xavier Reif. The issue of comparability is partially addressed in the introduction to this volume. Robert Lawrence advanced another way to look at the paper, focusing on agglomeration. The results show that even inside the United States, where firms are free to set up everywhere, they often choose not to, presumably because of the benefits of locating near one another. Clearly, if costs were different enough abroad, they would choose to relocate. But it may be that the more concentrated firms are now, the greater the agglomeration benefits and the less likely they are to move. From this perspective, we should see their concentration as comforting, not threatening. Lawrence also stressed that one should not jump from tradability in the sense of this paper to trade. For example, his work with Martin Baily finds considerable job loss in the computer industry, which is tradable. But their input-output table analysis concludes that this is overwhelmingly due to declines in domestic demand and that trade appears to have played a relatively minor role. Other participants elaborated on Robert Feenstra’s point that domestic tradability may be very different from international tradability. In particular, T. N.
J. Bradford Jensen and Lori G. Kletzer
129
Srinivasan noted that if transactions and transportation costs are much less for domestic than for international trade, then domestic tradability has no implications for international tradability. Srinivasan also raised the point that often the same industry can produce using different technologies. For example, steel, which is certainly tradable, can be produced using both integrated mills that are quite concentrated and the more recent electronic processing mini-mills, which tend to be quite dispersed. He argued that it is important to consider technology in assessing whether concentration provides a good indicator of tradability. He also pointed out that occupation, and perhaps to some degree industry, is a matter of choice. Thus he suggested controlling for selection when estimating the earnings regressions. Catherine Mann noted that regulations can play a very important role in some service sectors. This includes legal bar exams, state-specific insurance regulations, and others. There are also significant differences in cross-country regulations. Thus it would be interesting to explore whether changes in state-specific regulations that make a particular industry more easily traded have affected its occupational stratification or its concentration indicators. Changes in rules for interstate banking are one especially interesting recent example. Mann also asked what the results in the paper could tell us about the risk versus expected return associated with particular occupations. Job loss is certainly very costly. However, her casual impression was that the empirical estimates find a relatively large wage premium for jobs in risky service industries and occupations, and it was worth exploring how this compared with the probability and expected costs of job loss. In contrast, manufacturing jobs are also risky but have been commanding a much smaller wage premium. Lawrence Mishel raised concerns about drawing conclusions from simply comparing employment growth in traded and nontraded industries (or occupations) within a given time period. Because employment trends may be quite different, he thought it important to develop a more convincing counterfactual that incorporates information about previous trend behavior. David Richardson suggested that it would be interesting to consider other concentration measures. For instance, the Ellison-Glaeser measure comes very close to an indicator of revealed comparative advantage. He also noted that the authors should be looking for both industries and occupations with very low concentration and those with very high concentration, because unusually low ratios for production to state GDP are also an indicator of (domestic) tradability. Collins noted that it might be helpful to distinguish between different types of services, and that the domestic concentration approach could be more
130
Brookings Trade Forum: 2005
appropriate for some types than for others. The General Agreement on Trade in Services (GATS) distinguishes among four modes by which services are traded. For example, mode 1 includes services supplied from one country to another, such as telephone calls, while mode 2 includes consumers who use a service in another country, such as tourists and students studying at a foreign university. It seemed to her that domestic concentration might be a better indicator of tradability for mode 1 services than for mode 2.
J. Bradford Jensen and Lori G. Kletzer
131
References Amiti, Mary, and Shang-Jin Wei. 2004. “Fear of Service Outsourcing: Is It Justified?” Working Paper WP/04/186. Washington: International Monetary Fund. Arora, Ashish, and Alfonso Gambardella. 2004. “The Globalization of the Software Industry: Perspectives and Opportunities for Developed and Developing Countries.” Working Paper 10538. Cambridge, Mass.: National Bureau of Economic Research. Bardhan, Ashok Deo, and Cynthia A. Kroll. 2003. “The New Wave of Outsourcing.” Working Paper 1103. University of California, Berkeley: Fisher Center for Real Estate and Urban Economics. Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott. Forthcoming 2006. “Survival of the Best Fit: Exposure to Low Wage Countries and the (Uneven) Growth of U.S. Manufacturing Plants.” Journal of International Economics 68 (1): 219–37. Bhagwati, Jagdish, Arvind Panagariya, and T. N. Srinivasan. 2004. “The Muddles over Outsourcing.” Journal of Economic Perspectives 18 (Fall): 93–114. Blum, Bernardo S. 2004. “Trade, Technology and the Rise of the Service Sector: An Empirical Assessment of the Effects on U.S. Wage Inequality.” University of Toronto (September). Brainard, Lael, and Robert E. Litan. 2004. “Offshoring Service Jobs: Bane or Boon— and What To Do?” Brookings Policy Brief 132 (April). Bronfenbrenner, Kate, and Stephanie Luce. 2004. “The Changing Nature of Corporate Global Restructuring: The Impact of Production Shifts on Jobs in the U.S., China, and around the Globe.” U.S.–China Economic and Security Review Commission. October. Dossani, Rafiq, and Martin Kenney. 2004. “The Next Wave of Globalization? Exploring the Relocation of Service Provision to India.” Working Paper 156. Stanford University: Asia Pacific Research Center. ———. 2003. “Went for Cost, Stayed for Quality? Moving the Back Office to India.” Stanford University: Asia-Pacific Research Center. Duranton, Gilles, and Henry G. Overman. 2004. “Testing for Localisation Using MicroGeographic Data.” London School of Economics. Ellison, Glenn and Edward L. Glaeser. 1999. “The Geographic Concentration of Industry: Does Natural Advantage Explain Agglomeration?” American Economic Review, Papers and Proceedings 89 (May): 311–16. ———. 1997. “Geographic Concentration of U.S. Manufacturing Industries: A Dartboard Approach.” Journal of Political Economy 105 (October): 889–927. Farber, Henry S. 2005. “What Do We Know about Job Loss in the United States? Evidence from the Displaced Worker Survey, 1984–2004.” Working Paper 498. Princeton University: Industrial Relations Section. ———. 2003. “Job Loss in the United States, 1981–2001.” Working Paper 471. Princeton University: Industrial Relations Section. ———. 1993. “The Incidence and Costs of Job Loss: 1982–1991.” BPEA, no. 1: 73–132. Fernandez, Raquel, 1989. Comment on “Industrial Wage Differential, International Competition and Trade Policy,” by Lawrence F. Katz and Lawrence H. Summers. In
132
Brookings Trade Forum: 2005
Trade Policies for International Competitiveness, edited by Robert Feenstra, pp. 117–22. University of Chicago Press. Gardner, Jennifer M. 1993 “Recession Swells Count of Displaced Workers.” Monthly Labor Review 116 (June): 14–23. Gartner Research. 2004. “Worldwide IT Services Market Forecast, 2002–2007” (January 12). Gentle, Chris. 2003. “The Cusp of a Revolution: How Offshoring Will Transform the Financial Services Industry.” Deloitte Research. Harrigan, James, and Rita A. Balaban. 1999. “U.S. Wages in General Equilibrium: The Effects of Prices, Technology, and Factor Supplies, 1963–1991.” Working Paper 6981. Cambridge, Mass.: National Bureau of Economic Research. Helpman, Elhanan, and Paul R. Krugman. 1985. Market Structure and Foreign Trade: Increasing Returns, Imperfect Competition, and the International Economy. Cambridge, Mass.: MIT Press. Imbs, Jean, and Romain Wacziarg. 2003. “Stages of Diversification.” American Economic Review 93 (March): 63–86. Katz, Lawrence F., and Lawrence H. Summers. 1989a. “Can Interindustry Wage Differentials Justify Strategic Trade Policy?” In Trade Policies for International Competitiveness, edited by Robert Feenstra, pp. 85–116. University of Chicago Press. ———. 1989b. “Industry Rents: Evidence and Implications.” In Brookings Papers on Economic Activity: Microeconomics 1989, edited by Martin N. Baily and Clifford Winston, pp. 209–75. Brookings. Kirkegaard, Jacob F. 2004. “Outsourcing—Stains on the White Collar?” Washington: Institute for International Economics (February). Kletzer, Lori G. 2001. Job Loss from Imports: Measuring the Costs. Washington: Institute for International Economics. ———. 1998. “Job Displacement.” Journal of Economic Perspectives 12 (Winter): 115–36. ———. 1995. “White Collar Job Displacement, 1983–91.” In Proceedings of the 47th Annual Meeting, pp. 98–107. Madison, Wis.: Industrial Relations Research Association. Krugman, Paul R. 1991. Geography and Trade. Cambridge, Mass.: MIT Press. Landy, Stephen D. 1999. “Mapping the Universe.” Scientific American 280 (June): 38–45. Mann, Catherine L. 2003. “Globalization of IT Services and White Collar Jobs: The Next Wave of Productivity Growth.” Policy Brief 03-11. Washington: Institute for International Economics. McCarthy, John C. 2002. “3.3 Million U.S. Services Jobs To Go Offshore.” San Francisco: TechStrategyTM Research, Forrester Research (November). Naughton, Barry J. 2003. “How Much Can Regional Integration Do to Unify China’s Markets?” In How Far across the River? Chinese Policy Reform at the Millennium, edited by Nicholas Hope, Dennis Tao Yang, and Mu Yang Li, pp. 204–32. Stanford University Press.
J. Bradford Jensen and Lori G. Kletzer
133
Podgursky, Michael. 1992. “The Industrial Structure of Job Displacement 1979–89.” Monthly Labor Review 115 (September): 17–25. Poncet, Sandra. 2003. “Measuring Chinese Domestic and International Integration.” China Economic Review 14 (1): 1–21. Sachs, Jeffrey D., and Howard J. Shatz. 1998. “International Trade and Wage Inequality: Some New Results.” In Imports, Exports, and the American Worker, edited by Susan M. Collins, pp. 215–40. Brookings. Samuelson, Paul A. 2004. “Where Ricardo and Mill Rebut and Confirm Arguments of Mainstream Economists Supporting Globalization.” Journal of Economic Perspectives 18 (Summer): 135–46. Schultze, Charles L. 2004. “Offshoring, Import Competition, and the Jobless Recovery.” Brookings Policy Brief 136 (August). Tilton, Andrew. 2003. “Offshoring: Where Have All the Jobs Gone?” U.S. Economic Analyst 03/38 (September 19). Goldman Sachs & Co. Van Welsum, Desirée, and Graham Vickery. 2005. “Potential Offshoring of ICTIntensive Using Occupations.” DSTI Information Economy Working Paper DSTI/ICCP/IE(2004)19/FINAL. Paris: Organization for Economic Cooperation and Development. Young, Alwyn. 2000. “The Razor’s Edge: Distortions and Incremental Reform in the People’s Republic of China.” Quarterly Journal of Economics 115 (November): 1091–35.
This page intentionally left blank
MARIA BORGA U.S. Bureau of Economic Analysis
Trends in Employment at U.S. Multinational Companies: Evidence from Firm-Level Data
B
ecause of their networks of foreign affiliates and their well-established trade channels, U.S. multinational companies (MNCs) would be expected to play a leading role in the offshoring of production of goods and services. As a result, there has been much discussion about the impact of offshoring on employment at U.S. MNCs. In this paper, I use firm-level data from the Bureau of Economic Analysis (BEA) to examine the evidence on the extent of offshoring by U.S. parent companies to their foreign affiliates and then to determine if, and how, offshoring is associated with changes in U.S. parent employment. While no standard definition of offshoring exists, the Government Accountability Office (GAO) recently defined the offshoring of services as generally referring to “an organization’s purchases from abroad (imports) of services that it previously produced in-house or purchased from another domestic source.”1 This definition could be adapted to apply equally well to goods. Such an expanded definition is applicable to all U.S. firms and is broad enough to encompass several different types of offshoring behavior. For example, it covers the case in which a U.S. parent company begins to purchase an input, either a good or a service, from a foreign affiliate or from an unaffiliated foreign supplier that it previously either produced in-house or purchased from a domestic supplier. It
The author wishes to thank this paper’s discussants, Robert Lawrence and Catherine Mann, and participants at the Brookings Trade Forum for helpful comments. The author also thanks her colleagues at the U.S. Bureau of Economic Analysis for their valuable suggestions and comments. The views expressed in this paper are those of the author and do not necessarily reflect those of the Bureau of Economic Analysis. 1. GAO (2004, p. 2). This report goes on to observe, “The term offshoring has also been used in the public debate to include several other types of international trade and foreign investment activities.”
135
136
Brookings Trade Forum: 2005
would also cover the case of a U.S. parent company choosing to transfer to a foreign affiliate the production of a good or service produced domestically and exported to foreign markets. This paper does not attempt to deal with all types of offshoring, but instead focuses on a very specific form of offshoring: the case in which a U.S. MNC chooses to produce a good or service at a foreign affiliate and then to import that good or service back to the U.S. parent company to use in its production or to resell. This setup is a specific form of vertical foreign direct investment (FDI), with production located in other countries to take advantage of differences in relative factor costs or to exploit economies of scale.2 On the opposite end of the spectrum is horizontal FDI, in which MNCs establish foreign affiliates whose production processes replicate those of the parents.3 A single MNC may demonstrate characteristics of both vertical and horizontal FDI. For example, it may expand horizontally by establishing an affiliate in Europe to serve the local market, and also vertically by establishing an affiliate in Mexico to assemble and reexport parts fabricated in the United States or by establishing an affiliate in India to provide accounting services to the U.S. parent. Much of the discussion about offshoring by U.S. MNCs has centered on its impact on employment in the United States. It has been conjectured that as U.S. MNCs offshore there will be a negative impact on employment in the United States, either from production lost in the United States to foreigners or from forgone job creation. This conjecture, however, ignores the potential positive impact that offshoring can have on the MNC as a whole. If offshoring achieves cost savings, then the MNC’s total sales may expand, resulting in increased demand for the goods and services still provided by the U.S. parent as well as by its foreign affiliates. In an extreme case, the cost savings from offshoring may enable the MNC to survive. BEA has examined long-term trends in the aggregate data on U.S. MNCs.4 This analysis showed that worldwide operations of U.S. MNCs are concentrated in the United States, that the foreign operations of U.S. MNCs are centered in high-income countries, and that most of the output of foreign affiliates is sold in local or other foreign markets rather than exported to the United States. However, the aggregate data used in this analysis could mask diversity at the firm level by giving greater weight to larger firms. This paper builds on the earlier 2. Another form of vertical FDI occurs when the foreign affiliate of an MNC sells its output to another affiliate of the U.S. parent for use as an intermediate input rather than selling it back to the parent. For a discussion of vertical FDI, see Helpman and Krugman (1985). 3. See Markusen (2002). 4. U.S. Bureau of Economic Analysis (2004, pp. 52–56).
Maria Borga
137
work by examining the firm-level data. Whereas the earlier work portrayed aggregate patterns in the data, the present analysis is more directed at characterizing and explaining typical firm behavior. In general, it confirms the earlier findings and makes some additional ones as well: —While U.S. parents increased their reliance on purchased goods and services between 1994 and 2002, there was no significant association between increased reliance on purchased inputs and decreases in parent employment. However, changes in parents’ reliance on imports of goods from their foreign affiliates was significantly and negatively associated with changes in parent employment. In contrast, there was no significant correlation between changes in parents’ reliance on imports of services from their affiliates and changes in parent employment. —More of the change in parent employment between 1994 and 2002 is attributable to changes in output and in labor productivity than to changes in the use of purchased inputs. —Growth at U.S. parents and at their foreign affiliates is closely and positively linked. This growth can result from mergers and acquisitions as well as the expansion of existing operations. —On average, the share of sales by foreign affiliates to local markets increased between 1994 and 2002, suggesting that market access is an increasingly important reason for investing overseas. The increase in the share of sales sold locally coincided with an increase in the share of affiliate employment in low-income countries. This result suggests that for investment in low-income countries, market access, not just factor cost differences, is an important consideration. In the remaining sections of the paper I first describe the data set. I then present descriptive statistics from the panel data set. Next I estimate the relative impacts of the main sources of change in parent employment. I then seek to explain associations between changes in employment at U.S. parents and changes in selected characteristics of the U.S. parents, their affiliates, and the entire MNC. Finally, I offer conclusions and suggest some avenues for future research.
The Data Set The analysis is based on a panel of 1,117 U.S. parent companies that responded to both BEA’s 1994 benchmark survey and its 2002 annual survey of U.S. direct investment abroad. The panel thus excludes firms that entered or dropped out of the universe after 1994. By looking at a sample of long-lived
138
Brookings Trade Forum: 2005
parents over an extended period, it is possible to distinguish changes in behavior that may be obscured in the aggregate data covering all parents, particularly because of the entry of new parents into the universe. It is mandatory for firms to respond to BEA’s direct investment surveys, which obtain data on the financing and operations of U.S. parents and their foreign affiliates. The analysis that follows covers only majority-owned foreign affiliates because U.S. parents are likely to exhibit a higher degree of management influence and control over majority-owned affiliates than minority-owned affiliates. Also, only nonbank affiliates are included because bank affiliates are not required to report financial and operating data to BEA for years not covered by benchmark surveys. One of the variables that is particularly relevant to a discussion of offshoring is the extent to which U.S. parents’ reliance on purchased goods and services has changed over time. Because the resale of purchased goods is the principal business of wholesalers and retailers, including them in the data set could obscure any changes in the reliance of parents in other industries on purchased inputs.5 Therefore all parents whose primary industry is wholesale or retail trade have been excluded from the data set.6 Mergers and acquisitions affected the companies that were included in the panel data set. Take the case in which a U.S. parent company (company A) that reported on the 1994 benchmark survey was acquired by another U.S. parent company (company B) before 2002. Company A (and its affiliates) was not included in the panel data set because it did not report on the 2002 annual survey. Company B was included in the panel data set because it did report in both 1994 and 2002. Company B’s reported data in 2002 would reflect the acquisition of company A and its affiliates. The acquisition would manifest itself as growth in employment (at both the parent and the affiliates), in sales, and so on. The variables included in the data set are employment, employee compensation, value added, sales, expenditures on research and development (R&D), and
5. In estimating the gross output of the wholesale and retail trade industries, the goods for resale are excluded from the value of intermediate inputs consumed in production by wholesalers and retailers because these goods are subject to only minimal processing, such as cleaning or packaging. 6. U.S. parents usually have operations in multiple industries but are classified in a primary industry on the basis of their sales data. Throughout the paper, the parent’s primary industry classification in 2002 will be used to classify the data. Because parents can have operations in multiple industries, those that have secondary activities in wholesale or retail trade remain in the data set.
Maria Borga
139
exports and imports of goods. Sales are disaggregated into goods, services, and investment income. Affiliates’ sales are disaggregated into six destinations: to the host country, to the United States, or to third countries, and then further for each of these into sales to either affiliated or unaffiliated parties. The sample accounts for 69 percent of employment and 68 percent of sales of all parents in 1994; by 2002 the percentage of employment covered by the sample had fallen to 63 percent, but the percentage of sales covered had increased to 69 percent. Most of the characteristics for the sample match the population. However, there are some differences. The sample shows slower employment growth than the population. Also, it slightly underrepresents services firms and overrepresents firms that rely on imports from their affiliates in their production. For further information about the data set and a complete comparison of the panel data set with the population, see the appendix.
Descriptive Statistics This section begins with an examination of changes between 1994 and 2002 in the mean values of selected characteristics of the firms in the sample data set. The characteristics examined describe the U.S. parents, their foreign affiliates, and the MNC as a whole, such as the distribution of employment between the parents and their affiliates. Several of these characteristics are expected to be associated with changes in U.S. parent employment, such as changes in sales, in labor productivity, in the share of purchased inputs in the parents’ production, and in the sourcing of inputs by parents from their foreign affiliates. Other characteristics, such as R&D spending, are less closely associated with changes in U.S. parent employment but are still of interest. Then the sample is divided between parents that lost employment and those that gained employment, and the characteristics of those two subsets are contrasted. Finally, the characteristics are examined by industry of the U.S. parent. Characteristics of the Firms in the Sample Table 1 shows the mean values of selected sample characteristics in 1994 and 2002. Average employment growth for the parents in the sample was 39 percent. Note that an increase in employment for a parent could result not only from an expansion of existing operations, but also from the merger with, or acquisition of, an existing U.S. company. By the same token, a decrease in a parent’s
140
Brookings Trade Forum: 2005
Table 1. Characteristics of the MNCs in the Panel: Mean Values of Selected Variable, 1994 and 2002 1994 Variables U.S. parent variables Employment change (percent) Salesa (millions of dollars) Ratio of purchases to sales Share of services in sales International share of sales R&D intensity Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars) Affiliate variables Employment change (percent) Share of services in sales Share of sales to local market Share of sales to third countries Share of sales to U.S. Value added per employeeb (thousands of dollars) Compensation per employeeb (thousands of dollars) MNC variables Parent share of employment Parent share of sales Parent share of value added Share of affiliate employment in low-income countries Affiliated goods imports as a share of U.S. parent purchases Unaffiliated goods imports as a share of U.S. parent purchases Affiliated services imports as a share of U.S. parent purchases Number of observations
2002
Mean
Standard deviation
Mean
Standard deviation
… 2,378 0.60 0.22 0.14 0.031
… (7,650) (0.17) (0.39) (0.17) (0.05)
39 3,190 0.63 0.24 0.14 0.029
(1.02) (9,712) (0.18) (0.40) (0.18) (0.07)
86.58
(59.99)
89.43
(70.30)
55.00
(25.44)
61.65
(28.99)
… 0.21 0.70 0.19 0.11
… (0.39) (0.31) (0.26) (0.21)
231 0.24 0.78 0.14 0.08
(7.87) (0.41) (0.27) (0.08) (0.16)
n.a.
n.a.
89.60
(580.65)
n.a.
n.a.
44.59
(30.49)
0.77 n.a. n.a.
(0.20) n.a. n.a.
0.72 0.73 0.80
(0.22) (0.20) (0.98)
0.20
(0.32)
0.27
(0.32)
0.035
(0.09)
0.046
(0.11)
0.030
(0.08)
0.030
(0.09)
0.004
(0.04)
0.002
(0.02)
1,117
1,117
a. For 1994, rough estimates of U.S. parents’ sales, value added, and employee compensation in 2002 dollars were derived using the U.S. GDP deflator. b. For 1994, estimates of affiliates’ sales, value added, and employee compensation in 2002 dollars were not derived because of the difficulties of accounting for changes in exchange rates from 1994 to 2002. n.a. = Not available.
Maria Borga
141
employment could result either from the contraction of existing operations or from the selling off of some of the parent’s operations. U.S. PARENT CHARACTERISTICS. The ratio of purchased goods and services by U.S. parents to their sales measures the reliance of parents on inputs purchased from outside the firm rather than on production inside the firm.7 These purchases could be sourced from domestic companies or imported from foreign suppliers, both affiliated and unaffiliated. Thus, this measure encompasses both the outsourcing (inputs provided by a third party, either domestic or foreign) and offshoring (inputs provided by a foreign company, either affiliated or unaffiliated) behavior of the firms in the sample. Parents’ offshoring behavior is examined by looking at imports as a share of their purchased inputs. The data show that reliance on purchases of goods and services increased between 1994 and 2002, from 60 percent to 63 percent. The share of sales accounted for by services increased slightly between 1994 and 2002, as sales of services increased more rapidly than sales of goods. The share of U.S. parents’ sales accounted for by international sales remained constant, indicating that U.S. parents’ international orientation did not change between 1994 and 2002. R&D intensity (measured as the ratio of R&D expenditures to sales) fell slightly between 1994 and 2002. Mean labor productivity, as measured by value added per employee, grew between 1994 and 2002; however, its growth was less than the growth in mean compensation per employee. FOREIGN AFFILIATE CHARACTERISTICS. The mean change in employment at the foreign affiliates is 231 percent. There are some parents in the panel whose foreign operations were quite small in 1994 and whose expansion resulted in employment increases at their foreign affiliates of several orders of magnitude by 2002. These few outliers raise the mean significantly. The median increase in affiliate employment was 47 percent—much lower than the mean increase in affiliate employment, but still higher than the mean increase in parent employment. Furthermore, not all of these increases in affiliate employment are due to growth at existing affiliates or to the establishment or acquisition of new affiliates overseas. For example, some affiliate employment growth is due to the acquisition of an existing U.S. parent by another U.S. parent. Such acquisitions would result in the affiliates of the acquired U.S. parents being transferred to the 7. Purchases of goods and services are calculated residually as sales minus value added. Parents’ output is measured as their sales. Ideally, to accurately reflect output, sales should be adjusted for inventory change, but because BEA collects data on parents’ inventories only for benchmark years, and then only as of the end of the year, it is not able to estimate the annual change in inventories for parents.
142
Brookings Trade Forum: 2005
acquiring U.S. parents and would be manifested as employment growth at the affiliates of the acquiring U.S. parents. The mean share of affiliates’ sales accounted for by services increased between 1994 and 2002, indicating that, as for parents, sales of services grew faster than sales of goods for affiliates. In 2002 an average share of 78 percent of affiliates’ sales was to the local market. The high share of sales to local markets demonstrates the importance of serving local customers in the parent’s decision to invest abroad. For an affiliate, the greater the share of local sales, the more likely the affiliate was established as part of horizontal expansion abroad by its parent rather than vertical expansion. Because offshoring is defined in this paper as the parents sourcing inputs from their affiliates, horizontal expansion is less associated with offshoring than vertical expansion. Between 1994 and 2002, the share of sales to the local market increased while the share of sales to third countries and to the United States fell. This suggests that serving local markets was an increasingly important motive for U.S. MNCs’ overseas expansion. MNC CHARACTERISTICS. For the MNC as a whole, the typical parent accounts for a majority share of MNC employment, sales, and production. However, the average parent share of MNC employment fell between 1994 and 2002, reflecting the faster growth rate in employment at affiliates than at parents. The average share of affiliate employment in low-income countries increased between 1994 and 2002.8 Countries in Asia and the Pacific and Latin America and Other Western Hemisphere accounted for the largest share of growth in the low-income countries. In Asia and the Pacific, China and India accounted for the largest increases in employment. In Latin America and Other Western Hemisphere, the largest increases were in Mexico and Brazil. Notably, the increase in the average share of affiliate employment in low-income countries coincided with an increase in the local share of sales for all affiliates, demonstrating that while factor costs differences may be important in the decision to invest in lowincome countries, serving those markets is also an important reason, and perhaps the paramount one. The three characteristics that describe parents’ offshoring behavior are the share of their purchased inputs accounted for by imports of goods from their affiliates, by imports of goods from unaffiliated parties, and by imports of services from their affiliates. Parents’ use of goods imported from their foreign affiliates grew more intensive between 1994 and 2002; the average share of par8. High-income countries are defined as all members of the Organization for Economic Cooperation and Development except the Czech Republic, Hungary, Mexico, Poland, the Slovak Republic, the Republic of Korea, and Turkey. All other countries are considered low-income countries.
Maria Borga
143
ents’ purchased inputs accounted for by affiliated imports of goods increased from 3.5 percent to 4.6 percent. In contrast, there was little change in the share of parents’ purchases accounted for by imports of goods from unaffiliated parties. The share of parents’ purchased inputs accounted for by imports of services from their affiliates fell from 0.4 percent in 1994 to 0.2 percent in 2002. Summing over the share of purchased inputs accounted for by imports that can be identified in the data shows that parents’ reliance on imports increased from 6.9 percent in 1994 to 7.8 percent in 2002. Even though data on unaffiliated imports of services are not available, it is clear that the vast majority of U.S. parents’ purchases were from domestic suppliers.9 Characteristics of MNC Parents That Gained or Lost Employment U.S. PARENT CHARACTERISTICS. Parents that gained employees outnumbered parents that lost employees by 689 to 428 (see table 2). For parents that gained employees, the average employment increase of 82 percent was slightly below the average increase in sales of 84 percent. In contrast, for those parents that lost employees, the average employment decrease of 30 percent outpaced the average decrease in sales of only 10 percent. For parents that hired additional employees, in order to reach levels of sales that outpaced their employment growth, either their workforce had to become more productive or the parents had to rely to a greater extent on purchased inputs. While labor productivity was virtually constant for those parents that gained employees, they did increase their reliance on purchased inputs from 59 percent to 62 percent of sales. For parents that lost employees, the much greater reduction in employment than in sales implies either increased labor productivity or greater reliance on purchased inputs. Unlike for the parents that gained employees, both labor productivity and reliance on purchased inputs increased for parents that lost employees— labor productivity increased 8.2 percent, and purchased inputs increased from 62 percent of sales to 66 percent of sales. Employees of parents that reduced the size of their workforce had larger increases in average compensation than employees of parents whose workforce grew. Finally, services accounted for a larger share of sales for those that gained employees. FOREIGN AFFILIATE CHARACTERISTICS. While affiliate employment grew for both sets of parents, those parents who gained employees had greater average
9. Purchases from domestic suppliers could embody imported goods and services. Currently, it is not possible to estimate the import content of purchases from domestic suppliers. However, BEA is exploring the possibility of estimating these indirect imports with the use of input-output tables.
144
Brookings Trade Forum: 2005
Table 2. Characteristics of MNCs in the Panel for U.S. Parent Firms That Gained or Lost Employment: Mean Values for Selected Variables, 1994 and 2002
U.S. parent variables Employment Salesb (millions of dollars) Employment change (percent) Ratio of purchases to sales Share of services in sales International share of sales R&D intensity Value added per employeeb (thousands of dollars) Compensation per employeeb (thousands of dollars) Affiliate variables Employment change (percent) Services as a share of sales Share of sales to local market Share of sales to third countries Share of sales to the United States Value added per employeec Compensation per employeec
Parents with employment gains
Parents with employment losses
1994
2002
1994
2002
7,554.4 (19,576.0)a 1,815 (5,106) …
11,437.3 (26,258.3) 3,342 (9,353) 0.82 (1.09) 0.62 (0.18) 0.27 (0.42) 0.14 (0.18) 0.030 (0.07) 88.93 (74.10) 59.45 (29.40)
12,184.4 (33,641.3) 3,286 (15,106) …
7,566.4 (17,859.0) 2,945 (13,312) –0.30 (0.23) 0.66 (0.18) 0.20 (0.37) 0.14 (0.17) 0.028 (0.06) 90.23 (63.77) 65.18 (27.97)
0.59 (0.18) 0.24 (0.41) 0.14 (0.17) 0.031 (0.06) 88.54 (59.96) 56.01 (28.38) … 0.24 (0.41) 0.71 (0.32) 0.19 (0.26) 0.11 (0.21) n.a. n.a.
3.07 (9.24) 0.27 (0.42) 0.80 (0.27) 0.13 (0.22) 0.07 (0.15) 72.61 (549.19) 45.11 (32.88)
0.62 (0.16) 0.18 (0.35) 0.14 (0.15) 0.030 (0.05) 83.42 (59.97) 53.39 (19.66) … 0.17 (0.35) 0.68 (0.31) 0.20 (0.25) 0.11 (0.21) n.a. n.a.
1.09 (4.65) 0.21 (0.38) 0.74 (0.28) 0.16 (0.23) 0.10 (0.17) 116.97 (627.71) 43.74 (26.19) (continued)
growth in employment at their affiliates than parents who lost employees. All affiliates increased their focus on selling to the local market, but the increase in the share of local sales was greater for affiliates of parents with employment gains. Also, there was a decrease in the share of sales to the United States for affiliates of parents that gained employees, while that share was little changed for the other affiliates.
Maria Borga
145
Table 2. Characteristics of MNCs in the Panel for U.S. Parent Firms That Gained or Lost Employment: Mean Values for Selected Variables, 1994 and 2002 (Continued)
U.S. parent variables MNC variables Parent share of employment Share of employment in lowincome countries Affiliated goods imports as a share of parent purchases Unaffiliated goods imports as a share of parent purchases Affiliated services imports as a share of parent purchases Number of observations
Parents with employment gains
Parents with employment losses
1994
2002
1994
2002
0.78 (0.20) 0.19 (0.33) 0.034 (0.09) 0.029 (0.09) 0.006 (0.05)
0.76 (0.20) 0.26 (0.32) 0.039 (0.11) 0.027 (0.08) 0.002 (0.02)
0.76 (0.20) 0.21 (0.31) 0.037 (0.08) 0.031 (0.07) 0.0015 (0.01)
0.66 (0.24) 0.28 (0.33) 0.058 (0.13) 0.033 (0.09) 0.002 (0.01)
689
428
a. Standard deviations are in parentheses. b. Rough estimates of U.S. parents’ sales, value added, and employee compensation in 2002 dollars were derived using the U.S. GDP deflator. c. For 1994, estimates of affiliates’ sales, value added, and employee compensation in 2002 dollars were not derived because of the difficulties of accounting for changes in exchange rates from 1994 to 2002. n.a. = Not available.
MNC CHARACTERISTICS. Those parents that lost employees strengthened their ties to their affiliates by increasing their reliance on imported goods from them, from 3.7 percent of purchases to 5.8 percent of purchases. To a lesser extent, they increased their reliance on services imported from their affiliates, from 0.15 percent to 0.2 percent. These parents also increased their reliance on goods imported from unaffiliated parties. On the other hand, parents that gained employees had a smaller increase in reliance on imports of goods from their affiliates, from 3.4 percent to 3.9 percent of purchased inputs, and reduced their reliance on imports of services from their affiliates, from 0.6 percent to 0.2 percent. They also reduced their reliance on imports of goods from unaffiliated parties.
Characteristics by Industry of the U.S. Parent Tables 3, 4, and 5 show the mean values for the parent, the affiliate, and the MNC characteristics, respectively, using three broad industry categories for the U.S. parent: manufacturing, services, and “other” industries.10 10. For this paper, services are defined as the following NAICS sectors: information; professional, scientific, and technical services; and finance and insurance (except depository institutions because the data set consists of only nonbank parents and affiliates). “Other” industries are all remaining NAICS sectors except manufacturing.
146
Brookings Trade Forum: 2005
Table 3. Characteristics of Parent Firms in the Panel by Industry: Mean Values of Selected Variables, 1994 and 2002 1994 Characteristics Parents in manufacturing Employment change (percent) Ratio of purchases to sales Salesa (millions of dollars) Share of services in sales International share of sales R&D intensity Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars) Number of observations Parents in services Employment change (percent) Ratio of purchases to sales Salesa (millions of dollars) Share of services in sales International share of sales R&D intensity Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars) Number of observations Parents in “other” industries Employment change (percent) Ratio of purchases to sales Salesa (millions of dollars) Share of services in sales International share of sales R&D intensity Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars) Number of observations
2002
Mean
Standard deviation
Mean
Standard deviation
… 0.62 2,080 0.02 0.16 0.033
… (0.14) (7,647) (0.10) (0.16) (0.04)
31 0.65 2,770 0.02 0.16 0.034
(0.93) (0.16) (9,812) (0.08) (0.16) (0.07)
(46.72)
85.51
(60.29)
(16.50)
57.97
82.18 52.26 822 … 0.55 4,113 0.83 0.09 0.039
… (0.25) (9,670) (0.33) (0.16) (0.10)
100.17 73.12
57 0.58 5,942 0.94 0.10 0.028
(86.78)
106.55
(48.40)
81.88
163 … 0.57 2,097 0.68 0.10 0.005 101.83 50.45 132
(23.50) 822 (1.23) (0.24) (11,903) (0.19) (0.21) (0.08) (83.90) (40.46) 163
… (0.22) (3,477) (0.42) (0.20) (0.02)
81 0.61 2,405 0.74 0.08 0.003
(91.59)
94.74
(21.03)
59.62
(2.63) (0.22) (3,926) (0.38) (0.19) (0.01) (101.00) (32.52) 132
a. Rough estimates of U.S. parents’ sales, value added, and employee compensation in 2002 dollars were derived using the U.S. GDP deflator.
Maria Borga
147
U.S. PARENT CHARACTERISTICS. There are 822 manufacturing parents in the panel, far outnumbering the 163 parents in services and the 132 parents in “other” industries (see table 3). Parents in “other” industries had the fastest average employment growth between 1994 and 2002, at 81 percent, but the smallest increase in average sales, at 15 percent. While their reliance on purchased inputs increased, from an average of 57 percent of sales to 61 percent, their average labor productivity fell. Parents in services had the second fastest average employment growth, at 57 percent, but the largest increase in average sales, at 44 percent. These parents increased their reliance on purchased inputs, from an average of 55 percent of sales to 58 percent, and the average productivity of their workers increased. Parents in manufacturing had the slowest average growth in employment, at 31 percent, but the second largest increase in average sales, at 33 percent. These parents increased their reliance on purchased inputs, from an average of 62 percent of sales to 65 percent, and their average labor productivity increased. Thus, for parents in each industry group, employment, sales, and reliance on purchased inputs increased. For parents in manufacturing and services, labor productivity increased, while it fell for parents in “other” industries. FOREIGN AFFILIATE CHARACTERISTICS. Affiliates of parents in services had the fastest employment growth between 1994 and 2002, followed by affiliates of parents in “other” industries and in manufacturing (table 4). The share of sales sold locally increased for affiliates of parents in manufacturing and in services. It fell for affiliates of parents in “other” industries, but it was the share of sales to third countries, not to the United States, that increased for affiliates of parents in “other” industries. Indeed, the share of sales to the United States fell for all three industry groups. MNC CHARACTERISTICS. Parents in manufacturing accounted for a smaller share of MNC employment and sales than other parents (see table 5). Parents in “other” industries had the largest share of affiliate employment in low-income countries, but parents in services had the largest increase in the share of employment in low-income countries. For parents in services and in manufacturing, the increases in affiliate employment in low-income countries coincided with increases in the share of sales to the local market, suggesting that serving these markets, rather than exploiting differences in factor costs, was the principal reason for investing in them. Turning to parents’ offshoring behavior, parents in manufacturing used imported goods the most intensively—the share of purchased inputs accounted for by goods imported from affiliates increased from 4.6 to 5.9 percent, and the share accounted for by goods imported from unaffiliated parties was constant at
148
Brookings Trade Forum: 2005
Table 4. Characteristics of the Majority-Owned Foreign Affiliates in the Panel by Industry of the Parent: Mean Values of Selected Variables, 1994 and 2002 1994 Characteristics Affiliates of parents in manufacturing Employment change (percent) Share of services in sales Salesa (millions of dollars) Share of sales to local market Share of sales to third countries Share of sales to U.S. Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars) Number of observations Affiliates of parents in services Employment change (percent) Share of services in sales Salesa (millions of dollars) Share of sales to local market Share of sales to third countries Share of sales to U.S. Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars) Number of observations Affiliates of parents in “other” industries Employment change (percent) Share of services in sales Salesa (millions of dollars) Share of sales to local market Share of sales to third countries Share of sales to U.S. Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars) Number of observations
2002
Mean
Standard deviation
Mean
Standard deviation
… 0.03 n.a. 0.65 0.23 0.12
… (0.12) n.a. (0.31) (0.26) (0.22)
1.89 0.04 1,858 0.75 0.16 0.10
(6.91) (0.15) (9,822) (0.27) (0.22) (0.17)
n.a.
n.a.
79.96
(215.45)
n.a.
40.52
n.a. 822
(17.13) 822
… 0.79 n.a. 0.83 0.12 0.05
… (0.38) n.a. (0.26) (0.22) (0.13)
3.87 0.91 1,623 0.88 0.10 0.02
(10.37) (0.24) (6,146) (0.23) (0.21) (0.07)
n.a.
n.a.
100.22
(161.15)
n.a.
62.75
n.a. 163 … 0.64 n.a. 0.85 0.07 0.08
… (0.45) n.a. (0.27) (0.18) (0.20)
n.a. n.a. 132
(44.36) 163
2.99 0.69 731 0.83 0.10 0.07
(9.52) (0.43) (1,802) (0.28) (0.22) (0.17)
n.a.
162.28
(1,599.48)
n.a.
47.82
(56.23) 132
a. For 1994, estimates of affiliates’ sales, value added, and employee compensation in 2002 dollars were not derived because of the difficulties of accounting for changes in exchange rates from 1994 to 2002. n.a. = Not available.
Maria Borga
149
Table 5. Selected Characteristics of MNCs in the Panel by Industry of the Parent: Mean Values of Selected Variables, 1994 and 2002 1994 Characteristics MNCs with parents in manufacturing Parent share of employment Parent share of salesa Parent share of value addeda Share of employment in low-income countries Affiliated goods imports as a share of U.S. parent purchases Unaffiliated goods imports as a share of U.S. parent purchases Affiliated services imports as a share of U.S. parent purchases Number of observations MNCs with parents in services Parent share of employment Parent share of salesa Parent share of value addeda Share of employment in low-income countries Affiliated goods imports as a share of U.S. parent purchases Unaffiliated goods imports as a share of U.S. parent purchases Affiliated services imports as a share of U.S. parent purchases Number of observations MNCs with parents in “other” industries Parent share of employment Parent share of salesa Parent share of value addeda Share of employment in low-income countries Affiliated goods imports as a share of U.S. parent purchases Unaffiliated goods imports as a share of U.S. parent purchases Affiliated services imports as a share of U.S. parent purchases Number of observations
2002
Mean
Standard deviation
0.76 n.a. n.a.
(0.20) n.a. n.a.
0.71 0.72 0.77
(0.22) (0.19) (0.24)
0.20
(0.31)
0.26
(0.32)
0.046
(0.10)
0.059
(0.13)
0.038
(0.10)
0.038
(0.10)
(0.01)
0.0004
0.001
Standard deviation
Mean
822
(0.002) 822
0.83 n.a. n.a.
(0.17) n.a. n.a.
0.77 0.78 1.00
(0.20) (0.18) (2.50)
0.13
(0.25)
0.22
(0.29)
0.004
(0.03)
0.005
(0.04)
0.005
(0.03)
0.003
(0.03)
(0.09)
0.007
0.017 163
(0.04) 163
0.79 n.a. n.a.
(0.23) n.a. n.a.
0.75 0.75 0.74
(0.26) (0.26) (0.30)
0.33
(0.42)
0.32
(0.39)
0.007
(0.03)
0.014
(0.07)
0.009
(0.03)
0.009
(0.06)
(0.05)
0.008
0.008 132
(0.03) 132
a. For 1994, estimates of affiliates’ sales, value added, and employee compensation in 2002 dollars were not derived because of the difficulties of accounting for changes in exchange rates from 1994 to 2002. n.a. = Not available.
150
Brookings Trade Forum: 2005
3.8 percent. The share of purchases accounted for by services imported from their affiliates fell, from 0.1 percent to 0.04 percent. The share of purchases accounted for by goods imported from affiliates increased for parents in “other” industries, from 0.7 percent to 1.4 percent. For parents in “other” industries, the share accounted for by imports of goods from unaffiliated parties and by imports of services from affiliates did not change. For parents in services, the share of purchases accounted for by services imported from their affiliates fell from 1.7 percent in 1994 to 0.7 percent in 2002.
Decomposition of Changes in Parent Employment To provide a starting point for a more formal analysis of, and a framework for, examining the impact of changes in offshoring behavior on employment at U.S. parent companies, the data reported by U.S. parent companies are used to apportion the actual change in U.S. parent employment among three factors: the change in output, the change in labor productivity, and the change in the use of purchased inputs in production. This framework allows changes in sourcing behavior, which are central to the issue of offshoring, to be viewed in isolation from other sources of change in employment and allows its importance relative to that of the other two factors to be gauged. U.S. parent employment can be expressed as an identity involving output, labor productivity, and the share of production performed in-house S(VA/S) E = —————— (VA/E) where E = employment, S = output (as measured by sales), and VA = value added. Output can be expressed as the sum of value added and purchased inputs, P, or, S = VA + P. Substituting this expression into the identity above yields S(1 – P/S) E = —————— (VA/E) Thus, employment is positively associated with changes in output and negatively associated with changes in the share of purchased inputs in output and with changes in labor productivity. Although each variable has determinants of its own, which would have to be investigated in seeking the root causes of
Maria Borga
151
Table 6. Decomposition of Changes in U.S. Parent Companies Employment between 1994 and 2002 Employment change
Share of total employment change
All U.S. parents Total change in employment Due to changes in output Due to changes in labor productivity Due to changes in the use of purchased inputs Residual
703.0 3,648.1 –1,436.6 –903.8 –604.7
5.19 –2.04 –1.29 –0.86
U.S. parents that lost employment Total change in employment Due to changes in output Due to changes in labor productivity Due to changes in the use of purchased inputs Residual
–1,972.3 –657.7 –825.6 –601.5 112.5
–0.33 –0.42 –0.31 0.06
U.S. parents that gained employment Total change in employment Due to changes in output Due to changes in labor productivity Due to changes in the use of purchased inputs Residual
2,675.3 4,305.8 –611.0 –302.3 –717.2
1.61 –0.23 –0.11 –0.27
changes in employment, they provide a framework for an initial examination of the changes in employment at U.S. parents. Table 6 attributes the change in parent employment to each variable, both for all parents and, separately, for parents that gained employment and parents that lost employment. In the decomposition, each variable is allowed in turn to change as it did between 1994 and 2002, while holding the other two variables constant at their 1994 levels. The residual is the portion of actual employment change not explained by the three calculations described above. The residual is due to the finite, rather than infinitesimal, changes in the reported data and can be viewed as reflecting the net interactions among the three variables. For all parents, employment increased by 703,040 employees. Increased output more than accounted for the increase in employment. Increases in labor productivity and in the use of purchased inputs had negative, and much smaller, impacts on parent employment. Parents that lost employees had a total decrease of 1,972,305 employees. The increase in labor productivity was the most important of the three variables, accounting for 42 percent of the total decline in employment, followed by the loss of output (33 percent) and by the increased
152
Brookings Trade Forum: 2005
use of purchased inputs (31 percent).11 Those parents that gained employees added 2,675,345 employees. The increase in their output more than accounted for the increased employment. The increase in labor productivity had a much larger negative impact on their employment than the increase in the use of purchased inputs. Overall, increased reliance on purchases by U.S. parents, from 60 percent to 63 percent of their sales, reduced their workforce by 903,800 employees between 1994 and 2002. Because most of the inputs purchased by parents are supplied by domestic firms, rather than imported, the loss of jobs due to the increased use of imports is less than that. Between 1994 and 2002, the share of purchases accounted for by imports of goods from affiliates increased about one percentage point, from 3.5 percent to 4.6 percent. Thus, the increased use of imports of goods from affiliates accounted for about one-third of the total increase in purchases. The share of unaffiliated imports of goods was flat, and the share of affiliated imports of services—small to begin with—fell slightly, so the use of these imports had little impact on the change in parent employment.
Correlations Table 7 shows the correlations between changes in selected variables. The change in parent employment was significantly positively correlated with the change in output, as measured by sales, and significantly negatively correlated with the change in labor productivity, as measured by value added per employee. The change in parent employment was negatively correlated with the change in the ratio of purchases to sales, but this correlation was not significant at the 10 percent level. However, a change in the share of goods imported from affiliates in U.S. parent purchases was significantly negatively correlated with a change in employment at the parent. Therefore, while a change in the use of purchased inputs overall was not significantly associated with a change in employment, there was a significant association between a change in parents’ use of goods imported from their foreign affiliates and its employment. In contrast, a change in the use of goods imported from unaffiliated parties was significantly positively correlated with employment change. Changes in the use of services imported from affiliates were not significantly correlated with changes in parent employment. 11. The residual term was positive and accounted for 6 percent of the entire change in employment.
Maria Borga
153
A change in the ratio of purchases to sales was significantly negatively correlated with a change in the use of goods imported from affiliates. This indicates that when parents increased their reliance on purchased inputs they tended to increase their reliance on inputs from outside the firm more than on goods purchased from their foreign affiliates. Conversely, it suggests that when parents reduced their reliance on purchased inputs, they continued to rely on imports of goods from their foreign affiliates and reduced their reliance on inputs from other sources. There was a significant positive correlation between changes in parent employment and changes in affiliate employment, reflecting a close link between growth at affiliates and growth at their parents. The growth in employment at both the parents and their affiliates could be due to mergers and acquisitions as well as to the expansion of existing operations. This finding suggests that a complementary relationship exists between a U.S. parent’s employment and its affiliates’ employment.12 The change in the share of affiliate employment in low-income countries was also positively correlated with the change in parent employment, suggesting that the expansion of operations in low-income countries is associated with gains, not losses, in employment at the parent. The change in affiliate employment was significantly positively correlated with the change in the share of sales to the United States and significantly negatively correlated with the share of sales sold locally. Changes in the share of affiliates’ sales to the local market were significantly negatively correlated with changes in their parents’ sourcing of inputs from them. Changes in compensation per employee at parents were highly negatively correlated with changes in parent employment, and highly positively correlated with changes in labor productivity, confirming that those parents with the greatest employment losses are those with the greatest compensation and productivity per employee increases. It is interesting to note that changes in the use of purchased inputs were significantly negatively correlated with changes in value added per employee. If firms were choosing to purchase inputs that they were relatively inefficient at producing, then an increase in purchased inputs would result in an increase in labor productivity. However, there was no evidence of this in the sample data set. There was a significant positive correlation between an increased share of imports of goods from affiliates in parent purchases and increased compensation and value added per employee, indicating that employees of firms that used inputs from their affiliates more intensively tended to 12. For a discussion of the complex issue of whether U.S. parent and foreign affiliate labor are substitutes or complements and a formal statistical analysis, see Brainard and Riker (1997).
154
Brookings Trade Forum: 2005
Table 7. Correlations between Changes in Selected Variables between 1994 and 2002 Change in:
Change in: Parent employment R&D intensity Real compensation per employeea Real value added per employeea Real salesa Share of services Ratio of purchases to sales Affiliate employment Local sales Sales to the United States Affiliates’ share of services Employment in low-income countries Affiliated goods imports Unaffiliated goods imports Affiliated services imports
Parent employployment
R&D intensity
1.00 0.02 (0.55)b –0.19** (0.0001) –0.08** (0.01) 0.23** (0.0001) –0.03 (0.33) –0.05 (0.12) 0.17** (0.0001) –0.01 (0.74) –0.01 (0.82) –0.02 (0.47) 0.06* (0.06) –0.12** (0.0001) 0.06** (0.04) –0.04 (0.23)
1.00 0.13** (0.0001) –0.05 (0.10) –0.01 (0.78) –0.004 (0.90) 0.02 (0.49) 0.01 (0.63) –0.08** (0.01) –0.001 (0.98) –0.01 (0.78) 0.04 (0.22) 0.11** (0.0003) 0.01 (0.65) –0.02 (0.58)
Real Real compen- value sation added per em- per employeea ployeea
1.00 0.43** (0.0001) –0.01 (0.72) 0.05 (0.12) –0.17** (0.0001) 0.04 (0.15) –0.07** (0.02) 0.03 (0.37) 0.03 (0.29) 0.02 (0.47) 0.08** (0.01) 0.02 (0.49) 0.09** (0.003)
Real salesa
Share of services
1.00 0.05* (0.08) 0.02 (0.60) –0.57** (0.0001) 0.02 (0.50) –0.05 (0.12) 0.01 (0.69) –0.03 (0.31) 0.05* (0.07) 0.09** (0.002) –0.01 (0.76) 0.01 (0.77)
1.00 –0.01 (0.69) 0.05 (0.13) 0.20** (0.0001) –0.06** (0.04) –0.01 (0.83) –0.04 (0.15) 0.03 (0.28) –0.01 (0.66) –0.02 (0.48) 0.01 (0.79)
1.00 –0.13** (0.0001) –0.03 (0.14) –0.001 (0.99) –0.01 (0.66) 0.53** (0.0001) –0.02 (0.42) –0.0001 (0.99) –0.01 (0.69) 0.03 (0.29)
* Significant at the 10 percent level. ** Significant at the 5 percent level. a. Rough estimates of U.S. parents’ sales, value added, and employee compensation in 2002 dollars were derived using the U.S. GDP deflator. b. Standard deviations are in parentheses.
become more productive and received higher compensation than employees of other firms. Finally, changes in the share of parent purchases accounted for by imports of services from affiliates were significantly positively correlated with changes in compensation per employee. Changes in R&D intensity were significantly positively correlated with changes in compensation per employee at parents but not with value added per employee at parents. Changes in R&D intensity were also significantly posi-
Maria Borga
Ratio of purchases to sales
Affiliate employment
155
Local sales
Sales to the United States
EmployAffiliment in Unaffiliates’ low- Affiliated ated Affiliated share of income goods goods services services countries imports imports imports
1.00 0.02 1.00 (0.61) –0.04 –0.10** 1.00 (0.14) (0.001) 0.07** 0.07** –0.48** 1.00 (0.02) (0.03) (0.0001) –0.08** –0.05* 0.02 0.01 (0.01) (0.07) (0.49) (0.77) 0.02 0.17** –0.04 0.11** (0.57) (0.0001) (0.18) (0.0003) –0.10** 0.07** –0.12** 0.15** (0.001) (0.01) (0.0001) (0.0001) –0.01 0.01 0.04 –0.02 (0.75) (0.81) (0.15) (0.58) –0.02 0.02 –0.13** –0.17** (0.50) (0.43) (0.0001) (0.0001)
1.00 0.01 1.00 (0.78) –0.004 0.01 (0.88) (0.79) –0.01 0.02 (0.71) (0.51) 0.07** –0.03 (0.02) (0.27)
1.00 0.03 (0.33) 0.05 (0.11)
1.00 0.01 (0.78)
1.00
tively correlated with changes in the share of parents’ purchases accounted for by affiliated imports of goods. This finding concurs with many studies of intrafirm trade that have found higher R&D intensity to be associated with greater intrafirm trade due to concerns about protecting intellectual property.13 13. See, for example, Lall (1978); Buckley and Pearce (1979); and Andersson and Fredriksson (2000).
156
Brookings Trade Forum: 2005
Conclusions A panel data set consisting of U.S. parent companies and their foreign affiliates was used to examine evidence on the extent of offshoring by U.S. parent companies to their foreign affiliates, and then to determine if this offshoring is associated with changes in employment at U.S. parents between 1994 and 2002. The data were used to apportion the changes in U.S. parents’ employment to three sources: changes in output, changes in labor productivity, and changes in their reliance on purchased inputs. This decomposition showed that changes in output and in labor productivity had relatively larger impacts on parents’ employment than changes in the use of purchased inputs. Correlations between changes in employment at U.S. parents and changes in various characteristics of the parents, their affiliates, and the MNC were examined. It was found that parents have increased their reliance on purchased inputs, but there was no significant correlation between changes in their reliance on purchased inputs and changes in the size of their workforce. Indeed, the most significant factors affecting parent employment were found to be changes in output and in labor productivity. While there was no significant association between changes in parents’ reliance on purchased inputs and changes in their employment, there was a significant negative correlation between changes in the share of parent purchases accounted for by imports of goods from their affiliates and changes in their employment. However, the vast majority of U.S. parents’ purchased inputs are acquired from domestic sources, not imports. It was found that the fortunes of parents and their foreign affiliates were closely linked, with changes in parents’ employment being positively correlated with changes in employment at their affiliates. Overall, the majority of foreign affiliate sales were local, indicating the importance of market access in the decision to invest abroad. The share of sales sold locally increased between 1994 and 2002, indicating that market access may be becoming an increasingly important reason for investing abroad. Finally, the increase in the share of local sales coincided with an increase in the share of employment in low-income countries, suggesting that market access, and not just factor cost differences, is an important reason for investing in these countries. One drawback of the approach used in the paper is that each factor has been examined individually. The correlations between changes in the variables measure the associations between any two of the variables. However, it is not possible to say with any certainty whether any of those associations found to be significant would still be if other variables were taken into account. Likewise, it is
Maria Borga
157
possible that variables examined above that showed no significant relationship would become significant if other variables were taken into account. Thus, an obvious avenue for future research would be to adopt a more rigorous statistical framework for analyzing the data that would allow more definitive conclusions to be drawn. Another avenue for future research would be to investigate the impact that industry-specific factors and mergers and acquisitions have had on the changes in U.S. parents’ employment. In addition it would be useful to link BEA’s data on direct investment with BEA’s data on imports of services from unaffiliated foreigners. Linking these two data sets would allow a more complete portrait of the direct imports by U.S. parents to be drawn.
Appendix: The Data Set BEA’s financial and operating data provide a picture of the overall activities of foreign affiliates and U.S. parent companies using a wide variety of indicators of their financial structure and operations. These data cover items that are needed in analyzing the characteristics, performance, and economic impact of MNCs, such as sales, value added, employment, compensation of employees, and exports and imports.
Definitions A U.S. parent is defined as a U.S. resident that has a 10 percent or more ownership interest in a foreign business enterprise, where U.S. resident is defined in the broad legal sense to include individuals, business enterprises, trusts, and other entities. However, most U.S. parents are businesses. A foreign affiliate is any foreign business enterprise in which there is a U.S. direct investment. Employment is defined as the number of full-time and part-time employees on the payroll at the end of the company’s fiscal year. Employee compensation consists of wages and salaries and employee benefits. Value added is measured as the sum of costs incurred and the profits earned in production. The costs incurred fall into four categories: compensation of employees, net interest paid, indirect business taxes, and capital consumption allowance. Sales, or gross operating revenues, are disaggregated into goods, services, and investment income. Sales of goods are those typical of establishments in any of the following NAICS sectors: agriculture, forestry, fishing, and hunting
158
Brookings Trade Forum: 2005
(except support activities for agriculture and forestry); mining (except support services); construction; manufacturing; wholesale trade; and retail trade. Sales of services are defined as those typical of establishments in the other fourteen NAICS sectors and the support activities for agriculture, forestry, fishing, hunting, and for mining.14 Information on investment income was collected to ensure that if it were included in sales or gross operating revenues it would not also be included in sales of services. Sales are also disaggregated into six destinations: the host country, the United States, third countries, and then further for each of these, into sales to either affiliated or unaffiliated parties. Expenditures for research and development (R&D) measure expenditures for R&D conducted by parents, whether for themselves or for others under contract. These expenditures exclude those for R&D conducted by others for the parents under contract.
Import Data The data on trade in goods collected on BEA surveys are generally comparable to the concepts and definitions used by the Census Bureau in compiling the data on U.S. trade in goods. Data on the imports of goods from both affiliated and unaffiliated parties are reported on forms covering the activities of the U.S. parent. In addition, trade data are reported on forms covering the activities of majority-owned foreign affiliates, including imports of goods to their U.S. parents. The data on parents’ imports from affiliated parties used in this paper are from the U.S. parents’ reports.15 BEA collects data on imports of services by U.S. parents from their foreign affiliates on its quarterly surveys of transactions between U.S. parents and their foreign affiliates. Because of the difficulties of matching data collected on those surveys with data on the 1994 benchmark and 2002 annual surveys (which were 14. For 1994, sales were divided into goods or services based on establishments using industry classifications derived from the 1987 Standard Industrial Classification system (SIC). For that year, sales of services were defined as those typical of establishments in the following SIC-based industry categories: services; finance (except depository institutions), insurance, and real estate; agriculture, mining, and petroleum services; and transportation, communication, and public utilities. 15. The only difference between the two data series is the inclusion of imports from minorityowned foreign affiliates in the data reported on the U.S. parents’ forms. These tend to be relatively small in value. For example, in 2002, imports from minority-owned foreign affiliates were estimated to be $11.7 billion while imports of goods from majority-owned foreign affiliates were estimated to be $171.6 billion.
Maria Borga
159
Table A-1. Imports of and Sales of Services to U.S. Parent Firms from Affiliates, 1994 and 2002 Millions of dollars
Total U.S. parent imports of other private services from their foreign affiliates Total sales of services to U.S. parents by majority-owned foreign affiliates Sales of services to U.S. parents by majority-owned foreign affiliates in panel data set
1994
2002
6,538
17,006
6,955
15,272
5,152
12,566
the source of the data used in this paper), data on imports of services from foreign affiliates were not used. Instead, the data on sales of services by majorityowned foreign affiliates to their U.S. parents from the annual and benchmark surveys were used. While there are some differences in definitions and coverage between the two data series, they are closely related.16 The first two rows of table A-1 compare the two data series, from BEA’s aggregate data, for the years 1994 and 2002. One concern about the composition of the panel used in this paper is that by excluding U.S. parents that entered the universe after 1994, the panel excluded parents that decided to invest abroad because of advances in technology (for example, better and cheaper telecommunications) that allow them to rely on services produced at their affiliates. To check this possibility, the last row of table A-1 shows the sales of services to U.S. parents by majority-owned foreign affiliates included in the panel data set. Comparing rows 2 and 3, the share of sales of services to U.S. parents accounted for by foreign affiliates included in the panel increased from 74 percent in 1994 to 82 percent in 2002. Thus, the panel does not appear to be missing large numbers of new entrants that rely to a greater extent on services from their affiliates than existing firms. Comparison of the sample data set with the population of firms. Differences due to sample selection can be identified by comparing the aggregate values for the sample data set with the aggregate values for the population of U.S. direct
16. An example of a difference in coverage is that the data on imports from affiliates include all foreign affiliates while those on sales of services include only majority-owned foreign affiliates. An example of a difference in definition is that the data on sales of services include settlement transactions between affiliated telecommunications carriers, but these transactions are excluded from the affiliated imports of services data.
160
Brookings Trade Forum: 2005
Table A-2. Aggregate Values of Selected Characteristics for the Panel Data Set and the Population, 1994 and 2002 1994 Characteristics U.S. parent variables Employment (in thousands) Sales (millions of dollars)b Employment change (percent) Ratio of purchases to sales Share of services in sales International share of sales R&D intensity Value added per employeeb (thousands of dollars) Compensation per employeeb (thousands of dollars) Affiliate variables Employment (in thousands) Share of services in sales Share of sales to local market Share of sales to third countries Share of sales to U.S. MNC variables Parent share of employment Parent share of sales Parent share of value added Share of affiliate employment in low-income countries Affiliated goods imports as a share of parent purchases Unaffiliated goods imports as a share of parent purchases Affiliated services imports as a share of parent purchases
2002 Sample
Population a
15,059.9 3,885,676 … 0.65 0.30 0.13 0.026
11,118.3 3,563,318 0.07 0.70 0.31 0.13 0.028
17,531.0 5,130,846 0.16 0.69 0.38 0.11 0.026
89.56
91.38
97.00
91.66
56.80
56.19
61.51
57.65
3,894.9 0.13 0.67 0.23 0.10
5,153.3 0.13 0.67 0.23 0.10
5,104.9 0.16 0.62 0.27 0.11
7,138.2 0.18 0.62 0.27 0.11
0.73 0.70 0.73
0.75 0.71 0.75
0.69 0.65 0.70
0.71 0.69 0.74
0.34
0.35
0.40
0.39
0.059
0.047
0.056
0.046
0.038
0.037
0.033
0.033
0.003
0.003
0.004
0.004
Sample
Population
10,415.3 2,656,980 … 0.65 0.28 0.11 0.031
a
a. Aggregate values shown exclude U.S. parents in wholesale and retail trade because those parents were excluded from the sample data set. b. Rough estimates of U.S. parents’ sales, value added, and employee compensation in 2002 dollars were derived using the U.S. GDP deflator.
investment abroad. Table A-2 shows the values of selected variables for the sample data set and the population of US MNCs. The percentage change in aggregate employment of U.S. parents for the sample was about half that for the population as a whole, partly because the employment of new entrants to the population exceeded that of those parents that
Maria Borga
161
Table A-3. Aggregate Values of Selected Characteristics for MNC Deaths for 1994 and Births for 2002 Deaths (values in 1994)
Births (values in 2002)
3,740.5 3,740,504 0.63 0.31 0.10 0.020 76.61 45.32
5,913.9 1,355,504 0.70 0.41 0.06 0.020 68.79 43.18
Affiliate variables Employment (in thousands) Services as a share of sales Share of sales to local market Share of sales to third countries Share of sales to the United States
901.8 0.11 0.70 0.21 0.09
1,469.0 0.28 0.68 0.23 0.09
MNC variables Parent share of employment Parent share of sales Parent share of value added Share of affiliate employment in low-income countries Affiliated goods imports as a share of parent purchases Unaffiliated goods imports as a share of parent purchases Affiliated services imports as a share of parent purchases
0.81 0.74 0.79 0.25 0.030 0.037 0.003
0.80 0.82 0.86 0.34 0.023 0.052 0.002
Characteristics U.S. parent variables Employment (in thousands) Salesa (millions of dollars) Ratio of purchases to sales Services as a share of sales International share of sales R&D intensity Value added per employeea (thousands of dollars) Compensation per employeea (thousands of dollars)
a. Rough estimates of U.S. parents’ sales, value added, and employee compensation in 2002 dollars were derived using the U.S. GDP deflator.
dropped out. Table A-3 presents the aggregate values for 1994 for deaths and the aggregate values for 2002 for births.17 One issue that arises with a data set consisting only of parents that responded to both the 1994 and 2002 surveys of U.S. direct investment abroad is that parents that dropped out of the universe (deaths) or entered it (births) between 1994 and 2002 are excluded. The exclusion of deaths raises the possibility of survivor bias. However, it must be emphasized that the death of a parent due to the liquidation of the firm is rare. The more 17. Data from BEA’s 2002 preliminary estimates were used in constructing the sample data set. Some parents’ responses were still being edited at the time the panel data set was assembled, and these parents were dropped from the sample even though they were in the universe in both 1994 and 2002. Therefore the aggregates from the sample plus the aggregates from deaths or births do not equal the population aggregates.
162
Brookings Trade Forum: 2005
common causes are that the parent is acquired by another U.S. parent company (in which case the first parent is included in the data but is now consolidated with the acquiring company); the parent is acquired by a foreign company that chooses to hold the foreign affiliates directly; the parent sells off its foreign operations; or, the foreign operations of the parent fall below BEA’s reporting requirements. For most of the other parent variables, the sample and the population characteristics are similar. One difference between the sample and the population is the share of parents’ sales accounted for by services, which is lower for the sample in 1994 and 2002, indicating an underrepresentation of services firms. The affiliates of parents included in the sample account for 76 percent of all affiliate employment in 1994, falling to 72 percent by 2002. The sample and the population are similar for most of the affiliate and MNC characteristics. One difference is that imports of goods from foreign affiliates account for a larger share of purchased inputs for parents in the sample than for the population, indicating an overrepresentation in the sample of parents that source inputs from their affiliates. Comparison of the aggregate and mean values for the panel data set. Comparing the aggregate values for the sample in table A-2 and the mean values from the sample in table 1 identifies differences that are due to the greater weight given to larger firms in the aggregate values. The unweighted data show much higher average employment growth, at 39 percent, than the aggregate value because the employment of smaller parents tended to grow faster than that of larger firms. In addition, smaller firms used purchased inputs less intensively than larger firms. The foreign affiliates of smaller parents tend to focus more on selling to the local market and less on exporting from the host country, and this tendency increased between 1994 and 2002. For the MNC as a whole, the data indicate that smaller parents have less extensive foreign operations than larger parents, and these operations tend to be more concentrated in high-income countries. In addition, smaller parents use imports of goods less intensively than larger firms.
Maria Borga
163
References Andersson, Thomas, and Torbjorn Fredriksson. 2000. “Distinction between Intermediate and Finished Products in Intra-firm Trade.” International Journal of Industrial Organization 18 (July): 773–92. Brainard, Lael S., and David A. Riker. 1997. “Are U.S. Multinationals Exporting U.S. Jobs?” Working Paper 5958. Cambridge, Mass.: National Bureau of Economic Research. Buckley, Peter J., and R. D. Pearce. 1979. “Overseas Production and Exporting by the World’s Largest Enterprises: A Study in Sourcing Policy.” Journal of International Business Studies 10 (March): 9–20. Government Accountability Office (GAO). 2004. “Current Government Data Provide Limited Insight into Offshoring of Services.” Washington: GAO-04-932 International Trade. Helpman, Elhanan, and Paul R. Krugman. 1985. Market Structure and Foreign Trade. Cambridge, Mass.: MIT Press. Lall, Sanjaya. 1978. “The Pattern of Intra-firm Exports by U.S. Multinationals.” Oxford Bulletin of Economics and Statistics 40 (3): 209–22. Markusen, James R. 2002. Multinational Firms and the Theory of International Trade. Cambridge, Mass.: MIT Press U.S. Bureau of Economic Analysis. 1998. U.S. Direct Investment Abroad: 1994 Benchmark Survey, Final Results, U.S. Department of Commerce. ———. 2004. “A Note on the Patterns of Production and Employment by U.S. Multinational Companies.” Survey of Current Business (March): 52–56. ———. 2005. U.S. Direct Investment Abroad: Operations of U.S. Parent Companies and Their Foreign Affiliates, Preliminary 2002 Estimates. U.S. Department of Commerce.
This page intentionally left blank
D E S I R É E VA N W E L S U M X AV I E R R E I F Organization for Economic Cooperation and Development
Potential Offshoring: Evidence from Selected OECD Countries
R
apid advances in information and communication technologies (ICTs), combined with continuing efforts to liberalize international trade and investment in services, have increased the tradability of services and created new types of tradable services. This, in turn, has led to a new wave of globalization in the services sector, with the offshoring of particular types of services activities now becoming increasingly common, as it has been for many years in manufacturing. New technological developments now allow many service activities to be carried out regardless of their geographic location, and their production and delivery no longer have to take place in the same location. As a result, many white-collar jobs that were shielded from international competition now face competition from abroad. Despite the widespread media attention given to the apparent offshoring of service sector jobs, little is known about the extent of this phenomenon, or the extent to which it is related to other economic and structural developments. This paper draws on and extends a previous detailed analysis of occupational data for
The opinions expressed and arguments employed in this paper do not necessarily reflect the official views of the organization or of the governments of its member countries. This paper draws on a larger body of work and a longer paper entitled “The Share of Employment Potentially Affected by Offshoring—An Empirical Investigation,” which will be published by the OECD (www. oecd.org/sti/offshoring). We are grateful to Nigel Pain of the OECD Economics Department and Ron Smith of Birkbeck College at the University of London for their help and advice in preparing this paper, as well as our colleagues in the Directorate for Science, Technology and Industry, in particular Graham Vickery; in the Directorate for Employment, Labour, and Social Affairs; and in the Trade Directorate. Finally, our thanks go to participants in the conferences at which earlier parts of this work have been presented, in particular Catherine Mann and Robert Lawrence.
165
166
Brookings Trade Forum: 2005
selected OECD countries that sought to determine the share of total employment that could potentially be affected by the international sourcing of IT and ICTenabled services (van Welsum and Vickery 2005a). Including both the low- and the high-skill white-collar occupations potentially affected by global services sourcing, that analysis suggested that close to 20 percent of total employment could potentially be affected by offshoring. The work also found that sectors such as business services (for example, accounting and consulting), financial services, and research and development have a relatively high share of such employment. It is important to keep in mind that “potentially affected by offshoring” refers to activities that could be coming into a country as well as those that might leave a country. Incoming offshored services activities would bring about an increase in the share of employment potentially affected by offshoring, whereas services activities that leave a country would bring about a relative decline in the share. This paper takes this analysis one step further by examining the relationship between the share of employment potentially affected by offshoring and other economic and structural developments, using some simple descriptive regressions on a panel of OECD economies between 1996 and 2003. In particular, first estimates are provided of the statistical association between the share of employment potentially affected by service sector offshoring, trade in business services, and foreign direct investment.The analysis in this paper does not find any systematic evidence to support the popular belief that net outward investment or imports of business services are associated with significant declines in the share of employment potentially affected by offshoring, at least at the aggregate level. Exports of business services are found to have a positive statistical association with the share of employment potentially affected by offshoring, suggesting that increases in demand and production have also raised demand for these types of ICT-using occupations. Other factors positively associated with the share of employment potentially affected by offshoring are found to be the comparative size of the service sector, the growing share of ICT investment in total fixed investment, and human capital. Although there are no direct official data measuring the extent of offshoring, it is commonly believed that it has the potential to grow substantially beyond its current relatively small level. This paper aims to contribute to the debate surrounding offshoring by looking in detail at some of the trade in services and employment data that may reveal further insights about its current extent, as well as by performing a simple descriptive econometric analysis of the factors statistically associated with movements in the aggregate share of employment that could potentially be affected by offshoring.
Desirée van Welsum and Xavier Reif
167
The paper is organized as follows. The next section looks at what can be learned about offshoring on the basis of data on trade in services and occupational employment data, respectively. The subsequent section sets out the simple descriptive empirical model employed to examine what factors are associated with the share of employment potentially affected by offshoring and gives some initial results. The final section offers concluding remarks.
The Extent of Potential Offshoring Offshoring includes both international outsourcing (where activities are contracted out to independent third parties abroad) and international insourcing (to foreign affiliates). The cross-border aspect is the distinguishing feature of offshoring—that is, whether services are sourced within the domestic economy or abroad—not whether they are sourced from within the same company or from external suppliers (outsourcing). Offshoring is often confused with outsourcing, but only a part of offshoring consists of outsourcing. Offshoring is also often interpreted as referring to the purchase of intermediate services, even though the distinction between final and intermediate services is a difficult one to make in some instances and may not be very meaningful for certain types of services. To date there are no official data measuring the extent of offshoring, so it is necessary to use indirect measures such as data on trade in services, employment data, input-output tables, and trade in intermediates. Evidence from company surveys can also be a useful complement.1 However, while the offshoring of services activities should result in a flow of trade in services, not all trade in services is related to offshoring, and it is not possible to distinguish which part of it is. Similar problems apply to the analysis of foreign direct investment (FDI) because it is not possible to determine what share of FDI is directly related to offshoring. There are also no official and internationally comparable data on changes in employment resulting from the offshoring of services activities. This paper uses both trade and employment data to examine what is known about the extent of services offshoring and what can be said about its potential. Trade Data Trade in services provides one possible proxy for measuring ICT-enabled services offshoring. Because offshoring of activities takes place between residents of 1. See, for example, Marin (2004).
168
Brookings Trade Forum: 2005
different countries, some of it should result in a flow of trade in services, exports from the country receiving the offshored activities, and imports for the country from which the offshoring originates. OECD (2004a) and van Welsum and Vickery (2005a) examine exports of services, while Schultze (2004) and van Welsum (2004) analyze offshoring and imports of services. It is important to bear in mind, though, that not all trade in services is related to offshoring, and it is not possible to identify the share of trade in services that is directly related to offshoring. The extent of international trade in IT and ICT-enabled business process services in international statistics is approximated by summing the IMF balance of payments categories “computer and information services” and “other business services” (see table A-1 for details on which services are included in these categories). These data contain information on international outsourcing and international insourcing combined, but it is not possible to identify the proportion of this trade that results directly from offshoring. Data on computer and information services are not available for all countries. For some, such as India, they are included under “other business services,” along with other services.2 The “other business services” category may have variable shares of IT and ICT-enabled services in different countries. Moreover, the data are reported in current U.S. dollars and will be affected by currency movements. Most exports of other business services and computer and information services still originate in OECD countries, although their share is slowly declining (from 80.3 percent of the total reported value in 1995 to 79.1 percent in 2002).3 The twenty countries that accounted for the largest-value shares in 2003, as well as some selected other economies, are shown in figure 1. OECD countries have the top seven shares of these services exports; China is in tenth and India is in fourteenth position. Nevertheless, some nonmember developing economies are experiencing rapid growth in exports (see figure 2), although most are starting from very low levels. Only Ireland is among the ten countries with the largest shares (in 2003) and the fastest export growth rates (1995–2003). The average annual growth rate of exports and imports of other business and computer and information services (in current U.S. dollars) over the period 2. For India, the category “other business services” includes all services except travel, transport, and government services. However, Indian firms are now extensively exporting ICT-enabled services and business process services, and the remaining services included in the category are likely to be small in comparison. Furthermore, data on overseas revenues from annual reports of top Indian export firms show patterns similar to the IMF data. 3. The share of some services-exporting countries may be understated because the data on trade in services they report to the IMF are not very accurate. Furthermore, other countries that export services may not be members and report to the IMF.
Desirée van Welsum and Xavier Reif
169
1995–2003 is given in figure 2 (the top twenty of selected countries). It shows that many of the countries often mentioned in the offshoring debate (such as India, China, and Brazil, but also eastern European countries such as Estonia) have indeed experienced rapid growth of these exports, which may confirm their emergence as offshoring locations in recent years. However, some of these countries are growing from a very low level, and some of the rapid growth is explained by their economic development. Furthermore, many countries experiencing rapid exports growth are also seeing rapid growth of their imports of these services (fifteen countries are among the twenty countries with the fastest growth of both exports and imports of these services). The trade balance (in current U.S. dollars) in the sum of the categories “other business” and “computer and information services” as a percentage of GDP is shown in figure 3. The balance share, though positive, is smaller and more stable in the United States than in the United Kingdom and Denmark. The share in the United Kingdom is increasing, in spite of the impression that may be given by the many reports, media reports in particular, on the extent of offshoring and related imports. Surprisingly, the deficit in Ireland is quite large. It remains difficult, however, to interpret these data and link them to different sourcing activities. It is not possible to tell what share of this trade results from international sourcing activities. Offshoring can include unaffiliated trade in services (from international outsourcing) and affiliated trade (from international insourcing), but some of it is also related to foreign direct investment and temporary migration, mode 4 trade in services (the temporary movement of natural persons) under the General Agreement on Trade in Services (GATS). But temporary migration is not captured by balance of payments trade data.4 Furthermore, the quality of the data may vary, and there can be very large discrepancies between reported exports and imports.5 Employment Data To get an idea of the “outer limits” of employment potentially affected by offshoring, van Welsum and Vickery (2005a) calculate the share of people employed who are performing mainly the types of functions that could potentially be carried out anywhere, using data on employment by occupation by industry. The classifications were not harmonized internationally, but the same 4. See van Welsum (2003) for a discussion. 5. See OECD (2004a, chap. 2) for an example using Indian data.
Figure 1. Share of the Value of Reported Total Exports of Other Business Services and Computer and Information Services, Selected Countries, 1995 and 2003a Percent in decreasing order of the total reported value share in 2003 1995
2002
14 12 10 8 6 4
Finland
Australia
Brazil
Korea
Singapore
Sweden
Canada
India
Austria
Spain
China
Japan
Ireland
Italy
Netherlands
France
Germany
United Kingdom
United States
2
Source: Authors’ calculations based on IMF balance of payments database (August 2005). a. The reported total for all countries does not necessarily correspond to a world total. For some countries, such as India, it is not possible to isolate other business services and computer and information services. As a consequence, for India the category includes total services, minus travel, transport, and government services (that is, including construction, insurance, and financial services as well as other business services and computer and information services).
Desirée van Welsum and Xavier Reif
171
Figure 2. Average Annual Growth in the Value of Exports and Imports of Other Business and Computer and Information Services, Selected Countries, 1995–2003a Compound annual growth rate Latvia Croatia Ireland Argentina Romania Lithuania Estonia India Peru Sweden China Brazil Israel Colombia Spain United Kingdom Morocco Norway Iceland Denmark 0
10
20
30 Exports
40
50
60
0
10
20
30 Imports
40
50
60
Latvia Estonia Lithuania Ireland Sweden Iceland Cyprus Switzerland Croatia India Brazil Denmark Ghana Morocco Spain United States Turkey Romania United Kingdom Israel
Source: Authors’ calculations based on IMF balance of payments database (August 2005). a. OECD member countries in dark shading. Data for Belgium, Luxembourg, and Mexico are not included.
Figure 3. Trade Balance in the Categories “Other Business Services” and “Computer and Information Services” as a Percentage Of GDP, Selected Countries, Various Years Percent 4.0 1995
2000
2001
2002
2003
3.0 2.0 1.0 0.0 –1.0 –2.0 –3.0 –4.0 –5.0
Source: Authors’ calculations based on IMF balance of payments database (February 2005).
ed
en
l ga rtu
s
ate
Sw
he et
Po
rla
nd
ur bo m N
xe Lu
s
g
ly Ita
nd la Ire
e ec re
an
d
ite
Un
n Ki
G
ce
d
om
gd
Fr
n
an nl Fi
ai
m
Sp
k ar
y en D
G
er
m
na
an
da
m Ca
iu lg
tri
a Be
us A
A
us
tra
lia
–6.0
Un
i
ted
St
Desirée van Welsum and Xavier Reif
173
methodology and rationale were applied to the individual country data sources.6 Because this analysis was carried out in order to obtain an order of magnitude on the share of people employed performing tasks that could potentially be carried out anywhere, no additional assumptions were made as to what proportion of each occupational group was actually likely to be affected by offshoring in practice. Thus the whole of each selected occupation was then included in the calculations. Occupations were selected by examining detailed occupational and task descriptions on the basis of the following four criteria, or “offshorability attributes”: (1) intensive use of ICTs, (2) an output that can be traded/transmitted enabled by ICTs, (3) high codifiable knowledge content, and (4) no face-to-face contact requirements. The occupational selections that resulted from this exercise are reported in the appendix (tables A-2 through A-5).7 This analysis, using occupational data for several OECD countries, suggests that around 20 percent of total employment carries out the kinds of functions that are potentially geographically footloose as a result of rapid technological advances in ICTs and the increased tradability of services, and could therefore potentially be affected by international sourcing of IT and ICT-enabled services.8 Nevertheless, because 6. The European data are Labour Force Survey data provided by Eurostat. The occupational classification system in those data is the ISCO, the International Standard Classification of Occupations; NACE, the industrial classification system of the European Union, is used for sectoral classification. For the United States, data from the Current Population Survey were used. The Current Population Survey collects information on both the industry and the occupation of the employed and unemployed. However, beginning with data from January 2003, the 1990 Census Industrial Classification System was replaced by one based on the North American Industry Classification (NAICS), and the 1990 Census Occupational Classification was replaced by one derived from the U.S. Standard Occupational Classification (SOC). Further information is available on the website of the U.S. Bureau of Labor Statistics at www.bls.gov/opub/hom/pdf/homch1.pdf [November 2004]: Chapter 1: Labor Force Data derived from the Current Population Survey. For Canada, labor force data provided by Statistics Canada were used. The occupational classification is in SOC91. For Australia, data from the Labour Force Survey provided by the Australian Bureau of Statistics were used. The occupational classification is in Australian Standard Classification of Occupations (ASCO), second edition. 7. For further details on the methodological background see van Welsum and Vickery (2005a), van Welsum and Vickery (2005b) and OECD (2004a). 8. Other studies have taken a similar approach. For example, Bardhan and Kroll (2003) produced estimates of 11 percent of total employment in the United States in 2001 as potentially affected by offshoring, and Forrester Research, as reported by Kirkegaard (2004), up to 44 percent of total employment. The differences in these estimates can be explained by the selection criteria that are applied to the occupational data. Thus Bardhan and Kroll (2003) only included occupations in which at least some offshoring was already known to have taken place yielding a more conservative estimate of the share of employment potentially affected, whereas the Forrester study used less detailed occupational categories resulting in a larger estimate of jobs potentially affected.
174
Brookings Trade Forum: 2005
Figure 4. ICT-Intensive Using Occupations Potentially Affected by Offshoring as a Share of Total Employment: EU15, USA, Canada, and Australia, 1995–2003a Percent
19.5 19.0 18.5 18.0 17.5 17.0 EU15
USA
Canada
Australia
16.5
1995
1996
1997
1998
1999
2000
2001
2002
2003
Source: Authors’ calculations and van Welsum and Vickery (2005a), based on EULFS, U.S. Current Population Survey, Statistics Canada, and Australian Bureau of Statistics (2004/5). a. Includes estimates where a full data set was not available. Because of classification changes, the number for the United States for 2003 is also an estimate. There is a break in the data for Australia, with data for 1995 and 1996 in the ASCO first edition and subsequent data in ASCO second edition. Because of differences in classifications, the levels are not directly comparable.
classifications are not harmonized internationally, the country estimates are not directly comparable. The evolution over time of the share of employment potentially affected by offshoring is illustrated in figure 4. Even though the levels of these shares are not directly comparable, the evolution of the trends is interesting. The share of occupations potentially affected by offshoring in the EU15 increased from 17.1 percent in 1995 to 19.2 percent in 2003. For Canada it was more or less flat around 19.5 percent until 2001, after which it declined to 18.6 by 2003. For the United States the share declined by more than a percentage point from 19.2 percent in 1995 to 18.1 percent in 2002.9 In Australia, the share increased between 1997 and 2001, except in 1999, when it started to decline. Data for 1995 and 1996 were generated by a different classification system and therefore are not directly comparable. 9. The number for 2003 (just under 18 percent) is an estimate since both the occupational and industrial classification systems were changed in 2003 in the United States.
Desirée van Welsum and Xavier Reif
175
While it is difficult to draw inferences from these trends without further analysis, since the trends are affected by a multitude of factors, the evolutions shown in these trends are consistent with some casual observations on the ICTenabled offshoring that is taking place, such as Canada serving as an offshoring location, mainly from the United States, but less so more recently as other locations, such as India, have started to emerge. Similarly, Australia possibly also experienced competition for attracting, or keeping, activities that can be sourced internationally from India and other emerging locations in the region. Thus, the declining share in the United States, Canada, and Australia toward the end of the period could be consistent with the offshoring of IT-related and back-office activities (with some “potential offshoring” having become “actual offshoring”), for example, even though this is unlikely to account for all of the decline. Another possible explanation could be a differential pace of technological change with a relatively more rapid adoption and integration of new technologies, leading to relatively more jobs disappearing sooner as they become automated or digitized, or both.10 The increasing share for Europe is compatible with an overall increase in services employment as well as the finding from surveys that European firms tend to offshore within Europe.11 (At least one EU country, Ireland, is also a major destination for offshoring activities from the United States—IT-related activities in particular.) Other factors could also be important, such as cyclical developments and changes in labor supply and labor quality. The offshoring phenomenon does not necessarily have to result in a decline in services employment, though. Many existing services sectors have expanded, new services have emerged, and with ongoing technological developments and services trade liberalization it is likely yet more will be created. Furthermore, with the elasticity of demand of internationally traded services greater than one,12 rapid growth in countries such as India and China should also lead to reinforced exports from OECD countries. The offshoring phenomenon itself will also create new jobs in the domestic economy. However, certain types of occupations may experience slower growth than they otherwise might have. 10. A parallel can be drawn here with some of the work undertaken by Autor, Levy, and Murnane (2003) and Levy and Murnane (2004). These authors argue that the tasks most vulnerable to being replaced by technology are those where information processing can be described in rules. If a significant part of a task can be described by rules, this increases the likelihood of the task’s being offshored, since the task can then be assigned to offshore producers with less risk and greater ease of supervision. 11. See Millar (2002) and Marin (2004), for example. 12. See, for example, Pain and van Welsum (2004); van Welsum (2004); and Mann (2004).
176
Brookings Trade Forum: 2005
There are several possible explanations for the changes in these trends, which are expressed in shares (see figure 4). For example, a decline in the share could be explained by an absolute decline in the number of people employed in the categories identified as potentially affected by offshoring. Alternatively, this selection of occupations could be growing at a slower pace than total employment. The relatively slower growth of employment potentially affected by offshoring is in fact what explains most of the declines observed in the trends, except for the United States, where the absolute number of people employed in the categories identified as potentially affected by offshoring has declined. These observations would therefore tend to support the idea that offshoring may lead to slower growth of employment in occupations potentially affected by offshoring and not necessarily to actual declines in employment. UNDERLYING TRENDS. This section describes the employment data underlying figure 4 in more detail, and the detailed graphs for the EU15 countries are given in figures A-1 and A-2. One caveat about these data is that there may be differences in the ICT content of occupations within and between countries. Similarly, any possible dynamic adjustments or changes in qualifications, skill requirements, and task descriptions that may take place within occupations over time are not taken into account. For the EU15 as a whole, the trend increases in all years, except in 1998. The year-by-year rate of change shows that employment potentially affected by offshoring grew faster than total employment in EU15 in all years except 1998, when total employment grew faster than offshoring. There was no absolute decline in employment potentially affected by offshoring. Figures A-1 and A-2 show the underlying data for the countries that make up the EU15. For Greece and Portugal the data quality is poor, especially early in the sample period. Furthermore, there appears to be a break in the data for Ireland between the 1995 and 1997 period and between the 1999 and 2003 period, with a missing data point in 1998. Nevertheless, the increase observed for the EU15 aggregate appears to be representative of the evolution in most of the underlying countries and does not appear to be driven by outliers. Most countries see an increase in the share of employment potentially affected by offshoring, and while there were some incidences of absolute decline at the individual country level, these were isolated and do not represent a trend. For the United States there was a downward trend in the period 1995–98 and 2001–03 (although 2003 is an estimate since both the industry and occupational classifications were changed that year). The year-by-year rate of change shows that total employment grew faster than employment potentially affected by offshoring in all years except 1999 and 2000. The absolute number in employment
Desirée van Welsum and Xavier Reif
177
potentially affected by offshoring declined in the United States in 1996 and in 2001, 2002, and 2003. This absolute decline was fairly generalized and not limited to a specific type of occupation or level of skills. All of the occupations selected as potentially affected by offshoring experienced at least one annual period of decline. Furthermore, forty-five of the sixty-seven occupations included in the U.S. selection experienced an absolute decline between 2001 and 2002, as did the overall selection of occupations potentially affected by offshoring and total employment. Finally, forty-seven of the selected occupations experienced at least three absolute declines between 1995 and 2002. For Canada the trend fell in 1995–96 and 1998–2003, except in 2000. The annual rate of change shows that total employment grew faster than employment potentially affected by offshoring except in 1997, 1998, and 2000. There was no absolute decline in employment potentially affected by offshoring. Finally, for Australia the trend fell in 1999 and 2001–03. The annual rate of change shows that total employment grew faster than employment potentially affected by offshoring in 1999 and 2002–03. There was no absolute decline in employment potentially affected by offshoring. Data for 2004 indicate that the trend continues to decline. Data for 1995 and 1996 are not directly comparable with those for the rest of the period because 1995 and 1996 are in the first edition of the Australian Standard Classification of Occupations (ASCO) and subsequent data in the ASCO second edition. These observations support the idea that it is not so much a decline in certain types of employment that can be expected, but rather slower employment growth in these types of occupations. Even though technology may account for at least some of the relative decline in the occupations potentially affected by offshoring (and for absolute declines in the case of the United States), the possibility that some of these jobs have been offshored cannot be ruled out. For example, Baily and Lawrence (2005) argue that at least some of the declines in low-wage ICTenabled occupations, a concept close but not equivalent to the group of clerical workers identified in the occupational selections of tables A-2 through A-5, took place as a result of activities being shifted overseas. Looking at IT specialist occupations, they also find that the net loss of computer programmers in the United States was most likely the result of offshoring. Nevertheless, even the largest projections of jobs to be offshored, as often reported in the media, are relatively small in comparison with annual job churning in OECD labor markets.13 Having examined some of the underlying trade and employment data, the next section presents a simple descriptive empirical model to provide a first indication 13. See OECD (2004b).
178
Brookings Trade Forum: 2005
of the factors associated with aggregate changes in the share of potentially offshorable employment.
The Empirical Model and Some Results Using panel data estimation techniques, this paper attempts to identify those factors that are associated with the share of employment potentially affected by offshoring in total employment for the United States, Canada, Australia, and the EU15 countries (except Greece, Ireland, Luxembourg, and Portugal)14 over the period 1995–2003. The Model In the model, potentially offshorable employment as a share of total employment (OL) is a function of trade, investment, the industrial structure of the economy, a technology adoption/integration variable, a product market regulations indicator, an employment protection indicator, and human capital.15 The choice of variables is motivated by findings from a vast background literature, including studies of the factors determining the overall share of the service sector in the economy, studies of services sector employment, and studies of the effect of trade and technology on employment.16 Ideally, it would be appropriate to begin with a simple structural model of the factors affecting the relative demand for ICT-using occupations. Using the firstorder marginal productivity conditions from an (unknown) production function with two types of labor (ICT and non-ICT labour), such a model might be 14. These countries were excluded from the sample because of a lack of data. 15. Even though GDP per capita is a variable found to be an important determinant of the share of services sector employment (Messina 2004), it is not used here. In a time series context it does not make sense to include the level of GDP per capita in a regression of a bounded variable. The first difference in GDP per capita was found to be insignificant. This is not necessarily surprising since the countries in the sample all have relatively high levels of GDP per capita, so over the sample period (1995–2003) this variable is not found to have an impact on the share of employment potentially affected by offshoring. Nevertheless, with the exception of Austria, the countries with a relatively low share of employment potentially affected by offshoring were also those with the lowest levels of GDP per capita. The role of productivity growth is also not considered here. It is sometimes argued that the decline in certain types of employment, or the lack of new jobs (the jobless recovery), is the result of important productivity increases, but Baily and Lawrence (2004) argue that this is a mistake and that while productivity may have played some role, it should not be considered a fundamental cause. Time dummies pick up common cyclical effects. 16. See van Welsum and Reif (2005) for a review.
Desirée van Welsum and Xavier Reif
179
expected to include measures of the relative output and relative wages of ICTusing occupations. Control variables might also be included to pick up possible differences in the extent of (labor-augmenting) technical progress in the two broad types of occupations. As in the literature on the demand for skilled and unskilled labor, possible controls are indicators for both trade and technology. Unfortunately, although it is possible to control for output and technology effects directly, data on occupational wages are not readily available in most countries at the level of detail required. Their effect can be captured only indirectly by including variables that can be expected to have an influence on real wages. It should be noted that although it is not possible to estimate a full structural model, the estimates we show are not a pure reduced-form model either, since potentially endogenous current dated terms in output or in trade and technology (or both) remain in the model. OL = f(TRADE, FDI, STRUC, ICT, PMR, union, HK)
(1)
In particular, trade effects are approximated by including both imports and exports of other business and computer and information services as a share of GDP (current U.S. dollars, IMF balance of payments for trade data, OECD ANA database for GDP data). It is expected from the literature on trade-related displacement that imports may have a negative association with the share of employment potentially affected by offshoring, while exports are thought to have a positive relationship. Nevertheless, trade may not have an impact at the aggregate level but rather bring about shifts at the industry and occupation level.17 Net foreign direct investment is included as a share of GDP (current U.S. dollars, IMF balance of payments for stock data, and OECD ANA database for GDP data).18 The predictions from the literature are ambiguous about the direction of the relationship between these variables and the share of employment potentially affected by offshoring. Differential effects might be expected to occur for FDI in services and in manufacturing (similar to the way the relationship between trade and FDI depends on the level of aggregation),19 but such differences are hidden in the aggregate measures; only the net effect, which will be dominated by manufacturing FDI, can be picked up because much of the total inward and outward FDI stocks is in manufacturing, and there are relatively few detailed cross-country data that distinguish manufacturing from services FDI 17. See OECD (2005) for an overview. 18. This is done by imposing equal and opposite signs on outward and inward FDI, a restriction accepted by the data. 19. See Pain and van Welsum (2004) and van Welsum (2004).
180
Brookings Trade Forum: 2005
over a long time period. However, in further research it will be attempted to include separate indicators for services and manufacturing FDI. The share of services sector20 value added in total value added and the share of high-tech industries21 value added in total value added are included as indicators of the industrial structure of the economy (OECD STAN database; missing values have been estimated using the “60-Industry Database” from the Groningen Growth and Development Centre of the University of Groningen (Netherlands), available at www.ggdc.net/dseries/60-industry.html [April 28, 2005]). Other things being equal, the larger the share of the services sector in the economy, the larger the relative demand for ICT-using occupations can be expected to be. To approximate technology adoption or integration, ICT investment (capital expenditure)22 as a share of gross fixed capital formation and as a share of GDP are included separately in different versions of the model.23 The ICT investment data are from an unpublished OECD database based on national account sources. The indicator of product market regulation is an average of indicators of regulation in selected nonmanufacturing industries.24 These indicators measure, on a scale of 0 to 6 (from least to most restrictive), restrictions on competition and private governance. The original version of these data is described in Nicoletti and Scarpetta (2003). This indicator is used as a proxy for competitive pressures in the economy. The weaker such pressures are, the less incentive there is for companies to adopt efficient new technologies and new, more productive, ways of working. This would imply that a negative relationship can be expected between the importance of product market regulations in the economy and the 20. ISIC Rev.3 categories: 50–55: wholesale and retail trade; repairs; hotels and restaurants; 60–64: transport, storage, and communications; 65–74: finance, insurance, real estate, and business services; 75–99: community, social, and personal services. 21. ISIC Rev.3 categories: 2423: chemicals excluding pharmaceuticals; 30: office, accounting, and computing machinery; 32: radio, television, and communication equipment; 33: medical, precision, and optical instruments; 353: aircraft and spacecraft. 22. ISIC Rev.3 categories: 30: office, accounting, and computing machinery; 3130: insulated wire and cable; 3210: electronic valves and tubes and other electronic components; 3220: television and radio transmitters and apparatus for line telephony and line telegraphy; 3230: television and radio receivers, sound or video recording or reproducing apparatus, and associated goods; 3312: instruments and appliances for measuring, checking, testing, navigating, and other purposes; 3313: industrial process control equipment; 5150: wholesale of machinery, equipment, and supplies; 6420: telecommunications; 7123: renting of office machinery and equipment (including computers); 72: computer and related activities. 23. Results from the regressions using ICT investment as a share of GDP are not reported here. See van Welsum and Reif (2005) for details. 24. We use a preliminary unpublished version of this product market regulation indicator.
Desirée van Welsum and Xavier Reif
181
share of employment potentially affected by offshoring. Similarly, Messina (2004) includes a measure of entry barriers to the creation of new firms in the economy as an indicator of product market regulations and finds a significant and negative effect on the share of services sector employment. Two variables are included to capture institutional and supply-side influences on (unobserved) real wages: union density and human capital. Trade union density indicators may of course provide information about the degree of flexibility in national labor markets, as well as the relative strength of workers in wage bargaining.25 A number of existing papers suggest that union density rates are related to the growth of service sector occupations. For example, Messina (2004) finds that a fall in union density rates is associated with an increase in services sector employment. Similarly, Nickell, Redding, and Swaffield (2004) find evidence that countries with higher levels of employment protection were slower to reallocate resources from declining sectors (agriculture, manufacturing, and other production) to the services sector, possibly because stronger employment protection makes shedding labor in declining sectors more costly. The analysis in this paper does not consider employment at the sectoral level, but an analogy can be drawn since labor market inflexibilities are likely to affect occupational shifts as well as sectoral changes. The a priori effect of this variable is ambiguous, though, because it can both prevent a reallocation of resources into ICTintensive using occupations and hinder the speed at which existing ICT-intensive using jobs can be transferred abroad, maintaining the share at a higher level than it would otherwise have been. Finally, human capital is approximated by the average years of education per person.26 It is expected that this variable is positively related to the share of potentially offshorable occupations since increases in human capital are positively correlated with increases in the supply of ICT-literate people in the workforce. Such increases in supply should help to restrain the growth of real wages of workers in ICT occupations and hence support demand. Nickell, Redding, and Swaffield (2004) find a strong positive effect of increases in educational attainment on the output share of the “other services” sector in the economy in Australia, Canada, France, Italy, Japan, the Netherlands, Sweden, West Germany, the United Kingdom, and the United States.27
25. The data on trade union density rates come from OECD Labor Force Statistics Indicators and OECD 2004c (table 3.3). Factors other than union density rates, including union coverage and hiring and firing restrictions, are also important. 26. See de la Fuente and Doménech (2002a, 2002b) and OECD (2003). 27. But in the sector “business services” they found a greater role for changes in relative prices.
182
Brookings Trade Forum: 2005
Some Results The results using fixed-effects and instrumental variables estimation techniques on a sample excluding Greece, Ireland, Luxembourg, and Portugal are reported in table 1.28 The estimation for the basic fixed-effects models is for a sample of fourteen countries over the period 1996–2003. The instrumental variables estimates are for the same countries, but over the period 1997–2003. Columns 1 and 3 of table 1 show the standard fixed-effects results, and columns 2 and 4 show the results obtained when reestimating these models using instrumental variables. A year is dropped from the estimation period for these latter regressions to allow higher-order lagged variables to be used as instruments. All current dated terms, with the exception of the product market regulation indicator, are instrumented in columns 2 and 4. For these variables only instruments dated t-2 are included in the instrument set. The Sargan tests of the overidentifying restrictions provide support for the validity of the instrument set employed in both models. In each of the four models (columns 1 to 4), exports are found to have a positive and significant association with the share of employment potentially affected by offshoring, as expected. The coefficient on imports is negatively signed, as expected, but is not significant at the conventional 5 percent level in any of the models. Thus there is no significant evidence that increasing imports of other business and computer and information services are associated with a reduction in the share of employment potentially affected by offshoring at the aggregate level. Care is needed in drawing strong conclusions from these results, though, as the trade variables may be endogenous, especially if companies’ decisions about international sourcing and employment are made simultaneously. However, as shown in columns 2 and 4 of table 1, and in van Welsum and Reif (2005), the basic findings remain even when an instrumental variables estimator is employed. Net FDI is found to have a positive and significant association with the share of employment potentially affected by offshoring. Thus, contrary to popular belief, there is no evidence that outward investment or net FDI reduces the share of this type of employment at the aggregate level. This effect can probably be explained by the fact that manufacturing activities are much more important in total FDI than they are in the overall share of activities in host and home economies. An increase in the outward stock of FDI can also be expected to increase the relative share of occupations in support functions, as well as mar28. Country fixed-effects and year time dummies are included in all models. See van Welsum and Reif (2005) for further results.
Desirée van Welsum and Xavier Reif
183
Table 1. Estimation Results Using Fixed-Effects and Instrumental Variables Dependent variable: employment potentially affected by offshoring as a share of total employment (OLt ) 1 (X/GDP)t (M/GDP)t (NETFDI/GDP)t-1 (ICTI/INV)t-1 SERVICESt-1 HTECHt-1 PMRt UNIONSt-1 HKt-1 Sample period Observations Log likelihood R2 Standard error Time dummies (p value of joint deletion) Sargan test (p value)
2
3
4
0.9086 (5.8)* 0.9298 (2.6)* 0.8977 (5.6)* 1.3139 (3.0)* –0.2246 (1.4) –0.1309 (0.3) –0.3099 (2.0)* –0.7119 (1.4) 0.0384 (3.3)* 0.0435 (3.2)* 0.1132 (1.8)† 0.0984 (0.8) 0.0968 (1.5) 0.0992 (0.8) 0.1649 (3.6)* 0.1716 (3.5)* 0.1852 (3.5)* 0.1961 (3.6)* 0.1592 (0.7) 0.1760 (0.6) 0.2382 (1.1) 0.3056 (1.1) –0.1614 (0.7) 0.0171 (0.0) –0.0348 (0.1) –0.0105 (0.0) –0.1252 (2.9)* –0.1298 (2.6)* –0.0952 (2.1)* –0.1145 (2.1)* 1.1719 (3.7)* 1.2913 (3.2)* 1.3954 (4.2)* 1.4404 (3.3)* Fixed effects Fixed effects IV Fixed effects Fixed effects IV 1996–2003 1997–2003 1996–2003 1997–2003 112 98 112 98 –70.145 –74.863 0.963 0.960 0.960 0.957 0.542 0.563 0.562 0.583
0.193
0.795 0.112
0.609
0.853 0.611
Notes: (X/GDP) is the share of exports of other business and computer and information services in GDP; (M/GDP) is the share of imports of other business and computer and information services in GDP; (NETFDI/GDP) is the net stock of foreign investment (outward-inward) as a share of GDP; (ICTI/INV) is the share of ICT investment in total fixed investment; SERVICES is the share of the services sector in total value added; HTECH is the share of high-tech industries in total value added; PMR is a product market regulations indictor; UNIONS are trade union density rates; and HK is the average years of education per person. The additional instruments used are drawn from a set comprising (X/GDP)t-2, (M/GDP)t-2, OLt-2, (ICTI/INV)t-2, PMRt-1, PMRt-2, UNIONSt-2 and (NETFDI/GDP)t-2. * Significant at the 5 percent level. † Significant at the 10 percent level.
keting, design, and general headquarters services. Inward investment is found to be negatively related to the share of employment potentially affected by offshoring. With manufacturing also having a comparatively heavy weight in the activities of inward investors, it is not necessarily surprising that the relative share of employment in the types of occupations identified as potentially affected by offshoring is reduced. Further research will attempt to disentangle the effects of services investment from manufacturing investment. There are many different factors that might be reflected in the coefficients on the FDI variables. It is also true that FDI data can, at times, be a poor measure of the scale of activities that multinational companies undertake. Although this in itself is not a reason for omitting the FDI variables, it is prudent to repeat the regressions without them to ensure that their inclusion is not serving to significantly bias the coefficients on the other explanatory factors. The results, shown
184
Brookings Trade Forum: 2005
in columns 3 and 4 of table 1, suggest that the net FDI variable is largely orthogonal to the remaining regressors, with the possible exception of the imports term, whose coefficient becomes more negative. However, it remains insignificant, at least at the 5 percent level. The share of ICT investment in gross fixed capital formation is positively signed, but is not especially significant. The share of services sector value added as a percentage of total value added has a significant positive association with the share of employment potentially affected by offshoring, as expected, with many services having high shares of ICT-using occupations, but there is no significant relationship between the dependent variable and the high-tech industry value added as a percentage of total value added (though the coefficient is positively signed). The indicator of the importance of product market regulations in the economy is negatively signed (except in column 3) but is not significant. The two variables that are most likely to affect wages—union density and human capital—both have coefficients of the sign expected given the assumption that wages have a negative effect on employment. Higher levels of union density are associated with slower adjustment into the types of occupations potentially affected by offshoring, and the average years of education per person is significantly positively associated with the share of potentially offshorable employment, consistent with the observation that many such occupations are comparatively skill-intensive. Overall, the results appear fairly robust to different estimation techniques and specifications of the model. The most stable coefficients appear to be those on the ratio of exports of other business and computer and information services to GDP, net foreign direct investment stocks as a share of GDP, the share of the services sector in value added, and the average years of education per person. The full interpretation of these results must await further study. In particular, the development of corresponding data on relative wages should help to separate out demand and supply influences more clearly. Nevertheless, the results from the descriptive regressions in the present paper provide some useful indications of the statistical associations that are found between the variables examined and provide guidance for further work in this area.
Conclusions Despite the widespread media attention given to the apparent offshoring of service sector jobs, little is known about the extent of this phenomenon, or the
Desirée van Welsum and Xavier Reif
185
extent to which it is related to other economic and structural developments. In particular, an explicit link is often made between trade, the activities of multinational firms, and changes in employment, but this has not been founded on any solid evidence. This paper builds on previous detailed analysis of trade and occupational data. Trade data show that many of the countries frequently cited as beneficiaries of offshoring have seen rapid growth of their exports of other business and computer and information services. However, many have also seen rapid growth of imports of these services, and the bulk of exports of these types of services still come from OECD countries, although their export share is declining. The analysis of occupational employment data for selected OECD countries sought to determine the share of total employment that could potentially be affected by the international sourcing of IT and ICT-enabled services, drawing on van Welsum and Vickery (2005a). It suggested that close to 20 percent of total employment could potentially be affected by offshoring. This paper also makes an initial attempt to examine the relationship between the share of employment potentially affected by offshoring and other economic and structural developments using some simple descriptive regressions on a panel of selected OECD economies between 1996 and 2003. In particular, first estimates are provided of the statistical association between the share of potentially offshorable employment and trade in business services and international direct investment. The results indicate that exports of other business services and computer and information services are positively associated with the share of employment potentially affected by offshoring. This suggests that increases in demand and production have led to a relative increase in the types of ICT-using occupations identified in the analysis. Furthermore, contrary to popular belief, no evidence is found of a significant negative association between imports of these services and the share of employment potentially affected by offshoring. Similarly, no evidence is found that net outward investment reduces the employment share of the ICT-intensive using occupations identified as potentially affected by offshoring. Other key factors associated with cross-country differences in the employment share are found to be the comparative size of the service sector, the growing share of ICT investment in total fixed investment, and human capital. These results suggest that, in the OECD countries analyzed, ICT-enabled services offshoring (as proxied by trade and investment) has not yet led to a relative decline in the occupational share of location-independent ICT-using occupations. This implies that in the long run the positive benefits of services off-
186
Brookings Trade Forum: 2005
shoring outweigh the costs, even though the adjustment process may occasionally be difficult in the short run. Policy responses to services offshoring should reflect these positive aspects. These would include policies that seek to contribute to the overall competitiveness of the economy and improve the macroeconomic framework, to contribute to a sound investment climate, and to improve the skills base and flexibility of the workforce. It is important to interpret these results with caution, though, because they are not drawn from the empirical testing of a formal theoretical model of the underlying structural relationships. Thus, it is not possible to separate out completely the effects from demand and supply-side developments. However, the results provide guidance on the statistical associations between the variables included in these descriptive regressions and to this extent can be used to shape further work and analysis. This could include improvements to the underpinnings of the empirical model, such as the use of separate indicators for services and nonservices FDI, and examination of whether different factors affect different groups of ICT-using occupations, such as clerical and nonclerical occupations. It would also be useful to develop an indicator of business adoption of ICTs to try to control for differences in “the use of ICT” or the “ICT content of occupations” across countries.
Desirée van Welsum and Xavier Reif Table A-1. IMF Balance of Payments Categories 7 7.1 7.2 7.2.1 7.2.2
Computer and information services Computer services Information services News agency services Other information provision services
9 9.1 9.1.1 9.1.2 9.2 9.3 9.3.1 9.3.1.1 9.3.1.2 9.3.1.3 9.3.2 9.3.3 9.3.4 9.3.5 9.3.5.1 9.3.5.2 9.3.6 9.3.7
Other business services Merchanting and other trade-related services Merchanting Other trade-related services Operational leasing services Miscellaneous business, professional, and technical services Legal, accounting, management consulting, and public relations Legal services Accounting, auditing, bookkeeping, and tax consulting services Business and management consulting, and public relations Advertising, market research, and public opinion polling Research and development Architectural, engineering, and other technical services Agricultural, mining, and on-site processing services Waste treatment and depollution Agricultural, mining, and other on-site processing services Other business services Services between related enterprises, n.i.e.a
Source: OECD (2002). a. Not included elsewhere.
187
188
Brookings Trade Forum: 2005
Table A-2. Europe: Occupations Potentially Affected by Offshoringa 3-digit ISCO-88 categories 123 211 212 213 214 241 242 243 312 341 342 343 411 412 422
Other specialist managers Physicists, chemists, and related professionals Mathematicians, statisticians and related professionals Computing professionals Architects, engineers, and related professionals Business professionals Legal professionals Archivists, librarians, and related information professionals Computer associate professionals Finance and sales associate professionals Business services agents and trade brokers Administrative associate professionals Secretaries and keyboard-operating clerks Numerical clerks Client information clerks
Source: Van Welsum and Vickery (2005a), based on European Union Labor Force survey (www.eds-destatis.de/en/microdata/ microlfs.php [December 2004]). a. Shaded occupations are classified as clerical.
Desirée van Welsum and Xavier Reif
189
Table A-3. United States: Occupations Potentially Affected by Offshoringa CPS categories 23 24 25 26 43 44 45 46 47 48 49 53 54 55 56 57 58 59 63 64 65 66 67 68 69 73 74 75 76 77 78 79 83 164 165 166 173
Accountants and auditors Underwriters Other financial officers Management analysts Architects Aerospace engineers Metallurgical and materials engineers Mining engineers Petroleum engineers Chemical engineers Nuclear engineers Civil engineers Agricultural engineers Engineers, electrical and electronic Engineers, industrial Engineers, mechanical Marine and naval architects Engineers, n.e.c.b Surveyors and mapping scientists Computer systems analysts and scientists Operations and systems researchers and analysts Actuaries Statisticians Mathematical scientists, n.e.c. Physicists and astronomers Chemists, except biochemists Atmospheric and space scientists Geologists and geodesists Physical scientists, n.e.c. Agricultural and food scientists Biological and life scientists Forestry and conservation scientists Medical scientists Librarians Archivists and curators Economists Urban planners
183 184 195 227 229 233 243 253 254 255 257 304 305 306 308 309 313 315 318 335 336 337 338 339 343 344 348 383 385 386
Authors Technical writers Editors and reporters Air traffic controllers Computer programmers Tool programmers, numerical control Supervisors and proprietors, sales occupations Insurance sales occupations Real estate sales occupations Securities and financial services sales occupations Sales occupations, other business services Supervisors, computer equipment operators Supervisors, financial records processing Chief communications operators Computer operators Peripheral equipment operators Secretaries Typists Transportation ticket and reservation agents File clerks Records clerks Bookkeepers, accounting, and auditing clerks Payroll and timekeeping clerks Billing clerks Cost and rate clerks Billing, posting, and calculating machine operators Telephone operators Bank tellers Data-entry keyers Statistical clerks
Source: van Welsum and Vickery (2005a), based on U.S. Current Population Survey. a. Shaded occupations are classified as clerical. b. Not elsewhere classified.
Table A-4. Canada: Occupations Potentially Affected by Offshoringa SOC91 Canada categories A121 A122 A131 A301 A302 A303 A311 A312 A392 B011 B012 B013 B014 B022 B111 B112 B114 B211 B212
Engineering, science, and architecture managers Information systems and data processing managers Sales, marketing, and advertising managers Insurance, real estate, and financial brokerage managers Banking, credit and other investment managers Other business services managers Telecommunication carriers managers Postal and courier services managers Utilities managers Financial auditors and accountants Financial and investment analysts Securities agents, investment dealers and traders Other financial officers Professional occupations in business services to management Bookkeepers Loan officers Insurance underwriters Secretaries (except legal and medical) Legal secretaries
C012 C013 C014 C015 C021 C031 C032 C033 C034 C041 C042 C043 C044 C045 C046 C047 C048 C051 C052 C053
Chemists Geologists, geochemists, and geophysicists Meteorologists Other professional occupations in physical sciences Biologists and related scientists Civil engineers Mechanical engineers Electrical and electronics engineers Chemical engineers Industrial and manufacturing engineers Metallurgical and materials engineers Mining engineers Geological engineers Petroleum engineers Aerospace engineers Computer engineers Other professional engineers, n.e.c. Architects Landscape architects Urban and land use planners
B213 B214 B311 B312 B412 B512 B513 B514 B521 B522 B523 B524 B531 B532 B533 B534 B553 B554 C011
Medical secretaries Court recorders and medical transcriptionists Administrative officers Executive assistants Supervisors, finance and insurance clerks Typists and word processing operators Records and file clerks Receptionists and switchboard operators Computer operators Data entry clerks Typesetters and related occupations Telephone operators Accounting and related clerks Payroll clerks Tellers, financial services Banking, insurance and other financial clerks Customer service, information and related clerks Survey interviewers and statistical clerks Physicists and astronomers
Source: van Welsum and Vickery (2005a), based on Statistics Canada. a. Shaded occupations are classified as clerical.
C054 C061 C062 C063 C152 C172 E012 E031 E032 E033 F011 F013 F021 F022 F023 F025 G131
Land surveyors Mathematicians, statisticians, and actuaries Computer systems analysts Computer programmers Industrial designers Air traffic control occupations Lawyers and Quebec notaries Natural and applied science policy researchers, consultants, and program officers Economists and economic policy researchers and analysts Economic development officers and marketing researchers and consultants Librarians Archivists Writers Editors Journalists Translators, terminologists, and interpreters Insurance agents and brokers
192
Brookings Trade Forum: 2005
Table A-5. Australia: Occupations Potentially Affected by Offshoringa ASCO 4-digit categories 1221 1224 1231 1291 2111 2112 2113 2114 2115 2119 2121 2122 2123 2124 2125 2126 2127 2211 2212 2221 2231 2292 2293 2294
Engineering managers Information technology managers Sales and marketing managers Policy and planning managers Chemists Geologists and geophysicists Life scientists Environmental and agricultural science professionals Medical scientists Other natural and physical science professionals Architects and landscape architects Quantity surveyors Cartographers and surveyors Civil engineers Electrical and electronics engineers Mechanical, production, and plant engineers Mining and materials engineers Accountants Auditors Marketing and advertising professionals Computing professionals Librarians Mathematicians, statisticians, and actuaries Business and organization analysts
2299 Other business and information professionals 2391 Medical imaging professionals 2521 Legal professionals 2522 Economists 2523 Urban and regional planners 2534 Journalists and related professionals 2535 Authors and related professionals 3211 Branch accountants and managers (financial institution) 3212 Financial dealers and brokers 3213 Financial investment advisers 3294 Computing support technicians 3392 Customer service managers 3399 Other managing supervisors (sales and service) 5111 Secretaries and personal assistants 5911 Bookkeepers 5912 Credit and loan officers 5991 Advanced legal and related clerks 5993 Insurance agents 5995 Desktop publishing operators 6121 Keyboard operators 6141 Accounting clerks 6142 Payroll clerks 6143 Bank workers 6144 Insurance clerks 6145 Money market and statistical clerks 8113 Switchboard operators 8294 Telemarketers
Source: van Welsum and Vickery (2005a), based on Australian Bureau of Statistics. a. Shaded occupations are classified as clerical.
Desirée van Welsum and Xavier Reif
193
References Autor, David H., Frank Levy, and Richard J. Murnane. 2003. “The Skills Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics 118, no. 4: 1279–1333. Baily, Martin N., and Robert Z. Lawrence. 2004. “What Happened to the Great U.S. Job Machine? The Role of Trade and Electronic Offshoring.” BPEA, no. 2: 211–84. Bardhan, Ashok D., and Cynthia Kroll. 2003. “The New Wave of Outsourcing.” Fisher Centre Research Report 1103. University of California Berkeley, Fisher Centre for Real Estate and Urban Economics. de la Fuente, Angel, and Raphael Doménech. 2002a. “Educational Attainment in the OECD, 1960–1995.” Discussion Paper 3390. Centre for Economic Policy Research, London. ———. 2002b. “Human Capital in Growth Regressions: How Much Difference Does Data Quality Make? An Update and Further Results.” Discussion Paper 3587. Centre for Economic Policy Research, London. Kirkegaard, Jacob Funk. 2004. “Outsourcing—Stains on the White Collar?” Washington: Institute for International Economics. Levy, Frank, and Richard J. Murnane. 2004. The New Division of Labor. Princeton University Press and the Russell Sage Foundation. Mann, Catherine L. 2004. “The U.S. Current Account, New Economy Services and Implications for Sustainability.” Review of International Economics 12, no. 2: 262–76. Marin, Dalia. 2004. “‘A Nation of Poets and Thinkers’—Less So with Eastern Enlargement? Austria and Germany.” Department of Economics Discussion Paper 2004-06. University of Munich. Messina, Julian. 2004. “Institutions and Service Employment: A Panel Study for OECD Countries.” Labour: Review of Labour Economics and Industrial Relations 19, no. 2: 343–72. Millar, Jane. 2002. “Outsourcing Practices in Europe.” STAR Issue Report 27, www.databank.it/star/list_issue/e.html. Nickell, Stephen, Stephen Redding, and Joanna Swaffield. 2004. “The Uneven Pace Of Deindustrialisation in the OECD.” Paper prepared for the OECD Workshop on Services, November 15–16, based on CEPR Discussion Paper 3068. Nicoletti, Giuseppe, and Stefano Scarpetta. 2003. “Regulation, Productivity and Growth: OECD Evidence.” Economic Policy (April): 9–72. Organization for Economic Cooperation and Development (OECD). 2002. The Manual on Statistics of International Trade in Services, joint publication of the United Nations, the International Monetary Fund, the OECD, the European Commission, the United Nations Conference on Trade and Development, and the World Trade Organization. An electronic version of the manual is available free of charge at www.oecd.org/std/trade-services.
194
Brookings Trade Forum: 2005
———. 2003. The Sources of Economic Growth in OECD Countries. Paris. ———. 2004a. OECD Information Technology Outlook 2004. Paris. ———. 2004b. OECD Economic Outlook 2004/1, no. 75. Paris. ———. 2004c. OECD Economic Outlook 2004. Paris. ———. 2005. OECD Employment Outlook 2005. Paris. Pain, Nigel, and Desirée van Welsum. 2004. “International Production Relocation and Exports of Services.” OECD Economic Studies 2004/1, no. 38. Schultze, Charles L. 2004. “Offshoring, Import Competition, and the Jobless Recovery.” Policy Brief 36. Brookings. van Welsum, Desirée. 2003. “International Trade in Services: Issues and Concepts.” Birkbeck Economics Working Paper 2003/4. Birkbeck College, University of London. ———. 2004. “In Search of ‘Offshoring’: Evidence from U.S. Imports of Services.” Birkbeck Economics Working Paper 2004/2. London: Birkbeck College, University of London. van Welsum, Desirée, and Xavier Reif. 2005 (forthcoming). “The Share of Employment Potentially Affected by Offshoring—An Empirical Investigation.” DSTI Information Economy Working Paper DSTI/ICCP/IE(2005)8/FINAL. Paris: OECD. www.oecd. org/sti/offshoring. van Welsum, Desirée, and Graham Vickery. 2005a. “Potential Offshoring of ICTIntensive Using Occupations.” DSTI Information Economy Working Paper DSTI/ICCP/IE(2004)19/FINAL. Paris: OECD. www.oecd.org/sti/offshoring. ———. 2005b. “New Perspectives on ICT Skills and Employment.” DSTI Information Economy Working Paper DSTI/ICCP/IE(2004)10/FINAL. Paris: OECD. www. oecd.org/sti/ICT-employment.
Comments and Discussion
Robert Z. Lawrence: The strength of these two papers is that they both carefully scrutinize the data to see what we can learn about offshoring. Their weakness is that they both do so without adopting a theoretical framework that could help them sort out what they find. Data mining can sometimes be useful when the data clearly allow us to reject hypotheses, but when they do not, the limitations become apparent and the need for a framework more obvious. I will elaborate on these points in discussing each paper in turn. In the first paper, Maria Borga uses firm-level data to explore trends in employment in the parents and affiliates of U.S. multinational companies. Her strategy is to explore the degree to which firms changed their offshoring behavior between 1994 and 2002 and then to explore the association between offshoring and other variables, in particular employment. As the author herself points out, this approach allows only for a consideration of bivariate relationships and excludes the possibility of detecting more complex interactions. The paper does yield a number of insights. Let me point out three I found striking. First, the most relevant for our topic is the finding that between 1994 and 2002 imports of services from majority-owned foreign affiliates actually played a very small role in the growth of parent companies’ overall purchases of goods and services. In fact, services imports actually fell from 0.4 percent of parent purchases in 1994 to 0.2 percent in 2002. If these data are correct, it suggests that whatever may happen in the future, offshoring has had very little to do with U.S. employment changes in the past. This would be a very surprising result for anyone listening to the nightly news in 2004. Second, the data also suggest less direct outsourcing in goods than might have been expected. The growth in parents’ use of merchandise imports from affiliates (up from 3.5 to 4.6 percent of purchases) was relatively small, while 195
196
Brookings Trade Forum: 2005
there was no growth in the share accounted for by imports from nonaffiliated firms. Firms are outsourcing more (up from 60 to 63 percent of sales), but apparently most of this rise is from domestic firms. Thus, there is some support for the view that firms are slimming down to their core competencies, but not for the view that the globalization of production is a hugely important phenomenon. Third, a conspicuous feature of the recent U.S. recession was the sustained growth in labor productivity. This could be due either to the cold bath effect of eliminating the least productive parts of the firm—that is, downsizing—or to technological innovation. Borga’s paper suggests downsizing could be important. The picture we obtain from the paper is that the firms that were reducing employment were disproportionately laying off their low-wage (least productive) workers. We see this in the rise in average compensation and relatively faster productivity growth in these firms. These firms also significantly increased their purchases of goods and services as a share of output (increasing purchased inputs from 62 to 66 percent of sales—much of which represented purchases from the rest of the world up from 6.95 to 9.3 percent of purchases). The foreign affiliates of firms with declining employment growth also have slower employment growth, suggesting complementarity rather than substitutability in employment at home and abroad. This is a useful paper, but it is just a start because it does not allow us to explain why we obtain these results. As we know, correlations do not tell us about causation. Consider the finding that quantitatively the most significant factors affecting parent employment were changes in output and in labor productivity and that outsourcing in general, and offshoring in particular, appear much less important. We know that all three of these variables are jointly determined and are not exogenous. It is certainly possible that the ability to source foreign inputs actually stimulates output and helps to induce greater productivity. In this case, even though an ex post accounting exercise of the type performed in the paper would attribute only a small part of the change to outsourcing, it would clearly be wrong to infer that outsourcing is not important. In fact, the more powerful the impact of outsourcing on output and productivity, the less important it would appear to be in the decomposition exercise that is performed here. On the other hand, if the variables have predominantly independent effects, it is also possible that the conclusions drawn here are correct. It is therefore appropriate that Maria Borga is cautious in the claims that she makes about these findings and that she calls for further research on these issues. The paper by Desirée van Welsum and Xavier Reif uses international occupational data to construct an estimate of jobs that could potentially be offshored.
Borga / Van Welsum and Reif: Comments
197
As the authors are aware, these estimates are subject to considerable qualification because national classification methodologies differ. In addition, even a casual inspection of the occupational categories suggests that there is considerable misclassification. Even though their jobs have been identified here as all potentially affected by outsourcing, for example, it is highly likely that many secretaries, librarians, bank tellers, real estate agents, air traffic controllers, editors, and reporters need to be in the location where their services are used. This means that measurement error is introduced into the dependent variable and could seriously impair the results. The authors report on the behavior of this variable (let’s call it the PO variable) over the period 1995 to 2003 in different OECD countries. The PO employment experience has been mixed. There are pronounced declines in the United States and Canada and more recently in Australia, but strong upward growth in the European Union. This could be evidence that the offshoring of ITrelated and back-office activities has had an important impact on North American employment. But this is not something the authors are able to pin down. The reason is that they are interested in explaining PO, not actual offshoring. The main contribution of the paper is to explain the behavior of the PO variable using a cross section regression analysis of fourteen OECD countries. The results indicate that PO employment is significantly and positively affected by exports of information services, net FDI, the employment share of services, and human capital, and negatively affected by a high share of unions. Think about what these findings imply for the United States. I presume, for example, that the United States exports lots of information services, has a high share of services employment, a highly educated labor force, and a low share of unions. This would lead me to expect, on the basis of the regression results in the paper, that the United States should have a high share of PO employment! But apparently, as of 2003, the United States actually had a low share of PO. This result could mean something else; perhaps, as the authors suggest, it is that offshoring has caused the U.S. share to be low. But they do not prove or even explore this hypothesis. In fact, taking these results at face value increases the puzzle as to why the share of employment potentially affected by offshoring has declined in the United States. My main problem with the paper is that it gets to the regressions too quickly and does not lay out an adequate theoretical framework. If this is to be a paper about PO, I would like a step-by-step explanation of the relationship between PO and the variables that are introduced as independent. Assume, for a moment, that there is no offshoring—that all information services have to be provided locally.
198
Brookings Trade Forum: 2005
We could then build and estimate a model that explained the share of employment in such services. The reduced form of such a model would probably have several of the independent variables that are in the regression the authors use; some would relate to the demand for these services (specific types of spending such as investment, exports, services in general, and imports) and some would relate to the supply of these services (such as human capital). As long as there was no offshoring we would be able to estimate this model using data for actual information technology services employment. Now assume that because of the Internet or improvements in communication offshoring can take place. We would then have to come up with a model that reflected the determinants of this offshoring. (Presumably this would require modeling the variables driving relative costs of undertaking services activities at home and abroad—variables that seem to be missing in this paper.) Actual services employment would then reflect potentially affected employment—the variable explained by the previous model—minus employment that is actually offshored. If we now want to explain actual services employment, therefore, we see that we have to include the independent variables from both the first and the second model. Which of these two worlds do we live in? Presumably the one in which outsourcing is already taking place. This means that the variable used by the authors as the dependent variable is also affected by the determinants of offshoring, but this is not explicitly taken account of in the regression. To the degree that offshoring is already an important source of variation in IT service employment, this could be a serious omission, and it could introduce bias into the results. Indeed, the essence of the debate about offshoring is that the relationship between certain types of domestic production and (actual) employment in IT services is changing. If offshoring has become important, exports, investment, and other types of spending now lead to fewer domestic employment opportunities in IT. But that is not something we really can learn about from these regressions. In sum, both these papers take first steps to deal with the questions we are interested in at this conference, but neither really succeeds in presenting a convincing explanation for its findings. In both cases, it seems, the main problem stems from moving too quickly to the data, before laying out the theory of what really drives offshoring in the first place. Catherine L. Mann: Recent media and political hype has focused on the negative impact on white-collar employment in industrial countries of cross-border
Borga / Van Welsum and Reif: Comments
199
trade in services, particularly by multinational corporations. These papers address this theme, but try to evaluate the simplistic stories through the more robust and well-founded research lens of how multinational corporations engage with the global economy through trade and direct investment, and how the decision to go abroad is importantly linked to both the domestic and foreign environments. These two papers are complementary in their research focus and method. Borga focuses on U.S. multinational firms and on types of outsourcing in both manufacturing and services activities. Van Welsum and Reif address a range of industrial country experience, paying attention primarily to services occupations. In keeping with new availabilities, each uses some data that are quite detailed. Borga uses firm-level data for trade, direct investment, and employment. Van Welsum and Reif use detailed occupation data. The authors of both papers recognize that their results would be enhanced by more rigorous matching at the detail level of the data. For example, Van Welsum and Reif need to use FDI disaggregated at least to the level of manufacturing and services and could stratify occupations by above- and below-average wage. Borga could match the sector of the firmlevel data to occupational data, which might give a glimmer of the different impacts of outsourcing on production and on nonproduction employment. Each paper uses some statistical methods to carve the data into subsamples (by industry, or by occupation, or by expanding and contracting firms). A more systematic strategy of fixed-effects or random-effects econometric analysis (by country or occupational group in the case of Van Welsum and Reif, and by industry sector and occupational group in the case of Borga) would perhaps appear a bit less ad hoc to the reader and would highlight important differences across sectors or countries or occupations. On balance, however, the findings of these two papers are important additions to our understanding of both “old” outsourcing and “new” offshoring. In keeping with most previous work on global engagement (which could have been more liberally referenced by the authors), these papers find that trade, FDI, and domestic employment are positively related, particularly for competitive and expanding industries. Moreover, factors that underpin a rising standard of living (such as IT and human capital investment) go hand-in-hand with greater exposure to international market forces. More detailed comments on the findings and methods of each paper individually follow. A key contribution and a key aspect of the Borga paper is in distinguishing between outsourcing and offshoring. The value of a multinational parent’s “purchased inputs” plays a key role in the paper. Purchased inputs include outsourcing (purchasing from a domestic or foreign unrelated party) and offshoring
200
Brookings Trade Forum: 2005
(purchasing from a foreign party, either affiliated or unrelated). The data show that dependence on purchased inputs rose from 60 to 63 percent (1994 to 2002). But the imported share of the purchased inputs rose only from 6.9 to 7.8 percent. Hence, even for multinationals, the bulk of purchased inputs (such as outsourcing) comes from home, not from abroad. Borga stratifies the sample into growing parents and shrinking parents, as measured by employment over the period 1994 to 2002. It may come as a surprise, but of the 1,117 multinational parent firms, 689 expanded and 428 shrank. The most notable difference between expanding parents and contracting ones was in sales. Sales grew in expanding parent firms and declined in contracting ones. In many other respects the two are similar: Both increased employment at their affiliates abroad as well as the share of those affiliates’ sales in their own markets, although expanders increased by more both the share of employment abroad and the share of affiliate sales abroad. Both relied more on purchased inputs. So, other than in sales growth and decline, how do expanders and shrinkers differ? Expanders reduced their reliance on imported purchased inputs, whereas shrinkers increased their reliance on imported inputs, particularly from their own affiliates (although keep in mind that the affiliate imports as a share of purchased inputs was still small, just 6.0 percent in 2002). When all data are aggregated, the finding that FDI and domestic employment are complements, not substitutes, is affirmed. However, when the expanders and the shrinkers are disaggregated, the increase in imported inputs is statistically negatively correlated with employment, both because expanders reduced imports and because shrinkers increased imports. All told, however, the shrinkers look to be in sectors experiencing secular decline in demand for their products that no form of outsourcing (either purchasing domestically or from affiliates abroad) can remedy. It would be very interesting to know the sectors of the expanders and shrinkers, even if only at the level of textiles and apparel, electronics, and so on. In this paper, Borga only pursues the limited classification of manufacturing, services, and “other,” which is not very enlightening. The most important attribute of the van Welsum and Reif paper is the detailed work with occupation-level data across several different OECD countries. Occupations are stratified by characteristics including ICT intensity in use; digitizable output; codifiable knowledge content; face-to-face contact not required. The stratification requires a detailed assessment of each country’s labor statistics. It would be nice if a more econometrically based approach could back up the stratification, but the necessary data (such as IT by sector, or wages by sector and occupation) are not available. At the end of the exercise, van Welsum and Reif find that about
Borga / Van Welsum and Reif: Comments
201
20 percent of employment is exposed to technology-based tradability. Note that this tradability could yield domestic outsourcing or foreign offshoring. Van Welsum and Reif find that the share of such exposed employment rises in the European Union and is stable or falling in the United States, Canada, and Australia. They undertake preliminary econometric analysis to explain the time series evolution of the share using country-specific international data on services trade flows and net total FDI, and domestic structural characteristics including intensity of ICT investment, product and employment regulations, and human capital. As in the Borga paper, exports and FDI are positively associated with the share of employment exposed to tradability. The FDI variable, being both net and inclusive of manufacturing and services, needs additional disaggregation to better match the detail of the data, as the authors agree. With respect to the structural variables, it is notable that the factors that underpin a rising standard of living (such as IT and human capital investment) go hand-in-hand with greater exposure to international market forces. The results of both papers yield the observation that technological change and global forces engender trend and structural change for businesses and labor. Gaining from these forces may require that workforce and business change what they do to take advantage of new growth and occupational opportunities, whether those be at home or abroad. General Discussion: Robert Litan reflected that some of the results reported by Desirée van Welsum and Xavier Reif may come from the vulnerability of particular occupations to technological changes, and not to their offshorability. He suggested this as a possible explanation for the finding that the share of employment vulnerable to offshoring in the United States has been declining, a finding that he saw as counterintuitive. If so, these jobs would be disappearing, with or without offshoring. Perhaps not surprisingly, some participants had questions about the particular classification of some occupations in the paper by van Welsum and Reif. For example, Litan had not thought of air traffic controllers as vulnerable to offshoring. Van Welsum explained that there had been extensive discussions about how to classify some of the occupations. In some countries, air traffic controllers can be located up to a thousand kilometers away from the airport for which they work, hence the classification of this occupation as offshorable. Litan also pointed out that the United States uses somewhat different occupational classifications than the European countries reported, which makes cross-country comparisons difficult.
202
Brookings Trade Forum: 2005
Brad Jensen noted that his look at the data does suggest that importing from affiliates can help a multinational firm to survive. In this sense, he concurred with points made by Maria Borga. A number of participants made suggestions regarding the empirical analyses in the two papers. Some expressed concerns about possible sample selection issues such as survivor bias, in the paper by Borga. An appendix has been added to the revised version of the paper to address these concerns. Others noted that it would be interesting to explore recent developments using price and wage data.
T. N . S R I N I V A S A N Yale University
Information-Technology-Enabled Services and India’s Growth Prospects
D
uring the first three decades (1950–80) of India’s planned insular economic development, real GDP grew at an annual average rate of around 3.75 percent. The 1980s saw a limited opening of the economy and hesitant reforms. The growth rate accelerated to 5.7 percent, fueled by fiscal profligacy financed in part by external borrowing at high interest rates. A severe macroeconomic and balance-of-payments crisis in 1991 following the first Gulf War, the collapse of the Soviet Union (which was not only India’s model for planned economic development but also its arms supplier, a partner for barter trade, and a supporter of India’s interests in the Security Council of the United Nations), and the fear of being left behind by the rapid growth of China since its opening in 1978 led Indian policymakers to break away from its inward-oriented, state-directed, and controlled development strategy and open the economy to external competition and investment. After addressing the crisis with the assistance of the International Monetary Fund and the World Bank, the policymakers launched a process of systemic economic reforms that is still in progress. The economy responded to the reforms and quickly rebounded from the crisis-induced fall in growth of GDP to 1.3 percent in 1991–92. The growth rate accelerated, peaking at 7.8 percent in 1996–97. Subsequently it has fluctuated, falling to a low of 4.0 percent in 2002–03, largely because of a severe drought-induced decline in agricultural output, and rising to a peak of 8.5 percent the very next year, in large part owing to the recovery of agricultural output (MOF 2005, appendix table 1.6). The latest available data show that in the fiscal year of 2004–05, GDP growth was estimated at 6.9 percent,
I thank Lael Brainard, Susan Collins, Kanwal Rekhi, AnnaLee Saxenian, and Nirvikar Singh for their comments.
203
204
Brookings Trade Forum: 2005
and in fiscal year 2005–06 it is also expected to be around that level (RBI 2005a, table 1). India’s record of sustained growth since 1980 is second only to China’s among large economies, so much so that, in discussions about global economic prospects generally or about global demand for natural resources (including, most important, fossil fuels), the impact of Chinese and Indian growth is explicitly mentioned. However, India has succeeded only modestly in raising its share of world merchandise trade to 0.8 percent in 2004 from a low of 0.5 percent in 1983, while China’s share has more than quintupled, from 1.2 percent to 6.4 percent, during the same period. China ranked third from the top in the share of world merchandise trade in 2004, while India ranked a distant thirty-first in 2003 (WTO 2005, appendix tables 1 and 3; WTO 2004, table I.3). In global trade in commercial services (in which trade in information-technology-enabled services [ITES] and business process outsourcing [BPO] are included),1 India has done relatively better, with a share of 1.5 percent and a rank of twenty-one in 2004 as compared to China’s 2.8 percent share and ninth rank (WTO 2005, appendix tables 2 and 4). Both China and India are expected to gain a significant share of the global market of textiles and apparel with the expiration of the infamous Multifibre Arrangement. Apparel imports from China are already being targeted for restrictions by the United States and the European Union, and China itself is restraining exports in anticipation. Similar actions against Indian exports are possible, although India as a founding member of the World Trade Organization (WTO) is not subject to the special provisions of China’s Agreement of Accession to the WTO that have been used to restrict China’s exports. India’s software services exports in 2003–04 amounted to $12.2 billion, or nearly half the total services exports of $24.9 billion. Earnings from ITES and BPO accounted for another $3.6 billion (RBI 2005b, p. S343; MOF 2005, p. 111).2 A high-level strategy group set up by the All India Management Association (AIMA), comprising leaders from industry, academia, and the government, deliberated on the opportunities for providers in India in the growing global market for remote services (IT services such as software, ITES, telemed1. As will be clear from my subsequent discussion, a range of services is included under the umbrella of ITES. BPO is a significant enough category of ITES to be broken out separately; IT services themselves, which are treated in a different category, are also inherently “IT enabled.” 2. Data from Indian sources on India’s exports of computer and information systems to the United States and data on imports of the same services by the United States differ. On the reasons for this difference and adjustments to narrow it, see WTO (2005, box 2, p. 280). There is a presumption that Indian data on exports include the earnings of its nationals working in the United States on a temporary basis. They are also apparently included in the total employment of the sector in the Indian data.
T. N. Srinivasan
205
icine, and learning) and services sought by visitors to India such as conventional tourism, health care, and education.3 The group consulted wide segments of society and decisionmakers, and its findings were processed by a task force comprising AIMA, the Confederation of Indian Industry, and the Boston Consulting Group to formulate an action program. The task force concluded that lack of coordination among various stakeholders toward achieving a common goal was why India had been relatively slow in availing itself of the emerging opportunities (AIMA 2003). It views its report as a first step in a process of bringing about coordination and aligned actions. For the purposes of this paper, it is enough to note the growth and employment implications of the opportunities that the task force identifies. It finds that, by 2020, India can hope to generate $139–$365 billion of additional revenue from the supply of remote services for foreign residents and in situ services for visitors, pushing up the GDP growth rate by an additional 0.6–1.5 percent per year between 2002 and 2020. It estimates the direct and indirect employment generated by this additional growth to be between 20 million and 72 million.4 To put these numbers in perspective, India is targeting GDP growth of at least 8 percent per year in the next two decades. Its labor force, estimated at 363 million persons in 1999–2000 (MOF 2004, table 10.7), is expected to grow by 1.5 percent per year in the next two decades. If that happens, 125 million persons would be added to the labor force during the period. IT job growth projected by AIMA could provide jobs for a significant share of these additions to the labor force, assuming that each IT worker is fully employed. The methodology of projection by the task force is not explained by AIMA. The possibility that the projections of revenues and employment are very optimistic cannot be ruled out. However, for the very near term, official projections are also available (MOF 2005, box 6.2). The Ministry of Finance expects value added by the IT sector (including ITES) to grow to 7 percent of GDP by 2008 from around 2.64 percent in 2003–04. Exports of this sector are expected to be between $57 billion and $65 billion, accounting for 35 percent of total exports 3. The last category does not directly come under the umbrella of ITES, but just as the development of IT services created positive spillovers for ITES, the latter, by improving information flows and infrastructure, support the development of tourism and health and education services for nonresidents. Thus this third category is also discussed in this paper. 4. AIMA (2003, p. 16). WTO (2005, p. 283 and appendix table 8, p. 301) cites India’s National Association of Software Service Companies (NASSCOM) as its source for data on employment in India’s software industry. It reports employment of 813,000 in 2003–04 and employment growth of 21 percent in that year alone. If this rate of growth is sustained over sixteen years, employment would grow to 17.2 million, generating additional employment of 16.4 million in the software industry alone between 2003–04 and 2019–20.
206
Brookings Trade Forum: 2005
in 2009, a rise of 14 percent from 2003–04. ITES/BPO exports rose from $0.57 billion in 1999–2000 to $3.6 billion in 2003–04 and are expected to rise to $21–24 billion by 2008. The fact that India’s share of world IT spending was only 3.4 percent in 2003–04 suggests that there is still considerable potential for significant further growth. Services imports from and BPO to India have attracted and continue to attract media attention in the United States. They have also evoked a protectionist response: “State legislators have introduced at least 112 bills in 40 states to restrict outsourcing till March 17 this year [2005]. . . . The New Jersey bill, which awaits the governor’s decision to sign or veto, would be the most farreaching anti-outsourcing measure in the country by prohibiting all state contract work from being performed overseas.”5 All this is indicative of the fact that not only within India, but also in the rest of the world, India is expected to be a major player in the IT industry in general and in BPO in particular. The emerging consensus is that India will continue to grow rapidly in the next several decades, and that its IT sector, broadly speaking, will contribute significantly not only to GDP growth but also to employment generation and poverty alleviation. In what follows, I trace the development of India’s IT sector and the continuing role of the Indian IT diaspora in the Silicon Valley in the United States. I then briefly discuss the possible role of IT in the growth process and as a source of dynamic comparative advantage and look in some detail at the prospects of and possible constraints on India delivering the high expectations about its IT sector.
The Development of India’s IT Sector Prima facie, it is a surprise that India has been able to achieve as much as it has in IT development.6 India is still a low-income country with gross national income per capita of $540 (ranked 159th from the top) using World Bank’s Atlas method of calculating exchange rates, or $2,880 at purchasing power parity exchange rates in 2003 (146th from the top). Even the IT indicators for India are not impressive: with 7.2 personal computers per 1,000 people in 2002, India is 5. Suman Guha Mozumdar, “BPO Scare Intact: 112 Anti-Outsourcing Bills Moved,” India Abroad, May 6, 2005, p. A26. The United States is among thirty or more members of the WTO who are signatories to the plurilateral code on government procurement. The New Jersey measure could be in violation of the code. However, since India is not a signatory to the code, it cannot avail itself of the provisions of the code to dispute the measure. 6. This section draws on Kapur (2002) and Saxenian (2002a).
T. N. Srinivasan
207
at just about the average of 6.9 for low-income countries and one-quarter of China’s 27.6. Indian has 17 Internet users per 1,000 people, just about the average of 16 for low-income countries, but still only one-quarter of China’s 64. India spent 3.7 percent of its GDP on IT in comparison with China’s 5.3 percent (World Bank 2005, tables 1.1 and 5.11).7 Given this apparent backwardness, Kapur rightly asks why India emerged as a leader “in a leading edge industry when, despite strenuous (and, in retrospect, misguided) policies, it failed to achieve such leadership in any other technology-intensive sector” (2002, p. 93) (with the possible exception of pharmaceuticals). He is again right in rejecting as inadequate and incomplete the explanation that the onerous economic control regime that was in place, certainly during 1950 –80 and arguably until the reforms of 1991, had not intervened in the software sector, and given India’s endowment of science and technology manpower, comparative advantage considerations would have enabled the development of the IT sector anyway. At a deeper level, why indeed the state’s role eventually became more facilitative than constraining in this sector remains to be answered. Saxenian (2002a) points out that it is not entirely appropriate to conclude that the self-sufficiency-oriented, insular development strategy of India since 1950 did not affect the development of the IT sector adversely; indeed it did, by restricting imports of computer hardware (even if the importer committed to exporting a certain amount of software) through high tariffs and limits on foreign exchange allocations, and above all by insulating Indian industry from its global counterparts. She notes that IBM was forced to depart from India in 1978, primarily because of its refusal to comply with the requirements of India’s draconian Foreign Exchange and Regulation Act. Kapur (2002) suggests that the departure of IBM and heavy protection of domestic hardware raised the relative costs of hardware and technology acquisition.8 The rising costs induced the industry to develop software skills in 7. According to UNDP’s (2005) technology achievement index (a composite of disparate indexes ranging from patents granted to residents to gross tertiary education enrollment ratio), India ranked sixty-third, behind Trinidad and Tobago with a rank of forty-one and China with a rank of forty-five! India, though included among the group of dynamic adopters, barely managed to escape being included in the groups of marginalized countries (Nicaragua, ranked sixty-fourth, leads the group of marginalized). This says more about the dubious value of this index and others (such as the Human Development Index) put together by the UNDP than about the technological achievement of the countries ranked and groups. 8. It is interesting that the Chinese prime minister started his April 2005 visit to India at Bangalore, India’s IT capital, and there were euphoric statements at official and unofficial levels of the possibility of joint efforts to capture a large share of the global market by capitalizing on China’s capabilities in hardware and India’s strength in software.
208
Brookings Trade Forum: 2005
response. While this is plausible, until the disincentives of the autarkic system were attenuated (and indeed reversed and turned into incentives), the skills so developed would not have led to the spectacular growth of software exports in particular, and of the IT sector in general. Both Kapur and Saxenian note that there was a dramatic turning point in the policy environment for India’s software and IT industries after Rajiv Gandhi became prime minister in 1984. The major policy changes made by his regime included recognition in the Computer Policy announcement of November 1984 of software as an “industry” entitled to the investment and other incentives available for domestic industries; lowering of import tariffs (from 100 percent to 60 percent) on software and personal computers; and the announcement in 1986 of the Computer Software, Development and Training Policy, which liberalized access to the latest technologies and software tools for promoting the domestic software industry, with the expectation of its becoming globally competitive, moving up the value-added ladder, and capturing a significant share of global software exports. The policy allowed import of software in any form, invited foreign investment, and promised access to venture capital. There cannot be a more dramatic departure than this policy from the strategy of technological selfreliance, import substitution across the board (from intermediates to capital goods), and export pessimism. Another important event of the 1980s was the visit to New Delhi in September 1989 by Jack Welch, then chairman of General Electric (GE), and his breakfast meeting with Sam Pitroda, the chief technology adviser to Prime Minister Rajiv Gandhi. It led to GE’s technology partnership with India, which began in 1991.9 The 1984 and 1986 policy initiatives were enabling in the sense of removing policy-created barriers to the growth of the software sector, but the policies did not become proactive until after the reforms of 1991. Saxenian quotes an indus9. “India today earns more than $17 billion from corporations world-wide seeking low-cost overseas talent to do everything from write software to collect debts to design semiconductors. GE in large measure stoked the phenomenon, playing an unheralded role as the Johnny Appleseed on India Inc. and reaping billions in savings for itself along the way. But the strategy has been pivotal for GE. In 2000, it inaugurated a Jack F. Welch Technology Centre in Bangalore that employs thousands of researchers working on everything from new refrigerators to jet engines. This year, the conglomerate plans to spend about $600 million on computer-software development from Indian companies, according to a recent company report. The company estimates that similar products would cost it as much as $1.2 billion in the US. GE also recently unleashed a big new player in Indian outsourcing. In November, it sold a controlling interest in GE Capital International Services, or Gecis, a company with about 17,000 employees that GE started in 1997 to answer mail from its credit-card customers.” Jay Solomon and Kathryn Kranhold, “Early Investments Helped Fuel Tech and Service Sectors,” Wall Street Journal, March 23, 2005.
T. N. Srinivasan
209
try analyst: “Until 1991–92 there was virtually no policy support at all for the software sector. Even the term ‘benign neglect’ would be too positive a phrase to use in this connection” (2002a, p. 172). She notes the lack in the 1980s of international communication links for software exports and cites the example of Texas Instruments which, in setting up the first earth station in Bangalore in 1986, had to negotiate with the authorities for the removal or breaking of twentyfive different government rules relating to export of data via satellite links. The absence of reliable telecommunications links forced Indian firms to be primarily “body shoppers,” who provided programming services on-site, typically in the United States, to customers under contract. All of this changed in the 1990s. Telecommunications reforms, which began in the mid-1980s and included privatization and the creation of a regulatory agency, were very successful, although the process was not smooth because of the resistance of state-owned monopolies. Telephone calls within and from India are perhaps the least expensive in the world. The Department of Electronics introduced the scheme of Software Technology Parks (STPs) in the early 1990s. An STP is the analogue of an export-processing zone. Firms in STPs were allowed tax exemptions, guaranteed access to high-speed satellite links, and provided with reliable electric power and basic infrastructure, including core computer facilities, ready-to-use office space, and communications facilities. They were allowed to import equipment duty-free and without import licenses. Full (100 percent) foreign ownership was permitted in exchange for an export obligation. Firms were also allowed to repatriate capital investment, royalties, and dividends freely once they paid the taxes due.10 The STPs played a major role in the development of the IT sector in the 1990s. The share of units located in STPs in India’s software exports rose dramatically, from 8 percent in 1992–93 to 81 percent ten years later (WTO 2005, box 1, p. 274). However, there is no hardheaded social cost-benefit analysis of the use of public resources in the creation of STPs and the provision of other incentives. There were also several somewhat fortuitous factors. First, as Kapur notes, the restrictions on large business houses entering new fields under the Monopolies and Restrictive Trade Practices Act (which was repealed de facto only after 1991) prevented most of them from exploiting the newly emerging opportunities in the IT sector. Only one of the very successful IT enterprises (namely, Tata 10. China provided all of these incentives and more (in particular, complete flexibility in hiring and firing) in its special economic zones (SEZs) to attract investment, particularly foreign direct investment that was oriented toward export markets. But India’s STPs focused only on software. India’s later embrace of SEZs did not attract much FDI or lead to rapid growth of exports, in contrast to what happened in China.
210
Brookings Trade Forum: 2005
Consulting Services) belonged to a large industrial house. Although large houses had substantial capital and other resources at their command, and precluding or restricting them could have been crippling, this did not happen in the emerging software and IT service industry since it was not very capital-intensive. Besides, small and medium-sized enterprises are usually the most innovative elsewhere in the world, and the Indian IT enterprises were no exception. Second, India had invested disproportionately in higher education in general and, in particular, had established the elite Indian Institutes of Technology (IITs). Although IIT graduates emigrated in significant proportions to the United States and accounted for less than 5 percent of the engineering graduates in India, the impact of émigrés on the development of India’s IT sector, particularly in the 1990s, was significant (more on this in a later section). As in the United States, India’s IT sector is also concentrated, located in clusters in Bangalore, Chennai, Hyderabad, Mumbai, New Delhi, and Pune. These cities also had the highest concentration of public sector R&D establishments (especially defense) as well as publicly funded engineering colleges. Five of these (other than New Delhi) are in the West and in South India. The states in which the five are located, namely Maharashtra (Mumbai and Pune), Karnataka (Bangalore), Tamil Nadu (Chennai), and Andhra Pradesh (Hyderabad), together accounted for 64 percent of the annual intake of engineering colleges in 2003, even though their share of India’s population is only 27 percent (Forbes 2003, table 3). Indeed, these were the states that had the largest expansion of engineering colleges since 1983 when, in a major liberalization, privately funded institutions without state aid were encouraged. In Forbes’s view: “It is this expansion of engineering education that fueled India’s software boom, and it is no accident that the states with massive private expansion of engineering education are precisely those where the software industry is located” (Forbes 2003, p. 8). The Y2K crisis brought prominent global attention to the skills of Indian software engineers. Indians had become good at converting old mainframe and minicomputer applications to Unix applications in the 1980s and early 1990s. It was essentially a tedious task that Americans were not willing to undertake. However, it directly prepared Indians for Y2K by giving them expertise in applications and also the reputation for reliable work. Y2K work accelerated the process, but the previous application conversion work was the base on which the whole thing was built.11 According to DataQuest, 11. I thank Kanwal Rekhi for pointing this out in a private communication.
T. N. Srinivasan
211
Y2K . . . triggered off a chain of events. Exports grew on the strength of Y2K and never looked back. The local training industry boomed partially because of it. . . . Locally, full-page government ads in leading dailies, asking companies to become “Y2K-compliant,” helped the paranoia along. By the end of 1999, the industry was on an all-time high. IPOs of software companies were getting oversubscribed several times. . . . This gave rise to a minor scandal that rose and fell—fly-by-night operators who often had no business at all started putting IT in their names and entering the IPO market. They made money; a lot of investors lost theirs.12 Whether or not it was fortuitous, the 1986 report of the Rangarajan Committee on Modernization of India’s largely state-owned banking sector recommended “standardizing banking systems on Unix, then an unperfected operating system when compared with MS-DOS. The government floated a tender for 400 Unix systems and set off a scramble among Indian companies to come up with a Unix platform. Though the local part of the contract eventually went to Sunray Computers, the report led local vendors into the Unix arena and eventually India’s transformation into a ‘Unix country.’”13 It is possible that Unix would have been attractive to a developing country such as India even had the committee not recommended it: it was a semi-open source developed for larger, more powerful computers than PCs, for which the totally proprietary MS-DOS was developed. Other competitors for Unix were also totally proprietary. Later, in the 1990s, Unix turned out to be ideal for networked computing, and Unix-based systems still dominate the Internet server realm.14 The sources of the spectacular development of India’s IT sector are diverse. There is no doubt that the foundation of a skill base existed for its development in the 1980s, in large part due to public investment in higher education and the creation of elite engineering schools. The public policy regime mattered, both negatively in restricting the potential development of the sector until the mid-1980s, and positively when it changed gradually from enabling its development before the reforms of 1991 to proactively supporting its growth thereafter. Fortuitous or serendipitous factors, depending on one’s point of view, contributed as well. From its beginning in body-shopping and routine programming, the industry has grown
12. DataQuest, “The Hot Verticals: The Great Indian Software Revolution,” December 23, 2002, p. 5 (www.dqindia.com/content/20years/102122306.asp [September 13, 2005]). 13. DataQuest, “The Hot Verticals,” p. 3. 14. I thank Nirvikar Singh for pointing this out.
212
Brookings Trade Forum: 2005
Table 1. The Indian IT Industry, 2003–04 Domestic 2003–04
Growth, 2003–04 (percent)
Exports Value, Growth 2003–04 a (percent)
Units
Valuea
Units
Value
Hardware Systems: Servers Total non-PC servers Total PC servers Total servers
4,393 52,609 57,002
1,170 912 2,082
14 50 46
13 34 21
… … …
… … …
Systems: Workstations Personal workstations Traditional workstations Total workstations
12,078 3,100 15,178
99 122 221
73 29 61
13 44 28
… … …
… … …
Systems: Single-user systems Desktops 2,691,823 Notebooks 90,680 Total PCs 2,782,503
8,014 736 8,750
21 88 22
8 80 12
… … …
… … …
Total systems
2,854,683 11,053
23
14
…
–100
Peripherals Total printers/MFD Other peripherals Total peripherals
1,382,993 … …
1,333 3,163 4,496
21 … …
33 54 47
… 2,200 …
… 72 …
… 2,978 … 18,527 … 3,142
… … …
32 23 39
… 2,300 10
… 59 –67 (continued)
Networking Total hardware Total hardware services
in depth and scope. I conclude this section with a description of the industry as of 2003–04. The hardware component of the industry is still small, accounting for a little over one-fifth of the total value of output (see table 1). Exports accounted for nearly two-thirds of total output, of which software and BPO services had shares of 67 percent and 28 percent, respectively. It employs fewer than a million workers out of a labor force of 363 million. According to India Today, “Still, the buoyant BPO sector is absorbing English-speaking graduates in the thousands. In 2005, IT & ITES will be the biggest job generator, creating more than 2.75 lakh [275,000] jobs. India’s huge cost advantages with quality assurances and large pool of skilled manpower will keep the going smooth. More than 250 of the Fortune 500 firms outsource their IT needs to India. There is more growth
T. N. Srinivasan
213
Table 1. The Indian IT Industry, 2003–04 (Continued) Domestic 2003–04
Growth, 2003–04 (percent)
Exports Value, Growth 2003–04 a (percent)
Valuea
Units
…
1,710
…
–10
27,755
18
… … …
3,300 1,951 6,961
… … …
30 69 24
1,586 10,309 39,650
–8 11 15
…
1,450
…
53
16,380
45
Training Services Total training services Total services
… 992 … 12,545
… …
–12 26
110 56,150
29 22
Grand total
… 33,374
…
24
59,550
24.5
Units Software Services Customized software Turnkey projects Consulting/others Total software services Business process outsourcing (BPO)
Value
Source: DataQuest, “Intellectual Property: India: Sleeping IP Giant” (www.dqindia.com/dqtop 20/2004 [December 2005]). a. Rupees Crore (107 rupees); US$1= 48 Rs.
in store” (India Today, March 2, 2005, p. 14). The same story reports that the top five IT and ITES firms pay an annual salary of $4,500–$6,250 for an engineering graduate, roughly ten to twelve times the per capita income of the country, one-sixth of the annual average U.S. salary in 2000, but less than that of an employee of the occupations at risk of being offshored in the United States (see also Bardhan and Kroll 2003, table 4). The Indian industry is no longer confined to producing and exporting low-end software products and services. Several multinational companies (MNCs), including many leading ones, have established software development centers in India. DataQuest reports that such MNC centers are filing for patents in large numbers (1,108 in 2002–03).15 It suggests that intellectual property revenues would constitute a major chunk of a software company’s revenue in the future, and Indian companies (other than MNCs), including some of the large ones, have not yet started preparing for it. Leading Indian IT firms, such as Infosys and Wipro, are multinational with offices around the world and employ nationals in these countries. Infosys has alliances with the world’s leading firms, including 15. DataQuest, “Intellectual Property: India: Sleeping IP Giant” (www.dqindia.com/dqtop20/ 2004 [December 7, 2005]).
214
Brookings Trade Forum: 2005
IBM, Intel, Microsoft, and Oracle, and also has made strategic acquisitions of foreign firms. The story is similar for Wipro; it is aggressively looking at companies to acquire in the enterprise, finance, and consulting areas. NASSCOM (2004, p. 9) documents the increasing maturity of the industry following a large number of mergers and acquisitions in 2002. It noted that traditional IT service players have added ITES-BPO portfolios to their existing offerings in order to provide a complete umbrella of end-to-end services. Multivendor and build-operate-transfer (BOT) contracts that offer customers advantages such as low risk, scalability, and competitive pricing have increased. Indian vendors are expanding the spectrum of their service offerings in client locations and even setting up facilities in other low-cost ITES-BPO destinations such as China and the Philippines in order to tap those markets. They are also moving up the value-added ladder to offer high-end services such as equity research and analytics, as well as insurance and technology support and development. Moreover, Indian vendors have moved far beyond call centers into financial services, telecom, retailing, and automotive segments of the ITES-BPO sector. In financial services, Indian companies are offering customers services centered on accounting, billing and payment services, and transaction processing. Over the past few years, some Indian service providers have also been offering highervalue services to customers in the areas of insurance claims processing and equity research support. They expect to gain from offshore outsourcing of customer and technical support and product development by the global telecom industry; transaction processing, billing, telemarketing, and inventory management for large retailers; and engineering activities, such as computer-aided product and tool design, claims processing, and accounting processes for the automobile industry (NASSCOM 2004, p. 10). The report also benchmarked the performance of Indian industry on key operational issues with global benchmarks. It found that Indian industry is able to deliver at levels comparable to their international counterparts on parameters such as quality, customer satisfaction, and between quality and customer satisfaction. Finally, let me turn to some relatively recent developments in the provision of services to tourists and visitors to India. As the home of an ancient civilization and of several ethnic and religious groups (Hindus, Buddhists, Jains, Sikhs, Muslims, Christians, Jews, and Zoroastrians) over millennia, India has many ancient monuments, including churches, mosques, synagogues, and temples besides the Taj Mahal. Of course, the diversity of its flora and fauna, the peaks of the Himalayas, and other natural beauty also attract tourists. However, because of the inadequacy of affordable quality hotel rooms, transport, and communications, India failed to attract as many tourists as even much smaller
T. N. Srinivasan
215
countries of the region. This is changing: tourist arrivals grew by 24 percent in 2004, and India is fast emerging as one of the top ten tourist destinations in the world, according to the Reserve Bank of India (RBI 2005a, p. 80). However, unconventional tourism is also on the rise. In particular, medical tourism is growing. According to several recent newspaper reports, the availability of worldclass treatment at a fraction of the cost in the United States or Europe is attracting up to 100,000 to 150,000 foreign patients to quality hospitals across India, representing a tenfold increase over just five years ago. Many of these hospitals were established and are staffed by Indian doctors who had emigrated to and practiced for a long time in the United States and Europe. Another unconventional service is education—institutions of higher education are attracting students from other South Asian countries. Recently, two premier institutions, the Indian Institute of Management at Bangalore and IIT in Mumbai, have started programs in Singapore in collaboration with the National University of Singapore. Also, drug trials and sponsored research in the pharmaceutical industry are being outsourced to India.16
The Indian Diaspora and ITES Sector The IT revolution, particularly the development of the Internet, has spawned networks of engineers and scientists who are transferring technology, skill, and know-how between distant locations, and more flexibly than most corporations.17 Among foreigners in the United States, Indians were second only to South Koreans as recipients of U.S. Ph.D.’s in engineering and science in 2003. Indian engineers were at the helm of a significant and growing number of Silicon Valley–based technology companies. The proportion of companies run by Indians has grown from 3 percent of those started between 1980 and 1984 to 10 percent of those started between 1995 and 2000, and is probably even higher 16. India’s success in producing generic equivalents of patented drugs is owed in part to India’s decision since 1970 to grant only process and not product patents in the pharmaceutical industry. This success became internationally visible when Indian companies became able to supply generic antiretroviral drugs to treat AIDS far more cheaply than the multinational companies. With India becoming a signatory of the TRIPS (Trade Related Aspects of Intellectual Property Rights) agreement, Indian patent law had to be amended in March 2005 to grant product patents as well. There is a real danger that the generic industry will be destroyed by this amendment (see Abbott, Kapczynski, and Srinivasan 2005). On the other hand, because the Indian pharmaceutical industry has built a reputation for quality, drug trials and sponsored research are being outsourced to India. 17. This section draws on Saxenian (2002b).
216
Brookings Trade Forum: 2005
among those started after 2000. Further, data from the U.S. Bureau of the Census show that the Indian share of Silicon Valley’s science and engineering grew from 2 percent (400 workers) in 1970 to a significant 13 percent (20,000 workers) in 2000, most of the increase having taken place in the 1990s.18 Members of the two professional associations of Silicon Valley, The Indus Entrepreneur (TIE) and Silicon Valley Indian Professionals, and those who held senior positions in U.S. companies, were instrumental in convincing their senior management to establish operations in India. As India built a reputation as a supplier of software, most Silicon Valley technology companies established their own development centers in India. The dot.com bust resulted in some Silicon Valley–based Indians returning to India and setting up enterprises of their own. Saxenian (2002b) rightly stresses the growing influence of the Silicon Valley Indian community in Indian policymaking. One of the doyens of the community, often called the “sage of Silicon Valley,” Kanwal Rekhi, former chief technology officer at Novell, has been a vocal advocate of policy change in India. In his regular visits to India, he has met with senior policymakers at the central and state levels. He deserves much of the credit for the breathtaking scope of India’s telecommunications reform and its success. But for his very public attack on the entrenched bureaucracy of the Department of Telecommunications (DOT), it is unlikely that the reform would have proceeded very far.19 K. B. Chandrasekhar, another Silicon Valley entrepreneur, led a committee in 1999 on venture capital (VC) for the Securities and Exchange Board of India. The committee’s report provided a comprehensive vision of the growth of India’s VC industry. The industry has had remarkable growth since the report. In 2004 nonresident Indian (NRI) entrepreneurs led by the Silicon Valley giants prepared an action plan for attracting substantial foreign direct investment (FDI) to India and presented it to the government in January 2005. It has been well received and some of its recommendations have already been implemented. There is no doubt that their success in the global market is the main, if not the only, reason for their not only being heard with respect but actively sought after by policymakers. Three prominent Silicon Valley entrepreneurs, Vinod Khosla (of Sun Microsystems fame), Kumar Malavalli (of Brocade Communications Systems), and Kanwal Rekhi, realized that to change policy successfully, three ingredients are essential: sound, policy-oriented research from academics to provide the foun18. I thank AnnaLee Saxenian for drawing my attention to these data. 19. He coined the Hindi slogan “DOT Hatao-Desh Bachao,” which means “Curb the DOT and Save the Nation!”
T. N. Srinivasan
217
dation for policy advice; discussion and debate of the proposed policy advice in a forum consisting of academic researchers, prominent politicians, bureaucrats, and media representatives; and monitoring of the implementation of the advice. The three took the initiative to provide significant resources to Stanford University’s Center for International Development for sponsoring and undertaking policy research on India’s economic reform and organizing a conference at which research findings and policy recommendations following from them are presented. Five annual conferences have been held since 2000; two conference volumes have been published, and a third is in press. The sixth conference was held in June 2005. Besides supporting research at U.S. universities, the Silicon Valley community has also been active in upgrading Indian institutions, starting from the IITs, of which many of them are alumni, and also creating new institutions, such as a business school in Hyderabad whose faculty includes regular visitors from top U.S. business schools. Some Silicon Valley entrepreneurs, hailing from different states in India, are realizing that policy reforms are urgently needed at the state level in India. A group of TIE members interested in Kerala have formed a Kerala Support Group. Others are likely to follow. Finally, at the request of India’s prime minister a group of entrepreneurs and business persons in the Indian diaspora, including veterans of Silicon Valley, got together in fall 2004 and came up with practical suggestions for improving FDI flows to India. Their proposals were presented to the prime minister and the Planning Commission in January 2005. The subsequent removal of some dysfunctional restrictions on FDI and the legal action creating special economic zones were in part the result of these proposals. It is clear that the influence of the Indian IT diaspora is spreading beyond the narrow confines of the Indian IT industry to influence India’s economic development and growth more broadly. The demonstrable success of India’s IT and software sector in the global market is having a profound impact on Indian industry and also, importantly, on the educated youth. From a fear of competition, as evidenced, for example, by the demand for antidumping measures (ADMs) against cheap imports, particularly from China (a demand that the government was too willing to agree to, making India the largest user of ADMs among members of the WTO in the last couple of years), there is now a growing sense of confidence among Indian industries of being able to compete and a desire to view China as a large and growing market to export to, with the result that trade with China has grown rapidly. An important aspect of achieving and maintaining global competitiveness is attention to product and service quality. Again, the reputation for quality that IT vendors have built has not gone unnoticed. India is now poised to become a major destination for outsourcing of
218
Brookings Trade Forum: 2005
manufactured products, as China is already. This is already happening in auto components—Indian suppliers, like the Chinese, are competitive both in cost and quality (Sutton 2005). Indian automakers have recently begun exporting passenger cars: from 25,000 cars in 1998–99, exports grew to 121,000 in nine months of 2004–05 (MOF 2005, table 7.5). Last, there is a sense of confidence among the educated youth about the future and in their ability to compete successfully with the best of their cohort from anywhere in the world, a confidence that was noticeably absent not so long ago. These intangible effects are no less important than their tangible ones on the growth of exports, GDP, and employment. Among the most important intangible effects of India’s perceived economic success broadly, and in high-technology sectors including IT and the pharmaceutical sectors in particular (both achieved without in any way compromising its vibrant democratic political system), one must include a vast improvement in India’s international standing. There is a recognition that India is a major Asian power, along with China and Japan. It is no coincidence that, in a short span of six weeks in March and April of 2005, high-level visitors to India included the Chinese prime minister, the U.S. secretary of state, the president of Pakistan, and the UN secretary general. India is a serious contender for permanent membership on the UN Security Council if the proposed expansion of the council comes about, though this seems unlikely in the foreseeable future.
The Analytics of the IT Sector in the Growth Process There are many possible channels through which IT could affect an economy’s output (its level and possibly rate of growth in a steady state). Singh (2004) focuses on one, namely the reduction of transaction costs through the use of information technology.20 In his model, a reduction in transaction costs increases the number of intermediate goods that are produced and in turn influences growth. A representative household supplies labor inelastically and has a logarithmic instantaneous utility function for a single consumption good produced by identical competitive firms with constant returns to scale technology using labor and a symmetric composite of differentiated varieties of intermediate goods best viewed as perishable producer services. The composite is a con20. Information technology can reduce transaction costs in several ways, for example, by reducing search and matching costs, by speeding up and making more reliable the completion of transactions, by substituting long-distance communications for physical transportation, and by improving tracking and logistics of delivery. These benefits can be obtained in transactions for intermediate or final goods and services.
T. N. Srinivasan
219
stant elasticity of substitution (CES) aggregate of available varieties at each point in time, with an elasticity of substitution greater than one. This implies that aggregate production must be increasing in the number of varieties of intermediates (in a symmetric monopolistically competitive equilibrium in which all the available varieties cost the same and are used in the same amount). All varieties are produced with the same technology with a fixed start-up cost and a constant variable cost, both in units of labor. Finally, transaction costs are modeled à la Samuelson’s iceberg transportation costs: intermediate goods producers receive only an exogenously set fraction of what the producers of the consumer goods pay them; the remaining fraction melts away as a transaction cost. Thus any fall in this exogenous fraction is a reduction in transaction cost. There are no forces in this model that can generate sustained growth, since the labor force is given, and in contrast to Helpman’s (1990) model, there is no learning effect that continuously reduces the fixed cost of producing a variety of intermediates as their cumulative output grows. Thus Singh focuses on the properties of a steady state, if it exists. It turns out that the relative value of the elasticity of substitution between labor and the intermediates aggregate in the production of the consumption good and that between different varieties in the intermediate aggregate, , is what matters. Both are assumed to exceed one. If ≤ , there is a unique value n* for the number of intermediates n, such that if the economy starts with n0 < n*, it converges to n* over time, and if it starts with n0 > n*, it stays at n0. Thus initial conditions do not matter too much, in the sense that any economy starting from n0 below n* converges to it. However, n* is a decreasing function of the transaction cost. If > , there are three possibilities depending on other parameters, including the utility discount rate and transaction costs: (i) n stays at n0 , so that any initial n0, is also a steady state; (ii) there are two values n *L and n *H with n *L < n *H such that if n0 ≤ n*L, it stays at n0 ; if n *H ≥ n0 > n *L, n converges to n *H ; and if n0 > n*H, n stays at n0 . Thus any initial n0 in the interval (0, n *L ] or [ n *H ,) is a possible steady state. However, for an initial n0 in ( n *L , n *H) the steady state is n *H ; and (iii) again, there are two values n *L and n *H as in (ii), but in this case, even if n 0 < n *L , the n can converge to n *H so that for any initial n0 in (0, n *H ] including the value n *L, the steady state is n *H . Summarizing, transaction costs have three impacts: first, the standard deadweight loss; second, with higher transaction costs, few intermediates are produced, leading to lower output of the consumer good and hence lower welfare; and third, they may arrest the process of change in the sense of keeping the number of varieties of intermediates at their initial level and/or reduce their long-run level. Although the model is useful in illustrating the possible cost of high transaction costs, as noted earlier, its structure precludes the analysis of growth effects.
220
Brookings Trade Forum: 2005
The model in Singh (2002), though not as fully worked out as the above, addresses the growth impacts of IT (including nondigital methods of storing and communicating information) through an extension of a model of recombinant growth developed by Weitzman (1998). The Weitzman model captures the simple concept that new ideas are formed through combinations of old ideas. Singh follows Weitzman in focusing on this concept to the case in which ideas can be combined in pairs. Given the possible combinations of pairings, the number of ideas grows exponentially. By decomposing the total stock of knowledge or ideas into the stock of IT knowledge and non-IT knowledge, Singh captures the special role of IT knowledge. He specifies that the stock of IT knowledge affects the growth in stocks of both types of knowledge in the same manner, so that IT stock gives the growth process an “extra kick” even beyond the exponential growth that the Weitzman model produces.21 Singh conjectures that the model has a steady-state growth rate, but its comparative statics with respect to the parameters of the “extra kick” term are yet to be worked out. Let me conclude this section by listing the mechanisms through which IT could affect the level and growth of output. First, as in the Singh models, IT services are in effect universal intermediates (like energy) that are essential to any production activity and possibly most, if not all, consumption activities. Thus any technical progress in the IT sector reflects itself first in productivity gains (or cost reductions) in the IT sector. Then, as the changes in the IT technology diffuse, as the rest of the economy makes the appropriate investments in equipment and processes to take advantage of the new lower-cost IT technology, other sectors of the economy will experience productivity gains. Because the diffusion process is likely to be gradual, the total factor productivity (TFP) gains will be spread over time. The debate in the United States about the contribution of computers and IT to productivity growth is in large part about this diffusion effect (see Jorgenson 2005; Gordon 2000). In India also, the diffusion that has begun would yield TFP gains over an extended period of time. The mechanisms through which IT affects the level and growth of output are essentially aspects of the diffusion process. First, as IT diffuses, improvements in the efficiency of resource use would affect the level of output growth both sec21. The justification of this special role for IT comes from the importance of being able to store, process, and communicate information effectively: without writing, without telephones, without the Internet, the success rate of converting potential new ideas into actual additions to the stock of knowledge would be lower. As an illustration, the use of IT makes the IT sector itself more efficient and innovative (for example, using software for automated testing during software development) as well as providing this benefit to other sectors (for example, using IT to improve the workings of call centers, or to reduce mistakes in medical transcription).
T. N. Srinivasan
221
torally and, even more important, intersectorally, as gainful transactions (market and nonmarket) that were previously unavailable became available. Already, improvement in market price information, and hence higher revenues, are being realized in some parts of rural India. Another area that positively affects both the level and growth of output is IT-based delivery of education at all levels.22 In principle, an extensive and effective use of IT in public administration could enhance transparency, as well as reduce avoidable time lags and transaction costs. It is also possible that increased transparency in itself could reduce administrative corruption and malfeasance. It is too early to judge from several ongoing e-governance pilot projects whether these expected beneficial effects are being realized (Singh 2002, p. 18). However, it is early enough to recognize the existence and rapid growth of several private (profit-oriented as well as nonprofit) initiatives in addition to publicly funded schemes in India. Singh mentions several of them. There is reason to hope that the diffusion of IT will gather steam and contribute significantly to accelerating growth in the not too distant future. Because IT goods and services are tradable internationally, productivity gains from their diffusion to the entire economy should occur regardless of whether such goods and services are locally produced. Obviously, whether a country produces an intermediate good domestically or imports it is the aggregate outcome of the decisions of its producers of final products to buy the good from local or foreign suppliers. Indeed, these decisions are related to the organization of firms in the sense that a vertically integrated firm producing a final product produces the intermediate products it needs, whereas less integrated firms would purchase some of the intermediates they need from others (in particular from suppliers who specialize in the production of intermediates); that is, they would outsource such intermediates. When such outsourcing leads to offshoring, the intermediates would be imported. The offshore producer could also be a subsidiary of the offshoring multinational firm, so the purchase transaction would be internal to the firm. Thus, in general the problem of whether to outsource or offshore has to be embedded in an analysis in which the organization of firms (whether to be vertically integrated, or whether to become multinational—that is, to engage in FDI) is endogenous. In a series of papers Grossman and Helpman model the determinants of the location of subcontracted (that is, outsourced) activity in a general equilibrium model of outsourcing as trade, integration versus outsourcing in the equilibrium 22. Singh (2002) cites the projects “e-choupal,” linking the Indian farmer with national and international markets, and TARAhaat as examples.
222
Brookings Trade Forum: 2005
organizational structure of industries, and outsourcing versus FDI (2002a, 2002b, and 2003). An essential feature of these models is that outsourcing of an intermediate requires searching for a partner and making relation-specific investments that are governed by incomplete contracts. Naturally, finding a suitable partner is easier if the market for intermediates is thick (and thickness can differ between domestic and foreign markets). Also, whether the intermediate has to be customized to the specification of a single user (in which case, once it is produced, the supplier has no other user for it than the one for whom it was customized) or is usable by many, matters for contracting, since the possibility of the supplier, once he has produced the customized intermediate, being “held up” by the user can arise. IT products, such as some software, have characteristics that often involve customization. Moreover, such characteristics are at best observable by the supplier and purchasers, but not verifiable by third parties, thus precluding contracts between suppliers and purchasers that stipulate a given price for an agreed quantity to be purchased. For a closed economy Grossman and Helpman (2002b) analyze domestic outsourcing in such a context, based on the tradeoff between costs of running a larger and less specialized organization in a vertically integrated firm and the costs arising from the search for a suitable partner to outsource and imperfect contracting, were it to outsource. Turning to FDI, Grossman and Helpman (2003) set forth three possible equilibria. In one, all firms choose to invest in a subsidiary abroad to produce their intermediates there. In another, all choose to purchase them from independent suppliers in the foreign country. In the third, some firms opt to engage in FDI and others in purchase. Which of the three is relevant depends on the values of three parameters representing, respectively, the price elasticity of demand for any variety of the differentiated final product (assumed to be greater than one), the marginal cost of producing the substitute in a subsidiary abroad, and the difficulty of contracting with an independent foreign supplier. With marginal cost sufficiently low, elasticity of demand sufficiently low (that is, close to one), and the difficulty of contracting sufficiently great, in equilibrium all firms engage in FDI. By contrast, with marginal costs neither too low nor too high, elasticity of demand above 1 but by not too much, and a moderate difficulty in contracting, in equilibrium some firms do engage in FDI and others purchase. It is conceivable that the search and contracting costs will be lower if both supplier and purchaser happen to be national firms. If this is the case, the gains from having domestic firms supplying and other domestic firms purchasing IT products could be large in comparison with domestic purchasers buying and contracting with foreign firms. For this reason, it is likely that firms in non-IT sectors in India would pur-
T. N. Srinivasan
223
chase their IT services and products from existing domestic IT firms rather than import them. Also, if contracting difficulties are relatively fewer and marginal cost of production relatively higher in India than in China, the Grossman and Helpman (2003) model would predict more FDI in China than in India.
Future Prospects of the Indian IT Sector: Conclusions The spectacular growth of software exports as well as ITES-BPO has been underpinned both by the relative abundance of skilled workers and by their relatively low cost. NASSCOM (2004, p. 15) points out that the success of the Indian ITES industry is in large part due to the country’s immense pool of English-speaking skilled workers. NASSCOM (2005) estimates the stock of graduates (engineering degree and diploma holders, degree holders in the arts, commerce, and science) at 22 million in 2003, with approximately 2.5 million graduates in 2004 from about 275 universities and 14,000 colleges in the country. Of course, those numbers do not adjust for differences in the quality of skills and are not broken down by the skills required for working in different segments of the IT sector. For example, the skills required to be employed in a call center are surely different from those for developing software. NASSCOM (2005) estimates that 250,000 engineering degree and diploma holders entered the workforce during 2003–04, with a large segment believed to have joined the IT industry. The industry is estimated to have employed 841,500 professionals in 2003–04, of which 270,000 worked in the export sector and 253,000 in the BPO sector. Still, if one assumes that the share of the stock that is employed in the industry does not change, the implied annual growth rate of the stock at 11.3 percent (the ratio of graduates to stock) is far below the growth of 35 percent that NASSCOM (2004, p. 15) projects for the value of India’s ITES/IT services output for the period 2003–12. Thus, unless the ITES/IT sector’s share of the stock grows or the stock itself grows more rapidly, there could be excess demand for labor in the sector, and the wages and salaries of ITES/IT workers would have to increase to wipe out the excess demand: While demographic studies have suggested that India could be one of the few countries with a surplus of personnel within the employable age group by 2020, there is a possibility of a shortage in terms of availability of skilled personnel for ITES/IT, even in the medium term. This gap could be to the tune of 235,000 for IT services and 262,000 for IT-enabled services and could increase in 2012 in the absence of any
224
Brookings Trade Forum: 2005
special efforts to meet the manpower requirements. (Government of India 2003, p. 6) Any increase in emoluments of ITES/IT workers in India with no change in their productivity would cut into the cost competitiveness of India. It is not easy to obtain reliable and well-documented unit costs of IT sources and products for Indian suppliers relative to their competitors in other countries. The following data from reports of Evalueserve, a full-service business intelligence, market research, and intellectual property services firm, are suggestive: the costs reported do not include profit margins, which run around 10–15 percent in India, but do include marketing and sales costs: For IT services, costs range from $11 per hour to $19 per hour for work being done in India. $11 per hour essentially corresponds to the lower end IT work that is being done by a college graduate (with a bachelor’s in computer architecture) or a graduate engineer. On the other hand, $19 per hour is for higher-end work being done by people with four or five years of experience. For ITES the costs range from $9 per hour to $22 per hour (for work being done in India); $9 per hour is essentially credit card processing and other low-end, nonvoice work. For low-end voice work (such as call centers), this goes up to $11 per hour. On the other hand, $22 per hour is for higher-end work like investment banking research, intellectual property research, etc. In either case, Indian companies typically charge for 2,050 to 2,100 working hours per year per full-time-equivalent (FTE), although an average FTE in India is currently working 2,300 hours, which is the same as in South Korea (Alok Aggarwal of Evalueserve, private communication with the author, August 15, 2005). A significant component of unit cost is labor. Table 2 puts labor cost differences in perspective. At the low end of the skill spectrum (the first three rows of occupations in table 2), India is likely to remain competitive for the foreseeable future. At the higher end of the spectrum (the last three rows), India’s competitive edge is likely to be eroded, with Indian wages rising relative to the U.S. wage. On the other hand, productivity and quality improvements that will mitigate, if not completely offset, cost increases are also taking place. Saxenian, in a private communication of May 5, 2005, points out that in interviews Indian IT professionals informed her of “the process and quality improvement in the largest Indian firms (TCS, Wipro, Infosys, etc.), especially in the period since
T. N. Srinivasan
225
Table 2. Hourly Wages for Selected Occupations, United States and India, 2002–03 U.S. dollars Occupation
Hourly wage, United States
Hourly wage, India
Telephone operator Health record technologists/medical transcriptionists Payroll clerk Legal assistant/paralegal Accountant Financial researcher/analyst
12.57 13.17 15.17 17.86 23.35 33.00 –35.00
Under 1.00 1.50 –2.00 1.50 –2.00 6.00 –8.00 6.00 –15.00 6.00 –15.00
Source: Bardhan and Kroll (2003, table 3).
2000 when substantially more work was shifted to India from the U.S., allowing for the accumulation of skill and learning on the job.” Her own research found that there are a “surprisingly large number of Indian firms that are CMM [capability maturity model, developed by the Software Engineering Institute at Carnegie Mellon University] Level 5 certified. There are none in China at Level 5 yet, and few even in the U.S.” Another interesting set of cost comparisons relates to medical tourism. Jay Solomon reported in the Wall Street Journal (April 26, 2004) that India’s “Apollo [hospital] offers cardiac surgery for about $4,000, compared with at least $30,000 in the U.S. Apollo’s orthopedic surgeries cost $4,500, less than one-fourth the U.S. price.” John Lancaster reported in the Washington Post (October 21, 2004, p. A01) on a patient with a life-threatening heart condition who would have had to undergo surgery at a cost in the United States of $200,000; instead he flew to New Delhi and had it done at Escorts Heart Institute and Research Centre. It cost him $10,000, including airfare and a side trip to the Taj Mahal. Not only are costs lower, but quality may be higher as well at India’s private centers of excellence, such as Escorts. Lancaster quotes Naresh Trehan (a former assistant professor at New York University’s Medical School) of the Escorts Centre as saying that the “death rate for coronary bypass patients at Escorts is 0.8 percent. By contrast, the 1999 death rate for the same procedure at New York-Presbyterian Hospital, where former president Bill Clinton recently underwent bypass surgery, was 2.35 percent, according to a 2002 study by the New York State Health Department.” A third report, by Saritha Rai (New York Times, April 7, 2005) speaks of a patient from England who needed a coronary bypass operation but would have had to wait six months to get it from British National Health Services. He traveled to Wockhardt Hospital in Bangalore for the surgery following a chance
226
Brookings Trade Forum: 2005
meeting with a businessman who had gone to India for surgery. His surgeon there had trained in London, and the surgery cost $8,400, including travel. Moreover, according to the patient, his surgeon gave him his cell phone number and was available twenty-four hours a day. He concluded that in the British Health System “you are just a number, but here you are a person.” Such stories are frequently reported in the U.S. media. As I noted earlier, 100,000–150,000 foreigners currently visit India for medical treatment annually. A report by McKinsey & Co. projects a revenue of $2.3 billion by 2012 from medical tourism to India. It is not out of the realm of possibility that India could go beyond offering inexpensive but high-quality surgery at hospitals and develop as a destination for the elderly to live out their retirement years in a warmer climate and with better health care than they could obtain in the United States or Europe. Although the prospects for substantial growth in the ITES/IT sector, as well as in in situ services such as medical tourism, are very bright, realistically speaking there are several constraints besides a potential manpower shortage that could preclude their full realization. Reliable electric power, efficient and inexpensive telecommunications, and access to venture capital are essential infrastructures for the IT sector. Although telecommunications infrastructure has vastly improved, as noted earlier, there are still some unresolved issues relating to the authority of the regulatory agency (Telecommunications Authority of India, TRAI) vis-à-vis the Department of Telecommunications and the stateowned providers. The electric power situation continues to be abysmal. In fact, the large IT firms, like other large enterprises, have had to invest in their own captive power generation facilities. To the extent that the unit cost of power from small-scale captive plants is much higher than it would be from an efficient large-scale utility, the failure of India’s public power system adds an avoidable cost to doing business and dampens the competitiveness of its IT sector. Narayana Murthy, the CEO of India’s leading IT firm, Infosys, and Sandeep Raju of the same firm pointed out in 2002: Efficient commercialization of cutting-edge output from research labs, entrepreneurship forums at universities, highly efficient alumni networks, close links between leaders in academia and business, risk appetites of venture capitalists, synergies between science/engineering schools and business schools, collaborative research among universities, keiretsus bringing together businesses and venture capitalists, angels with the willingness to nurture talent, the abundance of forums where youngsters may put forth their ideas and interact with industry leaders, opportunities for
T. N. Srinivasan
227
collective learning—all these are differentiators that put the Valley several notches above other hi-tech habitats. In sum, Silicon Valley operates in a vibrant market economy that reveres innovation. That Bangalore entrepreneur, on the other hand, does not have easy access to all these resources. However, it must be borne in mind that the information revolution is a fairly recent phenomenon in Bangalore. (Murthy and Raju 2002, p. 201). Things have improved substantially since they wrote. The venture capital supply has increased, and some of the networks they mention are beginning to emerge. But significant close links between academia and business are yet to be forged. Young workers in the IT industry are conscious of their market value, so much so that turnover of workers in the ITES sector, particularly in smaller firms operating at the lower end of the quality spectrum, is high. In call centers, annual turnover is reported “to exceed 50 percent. High staff turnover is reported even among the more established, employee-friendly IT companies, some of whom offer stock options and residential accommodations to entice employees to stay on” (Chithelen 2004, p. 1023). High turnover indicates an excess demand and makes it more expensive for Indian firms to maintain and improve their competitiveness and quality of service.23 Although India’s labor force is very large, and more workers would like to acquire the skills to enter the IT sector, setting up training facilities would be costly and involve years of planning and implementation. Chithelen notes that the explosive growth in BPO demand has attracted new entrants to the business, and the resulting competition is reported to have bid down fees from $20 per programming hour for U.S. entrants currently from over $60 in 2000. He fears that greater competition among vendors, some of whom are aggressive but inexperienced new entrants, coming on top of labor supply constraints, could compound the decline in quality of service. India’s labor and bankruptcy laws could be counterproductive in the IT sector, as they are in other sectors of the economy. A report in 2000 by the Subject Group of Knowledge-Based Industries in the Prime Minister’s Council recommended exempting the IT sector from some of the draconian provisions of labor laws. Whether it is wise to exempt one sector from a dysfunctional law rather than repealing it is arguable. In any case, political support for a repeal is not there
23. Kanwal Rekhi points out in his private communication that the high turnover in the ITES sector is in the nature of the beast. This industry essentially employs youngsters who do not consider it a career. Many of them go on to higher learning. He cites the increase of Indian students in U.S. universities over the past several years as evidence of that.
228
Brookings Trade Forum: 2005
yet. However, some de facto exemptions do exist. For example, in ITES, states often exempt call centers from working-hour restrictions, allowing women to work at night. Also, it is likely that programmers in large firms are not subject to the same provisions as industrial workers.24 Moving away from domestic to external constraints, I have already mentioned the protectionist backlash against outsourcing in the United States and elsewhere. Apart from this, the outcome of negotiations on services in the Doha Round of world trade talks is also important. In the parlance of the General Agreement on Trade in Services (GATS), there are four modes of services trade. From the perspective of outsourcing and trade in ITES/IT services more generally, the two modes that are particularly relevant for India are mode 1, which covers outsourcing—that is, the supply of a service from the territory of one member of the WTO to the territory of another member without movement of the use and provider of the service from where they are located; and mode 4— that is, trade by a service supplier of one member, through the presence of natural persons of a member in the territory of any other member. In plain language, mode 4 involves temporary migration of labor from the territory of one member to that of another. Members undertake commitments under each mode of supply. Thus far, members have made most cross-border commitments for mode 3, relating to foreign establishment. Commitments for mode 1 are much narrower and more limited, with a range of restrictions involving nationality, residency, authorization, and local authentication requirements. Commitments on mode 4, perhaps the most relevant mode for most developing countries, are the fewest and most restrictive. Many of the IT professionals from India in the United States have been admitted to the United States as temporary immigrants under H-1B visas. The annual number of such visas to be issued is determined by Congress, in part on the basis of market conditions for labor with the requisite skills and in response to lobbying. It is to be hoped that before the WTO ministerial meeting in Hong Kong in December 2005 the lacunae in GATS will be addressed. India is not the only country with a pool of English-speaking workers available for employment in the ITES/IT sector. Other countries with such populations include Bangladesh, Ireland, Pakistan, Sri Lanka, and the Philippines. Except in Bangladesh, the wage costs are higher than India’s in the other countries. The ability to speak English can, of course, be acquired, and as such, potential future competition for India from countries currently without a significant pool of English-speaking workers cannot be ruled out. Prominent among such 24. I thank Nirvikar Singh for pointing this out.
T. N. Srinivasan
229
countries is China. Yahya (2002) notes that there are two competing schools of thought among Indian policymakers about China’s efforts to expand its software sector: “One school believes that China is a serious competitor and that India should not assist its progress in any way. . . . The other school argues that India has a lead of two to three years over China and notes that India should use China as a market and collaborate with it” (Yahya 2002, pp. 114–15). I have already noted that, given the various agreements signed and the press releases during Chinese premier Wen Jiabao’s visit to India in April 2005, the collaborationist school seems to have won. In fact, as Yahya points out, China’s former premier, Zhu Rongji, during his visit to India in January 2002, also expressed the view that China and India could dominate the world IT market if they combined forces. The report of the India-China Joint Study Group on Comprehensive Trade and Economic Cooperation, presented to the prime ministers of the two countries by their commerce and industries ministers during Wen Jiabao’s visit, recommended the following: Companies from India and China should continue to explore each other’s markets. The two countries can use their core competencies in hardware and software to increase their share of [the] world’s trade and gain greater access in third country markets. Industry association such as NASSCOM may closely interact with its counterpart organization in China to promote co-operation in this sector. Both countries should work together to enforce copyright and reduce piracy. Moreover, the software enterprises of both countries can easily adapt to the ever changing high-tech market and strengthen their status in [the] global market by having more joint research projects, enhancing the exchange of technological personnel and cooperation in training.25 The report also noted significant scope for collaboration between the two countries in the market for services in several areas, including health, accounting and auditing, education, finance, and advertising. India’s IT and ITES-BPO sectors have firmly established a global reputation, and the potential for sustaining their rapid growth is bright. The success of the industry’s premier organization, the National Association of Software and Service Companies (NASSCOM), is being emulated by associations in the pharmaceutical, computer hardware, and auto parts manufacturing industries and 25. “Report of the India-China Joint Study Group on Comprehensive Trade and Economic Cooperation” (www.hindu.com/thehindu/nic/0041/report.pdf [May 4, 2005]).
230
Brookings Trade Forum: 2005
others. Importantly, like NASSCOM, they are succeeding not only in lobbying but also in raising standards and quality.26 The growth of the IT sector could contribute significantly to accelerating GDP growth and add a large number of wellpaid jobs. Faster GDP growth and the indirect employment created by IT sector growth would also contribute to poverty reduction. There are, however, some serious domestic and external constraints that, if left unaddressed, could substantially diminish the realization of the potential growth.
Conclusion The visible success of the IT sector raises two important questions. First, can India’s future growth be IT-led, in the sense that the IT sector would not only be the prime engine of growth but also, because of its increasing share of GDP, compensate for slower growth in the nonservice sectors such as agriculture and large-scale manufacturing? Second, to the extent that the success of the IT sector could be linked to policy reforms—in telecommunications, for example— would it broaden and strengthen political support for reforms in other sectors? In attempting to answer the first question it is useful to think about the backward and forward linkages in IT sector growth. Its backward linkages arise from the growth in demand for inputs of goods, services, and factors. In particular, the demand will rise rapidly for labor with appropriate skills, for institutions that can educate and train workers for employment in the IT sector, and for telecommunication services and power. Its forward linkages could be significant as well, if other sectors of the economy invest in hardware and software to reap the productivity advantages of using IT services in their operations. It was noted earlier that the Ministry of Finance projects that value added by the IT sector will reach 7 percent by 2008, and at that rate a share of 25 percent is likely by 2020. Taken together, the prospects for the IT sector’s becoming the leading sector of the economy in the next decade or so are bright. However, accelerating the rate of GDP growth to 8 percent or more per year and sustaining it for several decades is a necessary, though not sufficient, condition for achieving the overarching objective of India’s development, namely, the eradication of poverty. As is well known, a large majority of India’s population is rural, and more that half of the country’s labor force still depends on agriculture and informal sector employment for a livelihood. And among the states 26. Kanwal Rekhi points this out and rightly emphasizes that such quality consciousness bodes well for the future of Indian industry.
T. N. Srinivasan
231
there is a significant variation since 1980 in economic performance, as measured by growth of state domestic product. The IT sector is concentrated in the cities of only a few states. Although the IT sector’s spectacular past and likely future growth will almost surely help to accelerate aggregate growth, translating faster growth into more rapid eradication of poverty would require widening, deepening, and accelerating economic reforms and liberalization. The response to the second question, whether the prospects for bringing about the required reforms are likely to be enhanced by the contribution of past reforms to the visible success of the IT sector, is a qualified yes. The reasons for qualification are many. First, the link between reforms and success of the IT sector is not clearly seen even by reform-minded politicians. Second, a vast majority of the population, particularly those rural areas, has yet to experience the benefits of efficient and inexpensive IT services in production and consumption activities, in part because IT services have not yet reached this population. Although this is changing, it is not changing rapidly enough to affect a large share of the population. Third, although there is no political pressure to reverse the reforms enacted thus far, it is evident that the remaining items of the reform agenda to be adopted would require persuasion of and bargaining with the parties not in the ruling coalition and its allies. Unlike in China, with its authoritarian, single-party government, success in this effort of persuasion in the competitive democratic polity of India requires foresighted leadership and the commitment of all parties. Just as the evident success of early agricultural reforms in China and the participation of a large share of the population in the fruits of its success created a constituency for later reforms, one can expect a similar effect in India once the benefits of IT success, and reforms more broadly, diffuse to the larger Indian population. It is encouraging and augurs well for the future that no political party (except perhaps the unreconstructed Stalinist elements of the left) is demanding a reversal of the reforms already enacted, and the differences among parties are more on the pace and sequencing of further reforms than on their content.
Comment and Discussion
Anne Krueger: As always, T. N. Srinivasan has provided an excellent analysis of the Indian economy’s performance—this time of the role of ITES (information-technology-enabled services) in past and prospective economic growth. His analysis of the growth of the IT sector is masterly, and I have only a few comments on it. I will then focus on the insights that the experience of the IT sector provides for Indian economic policy and growth more broadly. Turning first to the IT sector, its performance has been spectacular by any standard, especially in contrast to the rather sluggish performance of many Indian economic activities. But it should be noted that there are several unique characteristics of the IT sector that may have enabled it to grow so rapidly. First and foremost, IT is less heavily dependent on infrastructure than most industries are. Srinivasan notes that twenty-five different government rules had to be changed or removed in order to set up the first earth station in Bangalore in 1986. But the fact is that the industry could rely on an earth link and avoid many of the cumbersome aspects of Indian infrastructure: it hardly needed Indian roads or railroads, telecommunications (minimal as they then were), or Indian ports. For many other industries, rules would have had to be altered and infrastructure and other bottlenecks removed. While the IT sector had to live with the same constraints regarding other aspects of infrastructure as other industries, the effect on their cost was arguably considerably less. A significant feature of the IT sector is that most of the major firms have campuses in Bangalore and a few other cities, and these campuses are virtually self-sufficient. They have their own generators and do not depend on public provision of power. Company-owned buses even take people to and from work so 232
T. N. Srinivasan
233
they don’t have to rely on public transportation. Indeed, in Bangalore, some of the large firms are concerned that they may be unable to sustain their rapid growth because, as their employment increases, road congestion will be too great to absorb additional buses. The IT firms perceive the need to fight with the local authorities for capacity expansion for virtually every publicly provided urban service for which private investment cannot substitute. Infrastructure was less of a constraint for IT, but another factor was important as well. The IT sector was “encouraged,” as Srinivasan notes, but mostly by removing the disadvantages that other industries have. The sector was “discriminated for” in the sense that it did not suffer the disadvantages the government imposes. For one thing, the industry was new, and in a sense it outran the government by growing rapidly before regulations could be put in place. But, as Srinivasan also notes (all too modestly, since he had a significant role in it), the telecommunications reforms of the 1990s and later were highly significant. The IT sector probably could not have grown as rapidly as it did in the absence of those changes. One can only wonder what industry or industries would experience comparably rapid growth if the transport, or the labor market, or the power, bottlenecks, and regulations were removed for other sectors of the economy. One final comment on the IT sector’s growth prospects. Srinivasan has a table on wage differentials and notes that wages are very low by international standards. The relevant comparison should, of course, be unit labor costs, and a comparison there would be highly worthwhile. During the NAFTA debate in the early l990s, for example, much was made of the fact that average factory wages in the United States were some ten times the average Mexican factory wage. But once the calculation of unit labor costs was undertaken, it was found—not surprisingly—that unit labor costs were quite similar, and indeed several percentage points lower in the United States than in Mexico. Clearly, there is great scope for productivity improvement in India, in IT and elsewhere, and that can enable rising wages and sustained economic growth. Moreover, one would have to guess that in many IT services (where the physical capital input is very low) unit labor costs are significantly lower in India than in the major industrial countries and that, for that reason, there is some scope for real compensation increases in India in the context of rapid IT expansion. But I want to focus on overall Indian growth. A first point relates to the excellent Indian Institutes of Technology, which have provided a steady flow of world-class engineers since they were first set up in the late 1950s. When they were established, it was in the belief that India had a “shortage” of engineers.
234
Brookings Trade Forum: 2005
The number of student places in the institutes was determined by estimating the number of “needed” engineers and providing sufficient places to generate the “needed” stock in a fairly short time. Not surprisingly, before too long there was an excess supply of highly qualified engineers. That is a partial explanation for the large diaspora of Indian engineers in Silicon Valley and elsewhere, as well as of the advantage the IT sector had in recruiting its personnel. A strong case can be made that India overinvested in higher education for engineers and that overall growth may have been significantly more rapid had some of those resources been allocated to increasing places in primary, secondary, and technical education. This brings me to my second point. India remains a country with a very large quantity of unskilled labor. No matter how successful the IT industry is, India is going to have to use its abundant supply of unskilled labor more productively and provide primary and secondary education and training for an increasing fraction of its labor force. To achieve rising living standards, that will be essential, especially in light of the fact that 70 percent of India’s population is still located in rural areas. To utilize its unskilled labor more productively, innumerable further policy reforms are needed. There are considerable labor market rigidities, which undoubtedly serve as a disincentive for firms to hire unskilled labor: in the “organized sector,” as it is called, firing workers is illegal (although some businesses have learned that if they do not pay their electric bills their electricity is shut off and they are forced to close, thus solving their labor problems). There are requirements for training workers, for provision of housing, and for other services. There was a “small-scale reservation” (SSR) policy, under which more than 800 small, labor-intensive industries were identified as eligible for privileges (such as tax exemptions) provided that they did not grow large. They engaged in activities such as candle-making, radio assembly, and production of batteries, and large firms were forbidden to enter these activities. Interestingly, these industries, and exports from them, grew much more slowly than would have been expected.1 Logically, though, small-scale firms cannot be expected to have the resources or the capacity to develop international markets. While enterprising businessmen were able to have many “companies” owned by various relatives side by side in one building, and thus circumvent the small-scale requirement to some extent, it was still a major barrier and disincentive to expansion and exporting. 1. See Mohan (2002).
T. N. Srinivasan
235
Ironically, when China began its rapid industrial expansion, based largely on unskilled labor-intensive goods, Chinese firms were able to penetrate the Indian market in many of the SSR industries. In the past several years, the government of India has begun to remove industries from the SSR list, but it still includes about 350 reserved industries. I have already noted the poor quality of Indian infrastructure. It would probably be absolutely insane to plan a business without including provision for the costs of generators. And transport costs are high; port delays are long; and there is still excess demand for airplane flights and services, though they have greatly improved in recent years.2 And although the government of India has invested in education, there are many problems. Teachers are not in school; they are absent and tutoring for higher pay than they could obtain in the schools (but still collecting their pay from the school). In a recent survey, the state with the lowest rate of teacher absenteeism, Maharashtra, had an absentee rate of 15 percent. In some states, it was more than 40 percent. With rates such as those, of course, parents may decide to have their children work at home, or in the field, since even if they go to school, they might simply have to return home.3 The litany of ills could continue, but I will stop after mentioning two more. The first is the regulatory environment that still exists (despite improvements) and that is surely a major productivity-reducing block for most private sector activities.4 Bureaucratic red tape and delays constitute a major obstacle and deterrent to efficient production and expansion.5 The second is the very large fiscal deficit of the general government (almost 10 percent of GDP in 2004) with India’s debt-to-GDP ratio already above 80 percent. Yet the fiscal deficit must be addressed at the same time that the Indian government finds means for improving infrastructure. That is in part because much of the existing pattern of expenditures is inefficient, with untargeted subsidies intended to benefit the poor going largely to the rich, and the need for thorough-going tax reforms.6 In my judgment, India has tremendous growth potential. Reforms have proceeded, albeit much more slowly than might have been desirable to attain significantly higher growth rates. They have been undertaken in a functioning 2. See Forbes (2002). 3. See Kremer and others (2005). 4. See World Bank (2006), p. 129. 5. See Shourie (2004). 6. See Srinivasan (2000).
236
Brookings Trade Forum: 2005
democracy, which is certainly a huge plus for India. But much more is needed by way of reform, and soon. While the IT sector will surely continue to contribute to growth, it cannot absorb the greater part of India’s abundant labor force. IT can increase growth somewhat over the coming decades, but the rapid increase India needs will depend on the success of other reforms. Certainly, Indian economic prospects for the coming decade or two are brighter than they were around 1990. A growth rate of around 6 percent is probably sustainable if reforms continue at their present rate. But with more reforms, the 8 or 9 percent growth that India needs would become attainable. The IT sector’s development is not only a major success story; it is also an indication of what could be achieved. General Discussion: Anne Krueger’s comments stimulated an animated discussion on the pros and cons of the enclave-driven growth strategy taken by India relative to the counterfactual of a broader-based growth path. Robert Litan pointed out that India’s growth story might not be deemed a success if the full opportunity cost of broader policy reform and social investment were taken into consideration. For instance, the significant resources allocated to the engineering schools and elite education might have had higher social returns if directed instead to broader education or more infrastructure. But Litan also allowed that this is not altogether evident, owing to the considerable waste that is likely to accompany broad social spending because of Okun’s leaky bucket. Alan Deardorff wondered whether we should credit India’s “awful policies” for having created a comparative advantage in software that would not have existed in a more benign overall economic policy environment. In a similar vein, Kimberly Clausing noted that she was struck by the evidence presented in T. N. Srinivasan’s paper of beneficial growth effects from a significant concentration of engineering graduates, against the backdrop of significant illiteracy in the overall population. This circumstance diverges sharply from the general emphasis in development economics on investing first in universal access to primary education and working up the education ladder only as a country gets richer. Lael Brainard asked whether the apparent economic spillovers from enclavedriven growth in this case were also apparent in the domain of political economy. India’s last election certainly did not appear to confirm that the IT sector had spawned a strong political voice for broader economic reform. Rafiq Dossani endorsed Anne Krueger’s notion that policy reforms had had a redemptive effect, noting that many of the reforms starting in the mid-1980s were mainly redeeming bad preexisting laws and policies. He cited a recent
T. N. Srinivasan
237
change to a 1952 law that prohibited women from working at night—the activity that is the mainstay of the entire call center industry. Robert Feenstra questioned the potential magnitude of economic spillover effects from the IT industry’s success into the critical agricultural sector—still the main source of employment in India. He cited a discussion with India’s secretary of agriculture, who mentioned that certain parts of the country could grow three crops a year, but were prevented from doing so by constraints on the water supply. Attention then shifted to the role of the diaspora and elite educational institutions in the success of India’s software industry. Rafiq Dossani raised doubts about the significance of graduates of the elite Indian technology and management institutes as well as the Indian diaspora in the rise of India’s software industry. He said that none of the top ten IT companies in 1970, 1980, 1990, or 2000 were started by someone who did his undergraduate degree at one of the Indian Institutes of Technology or by a member of the diaspora. Dossani refuted the notion that Silicon Valley Indians played a large role in India’s IT boom, noting that they invested little in India’s IT industry. Indeed, he said, it appeared Silicon Valley Indians did not have the right expertise, since success in the Valley depended on developing technology products, while the India IT industry initially prospered by delivering customized software to the financial services industry. As the Indian industry is becoming more technology oriented, however, the involvement of the diaspora is increasing. Martha Laboissiere largely agreed. She pointed out that the Indian diaspora is found mainly in senior positions in the United States or Europe, and has contributed to India’s software industry by elevating India as the location of choice for offshoring, rather than through direct participation in the Indian industry. This contrasts with members of the Chinese diaspora, who are more likely to return home as middle managers. Laboissiere also underscored concerns about demand outstripping the supply of suitable engineers in India. Despite a high number of engineering graduates each year in India, headhunters consider only about 15 to 20 percent of these suitable for recruitment by offshore service providers. As a consequence, she predicted that demand would outstrip supply in India sooner than the 2012 timeframe predicted by NASSCOM, and that the crunch would come as early as 2008 in some places, such as Hyderabad. Moreover, Laboissiere emphasized that middle managers are in particularly tight supply in India, in part because of recruitment in India by eastern European companies. Srinivasan responded by reiterating his preference for broad-based over enclave-driven growth—for making all of India a zone for economic development
238
Brookings Trade Forum: 2005
rather than relying on special economic zones. Nonetheless, he reasserted his optimism that the IT sector might be driving growth through important indirect as well as direct effects because of its role as an essential intermediate input. He noted that India probably should prefer South Korea’s path—6 to 7 percent growth for several decades followed by a short but sharp financial crisis—to the path it actually took: no financial crises and three decades of 3.5 percent growth only recently giving way to growth above 5 percent. Srinivasan observed that India’s investment in engineering education had been driven by Nehru—following the Soviet dictum that investing in engineering and electricity would solve the problems of development—rather than by any rational cost-benefit analysis. He agreed with Rafiq Dossani that the elite engineering and management schools were quantitatively small, and he believed they produce no more than 5 percent of the 250,000 engineers joining India’s labor market each year. But he contended that they have much greater impact on policymaking and the highest levels of business than at the managerial or entrepreneurial levels. Finally, Srinivasan agreed with Feenstra’s concerns about agriculture, noting that credit constraints and lack of domestic market integration are important impediments, along with policy distortions that lead to inefficient water use.
References Abbott, Frederick, Amy Kapczynski, and T. N. Srinivasan. 2005. “The Draft Patent Law,” Hindu, March 12, 2005 (www.thehindu.com/2005/03/12/stories/ 2005031201151000.htm). AIMA. 2003. India’s New Opportunity—2020. Report of the High Level Strategic Group. New Delhi: All India Management Association and Boston Consulting Group. Bardhan, Ashok Deo, and Cynthia A. Kroll. 2003. “The New Wave of Outsourcing.” Research Report. Berkeley, Calif.: Fisher Center for Real Estate and Urban Economics. Chithelen, Ignatius. 2004. “Outsourcing to India: Causes, Reaction and Prospects.” Economic and Political Weekly, March 6, pp. 1022–24. Forbes, Naushad. 2003. “Higher Education, Scientific Research and Industrial Competitiveness: Reflections on Priorities for India.” Paper presented at the Fourth Annual Conference on Indian Economic Reform held at Stanford University, June 5–7, 2003. http://scid.stanford.edu/events/India2003/Priorities_India.pdf. ———. 2002. “Doing Business in India: What Has Liberalization Changed?” In Economic Policy Reforms and the Indian Economy, edited by Anne O. Krueger, pp. 129–67. University of Chicago Press. Gordon, Robert J. 2000. “Does the ‘New Economy’ Measure Up to the Great Inventions of the Past?” Journal of Economic Perspectives 14 (4): 49–74.
T. N. Srinivasan
239
Government of India. 2003. Ministry of Communications and Information Technology Department of Information Technology. “Task Force on Meeting the Human Resources Challenge for IT and IT Enabled Services.” Grossman, Gene, and Elhanan Helpman. 2003. “Outsourcing versus FDI in Industry Equilibrium.” Journal of the European Economic Association 1 (2–3): 317–27. ———. 2002a. “Outsourcing in a Global Economy.” Working Paper 8728. Cambridge, Mass.: National Bureau of Economic Research. ———. 2002b. “Integrating versus Outsourcing in Industry Equilibrium.” Quarterly Journal of Economics 117 (1): 85–120. Helpman, Elhanan. 1990. “Monopolistic Competition in Trade Theory.” Special Papers in International Finance 16. Princeton, N.J.: Princeton University Department of Economics. Jorgenson, Dale. 2005 forthcoming. “Accounting for Growth in the Information Age.” In Handbook of Economic Growth, edited by Philippe Aghion and Steven Durlauf. Amsterdam: North Holland. Kapur, Devesh. 2002. “The Causes and Consequences of India’s IT Boom.” India Review 1 (2): 91–110. Kremer, Michael, Nazmul Chaudhury, F. Halsey Rogers, Karthik Muralidharan, and Jeffrey Hammer. 2005. “Teacher Absence in India: A Snapshot.” Journal of the European Economic Association 3 (2–3): 658–67. MOF. 2005. Economic Survey 2004–2005. New Delhi: Ministry of Finance, Government of India. ———. 2004. Economic Survey 2003–2004. New Delhi: Ministry of Finance, Government of India. Mohan, Rakesh. 2002. “Small-Scale Industry Policy in India: A Critical Evaluation.” In Economic Policy Reforms and the Indian Economy, edited by Anne O. Krueger, pp. 213–67. University of Chicago Press. Murthy, N., R. Narayana, and Sandeep Raju. 2002. “Comment on Chapters 4 and 5.” In Economic Policy Reforms and the Indian Economy, edited by Anne O. Krueger, pp. 194–203. University of Chicago Press. NASSCOM. 2005. “IT Industry Communiqué for the Academic Fraternity.” An IT Workforce Development Initiative 1. New Delhi: National Association of Software and Service Companies (www.nasscom.org/download/issue-2-july-sep-05.pdf. [September 29, 2005]). ———. 2004. Indian ITES-BPO Industry Handbook 2004. New Delhi: National Association of Software and Service Companies. RBI. 2005a. Macroeconomic and Monetary Developments in 2004–05. Mumbai: Reserve Bank of India. ———. 2005b. Reserve Bank of India Bulletin, April 13, 2005. Mumbai: Reserve Bank of India. Saxenian, AnnaLee. 2002a. “Bangalore: The Silicon Valley of Asia?” In Economic Policy Reforms and the Indian Economy, edited by Anne O. Krueger. University of Chicago Press.
240
Brookings Trade Forum: 2005
———. 2002b. “Transnational Communities and the Evolution of Global Production Networks: The Cases of Taiwan, China and India.” Industry and Innovation 9 (3): 183–202. Shourie, Arun. 2004. Governance and the Sclerosis That Has Set In. New Delhi: ASA Publications. Singh, Nirvikar. 2004. “Transaction Costs, Information Technology and Development.” Revised version of paper presented at a conference in honor of Pranab Bardhan’s contribution to the Journal of Development Economics and to the larger field of development economics, September 24–25, 2004, Harvard University. ———. 2002. “India’s Information Technology Sector: What Contribution to Broader Economic Development?” Paper prepared for the conference “The IT/Software Industries in Indian and Asian Development,” November 11–12, 2002, Chennai, India. Srinivasan, T. N. 2000. Eight Lectures on Indian Economic Reforms. New Delhi: Oxford University Press. Sutton, John. 2005. “The Globalization Process: Auto-Component Supply Chains in China and India.” In Are We on Track to Achieve the Millennium Development Goals? edited by François Bourguignon, Boris Pleskovic, and André Sapir. Washington: World Bank. UNDP. 2005. Technology Achievement Index. http://hdrc.undp.org.in/hds/rgnl/TAI.htm. Weitzman, Martin. 1998. “Recombinant Growth.” Quarterly Journal of Economics 113 (2): 331–60. World Bank. 2006. Doing Business in 2006. Washington: World Bank and IFC. ———. 2005. World Development Indicators. Washington: World Bank. WTO. 2005. World Trade Report. Geneva: World Trade Organization. ———. 2004. International Trade Statistics. Geneva: World Trade Organization. Yahya, Faizal. 2002. “The Dragon Arises: China’s Challenge to India in Software Development.” India Review 1 (4): 91–122.
RAFIQ DOSSANI Stanford University
Globalization and the Offshoring of Services: The Case of India
T
he overwhelming majority of exports from developing countries to developed countries consist of agricultural commodities and manufactured goods. Such goods are usually produced under contract to a buyer from a developed country, the buyer managing design, marketing, and sales, while the seller handles production. Intermediate steps, such as accessing finance, technology, and raw materials, managing currency risks, and maintaining quality control, are shared between the buyer and the seller in special arrangements. Typically, in the introductory stages, the buyer assumes more control and takes more risk than in later stages. As the seller matures, it shares more of the risks and rewards. Nevertheless, history shows that the share of rewards (that is, the economic rents, if any) tend to remain with developed-country buyers. A new phase of globalization, international trade in services, has been emerging for at least a decade, and by now it looms important within the total value of World Trade Organization (WTO) trade. Developing countries around the world, particularly in Asia, have become large producers of services for developed countries. The range of such services is impressive. It includes back-office services such as payroll; customer-facing services such as call centers and telemedicine; design services such as the design of application-specific integrated circuits; research services such as conducting clinical trials; venture capital provision, from Taiwan to Silicon Valley, for example; software services such as programming; and IT and infrastructure outsourcing such as the managing of
The author thanks Arvind Panagariya, an anonymous referee, and the participants at the 2005 Brookings Trade Forum for helpful comments.
241
242
Brookings Trade Forum: 2005
corporate e-mail systems and telecommunications networks. These new fields of service exports join the traditional fields of tourism and labor migration. As a result, services are becoming an important component of international trade for both developing and developed countries.1 Hence, studies of changes in the composition and trends of trade in services between developed and developing countries and the analysis of sharing of risks and rewards between buyers and sellers can provide insights into an important component of international trade and the new challenges and opportunities for economic development in the developing world. Some have even argued that the ability of countries like China and India to undertake high-end services work, such as in semiconductor design and information technology (IT), can threaten employment in developed countries, if productivity gains in developing countries are sufficiently high (for example, Samuelson 2004). In this paper, I use India as a case study of growth and value addition. India has catapulted itself in recent years into the leadership position in services exports from developing countries, and the study of India’s experience with exporting services in information technology could provide important insights into the workings of this new and promising area of international trade. Services appear to be key to its future growth (see table 1). Some components of the Indian experience (such as software services) are over three decades old, enabling us to draw useful insights about how a developing country can succeed in exporting services to developed countries. This is likely to have implications for policy in institutional development (such as educational and risk management institutions) and physical infrastructure. I argue that (1) local entrepreneurship and a high level of infant industry protection enabled the Indian IT industry to reach a high growth path and allowed local skills to develop rapidly to keep pace with global changes; (2) protectionism—in force until 1990—hurt the industry by forcing the work done to focus on low-value-added components; and (3) the rising importance of multinational firms from the late 1990s onward was the result of the country’s adopting a more welcoming legal and regulatory structure and led to a rise in the sophistication of work done. As a result, the contractual relationship between India-based IT providers and developed-country buyers has changed over the past three decades. Ini-
1. According to WTO statistics, services accounted for about 20 percent of global trade in 2003. The fastest growing category, commercial services, accounts for half of all services trade and grew at 15 percent in 2003, somewhat faster than merchandise trade (13 percent). WTO tables (www.wto.org/english/res_e/statis_e/its2004_e/its04_toc_e.htm).
Rafiq Dossani
243
Table 1. India’s IT Exports as a Share of Total Exports, 1999–2000 to 2003–04 Billions of U.S. dollars, unless otherwise indicated
Total exports IT-enabled services software exports Share of total (percent)
1999–2000 a
2000 –01
2001–02
2002–03
2003–04
37.54
44.89
44.91
52.51
63.2
4.52 12.0
7.14 15.9
9.14 20.3
12.04 22.9
15.8 25.0
Source: NASSCOM (2004), pp. 26, 63; Indian Economic Survey (2004), p.101, and www.dgciskol.nic.in/. a. The Indian financial year is from April 1 to March 31.
tially, it was characterized by “body-shopping”: the provider recruited labor and the buyer decided how to use it (Wortzel and Wortzel’s stage 1, see table 2). Later, applications programming became more prominent (stage 2). Today, the relationship is characterized by a mix of the two, though stage 1 work remains dominant.2 Do these findings suggest that developed countries are likely to be only marginally threatened by the globalization of services? After all, if the Indian software industry case is typical, then high-end work is likely to stay with developed countries. The problem for developed countries is that not everyone in developed countries can readily shift to high-end work. The shift of developed nations’ economies toward service-based economies beginning in the 1960s certainly increased the number of highly skilled service workers, but there was an even greater rise in the number of less-skilled service workers. This is partly a consequence of the nature of many services as linked, inseparable sets of activities provided with different levels of skill, with a pyramid of labor requirements— that is, lower-level work employs more people than high-end work. In manufacturing, the unemployment created by the reduction in demand for blue-collar labor in developed countries was offset by the absorption of much of the surplus labor into services, often with minimal training. But the shift from low-end service workers to high-end workers will require a longer period of reeducation and may have significant interim consequences on unemployment rates and costs. The threat to developed countries is increased by the fact that, apart from software, the largest growth in offshoring is happening in business services. These are also the sectors with the largest growth in U.S. employment. Goodman and Steadman (2002, p. 3) found that in 2002 more than 97 percent of the jobs added to U.S. payrolls were in services. Of these, business services and health care accounted for more than half of the growth. 2. See Wortzel and Wortzel (1981).
244
Brookings Trade Forum: 2005
The threat to developed countries may grow if India’s success with IT exports and with other business processes is replicable by other developing countries. India’s history of private sector entrepreneurship has continued uninterrupted for many years, and, along with it, many institutions have been created—legal institutions, stock markets, educational institutions, and others—that have helped it succeed.3 To the extent that such entrepreneurship and institutions are weak in some developing countries (including China), those countries will have difficulty replicating India’s success. In this paper, I first discuss the two main services that have been offshored to India, business processes and software services.4 I then analyze the factors that enabled India to succeed in the business and conclude by examining the impact of offshoring on employment patterns, organizational structure of work, and value retention in developed countries.
The Globalization of Services The novelty of services in international trade warrants some definitional classifications as basic as what a “service” is. Most would agree that manufacturing is a process that involves the transformation of a tangible good. It would also generally be agreed that manufacturing does not require face-to-face contact between buyer and seller. Usually, manufacturing creates a good that can be stored, thereby allowing a physical separation of the buyer and the seller. Services have been defined by the opposites to these qualities: as transactions involving intangible, nonstorable goods, requiring that client and vendor be face to face while the service is being delivered. For example, Gadfrey and Gallouj (1998, p. 6) define services as goods that are “intangible, cosubstantial (i.e., they cannot be held in stock) and coproduced (i.e., their production/consumption requires cooperation between users and producers).” This was obviously true 3. Whitman (1990), quoted in Schware (1992, p. 148), argues that the factors for success in software are: computerization in industry and schools; university R&D in software and direct interactions with industry; skilled labor; funding sources such as venture capital and government contracts; support services such as telecommunications infrastructure; social networks among players such as engineers, managers, marketers, and funders; an entrepreneurial culture; an attractive, low-cost work environment and access to market channels through joint ventures and cooperative arrangements. India has several of these factors in place (Dossani 2004) 4. I collaborated with several others on various parts of the research presented here: with Martin Kenney on business process offshoring and with Anita Manwani on the Agilent case (see references). These sections of this paper reflect joint work.
Rafiq Dossani
245
when the service required face-to-face experience, such as receiving a haircut (a “customized” service, in the phraseology of Yotopoulos [1996]), but also true when the service experience did not require customization, such as when a bank’s client wanted to check the bank’s home loan offerings, or did not require proximity, as when she wanted to check her bank balance. For these reasons services are intrinsically more difficult to offshore than manufactured goods. Indeed, examined as a totality, services appear to resist relocation. Very few service operations can be done only on the computer (the modern form of “mundane work” in the terminology of Thompson, Warhurst, and Callaghan [2001]). Most services require at least some level of face-to-face interactivity, either among coworkers or with persons outside the organization, such as vendors and clients. The new twist in the provision of services is that the required interaction between the seller and the consumer has been substantially limited. Advances in information technology made possible the parsing of the provision of certain services into components requiring different levels of skill and interactivity. As a result, certain portions of the serviced activity—which might or might not be skill-intensive but required low levels of face-to-face interactivity—could be relocated offshore. The sequence of events that enabled this process is the following. First, the digital age allowed (or at least revolutionized) the conversion of service flows into stocks of information, making it possible to store a service. For example, a legal opinion that earlier had to be delivered to the client in person could now be prepared as a computer document and transmitted to the client by e-mail, or better yet, encoded into software. Easy storage and transmission allowed for the physical separation of the client and vendor as well as their separation in time. It also allowed the separation of services into components that were standardized and could be prepared in advance (such as a template for a legal opinion) and other components that were customized for the client (such as the opinion itself) or remained nonstorable. Taking advantage of the possibility of subdividing tasks and the economies that come with the division of labor, these developments reduced costs by offering the possibility of preparing the standardized components with lower-cost labor and, possibly, at another location. The second fundamental impact of digitization was the conversion of noninformation service flows into information service flows. For example, a buyer can often examine virtual samples of tangible goods over the Internet instead of visiting a showroom. Once converted to an information flow, a service may then be converted into a stock of information, as noted earlier, and subjected to
246
Brookings Trade Forum: 2005
the forces of cost reduction through standardization of components and remote production. Third, by enabling low-cost transmission of the digitized material, the digital age accelerated the offshoring of services. Services such as writing software programs that were offshored to India in the early 1970s were enabled by digitized storage and, in the 1980s, by the standardization of programming languages. Still later, as digital transmission costs fell in the 1990s (just as digital storage costs had fallen earlier), even nonstorable services, such as customer care, could be offshored. The sequence of events that enabled offshoring did not happen all at once and may not even be standardized. Consider the evolution of sophistication in services as an analog to manufacturing. Table 2 provides a framework of changes in the relationships between a developed-country (DC) buyer and the less-developedcountry (LDC) seller. At each higher stage the process becomes increasingly decommodified with the increase in the exporter’s reputation. This and other similar frameworks, such as the “global value chain” framework,5 do not, however, (a) imply that moving to new stages is automatic, happens in the same sequence, or is time-bound; (b) provide conditions for movement to new stages; or (c) predict that the exporter will capture a rising share of the economic rents. At the very least, the need for costly global coordination can hinder movement to higher stages. Such coordination will initially be done by the developedcountry buyer and enable it to earn a rent for doing so. In addition, much of the market-related coordination and networking will occur through developedcountry institutions, enabling a further retention of value in the developed country. This may be why some countries in Asia have failed to go beyond the initial stages, such as the Philippine back-office industry. Even the much-vaunted Taiwanese semiconductor industry specializes in subcontracting for developedcountry designers of chips—that is, it lies in Wortzel’s stage 2. Examples of rising through stages are known, such as the Israeli software industry and the Korean cell phone manufacturers; and some East Asian economies have seen increased employment, rising wages overall, and reduced poverty as a result of exporting services;6 more commonly, though, developingcountry industries have not reached parity with their developed-country clients. In fact, few go beyond stage 3 work.7 The inference is that certain key aspects, such as deciding on the product and its specification, design, marketing, and 5. Gereffi and others (2001). 6. Humphrey (2004, p. 5). 7. Scholte (2000).
Rafiq Dossani
247
sales, are usually retained by the importer. When that happens it appears to hinder the exporter’s ability to rise to new stages of growth and the rewards are overwhelmingly gathered by the developed-country buyer. Note, however, that stagnation within a stage of growth and stagnation in value share are not identical. For example, OPEC’s members largely have stagnated in stage 1, but their value share has fluctuated with time relative to higherstage work such as refining and marketing petroleum products, owing to fluctuations in world oil prices. More generally, stagnation within a stage can result from a lack of skills, capital, infrastructure, and institutions relative to international competitors, technology, or other interdependence at higher stages that makes separation of higher stages from the developed-country location too costly (lack of “modularization”).8 Stagnation within a stage is likely to lead to a reduction in value share. This is because, to enable outsourcing, a production process first needs to be modularized; that is, the costs of coordination have to be manageable, and the technical and other relevant parameters need to be codified. Implicit in modularization is a shift from proprietary to standard inputs, design, and fulfillment techniques. To the extent that standardization (or, as it is often termed, “commodification”) lowers barriers to entry, it will reduce the share of value through competition. On the other hand, stagnation in value share can happen even if the developing country’s industry rises to higher stages of growth. This can happen if developed-country buyers have market power vis-à-vis their vendors (which may or may not be linked to owning nonstandard processes), so that they obtain better terms of trade regardless of their vendors’ level of sophistication. Hence, moving to higher stages of growth may not be associated with rising value share.
The Offshoring of Business Processes Business processes (BPs) is the catchall term used for the myriad white-collar processes that any bureaucratic entity undertakes in servicing its employees, 8. Modularization is defined as the conversion of a component of the production process with one or more proprietary inputs, design, or fulfillment techniques into a component with standardized inputs, design, and fulfillment techniques. For example, understanding end-user needs is a requirement for a developing-country firm to move from stage 3 to stage 4. If such work is only possible through pilot projects undertaken on clients’ sites, the management of the pilot project must be an integral part of the requirements analysis. In this case, modularization is not feasible, so it may not be possible for the developing country to do such work.
Table 2. Stages of Growth in Offshoring
Stage Stage 1 (assembly) Stage 2 (original equipment manufacture) Stage 3 (original equipment manufacture) Stage 4 (original design manufacture) Stage 5 (original brand manufacture)
Importer’s rolea (A)
Transferred to exporter (B)
Shared roles (C)
Exporter’s roles outside the initial vendor relationship (D)
Custom software industry’s analog to column B (E)
Design, input sourcing and quality control External design and specification for internal design External design
Production capacity
None
None
Body-shoppingb
Sourcing, application of internal design
Quality control
None
Programming of applications
Specification for internal design
Quality control
Add customers
System design and integration
Purchase from exporter’s catalog
Product selection, external design, quality control Product innovation
None
Add sophistication to product range
System architecture and R&D
Competitor
None
System consulting and business development
Source: Wortzel and Wortzel (1981) and Hobday (1995) for stages and columns A–D. Column E by author, based on table 6. a. The rows indicate the stages of product development; the columns show the sharing of work between the importer and exporter. Columns A–D are relevant to manufacturing; column E provides the analogs to service for the particular case of software production. b. Body-shopping refers to providing the client with programmers. The domestic firm recruits the programmers, and the client decides how to use them, often on the client’s own sites in the developed country.
Rafiq Dossani
249
Table 3. Exports of Information Technology–Enabled Services (ITES) from India, 1999–2000 to 2003–04
ITES exports (millions of $) Employment (number) Revenue/employee ($) U.S.-firm comparable revenue/employee ($)
1999–00
2000 –01
2001–02
2002–03
Average growth 2003–04 (percent)
565 42,000 13,452
930 70,000 13,286
1,495 106,000 14,104
2,500 171,000 14,620
3,600 245,500 14,664
…
…
…
…
58,598
59 55 2.2 …
Source: Nasscom (2004, pp. 63, 64, 186).
vendors, and customers.9 These include human resources, accounting, auditing, customer care, telemarketing, tax preparation, claims processing, document management, and many other chores necessary for the firm to function. In the 1990s, India became the largest provider of such offshored services. As table 3 shows, the cost savings to companies that use these services can be significant. The services most offshored are call center services, employing about 200,000 persons at the end of 2003, or about 70 percent of the BPs offshored to India since 2000. This is a large industry in the United States: one recent study estimated that call centers alone employ as much as 3 percent of the U.S. workforce; some estimate that the share will increase to 5 percent by 2010.10 Most of the offshored jobs are relatively low-skilled jobs. For example, the single largest category is outbound calling for the financial services industry to sell financial services such as mortgages or to collect on overdue receivables. The work is routine; workers follow scripts that pop up on the computer screen in response to prompts. However, there is evidence that higher-skilled jobs can be offshored or might evolve on their own. For example, the author interviewed an Indian firm that had initially been contracted by an American firm to call its clients with overdue
9. We define a business process as a complete service, such as handling a customer complaint, processing a medical claim, or processing a purchase order. Completing a process requires undertaking a set of activities. For example, in handling a customer complaint it is necessary to understand the complaint, decide on a course of action, undertake the action, and follow up to ensure that the action solved the complaint. Each of these activities is potentially separable from the others. 10. “The Customer Care Workforce: Driving More Profitable Customer Interactions,” CRM Project, vol. 3, October 30, 2002 (www.crmproject.com/documents.asp?grID=293&d_ID=1578 [July 2003]).
250
Brookings Trade Forum: 2005
Figure 1. India-to-United States Bandwidth Pricing, 1996–2000 Dollars per megabyte
50
40
30
20
10
1996
1997
1998
1999
2000
Source: Andy Grove, “The Coming Software X Curve,” presentation to the Business Software Alliance, October 9, 2003.
credit card payments. The Indian firm later graduated to purchasing the receivables from its client and assuming the collection risk itself. Another firm interviewed, Wipro Spectramind, managed the radiology services of Massachusetts General Hospital for its second and third shifts. American radiologists, who earn $315,000 a year on average, were replaced by Indian radiologists, who earn $20,000 a year on average. Offshoring such work began in 1993 when American Express, an American bank, started using its Indian operations to provide bookkeeping support to its other Asian operations. The business grew rapidly after 1999, when the telecommunications infrastructure in India improved and costs fell (Dossani 2002b) (see figure 1). With relatively few regulatory restrictions on ownership or types of work, a difference in work types related to ownership appears to be evolving. Multinationals that do their own work focus on the back-office, higher-stage, and more sensitive work (time-sensitive work such as payroll, or work involving confidential data). Outsourcers, whether owned by Indian nationals or foreigners, do mostly call center work.
Rafiq Dossani
251
Figure 2. Evolution of Financial Services at Agilent Gurgaon Business process reengineering; ERP support;b global projects; PO to payment process;c process redesigns; policy formulations; solutions and business integration
H e a d c o u n t
Invoice processing Basic accounting; data entry; basic troubleshooting
0
Mar 2002
Complex VP;a problem solving; payments
Supplier calls
Customer collections; accounts reconciliation; year- and month-end close; Six Sigma Program
Dec 2002 Mar 2003 Jun 2003
Dec 2003 TIME
Feb 2004
a. VP = Vendor payables. b. ERP = Enterprise resource planning. c. PO = Purchase order.
It is clear from Indian business process offshoring (BPO) that complex work can be efficiently done in India. A recent case study of a large multinational firm, Agilent, which in 2002 began offshoring about half of its head count in finance and accounting to Gurgaon, India, shows the evolution of complexity and these gains (see figure 2). The company began with the simplest work of data entry in March 2002. Up to December 2003, the work did not change in complexity as new staff were recruited to take over the work of Agilent’s offices around the world. However, by December 2003 Agilent felt that the Gurgaon office was sufficiently mature to enable a move to the next step. This was achieved rapidly and successfully; by February 2004 about half of Agilent’s global head count for the finance and accounting function were employed in Gurgaon. The complexity of work rose significantly to include customer collections and policy formulation support work, and finally to managing relations with suppliers. Several of these functions were unanticipated benefits when the offshoring relationship began. Also, as table 4 shows, the work was done with a gain in both costs and staff time; that is, the Indian operations were more efficient than the unconsolidated global
252
Brookings Trade Forum: 2005
Table 4. Productivity Gain in the Vendor Payables Division at Agilent Technologies Number of workers Head count Activity
January 2003
April 2004
Head count reduction (percent)
19 75 11 27 9 1 46 188
13 56 9 19 6 1 34 138
46 34 22 42 50 38 35 37
Order supplier processing Invoice processing Employee reimbursement Help desk/CRC Payment Reporting metrics Management and administration Total Source: Agilent documents.
operations were earlier by a factor of 37 percent. This was attributable to the benefits of consolidation—that is, economies of both scale and scope—as well as new technology platforms.11
The Offshoring of Software Services A more mature example than business process offshoring is the offshoring of software services (different from software products) to India, which began in 1974.12 The Indian software services exporting industry employs over 250,000 people (see table 5). How the global software services industry’s scale and components compare with India’s is shown in table 6. Different service processes require different skills and levels of interaction with others in the supply chain. Consulting and system integration require high
11. Dossani and Manwani (2005). 12. Software is either (1) written for general use and intended to be replicated in its original form for many users or (2) customized to a client’s needs. The former is termed a software product or package. The latter is termed custom software and is part of a larger category called software services (see table 5). Services are more constrained by geography than products since products can be shrink-wrapped and transported physically or over the Internet, whereas services are delivered to order. Hence, most countries export software products, India being an exception. There are three types of software: (1) system-level software: programs that manage the internal operations of the computer, such as operating systems software, driver software, virus scan software, and utilities; (2) tools software: programs that help applications to work better, such as data-
Rafiq Dossani
253
levels of interaction, as does IT education and training. However, and this was important for the development of India’s software industry, the work of applications development requires only programming skills and limited interaction. This makes it possible for one firm to design and specify an application and for another firm to develop it, with almost no interaction needed between the two. Thus, a user might contract with a U.S.-based consulting firm for system specification and then transfer the work to an Indian programming firm to develop the applications. This separability of programming from other processes has been key to the work being done in India since the early days. But the need for high levels of interaction with clients and their customers and vendors may make it difficult to do consulting and other complex work offshore even though the skills might be available. Initially, the work done in India largely kept pace with the U.S. software industry. However, with the United States now mostly doing the highly interactive and most sophisticated work, the gap has lengthened (see table 7). Unlike the BPO business in India, the software business began with an Indian firm, TCS, in 1974. However, this was due to protection: in 1973 the Indian government required multinationals operating in India to reduce their shareholding to 40 percent. Many multinationals, including IBM (India’s largest IT firm at the time) preferred to close shop rather than divest. This created an opportunity for Indian firms to provide software services. Multinationals were allowed to reenter beginning in 1985, although the playing field was not leveled for at least another decade. Texas Instruments was the first to enter, in 1985. Multinationals’ presence in India remained small until recently, although since 2001 their growth rate has been greater than that of Indian firms. Their 2004 market share has been estimated at 13 percent.13 As tables 2 and 6 show, Indian software services are at the early stages of work. The relatively low value of their work may be due to the state’s protectionist policies and the resulting absence of multinationals in the early years, though there is some indication that in recent years the presence of multinationals has contributed to a rise in Indian value added (see table 5). The entry of base management software; (3) applications software: programs that deliver solutions to the end user, such as word processing software and financial accounting software. Systems software is the most complex because it manages the interfaces with both hardware and higher-level software; applications software is the least complex. All system-level software are products (although this was not always the case). However, the more varied end users’ needs are, the more likely the software is to be customized. Since variations in needs appear most at the stage of applications, most customized software is applications software. 13. NASSCOM (2004)
Table 5. Software Services Sales of the Indian IT Industry, 1996–97 to 2003–04 Millions of U.S. dollars, unless otherwise indicated
Software Domestic Exports Total Export (percent) Exports employment (number) Export revenue/ employee
1996–97 a
1997–98
1998–99
1999–2000
2000 –01
2001–02
2002–03
2003–04
Average growth, 1996–97 to 2003–04 (percent)
759 1,100 1,859 59.2
1,177 1,759 2,936 59.9
1,411 2,600 4,011 64.8
1,575 3,399 4,974 68.3
2,081 5,287 7,368 71.8
2,311 6,152 8,463 72.7
2,769 7,045 9,814 71.8
3,374 8,600 11,974 71.8
23.8 34.2 30.5 ...
n.a.
n.a.
n.a.
110,000
162,000
170,000
205,000
260,000
24.0
n.a.
n.a.
n.a.
30,900
32,635
36,188
34,366
33,077
1.8
Source: NASSCOM (2004, pp. 23, 26, and 64). a. The Indian financial year is from April 1 to March 31. n.a. Not available.
Rafiq Dossani
255
Table 6. Global Software Services Spending and Indian Share by Category of Work, 2003
Work categorya Consultingb Applications developmentc System integration: hardware and software deployment and supportd System integration: applications, tools, and operating systems (O/S)e IT education and trainingf Managed servicesg Total
Global software services spending (billions of $) (1)
Percentage share (2)
U.S. wage rate ($/hour) (3)
Indian exports (billions of $) (4)
India’s global market share (percent) (5)
41.5
11.6
80 –120
0.11
<1
18.4
5.1
25
3.02
16.4
91.7
25.6
18–25
0.37
<1
62.4
17.5
40
0.10
<1
18.5 124.9 357.6
5.3 34.9 100.0
40 60 –120 …
0 1.94 5.54
0 1.6 …
Sources: NASSCOM (2004, p. 19) for columns 1 and 2; NASSCOM (2002, p. 24) and author interviews for column 3; NASSCOM (2004, pp. 36 and 106) for column 4. Indian figures are for the twelve months ending March 2003. Indian export figures exclude product development and design ($0.56 billion) and embedded software ($1.1 billion). a. Work categories are listed in order of implementation. Developing a customized application begins with an IT strategy, following which the system is conceptualized, with appropriate architecture and technology. This leads to the system’s design and specification, after which the operating systems, tools, and applications are developed (or purchased as products). It is unlikely that only customized software will be used in the final product. As noted above, the operating system and tools will usually be packages, while the applications programs will be a mix of customized and packaged programs. For example, word processing software will usually be a product. The exact mix depends on the system’s specification and design. Once the software components are ready, the systems integrator puts it all together into a working application, which may be web-enabled. Finally, the managing of the software may be outsourced. b. Consulting refers to work on IT strategy, system conceptualization, architecture, and design. The data reflect NASSCOM numbers for information systems (IS) consulting and network consulting and integration. c. Applications development refers to the creation of applications programs. The data reflect NASSCOM numbers for custom applications development. The category does not include spending on developing applications programs as products. However, the other numbers, such as for systems integration, include work done on making software products work. d. System integration: hardware and software deployment and support refers to the work of making the software and hardware components compatible and interoperable according to the required specifications. The data reflect NASSCOM numbers for (1) Hardware Deployment and Support and (2) Software Deployment and Support. e. System integration: applications, tools, and operating systems (O/S) refers to integration of the software components in a software project. f. IT education and training refers to training provided in software and hardware use. g. Managed services refers to services such as managing applications either onsite or remotely over the web, managing networks, etc. It is comprised of NASSCOM numbers for applications management, IS outsourcing, network and desktop outsourcing, applications service providers and system infrastructure service providers.
multinationals (and the diaspora) was induced by sweeping reforms in foreign ownership rules, telecommunications policy, and venture capital policy since 2000. These changes ought to lead to a rise in the domain skills available in India, with benefits for the large Indian software service firms, but it is too early to say. The contractual relationship between the Indian software industry and its clients
Table 7. A Comparison of Software Work Done in the United States and India in Common Timeframes Work typea By U.S. software services firms b Timeframe
More complex
Up to 1970
Support for in-house IT department (mainly conversion work)c Applications programs and electronic data processing (EDP) Systems integration (hardware with systems) and EDP
1971–80
1981–90
1991–2002
Consulting, systems integration (software)
Less complex
By India-based labor More complex
Less complex
O/S, software support for IT firmd Support for in-house IT department Applications and O/S programs; Unix conversion work
Applications programs, web services
Support for in-house IT department (mainly conversion work); EDP Offsite conversion work; offsite applications development; in-house product development by MNEs Offsite applications development in large projects; engineering services; EDP; in-house product development by MNEs
O/S, software support for global IT firm Onsite conversion work; onsite applications develdevelopment
Onsite conversion (including Y2K) work, website maintenance
a. The table’s cells in some cases present multiple kinds of work done. These are ordered by dollars spent. For example, in the period 1981–90, in the more complex category for India, offsite conversion work was the largest segment of the business, followed by offsite applications development and in-house product development by MNEs. b. Excludes work done by U.S. product developers since this was not significant to India; note that it shows up as the lowest ranked work by dollars spent from 1981 onward. c. “In-house IT” refers to work given by non-IT firms with in-house IT departments. Examples of such work are conversion of client applications software to new operating systems and platforms. d. O/S (operating system), software support for IT firm refers to work given by software companies to outsourcers.
Rafiq Dossani
257
is still largely one in which the client takes charge of IT strategy and system design, while the Indian firm provides programming services to specification. A key enabler was technology. Just as reduced data transmission costs enabled the business services industry’s offshoring, in the 1980s the introduction of the workstation, with its sophisticated graphics and numeric capabilities, and the standardization of Unix/C as the language of programming for all mainframes and workstations dramatically changed the economics of outsourced software development. An outsourcer could now own a workstation made by any manufacturer, yet write programs for a client whose installed hardware might be of a different brand.
Key Success Factors and Future Prospects What were the institutional and other factors that led to India’s success with offshoring services? Spending on education, the role of the diaspora, and knowledge of English due to its colonial past are often cited. Contrary to the popular view, however, spending by the government on education was not a key contributor to the success of offshoring. In fact, India’s education policy has been widely criticized as ineffective.14 Under the Indian constitution, education is a “concurrent” subject, subject to both state and central control. The central government is the main financier of tertiary educational services and has focused on expanding the university system to provide widespread access. But quality has suffered. According to a government report, “obsolescence of facilities and infrastructure are experienced in many institutions . . . the IT infrastructure and the use of IT in technical institutions is woefully inadequate . . . the barest minimum laboratory facilities are available in many of the institutions and very little research activity is undertaken . . . engineering institutes have not succeeded in developing strong linkages with industry . . . the curriculum offered is outdated and does not meet the needs of the labor market.”15 There were 247 universities and 11,549 colleges in India in 1999. Still, by 1997 only 7 percent of the eligible population attended university.16 As a result, India has 0.3 scientists and technicians per 1,000 population, ranking forty-two out of
14. India spent 3.2 percent of GDP on education in 1999; China, in comparison, spent 2.3 percent. Statistical Outline of India (2001–02, p. 243). 15. Ministry of Human Resource Development, Technical Education Quality Improvement Project of the Government of India, October 29, 2001, sections 2.1.2–2.1.6 16. NASSCOM (2004, p. 78).
258
Brookings Trade Forum: 2005
sixty-two countries ranked by the World Bank in 1998, below China (ranked twenty-fifth) and Ireland (ranked twentieth).17 Further, the interaction between university and industry is minimal. There are few research partnerships between academia and industry. For example, at IIT Delhi, the value of sponsored research and consultancy assignments in 1998 was only $4.5 million.18 This may be because faculty at the IITs are not expected to do research. According to India’s National Association of Software and Service Companies (NASSCOM), “Over the years, there has been a general decline in the quality of faculty in Indian universities.”19 The average number of citations over a five-year period for a faculty member at the Indian Institutes of Technology is fewer than three. This compares with forty-five for a faculty member at MIT and fifty-two for a faculty member at Stanford University.20 India produces only 300 master’s degree graduates and twenty-five Ph.D.’s in computer sciences each year, compared with U.S. numbers of 10,000 and 800, respectively.21 Data from NASSCOM show that only 27.12 percent have an undergraduate or graduate degree in computer sciences or electrical engineering.22 The role of the diaspora was also minimal. The global Indian diaspora, also called nonresident Indians (NRI), is estimated at about 20 million persons. Their investment in India as a group is minimal and is estimated at about $160 million annually, or 4 percent of the total annual foreign investment in India.23 The number who have been in a position to play a role in the development of India’s IT industry is considerably smaller. For example, there are probably no more than 30,000 engineers of Indian origin in Silicon Valley.24 Their role in startups has been different from their role in Silicon Valley. In a 2002 survey of 1,556 engineers born in the PRC, Taiwan, and India and living in Silicon Valley, only a small percentage said they had invested once or more 17. World Development Indicators (1999). 18. Parthasarathi and Joseph (2002, p. 32). 19. NASSCOM (2002, p. 73). 20. NASSCOM (2002, p. 73). 21. Ministry of Human Resource Development, section 2.1.12. This is despite excess capacity: only 50 percent of the available seats are filled. 22. Parthasarathi and Joseph (2002, p. 20, quoting NASSCOM data for 2000). 23. By contrast, overseas Chinese are believed to send much larger sums to China. Coomi Kapoor, “Hum apke hain Kaun?” Indian Express (www.indianexpress.com/full_story.php? content_id=16354 [January 2003]). 24. The figure of 30,000 is a guesstimate because there are no reliable numbers. This is a likely overestimate. Here are two calculations: (1) according to the U.S. Government Census of 2000, there were 66,741 persons of Indian origin residing in Santa Clara Valley, or 3.97 percent of the total population (by comparison, there were 115,781 Chinese, or 6.9 percent of the total population) (http://factfinder.census.gov/bf/_lang=en_vt_name=DEC_2000_SF1_U_DP1_
Rafiq Dossani
259
in startups in their home countries. Of those surveyed, 89 percent of Chinese, 84.1 percent of Taiwanese, and 77.7 percent of Indians said they had never done so.25 This might reflect better opportunities in Silicon Valley or the difficulty of building networks overseas. In fact, the Indian software and BPO industries were started by domestic Indian firms, and they continue to dominate the industry. U.S.-educated returnees led most of the successful firms, although they did not have significant work experience overseas (see table 8). A key advantage appears to be widespread education in the English language. India’s linguistic diversity has increased English’s role as the country’s lingua franca. Although the number of speakers of good English is not known, it is probably over 50 million, since almost all college graduates (a population estimated at 27 million in the 2001 census) and a significant proportion of high school graduates (28 million annually) would count English as a language in which they are fluent. Other institutional causes, less cited but perhaps as relevant, are India’s mature judicial system, conformance to WTO obligations, and a history of successful private enterprise that provided the talent for initiating and managing complex service projects. The situation in other developing countries differs from that in India, as might be expected. China, for example, has better infrastructure but lacks several other components of the enabling environment, such as a history of private entrepreneurship, knowledge of English, and a mature judicial system. Based on current evidence, too little is known about which factors are the most crucial to creating a successful environment for services offshoring. At the very least, India’s success
geo_id=05000US06085.html [January 2003]). There were 170,113 persons employed in the professional services and information industry in Santa Clara County, which encompasses Silicon Valley (http://censtats.census.gov/data/CA/05006085.pdf#page=3 [January 2003]). So it is unlikely that more than 30,000 Indians (45 percent of the Indian population) work in the software industry. (2) The estimate of 170,113 software workers may be too high because it includes all professional services, such as legal and financial services. On the other hand, many manufacturing operations contain a high percentage of workers who develop software. Assuming that this percentage is as high as 50 percent, we turn to 2000 data for Santa Clara County (www.census. gov/epcd/cbp/map/00data/06/085.txt [January 2003]). The employment recorded under NAICS (North American Industry Classification System) code 5415 (Computer Systems Design and Related Services, Including Custom Programming and Systems Design) was 46,836 workers; code 333295 (Semiconductor Machinery Manufacturing) had 7,383 workers; code 334 (Computer and Electronic Product Manufacturing) had 125,916 workers; and code 5417 (Scientific R&D Services) had 13,204. The adjusted total (assuming 50 percent of manufacturing employment was software related) is 126,690 workers. Since this is less than the base number of 170,113, the conclusion about the total remains. 25. Dossani (2002a).
Table 8. Founders’ Backgrounds at the Leading Indian Software Firms Rank by sales revenue Ranked by sales revenue
Firm (1990) a
1
TCS
2 3
Tata Infotech Citibank OSL
4
Datamatics
5
Texas Instruments
6
Digital Equipment India Ltd. (DEIL) PCS Mahindra-British Telecom (BT)
7 8
Founder’s name (education and previous company) Kanodia (MIT, Punjab U, Queen’s U. Canada); Kohli (MIT, GE Canada) TCS spinoff … Kanodia (MIT, Punjab U, Queen’s U. Canada, TCS) … Sonawala (U. of Washington) Patni (MIT) Mahindra (Harvard)
a. 1990 refers to the fiscal year ending March 31, 1991, and 2003, respectively. b. 2002 refers to the fiscal year ending March 31, 2003.
Firm (2002) b
Founder’s name (education and previous company)
TCS
Kanodia (MIT)
Infosys Wipro
Murthy (U. Mysore, IIT) Prejmi (Stanford); Soota (IISc, Bangalore) Raju (Loyola College, Chennai, Ohio U.) Nadar (PSG College, Coimbatore) Patni (MIT)
Satyam HCL PCS Mahindra-BT IFlex
Mahindra (Harvard) Hukku (BITS Pilani, TCS, Citicorp)
Rafiq Dossani
261
was hard to predict, and many forecasters were pessimistic (Schware 1992; D’Costa 2003).
The Choice of Organizational Form One way in which developed-country firms retain value is by controlling the work done, either by providing the risk capital or through subsidiaries. While it is difficult to predict which organizational types will dominate, a number of firm-specific factors that influence offshoreability and organization structure are summarized below. —The knowledge component of the activity. A higher knowledge component makes the firm more concerned about whether the quality of the service will change if the firm moves or the service is transferred offshore. —The interactive components of the process, as discussed earlier. —The level of separability of the process, or modularization, as discussed earlier. —Savings achieved by concentrating an activity in one location, which could involve benefits of scale and scope. —Reengineering as part of the transfer process. To transfer a business process, it is necessary to study it intensively and script the transfer. In the process of study, often there will be aspects of the current methodology for discharging the process that do not add value. Very often these aspects are legacies of earlier methodologies that were not eliminated as the production process evolved. During the act of transfer these are easier to abandon than at an existing facility where they have become an established part of the daily routine. Interviews conducted by my colleagues and me identified other unexpected benefits that can be reaped from the transfer process that go beyond the efficiency effects. This process of examining and the transferring a service can yield significant efficiencies. During the transfer process, these inefficiencies can be addressed without disrupting work patterns as the workers in the new location are met with a fait accompli. —The time-sensitive nature of the work. A popular classification of ownership and location is presented in GAO (2004), where ownership might be in-house or outsourced and location might be onshore or offshore (sometimes divided into nearshore and remote). Given the importance of domain skills resident in the multinational, my colleagues and I find that work types are linked to the following organizational forms: onshore inhouse; onshore outsource; offshore in-house; offshore outsource to subsidiary of
262
Brookings Trade Forum: 2005
onshore outsourcer; offshore outsource to indigenous firm. Although this list seems adequate for analytical purposes, our interviews with firms indicated that there are even more creative organizational forms. For example, some client firms had signed contracts with overseas vendors that allowed them the flexibility to move the outsourced operations in-house at a later date, perhaps once scale was achieved or comfort levels were high enough. Such BOT (build, operate, transfer) contracts sometimes included an option for the client to transfer key employees as well. The link between organizational form and the six risk types is shown in table 9. For many activities, offshoring is difficult and in many cases unlikely to happen. This was supported by our interviews with onshore call centers. For example, with the recent rapid development of online sales, the call center became the prime vehicle for business development and so could not be outsourced or easily offshored. Technological development is another reason for retaining work onshore. With the switch to VOIP (voice over Internet protocol) now rapidly under way, the modern call center can consist of a hosted service on the Internet while the employees may be located anywhere. This allows many of the advantages of scale to be gained without relocating a large enough number of employees to an offshore location. When offshoring is possible, the in-house multinational is a favored organizational form in most cases because it enables much of the rents from doing sophisticated work to be captured in the developed country even though the work may be done in India. The only exception appears to be when an independent firm can obtain economies of scale. It is therefore not surprising that, in practice, most of the offshored services done in India by domestic firms focus on work that bears economies of scale, such as the call center and programming projects, while work that does not have economies of scale, such as consulting, tends to be retained by the multinational firms. Of course, economies of scale can be achieved at relatively low capital costs (for example, the efficient scale for a call center is 1,000 persons, requiring a capital investment of only about $5 million), so even this does not really act as a barrier to entry and enable a domestic firm to charge rents.
Implications for Employment in Developed Countries Most of the economic debate about offshoring has been over the immediate job losses in developed countries and over the speed with which the economy
Rafiq Dossani
263
can develop new jobs to offset the losses.26 Offshoring may indeed require a fundamental shift in our understanding of employment patterns. The shift to a services economy in the second half of the twentieth century occurred in the context of a shift away from manufacturing that was relatively gradual given the speed with which services can move. The U.S. services economy was able to provide employment for workers at a wide range of skill levels to replace jobs lost in manufacturing. However, if low-end service jobs move overseas at high speed, American workers will have to not only retrain themselves more quickly but also acquire higher levels of skill than were required during the shift to manufacturing. The “lateral quality” of the shift that allowed a low-skilled assemblyline worker to shift to a low-skilled service job when manufacturing declined may not be an option today. The standard argument in favor of offshoring is illustrated in figure 3, in which it is assumed that all production costs are labor costs. Current onshore costs, ON1, can be reduced to OFF through offshoring. Assuming that retrenched workers will find work at some wage rate ON2, welfare gains from offshoring equal A+B+C+D, where A+B+C+D+F is the gain in consumer surplus and F is the lost wage when retrenched workers find alternative employment at ON2. Hence, offshoring is always welfare-positive (as long as A is nonnegative—that is, as long as the wage rate in the offshored country is always lower than the wage rate onshore). If some portion of the offshored income is spent on purchasing domestic goods and services, then the gain is further increased, but this is not a necessary condition for the transaction to be welfare-positive. This analysis may fit some services, such as, say, IT support in Silicon Valley, where the alternative wage for the displaced worker is likely to be significantly higher than the offshore wage. However, for some occupations, such as semirural call centers, the worker who is displaced may already be earning just the minimum wage and may have limited reemployment options. If the alternative for the displaced worker is to go on welfare at the reserve wage ON2, then the gains from offshoring are B+C+D–A–E, where (A+E) is the welfare cost, which may not be positive. The net welfare change could, therefore, be negative. This analysis provides fuel for the debate on the employment impacts of offshoring since the reserve wage, ON2, is usually higher in practice than the cost of fulfilling the service overseas. For example, in call center employment, the wage rate in India averages $200 per month, well below the welfare wage in the United States and other developed economies. 26. See, for example, Samuelson (2004) and Bhagwati, Panagariya, and Srinivasan (2004).
Table 9. Organization Form and Types of Work Performeda Type of work
Organization form Onshore–in-house Onshore-outsource Offshore–in-house Offshore-outsource MNE Offshore-outsource domestic
Use of proprietary knowledge within a process, such as research design
Process standardization, such as between consulting and coding
1 3 2 4 5
5 4 1 2 3
Source: Author’s survey, jointly with Martin Kenney, of forty BPO firms, 2003. a. The ranking is from 1 to 5, with 1 being the most important.
Interactive, such as with clients
Implementing economies of scale, such as managing peak load
Time-critical, such as payroll, software product development
1 3 2 4 5
5 3 4 1 2
2 1 3 4 5
Rafiq Dossani
265
Figure 3. The Welfare Impact of Offshoring
Wage rate
ON1
F
ON2
A
B C
D
OFF E Supply of labor
In some fields of work, the analysis above is incomplete. Consider software services, an area that comprises various linked layers of work: namely, consulting, development, and systems integration. The most standardized of these fields is development. It is based on technology that is two decades old (Unix and C) and has been subject to waves of automation and offshoring. With each wave, the costs of applications development have fallen. On the other hand, the consulting component is subject to rapid change due to the change in client requirements. For these reasons consulting is the least standardized of the activities in software. The various components are linked by process. Consulting is the first step, followed by development and then systems integration. They also have similar skill requirements. A computer science degree is a necessary and sufficient requirement for many components of consulting,27 such as architecture and system design, and is also considered a sufficient requirement for development, although many parts of the development process are possible with lesser qualifications in specific computer languages or software platforms. System integration (for the software component) also usually requires a degree in computer sciences. When all of the work was done onshore, this similarity of skill requirements created a close link between the wages of all three components. Initially, when the business of independent software development began in the 1970s, wage rates for programmers and consultants were the same. Over time, as users’ needs evolved at a faster pace than the underlying technologies needed to create them, 27. We use computer science as equivalent in qualification to the computer engineer and systems analyst.
266
Brookings Trade Forum: 2005
consultants, who dealt most closely with clients, began to receive higher wages than programmers. This trend was further concentrated in the mid-1980s, when the language of programming began to standardize around Unix/C. Nevertheless, the underlying qualifications were similar, so that it was possible for a programmer to be a systems integrator or consultant. As a result, the wage rates stayed close to each other. In 1990, for example, the wage for a programmer was 70 percent that of a consultant.28 Offshoring changed the wage relationships because only the development function could be offshored. Because the consultancy and systems integration functions required significant levels of face-to-face interaction with the customer and vendors, they could not be done overseas. The result of offshoring the programming function was a decline in programmer wages in relation to the other components. By 2003 the programmer wage had fallen to about a quarter of the wage earned by consultants (see table 6). The U.S. supply of labor also responded. The number of programmers in the United States did not change between the mid-1980s and 2003, holding steady at 500,000; and many programmers in the services industry have moved on to higher-end work or to programming software products; meanwhile, the supply of system engineers more than doubled, to 1.2 million, during this period. To summarize, interactivity has had a decisive influence on the type of employment in developed and developing countries. Highly skilled activities such as programming have been offshored because they do not require interactivity, while less-skilled work such as some parts of systems integration have stayed onshore and earn higher wages than more-skilled work that has been offshored. Thus offshoring may create a dichotomy between skills and wage rates.
Conclusion India is the only developing country with a large and successful hightechnology services exporting sector, a remarkable achievement considering the importance of a firm’s reputation and the fact that exporting services is intrinsically more difficult than exporting manufactured goods. Offshored jobs span the skill spectrum and include jobs that require high levels of training and skills and are highly valued in developed economies. However, the bulk of the work done by the Indian IT industry is low-value-added and relatively low-skilled, likely a result of protection in the early days. Other factors, such as the diaspora and 28. Author’s interviews.
Rafiq Dossani
267
spending on education, have been relatively unimportant, although Englishlanguage skills (though limited to a small proportion of the population) were important. Technology, interactivity requirements, the need to protect proprietary knowledge, time and mission criticality, and economies of scale were significant factors in determining both the speed of offshoring and the ownership forms through which offshoring would be done. In some cases, the speed of offshoring can be very rapid, far in excess of what has been possible in manufacturing, thus threatening employment patterns in developed countries, especially for low-end work. This threat was bolstered by the welfare system in many developed countries. However, by and large, these factors favored the retention of higher-stage work in the developed countries, despite some anomalies, and favored ownership by multinationals rather than domestic firms, thus threatening the development of entrepreneurship in India and allowing the transfer of rents back to the developed countries. As a consequence, value addition in offshored services has been low. For India, the initial impact of offshoring was to generate high and growing levels of employment. However, the kinds of work offshored had low barriers to entry and were subject to automation. Again, this was because of the protectionist conditions under which the industry developed. These factors have led to competition and price deflation, mimicking the situation in manufacturing exports in developing countries and raising the likelihood that the asymmetries of globalization would be repeated in services exports from developing countries. Nevertheless, some Indian firms were able to extract economies of scale and developed globally known brand names. This allowed them to expand their business to related fields such as business process outsourcing. Despite this, their ability to expand into value-added fields within the value chain is questionable because they lack the domain skills, again a consequence of protectionist policies. Recent reforms that have induced the entry of transnational corporations and a transnationally trained diaspora could change the environment by introducing those domain skills. The threat to developed countries may increase if India’s success with IT exports and other business processes is replicated by other developing countries. India’s success can be attributed to a long history of private sector entrepreneurship that developed along with legal institutions, stock markets, educational institutions, and others. Other developing countries (including China) will have difficulty replicating India’s success unless they too can foster entrepreneurship and similarly strong institutions.
Comment and Discussion
Arvind Panagariya: Rafiq Dossani has provided an interesting account of the impact of the Indian offshoring on the rest of the world, especially the United States. T. N. Srinivasan provides the account of how offshoring is affecting the Indian economy itself in a separate companion paper. The authors address some common subjects such as the contribution of the Indian diaspora to the Indian information technology (IT) industry, and on these they have largely contrasting views. I find myself agreeing more often with the view taken by Srinivasan, which should not be surprising since I too come from the perspective of an international trade economist. My remarks are organized around five topics that appear in the paper by Dossani: definition and classification of trade in services; the pattern of the services outsourced to India; the likelihood of the hollowing out of the middle class in the developed countries; the role of the Indian diaspora in the promotion of offshore services; and miscellaneous issues.
Trade in Services: Definition and Classification On the issue of defining trade in services, the paper misses some key contributions by trade economists Bhagwati (1984) and Sampson and Snape (1985) that eventually formed the basis of the WTO terminology, now widely used by economists and trade negotiators. The WTO General Agreement on Trade in Services (GATS) divides services into four categories according to the “mode” of delivery: 268
Rafiq Dossani
269
—mode 1: services supplied at arm’s length with the buyer and seller staying in their respective locations (offshoring: call centers, back-office services, radiology services, distance learning) —mode 2: services supplied by bringing the buyer to the location of the seller (tourism, education, health care) —mode 3: commercial presence of the seller at the buyer location (multinational banks, insurance companies, telecommunications) —mode 4: individual seller moving to the location of the buyer (H1B visa holders) Conceptually, modes 3 and 4 are similar since both require the seller’s presence in the buyer’s territory; the distinction is a legal one for the purposes of defining legal market-access commitments. Modes 1 and 2 are clearly distinct from each other and from modes 3 and 4. One reason why the distinction between modes 1 and 2 is important is that from the jobs viewpoint mode 2 services such as medical care described in detail by Srinivasan are less controversial than the mode 1 trade in services popularly called offshoring. Most services require some interface between the recipient and provider since they must be consumed as they are produced. But this is not always true. Some services can be stored and effectively turned into goods, making their cross-border trade easier. While digitization and the Internet have brought the costs of storage and shipping of such services down dramatically, the latter have been around for a very long time. In ancient India, pupils committed the Vedas to memory and took them wherever they went. Later the Vedas began to be written down in books and got shipped as far as China. In our times, lectures can be videotaped or digitized and either shipped across borders on DVDs or transmitted over the Internet.
The Pattern of Services Outsourced from India Just as we argued in Bhagwati, Panagariya, and Srinivasan (2004), the pattern of outsourcing to India exhibits some features of the celebrated product cycle of trade in goods pioneered by Raymond Vernon (1966). Recall that Vernon pointed out that products would be innovated in the country where they are potentially demanded and will be sold and debugged there first. Once the debugging process is complete and the demand for them abroad is identified, they will be exported. Eventually, the product and production process will be standardized and production will move to the location where costs are the lowest, with the original innovator country becoming an importer of the product.
270
Brookings Trade Forum: 2005
The information technology revolution has opened the door to trade at arm’s length in services that have been in existence for many years. From that perspective, one would be tempted to hypothesize that the necessary standardization of many services already existed and that they would move to the cheapest production site en masse. Yet that is not what we are observing. Only some specific components of services appear to be subject to standardization and capable of being moved to the cheapest production location. Other components continue to require closer buyer-seller interface and thus have remained nontraded. This comes out strongly in table 6 of Dossani. The table shows that only in applications development, which is a standardized component of many services, does India have a significant presence. Few complex tasks such as systems integration, IT education and training, consulting, and managed services have been outsourced to India. In effect, the nontradability aspect of those components seems to have been preserved, suggesting that the bulk of the components of the services may never leave the territory of their buyers.
The Hollowing out of the Middle Class The biggest fear in the United States is that outsourcing to India may lead to a hollowing out of the middle class. The specific form the fear takes is that the workers in U.S. manufacturing who lost jobs to Chinese imports were previously able to move up to low-level services jobs. But that option may no longer be available since the low-level service jobs themselves are moving to India. If one looks at the available evidence carefully, however, so far we do not have a reason to be pessimistic. There are at least three reasons why we may continue to be cautiously optimistic in this regard. First, many of the low-end service jobs simply cannot be outsourced. Jobs in health care, education, retailing, tourism, and restaurants and catering require the presence of the provider at the site of the service recipient and therefore cannot be outsourced. Even many computer tech support jobs require such a presence. Many among the elderly and not-so-savvy users of technology even among the young would find support at arm’s length as difficult to follow as I find the directions to assemble furniture bought at Staples or Office Depot. Second, only a tiny component of services is subject to outsourcing. The only significant presence India has in software is in applications development. And here, as demand for onshore computer programmers has begun to stagnate, supply has adjusted, with the number of programmers in the United States remaining steady at 500,000 since the mid-1980s, according to Dossani. At the same
Rafiq Dossani
271
time, the demand for systems engineers has continued to grow, and so also their supply. With regard to call center jobs in India, if they were to be moved back to the United States, machines would do a significant portion of them. The price at which customers are willing to use the services currently provided by Indian call centers is simply not high enough to induce U.S. workers to take up those jobs. New call centers, though, continue to be established in the United States for more complex services that also fetch higher returns. Again, according to Dossani, online sales have given rise to the use of call centers for business development. Dossani notes that as much as 3 percent of the U.S. labor force is currently employed in call centers, and this figure is expected to rise to 5 percent by 2010. Thus, even call centers are not slated to disappear from U.S. territory. Finally, the capacity of India to create the skills necessary to sustain high growth of offshoring is itself far more limited than commonly claimed. The shortage of skilled personnel in India is already being reflected in job turnover rates as high as 40 percent,1 and by far the highest wage increases in information technology–enabled services (ITES) of anywhere in the Asia-Pacific region.2 Only 6 percent of the college-age population (aged 17–23) is actually in college currently, compared with 27 percent in the Philippines and 19 percent in Thailand.3 And with the exception of some well-known centers of excellence and some engineering, medical, and management colleges, the higher education system in India is a shambles. According to the 2001 census, India had only 37.6 million workers with undergraduate or higher degrees. Of these, only 25.4 million were in urban areas, and of the latter, a tiny 3.6 million had technical degrees. Taking into account the fact that India needs workers with technical higher education for its domestic needs and that the medium of instruction in most Indian universities is not English, the potential workforce available for employment in outsourcing is small and cannot be expected to hollow out the middle class in the United States.
The Role of the Indian Diaspora Dossani takes the view that the Indian diaspora did not make a significant contribution to the growth of the software and ITES industry. For evidence, he 1. “The Bangalore Paradox: Outsourcing and IT in India,” Economist, April 23, 2005, pp. 67–70. 2. Stephanie Overby, “India Sees IT Wages Rise,” ITnetcentral, February 4, 2004 (www. itnetcentral.com/article.asp?id=13034&leveli=0&info=home). 3. Indian National Commission for Cooperation with UNESCO (1998).
272
Brookings Trade Forum: 2005
points to just $160 million annual foreign direct investment by overseas Indian nationals. While it is true that the Indian diaspora has played a far less significant role in bringing investment to India than the Chinese diaspora in Hong Kong and Taiwan, I will argue that in the software and ITES sector it has contributed more than Dossani gives it credit for. For one thing, we cannot adequately assess the contribution of overseas Indians to the software and ITES sectors by the magnitude of their direct foreign investment. These sectors are not particularly capital intensive, so small direct foreign investments by the diaspora noted by Dossani can still be associated with large contributions. More important, much of the contribution of the Indian diaspora in these sectors has taken intangible forms. Several sources of contribution by the diaspora can be identified. First, though neither Dossani nor Srinivasan recognizes this, Indian entrepreneurs have made significant contributions to the advancement of the global IT industry. In May 2000, Fortune magazine noted that without Indian entrepreneurs Silicon Valley would not be what it was at the time. It placed the wealth generated by them at $250 billion, more than half of India’s GDP.4 Insofar as the advancement of the global IT industry was crucial to the opening of the outsourcing, these entrepreneurs evidently made an important contribution to the growth of the phenomenon. Second, during the 1980s, the Indian diaspora also helped link the Indian labor market in computer programmers to the U.S. market. Taking advantage of temporary worker visas, many Indian entrepreneurs exploited their familiarity with both the U.S. and Indian markets and brought Indian programmers to the United States, thus promoting mode 4 trade in services. That process brought U.S. employers and Indian programmers face to face with each other and contributed to the confidence building that later proved useful when technological advancement opened the door to outsourcing in a big way. In the same vein, as Srinivasan notes, in the 1990s and beyond, many Indians working for U.S. companies have driven the sourcing of services from India. Even though Indians do not own the companies in question, they have been instrumental in driving the process. Finally, Indian economists abroad were the first to begin building the case for opening up the Indian economy. While the immediate impetus for the opening may be traced to such factors as the collapse of the Soviet Union and the rapid advance of China, the early advocacy of outward-oriented and pro-market policies in contributions such as those by Bhagwati and Desai (1970) and Bhagwati and Srinivasan (1975) had influenced the thinking of many in India. In the same 4. See Panagariya (2001).
Rafiq Dossani
273
vein, some of the key policymakers involved in the reform process in the 1990s were ex-diaspora. Even the confidence exhibited by Dr. Manmohan Singh in outward-oriented policies can be traced back to his doctoral thesis written at Oxford University in 1962, which formed the basis for his brilliant 1964 book.5
Miscellaneous Quibbles Dossani’s claim that infant industry via the forced departure of IBM in the mid-1970s “allowed the Indian IT industry to reach a high-growth path and allowed local skills to develop rapidly to keep pace with global changes” remains largely unsubstantiated in the paper. It is one thing to claim that the departure of IBM created some rents for local firms with computer programming capacity and induced some innovation, but quite another that it placed the industry on a high-growth path.6 There is little evidence that, after IBM exited, sales by Indian software firms took off in a major way either at home or abroad. Quite the contrary, it was not until Rajiv Gandhi took steps to liberalize hardware and software imports in the mid-1980s that the software industry began to show promise. Srinivasan (this volume) provides a list of the specific measures taken by Rajiv Gandhi during 1984–86 and concludes, “There cannot be a more dramatic departure than this policy from the strategy of technological self-reliance, import substitution across the board (from intermediates to capital goods) and export pessimism.” Let me conclude with two final, truly minor quibbles. In describing the Samuelson (2004) paper, Dossani states that it deals with the jobs issue. As I discuss in Panagariya (2004), this is inaccurate. Samuelson deals with the impact of productivity improvement abroad in the production of goods exported by a country in a model with full employment. Likewise, my paper with Jagdish Bhagwati and T. N. Srinivasan covers virtually all aspects of the outsourcing debate rather than focusing on just the jobs issue.7
5. See Singh (1964). 6. Similar innovations had also taken place when, in the early 1990s, the United States banned India from buying supercomputers. Having acquired access to enough hardware and software by then, Indian computer engineers managed to create their own supercomputer, Param, at a fraction of the cost India would have paid for U.S. supercomputers. It then proceeded to export those computers to countries such as Canada. 7. See Bhagwati, Panagariya, and Srinivasan (2004).
274
Brookings Trade Forum: 2005
General Discussion: The presentation by Rafiq Dossani stimulated a lively discussion. Ravi Aron started it off by commenting on the interesting contrast in India between the rapid acceleration of BPO in IT-enabled financial services and the much slower progress in software services. He suggested this could be explained in part by differences in operational risk in the two industries. He cited firm-level research he has undertaken that highlights the extent to which process can be qualified as a key determinant of operational risk. Since financial processes are on a continuum with respect to the variable—there are processes at both the high and low ends of qualifiability—Indian firms can pluck the low-hanging fruit first and continuously mature thereafter. In contrast, for software services, as the volume of new services has grown, the work performed has not migrated substantially up the supply chain to more complex activities of greater qualifiablility. Second, Aron highlighted the importance of differences between domain expertise and domain experience. If a position can be accurately qualified, it is possible to substitute expertise for experience. For processes that are not qualifiable, experience is critical. For high-end services that require domain experience, India is hindered in its efforts to develop global capability when it has not accumulated such domain experience in the domestic market. For instance, an Indian BPO services provider will not be able to develop a finely calibrated price for a financial services product that has not already been developed for the domestic market. On the issue of supply-side constraints,Aron argued that the number of middlelevel managers in India is a more binding supply-side constraint than the number of new graduates. His research showed that 65 percent of managers responsible for high-level contracts among seven large BPO companies in India had been poached from the existing multinational base, and that such poachable resources will begin to run out very quickly. T. N. Srinivasan argued that Dossani’s paper overemphasized the concentration of Indian software services exports in lower-value-added categories. Srinivasan’s reading of the data suggests that only about 60 percent of software exports—rather than the 90 percent shown in Dossani’s database—are in customized software and that a growing share are in higher-value second- and thirdstate products. In contrast, Srinivasan lent support to Rafiq Dossani’s pessimism about moving up the value chain in financial services given continued government ownership of roughly 70 percent of the commercial banking system. Claire Brown commended the paper for laying out clearly the different segments within the software services industry and explaining why some segments such as coding have been subject to offshoring while others have not. She interpreted Dossani’s paper as “good news” in its implication that some high-end ser-
Rafiq Dossani
275
vices jobs in the United States are very unlikely ever to be vulnerable to offshoring, since American companies still own the brand and marketing, which cannot be offshored and which commands the market power. In these segments, Indian companies are effectively producing intermediate services, enabling American companies to benefit from lower costs, expertise, and innovation in the intermediate segment. Catherine Mann noted the striking divergence of views on whether the Indian development strategy emphasizing business process outsourcing and IT-enabled services was a threat to the American economy. She asked whether the lack of domestic demand for complex, high-end services would prevent India from attaining a global comparative advantage in these services. The alternative view is that India could master high-end services despite the absence of a domestic market simply by establishing development centers or by operating as an export processing zone for such services. Dossani responded that the development centers are actually quite similar to the old body-shopping paradigm, performing much lower-value-added work than applications development. Nonetheless, this work provides good diversification of work for the software industry. On the issue of supply constraints on middle managers, Dossani added that the problem is in some senses worse than the numbers suggest because many graduates of management schools in India prefer to go into client-facing work rather than into the BPO and IT industries. On the other hand, the cost advantage from offshoring is so high that the supply of qualified managers will not become a significant factor for at least another five years. Finally, on the issue of whether India can develop a competitive advantage providing services to an industry where India has weak domestic demand, Dossani noted that it is possible to acquire relevant learning or domain experience through other channels. One important channel that companies such as Intel have used is to move people (most often of Indian origin) who have acquired such experience in developed countries to India. Strategic partnerships between local Indian companies and multinationals also provide a good channel for such learning.
276
Brookings Trade Forum: 2005
References Bhagwati, Jagdish. 1984. “Splintering and Disembodiment of Services and Developing Nations.” World Economy 7: 133–44. Bhagwati, Jagdish, and Padma Desai. 1970. India: Planning for Industrialization. Oxford University Press. Bhagwati, Jagdish, and T. N. Srinivasan. 1975. India. Vol. 6. New York: National Bureau of Economic Research / Columbia University Press. Bhagwati, Jagdish, Arvind Panagariya, and T. N. Srinivasan. 2004. “The Muddles about Outsourcing.” Journal of Economic Perspectives 18 (Fall): 93–114. D’Costa, Anthony P. 2003. “Uneven and Combined Development: Understanding India’s Software Exports.” World Development 31 (1): 211–26. Dossani, Rafiq. 2002a. “Chinese and Indian Engineers and Their Networks in Silicon Valley.” Working Paper. Stanford University, Asia-Pacific Research Center (March). ———. 2002b. Telecommunications Reform in India. Westport, Conn.: Greenwood Press. ———. 2004. “The Software Industry in India.” Working Paper. Stanford University. Dossani, Rafiq, and Martin Kenney. 2004. “Lift and Shift: Moving the Back-office to India.” Information Technology and International Development 1 (2): 21–37. Dossani, Rafiq, and Anita Manwani. 2005. “Agilent’s Supply Chain: A Locational Analysis of Its Indian Operations.” Paper presented at the Conference on Globalization of Services, Stanford University, June. Gadfrey, J., and F. Gallouj. 1998. “The Provider-Customer Interface in Business and Professional Services. Service Industries Journal 18 (2): 1–15. Gereffi, Gary, John Humphrey, Raphael Kaplinsky, and Timothy J. Sturgeon. 2001. “Introduction: Globalisation, Value Chains and Development.” IDS Bulletin 32 (3) July: 1–8. Goodman, Bill, and Reid Steadman. 2002. “Services: Business Demand Rivals Consumer Demand in Driving Job Growth.” Monthly Labor Review 125 (4): 3–16. Hobday, Michael. 1995. Innovation in East Asia: The Challenge to Japan. Cheltenham: Edward Elgar. Humphrey, John. 2004. “Upgrading in Global Value Chains.” Working Paper 28. Geneva: World Commission on the Social Dimension of Globalization, International Labor Organization. Indian Economic Survey. 2004. New Delhi: Ministry of Finance. Indian National Commission for Cooperation with UNESCO. 1998. “Higher Education in India: Vision and Action Country Paper.” Paper presented at the UNESCO World Conference on Higher Education in the 21st Century, Paris, October 5–9. Available at www.education.nic.in/htmlweb/unhighedu.htm. NASSCOM 2002. The IT Industry in India. New Delhi. ———. 2004. The IT Industry in India. New Delhi. Panagariya, Arvind. 2001. “The Indian Diaspora in the United States.” Economic Times. May 23.
Rafiq Dossani
277
———. 2004. “Why the Recent Samuelson Article Is NOT about Offshore Outsourcing.” Columbia University. Available at www.columbia.edu/~ap2231/. Parthasarathi, Ashok, and K. J. Joseph. 2002. “Limits to Innovation with Strong Export Orientation: The Case of the Indian Information Communication Technology Sector.” Science, Technology and Society 7 (1): 13–50. Samuelson, Paul A. 2004. “Where Ricardo and Mill Rebut and Confirm Arguments of Mainstream Economists Supporting Globalization.” Journal of Economic Perspectives 18 (3): 135–46. Sampson, Gary P., and Richard Snape. 1985. “Identifying the Issues in Trade in Services.” World Economy 8: 171–81. Scholte, Jan A. 2000. Globalization: A Critical Introduction. New York: PalgraveMacmillan. Schware, Robert. 1992. “Software Industry Entry Strategies for Developing Countries: A “Walking on Two Legs” Proposition.” World Development 20 (2): 143–64. Singh, Manmohan. 1964. India’s Export Trends. Oxford: Clarendon Press. Statistical Outline of India. 2001–2002. Mumbai: Tata Services. Thompson, Paul, Chris Warhurst, and George Callaghan. 2001. “Ignorant Theory and Knowledgeable Workers: Interrogating the Connections between Knowledge, Skills and Services.” Journal of Management Studies 38 (7): 923–42. U.S. Government Accountability Office (GAO). 2004. “Current Data Provide Limited Insight into Offshoring of Services.” September. Vernon, Raymond. 1966. “International Investment and International Trade in Product Cycle.” Quarterly Journal of Economics 80 (2): 190–207. Whitman, J. 1990. “Key Factors for Software Success.” Mimeo (Cambridge, Mass.: Oakland Group, December 26). World Development Indicators. 1999. Washington: World Bank. Wortzel, L. H., and H. V. Wortzel. 1981. “Export Marketing Strategies for NIC and LDC–Based Firms.” Columbia Journal of World Business 16 (1): 51–60. Yotopoulos, Pan A. 1996. Exchange Rate Parity for Trade and Development: Theory, Tests and Case Studies. Cambridge University Press.
This page intentionally left blank
CLAIR BROWN GREG LINDEN University of California–Berkeley
Offshoring in the Semiconductor Industry: A Historical Perspective
S
emiconductor design is a frequently cited example of the new wave of offshoring of service sector jobs.1 It is certainly a concern to U.S. design engineers themselves.2 The semiconductor industry already has rich experience with the offshoring of manufacturing activity. Semiconductor (or chip) companies were among the first to invest in offshore facilities to manufacture goods for import back to the United States. A review of these earlier manufacturing experiences and their impact on the fortunes of the domestic industry and its workers can help to illuminate the current debate over the offshoring of services. Because meaningful data about the impact of the offshoring of chip design (and even of manufacturing) are limited, we rely on a more qualitative analysis The authors would like to thank the Alfred P. Sloan Foundation; the Institute for Technology, Enterprise and Competitiveness (ITEC/COE); and the Omron Fellowship at Doshisha University for funding; and the Institute of Industrial Relations, U.C. Berkeley, for administrative support. We are grateful to Koji Ban, Ben Campbell, Michael Flynn, Ron Hira, Dave Hodges, Rob Leachman, Elena Obukhova, Devadas Pillai, Amy Shuen, Gary Smith, Strategic Marketing Associates (www.scfab.com), and Bill Van Der Vort for their valuable contributions to this paper. We would also like to thank Jeff Macher for his detailed and helpful comments, and Melissa Appleyard, Hank Chesbrough, Jason Dedrick, Rafiq Dossani, Richard Freeman, Deepak Gupta, Bradford Jensen, Ken Kraemer, Frank Levy, Tim Sturgeon, Eiichi Yamaguchi, and participants at the 2005 Brookings Trade Forum on Offshoring of White-Collar Work and the Doshisha ITEC seminar series for thoughtful discussions, which improved the paper. The authors are responsible for any errors. 1. See for example, Pete Engardio, Aaron Bernstein, and Majeet Kripalini, “The New Global Job Shift,” Business Week, February 3, 2003, p. 50; and Don Clark, “Another Lure of Outsourcing: Job Expertise,” Wall Street Journal, April 12, 2004, p. B1. 2. Robert Bellinger, “It’s an Outsourced World, EEs Acknowledge,” EETimes, August 30, 2004, p. 64.
279
280
Brookings Trade Forum: 2005
for our key points. We have conducted dozens of interviews with engineers and managers at semiconductor and related companies in the United States, Asia, and Europe over the past twelve years. Our research also incorporates the rich store of publicly available information in trade journals and company reports. Before addressing semiconductor design directly, we begin by analyzing the impact on the U.S. semiconductor industry of the offshoring of semiconductor assembly and fabrication. We argue that the initial concern about losing domestic jobs in both stages turned out to be unfounded as the industry used the situation to its competitive advantage by becoming cost-competitive (through assembly offshoring) and by developing the fabless sector (through foreign outsourcing of chip fabrication or manufacturing). We then analyze the ongoing offshoring of design jobs and compare this stage to the two that came before in order to explore the possible impact on domestic jobs and the U.S. semiconductor industry. The paper begins with a brief description of the stages of semiconductor production and our analytical framework. The following sections look at the offshoring of assembly jobs, analyze the offshoring of manufacturing, and explore the offshoring of design jobs. We conclude with a discussion of what this means for the United States.
The Industry and an Analytical Framework We begin by describing the stages of production in the semiconductor industry, which will help illuminate the industry’s offshoring activities. The most important type of semiconductor, and the one on which this study focuses, is the integrated circuit, or “chip,” which is a network of tiny wires fabricated on a surface connecting transistors that switch on and off for processing data in binary code.3 The development and manufacturing of chips involves three primary activities in the value chain: design, fabrication, and test and assembly. The semiconductor industry has successively offshored each of these activities—first assembly, then fabrication, and now design. During design, the desired electronic circuits progress through a series of abstract representations of increasing detail. During fabrication, the circuits of 3. Other types of semiconductors, such as single transistors and diodes, use different design and manufacturing methods not subject to the same economic forces discussed in this paper. These other categories constituted about 15 percent of the total semiconductor market in 2004. See World Semiconductor Trade Statistics, “WSTS Semiconductor Market Forecast, Autumn 2004,” press release, November 2, 2004 (www.wsts.org).
Clair Brown and Greg Linden
281
the chips are built up on the surface of a flat, round silicon wafer in successive layers. Assembly is, typically, the process of cutting the wafer into individual chips (or die), which can number in the thousands depending on die size, and packaging the delicate chip in a protective shell that includes connections to other components. The economic characteristics of each step of the process differ significantly. Design is skill-intensive and requires expensive EDA (electronic design automation) software. Fabrication requires a huge fixed investment (currently on the order of $2 billion) to build a plant (called a fab) that holds a wide variety of expensive equipment and that meets extreme requirements of cleanliness. Assembly also requires expensive equipment, but the overall costs of plant and equipment for assembly are much lower than for the fab, as are the average skill requirements. Worker skill requirements go down along the value chain (that is, design is more skill-intensive than manufacturing, which is more skill-intensive than assembly). However, equipment costs dominate labor costs, especially for fabrication, and this has limited the attractiveness of low-cost labor locations. Even the most labor-intensive activity, chip assembly, has become more automated over time. Other costs, including those relating to land, taxes, and government regulations, often affect decisions to invest offshore. The framework for this analysis of the offshoring of the stages of the industry value chain relies on the concept of competitive advantage, which is linked to offshore investing and outsourcing, and ultimately to domestic jobs.4 A sustained advantage over rivals can be built on product (the intellectual property that defines functionality), price (the cost of production), or market attributes (new customers, customer service, brand reputation, and links to legacy products). These sources of competitive advantage correspond to the three principal reasons that firms globalize their activities: access to location-specific resources, including engineering talent; cost reduction; and market development. When a firm with some nonimitable advantage moves an activity offshore to reduce its costs or improve access to resources or markets, it improves its competitive position against its rivals (barring cases where the move is mismanaged). In an expanding market like that for chips, the firm will grow and will hire more workers, some of whom will be in the home country and some offshore. However, some or all of the workers in the home country who were engaged in the activity that shifted offshore may lose their jobs, so that only the remaining home country workers benefit from the firm’s move offshore, along with the 4. Porter (1985).
282
Brookings Trade Forum: 2005
consumers of the lower-priced products.5 In addition, exports and imports are increasing with market growth, which clouds the impact of offshoring on jobs.6 In the longer term, numerous firm-level investments in a foreign location may transform the location in such a way that it presents a new set of opportunities that lead to a transformation of the industry. A foreign location that is initially little more than a source of lower costs, especially labor, might develop over time as a specialized supply base, particularly in the presence of incentives and infrastructure provided by the host-country government. The changes can increase the value of the location to the point that the industry eventually restructures around the new distribution of skills; offshoring then becomes the preferred mode for this part of the value chain. Later in the paper we discuss how this occurred for semiconductor assembly, but it has also taken place in other industries, such as hard disk drives.7 In other words, the pursuit of offshoring to gain competitive advantage in the context of a growing market initiated a dynamic process that has made the net employment effect difficult to estimate even after the fact. Table 1 summarizes our analysis of the three segments of the semiconductor value chain that is presented in the following sections of the paper.
Assembly: From In-house Offshoring to Offshored Outsourcing Assembly was the easiest stage of production to move offshore. It was functionally separate from the other stages of production even when performed in close proximity to fabrication. Furthermore, assembly began with a relatively high use of less-skilled direct labor, making it an attractive target for costreduction offshoring. During the 1980s the U.S. offshore chip assembly subsidiaries switched to automation in response to a combination of increasingly intricate packaging requirements beyond the skills of manual labor and rising wages in some Southeast Asian nations. A typical chip assembly plant employs 1,000 or more workers. By the mid-1990s, low-skilled workers made up about 80 percent of the staff of offshore assembly plants. The share of engineering and professional jobs was about 6 percent, and technicians made up another 13 percent.8 5. See, for example, Garner (2004) for further discussion. 6. Groshen, Hobijn, and McConnell (2005). 7. McKendrick, Doner, and Haggard (2000). 8. The Advanced Micro Devices chip assembly subsidiary in Penang, Malaysia, employed 160 engineers and 380 technicians out of 2,900 workers (“Firm to Make New Microchips,” Star,
Clair Brown and Greg Linden
283
The move to offshore assembly led to a “hollowing out” of the U.S. chip assembly sector but kept the U.S. chip industry cost-competitive as new rivals appeared in Europe and Japan. Over time, Asian suppliers appeared and took over a growing portion of the assembly business; as a result, the industry has moved from in-house offshoring toward offshored outsourcing. Most large chip companies still own assembly plants in Asia. The main lesson from this period of offshoring is that giving up domestic production of one part of the value chain offers one option to “save” the domestic industry.9 The second lesson is that the initial moves offshore can have unforeseen dynamic consequences such as the emergence of foreign suppliers who dominate the industry segment. Offshoring, Job Loss, and Competition Because of their high value-to-weight ratio, semiconductors could profitably be fabricated in the United States, air-freighted to Asia for assembly, and then returned to the United States for final testing and shipment to the customer. This system allowed the U.S. companies to take advantage of the specialized skilled and semiskilled labor in the United States for design, fabrication, and key managerial functions while tapping the lower-cost unskilled labor, land, and taxes of Asia for assembly. Today the transoceanic division of labor between fabrication and assembly still exists. Final testing was added to Asian assembly plants in the 1980s, which allows the finished chips to be shipped directly to customers from Asia, where a large share of the market is also located. From 1984 to 2004 the share of semiconductor sales in Asia, including Japan, rose from 38 to 63 percent of the world total.10 The earliest offshore investment in semiconductor assembly was made in 1961 by Fairchild Semiconductor in Hong Kong for the assembly of discrete transistors. Over the next fifteen years, this pioneering investment was followed by assembly investments by other companies in seven other economies of the
May 25, 1996). This ratio is similar to that for the whole semiconductor-dominated MalaysianAmerican Electronic Industry group in 1994 as reported in their Annual Survey of seventeen members over the preceding five years (MAEI 1995). 9. A complete analysis of the connection between the offshoring of assembly and industry competitiveness is beyond the scope of this paper. The move provided short-term breathing space for U.S. firms facing low-cost competition. This is not to say that other strategies might not have provided equal or better results. 10. Calculated from Semiconductor Industry Association market statistics available at www. sia-online.org/pre_statistics.cfm.
Table 1. Offshoring and the Semiconductor Value Chain by Industry Segment Value chain segment Share of engineers in workforce (percent)
Assembly
Fabrication
Design
6
24
85
Key economic characteristics
Moderately capital-intensive; requires access to low-cost direct labor
Highly capital-intensive; requires access to infrastructure and experienced process engineers
Highly skill-intensive; requires access to experienced designers and end users
U.S. experience with offshore investments
Shifted to developing countries beginning in 1960s
Cross-investments with other developed countries beginning in 1970s
Offshore investments to both highand low-wage countries beginning in the 1980s
Impact of offshoring on the U.S. industry
Offshoring helped U.S. firms respond to the initial Asian competitive threat at cost of “hollowing out” the domestic assembly sector.
Offshoring by U.S. firms has been largely offset by foreign investments in the United States.
Offshoring has entered a period of expansion while domestic employment was flat during the period of slow growth.
U.S. experience with offshore outsourcing
Asian outsourcing started in late 1960s.
Asian outsourcing started in mid-1980s.
Asian outsourcing started in the 1990s.
Impact of outsourcing on the U.S. industry
Outsourcing helps U.S. firms reduce their investment in capacity and in the variety of chip packages.
Outsourcing stimulated the emergence of a “fabless” chip sector and helps fab-owning firms reduce their risk of overcapacity.
Long-term result uncertain. May have curbed domestic hiring but allows start-ups to compete.
Source: Authors.
Clair Brown and Greg Linden
285
region. By the mid-1970s, there were dozens of U.S.-owned assembly plants throughout the region employing about 1,000 workers each.11 Developing economies in the Western hemisphere such as Mexico and El Salvador received assembly investments, but during the 1970s gradually receded from one-quarter to one-tenth of U.S. reimports in the face of political or social unrest, while East Asia made up the difference.12 The Asian countries offered plentiful low-skilled workers and stable governments that adopted proinvestment policies. All large U.S. “merchant” chip firms (those selling to other companies) invested in offshore assembly. The two primary “captive” producers (producing for internal use), IBM and AT&T, initially kept their assembly in the United States and adopted a higher level of automation than the offshore plants.13 For at least one U.S. company, Philco, an attempt to automate in the United States ended badly because of the rapid obsolescence of the equipment.14 AT&T opened chip assembly plants in Singapore and Thailand in 1985. IBM announced a large chip assembly investment in China in 2000, but in 2004 sold that operation plus a testing facility in Singapore to Amkor, a major Korea-based assembly subcontractor, as part of a plan to move from in-house offshore to 100 percent outsourced assembly.15 Several factors contributed to the movement of assembly offshore. First, Japanese manufacturers automated their assembly lines and provided stiff competition for American producers.16 Automation was a more feasible strategy for the Japanese because of their relatively greater reliance on high-volume memory chips, which involve long production runs. U.S. companies produced a wider range of products that were less economical to automate.17 Also, whereas the military had been the primary early adopter of semiconductors, its dominance was steadily replaced by the consumer electronics industry, with its attendant price pressures, during the 1960s.18 Furthermore, U.S. policy was permissive because tariffs were limited to the value added offshore, which in
11. Henderson (1989, p. 51). 12. Flamm (1985, table 3-7). 13. Flamm (1985, p. 52). 14. Flamm (1985, p. 69). 15. Mark LaPedus, “IBM Jettisons IC-Packaging Units, Sells Plants to Amkor,” Silicon Strategies, May 18, 2004. 16. Henderson (1989, p. 45). 17. Flamm (1985, p. 92). 18. Henderson (1989, p. 43).
286
Brookings Trade Forum: 2005
the case of assembly was a relatively small portion of the total—about 12 percent in the late 1970s.19 By 1977, U.S. companies employed close to 100,000 workers in offshore assembly plants, compared to 114,000 domestic employees, of whom 64,000 were directly involved in production.20 The overseas expansion resulted in a decline in domestic assembly jobs. U.S. factories were closed during recessions and new jobs added overseas during upturns. Job-loss data from the period of peak overseas expansion (1968–72) are not available; but from 1975 to 1982, 8,500 former chip assembly workers were certified for trade adjustment assistance, and another 3,000 applied but were refused.21 There was, however, an offsetting gain in consumer surplus from the low prices of the reimported chips. Flamm (1985) estimates the welfare cost of repatriating assembly and making it profitable with tariff barriers to be about $1 billion for 1983, which can be thought of as an upper bound for the consumer surplus created by the move offshore.22 In terms of assembly, the U.S. chip industry “hollowed out.” Flamm (1985) estimated that in 1978 around 80 percent of U.S. semiconductor production was assembled abroad.23 The figure in the early 2000s is probably above 95 percent, with most remaining U.S. facilities predominantly engaged in prototyping and military jobs. This, together with the fact that about a third of chip assembly is outsourced, implies that U.S. firms are doing around 60 percent of assembly in offshore subsidiaries. History suggests that the move offshore helped keep U.S. chip firms competitive with the new rivals from Asia, and this was important for protecting the remaining jobs in the industry. When Flamm published his landmark study, the U.S. chip industry was in a period of decline, but in spite of the move to offshore assembly rather than because of it. When the industry sought U.S. government help, the fabrication stage of production was where the U.S. capability was seen as deficient, and fabrication was the focus of the joint public-private research consortium, SEMATECH, launched in 1986. Thus, chip assembly provides an example where the offshoring of one part of the value chain to reduce costs was important for maintaining overall costcompetitiveness against international rivals, albeit at the expense of specific (primarily low-skilled) jobs in the short run. 19. Flamm (1985, table 3-10). 20. Flamm (1985, p. 91). 21. Flamm (1985, p. 96). 22. Flamm (1985, p. 96). 23. Flamm (1985, p. 82).
Clair Brown and Greg Linden
287
The Appearance of New Suppliers Offshore chip assembly offers an additional lesson that is worth noting briefly, namely that in-house offshoring can generate technology diffusion to foreign companies. In countries where entrepreneurial conditions are favorable, foreign investment stimulates the emergence of local companies, often started by ex-employees of the foreign company, that offer low-end versions of the foreign company’s technology. In the case of chip assembly, this has meant the emergence of a number of contract assemblers that complement, and arguably strengthen, the U.S. semiconductor industry. With Asian suppliers covering more mature technologies, the foreign subsidiaries are able to specialize in higher-value types of chip packaging. Beginning in the late 1960s, local Asian firms started offering contract chip assembly services to the U.S.-owned plants. Today roughly one-third of all chip assembly is foreign outsourced, and the figure has grown rapidly in recent years as the complexity and diversity of packages have increased.24 The top ten assembly contractors, with about 70 percent of total contracting revenue in 2003, are all Asia-based, which reflects a strong link to the initial locational choices of U.S. firms.25 The Asia-based outsourcing providers have become more technologically sophisticated and can serve as important technology partners for U.S. producers.26 The requirements for chip packages have become quite challenging with the growing complexity of chips, the need to fit into size-sensitive products like mobile phones, and an increasing danger of package-induced electrical problems. The engineers at the leading Asian assembly companies are able to participate with the chip designers at an early stage to avoid problems with the final product.27 The availability of multiple Asian suppliers allows chip companies access to a large array of package types, since few U.S. firms have the scale to 24. Amkor estimates that 32 percent of assembly and test activity was outsourced in 2004 (reported in “Amkor Technology Inc., Q1 2005,”Amkor First Quarter 2005 Corporate Presentation, www.amkor.com/IR/AMKR_Investor_presentation.pdf [June 9, 2005]).This is up from about 20 percent in 2002 (Amkor estimate reported in Richard Tortoriello, “Amkor: A Promising Play in Chips,” Business Week Online, January 13, 2003). 25. Subash Khadpe, “Outsourced Semiconductor Assembly and Test: Preparing for the Next Boom Cycle, 2006–2008,” Chip Scale Review, April 2005. 26. One of the largest assembly providers, Amkor Technology, was founded in South Korea in 1968 and technically became a U.S.-based firm through an initial public offering in 1998 during the Asian financial crisis, although control of the company did not change. 27. For a detailed case study, see Russ Arensman, “Special Delivery: How Amkor’s Packaging Proficiency Helped Cisco’s Switches,” Electronics Design Chain, Fall 2004.
288
Brookings Trade Forum: 2005
supply internally all the types they need. The availability of assembly services is also vital to the growing group of design-specialist “fabless” firms, which are discussed in the next section. The entry of companies from industrializing economies into the chip assembly business has not brought about the exit of U.S. companies from the activity. Many U.S.-owned offshore assembly plants from the industry’s early days are still in operation, and new ones are still being built, such as Intel’s assembly plants in China.28 The main reasons for keeping assembly in-house are technological, to deal with such problems as electrical interaction between the packaging and the circuitry (a growing problem as the circuits shrink), which can affect performance and reliability. There is also the strategic aspect of avoiding overdependence on an assembly contractor. Furthermore, because outsourced testing (often provided with assembly) exposes some of the chips’ embedded intellectual property, the chips most likely to be outsourced by the major chip firms are low-end or otherwise mature chips where electrical interference and intellectual property are less of a concern. Chip makers also use outsourcing to provide buffer capacity in the case of excess demand and to handle specialized packages for low-volume products, for which the outsourcing provider may achieve better scale economies by aggregating across multiple customers.
Offshore Fabrication: From Foreign Outsourcing to Industry Restructuring The case of wafer fabrication is very different from that of assembly because offshore investments were made, beginning in the 1970s, primarily for market access in Japan and Europe, where trade barriers made U.S. exports uneconomical, rather than to lower production costs.29 The employment impact of these market-seeking investments was to some extent offset by the cross-investments of European and Japanese producers in the United States. Cost reduction via offshore investments in low-wage countries was not a feasible strategy because fabrication is so capital-intensive. Labor typically accounts for 16 percent of costs (including depreciation) in U.S. fabs producing 200mm wafers, and less than 10 percent in the newer 300mm fabs, which undercuts the major labor cost advantage of most industrializing countries.30 28. “Intel to Build Second IC Assembly Plant in Chengdu,” EETimes, March 23, 2005. 29. Henderson (1989, p. 45). 30. Authors’ calculations based on data in appendix 2 of Howell and others (2003). Labor costs for 200 mm fabs are 8 percent in Taiwan and 3 percent in China.
Clair Brown and Greg Linden
289
In a survey of industry executives, Leachman and Leachman (2004) found that the top five reasons, rated very close together, for fab site selection were “Tax advantages,” “Supply of engineering and technical talent,” “Quality of water supply and reliability of utilities,” “Proximity to existing company facilities,” and “Environmental permitting process and/or other regulations.”31 Empirical research on fab investment data shows that host-country political institutions, the presence of other fabs, and a firm’s prior investment experience also affect the location of fab investments.32 This multiplicity of concerns surrounding such a major investment accounts for the relatively few cases of U.S.-built fabs in industrializing countries, even with the rich subsidies that have been offered by Singapore and others. Despite the limited fab investment in industrializing countries, some of them, especially Taiwan, have successfully fostered local chip fabrication with focused government programs. The most successful business model for these fabs is contract fabrication for chips designed elsewhere. Foreign Outsourced Fabrication The foreign outsourced fabrication of U.S.-designed chips to suppliers based mainly in Asia is a growth industry. These suppliers, known in the industry as “pure-play foundries,” manufacture chips to the designs of other companies and sell no chips of their own design. Although some integrated companies, most notably IBM, offer foundry services, the “pure-play” companies are the most important source of such services to the rapidly growing design-only (“fabless”) sector. The foundry-fabless business model, which emerged in the mid-1980s, was initially ridiculed by industry executives, most famously by Jerry Sanders, thenCEO of Advanced Micro Devices, who is reputed to have dismissed the phenomenon of outsourced fabrication with the claim that “real men have fabs.”33 The foundry model has, however, proved to be extremely successful, and the technology level of the leading foundries is now close to the industry “bleeding edge” of companies such as Intel and IBM.34 31. Leachman and Leachman (2004, p. 226). 32. Henisz and Macher (2004). 33. Although the phrase is universally attributed to Mr. Sanders, the exact date and wording are obscure. 34. Jeff Chappell and Jessica Davis, “Have the Foundries Caught Up? No Technical Laggards, Foundries Close In on the Leading Edge,” Electronic Business, June 1, 2005, p. 32. Companies at the industry’s “bleeding edge” are the first to put a new technology generation into production,
290
Brookings Trade Forum: 2005
Table 2. Six Pure-Play Foundries, 2004a Total revenue $16,695 million
Company Taiwan Semiconductor Manufacturing Corporation (TSMC) United Microelectronics (UMC) Chartered Semiconductor Semiconductor Manufacturing International Corporation (SMIC) Vanguard DongbuAnam
Country
2004 revenue (US$ millions)
2004–2003 growth (percent)
2004 share of total (percent)
Taiwan
7,648
31
45.8
Taiwan Singapore
3,900 1,103
42 52
23.4 6.6
China Taiwan S. Korea
975 474 435
166 66 32
5.8 2.8 2.6
Source: IC Insights, reported in “China Gains in 2004 Pure-Play Foundry Rankings,” EE Times, March 28, 2005. a. “Pure-play” foundries manufacture chips designed by other companies; they sell no chips of their own design.
The pure-play foundry model originated in Taiwan in 1987, when the government brought together investors, licensed mature production technology from the United States, and attracted Taiwanese engineers and managers with experience in the U.S. chip industry. The initial foundry, Taiwan Semiconductor Manufacturing Corporation (TSMC), remains the largest in an increasingly crowded field, as shown in table 2. TSMC was founded by Morris Chang, a Chinese-born, MIT-educated executive with twenty-five years’ experience at Texas Instruments who moved to Taiwan in 1985. At the time TSMC was created, a handful of U.S. chip companies, such as Xilinx and Chips & Technologies, were already outsourcing all of their manufacturing, primarily to integrated Japanese manufacturers. This arrangement entailed certain risks over access to capacity and control of intellectual property that the availability of a pure-play foundry alleviated. TSMC’s chief rival is an older Taiwanese government-backed company, United Microelectronics (UMC), which sold off its design activities in the late 1990s and adopted the pure-play foundry model. The third biggest foundry, Chartered Semiconductor, is located in Singapore and is part-owned by the government. which forces them to bear a relatively large burden of learning costs in exchange for potential firstmover advantages. Follower firms benefit from spillover knowledge flows. See Ham, Linden, and Appleyard (1998, p. 140) for an example of the adoption of new wafer sizes.
Clair Brown and Greg Linden
291
The newest entrant, China-based Semiconductor Manufacturing International Corporation (SMIC), was founded in 2000 by Richard Chang, a Taiwanese expatriate with experience in Taiwan’s foundry business following a U.S. graduate education and twenty years’ experience at Texas Instruments. SMIC’s investors include international venture capitalists and Chinese government entities. SMIC has successfully attracted a range of technology partners and customers, primarily from the United States, hired hundreds of Taiwanese engineers with foundry experience,35 and listed its shares on the New York Stock Exchange in 2004. The technology level of China’s fabrication capability is theoretically limited by the Wassenaar Arrangement of 1996, by which more than thirty countries agreed to restrict exports of dual-use technologies that might undermine international security. The agreement’s interpretation and enforcement, however, are left up to individual member states, and most countries with chip equipment industries other than the United States have been unwilling to curb exports of advanced chip-making equipment to China.36 In 2003 the U.S. government issued SMIC a special license specifying that it does not make chips for military use, which allowed U.S. equipment makers to compete on a more equal basis with their Japanese and European rivals.37 The advent of government-funded manufacturing in Asia raised concerns about a potential loss of manufacturing jobs in the United States, especially since the U.S. semiconductor industry had seen its fortunes slip during the 1980s as U.S. DRAM makers lost market dominance to Japanese companies. However, the Asian foundries contributed to the resurgence of the U.S. chip industry, since they facilitated the blossoming of design-only (fabless) chip companies, especially in California, during the 1990s.38 Since the mid-1990s, fabless revenue (with a compound annual growth rate [CAGR] of 20 percent) has been growing faster then the semiconductor industry as a whole (CAGR of 7 percent), and worldwide fabless revenue was $20.6 billion in 2003. Of the top thirty fabless firms that year, twenty were U.S.based and had a combined revenue of $13.5 billion. The next most important location for fabless companies is Taiwan, where six of the top thirty firms (with 35. “TSMC Sues SMIC,” Electronic News, December 29, 2003. 36. Mark LaPedus, “Chip-Equipment Export Rules to China Are Unclear and ‘Ineffective,’” Silicon Strategies, February 15, 2002. 37. Mark LaPedus, “SMIC Obtains Special License for Advanced U.S. Fab Gear,” Silicon Strategies, September 30, 2003. 38. Macher, Mowery, and Hodges (1998).
292
Brookings Trade Forum: 2005
Table 3. Top 10 Fabless Companies, 2004 Company (location) Qualcomm (California) Broadcom (California) ATI Technologies (Canada) Nvidia (California) SanDisk (California) Xilinx (California) MediaTek (Taiwan) Marvell Semiconductor (California) Altera (California) Conexant (California) Total revenue: $33 billion
2004 revenue (US$ millions) 3,224.0 2,400.6 2,140.9 2,010.0 1,777.1 1,588.7 1,252.5 1,224.6 1,016.4 914.6
Source: Fabless Semiconductor Association, cited in “Worldwide Fabless Revenue Grew 27% in 2004, FSA Revealed,” Nikkei Electronics Asia Online, March 18, 2005.
a combined revenue of $2.8 billion) in 2003 were located.39 Table 3 lists the top ten fabless firms in 2004. In addition to supporting the increasingly important fabless sector, the Asian foundries are also permitting integrated firms, from smaller players all the way up to the world’s third largest chip firm, Texas Instruments, to hedge the enormous risk of building new factories by using the foundries for buffer capacity and even for fabricating leading-edge chips that have a short product life or uncertain volume. The availability of buffer capacity in Asia has allowed chip producers to build less fabrication capacity, which reduces their risk of having unused capacity with a large fixed depreciation expense. Fab-owning companies (called integrated device manufacturers, or IDMs) can keep their own fabs fully booked and shift excess demand to the foundries as needed. The foundries adjust their prices as their capacity utilization varies. IDMs began shifting business to foundries in the mid-1990s and in recent years have accounted for approximately 45 percent of foundry revenue.40 Looked at another way, 20 to 25 percent of the value of the semiconductor industry is being manufactured on a (mostly foreign) outsourced basis.41 39. Data from IC Insights reported in Peter Clarke, “SanDisk, Silicon Labs Leap in 2003 Fabless Rankings,” Silicon Strategies, March 18, 2004. 40. Data reported by Semico Research Corp, reported in Joanne Itow, “System Houses Remain Weak Link for Silicon Foundries,” Silicon Strategies, May 11, 2004. 41. Authors’ calculations, assuming that foundry revenue represents about one-third the value of the final chip price.
Clair Brown and Greg Linden
293
Although the outsourcing trend shifts manufacturing to Asia, it seems unlikely that the U.S. semiconductor industry will ever entirely cease domestic fabrication. At the leading edge, companies such as Intel, IBM, and Texas Instruments derive advantage from implementing advanced process technologies for their flagship products at the earliest possible time. Companies such as Freescale (formerly Motorola) and Micron benefit from running nonstandard processes. In all these cases, fab ownership affords closer interaction between the design and manufacturing functions and helps to ensure the protection of key trade secrets. Nevertheless, as the foundries add capacity and smaller fab-owning firms decline to invest in new plants, the number of fab-owning firms will likely decline. There is a limit to the eventual size of the foundry sector. For example, Intel will continue internal (but not entirely domestic) fabrication of its PC microprocessors because they require leading-edge process technology that is part of the company’s competitive advantage.42 Samsung, the leading memory maker, will continue internal fabrication of its memory chips, since they require highvolume, low-cost production runs with short product life cycles. Intel and Samsung, the top two semiconductor companies worldwide, accounted for 21.4 percent of industry sales in 2004.43 The major category of chips manufactured by foundries are logic chips, including a range of general-purpose and applicationspecific products, and mixed-signal chips, which primarily use standard processes. All told, outsourced manufacturing, the bulk of which occurs overseas, will probably never exceed 50 percent of the semiconductor industry.44 Offshore Fab Investments Another way to look at semiconductor fabrication in the United States is to consider where fab investment takes place. The data reveal a growing shift away from U.S. investment in domestic fabs. These historical data come from our colleagues Leachman and Leachman (2004). Table 4 shows fab capacity by location in 1980, 1990, and 2001.45 The shift of capacity from Japan and the United States to the rest of Asia (primarily South Korea and Taiwan) is striking. Japan 42. Intel has multiple fabs in the United States and also fabricates microprocessors in Israel (since 1985) and Ireland (since 1993). 43. Gartner Dataquest data reported in Mark LaPedus, “Gartner Differs with Rivals in Top-10 Chip Rankings,” EETimes, March 23, 2005. 44. For a similar analysis see Ed Sperling, “More Changes Ahead for Foundries, Industry,” Electronic News, December 4, 2003. 45. Because older fabs use a range of wafer sizes and linewidths, the underlying data have been normalized using a capacity metric based on the number of functions, where a function is one memory bit or one logic gate.
294
Brookings Trade Forum: 2005
Table 4. Regional Location and Ownership of Worldwide Fabrication Capacity a Year
Asia (excluding Japan)
Europe/ Middle East
Japan
North America
1980 1990 2001b
4 (3) 12 (12) 38 (39)
16 (15) 13 (9) 13 (8)
38 (37) 45 (45) 20 (24)
42 (44) 30 (36) 29 (38)
Source: Leachman and Leachman (2004), tables 8.2, 8.4. a. For each year, the percentage of location capacity is shown first and ownership capacity is shown in parentheses. b. The ownership row total for 2001 adds to more than 100 because jointly owned capacity was credited in full to all owners.
and the United States accounted for 80 percent of fab capacity in 1980, but only 49 percent of capacity in 2001. However, the same data by region of ownership show that although the rise of capacity owned by companies in Asia (excluding Japan) mirrors the rise of location capacity, the decline of capacity owned by U.S. companies is less severe than the fall in capacity located in the United States. Although only 29 percent of fab capacity in 2001 was in the United States, U.S. companies had an ownership stake in almost 40 percent of global capacity. Because so many fabs owned by companies in one region are located in another, these data do not directly answer the question of how much U.S.-owned capacity is located outside the United States. Rob Leachman generously helped us to make this calculation.46 In 2001 approximately one-third of U.S.-owned capacity was located offshore (18.6 percent in Europe and the Middle East; 13 percent in Japan; and 3 percent in the rest of Asia). The high number of fabs in Japan and Europe reflects the rise of joint ventures to share risk as the cost of fabs increased. Conversely, about 22 percent of the fab capacity located in North America was owned by companies based in other regions. The trend toward Asian manufacturing is continuing. The largest—and potentially most efficient—fabs today use 300mm (12-inch) diameter wafers.47 Taiwan already has one of the largest concentrations of these mega-factories, which 46. The Leachman data do not include ownership shares for jointly owned fabs. We divided such fabs by the number of regions (two or three) involved in ownership to estimate the U.S. share. As much as 10 percent of U.S.-owned capacity was in joint venture fabs in 2001, but those fabs were spread across all regions, so our estimation error is not likely to be more than 1 or 2 percent up or down from the figures in the text. 47. Many of the 300mm generation of fabs are still being expanded to efficient scale. Projected cost savings of about 30 percent per chip are expected from these fabs when they are equipped to efficient scale and running at full capacity (Rob Leachman, personal communication, May 2005). When such fabs are not running at full volume, losses mount quickly because of the rapid equipment depreciation.
Clair Brown and Greg Linden
295
the foundries and memory chip producers need to keep their unit costs low, even as they greatly increase those firms’ exposure to the risk of excess capacity. The major Japanese companies have also made a significant commitment to the technology, although most of these fabs are not yet in full operation. Leading U.S. firms, such as Texas Instruments, were early adopters of the 300mm technology but are slowing down their commitment to new fabs, in part because of their ability to turn to foundries for buffer capacity. SMIC’s newest fab constitutes China’s entry in the 300mm list. As of October 2004, thirty-six 300mm fabs were in various stages of construction, in addition to the twenty-four already in production. They were located as follows: Japan (24 percent); the United States (24 percent); Taiwan (19 percent); Europe (14 percent); South Korea (12 percent); China (5 percent); and Singapore (3 percent).48 Each fab requires annual revenues of well over $1 billion to be profitable.49 As has occurred following previous construction cycles in the semiconductor industry, the new fabs will likely result in a period of overcapacity until demand grows sufficiently. Fabs and employment This section discusses the impact of globalized fabrication on U.S. white-collar employment. Although the advent of outsourced manufacturing in Asia did not represent a transfer of capacity that entailed a shutdown of U.S. facilities, it has very probably reduced the number of facilities that would otherwise have been built here, which represents the loss of a number of potentially high-skilled jobs. TSMC, which accounts for about half the pure-play foundry market, currently operates one 300mm and five 200mm fabs in Taiwan (plus one in Washington state). We have detailed staffing data on earlier-generation fabs gathered in the mid1990s by the Berkeley Competitive Semiconductor Manufacturing (CSM) Program.50 The data describe the average employment distribution at a sample of fabs running 150mm and 200mm wafers in four countries.51
48. Data from Strategic Marketing Associates (www.scfab.com). 49. Authors’ calculation, suggested by Toshihiko Osada, based on data in Appendix 2 of Howell and others (2003). Annual depreciation and operating expense for a U.S.-based 300mm fab running 6,000 wafers per week on a 90-nanometer process totaled $975,000. 50. The CSM program is a multidisciplinary study of the semiconductor industry established in 1991 by a grant from the Alfred P. Sloan Foundation and with additional support from the semiconductor industry. Further details are available at esrc.berkeley.edu/csm/. 51. The sample consisted of seven same-company pairs. See the note to table 5 for complete details.
296
Brookings Trade Forum: 2005
Table 5. Workforce Composition at Semiconductor Fabs in Four Countries a Type of workerb Operators Technicians Engineers Total
150mm (percent)
200mm (percent)
547 (73%) 91 (12%) 114 (15%) 752
470 (62%) 107 (14%) 181 (24%) 758
Source: Brown and Campbell (2001). a. Twenty-three fabs in four countries were part of the CSM survey. For this table, the 150mm fabs were matched to the 200mm fabs by company, so that the company human resource policies are comparable between the two groups, which reduced the sample to fourteen. Data collected during the period 1999 to 2001. b. Mean head count in matched 150mm and 200mm fabs.
As wafer size increases, output rises for a given level of wafer throughput, and both materials handling and information systems become more automated to more safely handle the increased weight and value of each wafer and to minimize human error. Automation changes the composition of the workforce as the need increases for engineers and declines for operators. In the CSM data, engineers increase from 15 percent to 24 percent of the total workforce between 150mm and 200mm generation plants, with a corresponding decline in operators from 73 percent to 62 percent (see table 5), even as the overall employment level of the fab stayed approximately the same at about 750 workers. The shifting of jobs from operators to engineers in the transition from 150mm to 200mm fabs resulted in the growth of engineering jobs paying from $29,000 to $56,000 per year and the decline in operator jobs paying $14,000 to $37,000 per year (see table 6). The initial pay of technicians and engineers was more than one-third higher in the high-tech 200mm fabs, and their pay premium over operators increased. A look at the returns to experience, which are proxied by the ratio of maximum to initial pay, shows that engineers fare less well than technicians and operators. The returns to experience for operators and technicians stayed the same in the 200mm as in the 150mm fab. Experienced engineers lost out: their mean maximum salary was actually lower in the 200mm fabs. In interviews, we learned that fabs liked having young engineers with knowledge of new technology, and they did not worry about losing older engineers. Over time, consequently, fabs were willing to increase wages of new hires without raising the wages of experienced engineers. Rapidly changing technology plus an ample supply of new hires and low turnover allowed the companies to flatten engineers’ career ladders with no adverse consequences.
Clair Brown and Greg Linden
297
Table 6. Work Force Compensation in Semiconductor Fabsa 150mm fab
200mm fab
Type of worker b
Initial pay
Maximum pay
Initial pay
Maximum pay
Operators (hourly) Technicians (hourly) Engineers (monthly)
5.88 6.68 1,785
15.47 11.50 5,019
7.12 9.12 2,381
18.44 15.83 4,689
Source: Brown and Campbell (2001). a. See note a, table 5. b. Mean wage or salary in matched 150mm and 200mm fabs, U.S. dollars
We do not have comparable data for the 300mm fab, which typically costs $2 billion–$3 billion (depending upon the size), has 100 percent automation of materials handling and wafer processing, and fabricates a wafer that is 2.25 times larger than the 200mm wafer. The cost per chip in the 300mm fab is more than 30 percent lower than in the 200mm fab. Because the new 300mm fabs are processing advanced circuits, such as those using 90 nanometer (nm) processes, the amount of inspection, metrology steps, and in-line engineering-related activities are significantly higher than those of their older 200mm counterparts for the same wafer throughput. As a result, most of the 300mm worker savings achieved with the automation of materials handling, often cited to allow approximately 30 percent less labor input, are now being reapplied to the new engineering tasks, which are much higher value added and more intellectually challenging and involve more troubleshooting. Therefore, the advanced factory automation has not resulted in a reduction of the number of workers; instead, there has been a shift in task composition. The percentage of workers with higher engineering and technical problem-solving skills has greatly increased, while the percentage of workers needed for wafer movement and equipment starting and stopping has greatly decreased. However, the proportion of engineers has not increased.52 So what is the net employment impact of fabrication offshoring? According to the Semiconductor Industry Association, U.S. chip firms employed 103,000 engineers in 2003, of which 30 percent were located offshore. The share of offshore employment had not grown since 1998, and in fact fell during the Internet/ telecom bubble before returning to 30 percent.53 52. E-mail exchange between a U.S. chip company executive and Brown, April 25, 2005. 53. Data are from David R. Ferrell, “SIA Workforce Strategy Overview,” presentation to Electrical and Computer Engineering Department Heads Association, annual meeting, March 2005 (available at www.ecedha.org [April 19, 2005]).
298
Brookings Trade Forum: 2005
As a back-of-the-envelope calculation, we estimate that if all foundry production were based in the United States instead of Asia, it might add 11,000 jobs, of which some 2,600 would be highly paid engineers.54 But it must be noted that not all foundry sales are to U.S. customers. In 2003, for example, half of TSMC’s gross revenue came from non-U.S. sources.55 As a point of comparison, the Fabless Semiconductor Association reported that publicly traded fabless companies in North America employed approximately 45,000 workers as of December 2004.56 A review of company information suggests that more than half of those were software or hardware engineers, although an unknown share of them were located offshore. In summary, offshore investments in fabrication were driven more by market access concerns than by cost reduction and have been at least partially offset by reciprocal investments from leading foreign producers. Meanwhile, the availability of outsourced fabrication in Asia played to the U.S. strength in design by facilitating the emergence of the fabless chip industry. The Asian foundries are probably contributing to a long-run reduction in U.S. chip manufacturing, but any loss of U.S. manufacturing jobs has been gradual, and the loss of chip manufacturing jobs to foundries has probably been offset to some extent by the increase in design jobs. The reliance of the U.S. semiconductor industry on highend design jobs is one of the reasons that chip design, the latest frontier for offshoring, may be a cause for greater concern.
Offshore Design The picture for offshore design by U.S. semiconductor companies is still taking shape. Although some design has been done offshore since at least the 1970s, the pace of offshoring has increased noticeably in the past few years. The available evidence of the impact to date on the U.S. market for chip design engineers is mixed. 54. TSMC, which accounts for about half the foundry industry, has one 150mm, one 300mm, and five-and-a-half 200mm fabs outside the United States. These fabs probably have different rated capacities, but we can approximate employment by calculating 750 workers per plant, which works out to 5,625. Doubling that to approximate the entire foundry sector brings us to 11,250. 55. Note 27c of Form 20-F filed by TSMC with the Securities and Exchange Commission for fiscal year ended December 31, 2003. 56. Fabless Semiconductor Association (FSA), “Global Fabless Fundings and Financials Report, Q4 2004” (available to members at www.fsa.org/publications/financials/index.asp).
Clair Brown and Greg Linden
299
The Economics of Chip Design Chip design is a highly skill-intensive job that employs only college-trained engineers. A couple of medium-sized chip designs will employ as many electrical engineers as a fab for a year or more (although the skills are not directly transferable). A complex chip design such as Intel’s Pentium 4, with 42 million transistors on a 180nm linewidth process, engaged hundreds of engineers for the full length of the five-year project.57 Design teams can also be as small as a few engineers, and project duration varies from months to years. Team size depends on the complexity of the project, the speed with which it must be completed, and the resources available. The design of an integrated circuit is a hierarchical procedure that passes through identifiable stages with feedback as needed. Considerably simplified, the stages are specification, logic design, and physical design. Once the chip has reached the prototype stage, it needs to be validated in a hardware simulation of a complete system. Parallel with this process, the design must be repeatedly verified, and the software that will be part of the chip, and that will run on it, needs to be written. The highest-level design stage is the general specification for how the chip as a whole will behave within the system of which it is a part. This is a high-valueadded function that applies the company’s market knowledge and intellectual property in deciding what feature set will be most profitable. The next stage, logic design, uses symbolic abstractions to describe how signals will be processed within the chip, first at the register level, then at the gate level. The final stage, physical design, involves the translation of the abstract version into a map of actual wires and devices interconnecting across multiple layers on the silicon surface. Electronic design automation (EDA) includes both the use of software tools by engineers to realize their designs and the automation of specific parts of the design with less engineering input, especially at the later stages of mainstream digital designs. Table 7 shows the change in the effort required at each stage of design over succeeding generations of process technology, from 350nm linewidths, which was in volume production in the mid-1990s, to 130nm, in volume production in 2003. The underlying project is assumed to be a digital logic design, the 57. Terry Costlow, “Comms Held Pentium 4 Team Together,” EETimes, November 6, 2000. “Linewidth” refers to the size of the features etched on a wafer during the fabrication process. Each semiconductor process generation is named for the smallest feature that can be produced.
300
Brookings Trade Forum: 2005
Table 7. Engineer Hours Required to Design 1 Million Logic Transistorsa
Task
350nm
250nm
180nm
130nm
Change from 350nm to 130nm (percent)
Specification Logic design Physical design Validation Software Total
23.0 714.2 311.0 103.7 378.4 1,530.3
29.8 738.4 357.2 127.6 672.4 1,925.4
91.4 756.4 391.7 164.5 985.7 2,389.7
271.6 837.7 473.5 197.4 1,798.3 3,578.5
1,081 17 52 90 375 134
Source: International Business Strategies (see note 85), used with permission. a. The table normalizes the hours required for 1 million transistors at each generation based on the assumption that the representative project is increasingly complex: 2 million transistors at 350nm, 5 million at 250nm, 20 million at 180nm, and 40 million at 130nm.
industry’s typical product. Other types of design, such as analog or memory chips, require different engineering inputs. The representative chip is assumed to contain more transistors at each linewidth as miniaturization allows more functions to be packed onto a chip. The greater complexity results in the design hours increasing faster than the number of transistors. In raw terms (transistors per engineer per year), design engineer productivity improved by a factor of more than twenty during the 1990s,58 which would have reduced the engineering hours required per design except for the enormous increase in both the number of transistors and complexity. The total design hours for the full, representative logic chip increased by a factor of almost fifty between 350nm and 130nm technology. The biggest change involves the importance of software in the design process, which now accounts for one-half of the total engineer hours. Although software is typically not considered part of chip design as such, software expertise is increasingly important for the competitive advantage of semiconductor firms.59 Chips today are often integrated to system-level complexity because of the size, reliability, and other advantages this brings. Greater integration means that the system software must be generated in parallel with the system-level chip for reasons of coherence and, especially, time-to-market. It is this need to plan carefully for the hardware-software codesign that has caused the specification portion of chip designs to explode by 1,081 percent over the past four technology generations. 58. Semiconductor Industry Association. “International Technology Roadmap for Semiconductors: Design” (public.itrs.net/Files/2003ITRS/Design2003.pdf [April 19, 2005]). 59. Linden, Brown, and Appleyard (2004).
Clair Brown and Greg Linden
301
The software effort itself has increased by 375 percent. According to one software executive, a typical chip in 1995 went into a stand-alone product and required 100,000 lines of code. In 2002 a typical chip for a networked programmable product required a million lines of code.60 The software, plus the greater complexity of chips themselves, has caused design validation hours to grow by 90 percent for each million transistors. By comparison, the number of logic and physical design engineering hours for each million transistors has grown by a relatively modest 17 percent and 52 percent, respectively. This is largely because, as chips have gotten more complex, the process of chip design has become more automated.61 The number of transistors that can be fabricated on a given area of silicon has doubled every eighteen months for roughly forty years, a phenomenon known as “Moore’s Law,” after Gordon Moore, one of the founders of Intel. In the early 1960s, digital ICs contained fewer than fifty transistors.62 The industry can now place some 100 million transistors on a chip, and Intel predicts a billion-transistor processor by 2007.63 The automation that enables the design of today’s complex chips has evolved in parallel with the rise in complexity. At the beginning of the industry, designs were hand-drawn and hand-transferred to a template that was used to make the actual circuit. In the 1970s the later stages of the process were computerized, and in the 1980s they were automated. The introduction of automation, along with the advent of high-bandwidth telecommunications, gave chip companies the ability to subdivide the design process across multiple locations. These advances pertain primarily to digital designs—that is, those that work on binary streams of data. Designs that use all, or mostly, analog circuits, which process continuous signals such as sound waves, are also done with EDA tools, but are not so easily automated and require more experienced designers with specific training. In-house Offshore Design Companies pursue the in-house offshoring of chip design for any of the three reasons pertaining to competitive advantage: closer contact with customers, access to specialized skilled labor, and cost reduction. Most early offshore 60. Jerry Fiddler, chairman of Wind River Systems, cited in Richard Goering, “Keynoter Says Chip Value Is in Its Intellectual Property,” EETimes, June 14, 2002. 61. Hemani (2004). 62. Borrus (1988, p. 75). 63. David Lammers, “Intel Readies Road Map for Billion-Transistor Processors,” EETimes, January 4, 2002.
302
Brookings Trade Forum: 2005
design investments by U.S. companies through the 1970s, like offshore fabrication investments of the same period, were market-driven and limited to Japan and western Europe.64 By the mid-1980s, a handful of offshore design investments had been made in Hong Kong, Taiwan, and Singapore, which are among the more advanced economies of East Asia outside Japan.65 These design centers were dedicated to adapting existing chips to local market needs.66 The prime example of the market access motivation for offshore investment is the “application-specific” IC (ASIC), a logic chip designed for a specific customer. U.S.-based ASIC producers like IBM and LSI Logic have established design centers in all major markets of Europe and Asia to facilitate the interaction of their engineers with their customers. These companies also maintain other offshore design centers that develop the “cells” or building blocks that are later combined in various ways, and these centers may be low-cost-seeking or skill-seeking. Specialized skills are another important reason that U.S. semiconductor companies invest overseas. Britain, for example, has developed expertise in consumer multimedia, and Scandinavian countries are noted for their skills in wireless network technology. Companies often gain access to these specialized skill bases by acquiring an existing company that continues as a subsidiary. Examples abound. In 2000, Broadcom acquired Element 14, a British fabless company with sixty-eight employees specializing in central office ADSL technology that became Broadcom UK Ltd.67 In 2001, Agilent acquired Sirius, a Belgian designer of cellular chips for the CDMA standard with nineteen employees, and made it a research and design center for next-generation cellular technology.68 In 2005, Intel acquired Oplus, a successful Israeli maker of chips for digital television with 100 workers; it will remain an independent subsidiary.69 As was true of fabrication, in-house design offshoring works both ways, and many foreign companies maintain a Silicon Valley or other U.S. design center, where they can take advantage of the high skills and productivity available there and have access to U.S. customers. Philips of the Netherlands, for example, bought VLSI Technology, a major ASIC company with over 2,000 employees 64. Henderson (1989, p. 48). 65. Henderson (1989, figure 4.2). 66. Henderson (1989, p. 58). 67. Peter Clarke, “Broadcom to Acquire Element 14 in $600M Deal,” EETimes, October 4, 2000. 68. “Agilent to Buy Belgium’s Sirius to Offer New CDMA Chip Solutions,” Silicon Strategies, May 21, 2001. 69. Peter Clarke “Intel Buys into Consumer Sector with Oplus Acquisition,” Silicon Strategies, February 24, 2005.
Clair Brown and Greg Linden
303
(about one-third of whom were fab workers), in 1999 for nearly $1 billion.70 Hitachi Semiconductor has a U.S. design group several hundred strong.71 Toshiba has a network of seven ASIC design centers around the United States.72 Even foreign start-ups may need to have a U.S. design team to work with U.S. customers or to access leading-edge analog design skills. The desire for cost reduction is perhaps the fastest-growing reason for inhouse design offshoring—or at least it is getting the most attention. For Silicon Valley firms, some cost reduction is available by opening satellite design centers elsewhere in the United States, where some locations have average engineering salaries that are up to 20 percent lower than salaries in Silicon Valley. But they are still much higher than salaries in India and elsewhere.73 The prospects for cost reduction offshore are better than ever because of changes in the past twenty years that have seen high-bandwidth infrastructure extended around the globe and the economic liberalization of large areas in eastern Europe, and especially Asia.74 Designing a chip in more than one location presents a number of managerial challenges, as we have learned from interviews and press reports. The sacrifice of face-to-face interaction between different parts of the design team can adversely affect productivity, and distance makes it harder to evaluate and reward individual contributions to team performance. Task assignments must be more carefully codified for offshore teams than for locally based engineers, and managers will need to travel periodically between locations. When the separation is across borders, cultural differences can also make communication less effective. An Intel engineer was reported to say that cultural differences were the single biggest problem in managing design teams between California and Israel, and this separation did not involve any language differences.75 Cost-driven in-house offshoring incurs other costs that partially offset the difference in salaries, especially during the early stages of establishing an offshore 70. “Philips to Acquire VLSI Technology for $953 Million,” Semiconductor Business News, May 3, 1999. 71. “Hitachi Forms North America Semiconductor Systems Solutions Unit,” Hitachi press release, September 2, 1998 (www.hitachi.com). 72. “Toshiba Expands Soc Design Support Network with Opening of San Diego Design Center,” Toshiba press release, November 26, 2002. 73. Robert Bellinger, “Mean Wages Edge Closer to Six-Figure Mark,” EETimes, August 25, 2004. 74. Dieter Ernst, “Internationalisation of Innovation: Why Is Chip Design Moving to Asia?” Working Paper 64. Honolulu, Hi.: East-West Center, 2003, rev. March 2004. 75. Richard Goering, “Global Chip Design Raises Promises and Challenges,” EETimes, January 11, 1999.
304
Brookings Trade Forum: 2005
design center. One that is often mentioned is the lower quality and productivity of inexperienced engineers. This raises monitoring costs, and offshore engineers may also require a longer training period than a U.S. team would need. Additional controls may also be needed to protect key intellectual property. According to a venture capitalist, the actual savings from going offshore is more likely to be 25 to 50 percent than the 80 to 90 percent suggested by a simple salary comparison.76 Design offshoring can run up against national security barriers. For example, the U.S. government has placed limits on the export of advanced encryption technology. Communications chips that employ such technology are difficult to design offshore. Either the chip design must be compartmentalized, with the encryption block designed only in the United States, or government approval, subject to possible delays, must be obtained in advance.77 Yet despite these pitfalls, the amount of offshore design in industrializing economies has noticeably expanded since the mid-1990s. Some companies value the opportunity to design on a twenty-four-hour cycle because of the enormous pressure to reach the market ahead of, or no later than, competitors. One established U.S. chip company adopted a rolling cycle between design centers in the United States, Europe, and India.78 More common is the binational arrangement used by a Silicon Valley start-up that had all of its design beyond the initial specification done by a Chinese subsidiary established only months after the head office was set up. Ten executives in the head office had to train the mostly inexperienced staff in Beijing, which was about thirty strong.79 The Silicon Valley staff would review Beijing’s work from the previous day, then spend up to three hours on the phone (starting around 5:00 p.m. California time) providing feedback and reviewing assignments for that day in Beijing. In a singlelocation firm, this work-feedback cycle would take two days. Venture capitalists have reportedly begun to require start-ups to include some offshoring in their business plans in order to better leverage their resources. A typical comment is, “We don’t fund chip designs that don’t outsource to India. If you rely on Indian contractors for the things they do well, you can get a chip out for under $10 million. If you don’t, you can’t, and you won’t be competitive. It’s that simple.”80 PortalPlayer, the company behind the key multimedia chip in 76. Interview with a U.S. venture capitalist, May 6, 2004. 77. Interview with a U.S. chip company, December 16, 2004. 78. Interview by Linden with a U.S. chip company, April 16, 1998. 79. Interview with a U.S. chip company, August 2004. 80. William Quigley, managing director at Clearstone Venture Partners (Menlo Park, Calif.), quoted in Ron Wilson, “Venture Capitalist Explains New Rules for IC Startups,” EETimes, January 16, 2003.
Clair Brown and Greg Linden
305
Apple’s iPod, is a recent example of a successful start-up that set up an Indian software and chip design subsidiary within a few months of its founding in 1999.81 Domestic and Foreign Outsourcing of Design Low-cost design engineering resources can also be tapped through international outsourcing, although to date most design outsourcing by U.S. companies has taken place domestically. All parts of a design, from specification to finished chips, can be outsourced. In addition to the traditional work-to-order model, companies can also license standardized functional sections (for example, a USB interface) of a system-level chip predesigned at the logic or physical level to save time. These reusable modules are known in the industry as “cores” or “IP blocks.”82 The easiest part of chip design to outsource is physical design because it is a relatively standardized task. It is also the least sensitive part of design in terms of revealing the customer’s intellectual property. However, for designs requiring leading-edge process technology such as 90nm linewidths, layout has become much less straightforward because of the sensitivity of the atomic-scale wiring. In such a case, physical design is likely to be outsourced only by small and mediumsized companies that lack the resources to develop the necessary expertise inhouse. On the other hand, we interviewed one (well-funded) start-up whose initial design was so complex that outsourcing any parts was not a viable option.83 Another design function that is frequently outsourced is logic verification, the resource-intensive task of making sure that the first stages of the physical implementation correctly translate the abstract logic. At the other extreme, architectural design and the design of key functional blocks containing proprietary algorithms are the least likely to be outsourced because of the risk of exposing proprietary knowledge.84 The availability of outsourcing (foreign or domestic) is particularly important for small companies and start-ups because of the relatively large fixed cost of EDA tools, which are usually licensed based on the number of engineers to use 81. Jude Pinto, “Designs for Digital Audio, Auto Electronics,” Nikkei Electronics Asia, October 2002. 82. Linden and Somaya (2003). 83. Interview with a U.S. chip company, November 9, 2004. 84. Richard Wallace and Nicolas Mokhoff, “Outsourcing Trend Proves: Complex by Design,” EETimes, January 31, 2005; and interview with a U.S.-based semiconductor industry consultant, April 26, 2004.
306
Brookings Trade Forum: 2005
them. One consultant estimated that the minimum annual software expense for a small company is $10 million.85 For the industry as a whole, EDA expense runs close to 1 percent of revenue. In that case, a company earning less than $1 billion in revenue would be below the efficient scale for in-house design. Only the nine largest fabless companies met that criterion in 2004. One consultant estimated that outsourcing even within the United States would save a small startup that does fewer than five designs a year up to two-thirds the cost of doing the work in-house.86 Systems companies, such as Apple Computer and Cisco, are another type of customer for outsourced design services. Although these companies often design chips in-house either to protect intellectual property or to reduce the cost of custom chips, they may turn to outside (and possibly offshore) service providers for part of the design process. A great deal of outsourcing takes place in the United States. Many interviewees reported that they outsource physical design to small local companies on an as-needed basis. The leading suppliers of design services worldwide are the leading design automation software vendors, Cadence Design Systems, Synopsys, and Mentor Graphics. Their annual services revenue is about $300 million out of a total outsourced design market estimated at $2.5 billion.87 As this suggests, the remaining market is highly fragmented. As might be expected from the increasing interaction of physical design with advanced processes mentioned previously, foundries work closely with design services providers. TSMC and UMC each has equity ties to a Taiwanese design service provider (Global UniChip and Faraday, respectively). In China, the emergence of low-cost foundries has also given rise to design services companies. The most advanced of these, IPCore and VeriSilicon, were founded in 2001 by executives with years of experience in the United States and Asia. In India, despite the lack of any significant chip manufacturing, large IT service providers such as Wipro and Tata Consultancy Services have expanded into semiconductor design for international clients. Elsewhere there are dozens of companies around the world able to help customers complete all or part of their chip designs. However, most of the concern about foreign competition from low-cost chip designers focuses on China, Taiwan, and India, which are the countries that 85. International Business Strategies, “Analysis of the Relationship between EDA Expenditures and Competitive Positioning of IC Vendors: A Custom Study for EDA Consortium,” 2002 (www.edac.org/downloads/resources/profitability/HandelJonesReport.pdf [April 15, 2005]). 86. Interview with a U.S.-based semiconductor industry consultant, April 26, 2004. 87. Richard Goering, “Complex Chips Reignite Demand for Design Services,” EETimes, October 11, 2004.
Clair Brown and Greg Linden
307
we believe will have the greatest impact on the industry in the years ahead because of their large and growing number of chip designers. Chip Design in Asia Offshore design is driven in part by the steady advance of semiconductor process technology, which has created vast areas of “silicon real estate” for complex chips that could potentially be designed. The inability of design automation to keep pace with Moore’s Law is sometimes referred to as a design “productivity gap.”88 The gap is to some extent exaggerated, since a relatively small percentage of designs are done at the leading edge in any one year.89 Nevertheless, there is an acknowledged need for more design engineers than are available in the United States, especially those with systems and analog skills, as the industry continues to grow amid the increase in design complexity. For a clearer picture of the global market for design engineers, table 8 provides rough estimates of chip designer salaries, the number of annual engineering (excluding software) college graduates, the number of active chip designers (excluding embedded software engineers), and a rating of the intellectual property protection regime in the United States and the four key Asian countries. The numbers, which are based on a combination of published sources and interviews, suggest that engineers in the United States and Japan earn much higher salaries than Asian engineers. These data are imprecise and intended as a general guide only. The salaries are rough estimates, and their variance is large. The salaries are for engineers with five or more years experience in the United States and for engineers age 40 in Japan, since that is the approximate age at which they exit the union and transfer to a system that allows more variation in rank and pay within cohorts. The design engineers in the other countries tend to be younger and less experienced, but wages are reportedly rising rapidly in China and India. For example, the salary range offered for a design engineer with one to three years experience by SanDisk in Bangalore at jobstreet.com in June 2005 was $9,200 to $18,400 (at 43.52 Indian rupees to the dollar).
88. Semiconductor Industry Association, “International Technology Roadmap For Semiconductors: Design” (public.itrs.net/Files/2003ITRS/Home2003.htm [April 19, 2005]). 89. Erwin Ofner, Jari Nurmi, Jan Madsen, Jouni Isoaho, and Hannu Tenhunen, “SoC-Mobinet, R&D and Education in SoC Design,” presentation at International Symposium on System-onChip 2004, Tampere, Finland, November 16–18, 2004 (www.cs.tut.fi/soc/Tenhunen04.pdf [April 19, 2005]).
308
Brookings Trade Forum: 2005
Table 8. Recent Engineering Statistics, Selected Countries
Country United States Japan Taiwan China India
Annual base salary for a chip design engineer (dollars)
Number of engineering bachelor’s degrees conferred annually (excluding software)
Number of chip designers
Intellectual property protection, 2002 (10 = high)
82,000 60,000 30,000 15,000 15,000
60,000 100,000 25,000 200,000 120,000
45,000 n.a.a 14,000 5,000 5,000
8.7 6.2 6.7 4.0 4.2
Sources: U.S. salary from 2004 BLS Occupational Employment Statistics website (average for electronics and software engineers in NAICS 334400, Semiconductor Industry); Japan salary (average for circuit designer and embedded software engineers age 40) from Intelligence Corporation’s data on job offers in 2003; Taiwan salary information from March 2005 interview with U.S. executive in Taiwan; China and India salaries are estimated from interviews, industry literature, and online job offerings; number of degrees, from 2000 or 2001, from National Science Foundation, “Science and Engineering Indicators 2004,” appendix table 2-33, except for India, which is a rough estimate reported in Carl Bialik, “Outsourcing Fears Help Inflate Some Numbers,” WSJ.com, August 26, 2005; number of chip designers in the United States from iSuppli as reported in “Another Lure of Outsourcing: Job Expertise,” WSJ.com, April 12, 2004; number of chip designers in Taiwan from interview with Taiwan government consultant to industry, March 2005; number of chip designers in China and India is estimated from industry literature and discussions with industry analysts; intellectual property protection data from World Economic Forum as cited in Economic Freedom of the World, 2004 Annual Report, chapter 3 (Vancouver, Canada: Fraser Institute). All numbers rounded to reflect lack of precision. a. We have been unable to obtain an estimate of the number of chip designers in Japan.
The salary gap is narrower for comparable key employees. A 1999 report claimed that the salary ratio between the United States and India for experienced design engineers and managers was only 3-to-1.90 Profit-sharing bonuses vary over the business cycle in the United States and Taiwan and can be an important part of compensation. Benefits, which include health insurance and Social Security, and options also cloud the picture in the United States. The other columns of table 8 also must be interpreted carefully. The number of engineering graduates is only an indicator of political and social commitment to the discipline and does not translate to chip design capability. According to some sources, the number of chip designers being added each year in India and China is on the order of 400 each.91 Even the stated number of chip designers in the third column can be misleading, since there is disagreement over the definition of “chip designer.” One industry executive claimed that the number of “qualified IC designers” in China
90. Richard Goering, “Special Report: India Awakens as Potential Chip-Design Giant,” EETimes, January 22, 1999. 91. For India, see Deepika Janardhan, “Designs on the Future,” Express IT People, February 10, 2003; for China, see PricewaterhouseCoopers, “China’s Impact on the Semiconductor Industry,” December 2004, p. 7.
Clair Brown and Greg Linden
309
is only 500.92 One Taiwan consultant did not even consider the later (and lowerskilled) stage of physical design, called “place and route,” to be part of chip design.93 This group amounts to about 30 percent of Taiwan designers as we count them in the table. Last, weakness in the ratings on intellectual property protection may be driven by lapses in specific sectors such as pharmaceuticals, trademarked goods, or recorded media, which are not relevant to the semiconductor industry. Despite their lack of precision, these data indicate that Asian design engineers, especially from the emerging giant economies of China and India, represent an important source of supplemental engineering talent as well as a possible competitive threat. To attempt to get a clearer picture of the worldwide availability of “qualified IC designers,” we consulted the Institute of Electrical and Electronics Engineers (IEEE), the leading professional organization for engineers; almost 40 percent of its 365,000 members are located outside the United States.94 Of IEEE’s several technical societies, the one most closely associated with chip design is the SolidState Circuits Society (SSCS). Among other benefits, membership in the SSCS reflects an interest in accessing the latest research in the field.95 We compared the geographic distribution of membership in 2001 with that in 1991 (estimated) for individual countries.96 We limited our attention to a mix of developing and developed countries with an active commercial chip-design sector (see figure 1).97 Because the United States dominates membership and requires a different scale, it is not shown in figure 1. In 2001 the SSCS had 19,715 members worldwide, of whom 56 percent were outside the United States. This was up from 13,788 in 1991, when only 45 percent were outside the United States. Among the countries shown, China, Taiwan, and India had three of the four fastest-growing SSCS memberships. The United 92. PricewaterhouseCoopers, “China’s Impact,” p. 7. 93. E-mail exchange between a Taiwan-based industry consultant and Linden, March 21, 2005. 94. Institute of Electrical and Electronics Engineers (IEEE), “IEEE Quick Facts” (www. ieee.org [January 2005]). 95. The importance of SSCS membership for access to IEEE journals has declined in recent years now that libraries at universities and employers offer electronic access to IEEE publications. The data presented here end in 2001, when electronic access was not yet widely available. 96. Solid-State Circuits was a “council” before 1997, when it became a society. Our colleague David Hodges, a member of the SSCS Administrative Council, suggested a means of estimating the 1991 membership by using a ratio involving two of the old council’s sponsoring societies. Details available on request. 97. The top three countries for growth during the period were Poland, Malaysia, and China, but the first two grew from very small bases (only four and five members, respectively, in 1991).
310
Brookings Trade Forum: 2005
Figure 1. IEEE Solid State Circuits Society Membership (2001) and Growth (1991–2001), in Selected Countries outside the United States Membership
Percent 2001 SSC members 350
2001-on-1991 growth
800
300 250
600
200 400
150 100
200
w an d K an ing d do Ir m Ch ela n H ina d on an g d K on g
a
U
ni
te
Ta i
di In
y m er G
Fr
an
an
ce
el ra Is
ea K
or
n pa Ja
Ca
na
da
50
Source: IEEE data and authors’ calculations.
States had an SSCS membership of 8,747 in 2001, which is estimated to have grown 16 percent over the preceding decade, less than the growth rate for all countries shown except Canada (14 percent). Chip design capabilities in the three Asian countries are following different paths. Taiwan has the most well established chip design sector of the three, having benefited from focused government programs and the return of U.S.-educated and -trained engineers during the late 1980s.98 The Taiwanese chip design sector is mostly locally owned, with a few multinational companies also operating design subsidiaries. Taiwanese companies have particularly embraced the fabless model, with some sixty fabless companies listed on the Taiwan Stock Exchange in December 2004, compared with about seventy on NASDAQ.99 The 2004 output 98. Saxenian (2002). 99. Fabless Semiconductor Association, “Global Fabless Fundings and Financials Report” (www.fsa.org).
Clair Brown and Greg Linden
311
value of all Taiwan’s locally owned design companies (including fabless and design service companies) was reported by the Taiwan Semiconductor Industry Association to be $8.15 billion.100 One advantage for these firms was the availability of an important local market: many Taiwanese companies design, assemble, and procure components for computers, communication equipment, and consumer electronics for world-famous brands, including Hewlett-Packard, Nokia, and Sony. In 1999, 62 percent of Taiwan’s chip design revenue came from local sales.101 Although many of those we interviewed praised Taiwan’s design teams for their execution, which is a vital trait in an industry where time-to-market is often the difference between profit and loss, Taiwanese companies were mostly fast followers. Ironically, they are locked in to some extent by their reliance on business from the local systems firms, who are themselves as much as a generation behind the leading-edge technology.102 From a U.S. perspective, Taiwanese competition for chips using last-generation technology has shortened the market windows during which U.S. chip companies can recoup their investments. Taiwan’s government has instituted several programs to improve the local design sector, including a plan to train several thousand new design engineers in Taiwan’s universities, the creation of an exchange where local chip design houses can license reusable functional blocks, and an incubator where earlystage start-ups can share infrastructure and services.103 Another initiative aims to attract chip design subsidiaries of major semiconductor companies; early takers include Sony and Broadcom (a major U.S. fabless company). In 2000 a government research institute created the SoC Technology Center (STC) to design functional blocks that can be licensed to local companies, a model Taiwan has used successfully in other segments of the electronics industry. The STC has more than 200 engineers, most of whom have a master’s degree or better.104 China appears to be following a similar pattern: government sponsorship, local access to system firms such as Haier, Huawei, and TCL that are increasingly engaged in world markets, and active involvement of expatriates returning from the United States.105 In 2003, China claimed to have more than 400 fabless 100. NT$260.8 billion converted at NT$32:US$1. Datum from “TSIA: Taiwan’s 2005 Semiconductor Production Value to Rise 6.7% on Year,” DigiTimes.com, March 18, 2005. 101. Data from Taiwan’s Industrial Technology Research Institute cited in Chang and Tsai (2002), table 5. 102. Breznitz (2005). 103. Steven Leibson, “Trends in SOC Design Unthaw at SOC 2004,” EDN, December 9, 2004. 104. Interview with executives of Taiwan’s SoC Technology Center, March 7, 2005. “SoC” is a common industry acronym for “system-on-a-chip,” meaning a complex semiconductor integrating multiple functions. 105. Saxenian (2002).
312
Brookings Trade Forum: 2005
design firms with total revenue of $547 million.106 Many of these are small, poorly managed, and rapidly running through their seed money before they can bring a product to market.107 One interviewee, echoed by others, claimed that many, perhaps most, firms outside the top ten—whose total revenue was $328 million in 2003—are engaged in various types of reverse engineering, which is used both legally and illegally.108 Foreign firms are often reluctant to bring lawsuits for fear of displeasing the authorities and because they are unlikely to win in Chinese courts, but at least two U.S. companies are suing Chinese rivals in export markets for intellectual property violations.109 China is not yet an important destination for design offshoring. Of the fifteen top U.S. semiconductor companies, a handful had opened research centers in China (thirteen have done so in India) as of June 2005, but most are targeting the local market for the time being, and, according to press reports, some are engaged in software or system design rather than chip design per se. Concerns over intellectual property protection in China appear to pose a greater barrier to foreign design activity than they do in India.110 Chip design in China is at an early stage. Its relatively young chip design engineers will steadily build their experience. Some companies, particularly those whose founders include expatriates returning with foreign experience, will likely begin to penetrate global markets by the end of the decade. It is too early to predict the eventual relative importance of domestically owned and foreignowned chip design activity, and whether domestic firms will be involved mostly with contract services or with creating and selling their own chips. The Chinese government has taken many steps in support of chip design firms, some of the largest of which are state-owned. Measures include tax reductions, venture investing, incubators in seven major cities, and special government projects.111 A value-added tax preference for domestically designed chips was
106. Chinese government data cited in “TSMC: China IC Design Industry Only a Few Years behind Taiwan,” DigiTimes.com, September 10, 2004. 107. Assessment of Byron Wu, iSuppli analyst, reported in Mark LaPedus, “China’s IC Design Houses Struggling for Survival, Says Isuppli,” Silicon Strategies, May 20, 2004. 108. A European chip executive, interview conducted by Elena Obukhova, Shanghai, December 2003. 109. See Bill Roberts, “An Offshore Test of IP Rights,” Electronic Business, May 1, 2004, p. 19; and “SigmaTel Sues Chinese Chipmaker over IP,” Electronic News, January 10, 2005. 110. “SIA Pushes Steps to Better IP Protection in China,” Electronic News, November 22, 2004. 111. See “Synopsys Teams with China’s Ministry of Science and Technology, SMIC,” Nikkei Electronics Asia, March 21, 2003; Ed Sperling, “An Uneven Playing Field,” Electronic News, July 3, 2003; Mike Clendenin, “China Nurtures Home-Grown Semiconductor Industry,” EBN,
Clair Brown and Greg Linden
313
phased out under U.S. pressure and will reportedly be replaced by a WTOfriendly R&D fund.112 India presents a very different picture, with benign neglect by the government, a lack of manufacturing for chips and systems, and weaker levels of brain circulation with its U.S.-based expatriates.113 Unlike Taiwan and China, India has no high-volume chip manufacturing, although construction has reportedly begun on a 200mm fab near Hyderabad with backing by a nonsemiconductor firm from Korea and the Andhra Pradesh state government.114 Perhaps because of India’s weakness in chip and electronics manufacturing India has no major fabless companies (companies designing chips for sale under their own brand), and its chip designers provide design services or work at the subsidiaries of foreign chip companies, especially U.S. and European firms. Design services revenues for Indian companies in 2001 were $149 million.115 The government is in the early stages of developing policies to support domestic chip design firms.116 It is in the foreign subsidiaries that most of India’s chip design is taking place. The foreign chip companies were attracted by Indian engineers’ use of English and the successful Indian software sector. Many of the early Indian investments by chip companies were focused on writing embedded software, the microcode that becomes part of the chip. Over time, the Indian affiliates have taken on a bigger role, eventually extending to complete chip designs from specification to physical layout. This transition can happen quite quickly. Intel, for example, opened a software center in Bangalore in 1999, then started building a design team for 32-bit microprocessors in 2002.117 A number of U.S. semiconductor companies have software and chip development operations in India, including Texas Instruments (1,000 employees), Freescale (200), Cypress Semiconductor (200), and National Semiconductor (80), as well as a host of fabless companies including Qualcomm and Nvidia.
December 8, 2003; “China Government to Support Solomon Systech, Actions and Silan,” DigiTimes.com. April 14, 2005. 112. Peter Clarke, “China to Form R&D Fund to Replace VAT Rebate, Says Report,” EETimes, April 15, 2005. 113. Saxenian (2002). 114. K. C. Krishnadas, “Work Begins on Hyderabad Fab, Doubt Cast on IBM Role,” EETimes, June 27, 2005. 115. Data from NASSCOM, India’s IT industry trade group, in Aarti Gupta, “Designed to Win,” Business India, September 1–14, 2003. 116. K. C. Krishnadas, “India Expands Tech Incubator Initiative, Seeks New Investment,” EETimes, August 12, 2004. 117. Terho Uimonen, “Taiwan Chip Maker to Build in China,” Wall Street Journal, August 30, 2002, p. B4.
314
Brookings Trade Forum: 2005
The range of activities at these centers is quite broad, and the training curve for domestically educated engineers can be steep. In one instance we studied, a chip design project took twice as long to complete as planned.118 The oldest and largest of these centers is that of Texas Instruments (TI), which opened a software center in 1985. Most other U.S. investments in India have been made since the mid-1990s. The case of TI India shows the potential for the other offshore chip design investments in India to develop over time. Texas Instruments was the first semiconductor company to invest in India when it opened an office in Bangalore to work on its design automation software for internal use.119 In 1988 the company added the design of mixed-signal (analog and digital combined) chips. In 1995, TI added design for DSP devices, the company’s flagship product line. In 1998, TI India announced that it had taken its first DSP core from specification to working silicon over the preceding two years and had integrated a controller with the DSP function for the first time. In 2003, TI India announced that it had created a highly integrated DSL chip that was the first to market to include significant analog elements on the same chip as the DSP and network processor. During the specification phase, a team of twenty engineers went to TI’s Dallas headquarters and worked for three months with TI system engineers and dealt directly with TI customers about their requirements. The 130nm linewidth, 13 million transistor design was completed in India over the next year by a team of seventy, worked the first time, and gave rise to eight patent applications for improvements to DSL technology. By August 2003, TI India had 225 U.S. patents. TI India also develops design library elements, the basic building blocks needed for physical design, for TI’s new processes. Although library elements are low-level intellectual property, they are critical inputs to the design process and used throughout TI’s R&D infrastructure. Moreover, the designers engaged in library construction are gaining valuable experience by designing for leadingedge process technology. Since 1999, TI India has won several awards from EDN Asia, a design industry publication, for its chips. In 2004 a very high performance analog-to-digital converter was touted during an interview by the company’s CEO, who mentioned in passing that it had been designed primarily in Bangalore.120 118. E-mail communication between a former Indian chip designer and Brown, June 22, 2005. 119. The following description is based on a compilation of published accounts and the corporate website. 120. David Lammers, “Texas Instruments Collects on Split Fab Strategy Bet,” EETimes, May 17, 2004.
Clair Brown and Greg Linden
315
Texas Instruments is the pioneer among the large, vertically integrated chip companies in in-house offshoring of chip design to India, but it is far from unique. The trade press regularly publishes announcements by other semiconductor companies of plans to expand their Indian design centers, most of which were created in the 1990s. A fabless company we interviewed started its Indian team in 2004 with logic design and plans to eventually have it develop complete derivative products.121 A quarter of the company’s 500-plus design engineers are now located in India. The Job Picture in the United States The job market for U.S. chip designers is unclear, but the short-term dynamic of expansion overseas with modest growth of domestic design centers is a cause for concern. However, past experience in the semiconductor industry serves as a reminder to be cautious in extrapolating from the present. Since 2000, the many forces affecting the semiconductor industry have included the severe recession during 2001, the recovery that stalled in 2004, the large decline in venture funding for start-ups that is only beginning to pick up in 2005, fewer H1-B (temporary specialist work) visas available, and a drop in foreign student applications to U.S. graduate engineering schools since 9/11. It is difficult to disentangle the effect of the business cycle from any underlying longrun trend in the offshoring of design jobs. This caveat should be borne in mind during the following analysis of the U.S. labor market for electronics engineers (EEs) using Bureau of Labor Statistics (BLS) data. Ideally, we would have data specific to chip design engineers. However, this level of occupation and industry classification is not available. Instead, the data presented here are for EEs in all industries over the period 1999–2004 and in a broadly defined semiconductor industry. The most detailed level of EE occupational detail is “Electronics Engineers, Except Computer” (occupational classification 17-2072). In addition to chip designers, this category includes product-level system designers and engineers who develop nonchip components. Within this category, 12 percent were in the semiconductor industry (NAICS 3344) in 2004. Another 13 percent were in electronicsrelated industries, and 29 percent were in telecommunications. The rest are scattered throughout other industries. According to non-BLS semiconductor industry
121. Interview with a U.S. chip company, December 16, 2004.
316
Brookings Trade Forum: 2005
Table 9. Employment and Earnings for U.S. Electronics Engineers, 1999–2004
1999 2000 2001 2002 Nov. 2003 May 2004
Total EE employment
Mean annual EE earnings (dollars)
Percentage change in EE employment
Percentage change in total nonfarm employment
106,830 123,690 123,210 126,020 131,240 135,560
63,410 66,490 69,710 71,600 74,800 77,450
... 15.78 –0.39 2.28 4.14 3.29
... 1.94 –1.36 –0.36 –0.08 0.56
Source: BLS Occupational Employment Statistics website, www.bls.gov/oes/home.htm, accessed April 15, 2005.
data for 2000–04, 4 to 10 percent of new EE graduates go into the semiconductor industry.122 Table 9 shows a relatively strong national demand for electronics engineers. The rapid buildup during the telecom and Internet bubble is clearly visible in the 16 percent growth for 2000, but after a small correction in 2001, the level continued rising significantly faster than total nonfarm employment. Mean annual earnings continued to rise throughout the period and were 16.5 percent higher in 2004 than in 2000, while the consumer price index for urban wage earners increased 9.2 percent over the same period. The BLS industry-specific data, based on the North American Industry Classification System (NAICS), are also imprecise for our purposes. Employment and wage data are available for “Semiconductor and other electronic component manufacturing” (NAICS four-digit level 3344), which includes relatively lowvalue components such as resistors and connectors. The most relevant subcategory, “Semiconductor and related device manufacturing” (NAICS 334413), accounted for 39 percent of employees (and 45 percent of nonproduction workers) in the 3344 category in 2003, but occupation-specific data are not available at this level of industry detail.123 The data in table 10 from the broadly defined semiconductor industry for the two occupations involved in chip design, electronics engineers and software engineers (applications and systems), show a strong, although more variable, labor market for the period 2002–2004, the only years publicly available. EEs 122. Data are from Ferrell, “SIA Workforce Strategy Overview.” 123. U.S. Census Bureau, “Statistics for Industry Groups and Industries: 2003,” Annual Survey of Manufactures, April 2005.
Clair Brown and Greg Linden
317
Table 10. Employment and Earnings for U.S. Semiconductor Engineers, 2002–04
Engineering specialty
Total employment
Mean annual earnings (dollars)
Percentage change in employment
Percentage change in earnings
2002 Software applications Software systems Electronics
7,460 5,140 13,690
78,710 81,060 72,680
... ... ...
... ... ...
May 2003 Software applications Software systems Electronics
8,500 5,670 13,480
80,970 84,790 75,870
13.9 10.3 1.5
2.9 4.6 4.4
May 2004 Software applications Software systems Electronics
7,880 6,070 16,580
83,060 90,240 78,350
–7.3 7.1 23.0
2.6 6.4 3.3
Source: BLS Occupational Employment Statistics website, www.bls.gov/oes/home.htm, NAICS 334400 occupations.
earn less than the software engineers in the four-digit semiconductor industry, and systems software engineers experienced the fastest earnings growth during the two-year period. Applications software engineers experienced a dip in employment in 2004 after strong employment growth in 2003, and EEs experienced a dip in employment in 2003 followed by very strong employment growth in 2004. This is consistent with the peak in the national unemployment rate for electrical and electronics engineers to 6.2 percent in 2003, as it converged for the first time in thirty years with the general unemployment rate, before falling back in 2004 to a more typical rate of 2.2 percent.124 Based on the available government data, the current wave of expansion in inhouse offshore design centers and the growth in international design outsourcing does not appear to have had a major negative impact on the semiconductor labor market in the United States to date, but results from a regional survey reveal underlying strains. Silicon Valley, considered the cradle and creative font of the semiconductor industry, is experiencing a more difficult job climate. Silicon Valley jobs in the semiconductor industry declined 3.6 percent in 2004 (second quarter) from a year earlier, although semiconductor earnings, which 124. Data were provided by Ron Hira. BLS redefined occupations beginning with the 2000 survey covering 1999, but there is no evidence that the redefinition has contributed to the post-bubble unemployment rise. See also Jean Kumagai, “It’s Cold Out There,” IEEE Spectrum, July 2003.
318
Brookings Trade Forum: 2005
averaged considerably above the national average at $120,000, rose 13.5 percent in 2003. Overall the number of jobs in the Silicon Valley has decreased continuously from 2001 to 2004.125 The situation is also more difficult for older engineers, who face rapid skill obsolescence. We heard in our research interviews that chip companies value new graduates, who are trained with the newest technology and command lower salaries. The 2004 salary survey by the EE Times found almost no difference between engineers’ average salary at age 40 –44 ($104,000) and at 55–59 ($105,000).126 Experienced design engineers are often forced to work on mature technologies, which pays less. For example, the EE Times survey found that the average annual salary for U.S. and European engineers skilled at designing for the latest chip process technology was $107,000, whereas engineers designing for the more mature analog technology averaged $87,000.127 Perhaps unsurprisingly, industry participants themselves are split on the significance of offshoring for the U.S. job market. A 2004 survey of more than 1,453 chip and board design engineers and managers by EE Times shows that about half saw foreign outsourcing as leading to a reduction in headcount. Qualitative opinions in the survey were also divided, with optimists noting that reduced costs made for a stronger company and a more secure job, while the pessimists bemoaned downward pressure on wages and employment plus a possible loss of intellectual property and, in the long run, industry leadership.128 We have observed some of the same dynamics of design job movement over the business cycle as occurred during the offshoring of assembly. A wave of design offshoring took place at the height of the dot.com bubble. When the cascading effect of the subsequent downturn reached the semiconductor industry, chip companies cut staff at home. Now that the recovery requires expansion of design operations, chip companies appear to be expanding design operations abroad faster than at home.129 It is too early to predict where this relative shift in the geographic distribution of employment will find a new equilibrium.
125. JointVenture, “Index of Silicon Valley, 2005” (www.jointventure.org/PDF/ JVIndex2005_FINAL.pdf). This index uses state unemployment insurance data, which are the basis for the census data. 126. Bellinger, “Mean Wages Edge Closer to Six-Figure Mark.” 127. “After 10-Year Surge, Salaries Level Off at $89K,” EETimes, August 28, 2003. 128. Bellinger, “It’s an Outsourced World.” 129. See, for example, Bill Roberts, “The Perfect Storm Brews Offshore: Moving R&D outside the U.S. Creates Opportunity, Outcry—and Risk,” Electronic Business, March 1, 2004, p. 46.
Clair Brown and Greg Linden
319
Synthesis and Conclusion This section revisits the lessons from the earlier periods of offshoring of the semiconductor value chain in light of what is known about the offshoring of design, before turning to a discussion of policy issues. Assembly, Fabrication, and Design Compared The first lesson from the experience of assembly offshoring was that a partial move offshore to reduce costs can keep the industry competitive while allowing it time to adjust to a changing market environment. From the 1960s to the 1980s, the greatest source of anxiety was low-cost competition from Asia and a shift in importance from military to consumer electronics applications. Today the offshoring of design is being driven by the rising fixed costs of chip design at a time when the industry’s key application market is changing from the corporate computing sector to the price-sensitive consumer multimedia market. Offshoring appears to be a vital tool for allowing U.S. firms to bring complex chips to market at competitive prices and to expand the market for lower-cost chips in the developing countries. A second lesson is that offshoring can lead to hollowing out, which was the case for U.S. chip assembly. As the example of Texas Instruments’ twenty-yearold Indian design subsidiary shows, the work being done offshore will increasingly resemble that being done at home. Some start-ups have already adopted a structure featuring minimal high-level design staff in the United States with the bulk of engineering done in low-cost Asian locations. We expect these virtual chip firms to become more widespread. Nevertheless, most of the vital intellectual property will be developed and retained close to headquarters. Furthermore, we expect the large firms that employ large design groups in the United States to continue doing so. Therefore, we do not anticipate a hollowing out of chip design comparable to what occurred with chip assembly. The third lesson from the offshoring of assembly is that in-house offshoring can give rise to international outsourcing as local suppliers spring up alongside U.S. and other subsidiaries. It is not yet clear how far this will go in chip design. Many of the founders of foreign start-ups have educational and work experience in the United States. Foreign start-ups also benefit from policy interventions, as in China; from the strength of related activity, such as the Indian emergence of chip design on the back of the successful software industry; or from some combination of the two, as in Taiwan. It seems probable that local companies will
320
Brookings Trade Forum: 2005
continue to spin off from U.S. subsidiaries to offer design services and reusable functional blocks. For the time being, these lower-level design activities are mostly complements rather than competitors to U.S. firms. Because design is more central to the firm than assembly was, the diffusion of design knowledge through in-house offshoring and international outsourcing might give rise to rivals that could ultimately threaten U.S. industry leadership and U.S. design jobs. The prospects vary greatly by the institutional environment in each location. As discussed above, Taiwan’s fabless sector, which did not arise as an outgrowth of U.S. design offshoring, is nearly a generation behind U.S. rivals in developing innovative products. For now, local firms in India have generally avoided the fabless model, but in China there is a small but increasing number of fabless firms targeting world markets. Although for now Chinese firms lack experienced engineers and managers and are behind Taiwan in their development of innovative products, this will gradually change in the years ahead. It is too early to know where this process will end. One observation made earlier about the offshoring of fabrication was that the offshoring of high-end activities in developed countries for market access was partially offset by cross-investments in the United States by European and Japanese firms. As discussed, this is definitely the case for many offshore design investments, with the addition that such cross-investments may also be made in pursuit of specialized resources. Even ambitious small firms from relatively lowcost countries such as Taiwan are often compelled to invest in a small design center in Silicon Valley to access the expert skills available there. This has helped to blunt the effect of the in-house offshoring of chip design by U.S. firms. Another lesson from the fabrication experience is that the development of the activity in new locations can lead to industry restructuring. In the case of fabrication, the emergence of chip foundries in Asia spurred the growth of the U.S.dominated fabless sector. This was a long process, and it is too early to know whether something similar will happen with design. But the phenomenon of virtual chip firms, with most design and other activities offshore, appears to be spreading. This will remain true to the extent that it confers cost advantages over less global, fabless rivals. Policy Issues Offshoring appears to be a largely positive development at the industry level. The reduced costs and the flexibility provided by offshore design centers have allowed both new entrants and incumbents in the United States to maintain their competitive advantage despite the rising cost of the typical chip design. The
Clair Brown and Greg Linden
321
lower costs have translated into growing consumer markets, both for advanced products in the developed countries and for scaled-down products in developing countries, especially fast-growing Asian markets. U.S. companies have also become savvy about how to develop products for regional markets and often locate design and marketing activities accordingly. In addition, U.S. companies have carefully considered what intellectual property they must protect and keep close to home for strategic reasons, and what activities can be sent offshore, often with new protections in place. The industry’s offshoring has gone well beyond the point where blunt instruments such as trade policy can help engineers without harming companies. Taxes or quotas on traded activities or goods would raise the cost structure of the many companies that have already invested offshore, whether they are designing primarily for the foreign or for the domestic chip market. Policy is thus unlikely to be able to improve the demand side of the labor market, and industry has been active in lobbying for changes on the supply side in the form of education and immigration changes. The winter 2005 newsletter of the Semiconductor Industry Association includes articles such as “Maintaining Leadership as Global Competition Intensifies” by the organization’s president and “America Must Choose to Compete” by the outgoing CEO of Intel. One target of these industry analyses is education.130 Higher education policies, which reflect both university decisions and government funding, determine the number and country of origin of engineering graduates at all levels. The number of foreign nationals in our M.S. and Ph.D. programs in electrical engineering has a direct impact on the supply of engineers both in the United States and in China and India. Foreign graduates of U.S. engineering schools must obtain temporary visas, usually H1-B visas for up to six years, before they can work in the United States after graduation. The complex immigration and education issues are controversial, and a thorough discussion of them is beyond the scope of this paper. Experts cannot even agree if the United States is educating too few engineers (and scientists) or is facing an engineer shortage.131 Government policies regulating immigration, especially the issuance of H1-B (temporary specialist work) and L-1 (intracompany transfer) visas, have an important impact on the number of foreign engineers engaged in semiconductor 130. Semiconductor Industry Association, SIA Quarterly Newsletter, Winter (www.siaonline.org/downloads/05_Winter.pdf). 131. See, for example, National Research Council (2000); National Research Council (2001); Freeman (2003); Butz and others (2004); and Task Force on the Future of American Innovation, “The Knowledge Economy: Is the United States Losing Its Competitive Edge?” (www.futureofinnovation.org [April 20, 2005]).
322
Brookings Trade Forum: 2005
work. Changes in the policy appear to have had an effect on in-house offshoring. When the number of H1-B visas issued was dramatically cut in 2002 and 2003 in response to the recession, many U.S. companies used the opportunity to send foreign nationals with U.S. education and experience back to India and China to help build operations there. Although their salaries are much lower, the purchasing power parity (PPP) comparisons indicate that engineers still enjoy a high standard of living after returning home. For example, the PPP-adjusted salary that compares to a design engineer earning $82,000 in the United States is $77,300 in India and $69,400 in China.132 A study of foreign-born U.S. Ph.D.’s in science and engineering shows the importance of personal values, work-related considerations, and formal and personal ties, as opposed to purely economic motivation, in the decision to return home.133 An area of policy that has received less attention is compensation to engineers who are harmed by offshoring. Thanks to the offshoring of chip design, consumers benefit from lower prices and new products (although much of that benefit is received outside the United States), but some of the short-term cost of the offshoring is borne by engineers in particular companies or in industry sectors whose companies are restructuring globally. Currently, white-collar workers like chip designers do not qualify for trade-adjustment assistance from the government when their jobs are sent abroad. It would make sense to help these workers with retraining and other forms of assistance that will keep these highly skilled individuals productive. Finally, more and better data are needed. As researchers in other industries have noted, more labor market data, for both the United States and our trading partners, are needed in order to properly understand offshoring and its effects. National policies affecting education, labor markets, and innovation will continue to be based upon informed speculation. The semiconductor industry is still in the early stages of a complex and dynamic process, and policy interventions need to be flexible. At this point it is hard to say what the impact of design offshoring will be on the competitive position of the U.S. semiconductor industry, how long it will take for the economy to adjust, and whether the new equilibrium will be acceptable. What happens to the U.S. semiconductor industry and its workers has important consequences for the country, including the jobs created and the technologies developed. 132. This calculation is based on the PPP of .194 for India and .216 for China reported in the testimony for the IEEE by Ron Hira to the Small Business Committee, available at www. cspo.org/products/lectures/061803.pdf. The PPP adjustment is applied to the estimated $15,000 salaries for India and China reported in table 8 of this paper. 133. Gupta (2005).
Comment and Discussion
Jeffrey T. Macher: Clair Brown and Greg Linden examine the impact and implications of offshoring and outsourcing of product design (mainly) within the U.S semiconductor industry. The academic perspective they take is mainly qualitative, relying significantly on interviews with semiconductor firms and extensive literature searches via industry trade journals and company reports. Their main findings are twofold: first, offshoring and outsourcing occurred in semiconductor assembly and fabrication and have subsequently strengthened U.S. competitiveness in semiconductor design; second, offshoring and outsourcing are increasing in semiconductor product design, but predictions of their impact on U.S. competitiveness are currently speculative. Building on previous historical examinations,1 their paper contributes to explaining the continuing evolution and vertical disintegration of the global semiconductor industry. The authors’ main research questions center on offshoring and outsourcing in three distinct semiconductor value chain activities: assembly, fabrication, and product design, with emphasis on the latter activity. They first examine the national and industry effects of offshoring and outsourcing in semiconductor assembly and wafer fabrication, and then the more recent offshoring and outsourcing developments in semiconductor product design. The paper makes several interesting comparisons between and among these value chain activities, but more detailed assessments are possible. The authors’ data and information gathering efforts are noteworthy. Based on extensive reviews of trade journals, company reports, and related academic 1. See, for example, Flamm (1985); and Macher, Mowery, and Hodges (1998).
323
324
Brookings Trade Forum: 2005
research, combined with dozens of in-person interviews with global semiconductor firms, the authors thoroughly describe the broad evolution of the semiconductor industry and the more recent changes in semiconductor product design in particular. However, some of the information collected (especially salary and employmentrelated data) is rough, and the usefulness of any conclusions drawn from that information may be questionable. To be sure, the authors note many times that their “back-of-the envelope” estimates are “rough” or “must be interpreted carefully.” The authors’ findings that the impact of offshoring and outsourcing of product design to the U.S. semiconductor industry will be significant and far-reaching are admittedly speculative, but seem reasonable. In particular, they claim that the direction and magnitude of these changes are likely to be influenced more by supply-oriented policies (affecting education and immigration, for example) than by demand-oriented policies, such as trade.
Areas of Discussion Brown and Linden indicate that the framework of their paper links the theory of competitive advantage to offshore investing and outsourcing.2 They argue in particular that semiconductor firms engage in the offshoring and outsourcing of production activities to achieve competitive advantage in product, price, or market attributes, or some combination of those three objectives. They say that the stated goals of offshoring or outsourcing of production activities are to provide access to location-specific resources and (new) markets and achieve certain cost reduction targets, mainly through low-cost labor. While I agree with these general propositions, their discussion of each production activity—assembly, wafer fabrication, and product design—is uneven, emphasizing semiconductor product design more than assembly and wafer fabrication in the discussion of competitive advantage. I am in broad agreement with the authors’ historical overview of semiconductor assembly. Offshoring in general (that is, irrespective of the semiconductor industry) might have unintended and dynamic consequences, such as the emergence of a technologically capable foreign supply base that could come to dominate an industry segment. The offshoring of assembly did lead to a “hollowing out” of the U.S. assembly sector. And offshoring of assembly does appear to have been a necessary but not sufficient condition for U.S. semiconductor firms in maintaining their competitiveness. Nevertheless, semiconductor manufacturers over roughly this same time period also made significant 2. See Porter (1985).
Clair Brown and Greg Linden
325
improvements in their manufacturing performance (in manufacturing yield, cycle time, and equipment efficiency, for example) and competitiveness with other (especially Japanese) semiconductor producers. Narrowing these performance gaps, in combination with product innovations and strategic repositioning (exiting unprofitable product markets such as DRAMs, for example), also helped to sustain U.S. competitiveness.3 Some of the reasons that foreign entry into assembly and the rise of specialist assemblers have not brought about a U.S. semiconductor firm exodus from this activity, including technological sophistication, supplier hold-up, intellectual property protection, and buffer capacity, are treated in toto more as an addendum than as an important finding. If one considers that this same two-tiered industry structure also exists in wafer fabrication, which I elaborate on below in a slightly broader context, I believe that this discovery is more meaningful. I am also in general agreement with Brown and Linden’s overview of the history of semiconductor wafer fabrication. The globalization of wafer fabrication is a story of offshoring and then outsourcing. Tilton notes that several semiconductor firms set their sights globally from the industry beginnings (including IBM, NEC, SGS, TI, and Motorola) and now possess decades of international manufacturing experience, while others are either entirely domestic or just beginning to venture abroad (such as many other Japanese firms). More recently, the emergence of the fabless-foundry model has led to increased outsourcing. The authors’ examination of wafer fabrication provides three points worthy of additional discussion. First, the emergence of the fabless-foundry business model has had a significant impact on semiconductor industry structure, and has accelerated differences in the geographic location of semiconductor production activities. While the United States continues its (somewhat tenuous) leadership position in semiconductor product design, the Asia-Pacific region has overtaken every other region as both the location of choice and the location of ownership of wafer fabrication. Second, foundries are of strategic importance not only for fabless semiconductor firms, but also for integrated device manufacturers (IDMs). Foundries level the playing field between these semiconductor firm types by reducing the level of investment required to build production facilities, while simultaneously providing buffer capacity for excess demand and hedges against product life or volume uncertainties. Third, there appear to be limits to the eventual size of the fabless-foundry business model. The advantages of integrated product design and manufacture appear to be greatest in product lines at the leading edge of technology and in certain product markets, such as analog 3. See Macher, Mowery, and Hodges (1999).
326
Brookings Trade Forum: 2005
and DRAMs.4 In these markets, the technological requirements for close coordination of design and process innovation mean that intrafirm management of these activities provides advantages in flexibility, responsiveness, and the “debugging” of new manufacturing methods. Brown and Linden’s claim that the eventual size of the foundry sector is limited therefore is entirely correct. The authors provide a very informative historical analysis of semiconductor product design in relation to offshoring and outsourcing, despite their need to simplify the discussion of this complex value chain activity. They discuss the main stages of semiconductor product design, and how and why these stages have changed and continue to change with respect to software, EDA and automation, cost reduction, and national security. They note that the globalization of semiconductor product design was and is driven by both market demand and supply considerations. Many major semiconductor producers establish design centers to improve their relationships with important global customers, and certain geographic regions possess engineers with specialized skill sets. In their comparison of the benefits and costs of offshoring and/or outsourcing semiconductor product design, Brown and Linden examine, in particular, the “hidden costs” associated with outsourcing, including distance (and lack of face-to-face interaction); different incentives, cultures and skill sets; and intellectual property protection in relation to the more obvious direct cost savings. In summary, the authors provide a nice description of how offshoring and outsourcing influence and determine competitive advantage in the global semiconductor industry. Braun and MacDonald, Tilton, and Langlois and Steinmueller provide slightly different perspectives on the evolution of the industry.
Additional Considerations The antecedents that drove the semiconductor industry to greater levels of vertical specialization is an additional topic worthy of discussion. For the first two decades of the industry, large integrated producers designed their own solid-state components, produced the majority of their production equipment requirements, and used internally produced components in the manufacture of electronic computer systems.5 During the late 1950s, “merchant” manufacturers entered the industry and gained market share at the expense of these system firms. Specialized producers of semiconductor manufacturing equipment 4. See Macher and Mowery (2004). 5. Braun and MacDonald (1978).
Clair Brown and Greg Linden
327
began to appear in the industry by the early 1960s. During the late 1980s, fabless firms entered into the design (and marketing) of semiconductor devices, while foundries provided the manufacturing services for these products. Finally, the continued growth in the offshoring and outsourcing of semiconductor product design is further separating the semiconductor value chain into specialized entities. The value chain activities of the global semiconductor industry of today therefore are increasingly controlled by specialist firms, rather than vertically integrated within the boundaries of individual firms. How and why these trends in vertical specialization have co-evolved with industry maturation and decline, as well as the importance and role of industry factors and business strategies that influenced these trends, are interesting questions worth exploring. At least three factors help explain this resultant industry structure.6 First, as knowledge linking the successive stages of the semiconductor value chain became increasingly codified within the industry, the feasibility and cost effectiveness of vertical specialization increased. Second and related, the development of de facto technical standards between successive value chain activities promoted stability and codification in important technical interfaces, which served as the basis for entry by specialized firms. The emergence of CMOS as a process standard facilitated the division of labor between product designers, who were able to operate with relatively stable design rules, and process engineers, who were able to incrementally improve process technologies,7 which aided the establishment of the fabless-foundry model. Significant improvements in information technology and (particularly) CAD software for the layout and simulation of semiconductor designs also facilitated the separation of product design from wafer fabrication,8 as well as the separation of different stages of semiconductor product design. Finally, vertical specialization improves both the strategic positioning and competitiveness of semiconductor firms by facilitating economies of specialization and allows others to exploit scale and scope economies. Fabless firms and integrated device manufacturers who outsourced some value chain activities (mainly assembly and fabrication) can concentrate on what they do best while simultaneously responding to changes in demand. At the same time, specialist manufacturing organizations are able both to reduce costs via scale economies and to operate at full capacity by aggregating the demands of several diverse customers. 6. See Macher and Mowery (2004). 7. Macher, Mowery, and Hodges (1999). 8. Macher, Mowery, and Simcoe (2002).
328
Brookings Trade Forum: 2005
Brown and Linden do not include any significant discussion of the two-tiered industry structures that have resulted in assembly and wafer fabrication and that appear increasingly likely in semiconductor product design. In each of these value chain activities, one segment, mainly populated by integrated firms, competes (and often collaborates) with vertically specialized firms in the other segment. In addition, in both assembly and wafer fabrication these segments are increasingly locating in different geographic regions, though in product design this is less common. In assembly, the entry of specialist assembly firms mainly located in industrializing economies provides an important supply base, but has not brought about exit by integrated firms (located in the United States, Europe, and Japan) from this activity. As the authors indicate, only one-third of current assembly is foreign outsourced. Important reasons why most IDMs maintain some in-house assembly, such as technological factors, supplier overdependence, buffer capacity, and intellectual property protection, suggest that the two-tiered structure for assembly activities will continue for the foreseeable future. In wafer fabrication, fabless firms and foundries compete head-on against IDMs, especially in well-established markets where CMOS is the manufacturing industry standard, while several IDMs remain vertically integrated in areas where coordination between value chain activities is important. Again, this twotiered structure appears likely for the foreseeable future, given the coordination requirements of bleeding-edge technologies and some product markets. As important, the majority of the largest and most successful fabless firms are U.S.based, while there is an increasing trend toward Asian-located and Asian-owned wafer fabrication. Finally, the advent of automation and high-bandwidth telecommunications has improved industry standardization for product design and has facilitated the offshoring and outsourcing of product design. Although the offshoring of product design has seen greater success, mainly for cost-reduction reasons, outsourcing of product design has largely remained a U.S. activity. This outcome is explained in part by specialist design firms that create, license, and trade “design modules.”9 Nevertheless, and as the authors indicate, the diffusion of product design knowledge through either in-house offshoring or outsourcing could lead to significant changes in industry structure that threaten U.S. competitiveness in the future.
9. Linden and Somaya (2003).
Clair Brown and Greg Linden
329
General Discussion: A number of participants thanked both the authors and the discussant for particularly interesting and informative presentations. A variety of issues and questions were raised, many focusing on the role of trade policies in the United States and industrial policies abroad. David Richardson saw the semiconductor industry as providing a cautionary perspective on traditional, at-the-border trade policies. Given the extent of crossborder integration at different stages in this industry, he observed that it was difficult to imagine how a firm could gain unambiguously from a traditional tariff, quota, or investment restriction. This raised what he saw as a relatively unexplored question: what new tools of global competitiveness could be used by Congress, or in a regional trading agreement, to advance a particular country’s interests relative to its rivals’ interests? Chad Bown noted that there have been a variety of recent trade policies affecting semiconductors, including antidumping laws, countervailing duty laws, WTO dispute settlements, and some newer, potentially quite discriminatory policy tools. The latter include the “China safeguard” (for the case of the United States, this is implemented as Section 421 of the U.S. trade law), which is like antidumping and the countervailing duty laws in that it can be applied on a discriminatory basis against a single exporting country (China alone). Given the degree of integration in the industry, the implications of such policies are far from transparent. For instance, they could induce unintended collusive behaviors that affect price. Richard Freeman raised the example of Sematech as a nontraditional policy intervention that encouraged companies to work together. To the extent that this was successful, it would suggest an alternative, more positive policy approach. These commentators all stressed the need for research to help us understand these issues better. Bown also commented that he had thought that the rise of the semiconductor industry in Taiwan and Korea could be directly related to U.S. trade policies during the 1980s. He recalled that some analyses concluded that U.S. antidumping measures against the Japanese, who had very low prices at the time, induced collusion among Japanese firms. These firms increased their prices, creating shortages in the U.S. market, raising profitability, and generating an opportunity for the Taiwanese and Korean industries to enter the market and get off the ground. Although government assistance in these countries also played a role, the U.S. semiconductor policy was arguably very important. Lael Brainard raised the issue of industrial policies. She noted that Claire Brown and Greg Linden’s analysis suggests that these were important in moving capacity to both Taiwan and China, but she wondered whether research had identified more specifically which interventions had been effective and which had not.
330
Brookings Trade Forum: 2005
Rafiq Dossani stressed that government policy had been extremely activist in a number of the Asian developing countries. He did not think the fabs would have been established in the absence of these policies. However, he wondered whether these governments had induced suboptimal location decisions in many instances. He referred to a recent interview he had conducted with executives at Intel in which he asked why they had not set up a fab in China. They replied that the capital costs of setting up a fab are so large relative to the labor costs that it simply does not make sense to set one up in a low-labor-cost country. Interestingly, their decision was not due to intellectual property rights considerations. Catherine Mann stressed that any assessment of these interventions should consider their costs, ideally from a general equilibrium perspective. She believed that some recent research on Malaysia’s subsidies has suggested that the costs outweighed the benefits. Mann also highlighted the points about design center location made in the paper. These centers serve heterogeneous and sophisticated consumers, and to do so they need to be close to them. Thus the existence of a critical mass of such consumers will influence both the global structure of the industry and where innovation and leadership occur. Referring to the discussion of Dan Trefler’s paper, Brainard also asked whether the authors believed that institutional constraints really are binding for growth abroad at the high end of the industry. Jeff Macher’s comments about intellectual property suggested that they were. She found particularly intriguing Macher’s description of Taiwan implementing a rule prohibiting investments in China that involve technology less than two generations behind what is used domestically. Brown responded that she sees the IP constraints as enormous—even in Taiwan. She explained that firms take great care to encrypt their IP and to keep it protected. In her view, IP problems will definitely constrain development of the semiconductor industry, particularly in China, and somewhat in India. Brainard also asked about labor market implications for electrical engineers. Brown and Linden concluded that offshoring was not fundamentally threatening the high-end and very skill-intensive part of chip design, but also noted that there has been a sharp decline in the returns to experience for electrical engineers. Brainard wondered how the authors would square these two findings and whether there are any policy implications. Is the problem that general engineering skills are becoming less useful? Is something going on at the firm-specific level? Or is the decline in demand for experienced engineers simply due to the recent recession? Brown noted that the returns to experience have been falling since the mid1990s, so it is neither a recent phenomenon nor one simply associated with recession. She agreed that it would be useful to examine this further, including the potential role for various types of training or retraining.
Clair Brown and Greg Linden
331
References Borrus, Michael. 1988. Competing for Control: America’s Stake in Microelectronics. Cambridge, Mass.: Ballinger. Braun, Ernest, and Stuart MacDonald. 1978. Revolution in Miniature: The History and Impact of Semiconductor Electronics. New York: Cambridge University Press. Breznitz, Dan. 2005. “Development, Flexibility, and R&D Performance in the Taiwanese IT Industry: Capability Creation and the Effects of State-Industry Co-Evolution.” Industrial and Corporate Change 14 (1): 153–87. Brown, Clair, and Benjamin Campbell. 2001. “Technical Change, Wages, and Employment in Semiconductor Manufacturing.” Industrial and Labor Relations Review 54 (2): 450–65. Butz, William, Terrence Kelly, David Adamson, Gabrielle Bloom, Donna Fossum, and Mihal Gross. 2004. Will the Scientific and Technical Workforce Meet the Requirements of the Federal Government? Los Angeles: RAND Corporation. Chang, Pao-Long, and Chien-Tzu Tsai. 2002. “Finding the Niche Position—Competition Strategy of Taiwan’s IC Design Industry.” Technovation 22 (2): 101–11. Flamm, Kenneth S. 1985. “Internationalization in the Semiconductor Industry.” In The Global Factory: Foreign Assembly in International Trade, edited by Joseph Grunwald and Kenneth S. Flamm, pp. 38–136. Brookings. Freeman, Richard B. 2003. “Trade Wars: The Exaggerated Impact of Trade in Economic Debate.” Working Paper W10000. Cambridge, Mass.: National Bureau of Economic Research. Garner, Alan C. 2004. “Offshoring in the Service Sector: Economic Impact and Policy Issues.” Federal Reserve Bank of Kansas City Economic Review (Third Quarter): 5–37. Groshen, Erica L., Bart Hobijn, and Margaret M. McConnell. 2005. “U.S. Jobs Gained and Lost through Trade: A Net Measure.” Current Issues in Economics and Finance 11 (8): 1–7. Gupta, Deepak. 2005. “Ties That Bind: An Econometric Analysis of the Return Home of Foreign-Born U.S. Ph.D.s.” Working Paper 2005-1. Ettimadai, India: Amrita School of Business. Ham, Rose Marie, Greg Linden, and Melissa M. Appleyard. 1998. “The Evolving Role of Semiconductor Consortia in the U.S. and Japan.” California Management Review 41 (1): 137–63. Hemani, Ahmed. 2004. “Charting the EDA Roadmap.” IEEE Circuits and Devices 20 (6): 5–10. Henderson, Jeffrey W. 1989. The Globalisation of High-Technology Production: Society, Space, and Semiconductors in the Restructuring of the Modern World. New York: Routledge. Henisz, Witold J., and Jeffrey T. Macher. 2004. “Firm- and Country-Level Trade-offs and Contingencies in the Evaluation of Foreign Investment.” Organization Science 15 (5): 537–54.
332
Brookings Trade Forum: 2005
Hira, Ron. 2003. Testimony prepared for the Committee on Small Business, U.S. House of Representatives, June18 (www.cspo.org/products/lectures/061803.pdf). Howell, Thomas R., Brent L.Bartlett, William A.Noellert, and Rachel Howe. 2003. China’s Emerging Semiconductor Market: The Impact of China’s Preferential ValueAdded Tax on Current Investment Trends. San Jose, Calif.: Semiconductor Industry Association. Langlois, Richard N., and Edward W. Steinmueller. 1999. “The Evolution of Competitive Advantage in the Worldwide Semiconductor Industry, 1947–1996.” In The Sources of Industrial Leadership, edited by David C. Mowery and Richard R. Nelson. New York: Cambridge University Press. Leachman, Robert C., and Chien H. Leachman. 2004. “Globalization of Semiconductors: Do Real Men Have Fabs, or Virtual Fabs?” In Locating Global Advantage: Industry Dynamics in the International Economy, edited by Martin Kenney with Richard Florida, pp. 203–31. Stanford University Press. Linden, Greg, and Deepak Somaya. 2003. “System-on-a-Chip Integration in the Semiconductor Industry: Industry Structure and Firm Strategies.” Industrial and Corporate Change 12 (3): 545–76. Linden, Greg, Clair Brown, and Melissa Appleyard. 2004. “The Net World Order’s Influence on Global Leadership in the Semiconductor Industry.” In Locating Global Advantage: Industry Dynamics in the International Economy, edited by Martin Kenney with Richard Florida, pp. 232–57. Stanford University Press. Macher, Jeffrey T., and David C. Mowery. 2004. “Vertical Specialization and Industry Structure in High-Technology Industries.” In Business Strategy over the Industry Lifecycle, edited by Joel A. C. Baum and Anita M. McGahan, pp. 317–56. Advances in Strategic Management, vol. 21. New York: Elsevier Press. Macher, Jeffrey T., David C. Mowery, and David A. Hodges. 1998. “Reversal of Fortune? The Recovery of the U.S. Semi-Conductor Industry.” California Management Review 41 (1): 107–36. ———. 1999. “Semiconductors.” In U.S. Industry in 2000: Studies in Competitive Performance, edited by David C. Mowery, pp. 245–86. Washington: National Academy Press. Macher, Jeffrey T., David C. Mowery, and Timothy S. Simcoe. 2002. “e-Business and the Disintegration of the Semiconductor Industry Value Chain.” Industry and Innovation 9 (3): 155–81. MAEI. 1995. “Malaysian-American Electronics Industry: Annual Survey 1994/1995.” Kuala Lumpur, Malaysia. McKendrick, David G., Richard F. Doner, and Stephan Haggard. 2000. From Silicon Valley to Singapore: Location and Competitive Advantage in the Hard Disk Drive Industry. Stanford University Press. National Research Council. 2000. Forecasting Demand and Supply of Doctoral Scientists and Engineers. Washington: National Academy Press. ———. 2001. Building a Workforce for the Information Economy. Washington: National Academy Press.
Clair Brown and Greg Linden
333
Porter, Michael E. 1985. Competitive Advantage: Creating and Sustaining Superior Performance. New York: Free Press. Saxenian, Annalee. 2002. “Transnational Communities and the Evolution of Global Production Networks: The Cases of Taiwan, China and India.” Industry and Innovation 9 (3): 183–202. Tilton, John E. 1971. International Diffusion of Technology: The Case of Semiconductors. Brookings.
This page intentionally left blank
R O S E M A RY B AT T VIRGINIA DOELLGAST HYUNJI KWON Cornell University
Service Management and Employment Systems in U.S. and Indian Call Centers
T
he explosive growth of call centers in India has gained widespread attention because of its potential impact on employment in the United States and other advanced economies. Media accounts report that Indian operations are more likely to use college-educated workers while paying one-tenth of U.S. wages. Some argue that these advantages may allow Indian centers to outcompete U.S. centers on both cost and quality.1 Nonetheless, complaints about poor quality and security, as well as consumer backlash, have led some firms to pull out of India, while leaders in the offshoring business such as General Electric have sold their Indian operations altogether. High turnover rates have become a particularly serious problem in recent years as an expanding number of employers compete for a small pool of educated employees, a trend that both increases costs and undermines service quality. With heated debate more prevalent than systematic empirical investigation, our understanding of this emerging sector is based largely on anecdotal evidence. National figures on employment, industry trends, and the percentage of centers operated in-house (as opposed to outsourced or offshore) are unreliable.2
We thank the Alfred P. Sloan Foundation, the Russell Sage Foundation, and the Cornell University Center for Advanced Human Resource Studies for generous funding that made this study possible. Thanks also to the Survey Research Institute, ILR School, Cornell University, for administration of the U.S. survey and to Priti and Mudit Nopany for conducting the Indian survey. This research is part of a broader international survey of call center establishments in twenty countries in North America, Europe, and industrializing economies, coordinated by Rosemary Batt, David Holman (U. Sheffield, UK), and Ursula Holtgrewe (U. Duisburg, Germany). 1. See Dossani and Kenney (2004). 2. Data on numbers of call centers and employment come largely from interested parties, such as India’s National Association of Software and Service Companies (NASSCOM), and industry
335
336
Brookings Trade Forum: 2005
Our own national survey of U.S. call centers suggests that after two decades of rapid growth, the outsourced sector represents less than 15 percent of the market; and Indian offshore centers cover a tiny fraction of the U.S. market. In addition, there has been little or no research on management and employment practices in this sector, either in the United States or in India. In this paper, therefore, we consider two questions. First, how similar or different are call center management strategies and employment systems in each country? Here our goal is to map the management practices adopted by three types of operations: in-house centers in the United States, outsourced centers in the United States, and offshore centers that are owned and operated by subcontractors in India and serve the U.S. market. Are there systematic differences in these practices, or is there a call center “production model” that has diffused across very different institutional and organizational contexts? Second, what are the implications of different management practices for outcomes such as turnover? In other words, which practices explain the high levels of turnover in the industry? To answer these questions, we draw on an original establishment-level survey of 330 call centers in the United States and India. We focus on customer contact rather than back-office operations such as check processing or online order fulfillment. For each center, the survey provides information on the customer base, market and ownership conditions, organizational characteristics, work functions, workforce skills and training, call center technology, work organization, compensation, and outcomes such as absenteeism and turnover. In the next section, we discuss prior research that informs our study. We then present the study methods and analytic strategy and our findings. Finally we outline the study’s limitations and implications for policy.
consultants such as Datamonitor in the United States. NASSCOM put the number of call center positions in India at 158,000 in 2004. For the United States in 2001, Datamonitor estimated a total call center workforce of 2.5 million, with 88.7 percent located in in-house centers and 11.3 percent in outsourced centers. It projected that by 2005 call center employment would grow by 14 percent, reaching a total of 2.86 million, with 13.4 percent located in outsourced centers (Datamonitor 2001). That estimate is close to the 14.6 percent of U.S. centers outsourced that we found in our 2004 national survey. Datamonitor bases its estimates on market research and the sale of call center work stations and other technology. The numbers of work stations may underestimate employment because they may be used for two or three shifts of workers. More recently, Datamonitor (2004) estimated that the U.S. call center employment would fall to 2.7 million positions in 47,500 call centers by 2008. Our calculations, based on Bureau of Labor Statistics data, suggest a U.S. call center workforce in 2004 of 3.97 million, or an upper limit of 3 percent of the workforce. These calculations are limited by the available data. See appendix for a technical note on these calculations.
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
337
Prior Research The first question we address in this study concerns the extent to which call center management practices vary across markets and institutional settings. Call centers represent a new industrial model driven by advances in information technologies that are now ubiquitous. These technologies facilitate the automation of services through interactive voice recognition units, standardize customer transactions through skill-based routing systems, create machine-paced operations through automated call distribution systems, and routinize work through widespread use of scripting and electronic monitoring. However, research shows that service management strategies and employment systems vary substantially across centers that serve different industry and customer segments, and that perform different work functions—from professional approaches to service to highly transactional or cost-driven ones.3 In this line of research, work and employment systems typically are defined to include three dimensions: (a) the level of education and training required; (b) the level of discretion and collaborative problem-solving embedded in the design of work; and (c) the level and type of compensation system designed to motivate effort.4 The professional service model includes a set of employment practices based on high skills and training, employee discretion and collaborative problem-solving, and high relative pay.5 This approach to service management is typically found in business-to-business centers and information technology (IT) help desks or technical service centers. By contrast, centers that focus on simple transactions, such as telemarketing, reservations, or credit card handling, require relatively low skills, and jobs are likely to be highly routinized with low pay. Quality control is ensured through extensive use of electronic monitoring systems.6 A more complex question is how to explain the variation in customer contact centers that fall between these two extremes: centers that target the mass market or a mixture of markets and that provide service and sales for products that entail some degree of complexity along with opportunities to bundle services and customize offerings. These represent the overwhelming majority of contact centers,
3. Frenkel and others (1998); Batt (2000); Shire, Holtgrewe, and Kerst (2002). 4. Appelbaum and others (2000); Batt (2002). 5. Heskett, Sasser, and Schlesinger (1997); Batt (2002). 6. Heskett, Sasser, and Schlesinger (1997).
338
Brookings Trade Forum: 2005
serving customers in such sectors as financial services, insurance, telecommunications, and a variety of manufacturing industries. Here, management strategies vary considerably in how much weight they give to competing on quality and mass customization7 versus focusing primarily on cost. In-house versus Outsourced Strategies How does the variation in call center management strategies and employment systems align with their ownership status—that is, with whether call centers are in-house, outsourced, or offshore operations? There are many reasons to believe that outsourced and offshore centers will adopt management strategies that focus more on controlling costs than on investing in employees. First, outsourcing allows firms to avoid paying the high wages associated with internal equity norms and internal labor markets or union contracts.8 Several studies have found that subcontractors hire workers at lower pay and benefits to do the same work.9 Erickcek, Houseman, and Kalleberg (2003) found that this is particularly true for low-skilled work, where subcontracting led to the loss of union representation as well as lower pay and benefits. Second, the literature on transaction cost economics suggests that outsourced centers will focus on cost reduction because, as work is turned over to a third party, the client firm must absorb the costs of monitoring and contract enforcement.10 Thus, client firms are likely to exert pressure on subcontractors to keep costs low in order to justify the additional transaction costs of managing the vendor relationship. In addition, client firms worry about the operational risks associated with third-party subcontracting and as a result are likely to outsource those processes that are easily standardized or codified and monitored through objective performance metrics. As research by Ravi Aron and Ying Liu (this volume) shows, the more work processes are codified and the higher the number of performance metrics agreed upon by the buyer and seller, the lower the operational risk. Other research also demonstrates that subcontractors drive efficiency through greater work intensity and capital utilization than in-house operations.11 Grugulis, Vincent, and Hebson (2003) examined outsourcing in three functions requiring radically different levels of skill and complexity and found that in each 7. Pine (1993). 8. On internal labor markets see Abraham (1990); on union contracts, see Pfeffer and Baron (1988). 9. See, for example, Davis-Blake and Uzzi (1993). 10. Williamson (1985). 11. Marsden (1999).
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
339
case the process of subcontracting led to higher levels of employee monitoring, adherence to specific performance metrics, and lower levels of employee discretion. For consumers, high levels of process standardization also reduce service quality by limiting options for customization and relying on menus and self-servicing. In the call center industry these issues are likely to be particularly salient because arm’s-length contracting and attention to the bottom line are widespread, and contract enforcement typically is ensured through ongoing monitoring and adherence to performance metrics.12 Performance management technologies such as electronic monitoring systems provide real-time measures of talk times, adherence to schedules and scripted texts, and sales productivity, allowing client companies to regularly monitor the employees of subcontractors. Thus, subcontractors are under intense pressure to contain costs and meet these efficiency goals. The work of Levy and Murnane (2004) on computers, skills, and the organization of work provides additional insights into the process of subcontracting. They have argued persuasively that computers are best able to automate jobs that require rules-based logic, such as data management and order processing—precisely the kinds of jobs frequently found in call centers. Automation does not eliminate all jobs, but creates standardized work processes that reduce operational risk and allow electronic monitoring of a wide range of performance metrics. Once these processes are computerized and standardized, they are more easily outsourced to third-party vendors. However, more complex processes with higher levels of uncertainty are more likely to be retained in-house, where companies have direct control over operations that require more tacit knowledge and entail more nuanced interactions with customers. The strategic management literature on core competencies provides another perspective on how and why outsourced work systems are likely to be more cost-focused and standardized than those managed in-house.13 Core competencies are defined as those that contribute value to customer benefits and end products, that provide access to a wide variety of markets, and that are difficult for competitors to imitate.14 In theory, firms should retain functions that they consider to be their core competency while outsourcing those functions that are noncore. When applied to the choice of employment systems, the theory suggests that firms should retain human capital that creates value for the firm, is rare 12. Kinnie and Parsons (2004). 13. See Prahalad and Hamel (1990); Quinn (1992). 14. Prahalad and Hamel (1990).
340
Brookings Trade Forum: 2005
or unique, and is difficult to imitate.15 For example, firms are likely to choose internal employment systems for operations that involve firm-specific knowledge and skills, team-based systems, or work processes that involve “social complexity,” “causal ambiguity,” or “idiosyncratic learning.”16 They are likely to externalize or subcontract work that is more generic, involves lower-order skills, or is transactional in nature. Much call center work appears to fall into this latter category, and thus would be viewed as a prime candidate for outsourcing. According to this argument, whether call center work is outsourced depends on whether customer relationship management is considered central to a firm’s competitiveness. If the products and services offered by a company are relatively complex, involving firm-specific knowledge of products, processes, or customers, then firms are likely to retain their customer service and sales functions in-house. Similarly, if companies seek to compete on quality service or customer loyalty, they are also likely to keep call center work in-house because they do not want to lose control of their customer base or have their customers treated generically—in the same fashion as the customers of their competitors, who may be using the same call center subcontractor. For high-value-added customers, such as business customers, firms are particularly likely to use a strategy of service quality, customization, and loyalty and therefore retain business-to-business channels in-house.17 For mass-market service channels, the costs and benefits of keeping operations in-house are more ambiguous from a strategy perspective; and there appears to be considerable variation in what companies actually do. Although the number of call center subcontractors grew dramatically in the 1990s in the United States, at least 85 percent of contact centers in this country continue to be in-house operations.18 This would suggest that a large majority of firms view their customer service and sales operations as central to their competitiveness—or at least have not yet become convinced that they should outsource them. The implications of these arguments for the design of work and employment systems are straightforward. Companies are more likely to retain in-house services that are complex, that involve customer transactions that are nuanced or uncertain, and that provide services to highly valued customers. In order to meet the demands of these types of products and customers, they are more likely to use a strategy of service quality and customization, and therefore to adopt a
15. Williamson (1981); Barney (1991). 16. Lepak and Snell (1999, p. 35). 17. Batt (2002). 18. See Datamonitor (2001, 2003); Batt, Doellgast, and Kwon (2004).
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
341
more professional approach to service. Centers that are operated by subcontractors, either in the United States or offshore, by contrast, are more likely to compete on costs through lower wages and benefits, more standardized work processes, and higher levels of performance monitoring. Outsourced versus Offshore Strategies The academic literature provides much less guidance for predicting the differences between U.S. outsourced and offshore centers operated by Indian subcontractors. On the one hand, arguments regarding the likelihood of a more costbased strategy in outsourced operations may be equally or more relevant to offshore subcontractors. U.S. companies have sent work overseas to take advantage of lower wages, but at the same time they are concerned about the level of service quality provided. They also worry about consumer backlash and the security and privacy of financial databases. A recent survey of U.S. executives reported that the top driver for moving operations offshore was cost savings, while the top reasons for staying onshore were security and service quality.19 For these reasons, U.S. companies may impose tighter constraints on managerial discretion in Indian centers and higher levels of performance monitoring and adherence to call center metrics. If so, then we would expect the work and employment systems in Indian call centers to be more tightly constrained and standardized than those found among U.S. subcontractors. But unique conditions in the Indian labor market suggest that both the reasons for moving work to this segment and the incentives for investing in employees may differ from those in the U.S. outsourced sector. First, the offshore workforce tends to be drawn from a relatively small pool of college-educated, middle-class Indians. We might expect these employees to be more self-motivated, allowing managers to rely on more professional, or at least quasi-professional, employment practices to motivate their workforce. Moreover, given the large cost advantages that Indian centers enjoy, there is opportunity to relax adherence to performance metrics such as talk time so that employees can use their skills to respond more effectively to customer requests. In addition, the growing competition for these employees has put pressure on employers to invest in benefits intended to promote commitment and reduce turnover. Many call centers serving the international market occupy sprawling complexes outfitted with gyms and canteens. They often provide employees with free lunches and door-to-door taxi services and seek to create 19. See Ventoro (2005).
342
Brookings Trade Forum: 2005
a “fun” environment with games and prizes. The additional investment in “accent neutralization” training required by many companies, which averages one to two weeks, makes it particularly costly to lose employees. Moreover, the use of fixed employment contracts in India also means that there is a long wait for new employees, which increases the expense of recruitment. One manager of a multinational third-party center explained: You have to think about hiring way, way ahead. . . . Let’s say I was trying to hire someone from another company in India; she has to give thirty days’ notice, so I have a delay for the thirty days. And once I get her, she has to do the normal products training, but she also goes through two and a half to three weeks of accent neutralization training. So there is a long, long wait for employees offshore. It’s a month longer than in the U.S., easy. (Interview, March 2005) In sum, the unique labor market conditions and cost advantages of Indian offshore centers suggest that they will adopt a less cost-driven approach to work and employment systems than subcontractors located in the United States. Management Practices and Turnover The second question we ask in this paper is how the different management strategies adopted across segments of the market translate into organizational outcomes. Empirical research on the performance effects of alternative approaches to service management has expanded in recent years. There is growing evidence that a more professional, or at least quasi-professional, approach is associated with higher employee satisfaction and customer satisfaction, higher sales productivity, lower turnover and higher sales growth, and higher service quality and higher net revenues.20 Low-cost systems, by contrast, typically are associated with high levels of employee dissatisfaction, absenteeism, and turnover; and these in turn often produce added costs, reduce options for customization, and lead to lower service quality. For example, several studies of call center workers have found that routinized work design and high levels of electronic monitoring lead to stress, anxiety, depression, emotional exhaustion, and burnout.21 Deery, Iverson, and Walsh
20. On those issues, see, respectively, Loveman (1998); Batt (1999); Batt (2002); and Batt and Moynihan (2004). 21. Carayon (1993); Singh (2000); Deery, Iverson, and Walsh (2002); Holman, Chissick, and Totterdell (2002); Holman (2004).
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
343
found that customer interactions, scripts, routinization, workloads, and managerial emphasis on quantity predicted emotional exhaustion, which in turn predicted absenteeism. Singh demonstrated that as worker burnout with customers increased, call center workers were able to maintain their productivity levels, but their self-reported quality was lower. In this paper, we focus on turnover because it is extremely high in the industry and viewed as a major problem by employers. Industry analysts estimate that it averages between 30 and 70 percent in the United States, but in our interviews some managers reported rates of 100 to 150 percent annually. In India, news reports suggest that turnover rates are often 50 percent or higher. Voluntary turnover, or the employee quit rate, is of particular interest to organizational researchers because it represents a large cost to employers. When employees leave, their experience and the firm’s investments in training are lost. Moreover, as noted earlier, the factors that influence turnover also influence other important outcomes, including employee motivation, service quality, and labor costs. Empirical studies of voluntary turnover have found that it is significantly related to human resource practices,22 particularly with respect to work design and compensation. Shaw, Delery, Jenkins, and Gupta (1998) found that quit rates were lower when monitoring and work intensity were lower and pay and benefit levels were higher. Similarly, Batt, Colvin, and Keefe (2002) found that greater discretion and collaboration at work coupled with high relative pay predicted lower quit rates while high levels of electronic monitoring and use of commission-based pay led to higher quit rates. Expected Findings We have argued above that ownership status is likely to be associated with particular approaches to work and employment practices. Based on the theoretical and empirical literature, we expect that in-house, outsourced, and offshore establishments will differ systematically in their service management and employment systems. In comparison with outsourced or offshore centers, inhouse establishments are more likely to adopt employment practices that involve a higher educated and better trained workforce, that provide employees with more discretion and problem-solving capability, and that offer higher relative pay. We also expect differences between outsourced and offshore centers, with the latter more likely to adopt a professional approach to employment management than the former. These differences in choice of employment system, in 22. Arthur (1994); Huselid (1995).
344
Brookings Trade Forum: 2005
turn, should explain variation in turnover rates, with the more professional approach associated with significantly lower turnover. In other words, work and employment practices should partially explain the relationship between ownership status and turnover. To examine these arguments, we developed a model of turnover that includes controls for market and organizational characteristics, while examining the independent variables of ownership status and employment system characteristics, as follows: Turnover = f(market and organizational characteristics, ownership status, education and training, work organization, compensation strategy)
Methods Sample The sample for this study is based on two identical establishment-level surveys conducted in the United States and India between mid-2003 and mid-2004. The U.S. survey was administered to a stratified random sample of 472 call centers drawn from the subscriber lists of Call Center Magazine (60 percent of the sample) and the Dun and Bradstreet listing of establishments in the telecommunications industry (40 percent of the sample). Using the two lists was necessary to identify call centers in different industries. A survey team conducted the survey by telephone with a forty-minute average interview, yielding a 65.4 percent response rate. The sample was reduced to 464 after eliminating outliers and observations that were missing substantial data. The Indian survey was administered to a nonrandom sample of sixty Indian call centers compiled from Internet sites and the membership list of the National Association of Software and Service Companies (NASSCOM) in India. The research team focused on six cities with large call center concentrations (Chennai, Kolkata, Bangalore, Mumbai, Hyderabad, and Delhi). In each city, the research team had one week to contact the call centers on the list, make appointments, and conduct the survey, which averaged ninety-five minutes in length. The team did not target any particular type of center, but rather conducted surveys on a first come, first-served basis as appointments were made. All survey respondents were asked to answer questions pertaining to the “core” workforce in their establishment: the largest group of customer contact employees who carry out the primary work activity at that location. Owing to variations in the sample, we use a portion of the full dataset in this analysis.
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
345
First, we restrict our sample to three market segments: large business, mass market, and all markets. We dropped 106 small business centers and ten operator services centers from the U.S. sample, since these segments were not present in the Indian sample. Second, we excluded sixteen call centers that serve only the Indian domestic market (located primarily in Kolkata) because only the international centers serve the U.S. market. These exclusions reduced the sample size to 392. Because of randomly missing observations in the dataset, our regression analyses are based on a sample of 310 call centers (237 U.S. in-house, 42 U.S. outsourced, and 31 Indian offshore). In both the United States and India, we conducted extensive site visits in different industry segments to aid with the design of the survey and the interpretation of results. In the United States, we visited twelve in-house call centers and six outsourced call centers, where we interviewed managers, supervisors, and employees on various aspects of their human resource policies and work design strategies. In India, each survey was administered onsite, allowing the researcher to cross-check responses and providing an additional test of the reliability of survey responses. Measures The independent variables of interest include the ownership status of the center (in-house, outsourced, or offshore), and the work and employment system, as defined along three dimensions: human capital (employee education and training), work design (opportunities for discretion and problem solving), and rewards (compensation practices). To determine whether an establishment was in-house or outsourced in the U.S. sample, respondents were asked how they would best describe the call center: as an in-house center providing services to their company or as a subcontractor providing services to other companies. The offshore segment includes Indian call centers that serve an international market. Almost all of the Indian centers were owned and operated by Indian subcontractors, with only a handful owned by U.S. subsidiaries or U.S. subcontractors. To measure human capital, we control for the sex composition of the workforce and use two measures of education and training: the years of formal education of the typical worker in the call center and employer investment in initial training (an additive index of the number of weeks of initial training an employee receives and the number of weeks to become qualified). For work design, three measures capture the extent to which employees have opportunities for discretion and problem solving. First, discretion over customer interactions is measured by the variable script use, based on a 1 to 5 Likert response to
346
Brookings Trade Forum: 2005
the question, “To what extent are core employees required to use scripts when talking to a customer?” where 1 is “not at all” and 5 is “a great deal.” Second, we used three measures to construct a work discretion index, again based on Likerttype questions. Respondents were asked to rate the extent to which core employees had discretion over their daily work tasks; tools, methods, or procedures; and pace of work. The three measures were highly correlated (p < .001) and were combined into a mean index. The third measure of discretion, percent in offline teams, is the percentage of employees who participate with supervisors in problem-solving groups or teams. The final group of variables measure compensation practices, including total compensation and percent commission pay. We were unable to use average annual salary in the analysis owing to the large differences in pay between the United States and India. While there are national statistics in the United States on average compensation for customer service and sales employees, it is difficult to find accurate information on the typical pay of a call center employee in India. We therefore constructed a pay ratio measure based on the ratio of a call center’s average gross annual pay to the median pay in each full country sample ($29,000 in the United States and $2,444 in India). Informal documentation from industry publications gave similar estimates for average pay levels in the Indian market. The U.S. median pay in our sample was also similar to estimates from the Bureau of Labor Statistics for the median pay of customer service representatives ($28,720). Percent commission pay is measured as the percentage of total annual pay that is based on individual commission. DEPENDENT VARIABLE. The dependent variable of interest is the average annual quit rate, as reported by managers for the previous calendar year. A square root transformation was used to correct for the non-normal distribution of the variable. CONTROL VARIABLES. We included additional controls for common turnover determinants. The primary customer segment served by employees has been found in several previous studies of front-line service workplaces to influence both management practices and turnover rates.23 Call centers serving highervalue-added segments, such as large business customers, can be expected to invest more both in the skills of the workforce and in employee retention, as well as to be more selective in hiring, which reduces quit rates. Call centers serving multiple market segments typically have a broader skill base and more diverse job requirements. We thus control for whether the establishment serves primarily large business, mass market, or multiple market segments. We also control 23. Batt (2000).
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
347
for union presence, which has been found in past studies to be negatively correlated with quit rates.24 Employees in unionized establishments are able to exercise “voice” versus “exit,” which leads to improved pay and working conditions and reduces turnover.25 In earlier analyses we tested the effects of several other control variables that have been used in past studies of turnover, including systematic selection procedures for hiring new employees, the ratio of applicants hired, whether the call center was part of a larger organization, and the age of the call center. We also analyzed variation in outcomes when controls for industry and type of call center work were added, including a control for whether the call center predominantly handled sales or customer service. None of these had a substantial effect on the coefficients of the independent variables of interest and either reduced or had a negligible effect on the overall Chi-square. Several of these additional controls were also highly correlated with other variables included in the model. For example, both the outsourced and offshore centers have a significantly lower average age than in-house call centers. Thus, in the final model we included a more parsimonious list of control variables that captured key measures of markets and organizational characteristics.
Results Comparison of Mean Characteristics Table 1 presents a comparison of organizational characteristics, workforce characteristics, employment system variables, and organizational outcomes for the in-house, outsourced, and offshore centers. We use a broader range of variables here than were included in our analysis of turnover antecedents to provide a more comprehensive picture of how organizational characteristics and management practices differ across the segments. In addition, in order to make comparisons more precise, we restricted the mean comparison in table 1 to nonunion call centers serving mass-market or multiple customer segments. We tested the significance of mean differences using one-way analysis of variance. In general, there are significant differences in most dimensions of organizational characteristics and work and employment systems across the three types of centers. The patterns are consistent with our expectations, but
24. Shaw and others (1998); Batt, Colvin, and Keefe (2002). 25. Freeman and Medoff (1984).
348
Brookings Trade Forum: 2005
Table 1. Mean Comparison: In-house, Outsourced, and Offshore Call Centers Percent unless otherwise indicated ANOVA p < .05
In-house
Outsourced
Offshore
Organizational characteristics Establishment age in years Part of a larger organization Sales-oriented call centers
15.7 79.6 5.0
9.4 75.9 13.8
3.4 78.9 29.4
Workforce characteristics Female Tenure of less than one year Part-time
68.9 28.1 17.6
71.5 36.8 35.6
43.1 61.8 1.0
Training and qualification Average years of education Typical education (high school) Days of initial training Days to become qualified Days of ongoing training per year
13.3 38.3 19.7 66.8 9.6
12.6 69.0 11.5 44.2 10.4
14.0 36.4 23.6 53.3 11.2
a,b,c
a,b
a,b,c
b
b,c a,b,c a,b,c
a,c a,c
Employee discretion Reliance on scripted textsd Discretion over workd Discretion over handling customer requestsd Participation in offline teams
9.9 9.9
48.3 3.4
32.4 5.9
39.2 36.2
17.2 22.2
2.9 6.9
a,b,c
Performance monitoring Work time electronically monitored Frequency of supervisor monitoringe Frequency of feedback and coachinge
49.5 49.7 46.0
67.7 67.9 55.2
91.7 82.4 94.1
a,b,c
27,713 8.4
23,881 4.1
2,635 18.5
a,b,c
Turnover and absenteeism Quits Total turnover (quits + dismissals) Absenteeism
15.8 24.6 5.5
25.6 41.2 8.9
24.5 29.6 5.3
a,b
Sample size
181
29
34
Compensation Average annual pay ($)f Pay based on commission
a. In-house and outsourced are significantly different. b. In-house and offshore are significantly different. c. Outsourced and offshore are significantly different. d. Percentage answering “a lot” or “a great deal” (4 or 5 on a 5-point scale). e. Percentage with weekly to daily performance monitoring f. Gross annual earnings
b,c
b b,c
b,c
a,c a,c
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
349
there are important exceptions and contradictory patterns as well. Overall, inhouse centers tend to adopt a more quasi-professional approach to employment than either outsourced or offshore centers. They offer jobs with substantially more opportunities for discretion and problem solving, make significantly less use of electronic monitoring and performance management systems, and offer higher pay. Associated with these patterns are significantly higher rates of organizational tenure and lower turnover rates than those found in either outsourced or offshore centers. For example, while 28 percent of the workforce in U.S. inhouse centers has less than one year of tenure, the comparable rate in outsourced centers is 37 percent, and in offshore centers 62 percent. The annual employee quit rate alone is reported at 16 percent in-house centers, 26 percent in outsourced centers, and 25 percent in offshore sites—that is, it is over 55 percent higher than in the in-house centers. The exception to this pattern is that offshore centers rely on workers with somewhat more formal education than those in in-house locations. The typical worker in an Indian center has 14 years of education (on average two years of college) compared to 13.3 years among U.S. in-house establishments. Yet these differences are not as great as often portrayed in the media. Close to 40 percent of managers in both types of centers (38 percent in-house and 36 percent offshore) reported that the typical worker in their establishment has a high school education. Similar patterns hold for initial training, with offshore call centers providing 4.7 weeks on average, and in-house centers 3.9 weeks. However, given that much of the initial training in Indian centers is focused on accent neutralization, it appears that Indian centers do not provide more initial training for other aspects of the job. The comparison between U.S. outsourced and Indian offshore centers yields results that do not match our expectations. On the one hand, the formal education levels of Indian centers are substantially higher than those found among U.S. subcontractors, where the typical worker has an average education of 12.6 years and almost 70 percent of managers report that the typical worker has a high school diploma only. Initial training in U.S. outsourced centers is less than half that found in Indian centers. However, the amount of on-the-job training to become qualified and the annual rates of ongoing training are not substantially different. On the other hand, despite relying on a more educated and full-time workforce, the Indian centers have work systems that are more tightly constrained and standardized than those found among U.S. subcontractors, contrary to our expectations. With the exception of reliance on scripts, which is higher in the U.S. outsourced centers, Indian managers report substantially lower levels of
350
Brookings Trade Forum: 2005
discretion in handling customer requests and use of problem-solving groups. For example, only 3 percent of offshore call centers report giving employees “a lot” or “a great deal” of discretion in handling customer requests, compared to 17 percent of outsourced centers and 39 percent of in-house centers. While inhouse centers have an average of 36 percent of employees working in teams, 22 percent of employees in outsourced centers and only 7 percent of those in offshore centers do so. Measures of performance monitoring illustrate a similar pattern. Most call centers adopt a mix of practices to track employee performance on adherence to talk time, whether they follow the scripts provided, and their effectiveness in both providing friendly service and resolving customer requests. In a sales environment, monitoring is also used both to control potential employee fraud and to provide coaching on selling techniques. Both electronic monitoring and supervisor monitoring and feedback are employed for this purpose, and the intensity of these practices varies substantially among the different sites. While about 50 percent of work time in in-house centers is electronically monitored, this average jumps to 68 percent in outsourced centers and 92 percent in offshore centers. Similarly, supervisors provide feedback and coaching on a weekly or daily basis in 94 percent of the offshore centers, but in only 46 and 55 percent of the U.S. in-house and outsourced centers. With respect to compensation, the average median annual pay reported by managers is $27,713 among in-house centers, $23,881 in outsourced centers, and $2,635 in offshore centers. Thus, in-house centers pay about 14 percent more than outsourced centers and 90 percent more than the offshore segment. The use of commission pay is surprisingly low across the in-house and outsourced segments, at 8 and 4 percent, but significantly higher in offshore centers (19 percent). This probably reflects the higher percentage of sales-oriented call centers in the offshore sample (29 percent) than in the in-house (5 percent) and outsourced (14 percent) sites. Finally, we compare turnover and absenteeism, both important organizational outcomes. High investments in training at many workplaces mean that turnover is costly, and the often tight scheduling practices based on predicted fluctuations in call volume mean that excessive absenteeism has an immediate negative effect on customer satisfaction and sales. As noted earlier, quit rates as well as total turnover are the lowest among in-house centers and higher in outsourced and offshore centers. Absenteeism, by contrast, is highest in the U.S. outsourced segment (9 percent) and lower in both in-house and offshore centers (6 percent and 5 percent respectively). These measures capture the motivation of the workforce to show up and meet performance expectations and are largely in line with our
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
351
other results that indicate that outsourced centers tend to adopt a low-commitment employment system that combines low pay with intensive monitoring and low discretion. Owing to variation in the industries represented in each sample, we checked to see whether these patterns held when the sample was further broken down. For example, we compared centers serving high-end customers as well as those in telecommunications and financial services and found similar patterns. That is, no particular sectors accounted for the variation found across in-house, outsourced, and offshore sites. Multivariate Analyses Table 2 provides the means, standard deviations, and pairwise correlations of the variables included in the final model. For our analyses of turnover, we estimate left-censored Tobit models because the dependent variable is truncated at zero.26 PREDICTORS OF TURNOVER. Table 3 reports estimates of models for quit rates at the establishments. The first equation, model 1, includes the market segment and organizational characteristics. The second equation adds controls for employee human capital, while the third and fourth add measures of work organization and compensation practices. In the first model, after controlling for market segment, outsourced and offshore centers have significantly higher quit rates (compared to the omitted variable, in-house centers), while unionized centers are associated with significantly lower quits. In model 2, both the length of initial training investment and years of education are significantly associated with lower quit rates. The percentage of the workforce that is female is positively associated with higher quits, but this relationship becomes insignificant in the full model. Offshore ownership status continues to be positive and significant at the p < .001 level, while the significance of outsourced status decreases but is still marginally significant. With the introduction of work design variables in model 3, neither outsourced nor offshore status remains significant, and human capital variables decline in significance. Work discretion and the use of problem-solving groups are significantly negatively associated with quit rates (p < .001), while script use is positively associated (p < .10). In the full model (4), union presence, training investments, work
26. Maddala (1992).
352
Brookings Trade Forum: 2005
Table 2. Means, Standard Deviation, and Pairwise Correlations
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Variable
Mean
SD
1
Square root of annual quit rate Large business segment Multiple market segments Mass-market segment Union presence U.S. in-house U.S. outsourced Indian offshore Percent female Years education Initial training investment Script use Work discretion index Percent in offline teams Pay ratio Percent of pay based on commission
3.23 2.22 0.34 0.47 –0.16 0.19 0.40 0.02 0.47 0.50 0.13 0.08 0.27 –0.22 0.76 0.42 –0.27 0.14 0.34 0.13 0.10 0.30 0.24 0.64 0.25 0.08 13.53 1.64 –0.19 19.01 17.73 –0.23 2.19 1.20 0.27 2.60 0.92 –0.37 0.36 0.38 –0.36 1.19 0.60 –0.31 0.11 0.20 0.04
2
3
4
–0.35 –0.67 0.00 –0.04 0.12 –0.08 –0.25 0.21 0.06 –0.03 0.15 0.17 0.44 0.11
–0.46 –0.05 –0.04 –0.12 0.19 –0.04 0.01 0.00 0.03 0.07 –0.05 –0.11 –0.02
0.04 0.06 –0.01 –0.08 0.27 –0.21 –0.05 0.01 –0.19 –0.12 –0.33 –0.09
Notes: For all correlations greater than .11, p <.05. SD: Standard deviation.
discretion, use of problem-solving groups, and the pay ratio are all significantly associated with lower quit rates, while script use and percent commission pay are associated with higher quits. We estimated the effect sizes of the Tobit coefficients by decomposing them into estimates of changes in outcomes above the left censored limit and changes in the probability of observing an outcome above the left limit.27 This provides an interpretation equivalent to OLS estimates.28 The Tobit coefficients in the model are 0.62 of the OLS coefficients. Thus when the work discretion index changes by one standard deviation, quit rates decrease by 0.45 percentage points (0.62 –0.73); a one-standard-deviation increase in the percentage of employees who participate in offline teams decreases the quit rate by 1.08 percentage points (p < .001).
27. McDonald and Moffitt (1980). 28. The adjustment based on the second term in the McDonald and Moffit (1980) decomposition is calculated by multiplying the Tobit coefficients by [1–z*f(z)/F(z) – f(z)2/F(z)2], where F(z) is the cumulative normal distribution function associated with the probability of cases being above the left limit, f(z), the first derivative of F(z) is the unit normal density associated with this probability, and z is the corresponding z score for this probability. See Roncek (1992).
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
5
6
7
8
0.13 –0.08 –0.10 0.24 –0.06 0.22 –0.01 –0.01 –0.01 0.09 –0.13
–0.71 –0.60 0.11 0.07 0.17 –0.43 0.23 0.27 0.17 –0.05
–0.13 0.12 –0.19 –0.15 0.26 –0.09 –0.12 –0.16 –0.05
–0.30 0.12 –0.07 0.31 –0.22 –0.25 –0.05 0.13
9
10
11
12
–0.46 –0.04 0.06 0.07 –0.18 –0.17 –0.16 0.24 0.04 –0.29 –0.18 0.19 0.15 –0.16 –0.46 0.50 0.30 –0.20 –0.29 0.08 0.08 –0.05
353
13
14
15
0.25 0.33 0.14
0.26 –0.04
0.39
Discussion, Limitations, and Policy Implications In this study, we examined the extent of variation in service management and employment strategies among in-house, outsourced, and Indian offshore call centers that provide similar services to U.S. customers. We found significant differences in the patterns of employment practices and related outcomes across these three settings, but not in ways that were entirely anticipated. In this sample of establishments, in-house centers tended to adopt a more coherent quasiprofessional approach to service interactions than outsourced and offshore sites, with in-house jobs characterized by relatively higher levels of initial investments in training and pay, discretion, and problem-solving opportunities. Offshore centers, by contrast, had somewhat higher levels of formal education and initial training than in-house centers, but significantly lower levels of employee discretion and problem solving opportunities, and higher levels of electronic monitoring and performance management. From a managerial perspective, U.S. outsourced centers seem to present the worst of both worlds: a workforce with lower levels of formal education and training than in-house or offshore centers, low levels of discretion and problem solving opportunities that closely resemble those of offshore centers, and levels of pay much closer to those found among in-house operations than among Indian centers.
Table 3. Tobit Estimates for Quit Rates Model 1
Organizational and market characteristics Large business segment Multiple market segments Union presence Outsourced Offshore Human capital Workforce: percent female Years of education Initial training investment Work design Script use Work discretion index Percent in offline teams
Model 2
Model 3
Model 4
Coefficient
Standard error
Coefficient
Standard error
Coefficient
Standard error
Coefficient
Standard error
–1.02d –0.62a –1.99d 1.22d 2.03d
0.31 0.38 0.54 0.41 0.47
–0.61a –0.50 –2.02d 0.78 a 2.30d
0.32 0.37 0.55 0.41 0.48
–0.32 –0.21 –2.15d 0.19 0.57
0.30 0.35 0.51 0.39 0.50
–0.17 –0.20 –1.91d 0.15 0.40
0.31 0.34 0.52 0.39 0.50
1.10 a -0.20b -0.02b
0.68 0.09 0.01
0.22 –0.06 –0.01a
0.64 0.09 0.01
0.19 0.01 –0.01a
0.67 0.09 0.01
0.22a –0.73d –1.84d
0.12 0.15 0.37
0.23b –0.73d –1.74d
0.12 0.15 0.37
Compensation strategy Pay ratio Percent of pay based on commission Constant Sample size Chi square likelihood ratio Probability > Chi square Pseudo R2 Note: Unstandardized Tobit estimates are reported a. p <.10 b. p <.05 c. p <.01 d. p <.001
2.27 310 50.02 0.00 0.04
5.47 310 67.73 0.00 0.05
6.28 310 124.89 0.00 0.09
–0.52a
0.32
1.47b
0.71
5.70 310 130.05 0.00 0.10
356
Brookings Trade Forum: 2005
In further multivariate analyses, we found that outsourced and offshore centers had significantly higher quit rates after controlling for the market segment served, union presence, and measures of employee human capital. Systematic differences in work design explained most of the variation in quit rates, so the significance of ownership status disappeared when these practices were included in our equations. That is, ownership status is an important driver in the choice of management and employment practices, with outsourced and offshore centers more constrained to use standardized operating procedures and performance monitoring. It is these practices, in turn, that explain the higher quit rates in these centers. There are several limitations to this study. One concerns the representativeness of our samples, which we discussed earlier. Because larger organizations are overrepresented, if anything the study overstates the level of workforce education, pay, and levels of employee participation in call centers. We have no reason to believe that the bias is greater in one sample or the other, but there is really no way to test this deficiency in the data. A second limitation is that these largescale surveys provide only single-sourced data, and external labor market data from India are not available to compare the relative value of call center pay in that country with pay levels in the United States. A third limitation is that we cannot determine whether differences in management and employment systems are due to differences in the complexity of work functions or differences in business strategies based on quality and cost. Complexity and quality service strategies are highly correlated, such that companies tend to adopt quality strategies for higher-value-added functions, which typically are more complex in nature. In our analysis of average differences across ownership types, we used various methods to compare centers by industry and customer segment as well as work function. In each of these analyses, we found systematic differences based on ownership type. However, sample size restrictions prevent us from determining whether these differences are due to business strategies or service complexity, or some combination of both. In addition, the outcome measured in this study is limited. On the one hand, turnover is a useful metric to analyze because the industry has unusually high levels of workforce churn, which is widely recognized to be problematic and costly. There is also considerable empirical evidence to show that turnover is associated with lower service quality and productivity. On the other hand, future research needs to examine a much wider array of performance measures that directly capture operational quality and productivity if we are to understand the relative costs and benefits of alternative service management strategies.
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
357
Despite these limitations, the findings are consistent with other research on subcontracting relations. For example, subcontractors are more likely to have standardized processes and to use more performance monitoring and metrics, a pattern that supports Ravi Aron and Ying Liu’s argument in this volume that these practices are central to reducing operational risk. Aron’s finding that workforce training does not have a large effect on reducing operational risk is also consistent with our analysis. Despite the fact that offshore centers in India hire college-educated workers and offer considerable initial training, the high levels of process standardization do not let employees use their human capital in ways that can improve operational performance. This point is reiterated in the work of Vivek Agrawal (this volume), who demonstrates that the return to investment in technology in Indian call centers is far below that found in the United States. He notes that this “cookbook” approach to management reduces the incentives to innovate and constrains the ability to move up the value chain. Our findings have several policy implications at the level of managerial strategy and broader public policy. For management, the evidence is clear that the extensive use of routinized work processes in call centers leads to high turnover, which limits options for customization and is associated with lower service quality and productivity. Moreover, to the extent that call centers hire collegeeducated workers, the highly constrained and monitored work system creates an inefficient use of human capital: a particularly bad fit between selection and recruitment policies on the one hand, and between selection and work design policies on the other. The underutilization of human capital represents a substantial loss for Indian subcontractors, who are paying for skills that they are not using. Thus, to the extent that companies have complex service offerings or want to compete on the basis of service differentiation, quality, or customer loyalty, they are likely to retain customer contact interactions in-house, consistent with the transaction costs perspective and core competency argument. To date this appears to be what most U.S. corporations are doing: after two decades of rapid growth of U.S. call centers, most industry estimates are consistent with our own survey that less than 15 percent of U.S. call centers are run by third-party subcontractors, and only a tiny fraction have moved offshore. However, for those transactions that are simple and codifiable, it is likely that companies will continue expanding their operations offshore. Our data suggest that the strategy of outsourcing operations to U.S. subcontractors is likely to be a transitory one because the modest reductions in labor costs (compared with those of subcontractors offshore) may be offset by the high costs of turnover and
358
Brookings Trade Forum: 2005
low levels of employee skill. According to this scenario, the U.S. subcontracting sector, which grew dramatically in the 1990s, will be the hardest hit by Indian competition. If these findings hold across a larger and more representative sample of establishments, then the shift in customer contact employment from the United States to India is likely to be considerable but remain confined to standalone work functions that are relatively simple or transactional. Under this scenario, the problem for both U.S. and Indian subcontractors is that stand-alone call centers appear to be failing (see Dossani, this volume). An alternative scenario is that Indian call centers will gain the ability to compete more fully on the basis of quality and customer service as well as price. In theory, this is possible. With an educated workforce and high relative pay for the Indian labor market, Indian centers could be poised to handle more complex and nuanced customer transactions and provide service that builds customer loyalty. However, the current work systems are not in any way geared toward that alternative, but rather contain fundamental contradictions that are reminiscent of the problems of high turnover among overqualified workers in the monotonous jobs found in U.S. manufacturing industries in the 1960s and 1970s. Current analyses of the potential for high-quality service in offshore centers give too much weight to the level of formal education among workers and too little weight to the organization of work and technology, which shape the effective use of that human capital.29 However, case study evidence by Rafiq Dossani (this volume) shows that some call centers have been able to move up the value chain and expand their operations to include increasingly complex processes. Whether these examples of best practice can expand to the majority of call centers in India remains to be seen. This question turns on whether the current approach to managing vendor relations—through tight control by client firms—is considered so fundamental to limiting costs and operational risk that it will not be abandoned, or whether it is a temporary phenomenon that will give way over time to closer supplier relations built on trust. In the former case, the Indian call center sector would continue to handle relatively simple, codifiable, low-value-added transactions. In the latter case, the offshore market could expand to cover a much larger portion of the U.S. customer contact business. Even here, however, companies will need to learn much more about what kinds of tacit knowledge and contextual understandings are needed for which types of customer interactions. In service settings where “bridging to sales” is a major source of revenues, for example, tacit
29. Jaikumar (1986).
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
359
knowledge of cultural norms may still be an important source of competitive advantage, thereby favoring U.S. in-house or outsourced locations. A third alternative could involve a combination of organizational forms, with companies using a number of in-house, outsourced, and offshore venues to manage similar types of customer interactions. In our field research, we found several instances of this emerging strategy; and Ravi Aron and Ying Liu in this volume demonstrate that this “extended” model of organization may hold the most promise for quality and productivity in the long run. This approach allows companies to create competition for cost and quality innovations among their own subsidiaries and vendors. It also allows for organizational flexibility, so that client firms can adjust volumes and vendor contracts to seasonal demand. Similarly, some U.S. multinational subcontractors are offering a variety of venues to client firms, including a combination of onshore and offshore call centers, with volumes able to fluctuate according to seasonal demand. These strategies may help U.S. subcontractors survive as client firms exert ongoing pressure to reduce costs. These scenarios also depend on the role that public policy plays in human resource development. In India, there is evidence that demand is outstripping the supply of skilled labor, at least in the short run, in call centers in cities such as Bangalore and Chennai. Thus, there is a need for the Indian government to invest in the skills and human resource infrastructure required to respond to external demand. In the United States, the question is whether subcontractors will be able to improve the skill base of the workforce. They may be able to do so in locations where they have access to certification programs and community college programs in customer service management. Because centers are often co-located in “call center cities”—such as Jacksonville, Tucson, San Antonio, Omaha, or Phoenix—there may be opportunities to build a skilled labor pool with access to ongoing education and opportunities for multi-employer job ladders that help stabilize employment. Our survey results suggest that public support for the industry is available, with 49 percent of outsourced call centers reporting that they use public training resources and programs. Nearly all of the managers we interviewed in the outsourced industry relied heavily on local universities, community colleges, and partnerships with welfare-to-work and public sector organizations to recruit employees. These resources offer the potential to improve the quality of the workforce. However, we found that they are often used to substitute for internal investments in employee skills and discretion rather than to support a more professional or high-commitment strategy. Thus, while these types of innovations could allow U.S. subcontractors to improve the quality of their workforce and employment practices, the limited evidence in our study suggests
360
Brookings Trade Forum: 2005
that public sector resources are being used to supplant, rather than complement, private investment in human resource systems. If this represents the future among U.S. subcontractors, then they are unlikely to remain competitive with their Indian counterparts.
Appendix. Estimates of U.S. Call Center Workforce, 2004 To estimate the number of call center jobs in the United States, we used the May 2004 Occupational Employment and Wage Survey of the Bureau of Labor Statistics. We chose “office and administrative support occupations” (NAICS 430000). Within that category, we chose the suboccupations that were most likely to be located in call centers, based on the BLS description of work tasks and our own knowledge of call center operations. We also included telemarketers from sales occupations. Table A-1 details the employment numbers, percent of sample, and mean wages for those suboccupations. Table A-1. Employment and Compensation in Typical Call Center Occupations, United States
Call center work tasks Switchboard, answering services Telephone operators Bill and account collectors Credit authorizers and checkers New accounts clerks Order clerks Reservation agents, travel clerks (excludes travel agents, hotel clerks) Insurance claims and policy processing clerks Customer service representatives Telemarketers Total call center workers
U.S. workforce (percent)
Mean hourly wage (dollars)
Mean annual wage (dollars)
206,370 38,500 445,180 66,010 96,560 289,830
0.15 0.03 0.32 0.05 0.07 0.21
10.81 14.53 13.95 15.15 13.55 12.85
22,490 30,220 29,010 31,520 28,180 26,730
159,910
0.11
14.48
30,120
239,250 2,021,350 410,360 3,973,320
0.17 1.45 0.29 2.85
14.70 14.01 11.29 13.53a
30,580 29,130 23,490 28,147a
Number employed
Source: Bureau of Labor Statistics, Occupational Employment Statistics, May 2004 (www.bls.gov/oes). a. Weighted average, weighted by number employed by occupational group. By this methodology, there were an estimated 3.97 million call center workers in the United States in 2004, representing 2.85 percent of the working population. This estimate, however, undercounts some workers while overcounting others. Overcounting may occur because some of the workers in the categories provide face-to-face service. Undercounting occurs because this tabulation does not include other sales agents besides telemarketers, and many call centers define their work as primarily sales. If one subgroup of sales agents is also included (“sales representatives, services, other” [NAICS 41-3099]), then the estimated number of call center workers rises to 4.33 million, or 3.11 percent of the workforce. By these calculations, a reasonable estimate of the U.S. call center workforce in 2004 is between 2.5 and 3 percent of the U.S. workforce. This estimate is considerably higher than that found in reports by industry consultants. It may overstate the current numbers of jobs in call centers, but it includes jobs that, if not now organized into call centers, are prime targets for call centers in the future.
Comment and Discussion
Vivek Agrawal: The empirical data presented by Professor Batt confirm what many observers of the call center industry have long believed. Companies are taking a highly prescriptive approach to managing offshore call centers, a stance that risks continuing low rates of talent retention and an inability of providers to move up the value chain. However, there is another important implication of Professor Batt’s findings: prescriptive management is also stifling innovation at these centers and therefore depriving them of a significant amount of economic value. McKinsey calculations show that if companies were to innovate and optimize call center processes, especially rethinking the proportion of labor and capital costs, they could capture an additional 20–30 percent savings on top of what they are achieving today. Reasons behind the Prescriptive Management Approach Professor Batt notes that companies are taking the prescriptive management approach for good reason: they believe that they must take action to manage the inherent risks of operating in an unknown environment. This is generally true for all offshored services, but is particularly true of call centers. Companies are more careful when offshoring call centers because of the real-time nature of their services. In call centers, the process itself becomes the product; that is, how a call center runs its processes is precisely how its customers experience the “product.” So an understandable nervousness and anxiety exists on the part of companies procuring call center services offshore about how that service is performed. This is not true in other areas—manufacturing, for example. When procuring offshored manufactured goods, companies are willing to accept 361
362
Brookings Trade Forum: 2005
improvisation in production techniques so long as the end product meets their end-product specifications. In call center services, however, companies have mandated a cookbook approach to managing procedures. Although Batt and her colleagues have studied restrictive management practices at the agent level, I would argue that restrictive management is also happening at the managerial level. The long-term implications of the cookbook approach to managing employees go beyond just talent quality and retention. Innovation could offer companies a major competitive advantage. Those that do not encourage it could soon find that their competitors are achieving significantly higher levels of productivity and cost savings. Achieve Savings by Innovating Processes According to the cookbook philosophy, in offshoring centers, managers’incentives (bonuses, promotions) are tied to their ability to adhere to the cookbook. Supervisors are given incentives to do exactly as they are told. This is starkly different from the managerial role in manufacturing, where managers are given incentives to innovate. In fact, supervisors at offshored services centers are “punished” if found to “innovate.” However, the need for innovating process templates is significantly more important in services than in manufacturing. This is because, unlike in manufacturing, the proportion of labor to capital costs in services performed onshore is fundamentally different from that in an offshore location. On shore, 70 percent of cost is labor, 30 percent is capital. Offshore, 70 percent is capital, 30 percent is labor. Yet companies mandate that offshore centers operate exactly as they would onshore. They plug in the cheap labor and are content with the resulting savings. Manufacturers think about these issues differently. For example, the paint shops in a BMW plant in South Africa are far less mechanized than in Munich, but achieve the same quality. As long as the end product meets the required specifications, BMW has adjusted its inputs to make the optimal labor-capital tradeoffs (see figure 1). But flexibility in changing the proportion of capital to labor costs is generally not permitted in the offshore call centers. As a result, offshore call centers have employees being paid a fraction of onshore wages, but with infrastructure that is identical to that of onshore call centers. The offshore worker is required to use the equipment, with the equipment idling the same amount of time as in onshore environments. As a result, total factor productivity at offshore centers is far lower than its potential.
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
363
Companies in low-wage environments can make capital-labor tradeoffs in a number of ways: (1) they can reduce capital inputs by reducing automation; (2) they can use cheaper indigenously developed technology such as locally developed programs in place of branded expensive software; or (3) they can utilize capital more intensively. Capital can be used more intensively in two ways: by increasing shift utilization—that is, by running the call center round the clock, or by changing a process in order to reduce the downtime for capital equipment (see figure 2). Redesigning a task or a process to use capital more intensively bears some elaboration. Consider the simple example of a call center agent who manages customer accounts. In a high-wage country, each customer call is routed to an agent who listens to the request, opens up a computer database, and updates the account in real time. Neither the computer nor the telephone is used efficiently, since the agent is either talking or typing, seldom doing both simultaneously. The call takes longer, tying up telecommunications time, and keeping other customers on hold longer. However, an offshore agent equipped with only a telephone could write the customer request by hand into a tracking log and move on to the next call. Telecom costs are reduced because the agent spends less time on calls and customers less time on hold. Another agent, working at a computer station used around the clock, could enter the information into the database. While the new process requires more agents to handle requests, expensive computer hardware and software and telephone lines are used more intensively. Added wages are more than offset by savings on computers, software licenses, and telephone connections. The economics of an Indian call center suggest that this simple change could actually boost current profit margins for offshoring vendors by as much as 50 percent (see figure 3). This approach of disaggregating the value chain and reengineering processes to use capital more intensively can be used across a whole range of processes beyond call centers, even for knowledge-based services such as research and other information jobs. Innovation Will Soon Be a Competitive Imperative Why do companies hesitate to change established process templates, preferring to lose these gains, and how long do we expect this trend to continue? The answer lies in achieving sufficient competitive intensity in offshore locations. Today, a company that decides to go offshore is ahead of its competitors. The senior country manager who sets up an operation in India, the Philippines, or another low-wage country is given a budget and is told to produce the
Figure 1. Savings from Offshoring and Reengineering 45–55 percent saving
100
30–40 percent saving on offshore cost base
100
90 80 70
60–65
60 45–55
50 40
30–35
30 20
10–15
10 Original cost base
Factor cost savings
5–10
5
Added telecom cost
Added management cost
Task and process migration
Source: McKinsey Global Institute.
5–7 Offshore location cost
Task reengineering
Process reengineering
Task level gains
Process level gains
New base cost
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
365
Figure 2. Call Center Operating Costa Dollars/seat/hour Variable costs 10
Fixed costs
8 6 4 2
One shift per day (eight hours)
Two shifts per day (sixteen hours)
Three shifts per day (twenty-four hours)
Source: McKinsey Global Institute. a. Author performed similar calculations for five other sectors (not shown). Call centers exhibited the largest potential savings.
Figure 3. Savings from Process Reengineering in Management of Customer Accounts Dollars/billable seat/hour
3.00
(1.20)
Current profit margin
Impact of increase in transactions processing time on labor (five minutes) Penalty on labor productivity
Source: McKinsey Global Institute.
2.60
0.20
4.60
Impact of process reengineering (increased shift utilization by five minutes)
Impact of task reengineering (reduction in software licensing costs)
Total
Improvement in capital productivity
Net impact
50 percent lift in profit margins
366
Brookings Trade Forum: 2005
expected 40–50 percent savings. The manager has little incentive to rock the boat by changing proven ways of doing things. For the company, 40–50 percent is significant enough savings. Seventy percent would certainly be great, but there is a risk involved in doing something completely unproven. And given that the company’s competitors have typically not achieved even this level of savings, there is limited pressure to go the extra mile. However, as providers in low-wage countries ramp up capabilities, the lowwage location advantage will become a commodity. More and more companies will come to view offshoring as less risky. To some extent this is already happening. As this trend accelerates, and the industry frontrunners can no longer take comfort in the 40–50 percent savings they are achieving today, they will be forced to innovate and capture the additional 20 percent. Qualified professionals in low-wage destinations for offshoring are bubbling with ideas to increase capital productivity. For now, though, companies are not allowing them the room to innovate and go beyond the 50 percent cost savings achieved simply by moving operations to low-wage environments. I expect this will soon change. General Discussion: The formal presentations stimulated a wide-ranging conversation that focused especially on whether offshore call center operations might change dramatically over time to better optimize for local conditions and on whether the strikingly higher rates of attrition in offshoring relative to in-house call center operations might signal the demise of that organizational form. Alan Deardorff was struck by the finding that the offshore call centers were using their skilled labor suboptimally. He conjectured that in India fluency in English is most likely a proxy for high education levels. Thus, the call centers might be deploying this labor suboptimally because they are hiring solely for English fluency and not optimizing over the remaining educational capabilities of their employees. Deardorff also wondered whether there might be useful survey data on the amount of time actually consumed by individual calls, conjecturing that the lower wage levels in India might reduce employers’ sensitivity to call times as a cost variable and thus loosen the time limits imposed on employees in offshore call centers. Susan Collins agreed that the information would be valuable but cautioned that it might be difficult to interpret different average call times if different classes of call centers are handling calls of different complexity. T. N. Srinivasan argued it is premature to judge the innovation capabilities of call centers. The precise amount of risk involved with an offshoring project remains largely unknown because Indian providers are still undergoing a learning process. Although the majority of offshore call center managers currently must
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
367
follow strict procedures without much leeway to innovate independently, this practice may be a means to assess the performance and risks associated with offshoring. Once it is better understood that the risk is relatively low, it will be more likely that the benefits from giving a center more freedom to innovate will be achieved. Srinivasan took issue with Vivek Agarwal’s projection that offshore call centers were likely in the future to choose a different mix of labor and capital to take advantage of the relatively lower cost of labor in India. He cited the poor overall empirical record of “appropriate” technology—technology tailored to local conditions. However, Gary Saxonhouse cited Japan, Korea, and Taiwan as examples of emerging markets that have successfully adapted technology developed for high-wage environments to local factor prices. Lael Brainard also focused on the potential for further cost savings from adjusting the technology to local conditions in the offshore market. She noted the striking similarity between Vivek Agarwal’s conjecture that there are substantial unrealized cost savings in offshore call center operations and Ravi Aron’s findings in financial services. The impetus to push costs down further in the offshore operations appears to be temporarily absent, she said, because the initial cost savings of migrating these processes offshore are so substantial and because of the perceived risk of departing from the highly codified initial model. But Brainard argued that this is likely to change and thus that the offshoring phenomenon may still be in the early stages of driving productivity increases and cost savings across a range of industries. As competition increases, the differentials will become even more dramatic. Catherine Mann asked whether the sample could be stratified by industry, conjecturing that the comparisons between in-house versus offshore call center operations could lead to very different results depending on the industry being represented in the available data. Rishi Daga from Reliance commented on the difficulty of accurately measuring attrition rates at offshored call centers and suggested the survey data may actually understate attrition rates. In his company’s experience, attrition rates averaged between 25 and 30 percent for back office processes, about 50 percent for inbound call centers, and 70 to 75 percent for direct marketing and telemarketing. The attrition rates are disguised because many vendors in India and the Philippines retain call center employees on the company payroll for twelve to eighteen months as temporary workers. Since these workers are not regular employees, attrition on their part may not show up in the official numbers. Rafiq Dossani noted that both presentations gave credence to the prediction that the model of the independent call center—recently hailed as the future of
368
Brookings Trade Forum: 2005
the industry—might actually be failing because of unmanageably high attrition. In turn, he speculated that the call center function may be far less separable from the overall business process than is widely believed. The successful call centers were integrated into in-house operations, and the greater stability and success of the in-house call centers may be attributable to the fact that they can do other tasks as well. In contrast, outsourced offshore call center employees have much less flexibility. Susan Collins requested that the authors add more information about how the relative importance of the different type of organizational forms had changed over time, and whether the data supported Rafiq Dossani’s provocative assertion that the independent call center model had failed. Richard Freeman noticed that unionization appeared to reduce attrition significantly and wondered whether this applied to unionization of offshore operations or only to U.S.-based operations. Chad Bown thought it was important to compare the offshore call center attrition rates with the average turnover rate in the local market rather than only with the U.S.-based call center operations. He noted that the authors’ findings might be even bigger if the outside options for call center workers in these countries have much lower turnover on average. Douglas Kaden of Oak Hill provided support for the paper’s inference that processes are more tightly controlled when they are offshored. He noted a company that selected only processes that could be tightly controlled for offshoring discovered during the migration process that only 50 percent of these highly controllable activities were actually subject to tight controls by the in-house domestic operation. Rosemary Batt first addressed the question of how to measure complexity in this industry and how to control for it. First, she and her coauthors eliminated the business-to-business segment of call center operations because they are much more complex and more concentrated in relationship management. The statistical analysis initially controlled for industry and work function within the mass market broadly. However, since it was found to have no effect, it was taken out. She noted that even within the mass-market segment, however, there are varying degrees of complexity depending on the quality of the customer interaction. Furthermore, the process might be complex even if a product itself is not complex. For example, in telecommunications packaging variety and customization may require a call center operator to negotiate and interpret. Understanding those nuances and selling product features introduces complexity. She also noted that she is still struggling with issues raised in previous sessions about how difficult it is to outsource sales.
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
369
Batt noted that unionization is low for call centers overall but ranges as high as 25 to 30 percent in the telecommunications sector. Wage rates for the unionized workers are $40,000 a year rather than the average $30,000 for in-house call center employees. She also noted that the communications workers union recently negotiated major reductions in wages in order to keep the frontline work in-house. Finally, Batt ventured that the independent call center model may be less fragile than it appears because consumers demonstrate an apparent high tolerance for poor quality.
370
Brookings Trade Forum: 2005
References Abraham, Katherine. 1990. “Restructuring the Employment Relationship: The Growth of Market-Mediated Work Arrangements.” In New Developments in the Labor Market, edited by Katherine Abraham and Robert McKersie, pp. 85–118. MIT Press. Appelbaum, Eileen, Thomas Bailey, Peter Berg, and Arne L. Kalleberg. 2000. Manufacturing Advantage: Why High-Performance Work Systems Pay Off. Cornell University Press. Arthur, Jeffrey. 1994. “Effects of Human Resource Systems on Manufacturing Performance and Turnover.” Academy of Management Journal 37 (3): 670–87. Barney, Jay B. 1991. “Firm Resources and Sustained Competitive Advantage.” Journal of Management 17 (1): 99–120. Batt, Rosemary. 1999. “Work Organization, Technology, and Performance in Customer Service and Sales.” Industrial and Labor Relations Review 52 (4): 539–64. ———. 2000. “Strategic Segmentation in Front-Line Services: Matching Customers, Employees and Human Resource Systems.” International Journal of Human Resource Management 11 (3): 540–61. ———. 2002. “Managing Customer Services: Human Resource Practices, Quit Rates, and Sales Growth.” Academy of Management Journal 45 (3): 587–97. Batt, Rosemary, and Lisa M. Moynihan. 2004. “Human Resource Practices, Service Quality, and Economic Performance in Call Centers.” CAHRS Working Paper 04-16. Ithaca, N.Y.: Center for Advanced Human Resource Studies, Cornell University (www.ilr.cornell.edu/cahrs/2004). Batt, Rosemary, Alex Colvin, and Jeffrey Keefe. 2002. “Employee Voice, Human Resource Practices, and Quit Rates: Evidence from the Telecommunications Industry.” Industrial and Labor Relations Review 55 (4): 573–93. Batt, Rosemary, Virginia Doellgast, and Hyunji Kwon. 2004. “The U.S. Call Center Industry 2004: National Benchmarking Report.” CAHRS Working Paper 05-06. Ithaca, N.Y.: Center for Advanced Human Resource Studies, Cornell University (www.ilr.cornell.edu/cahrs/2005). Carayon, Pascale. 1993. “Effect of Electronic Performance Monitoring on Job Design and Worker Stress—Review of the Literature and Conceptual Model.” Human Factors 35 (3): 385–95. Datamonitor. 2001. “U.S. Customer Relationship Outsourcing to 2005.” London: Datamonitor. ———. 2003. “Opportunities in North American Call Center Markets to 2007.” New York: Datamonitor. ———. 2004. “The Vertical Guide to Contact Centers in North America: Tracking Sector Needs in a Mature Market.” New York: Datamonitor. Davis-Blake, Alison, and Brian Uzzi. 1993. “Determinants of Employment Externalization: A Study of Temporary Workers and Independent Contractors.” Administrative Science Quarterly 38 (2): 195–223.
Rosemary Batt, Virginia Doellgast, and Hyunji Kwon
371
Deery, Stephen J., Roderick D. Iverson, and Janet P. Walsh. 2002. “Work Relationships in Telephone Call Centers: Understanding Emotional Exhaustion and Employee Withdrawal.” Journal of Management Studies 39 (4): 471–97. Dossani, Rafiq, and Martin Kenney. 2004. “Went for Cost, Stayed for Quality? Moving the Back Office to India.” Asia-Pacific Research Center, Stanford University. Erickcek, George A., Susan N. Houseman, and Arne L. Kalleberg. 2003. “The Effects of Temporary Services and Contracting Out on Low-Skilled Workers: Evidence from Auto Suppliers, Hospitals, and Public Schools.” In Low-Wage America: How Employers Are Reshaping Opportunity in the Workplace, edited by Eileen Appelbaum, Annette Bernhardt, and Richard J. Murnane, pp. 368–403. New York: Russell Sage Foundation. Freeman, Richard B., and James L. Medoff. 1984. What Do Unions Do? New York: Basic Books. Frenkel, Steve, May Tam, Marek Korczynski, and Karen Shire. 1998. “Beyond Bureaucracy? Work Organization in Call Centres.” International Journal of Human Resource Management 9 (6): 957–79. Grugulis, Irena, Steven Vincent, and Gail Hebson. 2003. “The Rise of the Network Organization and the Decline of Discretion.” Human Resource Management Journal 13 (2): 45–59. Heskett, James L., Earl W. Sasser, and Leonard A. Schlesinger. 1997. The Service Profit Chain. New York: Free Press. Holman, David. 2004. “Employee Well-Being in Call Centres.” In Call Centres and Human Resource Management, edited by Stephen Deery and Nick Kinnie, pp. 223–44. Basingstoke: Palgrave. Holman, David, Claire Chissick, and Peter Totterdell. 2002. “The Effects of Performance Monitoring on Emotional Labour and Well-Being in Call Centres.” Motivation and Emotion 26 (1): 57–81. Huselid, Mark A. 1995. “The Impact of Human Resource Management Practices on Turnover, Productivity, and Corporate Financial Performance.” Academy of Management Journal 38 (3): 635–72. Jaikumar, Ramchandran. 1986. “Postindustrial Manufacturing.” Harvard Business Review 64 (6): 69–77. Kinnie, Nick, and Jon Parsons. 2004. “Managing Client, Employee and Customer Relations: Constrained Strategic Choice in the Management of Human Resources in a Commercial Call Centre.” In Call Centres and Human Resource Management, edited by Stephen Deery and Nick Kinnie, pp. 102–26. Basingstoke: Palgrave. Lepak, David P., and Scott A. Snell. 1999. “The Human Resource Architecture: Toward a Theory of Human Capital Allocation and Development.” Academy of Management Review 24 (1): 31–48. Levy, Frank, and Richard J. Murnane. 2004. The New Division of Labor: How Computers Are Creating the Next Job Market. Princeton University Press. Loveman, Gary W. 1998. “Employee Satisfaction, Customer Loyalty, and Financial Performance: An Empirical Examination of the Service Profit Chain in Retail Banking.” Journal of Service Research 1 (1): 18–31.
372
Brookings Trade Forum: 2005
Maddala, Kamaswari. 1992. Introduction to Econometrics. 2nd ed. New York: Macmillan. Marsden, David. 1999. A Theory of Employment Systems: Micro-Foundations of Societal Diversity. Oxford University Press. McDonald, John F., and Robert Moffitt. 1980. “The Uses of Tobit Analysis.” Review of Economics and Statistics 62 (2): 318–21. Pfeffer, Jeffrey, and James Baron. 1988. “Taking the Workers Back Out: Recent Trends in the Structuring of Employment.” Research in Organizational Behavior 10: 257–303. Pine, B. Joseph. 1993. Mass Customization: The New Frontier in Business Competition. Harvard Business School Press. Prahalad, C. K., and Gary Hamel. 1990. “The Core Competence of the Corporation.” Harvard Business Review 68 (3): 79–91. Quinn, James Brian. 1992. Intelligent Enterprise. New York: Free Press. Roncek, Dennis W. 1992. “Learning More from Tobit Coefficients: Extending a Comparative Analysis of Political Protest.” American Sociological Review 57 (4): 503–07. Shaw, Jason D., John E. Delery, G. Douglas Jenkins Jr., and Nina Gupta. 1998. “An Organization-Level Analysis of Voluntary and Involuntary Turnover.” Academy of Management Journal 41 (5): 511–25. Shire, Karen, Ursula Holtgrewe, and Christian Kerst. 2002. “Re-Organising Customer Service Work: An Introduction.” In Re-Organising Service Work: Call Centres in Germany and Britain, edited by Ursula Holtgrewe, Christian Kerst, and Karen Shire, pp. 1–16. Aldershot: Ashgate. Singh, Jagdip. 2000. “Performance, Productivity, and Quality of Frontline Employees in Service Organizations.” Journal of Marketing 64 (2): 15–34. Ventoro. 2005. “Offshore 2005 Research: Preliminary Findings and Conclusions.” Hillsboro, Ore.: Ventoro. Williamson, Oliver E. 1981. “The Modern Corporation: Origins, Evolution, Attributes.” Journal of Economic Literature 19 (4): 1537–68. ———. 1985. The Economic Institutions of Capitalism. New York: Free Press.
R AV I A R O N YING LIU University of Pennsylvania
Determinants of Operational Risk in Global Sourcing of Financial Services: Evidence from Field Research
T
he practice of global business process outsourcing (BPO) has gone beyond call centers to expertise-intensive functions such as tax accounting, equity research, cash flow forecasting, fixed income asset pricing research, transaction processing, supply chain coordination, and even research and development (R&D).1 Worldwide BPO is projected to reach $133.7 billion in 2005, an 8 percent increase from 2004 revenue of $123.8 billion, according to the Wharton Gartner Research Forecast (2003).2 The production of business processes is different from the production of physical goods in some important ways. First, the production of business processes involves no movement of physical goods, raw materials, inventories, or shipping and delivery costs. Here, inputs, output, and work-in-process are all information. Second, there is often minimal latency between the production of a process by a service center and its being bundled into a service for the end customer. An offshore service provider (a BPO firm) may process all the savings and checking account–related transactions of customers, and these customers may well make decisions based on their account balances a few minutes after the We wish to thank the Wharton–Singapore Management University Research Center for financial support that made some of the data collection leading to this research possible. We also thank the Infocomm Development Authority of Singapore (IDA) and the Ministry of Telcom and Information Technology, Mauritius. We also wish to thank executives at the following firms for sharing information and insights with us during our visits to their offices and in other discussions: AllsecTech Ltd., Beredium International, HCL Ltd., OfficeTiger, I-OneSource Ltd., and Wipro Technologies. 1. See “R&D Jobs: Who Stays, Who Goes?” Business Week, March 21, 2005. 2. Note that this is the size of the outsourcing market, of which offshore outsourcing forms a smaller segment (about 16 percent).
373
374
Brookings Trade Forum: 2005
transactions have been processed and the account has been balanced. This makes operational excellence—the accurate and timely production of informationintensive processes—of paramount importance. A third aspect in which the production of information-intensive processes varies is in the cost of inspecting the finished output. The act of inspecting the finished process for errors often involves retracing all the steps involved in its production. For instance, if a customer’s account balance (processed by an offshore BPO firm) is to be inspected for accuracy, then it is necessary to actually process each transaction and tally all the debits and credits to make sure that the account total is correct, all of which amounts to redoing the original work involved in balancing the account. There are numerous other business processes where the cost of inspection is as much as the cost of production, including those in corporate banking, insurance, direct procurement, brokerage services, and retail banking. Even where it is possible to inspect a service for quality, there are only a few unambiguous metrics of quality that both the buyer of services and the provider can agree on, such as what level of research is adequate for forecasting a stock’s price. Agency theory suggests that this aspect of offshored process production is rife with problems of moral hazard and resulting possibilities for opportunistic behavior on the part of the client or the provider. These factors amplify the likelihood that the process output will be of suboptimal quality—which we call operational risk— when processes are sent offshore to a third-party provider. By operational risk we mean the possibility that errors in processing information, delays in completion of work, inadequate documentation of procedures, or other unforeseen circumstances will lead to an output (finished process) that is of less than acceptable quality. For instance, when claims processing in insurance is outsourced to an offshore BPO firm, the offshore firm may not understand how to prioritize the client’s (buyer’s) requests and treat all requests alike by, for example, placing them all in a process queue. Alternatively, a BPO firm with a poor understanding of compliance requirements may keep insufficient audit trails of its work. An inadequate grasp of the client’s markets and business context may lead the offshore BPO firm to act in ways that provoke adverse reactions from the client’s customers. Examples of such outcomes, wherein the output of the outsourced firm is suboptimal from the standpoint of the buyer (client), are what we refer to as operational errors; the resulting risk of operational errors is operational risk. The complexity associated with information work is yet another factor that could amplify the operational risks associated with offshoring work to a third-party firm. For offshore outsourcing of information work to produce the gains associated with wage arbitrage and specialization, it is essential that operational risks be minimized.
Ravi Aron and Ying Liu
375
The aim of this paper is to investigate the factors that affect the operational risks of offshore business process outsourcing. We conduct an empirical analysis using the longitudinal data we collected, which includes biweekly detailed information on thirty-one outsourced processes carried out by five outsourcing providers for over a year. We found that the nature of the process, the effort and experience of the provider, and the characteristics of the contract all have a statistically significant impact on the operational risk of the outsourced processes. In the next sections, we review relevant research and then introduce the idea of complexity arbitrage, examine its implications, and link it to a discussion of operational risk. We then motivate the research hypotheses. In the next sections, we provide an econometric model of risk and analyze the survey data, discuss the managerial implications of the research, and analyze which forms of governance best mitigate the risk associated with offshore process production.
Literature Review Operational risk plays an important role in the transaction cost economics (TCE) literature in determining the boundaries of the firm. The prevailing view that the natural boundaries of the firm were determined by technology, technological nonseparabilities, and economies of scale was first challenged by Coase (1937), who held that the firm and the market were alternatives for organizing the very same set of transactions and the costs associated with arranging to have work done, rather than the cost of doing the work itself. This new view offered the best justification for the existence of separate firms rather than the universal reliance on market transactions. TCE explained how the firm could economize on transaction costs—that is, it identified the most economically efficient governance structure and described the conditions under which the firm and not the market provided the ideal governance structure. Central to TCE are transaction frequency, investment idiosyncrasy, and uncertainty as the critical dimensions of contractual relations that yield transaction costs. The role of risk as a large component of transaction costs that is created by leaving the internal hierarchy of the firm was first explained by Klein, Crawford, and Alchian (1978) and Williamson (1979). Agency theory explained the role of principal-agent problems, mainly deliberate underperformance of contractual tasks under conditions of imperfect observability and misalignment of incentives between client and vendor (see Grossman and Hart 1983; Grossman 1986). Incomplete contract theory brought up the opportunity renegotiation and hold-up problem as another contract risk between client and vendor (Hart and Moore
376
Brookings Trade Forum: 2005
1988, 1990). TCE has been readily used by researchers in the field of information systems to explain the impact of information technology (IT) on the boundaries of the firm. Bakos and Brynjolfsson (1993) show that while having too many suppliers can result in significant coordination costs and greater risk of operational errors, relying on too few suppliers amplifies the threat of opportunistic behavior. To further understand the nature and extent of operational risk associated with a process, knowing what components really drive the transaction cost and risk is important to both practitioners and researchers. The earlier empirical work in TCE confirms the positive relationship between investment specificity and contract cost, and thus contract duration or vertical integration (see Palay 1984; Joskow 1987; Crocker and Masten 1988; Monteverde and Teece 1982; Masten 1984). Anderson and Schmittlein (1984) find that the most important variables that drive the contract risk are the difficulty of evaluating performance and the importance of noncontract activities in the sales department context.
Process Complexity Our interest in operational risk associated with processes led us to study the nature of complexity associated with processes. We surveyed several senior executives regarding their views on the nature of complexity associated with several kinds of processes. We spoke to executives both in the “buyer economies” (the United States and the United Kingdom, for example) and in the “provider economies” (India, Mauritius, Thailand, and Singapore). To our surprise, there was no consensus around the extent of complexity associated with a process, and the executives’ views of complexity were quite subjective. Equally interesting, there was some agreement between the U.S. and U.K. executives even as there was agreement among executives of India, Singapore, Mauritius, and Thailand. However, while assessments of process complexity were positively correlated within the same country,3 they were negatively correlated across the offshoring divide. This means not only that the executives in the United States and the United Kingdom shared an assessment of complexity but that they disagreed with their counterparts in India, Singapore, and Mauritius. This led us to analyze the components of complexity and what led to the divergence in views of complexity. We selected some processes, mostly from the financial services industry, that were being executed in the United States and the United Kingdom 3. The United States and the UK on the one hand and India, Singapore, Mauritius on the other.
Ravi Aron and Ying Liu
377
Table 1. Survey of Process Complexity: Elements of Data Collected Dimension of complexity Agents in involved in process execution Firm-level inputs
Process characteristics
Composite complexity attribute
Elements of data collected Training; education; experience; compensation For example, the extent of managerial supervision; investment in IT monitoring systems; the extent of quality inspection by client Cost of inspecting finished process for errors, extent of errors in process execution; number of agents required to process a single process cycle; extent to which the work involved in the process could be codified; number of subtasks associated with the process; extent of algorithmic computation and quantitative analysis required to execute the processes, and other process-specific characteristics Managers were asked to rate the process complexity on a Likert scale (1 to 7) where 1 would be not at all complex and 7 highly complex.
as well as in India and Singapore and described the work involved in executing these processes in exact detail (see table 1).4 Our survey included a comprehensive collection of attributes associated with process complexity.5 We captured details of process execution ranging from data about the information workers (IWs) who executed these processes (including their training, educational level, and compensation) to firm-level inputs (such as the extent of managerial supervision and use of real-time systems to monitor quality), and process attributes (such as the extent to which the work involved in the process could be codified, the number of subtasks associated with the process, and the extent of algorithmic computation and quantitative analysis required to execute the processes). Finally we also asked managers from the four countries to provide a single composite rating of process complexity—that is, to rate the complexity of the process on a Likert scale (1 to 7). The analysis of the data collected provided further evidence of considerable divergence in managerial assessment of complexity across regions. Table 2 shows the complexity rankings assigned by clients in the United States and vendors in India and Singapore to the
4. As a result of the nondisclosure agreements that we entered into, we are unable to provide any description of processes or firms that could reveal the identity of the firms that provided these services in India and Singapore or their clients in the United States and the United Kingdom. 5. In selecting these attributes we were guided by extensive interviews with executives from all four countries.
378
Brookings Trade Forum: 2005
Table 2. Complexity Rankings Assigned to Outsourced and/or Offshored Processes, by Client Managers in the United States and the United Kingdom and by Vendor Managers in India and Singaporea Processb P1 P2 P3 P4 P5 P6 P7 P8 P9 P10
U.S. firms
U.K. firms
Indian BPO
Singaporean BPO
2 3 6 5 7 3 2 2 4 2
3 2 5 5 7 4 1 2 5 3
1 3 2 3 3 4 7 6 3 5
2 4 3 4 2 5 6 5 4 4
a. Managers ranked complexity on a Likert scale from 1 (not at all complex) to 7 (highly complex). b. To protect the confidentiality of firms, we cannot identify the specific processes.
same ten processes in the survey instrument. As we suspected, tasks that were viewed as complex by the American executives were generally viewed as much simpler by their vendors. There is a very high level of agreement on complexity, as perceived by vendors both in India and in Singapore. Indeed, the correlation between the observations from these two markets is close to 0.85. We then note that there is a strong negative correlation between complexity as predicted by American and British clients, and complexity as observed by Indian and Singapore vendors. In particular, we found that multi-stage, computationally intensive analytical tasks can be executed by engineers and technical personnel in India and in Singapore at adequate levels of quality. Often their performance levels relative to their counterparts’ in the United States and the United Kingdom are actually superior; that is, in addition to the wage arbitrage gains that this labor pool offers, these IWs, when provided with comparable training and comparable support technology, deliver levels of quality that are equivalent or superior to the levels of quality of the clients’ in-house operations. Examining how Western executives estimate the complexity of tasks to be outsourced explains why their estimates are consistently different from those of their vendors, and why there are opportunities for outsourcing tasks that they consider to be complex. Table 3 provides correlation coefficients in process complexity ratings between the four regions. There is strong agreement between the managers of the buyer economies—the United States and the United Kingdom—and equally strong agreement between the managers of the provider
Ravi Aron and Ying Liu
379
Table 3. Divergence in Perceptions of Complexity: Correlations of Process Complexity Ratings between Regions
U.S. firms U.K. firms Indian BPO Singaporean BPO
U.S. firms
U.K. firms
Indian BPO
Singaporean BPO
1 0.89 –0.47 –0.54
1 –0.56 –0.63
1 0.84
1
economies—India and Singapore. However, it is clear that between the two regions there is a clear divergence of views on what constitutes a complex process. What factors contribute to complexity as seen by managers in the buyer countries? Our survey shows that: —The buyer country managers surveyed consistently assigned higher complexity rankings to tasks that were more algorithmically described and more analytically or computationally demanding. —Tasks that were considered subjective, context-dependent, and implicitly understood received lower complexity scores than those that could be verbally described; those that could be explicitly codified, dependent only upon analysis of data using formal techniques that could be specified in advance, received the highest complexity scores from managers in the buyer countries. On the other hand, tasks that were seen as complex by managers in the provider economies often involved a high degree of contextual interpretation, with implicitly understood rules embedded in the business context of the client. For instance, Indian managers rated work that went into sell-side equity analysis that involved understanding mathematical models and analytical modeling skills as low in complexity; they rated writing a report on the prospects of a basket of stocks as significantly more complex. The opposite was the case with the managers of firms in the United States and United Kingdom. They rated the actual analysis that went into the report as high-complexity work and writing a report around analytical estimates (that have been made available) as work involving lower complexity. This difference raises the question whether there could be a market for ‘complexity arbitrage’ between the two regions? To answer this question we need to address a related issue: could the more complex processes (as seen by the Western managers) be offshored and the less complex ones executed in the Western economies? For such an arbitrage to be feasible, it is necessary that the extent of
380
Brookings Trade Forum: 2005
operational risk associated with offshoring processes be manageable and low enough to not negate the gains from wage arbitrage. In other words, we need to determine the extent of operational risk associated with complexity arbitrage. This observation leads us to the analysis of operational risk in the section that follows.
Operational Risk BPO involves the outsourcing of information-rich processes. The offshoring of processes is prone to errors in execution because the transformation of data into knowledge that can support decisionmaking is achieved through the combination of human intervention by IWs and computer processing (automated routines that sort and classify large data sets). Intervention by IWs is needed at various levels to convert, translate, transform, and validate the data that are fed into corporate information systems, and the risk of outsourcing a business process is determined by both the extent and the nature of the required IW intervention. These work task qualities can be plotted along a spectrum that we call the “knowledge continuum.” The Knowledge Continuum Our “knowledge continuum” categorizes business processes according to the judgment and expertise required to produce them. The knowledge continuum spans work from routine data creation (such as transcription and transfer from one medium to another) at one end of the spectrum, through information analysis that calls for some expertise and analytical skills on the part of the IWs, to highly judgment-driven expert work such as generating equity research and bond pricing research reports,6 and providing litigation research support. For those processes at the left end of the knowledge continuum, such as inbound tech-support calls, transaction processing, document management, direct mailing, and e-mail query resolution, most of information that the IWs need and most of work that needs to be done is easy to specify via business rules and calls for very little judgment and expertise. These processes feature a higher degree of what is described as “rules-based work” by Levy and Murnane (2004). Further to the right on the knowledge continuum are more complex processes that involve discretionary decisionmaking, such as yield analysis, direct procure6. While doing the mathematical analysis that goes into the report is less judgment-driven, the construction of the report involves a substantial amount of judgment-driven and often contextsensitive work.
Ravi Aron and Ying Liu
381
ment, critical event simulation, SEC filing verification, contract research for mergers and acquisitions (M&A), and the provision of remote research support to the investment banking industry. These processes call for the exercise of judgment, inference, and the ability to call upon tacit knowledge (or expertise) to map a course of action to a desired set of outcomes. Thus we can see that the knowledge continuum has three kinds of information generation activities: the set of activities leading to data creation, collation, correction, and transformation, which we call the data phase; the set of activities leading to information extraction and creation, which we call the information phase; and the set of activities leading to the production of knowledge that can be used to support decisionmaking, which we call the knowledge phase. On the basis of our field research we have mapped a set of offshored and outsourced processes on the knowledge continuum in figure 1. The nature of the information work that is performed along the knowledge continuum affects the extent of operational risk. In particular, two aspects of the knowledge work stand out. The first is that work that is performed in the data phase can be easily described as business rules that specify the set of actions that agents must take when they encounter specific situations. The second aspect that has an impact on operational risk is the set of metrics used to measure the quality of process output; an analysis of the data collected in our field research makes it intuitively clear that the quality of data transcription, correction, and other such tasks can be measured fairly easily through error rates and related metrics; in contrast, the quality of output of work that is carried out in the knowledge phase, such as a research report for investment banking or litigation support, cannot be easily measured. These observations in turn lead us to a discussion of the analysis of operational risk. Factors That Affect Operational Risk: Research Hypotheses Operational risk manifests itself in the form of operational errors. These errors may be in the form of inaccurate (mistake-ridden) process delivery, delays in completing work, incomplete or inaccurate audit trails, among others. In general, a deviation from a predetermined form of output is an operational error.7 The magnitude of operational risk is therefore the extent to which the output of a process is error-ridden. There are two advantages to proceeding 7. Note that it is possible to specify what the end result of a process should be, even when it is not possible to specify exhaustively how agents should behave when they encounter different situations (or states of the world). Here we make a careful distinction between what is to be done— that is, the output of a group of workers—and how the actual work is to be performed.
Figure 1. The Knowledge Continuum Manual reconciliation of transactions
Paper invoices scanned
Transactions matched and sorted by an application
Verification against other information sources within the firm Outstanding transactions sorted by size and data
Scanned invoice data entered into a transaction database
Data phase
Unreconciled transactions moved to Stage II
Stage II Verification against sources outside the firm such as banks, insurance companies,and third parties involved with the transaction
Unreconciled accounts moved to exception handling
Verification of transaction from historical data and archives
Transaction reconciled through account-level judgment calls
Payment advice issued where necessary
MIS reports generated
General ledger entries passed
Nonstandard MIS reports manually generated
Unresolved entries transferred back to user
Information phase
Working capital analysis, cash flow forecast, and yield analysis performed
Knowledge phase
Ravi Aron and Ying Liu
383
with this definition: (1) the firm that buys the services specifies what an error is in the service level agreement (SLA), thereby restricting the ambit of analysis to what really matters; and (2) parsimony: this model of operational risk is a parsimonious construction that can include different kinds of operational errors—errors of execution, delay in delivery, suboptimal customer satisfaction outcomes in different process types (in financial services, direct procurement, legal research, bio-informatics, and the familiar call center work). When the provider of services (the firm that executes these business processes) is given a fully specified charter of work—a document that exhaustively specifies exactly and how an agent should respond in each situation encountered while working on a business process—the provider is likely to make fewer errors. We call this ability to specify the set of agent responses to states of the world that can be encountered while working the “codifiability” of a process. The idea of codifiability of two constructs was developed by Levy and Murnane (2004): codifiability of work enables a combination of computer substitution and pattern recognition by human agents leading to specific actions that need to be taken. Processes that involve routine work are far more codifiable than processes that require subjective judgments. We would expect to find therefore that the more codifiable a process, the fewer the operational errors associated with the process. This leads us to our first hypothesis. H1: The higher the codifiability, the lower the operational risk. A second factor revealed by our field research to have a significant impact on the extent of operational risk is the convergence of understanding between the buyer of services (the firm) and the provider on precisely how quality of output is to be measured. Again, it is clear that this too is driven by the nature of work. For processes that fall on the left end of the knowledge continuum— routine data creation and simple forms of information extraction—the measures of process quality are clear and unambiguously understood by both the buyer and the provider of services. For processes requiring judgment and expertise that result in the creation of knowledge that supports decisionmaking, buyer and seller may not have complete convergence in their understanding of process quality. In several countries and many firms we encountered this phenomenon. For processes such as litigation research support, financial research in investment banking, financial statement analysis and presentation, the quality of the finished process, and indeed the measures used to estimate quality were not always well understood. We define an important measure here—the objectivity quotient
384
Brookings Trade Forum: 2005
of process metrics (objectivity quotient, or OQ, from here on) as follows: this is the ratio of the number of parameters of process quality on which there is complete convergence between buyer and seller to the total number of parameters of process quality. This ratio will equal unity when there is complete convergence and will fall to a number less than unity when there are one or more parameters on which the buyer and seller do not share an unambiguous understanding of process quality. The smaller the value of the OQ, the greater is the extent of operational risk, and the more operational errors. There is theoretical support behind this intuition: according to multitask theory, assigning tasks for which performance is imprecisely measured will increase the propensity of agents to shirk work (see Holmstrom and Milgrom 1991). This leads us to our next hypothesis. H2: The lower the objectivity quotient, the higher the operational risk. A process often consists of more than one task and is often executed by more than one IW. Often IWs pass information to one another, and one IW’s output (within the same process) serves as the input for another IW. We call such intraprocess links information dependencies (IDs). It is intuitively clear that the complexity of coordination increases with the number of IWs working on the same process, thereby resulting in greater operational risk. In general, more agents involved in executing a process increase the complexity of the coordination task and therefore the likelihood of operational errors. These observations lead us to our next two hypotheses. H3: Operational risk increases with the number of information dependencies (within a process). H3A: Operational risk increases with the number of IWs (per process cycle). A final set of factors that we will investigate has to do with the profile of agents. There is a body of research in the field of labor economics that points out that training given to IWs improves the quality of their work and that more experienced IWs make fewer errors. This leads us to a final set of hypotheses. H4: Operational risk decreases with more training imparted to IWs. H4A: The greater the experience of the IW, the smaller the operational risk.
Ravi Aron and Ying Liu
385
Model and Analysis of Results Our research hypotheses determine the econometric model that we employ. The Econometric Model Operational risk (R) is modeled as a linear combination of the factors discussed above: codifiability (C), number of agents per process (A), number of process dependencies (PD), objectivity quotient (OQ), number of measures of quality (Q), agent training (T), and agent experience (EX). We write the econometric model as follows:8 R it = 1 + 2Cit +3Ait +4PDit + 5Qit +6OQit + 7Tit +8EXit + i Data Profile We collected data from several firms in several countries starting in December 2004 until February 2005. The providers were drawn principally from Singapore, India, Mauritius, and Thailand, and the buyers were drawn from India and the United Kingdom. We used the data from five firms and thirty-three processes for this project. For each process we took two observations six months apart of all the variables (dependent and independent), for a total of sixty-six observations. Variables were operationalized as shown in table 4. Analysis of Results Our analysis shows that in general there is robust empirical support for our principal hypotheses. We find that codifiability and the objectivity quotient are indeed the major drivers of operational risk associated with process quality. Table 5 shows the results of the regression analysis. We find that greater codifiability is strongly correlated with lower operational risk. A unit increase in codifiability results in a decrease of 10.9 percent in the operational error associated with the process. Similarly, we find that convergence in process quality metrics—that is, OQ—has a marginal impact of a decrease of 25.9 percent in the operational error rate. It is clear that there is considerable value in having an exhaustive set of rules to describe the nature of the 8. Where the subscripts i and t stand for the i th process at the t th instant.
386
Brookings Trade Forum: 2005
Table 4. Operationalization of Variables Variable Codifiability (C) Number of agents per process (A) Number of process dependencies (PD) Number of measures of quality (Q) Objectivity quotient (OQ) Agent training (T)
Agent experience (EX)
Operationalization Likert scale 1 to 7; collected from process owners (managers in charge of process production) Number of IWs working per process cycle— computed from billing records From the process flow specified in the SLAa Captured from the SLA Ratio computed by dividing provider’s number by the number captured from the SLA Number of weeks of training (average) received by agents working on a process—computed by the provider from work log records Number of months of experience (average) of agents working on a process—computed by the provider from personnel data
a. Service level agreement.
work and in creating better shared definitions of process quality between provider and seller. The impact of complexity of coordination on the extent of process risk was also significant, though of considerably lower magnitude. As the number of agents per process increases, the rate of errors in process output increases. Each additional agent per process cycle causes a 1.5 percent increase in process quality. The number of process dependencies did increase the extent of operational risk, but to a less significant extent. We were surprised to find that process dependencies did not have a strong impact on the extent of operational error. We discuss these in the next section. Finally, we found that while greater training and experience (of IWs) did decrease the extent of operational risk, neither had a strong impact; and of the two, training had the stronger impact on operational risk. Note that the pooled OLS regression produces the same outcomes as the Huber-White robust error procedure. The results of the regression are also consistent across cross sections of time periods. We can see that the regression coefficients have more or less the same values across time periods. Finally, the GLS regression with the forced zero intercept constraint produces results very similar to the OLS and the H-W procedure. The convergence of regression tests assures us that our results are robust and without significant operationalization bias.
Ravi Aron and Ying Liu
387
Table 5. Model of Operational Risk: Regression Analyses
Variable Codifiability (C) Number of agents per process (A) Number of process dependencies (PD) Number of measures of quality (Q) Objectivity quotient (OQ) Agent training (T) Agent experience (EX) Intercept F (Prob > F)
Pooled OLS
First period OLS
Second period OLS
HuberWhite robust error
GLS
–0.109*** (0.002) 0.015*** (0.003) 0.008** (0.003) –0.014*** (0.002) –0.259*** (0.01) –0.008*** (0.001) –0.003*** (0.000) 0.985*** (0.021)
–0.110*** (0.002) 0.014** (0.005) 0.009 (0.006) –0.015*** (0.002) –0.260*** (0.01) –0.009*** (0.002) –0.004*** (0.001) 0.991*** (0.015)
–0.110*** (0.002) 0.015** (0.005) 0.008 (0.006) –0.015*** (0.002) –0.261*** (0.01) –0.009*** (0.002) –0.003*** (0.001) 0.994*** (0.017)
0.109*** –(0.001) 0.015*** (0.003) 0.008* (0.004) –0.014*** (0.002) –0.259*** (0.008) –0.008*** (0.001) –0.003*** (0.000) 0.985*** (0.013)
–0.110*** (0.000) 0.010*** (0.001) 0.015** (0.001) –0.016*** (0.000) –0.264*** (0.002) –0.010*** (0.000) –0.003*** (0.000) 0.999*** (0.003)
804.13 p < 0.0001
2332.56 p < 0.0001
2304.03 p < 0.0001
4437.33 p < 0.0001
Wald (Prob > 2) R2 (percent) N
296643.8 p < 0.0001 98.9 66
98.9 33
99 33
98.9 66
66
Notes: Standard error in parentheses; *** p < .001, ** p < .01, * p < .05.
Managerial Implications There are a number of interesting managerial implications of the insights uncovered by this research. We discuss the principal implications for managers and policymakers in the next section. Process Specification and Codification It is clear from our analysis that one of the principal drivers of operational risk is the extent to which a process and the work that goes into executing the process can be codified and explained analytically to the agents of the provider. Table 6 shows how operational risk increases as processes become less codifiable. Note too that for a given level of codification operational risk increases as the objectivity quotient declines (see table 7 for a similar comparison).
388
Brookings Trade Forum: 2005
Table 6. Codifiability and Risk Codifiability
Objectivity quotient
Operational risk
1.00 0.75 0.67 0.50 1.00 0.75 0.67 0.50 1.00 0.67 1.00 1.00 0.50 1.00
0.009 0.046 0.042 0.097 0.075 0.116 0.168 0.148 0.163 0.180 0.230 0.346 0.448 0.433
7 7 7 7 6 6 6 6 5 5 4 3 3 2
The value of codification becomes even greater when the employees of the provider (IWs) and the buyer are located in different countries. The IWs may be unfamiliar with the context of the work, although they may have a high degree of domain skill (such as in financial analysis, molecular biology, or accounting and actuarial services). Thus investments made in transferring process knowledge pay considerable dividends in the form of lower operational risk (and therefore higher process quality). Indeed, our survey found that firms (both buyers and suppliers) that invested the most in process specification and knowledge transfer were the ones that reaped the benefits of consistently high quality of process productions. Firms such as Wipro and HCL in India and Beredium International in Mauritius are examples of firms that invest in detailed process specification (codification). Convergence in Process Metrics A second insight from this stream of research is the importance of buyer and provider having as accurate an understanding of the parameters of process quality as possible. If the provider does not fully understand the nature of the parameters of process quality he will fail to monitor his employees sufficiently to ensure that the parameter of quality is met. This is even more the case for processes that call for a high degree of expert judgment. Table 7 shows how the operational risk increases with a decline in the objectivity quotient.
Ravi Aron and Ying Liu
389
Table 7. Process Metrics Convergence (Objectivity Quotient) and Risk Objectivity quotient 1.00 1.00 1.00 1.00 1.00 1.00 0.75 0.75 0.67 0.67 0.67 0.50 0.50 0.50
Codifiability
Operational risk
7.00 6.00 5.00 4.00 3.00 2.00 7.00 6.00 7.00 6.00 5.00 7.00 6.00 3.00
0.01 0.08 0.16 0.23 0.35 0.43 0.05 0.12 0.04 0.17 0.18 0.10 0.15 0.45
Firms such as OfficeTiger and HCL in India and IT-One in Thailand have recognized this factor and have set up procedures to make the measures of quality as transparent as possible. An OfficeTiger mechanism called the “Program Office” allows managers of the buyer and the supplier to jointly define the parameters of process quality and make sure that there is a convergence of understanding between the client and OfficeTiger in establishing process quality measures. Finally, we found it somewhat surprising that information dependencies had only a weak impact on the magnitude of operational risk. One of the explanations that we have for this phenomenon is that information dependencies are a weak proxy for the complexity of coordination; a stronger proxy for this phenomenon is the number of IWs per process. Indeed, it is clear that this factor has a stronger impact on the extent of operational risk than the factor of information dependency. This would strongly point toward automation as the means of ensuring process quality. Clearly, operational risk falls as the number of IWs that work on a process decreases. This indicates that when firms offshore processes they should also redesign them and introduce automation where possible to lower the complexity of coordination between agents. Firms such as Allsec Tech and I-OneSource have been able to do this to a significant extent for processes that span such diverse functions as financial services and supply chain coordination. Some processes can be automated over time, gradually reducing the
390
Brookings Trade Forum: 2005
number of agents needed to execute the process, while other processes resist automation because the nature of work involves a high degree of human judgment and interpretation (what Levy and Murnane refer to as pattern recognition). This raises the question of what else firms could do to lower the operational risk associated with offshore outsourcing and the issue of what economists call the “governance form,” or the nature of the contract between the buyer and seller of services. When a buyer of services can be reasonably confident that inspecting the finished output of the provider is effective—that is, the cost of inspection is low and the accuracy of inspection is high—it can indulge in what economists often refer to as arm’s-length outsourcing. However, when inspection is costly (or is not an accurate measure of process quality) the buyer would want to combine the control that in-house production of processes offers with the efficiencies of an offshore outsourcer.9 This observation leads us to our analysis of an emerging governance structure that combines the benefits of an organization with the market.
The Extended Organizational Form The advent of the Internet and the relatively cheap availability of bandwidth have enabled firms in the United States and Europe to link themselves reliably to firms in China, India, Mauritius, the Philippines, and Singapore. Advances in software platforms and connectivity have also made it possible for firms to monitor projects, people, and the execution of processes across the globe in real time and in great detail. An example of deep links between firms can be found in the case of Allsec Tech, a Chennai-based firm that has placed its process execution specialists on its client’s premises across the city, inside the Ford Motor Company’s “trusted zone.” Allsec Tech’s agents resolve supply chain coordination problems by tracking invoice clearance, payment, and accounts receivables and payables (by querying data repositories that are located in Detroit and Chennai), and provide expert intervention when it is called for. The agents of the provider work under the direct supervision of the provider’s mangers and the virtual supervision of the buyer’s (client’s) managers. It is important to note that this arrangement is offered for those tasks where the cost of inspecting the provider’s (Allsec Tech’s) quality of output is expensive. For several routine processes 9. The two governance forms are referred to as “hierarchies” and “markets” in research on the theory of the firm.
Ravi Aron and Ying Liu
391
where the quality of output is easy to inspect, this form of governance is not featured in the outsourcing contract. Such arrangements are not uncommon in the financial services industry either. Lehman Brothers has set up similar arrangements with its service providers in India, where agents of the provider work closely with the managers of the buyer to jointly monitor and supervise those tasks where inspecting the quality of the provider’s output is costly. Two large British banks have arrangements with their offshore service providers in India for the execution of (only) those processes for which the provider’s output quality is not transparent. The operations of OfficeTiger (OT) in India provide further evidence of this phenomenon. The firm is in the business of providing expert judgment and analysis to its clients. Its “Program Office” allows senior managers of the client and those of OT to jointly review and manage the strategic long-term goals as well as dayto-day operational details. Further, OfficeTiger has developed real-time performance tracking and quality control systems that allow its clients to track the productivity and quality levels of teams of agents and even the output of individual agents (employees of OT). It is possible for OT’s clients—some of the leading investment banks in the United States and the United Kingdom—to use these systems to provide detailed monitoring of their projects and ask for modifications to operational procedures, agent assignments, QC mechanisms, and project priorities. About 40 percent of the firm’s deadlines must be met in less than an hour and often involve research support, real-time scenario analysis, or model building with very little margin for error or time to correct errors after execution. The firm and its clients work off the same files, spreadsheets, and data feeds, and where necessary they work iteratively. Again our survey of OT showed that only for those processes and tasks where the provider’s quality of output is not transparent or where inspection is costly (involves costly delay in the delivery of service to the client’s customers) are such fine-grained joint monitoring and control mechanisms created. For more routine tasks, whose quality of output is transparent or inspection is not costly, the governance of the contract resembles the traditional outsourcing contract with prespecified price and postinspection penalties. While OT was perhaps one of the earliest and most successful proponents of this hybrid governance structure, which we call the extended organizational form (EOF), it is not the only one now practicing this form of governance. GECIS in India too has evolved its own flavor of EOF (similar to OfficeTiger’s in many ways). When GECIS’s employees are working on a client’s processes— such as tasks surrounding commercial real estate finance—the two teams, the client’s managers and GECIS’s agents, work as though they are part of the same
392
Brookings Trade Forum: 2005
team. GECIS has configured the project teams so that the offshore employees are closely monitored and controlled through performance tracking systems in real time. The process owners (the buyer’s managers) set task priorities, track progress, and help define quality standards and operational measures to ensure adequate output quality. Other large offshore service providers too are beginning to deploy this mechanism. Wipro Technologies, a large Indian offshore provider with offices in India and China, has developed its own EOF. The IT service delivery capacity of Wipro and HCL allows them to set up sophisticated performance tracking and process monitoring systems. The strong ties between the client and Wipro have made it possible for Wipro, the IT major, to link its managerial structures seamlessly with its clients. An in-house system called Veloci-Q tracks the key success parameters of each team (cost, quality, productivity), down to each employee. Wipro’s clients are given access to this tracking system via extranets that allow them to drill deep into Wipro and monitor the progress of their projects. Other firms, such as HCL (based in India), permit their buyers to use realtime monitoring mechanisms to track how specific functions are executed. I-OneSource, a BPO service provider firm based in India, is deploying similar mechanisms to track the progress of work that it does for offshore clients. Two Asian firms, Beredium International based in Mauritius (which executes outsourced processes for European firms) and ITOne based in Bangkok (which executes several outsourced services for its Thai clients), use mechanisms that allow them to track the key performance indicators in great detail and jointly monitor these with their clients. In all of these examples, where the service (a set of processes) executed by the provider is expensive to inspect or the quality of output is not transparent, buyers have set up real-time monitoring and managerial controls. For more routine tasks or tasks for which the cost of inspection is minimal, more conventional contracts with postinspection penalties are in place. The study of hybrid mechanisms is particularly interesting in services. Unlike in manufacturing, where raw material and physical products change form, in information work it is possible to redo a task or start a process again without incurring any costs other than labor-related ones. Buyers can then graft a layer of managerial control via interorganizational information systems and exert control across the boundaries of the firm without waiting for a process cycle to be completed. These controls have created the extended organizational form, a hybrid organizational form that has the following features: —The buyer contracts the production of process to the provider. —The buyer can inspect the provider’s output quality after production.
Ravi Aron and Ying Liu
393
—The buyer’s managers can also exercise partial managerial control over the provider’s agents by monitoring the quality of process execution during production. The objective of managerial control is to diminish operational risk in the production of services by an offshore provider. The buyer’s managers exert an additional layer of control that amplifies the returns to the provider’s effort. The natural question here is whether the EOF is an efficient governance structure. Does it improve the overall efficiency and thereby lead to welfare gains? We compared the efficiency of the three principal modes of governance in the production of financial services offshore: (1) in-house production of services offshore, also called a “captive center”; (2) offshore outsourcing of services to a BPO firm; and (3) the EOF in which an offshore provider is monitored and managed by the buyer’s managers. We selected two kinds of services: those that are highly codifiable and for which the codifiability score was between 6 and 7, and those for which the work was characterized by low levels of codifiability (with a codifiability score between 3 and 4 on a scale of 1 to 7). For each of the two process types we selected firms that were executing nearly identical processes in that category and collected panel data on production costs and production quality data and compiled their total cost of process production and quality of output over a period of time.10 Figures 2, 3, 4, and 5 show the trends. Highly Codifiable Processes Figures 2 and 3 compare the efficiency of the three forms of governance. The captive center produces the highest quality of output. If we ignore the total costs of production, it is clear that the captive center is the best option. However, over a period of time the quality of process output of the EOF nearly equals that of the captive center. The BPO provides just as much quality as specified in the contract (or slightly more than the contract-specified level). Providing higher levels of quality is costly and would reduce the BPO’s profits. The situation changes when we consider the total costs of production—that is, the welfare levels of the three forms of governance (see figure 5).11 While the captive center starts out being the most effective form of governance, over time the EOF is the most efficient and the captive center the least efficient. The EOF is able to combine the strengths of the BPO (cost efficiency, due to its profit 10. In return for receiving the data, we agreed not to furnish any details of the firm that might identify the firm or any of its clients. 11. Total costs of production = cost of production + managerial oversight + errors detection and correction + costs of errors that go undetected.
394
Brookings Trade Forum: 2005
Figure 2. A Comparison of Quality of Process Output for Highly Codifiable Outsourced/Offshored Work under Three Forms of Governancea Process quality (percent adequate) 97 CC 96 95 EOF
94
BPO
93 92 91 90 89 2
3
4
5
6 7 8 9 10 11 Months since inception of work
12
13
14
15
Source: Data collected by the authors. a. EOF: extended organizational form; BPO: business process outsourcing; CC: captive center.
focus) with those of the captive center (managerial control and focus on quality of production). The EOF thus produces the highest levels of welfare. Low Codifiability Processes For processes that are difficult to codify, the work done by agents calls for judgment, interpretation, and expertise based on a formal body of knowledge (such as that possessed by financial analysts, tax accountants, and other experts). Again it is clear from figures 4 and 5 that the output quality of the captive center exceeds that of the BPO and the EOF. Over time the gap diminishes, and as before, the EOF’s output quality improves and the BPO’s output quality flattens out once it reaches the contract-specified level. When efficiency levels are compared, an interesting picture emerges: The captive center starts out being the most efficient structure, while the EOF displays significant variations. There is a period during which the EOF produces gross inefficiencies. This is the period when the agents of the provider are adjusting their work style to match the expectations of the managers of the buyer. Since the work is not easily codifiable, managers of the buyers are involved in hands-on management of the
Ravi Aron and Ying Liu
395
Figure 3. A Comparison of Efficiency for Highly Codifiable Outsourced/Offshored Work under Three Forms of Governancea Total costs per process cycles (U.S. dollars)
550 BPO 500
CC
450
400
EOF
2
3
4
5
6 7 8 9 10 11 Months since inception of work
12
13
14
15
Source: Data collected by the authors. a. EOF: extended organizational form; BPO: business process outsourcing; CC: captive center.
provider’s teams. As a result the buyer’s agents underperform: they have to meet the expectations of their own managers (employed by the BPO) and those of the client. The client’s managers too learn to calibrate their expectations better over time; they start out expecting the same level of quality as that provided by their employees and then realize that this will take time. In the interim, as the two sides perfect their terms of engagement through a process of trial, error, and resetting of expectations, the costs increase, producing low levels of efficiency. Over time, however, the two sides reach a steady state of performance, and the EOF produces sharp gains in welfare. Complexity Arbitrage and the Offshore Outsourcing of Services While divergence in perceptions of complexity offers buyers and sellers the opportunity to maximize their relative comparative advantage, what really determines the extent of gains from offshoring is the ability to measure, manage, and control operational risk associated with offshore outsourcing. Our analysis shows that where it is possible to codify work processes, U.S. and European
396
Brookings Trade Forum: 2005
Figure 4. A Comparison of Quality of Process Output for Outsourced/Offshored Work with Low Codifiability under Three Forms of Governancea Quality of execution High 6 5
BPO
4 CC
3 2
EOF
1 Low 1
2
3
4 5 6 Quarters since inception of work
7
8
9
Source: Data collected by the authors. a. EOF: extended organizational form; BPO: business process outsourcing; CC: captive center.
firms will try to benefit from complexity arbitrage with third-party firms in countries such as China, India, the Philippines, Singapore, and Mauritius. For such processes, the optimal governance structure is an EOF with stable monitoring systems built into the contract. Where establishing EOFs proves to be expensive, the buyer should look to third-party providers of services offshore rather than set up captive centers. Managers of the buyer’s firm should invest in technologyenabled mechanisms that help them inspect the quality of process production and enforce contracts more effectively. For processes that are not highly codifiable—report generation in equity research and fixed income asset research, corporate finance, contract research support for investment banks, and risk analytics in insurance are examples— firms should either set up captive centers or transition their BPO contracts to EOFs. Ideally, firms should establish contractual mechanisms that allow the buyer’s managers to monitor and manage the provider’s agents. Over time, these structures yield the best results. Managers of these firms should invest in technology-based monitoring mechanisms and establish control structures that allow them to intervene in real time and correct processes that are being executed at less than optimal quality. The monitoring systems established by OT
Ravi Aron and Ying Liu
397
Figure 5. A Comparison of Efficiency for Outsourced/Offshored Work with Low Codifiability under Three Forms of Governancea Total cost per cycle (execution + quality) (U.S. dollars) EOF
6,000
CC BPO
4,000
1
2
3
4 5 6 Quarters since inception of work
7
8
9
Source: Data collected by the authors. a. EOF: extended organizational form; BPO: business process outsourcing; CC: captive center.
and Wipro represent the kinds of mechanisms that are likely to be very effective in global sourcing of services.
Conclusion We investigated the drivers of operational risk in the offshore execution of business processes and disaggregated operational risk into the different elements that constitute it. Our survey spanned several countries over several years, and we drew on data from a diverse array of industries and functions. We used our concept of a knowledge continuum to show how the two principal drivers of risk, process codifiability and process quality metrics, affect the nature of work along that continuum. Further, our econometric model of operational risk provides robust evidence in support of our findings. Our survey of the different forms of governance found that a hybrid form of governance that combined the market and hierarchy, which we call the “extended organizational form,” produced the best results.
398
Brookings Trade Forum: 2005
References Anderson, Erin, and David C. Schmittlein. 1984.“Integration of the Sales Force: An Empirical Examination.” Rand Journal of Economics 15 (3): 385–95. Bakos, Yannis, and Erik Brynjolfsson. 1993. “From Vendors to Partners: Information Technology and Incomplete Contracts in Buyer-Supplier Relationships.” Journal of Organizational Computing 3 (3): 301–28. Coase, Ronald. 1937. “The Nature of the Firm.” Economica 4 (16): 386–405. Crocker, Keith J., and Scott E. Masten. 1988. “Mitigating Contractual Hazards: Unilateral Options and Contract Length.” Rand Journal of Economics 19 (3): 327–43. Grossman, Sanford. 1986. “The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration.” Journal of Political Economy 94 (4): 691–719. Grossman, Sanford, and Oliver Hart. 1983. “An Analysis of the Principal-Agent Problem.” Econometrica 51 (1): 7–45. Hart, Oliver, and John Moore. 1988. “Incomplete Contracts and Renegotiation.” Econometrica 56 (4): 755–85. ———. 1990. “Property Rights and the Nature of the Firm.” Journal of Political Economy 98 (6): 1119–58. Holmstrom, Bengt, and Paul Milgrom. 1991. “Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design.” Special Issue, Journal of Law, Economics and Organization 7 (0): 24–52. Joskow, Paul L. 1987. “Contract Duration and Relationship-Specific Investments: Empirical Evidence from Coal Markets.” American Economic Review 77 (1): 168–85. Klein, Benjamin, Robert G. Crawford, and Armen A. Alchian. 1978. “Vertical Integration, Appropriable Rents and the Competitive Contracting Process.” Journal of Law and Economics 21 (2): 297–326. Levy, Frank, and Richard J. Murnane. 2004. The New Division of Labor. Princeton University Press. Masten, Scott. 1984. “The Organization of Production: Evidence from the Aerospace Industry.” Journal of Law and Economics 27 (2): 403–17. Monteverde, Kirk, and David Teece. 1982. “Supplier Switching Costs and Vertical Integration in the Automobile Industry.” Bell Journal of Economics 13 (1): 206–13. Palay, Thomas M. 1984. “Comparative Institutional Economics: The Governance of Rail Freight Contracting.” Journal of Legal Studies 13 (2): 265–87. Wharton Gartner Research Forecast. 2003. BPO Market to Grow to $173 Billion in 2007. Stamford, Conn.: Gartner Inc. Williamson, Oliver E. 1979. “Transaction-Cost Economics: The Governance of Contractual Relations.” Journal of Law and Economics 22(2): 233–61.
ASHISH ARORA Carnegie Mellon University
The Emerging Offshore Software Industries and the U.S. Economy
T
he possibility that a significant number of skill-intensive service jobs might move from the United States to developing countries has generated widespread recent concern. The newness of the phenomenon has led to rampant speculation, by academics, in the press, and on Wall Street, about the potential scope of such “offshoring” and its likely impact on the U.S. economy.1 Drawing from Arora and Gambardella (2004 and 2005), this paper attempts to shed some light on the development of offshoring by focusing on one of the first skill-intensive industries to move to low-wage economies, namely software. I first provide an overview of the global software industry. Next I discuss how three emerging-market countries—India, Ireland, and Israel (the “3 I’s”)—separately developed dynamic software industries that in 2001 together accounted for more than $15 billion in exports. In the next section I examine the impact of this offshoring on the U.S. economy, as well as future prospects for offshoring software R&D, concluding that this is unlikely. Finally, I compare the software industry’s experience with that of other skill-intensive industries.
The Global Software Industry The first task in describing the global software industry is to define what is meant by software industry. At the risk of oversimplification, software-related Prepared with the assistance of Gabriel Chodorow-Reich. 1. “Offshoring” in this paper refers to offshore outsourcing, or the moving of functions previously performed in the same country as the product market to an arms-length contractor operating outside the nation’s borders. “Outsourcing” refers to moving functions previously performed
399
400
Brookings Trade Forum: 2005
Table 1. Software Employment by Industry in the United States, 2001 Industry Computer equipment Computer and software services All other industries Total
Employment (thousands) 72 977 2,816 3,865
Source: Based on 2001 national 2-digit SIC Industry-Specific estimates (data from Bureau of Labor Statistics, www.bls.gov/oes/ oes_dl.htm [October 24, 2005]). I have included in the definition of software workers the following occupational categories: Computer and Information Systems Managers; Computer and Information Scientists, Research; Computer Programmers; Computer Software Engineers, Applications; Computer Software Engineers, Systems Software; Computer Support Specialists; Computer Systems Analysts; Database Administrators; Network and Computer Systems Administrators; Network Systems and Data Communications Analysts.
activities generally fall into one of three categories: design, coding, or maintenance. Design, which translates approximately into R&D and product development, has the highest value added of the three activities. Coding and maintenance may be thought of as analogous to production in other industries and consequently entail lower-end tasks. As discussed in more detail below, most of the functions offshored (especially to India) involve production, while design has tended to remain local. A correlate distinction concerns the difference between people who work in the software industry and those who work with software. In the United States in 2001, for example, two out of three people who wrote computer code worked for companies outside the computer equipment or software services industries (see table 1). These people work in universities, for financial funds, and in myriad other sectors that rely heavily on information and computer technology (ICT). Moreover, the bulk of the demand for computer-related outsourcing comes from firms outside the “core” industry. Roughly speaking, banks outsource coding and maintenance; Microsoft keeps its product design activities in-house. Thus, in studying the impact of offshoring, the first place to look is in software occupations outside the software industry. Of course, the core industry is still quite large, and it is also the appropriate level of analysis for describing the rise of providers in low-income countries that focus exclusively on software-related services. Table 2 gives summary statistics by country for five “newcomers” to the software development process as well as for the three most important industrialized countries. The United States dominates the global market; it had $200 billion in sales in 2002. It is important to note that this figure excludes software written and used in-house. Although in-
in-house to an arms-length contractor, whether or not the same contractor is located in the same country or even on-site.
Ashish Arora
401
Table 2. Summary Statistics for the Software Industry, 2002 or latest available figures
United States Japana Germanyb Brazilb China India Ireland Israelb
Sales (billions of dollars)
Sales/ GDP (percent)
Exports as percentage of total sales
Employment (thousands)
Sales per worker (thousands pf U.S. dollars)
200 85 39.8 7.7 13.3 12.5 13.9 4.1
2.0 2.0 2.2 1.5 1.1 2.5 11.4 3.7
n.a. n.a. n.a. 1–2 11 80 85 70
1,024 534 300 160a 190a 250 27.9 15
195.3 159.2 132.7 45.5a 37.6a 50.0 498.2 273.3
Source: Various; see also Botelho, Stefanutu, and Veloso (2005). Notes: Data for Ireland may be unreliable due to multinationals taking advantage of low taxes to increase the share of revenues shown as originating in Ireland. a. Data for 2000. b. Data for 2001.
house-produced software is difficult to measure, estimates suggest that, if included, it could constitute an additional $300 billion in value. Four of the five newcomers have similarly sized software sectors, as measured by sales revenue; Israel’s sector is about half their size (see table 2).2 However, three of the newcomers, India, Ireland, and Israel, earn roughly threequarters of their revenue from exports. Accordingly, the 3 I’s have received most of the attention as offshore locales. Documenting how these three countries have developed their industries constitutes most of the following section. First, two more comments are in order on the present state of the software industry. The first comment concerns the other two newcomers, China and Brazil. These countries have pursued a very different model for development from the export-led approach employed by India and Ireland and, to a lesser extent, Israel. Brazil has relied on a very sophisticated banking system to generate demand for an endogenous industry that it hopes will become internationally competitive.3 Such a strategy runs the risk of shaping an industry too closely harmonized with the idiosyncrasies of the Brazilian market, and the jury remains out on whether the Brazilian industry will achieve international viability. The Chinese model appears to have more traditional characteristics of import substitution, raising 2. Because of the high annual growth rates of the software industry in these countries, comparing data across years as presented in table 2 distorts the relative sizes of the sectors. When viewed for the same year, the sales for four of the countries are almost identical, with Israel producing about half the output. 3. See Botelho, Stefanutu, and Veloso (2005).
402
Brookings Trade Forum: 2005
questions about its long-run position.4 In sum, even as they have fostered domestic software industries, the future prospects for countries with inward-looking sectors appear uncertain. The second comment relates to the significant disparity in sales per worker (again see table 2). This disparity reflects a basic economic fact: the more laborabundant countries have cheaper labor and therefore use more of it, driving down revenue per worker. Since most producers in the industry operate in a very competitive market, the prices underlying the revenue figures generally reflect marginal cost. The three aspects of the global software market highlighted in this section will come up again in the rest of the paper. First, most of the jobs moving overseas are for in-house code writers in companies outside the core software industry. Second, the United States still dominates the global market for production of computer and software services. Finally, three lower-wage countries have emerged recently as players in global software trade, with a combined $15 billion in exports. It is to the rise of these “software tigers” that I now turn.
How India, Ireland, and Israel Developed Software Industries How did three lower-wage countries develop internationally competitive, high-skilled industries? Two complementary reasons suggest themselves, each of which is developed further in this section. First, however, a more in-depth look at the emergence of these countries’ industries is in order. Figure 1 shows the evolution of the share of software production for export in each country from 1991 to 2002. Both India and Ireland began their development with export shares of around 50 percent, indicating a high export orientation, while Israel started out a little more inward-looking before rapidly catching up after 1995.5 Still, even Israel began with a relatively large export share in comparison with the shares of China and Brazil by the early 2000s. Important differences also exist among the industries that have developed in the 3 I’s. Returning to table 2, Israel’s sales per worker exceed those of Japan, Germany, and the United States, suggesting that the Israeli industry does highvalue-added activities such as product development and R&D. The Indian figure,
4. Arora and Gambardella (2004). 5. The presence of an already successful hardware industry in Israel as well as large national defense needs initially served as domestic motivators for software production. Clearly, though, the industry’s expansion has coincided with a shift toward outward orientation.
Ashish Arora
403
Figure 1. India, Ireland, and Israel: Software Export Shares, 1991–2002 Percent
0.8 0.7 India 0.6 0.5
Israel Ireland a
0.4 0.3 0.2 0.1 1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
Source: Arora and Gambardella (2005). a. Between 1991 and 1997 data for Ireland were available only for alternate years.
meanwhile, is at the low end of the countries presented, reflecting the heavy service component and low-value-added nature of the industry there. Finally, the Irish industry combines product-oriented firms, some consulting, and large multinationals.6 The tendency of large multinational firms to overstate profits in Ireland to take advantage of the country’s low corporate taxes accounts for the seemingly impossible sales per worker number in table 2. These differences, however, mask strong similarities in what allowed the industries to develop. The first explanation for the rise of software industries in the 3 I’s relates to their excess capacity of skilled labor over demand. All three countries experienced somewhat difficult growth periods just before the software boom. For India, per capita GDP growth averaged barely 2 percent per year from 1970 to 1990, while Ireland and Israel grew at 2.9 percent and 2.4 percent, respectively, over the same period.7 Yet for reasons that are not entirely clear, all three countries continued to invest heavily in human capital. As a result, when global IT demand jumped in the late 1980s, these countries had a large supply of underemployed, highly skilled people. 6. See Arora and Gambardella (2005). 7. United Nations statistics at unstats.un.org.
404
Brookings Trade Forum: 2005
Table 3. Distribution of European Union Structural Funds, 1989–99 Percent Human resources Country Greece Spain Ireland Portugal Italy Average EU11
Infrastructure
1989–93
1994–99
1989–93
1994–99
25.6 24.2 38.0 26.1 21.6 29.6
24.6 28.4 43.9 29.4 21.4 29.8
40.9 54.0 27.7 29.2 38.7 35.2
45.9 40.4 19.7 29.7 29.8 29.5
Source: First Report on Economic and Social Cohesion 1996 DG XVI EC Brussels (from Sands 2005).
Since the onset of the software boom, their commitment to human capital has continued. Ireland provides one model for sustaining its investment, relying on government funds to improve advanced education. Table 3 compares Ireland’s use of EU Structural Funds with that of some of its peers. From 1989 to 1999 Ireland devoted more than a third more of its allotment to human resources than the EU11 average. Ireland’s efforts have made it the third-highest-ranking country by tertiary degrees per capita, of which 30 percent are in science and technology. Israel has followed largely the same model of government investment, much of it related to the country’s defense needs.8 India, by contrast, has used mostly private investment to finance a remarkable rise in engineering capacity. Figure 2 shows the capacity for bachelor’s degrees in engineering over the period 1987–2003; IT degrees have increased more than tenfold. Private colleges that do not receive government funding or grants account for most of this increase. The willingness of students to pay full tuition for such degrees is a good indicator of the perceived return to human capital investments in India. The second component of the emergence of software industries in the 3 I’s involves successful management at the level of the firm. Recent work by Ricardo Hausmann and Dani Rodrik (2003) emphasizes the importance of “self-discovery” by firms, as they explore new areas in which an economy might have a comparative advantage. This process has played a key role in the development of the software industry. A few examples may help to illustrate the point. Firms in each of the 3 I’s have exploited strong links with the regions to which they export. The most obvious example is Ireland’s participation in the European Common Market. Multinational firms in particular have taken advantage of this relationship by locating their operations in Ireland, which has a rel8. On Ireland, see Sands (2005); on Israel, see Breznitz (2005).
Ashish Arora
405
Figure 2. Engineering Education Capacity in India Engineering degrees (B/tech/BS), by field and year Thousands IT 300
341
All engineering
258
200 90
100
61
47
25 1987–88
1995–96
2002–03
Share of privately financed colleges Percent private 1995–96
1987–88
2002–03
100 80 60 40 20 East
Central
North Region
West
South
Sources: Author’s calculations from All India Council for Technical Education (AICTE) data.
atively low-wage workforce and favorable tax laws. In addition, all three countries have strong ties to the U.S. market through emigrants (again related to the abundance of skilled labor). The large diasporic populations facilitate connections between the export firms and their product markets and create a natural inflow of expertise when expatriates return from stints with companies abroad.9 In one survey of Irish entrepreneurs, Sands (2005) found that two-thirds of the founders of software firms in her sample had worked in another country. 9. Kapur and McHale (2005).
406
Brookings Trade Forum: 2005
Production decisions at the firm level have also strengthened the industry. Here the most relevant example comes from India. Despite pressure in the late 1990s to move to higher-value-added production, the leading Indian firms instead took advantage of economies of scale and expanded horizontally to encompass a larger share of low-end production, expanding operations to more countries and taking on larger and more complex, though not necessarily technically more demanding, projects. Along with an abundance of high-skilled, underemployed workers, then, firm-level strategies proved central to the growth of export-oriented software industries in the 3 I’s.
Will the Labs Follow the Mills? The emergence of skill-intensive industries in relatively low-wage countries has provoked fears about its effect on high-skilled workers in the United States. One concrete way to study this issue is to pose the following question: does the emergence of viable software industries in developing countries threaten the technical leadership of the United States in software? Put another way, will higher-value-added processes such as R&D and product design follow coding and maintenance overseas? The short answer is no, for at least three reasons. The United States is the world’s largest software market. The lead users, especially of business software, are American firms. Much high-end software production, such as design, occurs with a specific end user (a company) in mind. The relationship between designer and user requires that the software designer have intimate knowledge of the company’s needs. In many instances, developing this knowledge involves frequent meetings with representatives of the business, a problem when the designer has moved overseas. Moreover, firms desire a long-term relationship with the software designer. Businesses tend to use code for a substantial period of time (think of the green screens still employed in some stores, and the thirty-year-old software used by the Federal Aviation Administration to track airplanes). Facing the prospect of using the same code for thirty or forty years, and anticipating the need for updates and fixes by the designer, firms want assurances that the designer they choose will both provide high quality and be around for a while. This familiarity requirement works against designers located abroad.10 10. The company Check Point provides an excellent illustration of this argument. The company, which provides computer security (firewalls and antivirus software) to corporations, originated as an Israeli firm but moved to the United States in order to be closer to its clients.
Ashish Arora
407
Table 4. Number and Characteristics of Selected Foreign-Born Populations in the United States Aged 25 and Over
Country of origin
1990
2000
Percentage of 2000 population entering post-1990
India Brazil China
304 54 405
837 154 847
55 49 66
Population (thousands)
Educational attainment, 2000 (percent) Primary
Secondary
Tertiary
5 9 20
15 36 26
80 55 54
Country of origin (percent) Year of entry
India
Ireland
Israel
Before 1960 1960 –69 1970 –79 1980 –89 1990 –95 1996–2001
1 3 14 24 23 36
32 19 8 23 13 5
4 1 28 35 18 14
Source: Kapur and McHale (2005) based on the 2000 census and March 2001 Current Population Survey (Washington: Bureau of Labor Statistics).
The status of the United States as an attractor of high-end human capital also suggests the country will retain its lead in high-value-added production. Table 4 shows some characteristics of foreign-born populations in the United States. Between 1990 and 2000, the over-25 Indian-born population increased by more than 200 percent. Eighty percent of these Indians had earned tertiary (collegelevel) degrees, whereas the U.S. average is 50 percent. This trend indicates a growing U.S.-based high-skill sector, not a world where software designers are fleeing the United States. Domestic institutions constitute a final reason why we should not expect the United States to suddenly lose its technical advantage. As Daniel Trefler points out in this volume, developing countries remain far behind the United States in forming the types of institutions that encourage innovation. For software development, venture capital remains an especially important component in the creation of new firms. Although foreign firms can in theory tap U.S. venture capitalists (as Israeli firms have begun to do), the well-known home bias in investment suggests that firms in the United States will retain an advantage. For all these reasons, the U.S. technical leadership in software appears safe, at least over the medium term.
408
Brookings Trade Forum: 2005
Conclusion I close by returning to the question of what lessons the software industry’s experience provides concerning the prospects of offshoring high-skilled jobs more generally. It took an extraordinary confluence of events for export-oriented software industries to develop in India, Ireland, and Israel. The abundance of highly skilled and underemployed workers, key innovations by firms, and the good fortune to have these conditions in place at the onset of the global IT boom all contributed to successful sectors. Other countries will have a difficult time replicating these circumstances, while in many cases having to overcome the additional challenge of not having an English-speaking workforce. The software experience suggests general difficulties for other countries in attracting offshore industries. But I see the broader implications for offshoring of high-end activities in particular as less certain. My conclusion that software design and R&D are unlikely to move offshore depended in large part on their requirements for proximity to the product market. However, there are a number of high-tech industries where this requirement need not exist. In drug development, for example, a firm could just as easily research a cancer drug for the American market using overseas scientists and subjects. For some skilled-laborintensive activities such as synthesis of molecules for testing as potential drugs, it is clear that the expertise and infrastructure available in places such as India and China is adequate. This type of R&D activity is much more likely to move abroad.
References Arora, Ashish, and Alfonso Gambardella. 2004. “The Globalization of the Software Industry: Perspectives and Opportunities for Developed and Developing Countries.” Working Paper 10538. Cambridge, Mass.: National Bureau of Economic Research. Arora, Ashish, and Alfonso Gambardella, eds. 2005. From Underdogs to Tigers: The Rise and Growth of the Software Industry in Brazil, China, India, Ireland and Israel. London: Oxford University Press. Botelho, Antonio J., Giancarlo Stefanutu, and Francisco Veloso. 2005. “The Brazilian Software Industry.” In From Underdogs to Tigers: The Rise and Growth of the Software Industry in Brazil, China, India, Ireland and Israel, edited by Ashish Arora and Alfonso Gambardella. Oxford University Press. Breznitz, Dan. 2005. “The Israeli Software Industry.” In From Underdogs to Tigers: The Rise and Growth of the Software Industry in Brazil, China, India, Ireland and Israel, edited by Ashish Arora and Alfonso Gambardella. Oxford University Press.
Ashish Arora
409
Hausmann, Ricardo, and Dani Rodrik. 2003. “Economic Development as SelfDiscovery.” Harvard University. Kapur, Devesh, and John McHale. 2005. “Sojourns and Software: Internationally Mobile Human Capital and the Software Industry in India, Ireland and Israel.” In From Underdogs to Tigers: The Rise and Growth of the Software Industry in Brazil, China, India, Ireland and Israel, edited by Ashish Arora and Alfonso Gambardella. Oxford University Press. Sands, Anita 2005. “The Irish Software Industry.” In From Underdogs to Tigers: The Rise and Growth of the Software Industry in Brazil, China, India, Ireland and Israel, edited by Ashish Arora and Alfonso Gambardella. Oxford University Press.
This page intentionally left blank
FRANK LEVY ARI GOELMAN Massachusetts Institute of Technology
Offshoring and Radiology It turns out that even American radiologists, with their years of training and annual salaries of $250,000 or more, worry about their jobs moving to countries with lower wages, in much the same way that garment knitters, blast-furnace operators and data-entry clerks do. . . . Radiology may just be the start of patient care performed overseas.
—andrew pollack “Who Is Reading Your X-Ray?” New York Times, November 16, 2003
T
he observation quoted above appeared when offshoring was rapidly entering the public consciousness. The author chose a dramatic example. At the time, economists argued that while technology made the offshoring of some work inevitable, the United States could still prosper with a sufficiently educated workforce. But skilled software jobs were already being offshored,1 and if offshoring could threaten the work of radiologists, an occupation requiring eight to ten years of post-college education, the economists’ argument appeared suspect. We argue that the situation of U.S. radiologists remains quite different from that of U.S. software engineers and other professionals whose jobs are being offshored. Media stories notwithstanding, only a tiny number of radiological images now are read by the medical equivalent of cheap foreign labor. These stories,
The authors wish to thank the Alfred P. Sloan Foundation and the MIT Industrial Performance Center for financial support, as well as Professor Howard P. Foreman of Yale University, Dr. Krishna Kandarpa of University of Massachusetts Memorial Hospital, Dr. Jonathan Sunshine of the American College of Radiology, Robert Hartman, and Professors Joseph Newhouse and Dani Rodrik of Harvard University, among numerous other people, for helpful comments. Any errors in the paper are ours alone. 1. Scott Thurm, “Lesson in India: Not Every Job Translates Overseas,” Wall Street Journal, March 3, 2004.
411
412
Brookings Trade Forum: 2005
however, have a kernel of truth since it has become technically feasible to receive and interpret radiology images from a distance. The current lack of offshoring reflects economic, social, and regulatory conditions as much as immutable technical constraints. Although low-cost offshore radiology reads now occur only rarely, the horizon is less clear. The differences between radiology and software programming begin with the cognitive structure of the radiologist’s work, which makes it particularly timeconsuming to monitor. Because work quality is hard to monitor, a U.S. doctor who refers an image to an unfamiliar radiologist relies heavily on the radiologist’s credentials, in particular his or her U.S. board certification. One could imagine international agreements that allowed radiologists credentialed in one country to practice in another. In fact, no such agreements exist for radiologists practicing in the United States, a reflection, in part, of U.S. doctors’ group power. U.S. radiologists’ power to restrict foreign competition is reinforced by malpractice insurance, Medicare reimbursement regulations, and, in all likelihood, consumer preference. Software professionals have few of these protections and so face strong foreign competition. Conversely, these same factors do little to protect radiologists against other U.S. medical specialties, and interspecialty competition to read images is sometimes quite intense. We tell the story of offshore diagnostic radiology in five brief sections.2 We begin by sketching a cognitive framework to describe the jobs (of any kind) that are most easily offshored. We then situate the radiologist’s job in this framework. Next we describe the economic conditions and regulatory factors that define the current U.S. market for radiologists. In the fourth section, we describe the “nighthawk” radiology industry, the industry that handles most offshored U.S. radiology reads. We close by speculating how some parts of offshore radiology might come to resemble the offshoring described in the opening quote and the implications of such a shift for the cost of medical care.
Offshoring in Cognitive Terms Frank Levy and Richard J. Murnane (2004) have argued that there are broad similarities between the work most vulnerable to offshoring and the work most
2. Radiology has several subspecialties. Diagnostic radiologists read images and typically do not deal directly with patients. Interventional radiologists both read images and perform procedures on patients—for example, the image-guided insertion of a stent—and so their work is not at issue here.
Frank Levy and Ari Goelman
413
vulnerable to computer substitution. Their argument can be summarized as follows: All workplace tasks involve processing information: an engineer reading a report, a chef tasting a sauce, a farmer looking to the sky to check for rain, and so on. The tasks most vulnerable to computer substitution are those where the information processing can be described in rules. When a task can be fully described in rules, it can be programmed for a computer. When significant parts of a task can be described in rules, it is vulnerable to offshoring since it can be assigned to offshore producers with reduced risk of miscommunication and lower costs of monitoring. When a task’s rules cannot be articulated—when the task involves extensive tacit knowledge—neither computerization nor offshoring is a readily available alternative.3 The rules to which the argument refers can be either deductive or inductive. Deductive rules arise from the logical structure of the process—for example, the rules that describe issuing an airline boarding pass in a self-service kiosk (“Does this credit card number match a number in the reservation database? Yes/No”). This kind of information processing is often described as rules-based logic. Inductive rules, which are more complicated, typically refer to the equations of probits, neural nets, and other statistical models whose parameters are estimated on “training samples” of data before the model is put into use. Familiar examples include models estimated on credit card purchase histories to flag the possibility of fraud and speech recognition software for a personal computer that must be trained by the user before it is used. This kind of information processing is usually described as pattern recognition. We will use the term “pure pattern recognition” to describe information processing that is too complex—at least for the moment—to construct even inductive rules. Tasks based on pure pattern recognition arise at both the high and low ends of the skill distribution. It is hard to infer the rules involved in writing a convincing legal brief. It is equally hard to infer the rules a janitor uses to convert a two-dimensional array of photons on his retina into a three dimensional understanding of an unfamiliar room. From a cognitive perspective, pure pattern recognition also rests on rules, but the rules are too deeply embedded to articulate. Figure 1 illustrates this typology. 3. See also Autor, Levy, and Murnane (2003). A second characteristic required for computerization is that the information being processed can be digitized. For simplicity, we do not discuss that characteristic here.
414
Brookings Trade Forum: 2005 Figure 1 Varieties of Information Processing Increasingly Hard to Program or Offshore
Conscious application of deductive rules (rules-based logic) Examples —Issuing an airline boarding pass
Application of inductive rules (pattern recognition)
Rules cannot be articulated (pure pattern recognition)
—Speech recognition —Cancerous cell cluster recognition in mammograms
—Writing a legal brief —Making visual sense of an unfamiliar room
Within this framework, the same structure that makes a task easier to computerize often makes the task easier to send offshore. A computer executes rules: a task cannot be computerized without instructions for every contingency (including, in the case of the boarding pass kiosk, “Unable to Process Your Request – See Desk Agent”). Similarly, when a firm assigns a task to an offshore contractor, the transaction is much simpler when the firm can specify, step by step, how the task is to be done. The absence of such instructions risks quality problems and complicates the assignment of responsibility in the case of errors. Alternatively, the absence of instructions requires a higher level of contractor skill. Multiple examples illustrate this overlap between computerization and offshoring. Call center work is subsumed by speech recognition software, and call center work is sent offshore using operator-read scripts (a kind of rule). Doctors’ dictated case notes are sent offshore for transcription, but speech recognition software is subsuming that work too. Production of Boeing aircraft modules is sent offshore using digitized machine tool instructions;4 other production jobs are lost in this country to assembly line robots. Basic tax preparation is sent to offshore accountants (who use the tax code’s rules) while other tax preparation work is done by TurboTax and TaxCut software. As other countries gain in expertise (that is, tacit knowledge), the need for fully specified rules will decline. Offshored software work is an example of this transition, an occupation that involves tacit knowledge but rests on the rules of 4. The machine tool instructions come from the computer-assisted design software used to design the aircraft. See Levy and Murnane (2004, chap. 3) for further discussion.
Frank Levy and Ari Goelman
415
programming languages. In the near term, however, the rules-based nature of both computerized and offshore work is a reasonable characterization.
Where Radiology Reads Fit In In contrast to these examples, reading most radiological images is pure pattern recognition, work that so far defies easy characterization in rules. Several U.S. centers work on computer-assisted diagnostic (CAD) software to scan radiological images for abnormal patterns.5 To date, only two applications have received approval from the Food and Drug Administration: software to scan mammograms for potentially cancerous cell clusters and software to scan lung images for cancerous nodules.6 A third application—to scan virtual colonoscopies for polyps—is on the horizon. Most other images are currently too complex to infer the underlying rules. This is not surprising. An abdominal CT scan can reveal many different abnormalities, and even normal abdominal scans vary significantly among individuals. MRIs are particularly complicated. Where all of a telemarketer’s calls are similar, each radiological image is essentially a special case. In organizational terms, the radiologist is usually the agent of the referring doctor who ordered the scan. For the sake of the patient and to minimize malpractice issues, the referring doctor has to be confident that a radiologist reads images correctly, but the absence of articulated rules makes this hard to determine. The problem is not limited to radiologists many miles away. As a surgeon in a large Boston area hospital told us: “When I get night call from ‘Bob’ or ‘Jim’ [the attending radiologists in the emergency room] and they say I have to come in to operate, I come right in. When I get a night call from a radiology resident and they say I have to come in, I want much more information. I’ve been burned too many times by residents misreading a film” (personal communication, April 2005). Not all doctors face a surgeon’s decisions, but learning to trust a radiologist 9,000 miles away—a relationship with no face-to-face contact—is a potentially significant problem. Since the task is not defined by rules, the problem’s current solution is to rely heavily on the radiologist’s qualifications. The radiologist must be U.S. 5. The Kurt Rossmann Laboratory at the University of Chicago Medical School is one such center. Note that the radiologist must both recognize an abnormal pattern (which is what the software does) and identify the abnormality, a potentially harder job. 6. One developer of such software is R2 Technologies in Santa Clara, California; see www.r2tech.com.
416
Brookings Trade Forum: 2005
board-certified, which in turn requires the completion of a residency in an approved U.S. program.7 The radiologist must also be licensed in the state and accredited in the hospital in which the images were generated. As with most professional standards, these standards simultaneously address quality and limit competition.
The U.S. Radiologist Market U.S.-trained radiologists face a very tight market for their services. The website Salary.com reports a median cash compensation for diagnostic radiologists of $277,000 and 75th percentile compensation of $331,000. The RSNA News, a publication of the Radiological Society of North America, reports that median 2003 compensation for a diagnostic radiologist in a group practice was $346,000.8 Radiologists’ income has also increased faster than the income of many other doctors: between 1998 and 2000, for example, radiologists reported real annual increases of 9.1 percent in their income, while general internal medicine physicians saw a real annual decrease of 1.3 percent (Kane and Loeblich 2003). This tight market reflects both supply and demand factors. In classic cobweb fashion, today’s limited supply of radiologists has its roots in the mid-1990s. The Balanced Budget Act of 1997 capped the total number of a hospital’s residency slots that could receive federal support. At that time, relatively few residency slots were devoted to radiology because medical students perceived a weak job market for radiologists and federal policy was being designed around the primacy of the family practitioner (Grumbach 2002). Although it was unclear then, multiple factors were beginning to expand demand for radiologists’ services. Continued improvements in scanning equipment increased both the regions of the body that could be scanned and the number of images produced in a given case. One doctor notes that fifteen years ago, a CT scan could produce twenty “slices” (cross-sectional body images) in ten or fifteen minutes. Today a CT scan can turn out several hundred slices of similar resolution in less than a minute, producing “a flood of data to analyze” (personal communication, July 2004). As Goelman (2005) describes, imaging also became more important to diagnosis, in part due to concerns about malpractice liability. A final factor in increased demand for radiologists’ services was the rapidly expanding supply of scanning equipment. After brief governmental attempts to ration equipment pur7. But a U.S. residency does not require having attended a U.S. medical school, a point we return to below. 8. RSNA News, October 2004 (www.rsna.org/publications/rsnanews/oct04/salary-1.html).
Frank Levy and Ari Goelman
417
chases—through the “certificate of need” programs—many community hospitals now have the kinds of advanced equipment that used to be found only in regional medical centers. Several doctors have suggested to us that this increased access and convenience tilt the referring doctor’s decision toward ordering a scan. Taken together these factors improved radiologists’ job prospects and increased demand for radiology residencies. But under the federal funding cap, more radiology residencies required reallocation of slots from other specialties, an arduous process. In the 1990s, the software industry responded to a skill shortage by using H1B visas to import foreign labor.9 The U.S. health care system pursues a limited version of the same strategy. Each year, U.S. hospitals hire more residents than the number of students who graduate from U.S. medical schools. The system makes up the difference by hiring foreign-educated medical students who have passed a series of U.S. certification exams. Once through residency, these students are eligible to take U.S. board certification exams, with the result that one quarter of the physicians currently practicing in the United States received their medical training elsewhere in the world. While the cap on funded residency slots means that the increment of foreign students is more limited than previously, Mullan estimates that every year one-quarter of roughly 100,000 firstyear residents are foreign medical school graduates (Mullan 2002, 2003). From the perspective of offshore radiology, interviews conducted in India by our colleague Kyoung-Hee Yu indicate that the market for U.S. board-certified radiologists in India is tight largely because many Indian students who have received U.S. board certification choose to remain in the United States.10 Board certification itself points to the overriding difference between doctors and software professionals: U.S. doctors have a professional dominance in which doctors themselves are allowed to determine who qualifies as a doctor (Freidson 1970)—a power that software professionals, call center operators, and production workers do not have. As noted above, radiologists who read images generated in the United States must be board-certified, licensed in the state where they are working, and credentialed at the hospital in which they practice. While radiologists, like other doctors, detest malpractice litigation, fear of that litigation helps to enforce these requirements. A doctor who must defend his treatment in court does not want to explain why he referred an image to an uncertified or unlicensed radiologist. 9. The H1B visa is the primary USA work visa (permit) for foreign professionals who want to live and work in the United States. It is typically valid for up to six years. 10. Interviews conducted in 2005 by Kyoung-Hee Yu, a Ph.D. candidate in MIT’s Sloan School of Management.
418
Brookings Trade Forum: 2005
Outsourcing and Offshoring Radiology Reads Given these strict limits on who can read images, the economics of outsourced radiology involves not cheap labor, but economies of scale. Outsourced reads are sent to “nighthawk” radiology services (so named because their services are in the greatest demand at night), usually provided by a large hospital or a private firm. The typical client for nighthawk services is a small private radiology practice at a hospital with an emergency room that generates several images a night requiring immediate reads. A small private practice cannot afford a full-time night radiologist to read half a dozen images. Putting a daytime radiologist on night call—potentially waking him or her up at 3:00 in the morning—risks errors, reduces the number of the more remunerative day shifts the radiologist can work, and makes the practice a less desirable place to work at a time when the radiology market is very tight. In this situation, sending images to a nighthawk service is an attractive alternative. The nighthawk service, in turn, keeps its rates competitive by serving multiple hospitals in order to keep its radiologists busy for the duration of their shifts. All of the large hospitals that provide nighthawk services are located in the United States. The non-hospital nighthawk firms are all headquartered in the United States, but many locate their radiologists at offshore sites, including Sydney, Bangalore, Tel Aviv, and Barcelona. These remote locations allow radiologists to work daytime schedules as they read U.S. images generated at night. As one provider’s website says, “When it’s the middle of the night in Boston, it’s daytime down under.”11 To our knowledge, all nighthawk firms (and all hospitals) employ U.S. boardcertified and licensed radiologists. This is partially in order to reassure potential clients concerned about quality, but once again, malpractice insurance plays a role. A referring practice risks its own malpractice insurance if it uses a nighthawk firm that does not carry malpractice insurance, and a nighthawk firm cannot purchase malpractice insurance unless it can prove it uses board-certified and licensed radiologists. In her interviews, Kyoung-Hee Yu heard occasional stories of “ghost reads,” where individual U.S. doctors send films to uncertified radiologists abroad and then sign the reads themselves. This practice appears to be rare, in part because of the significant financial risk it entails. The first radiology firms that focused solely on remote nighttime readings opened in 2001. Since then they have grown rapidly, with the leading firms currently reading images from roughly 1,000 hospitals, almost 20 percent of the 11. www.nighthawkrad.net.
Frank Levy and Ari Goelman
419
5,764 hospitals registered with the American Hospital Association in 2003 (Goelman 2005). Given the recent inception of the market, it is unsurprising that the number of firms continues to fluctuate; but three firms now divide the majority of the market. Despite using board-certified radiologists, nighthawk services chiefly perform what are called “wet” or preliminary reads. These wet reads inform the treatment the emergency room patient is given that night. Then, the following morning, the referred images are given a second read—a “dry” read—by a staff radiologist at the referring practice who signs off on the report. The two-read system reflects both quality control and regulatory considerations, including the fact that Medicare will not reimburse medical procedures done outside the United States (another reason why U.S. doctors put their own names on “ghost reads”).12 In addition, it assuages any fear among physicians in the referring practice that they might lose control over their practice (Goelman 2005). Regulatory barriers aside, the supply of qualified Indian radiologists is uncertain. Kyoung-Hee Yu has collected indirect evidence on this point by assessing to what extent Indian radiologists are supplying radiology services to countries other than the United States. She reports that as of 2005, Indian firms are starting to negotiate entry into the British market and have been invited to begin discussions with Singapore. It is unclear how fast these activities can expand before running into supply constraints. Given this situation, U.S. radiologists are wary of foreign radiologists but, contrary to the opening New York Times quote, they currently do not see offshoring as their main competitive threat.13 That honor goes to members of other U.S. medical specialties: cardiologists who want to read heart images, obstetricians/gynecologists who want to read ultrasound images, and other specialists. Since this competition comes from certified and licensed U.S. physicians, neither existing professional requirements nor malpractice fears offer radiologists much protection. An example of these turf battles is the controversy surrounding a recent policy statement by Mark Miller, executive director of the Medicare Payment Advisory 12. This restriction was adopted a number of years ago to guard against, for example, people going to Mexico or Canada for treatment. 13. For example, a principal website for radiologists, www.AuntMinnie.com, holds an annual poll to choose “The Minnies,” the leading people and most significant developments and events in the field. In 2004, competition from foreign doctors was neither the first- nor second-ranked “biggest threat to radiology.” (It had been ranked second in the semifinal voting.) The Minnie for biggest threat went to “increased use of medical imaging by physicians in other specialties (turf battles),” discussed later in this section. See www.auntminnie.com.
420
Brookings Trade Forum: 2005
Commission (MedPAC), before the House Ways and Means Committee (Miller 2005). Medicare currently reimburses any physician for taking and interpreting images even if he or she has no training in image interpretation. Miller argued that some interpretation training should be required before a specialist—such as an orthopedic surgeon—could receive reimbursement. Miller based his recommendation on cost and quality grounds, but the recommendation was seen as favoring radiologists over other specialists. Michael Wolk, president of the American College of Cardiology, quickly challenged Miller’s statement, telling his membership: “Radiology leadership is directly challenging our ability to use the best and latest technology to care for our patients in our offices and in the hospitals in which we work.”14 Radiologists argue that these turf battles reflect a basic asymmetry. As they see it, radiologists’ patients are referred to them by other doctors. Other specialists who install scanning equipment in their offices can self-refer. Total imaging expenditures are now on a par with pharmaceuticals as drivers of rising medical expenditures,15 and many radiologists believe, perhaps incorrectly, that much of the growth reflects self-referrals done by other doctors to increase income (see, for example, Thorpe and others 2004). The MedPAC statement coincided with radiologists’ desire to avoid restrictions that are triggered by rising imaging costs. But recent evidence suggests that some radiologists have discovered how to self-refer by offering “discounts” (that is, kickbacks) to doctors who refer patients to them.16
The Future of Offshore Radiology How will offshore radiology evolve? A speculative answer begins by first reviewing how radiologists are paid. The insurance reimbursement for a scan typically involves two pieces. The professional fee covers the radiologist’s interpretation. The technical fee covers the cost of the scanning equipment, the technologist who operates the equipment, and so on. For example, at this writing in 2005, the Medicare schedule for the Boston metropolitan area reimburses a chest X ray (frontal and lateral) for $43.78, of which the professional fee is $12.36 14. As reported in Tracie L. Thompson, “Cardiology Leader Slams ACR Imaging Initiatives” (www.auntminnie.com [March 10, 2005]). See also Tracie Thompson, “U.S. Congress Hears Debate over Federal Imaging Standards” (www.auntminnie.com [March 17, 2005]). 15. Charles Stein, “Partners Program Aims to Rein in Skyrocketing Costs of Diagnostic Imaging,” Boston Globe, June 27, 2003. 16. See David Armstrong, “MRI and CT Centers Offer Doctors Way to Profit on Scans,” Wall Street Journal, May 2, 2005.
Frank Levy and Ari Goelman
421
(28 percent). A CT scan of the spine, including a contrast agent, receives reimbursement of $424.49, of which the professional fee is $68.92 (16 percent). An MRI of the spinal canal with contrast agent is reimbursed at $785.16, of which the professional fee is $103.33 (14 percent).17 Although the reimbursement per read is modest, a private practice radiologist (one who does not teach or do research) can read in excess of 11,000 studies per year, and thus his or her potential income is quite large. From an insurer’s perspective, the cost of interpretation is a relatively small share of the scan’s total cost, particularly for the more expensive CT or MRI. Correspondingly, an insurer seeking to control aggregate imaging costs would likely focus on limiting the number of scans through benefits management before they would consider mechanisms for hiring cheaper radiologists. Ironically, it is doctors themselves—both nonradiologists and radiologists—who may have the greater incentives to explore offshore radiology. Because doctors often own their own scanning equipment, they can collect technical fees regardless of who interprets the image. At one extreme, some self-referring specialists (nonradiologists) may recognize they are billing for images that they cannot accurately interpret and may turn to non-board-certified foreign radiologists for assistance. At the other extreme, some radiology practices may openly work to certify offshore radiologists to handle low-profit work. Today, for example, Medicare reimburses $101.65 for a screening mammogram, including a professional fee of $39.61. For many private practices, this fee makes the screening mammogram a loss leader: a frequently requested scan that can displace more profitable work but must be offered as part of a full array of services. More precisely, the large volume of normal screening mammograms is a loss leader: abnormal screening mammograms can lead to additional scans that can generate a profit. In this situation, many U.S. radiologists might welcome a mechanism that would triage the normal scans, allowing the radiologist to focus on the abnormalities. Such triage may be plausible. A mammogram is scanned for only a limited number of abnormalities (though these abnormalities, such as microcalcifications, are very small and often hard to detect). This helps explain why mammograms are one of the two kinds of images receiving FDA approval for computerized scanning. As one radiologist suggested, it is possible to imagine a private offshore firm that offers to screen mammograms twice—once by an offshore 17. Information available from the Center for Medicare and Medicaid Services (www. cms.hhs.gov/regulations/pfs/2005/1429fc.asp).
422
Brookings Trade Forum: 2005
radiologist and again by computerized scanning—flagging those mammograms that show any sign of abnormality. The arrangement might build on previous, unsuccessful attempts to redefine offshore foreign radiologists as “virtual residents”—a reference to the way that radiology residents in U.S. hospitals (who are not yet board certified) do supervised reads as part of their training. The redefinition would require a significant institutional shift, including coordination with U.S. malpractice insurance. Given the status of radiologists however, this scenario is one of the easier offshoring scenarios to imagine.
Conclusion In the first news reports about the offshoring of high-skilled jobs, radiologists were often paired with software engineers. We have argued in this paper that these occupations, while both high skilled, are distinguished by several key differences. Unlike the work of software engineers, the quality of a radiologist’s work is very difficult to gauge. This has contributed to the radiology profession’s continuing ability to exercise power over decisions about who is permitted to interpret U.S. radiological images. The radiologists’ control of their profession is reinforced by malpractice concerns that in part reflect U.S. consumer preferences. U.S. consumers may not care who wrote the code in their PC, but they do not favor having medical treatments influenced by an anonymous benefits manager or an anonymous foreign doctor. In most markets, consumer behavior is determined by price as well as preferences, but U.S. health insurance offers little price incentive for consumers to reconsider their view. For all these reasons, pairing the threats faced by radiologists and software engineers makes for a terrific headline but not much else. References Autor, David, Frank Levy, and Richard J. Murnane. 2003. “The Skill Content of Recent Technical Change: An Empirical Investigation.” Quarterly Journal of Economics 118 (November): 1279–1334. Freidson, Eliot. 1970. Professional Dominance: The Social Structure of Medical Care. New York: Atherton. Goelman, Ari. 2005. “What’s Space Got to Do with It?” Ph.D. dissertation, Department of Urban Studies and Planning, Massachusetts Institute of Technology. Grumbach, Kevin. 2002. “Fighting Hand to Hand over Physician Workforce Policy; The Invisible Hand of the Market Meets the Heavy Hand of Government Planning.” Health Affairs 21 (5): 13–27.
Frank Levy and Ari Goelman
423
Kane, Carol, and Horst Loeblich. 2003. “Physician Income: The Decade in Review.” In Physician Socioeconomic Statistics, edited by John Wassenaar and Sara Thran, pp. 5–11. Chicago: American Medical Association. Levy, Frank, and Richard J. Murnane. 2004. The New Division of Labor. Princeton University Press and the Russell Sage Foundation. Mullan, Fitzhugh. 2002. “Some Thoughts on the White-Follows-Green Law.” Health Affairs 21 (1): 158–59. ———. 2003. “The Future of Medical Education: A Call for Action.” Health Affairs 22 (4): 88–90. Miller, Mark E. 2005. “MedPAC Recommendations on Imaging Services.” Statement prepared for the Ways and Means Committee, U.S. House of Representatives, March 17 (http://www.medpac.gov/publications/congressional_testimony/031705_ TestimonyImaging-Hou.pdf). Thorpe, Kenneth E., S. Curtis and Florence and Peter Joski. 2004. “Which Medical Conditions Account for the Rise in Health Care Spending? Health Affairs, Web Exclusive, August 25 (http://content.healthaffairs.org).
Discussion
Despite the fact that it came at the end of a very long day, this panel generated a very active discussion. Nearly all speakers commended the panelists for their richly informative and thought-provoking presentations. The first commentators related the industry experiences back to points made in the theoretical sessions of the conference. Lael Brainard asked whether institutional development constrains offshoring in reality or whether this is simply a theoretical hypothesis. In this context, she pointed out that the discussion of India had focused on labor supply not institutional constraints. Alan Deardorff suggested that the “complexity of the symbolizers” was one interpretation of the K in Jim Markusen’s model. Reflecting on Frank Levy’s discussion of radiology, Brainard expressed the view that professional licensing is one of the channels through which protectionism seems most likely to appear. She also suggested that the skill requirements for doing particular kinds of work help to explain the traditional localization of services in the United States. Levy responded that, in the case of radiology, professional standards reflect the difficulties of acquiring and publicizing information about skills. Thus it is very hard to tell whether the restrictions are simply guild protection. Much of the discussion related to the issue and implications of task complexity. Catherine Mann asked how Ravi Aron’s point about differences in views of what is complex relates to the work Frank Levy has done with Richard Murnane on which tasks can be codified. Levy noted that this earlier work focused on which tasks can be easily done by computer. If people have different skills in different countries, the implications for offshoring are not straightforward. Aron explained that his research has examined 316 specific processes, studying which 424
Aron and Liu/Arora /Levy and Goelman: Discussion
425
can be codified. He sees his findings as closely related to Levy’s distinction between deductive and inductive ways of specifying work. But perception matters too. What makes work offshorable, he said, is whether it can be described by codifiable rules. But what makes managers perceive a task as difficult (or complex) is whether they believe they have workers who can perform that task successfully. Furthermore, their research suggests that when U.S. managers offshore work they consider complex (such as number crunching), they find fewer errors and higher quality than they expected. But when they offshore work they consider easy (such as writing a corporate banking report), they experience lower returns than expected—sometimes negative returns. Thus, his analysis suggests that U.S. and British companies are paying a premium (14 to 25 percent) for work that they consider complex. Martha Laboissiere reported on some of the findings from ongoing work at the McKinsey Global Institute (MGI) to study both the supply and the demand sides of offshoring. They study a set of representative sectors, including packaged software, as well as health care and pharmaceuticals. Like Brad Jensen and Lori Kletzer, they also examine occupational breakdown. They find the three main factors that inhibit offshoring are the need for a physical presence, the need for local knowledge, and the existence of complex interactions. The latter is most important at the upper management level. She noted that complex interactions are also reflected in the agglomeration effect raised by Robert Lawrence in his discussion of the Jensen-Kletzer paper. This helps to explain why so many IT service companies moved to Silicon Valley in California and to Bangalore, the even bigger Silicon Valley of India. Similar concentrations are evident in the pharmaceuticals industry. Thus she stressed the distinction between process complexity confined to a single company and network complexity between companies, which adds another layer and is also important. Finally, she reported MGI estimates that only about 3 percent of occupations in health care are potentially offshorable, while 49 percent are potentially offshorable in packaged software. She saw differences in complexity as the main reason these estimates are so different. Lori Kletzer asked whether cross-country differences in judgment about what constitutes complexity correlate with international rankings of numeracy. She noted that the United States ranks relatively poorly, and that she believed India ranked somewhat more highly. Aron replied that it is important to consider selection issues. Roughly half of the Indian population is not literate. But he believes there is a strong bias among the Indian middle class toward occupations that call for mathematic and scientific skills. Numeracy among this group is probably quite high.
426
Brookings Trade Forum: 2005
There was also discussion about the extent to which restrictions on defenserelated work constrained the offshoring of software work. Mann noted that, to her knowledge, all employees of firms in the Northern Virginia software complex have a Department of Defense (DOD) clearance, and that those firms cannot offshore to firms whose employees do not have such clearance. She also noted that, since 9-11, the range of areas considered sensitive has expanded—for instance, to include programming for some types of mapping technologies.
LAEL BRAINARD Brookings Institution R O B E R T E . L I TA N Brookings Institution and Kauffman Foundation N I C H O L A S WA R R E N Brookings Institution
A Fairer Deal for America’s Workers in a New Era of Offshoring
P
ublic anxiety has surged over a new wave of offshoring that for the first time puts white-collar jobs at risk from competition with low-wage foreign providers. White-collar offshoring burst into the public consciousness in the middle of a peculiarly unbalanced recovery. The 2001–03 recovery stands out on two counts: the unusually low rate of job creation relative to job destruction, as highlighted by Erica Groshen and Simon Potter (2003), and the “decline in the proportion of that national income going to compensation of employees,” as emphasized by Federal Reserve Chairman Alan Greenspan (2004). Critics were quick to point to the new wave of white-collar offshoring as a major contributor to the poor performance of the U.S. labor market, although the importance of offshoring relative to productivity growth, the bursting of the IT bubble, and other forces remains murky owing to incomplete data.1 Much more can and should be done to help safeguard the livelihoods of American workers in the face of structural shifts of whatever form—while preserving the benefits of an open economy. Whether lauded for its remarkable fluidity or lamented for its heartless insecurity, one of the most striking characteristics of the American job market is high job churning. As our colleague Charles Schultze has pointed out, apart from cyclical ups and downs, roughly 15 million new jobs are created each year, while
The authors wish to thank Chad Bown, Lori Kletzer, Robert Lawrence, Lawrence Mishel, David Richardson, and Howard Rosen for helpful suggestions. 1. Barry Bosworth, Lael Brainard, and Susan Collins, “Services Offshoring: What Do the Data Tell Us?” June 22, 2004 (www.brookings.edu/dybdocroot/pge/20040622summaryfinal.pdf).
427
428
Brookings Trade Forum: 2005
another 13 million are destroyed (Schultze 2004). No other OECD economy comes close (see OECD 1994). This high turnover rate reflects the vigorous forces of competition in the economy, the creation and death of firms, the growth of some, and the decline of others. The latest wave of white-collar offshoring is the most recent in a longer list of drivers, which includes shifts in consumer tastes, innovation that creates new products and services and makes it possible to produce more with less, new opportunities to outsource elements of the business system domestically and overseas, competition from imports, and job opportunities associated with rising exports. To a greater degree than many other advanced economies, the United States is characterized by the process of “creative destruction” so memorably described by the late, great Harvard economist Joseph Schumpeter. The term succinctly captures the two faces of a market economy. Living standards will not grow, according to Schumpeter and a long line of economists who have made similar arguments since, unless change is not only tolerated but actively nurtured. Indeed, alongside high job turnover, the U.S. economy has enjoyed a surge of productivity growth over the past decade. Productivity has increased at a roughly 3 percent annual pace since 1995, more than double the anemic 1.4 percent pace of the preceding two decades. But as Schumpeter’s term also acknowledges, the downside of market-driven dynamism is “destruction.” Firms that do not pass the market test shrink or go out of business, destroying the livelihoods of employees and owners alike. Since the Great Depression, America has recognized some collective responsibility to help those who, through no fault of their own, have been thrown of out a job. The main instrument is federally mandated unemployment insurance (UI), which replaces a portion of an unemployed worker’s previous wage for up to twenty-six weeks. Since 1962 the social contract has also included special protections for those displaced by trade, including extended unemployment insurance and retraining benefits. The Trade Adjustment Assistance (TAA) program was intended to partially offset the distributive consequences of trade liberalization, which exposes workers in import-competing sectors to job loss and often permanent reductions in lifetime earnings, even as it delivers benefits to consumers as well as workers and businesses in export sectors. For too many, however, the nation’s safety net has more holes than netting. For example, the main UI program has so many restrictions that today only about 40 percent of all unemployed workers actually receive benefits. Meanwhile, it has long been difficult, time-consuming, and expensive for workers to prove they are entitled to extended unemployment benefits under TAA. Since the
Lael Brainard, Robert E. Litan, and Nicholas Warren
429
Table 1. TAA Certification, Denials and Enrollment, 2000–04a Units as noted
Petitions (number) Petitions certified (percent) Workers certified (number) Workers denied (number) New income support recipients (number) Newly certified receiving income support (percent) New training recipients (number) Newly certified receiving training (percent) New ATAA recipients (number)b
2000
2001
2002
2003
2004
1,382 61 98,007 53,433
2,353 44 139,587 59,067
2,404 69 235,072 74,760
3,564 53 197,264 91,585
2,918 59 147,956 55,295
32,808
31,459
37,426
43,007
74,865
34 22,665
23 24,843
16 37,186
22 43,444
51 46,536
23 …
18 …
16 …
22 288
31 1,115
Source: Authors’ calculations based on data from the Department of Labor (www.doleta.gov/tradeact/taa_stats.cfm). a. Not all workers certified under an approved TAA petition are individually eligible for TAA benefits and services. b. ATAA = Alternative TAA.
program was last reformed in 2002, for example, it has helped fewer than 75,000 new workers a year, while denying about 40 percent of all petitions, on average (see table 1). Further, there is little evidence that the training requirement under the program is effective. To be sure, some workers get new jobs after retraining, but many others are required to enter retraining to receive extended income support only to find no job in their new specialty at the end of the program. For some, TAA may actually delay reentry into the workforce without providing a commensurate improvement in earnings prospects. And remarkably, the Department of Labor has interpreted the TAA statute as excluding the growing number of services workers displaced by trade, an interpretation that has been challenged in court in recent years. With offshoring accelerating the pace at which workers’ investments in jobspecific skills lose value, the time has come for the federal government to supplement existing efforts with a new insurance program that encourages rapid reemployment and insures wages, not just unemployment, for permanently displaced workers. By providing insurance against wage losses, wage insurance encourages workers to move back into employment more quickly, while defraying the cost to employers of hiring and training a new employee. Rather than paying for compulsory retraining that is not directly connected to a job opening and may never be put to use, wage insurance defrays the cost to employers of providing on-the-job training to new hires. The economy benefits from lower unemployment durations and more efficient retooling for workers. We and others have
430
Brookings Trade Forum: 2005
made the case for wage insurance previously, and we are pleased that it has been endorsed by the Committee for Economic Development.2 Here we make the case that wage insurance is particularly well suited to address the new wave of offshoring of white-collar work and explain why it must be publicly mandated to be effective rather than left to private provision. We make recommendations on program design, compare per worker costs for wage insurance relative to existing programs and alternative proposals, and provide aggregate cost estimates for different versions of the program, using the latest data on displaced workers published by the Department of Labor. Finally, we argue that wage insurance should be attractive to both sides of the political spectrum even in these times of budgetary stress.
An Assessment of Existing Unemployment and Trade Adjustment Programs America’s safety net is miserly in comparison with those of almost all other advanced economies. Not only are U.S. unemployment benefits of relatively shorter duration, but America’s heavy reliance on employer-based insurance means that displaced workers face the prospect of losing health and retirement benefits along with income when they lose their jobs. The consequences of job loss are particularly damaging to workers with some seniority in importcompeting industries, where it has been documented that workers face more protracted spells of unemployment and greater permanent earnings declines than other displaced workers. The UI program is the mainstay of America’s safety net, providing benefits of roughly $260 per week on average for up to twenty-six weeks—a period that is often extended during recessions through temporary legislation.3 It is funded through a combination of federal and state payroll taxes. President Kennedy established the TAA program in 1962 as a central part of the social contract on trade. It is intended to compensate workers who suffer job loss as a result of trade liberalization that otherwise brings gains to the economy. Whereas the theory of economic trade is broadly reassuring that the gains from trade liberalization are sufficient to compensate the losers in principle, TAA—
2. For earlier presentations of the idea, see Kletzer and Litan (2001), as well as Burtless and Litan (2001); Brainard and Litan (2004); and Brainard (1991). See also Committee for Economic Development (2001). 3. See Congressional Budget Office (2004).
Lael Brainard, Robert E. Litan, and Nicholas Warren
431
however imperfect—is the closest mechanism we have to undertake such compensation in practice. In 2002 Congress enacted a major overhaul and expansion of TAA that added a health-care tax credit, doubled the training budget, and substantially raised budget outlays for income support. Key changes included expanding eligibility beyond manufacturing to agricultural workers and to workers displaced by a shift of production overseas, as well as to so-called secondary workers—those indirectly affected by international trade by virtue of being suppliers to plants directly hurt by trade. Coverage of workers enrolled in the program was extended by twenty-six weeks, offering up to 104 weeks of income support and up to seventy-eight weeks of training, job search, and relocation support to eligible participants. Despite these changes, implementation of the TAA program continues to disappoint. The certification process is burdensome and unpredictable. For those few workers who do participate, training is underfunded and often ineffective, and program participants experience long delays before securing jobs that often pay substantially less than previous positions.4 Even after the important 2002 expansion, participation in the program has remained surprisingly low. Although there was a sharp increase in certifications in 2002, the U.S. Government Accountability Office (GAO) concluded that it was caused by the sharp decline in manufacturing employment that preceded the implementation of the 2002 act rather than by the legislative changes. As shown in table 1, nearly half of all petitions were denied in 2003, and fewer than onequarter of those certified eligible actually were granted income support. In part, the low participation rate reflects confusing Department of Labor practices that ultimately deny benefits to roughly three-quarters of the workers whom they certify as eligible for TAA. In eight cases involving hundreds of workers who were denied eligibility for TAA by the Department of Labor, U.S. judges have ruled that the department’s interpretation of eligibility requirements was overly restrictive.5 And the language of court decisions has become increasingly critical. In 2003 the U.S. Court of International Trade stated, “This case stands as a monument to the flaws and dysfunctions in the Labor Department’s administration of the nation’s trade adjustment assistance laws.”6
4. See U.S. Government Accountability Office (2004b), Kletzer and Rosen (2005) for in-depth evaluations of the changes to TAA. 5. U.S. Government Accountability Office (2004b, p. 53). 6. AFL-CIO (2004, p. 8).
432
Brookings Trade Forum: 2005
Moreover, funding for training under TAA remains woefully inadequate. As shown in table 1, the number of participants entering training nearly doubled between 2000 and 2003 owing to the plunge in manufacturing employment. Despite the 2002 increase in training caps, demand continues to exceed the budget cap significantly. According to the GAO: “States have struggled to meet higher demand with the TAA training funds available to them, even though TAA training funds available nationally doubled between fiscal years 2002 and 2003. . . . 19 states temporarily discontinued enrolling TAA eligible workers at some point between 2001 and 2003 because they lacked adequate training funds, and six states have taken this step during fiscal year 2004.”7 Interestingly, this shortfall was projected during the congressional debate on TAA in 2002, but efforts to raise the cap adequately were rejected by Congress. Partly as a result, the number of workers not receiving training benefits rose by one-third in 2003 and, as shown in table 1, now stands at two-thirds of newly certified workers.8 This trend effectively makes the TAA program one of extended unemployment insurance rather than an active reemployment program. The Department of Labor’s restrictive interpretation of eligibility requirements for income support coupled with the ongoing inadequacy of training funds helps to explain why recent outlays for TAA income support are far below available authority, while training outlays fully exhaust the available budget authority, as shown in table 2. Finally, despite the laudable goals of the TAA program, beneficiaries continue to achieve disappointing rates of reemployment and levels of earnings. As shown in table 3, between 2001 and 2004 an average of only 64 percent of participants found new jobs during their participation in TAA. Job retention by workers—defined as those employed in the first quarter after program exit that remained employed for at least two additional quarters—remained constant between 2001 and 2004. And the wage loss for those workers who secured reemployment rose sharply from 13 percent in 2001 to 26 percent in 2004. The hardest fought expansion in the 2002 TAA reform was the inclusion of a tax credit for health insurance, the Health Coverage Tax Credit (HCTC), in recognition that trade-displaced workers often suffer not only permanently lower earnings but also loss of health care coverage. The HCTC pays 65 percent of health insurance premiums through a fully refundable tax credit, so even individuals who owe little or no federal income tax get some benefit. The tax credit was also made “advanceable” beginning in August 2003, meaning that direct 7. See U.S. Government Accountability Office (2004b, p. 4). 8. Department of Labor (www.doleta.gov/tradeact/taa_stats.cfm).
Lael Brainard, Robert E. Litan, and Nicholas Warren
433
Table 2. Trade Adjustment Assistance Budget Authority and Outlays a Millions of dollars 2000 Income support and benefits Authority 283 Outlays 272 Training Authority 132 Outlays 133 ATAA c Authority … Outlays …
2001
2002
2003
2004
2005 b
2006 b
275 259
284 249
713 339
1,079 520
798 637
707 707
132 141
132 142
259 212
259 179
259 243
259 259
… …
… …
… …
10 2.2
48 48
52 52
Source: Office of Management and Budget. a. Actual budget authority, estimated outlays. Includes NAFTA-TAA training, income support, and benefits after the program was merged with TAA in 2003. b. Administration request. c. ATAA was initiated in 2003.
payment of the credits to insurers could be made monthly, when the premiums are due, rather than postponed until workers received an end-of-the-year tax refund, at which time payment would be an unmanageable burden for many displaced workers. Early implementation of the HCTC revealed a daunting set of obstacles and a disappointing participation rate. The GAO reports that fewer than 8,000 TAA beneficiaries were enrolled in the HCTC as of July 2004, less than 6 percent of the individuals certified under TAA. The GAO cites a number of reasons for the low participation rate, including the “fragmentation and complexity of eligibility determination and enrollment process, which required individuals to navigate steps that involve multiple federal and state agencies and to meet specific tax, labor, and health coverage requirements.” For most enrollees there is a three- to six-month process for completing enrollment requirements, during which time Table 3. Trade Adjustment Assistance Reemployment and Earnings Outcomes, 2001–04a Percent unless otherwise noted Outcome Wage loss Estimated salary following TAA enrollment (dollars) Reemployment rate Job retention rate
2001
2002
2003
2004
13
20
27
26
24,512 63 89
25,600 66 89
27,260 62 86
29,668 63 89
Source: Department of Labor. a. Combines results for TAA and NAFTA TAA until the programs were merged in 2003.
434
Brookings Trade Forum: 2005
unemployed individuals must pay out of pocket for coverage or risk a lapse of more than sixty-three days, which would invalidate consumer legal protections such as guaranteed acceptance by a health plan despite preexisting medical conditions. And finally, it quickly became apparent that even a 35 percent share of the annual premiums was out of reach for a large number of displaced workers. According to the GAO, the 35 percent share of premiums absorbs about 25 percent of the average monthly TAA income support for a couple and about 13 percent for an individual.9 Finally, despite the many laudable changes in the 2002 TAA reform, it quickly became clear that TAA remained woefully out of step with current economic realities as the debate over white-collar offshoring of jobs heated up. During the 2002 debate over TAA reform, the congressional majority rejected a strong push by some far-sighted members to explicitly expand coverage to services workers. And the Department of Labor hewed to a restrictive interpretation of the statutory eligibility criteria in rejecting a petition for coverage by IT workers—a decision that is being contested. With the rapid spread of globalization through the hitherto largely nontradable services sector, these decisions effectively shrank the nation’s safety net to an even smaller portion of at-risk workers.
Why Wage Insurance? A main purpose of wage insurance is to accelerate the pace at which permanently displaced workers are reemployed. Wage insurance is more likely to have positive economic benefits if it is targeted to workers who would otherwise suffer a significant earnings loss due to an exogenous drop in the value of their jobspecific skills. These workers have the greatest incentive to prolong their search before accepting new employment at a lower wage scale in the hope of regaining their former earning power—and possibly at the margin because their unemployment benefits are higher relative to earnings forgone. As documented in table 4 (see page 441), trade-displaced workers tend to earn nearly one-fifth less in annual wages following reemployment, compared with about one-sixth less for displaced workers overall, and they remain unemployed more than three times longer. The spread of offshoring to white-collar work would seem to accelerate the pace at which the firm-specific skills of affected workers diminish in value. This development puts a premium on the second critical value of wage insurance: it acts as a training subsidy for new employers. The retraining that displaced work9. U.S. Government Accountability Office (2004a, p. 5).
Lael Brainard, Robert E. Litan, and Nicholas Warren
435
ers receive on a new job is the best kind—it provides new skills that contribute directly to performance in the new job and that are directly useful to both the new employer and employee. This implicit on-the-job training subsidy may be particularly relevant to workers who possess valuable general skills even when their firm-specific skills decline in value. In contrast, generalized retraining programs, such as those available under TAA, cannot guarantee a worker a job at the end and cost the worker the wages that he or she might otherwise have earned if reemployed sooner. Finally, wage insurance has been shown to encourage workers to seek out new types of jobs and jobs in different sectors. This is particularly valuable for permanently displaced workers who have suffered a loss in the return to skills that were specific to their previous job type or sector. Carl Davidson and Stephen Woodbury (1995) provide a formal model to explore the effects of a generalized wage subsidy. They model a wage subsidy program in which a dislocated worker who becomes reemployed would receive a payment equal to half the difference between the wage previously earned and the wage currently earned—examining the cases where the subsidy is paid in perpetuity and one that is limited to the two years following reemployment. The wage subsidy provides incentives for dislocated workers to search harder for jobs and accept employment that they might otherwise refuse, thereby shortening their duration of unemployment and increasing their employment and lifetime earnings. The policy creates both private and social benefits: output increases, workers maintain their general skills, and they acquire new skills on the job. The costs of the wage subsidy are (at least partially) offset by reduced spending on public income support and generalized training programs. For the economy overall, wage insurance results in a small increase in steady-state employment: more of the total available jobs are filled as dislocated workers are induced to search harder for and accept jobs that would otherwise have remained vacant. Davidson and Woodbury also find that rising reemployment levels for dislocated workers are only partially offset by lower employment levels for other workers whose search intensity falls.
How Has Wage Insurance Performed in Practice? Several labor market programs contain elements of wage insurance and provide useful lessons for program design. Perhaps the closest analogue of the program we propose here is a pilot program undertaken in Canada in 1995–96. Approximately 6,000 displaced workers received supplement payments of
436
Brookings Trade Forum: 2005
75 percent of their loss in earnings for up to two years up to a maximum compensation of $13,000 per year. An assessment based on randomized trial methodology found that the program reduced unemployment durations by 4.4 percent on average.10 This seemingly modest result would amount to $450 million in annual savings on unemployment insurance payments in the U.S. context. The incentive effect could be amplified by counting the weeks of unemployment against the two-year maximum compensation period as discussed below, which was not done in Canada. Second, there was strong evidence that the presence of wage insurance did not make workers any more likely to accept the first job they found, allaying potential concerns about the efficiency and durability of job matches. Third, there was clear evidence that wage insurance encouraged workers to consider new types of jobs, including those in other sectors, and thus broadened the job search. In the United States, perhaps the best-known ongoing program with an earnings insurance dimension is the Earned Income Tax Credit (EITC), which effectively subsidizes the earnings of low-income households through refundable tax credits. It is an employee-based subsidy, implemented through the tax code, and strictly targeted on poverty. The substantial literature assessing the effectiveness of the EITC concludes that it has been effective at expanding the labor force participation of low-income workers and at moving several million households out of poverty. However, since the program is not limited to full-time workers, and since the subsidy phases out as income rises, it has also led to some reduction in hours for workers with earnings near the phase-out.11 Research on the Targeted Jobs Tax Credit (TJTC) by Lawrence Katz and others highlights participation rates as low as 9 percent of eligible workers. The literature suggests the low participation rates are primarily a function of stigma and reluctance to self-identify as a member of the target population for fear that they will be perceived to have fewer skills and to be less desirable employees. Second, burdensome certification and eligibility requirements restricted the pool of employers applying for the credit. Despite low participation rates, the research finds an improvement in labor force participation rates of the targeted populations.12 The New Jobs Tax Credit wage subsidy program of 1977–78 also provides some interesting lessons. This employer-based program provided a one-year wage subsidy for new hires. It was intended as a countercyclical measure and
10. See Bloom and others (1999). 11. See Dickert-Conlin and Holtz-Eakin (1999); Dickert, Hauser, and Scholz (1995); Eissa and Liebman (1996); Meyers and Rosenbaum (1997); Eissa and Hoynes (1998); and Liebman (1993). 12. See Katz (1996) and Bishop and Kang (1991).
Lael Brainard, Robert E. Litan, and Nicholas Warren
437
thus was limited to two years. Research by Jeffrey Perloff and Michael Wachter is consistent with findings on other programs that implementing subsidies through employers makes participation highly dependent on employers’ knowledge of the program and capacity for establishing workers’ eligibility. They conclude that the NJTC was not particularly effective as a countercyclical program and attribute its limited impact to its temporary nature. However, they also find that it significantly boosted employment for those firms that participated relative to those that were unaware of the program.13 It is likely that the NJTC program underestimates the economic benefits of a wage insurance program. The pilot program was in place for only a year, and few organizations knew about it beforehand. In our view, a permanent program, known widely, would be viewed differently, and participation rates and responses to the incentives under the program would be greater. There are also a number of wage subsidies currently in place that are implemented through employers at the state and federal levels, such as the Work Opportunity Tax Credit (WOTC) and the Welfare to Work Tax Credit (WTWTC). These programs share some of the features of the wage insurance program we will propose in this paper, namely that employers must hire the workers within a specified time period and that those employees cannot be rehires. However, they differ importantly in targeting groups of workers that are at risk for poor labor force outcomes, such as welfare recipients, felons, veterans, and at-risk youths. One key lesson from both the tax credit and wage insurance experience is that a tradeoff exists between targeting and participation rates. The more targeted the program, the more cost-effective it should be at moving the target population into employment faster. However, the targeting comes with burdensome eligibility and compliance requirements as well as possible stigma, which reduce participation rates to strikingly low levels. The only exception to this appears to be targeting by income level, which can be implemented through the personal income tax system rather than through a system that depends on employer certification. This argues strongly for a less targeted program that is implemented through an existing system with proven efficacy, such as the UI system, rather than the burdensome TAA system. Alternative TAA Most recently, in overhauling the Trade Adjustment Assistance program of 2002, Congress adopted a small and quite restrictive wage-insurance benefit 13. See Perloff and Wachter (1979).
438
Brookings Trade Forum: 2005
targeted at trade-displaced workers. Although the original proposal was laudable, strong resistance resulted in a program that was overly restrictive in scope, and implementation has been nothing short of embarrassing. The so-called Alternative TAA (ATAA) program is supposed to provide limited wage insurance for workers who are older than fifty with incomes at or below $50,000 and who obtain reemployment within twenty-six weeks at a lower rate of pay. The benefit provides 50 percent of the difference between the participants’ earnings in their old and new jobs up to a maximum of $10,000 over two years. Early implementation has been disappointing. As shown in table 1, only 288 participants were enrolled in ATAA in 2003, compared with over 43,000 newly eligible participants in TAA. This reflects a combination of poor implementation—fewer than half of the states implemented the ATAA program by 2003— and poorly defined eligibility criteria. The ATAA program should not be viewed as a defining pilot wage insurance program for trade-displaced workers, because the eligibility criteria are so badly defined. Eligibility is restricted to workers who “lack easily transferable skills” and nevertheless find reemployment within twenty-six weeks.14 The objective is to help workers “for whom retraining may not be appropriate” return to work as quickly as possible.15 In contrast, we believe that wage insurance is particularly well suited for workers who retain valuable general skills but may no longer earn a premium on a set of occupation- or job-specific skills that have lost value because of changes in demand, technology, or foreign competition. While the ATAA’s biggest flaw is its stipulation that eligible workers must “lack easily transferable skills,” which complicates eligibility enormously, we would contend that age and income restrictions are also counterproductive. There is a compelling case for making wage insurance available to mid-career workers with valuable general skills and those in higher income ranges (albeit with a cap). Private Provision? A widely cited study of offshoring in 2003 by McKinsey & Co. suggested that firms voluntarily offer wage insurance as a benefit to their workers and estimated it would cost no more than 5 percent of the savings firms realize from offshoring.16 Effectively, this proposal asks firms to promise a kind of severance program to workers as a condition of the employment contract. 14. U.S. Government Accountability Office (2004b). 15. See Department of Labor (www.doleta.gov/tradeact/benefits.cfm). 16. Agrawal and Farrell (2003).
Lael Brainard, Robert E. Litan, and Nicholas Warren
439
We are not opposed to the idea but suggest its limitations. For one thing, wage insurance is a cost-effective response not only to offshoring but also to all sources of permanent displacement, many of which have nothing to do with cost savings. For example, firms may lay off workers because demand for their products has weakened, or because a new technology has displaced their core business model. These events provide no “savings” from which firms could finance wage insurance. More broadly, a firm facing intense competition—and which firms aren’t these days?—almost certainly would “pay” for any wage insurance it offers by reducing workers’ cash salaries by the estimated cost of the program. Workers who could choose between firms that offered the insurance and firms that didn’t might not view the insurance to be of sufficient value, thinking like many who live in earthquake zones: displacement is unlikely to happen to me, so why should I accept lower wages to pay for it? Fearing this outcome, firms may be reluctant to add the benefit, even though some potential job seekers would value it. Moreover, if providing wage insurance were voluntary on the part of employers, they might be reluctant to offer it for fear of signaling to potential employees that their tenure could be abruptly curtailed. For both of these reasons, we are doubtful that the market, on its own, will provide wage insurance to any more than a small portion of the labor force. As a result, a broad range of workers who might lose their jobs for any number of reasons during their careers—and many if not most workers will at some point—will not be able to obtain wage insurance of the type we have outlined. Might insurance companies step in to fill this market need? We suspect not, for a simple reason that economists call “adverse selection.” Insurers are likely to surmise that those who most want the insurance are also those most likely to be serially unemployed and charge premiums accordingly. How can one otherwise account for the fact no insurer has yet stepped forward to offer the insurance? Of course, the same was true of unemployment insurance: before the government provided unemployment insurance, insurers were not providing it either. Only government, therefore, is likely to address these “market failures” in the labor market.
Key Design Choices The key design variables for a wage insurance program are as follows: the target population; the duration of the insurance payments; the wage loss replacement rate; and the maximum total compensation.
440
Brookings Trade Forum: 2005
We recommend targeting the program to permanently displaced workers who have achieved some seniority in their previous jobs, for instance, at least two years’ experience; these are defined as “long-tenure” workers in Bureau of Labor Statistics data. We use the definition of “displaced workers” from the Labor Department Displaced Worker Survey: those who lose their jobs because their plant or company closed or moved, there was insufficient work for them to do, or their position or shift was abolished. We also recommend restricting the program to workers who were displaced from full-time jobs and reemployed full-time, so as to avoid giving them any possible incentive to reduce the number of hours worked. There are compelling reasons to offer wage insurance to all full-time permanently displaced workers rather than restricting it to trade-displaced workers. First, most job displacement occurs for reasons other than trade—such as technology change and shifts in consumer demands. Indeed, for many occupations, it is difficult to disentangle the effects of trade and technology. In the most recent wave of offshoring, Frank Levy and Richard Murnane (2004) conclude that the jobs most vulnerable to offshoring are also the most codifiable and thus susceptible to automation. Moreover, it matters little to the displaced worker what has caused his or her misfortune. Nor should it matter for social policy; technological shifts can be as redistributive as shifts in trade. Finally, the administrative requirements for limiting benefits to trade-displaced workers, as well as the process of establishing eligibility, would severely undermine the program’s effectiveness, as with TAA. We further recommend limiting the compensation period to the first two years, when on-the-job-training is arguably most concentrated; and compensation would begin only once a worker actually landed a new job. (Below we also show cost estimates for a one-year program.) A displaced worker would thus have an incentive to find a new job as quickly as possible (which might slightly reduce his or her cost of unemployment insurance), even if it paid somewhat less than what the worker was earning before. For TAA-eligible workers, the incentive to use the wage insurance program instead of entering or continuing in TAA could be reinforced by triggering the two-year period for compensation when they find a new job or at twenty-six weeks, whichever is earlier. By doing this, the total remaining compensation under wage insurance would decline with each week of unemployment outside the twenty-six-week window. For other workers, the twenty-six-week trigger would simply reinforce the benefit of accepting a job more rapidly. Policy decisions about the replacement rate and the annual cap on compensation will determine what kinds of workers are most likely to benefit from the program. For instance, the combination of a high replacement rate and low
Lael Brainard, Robert E. Litan, and Nicholas Warren
441
Table 4. Displaced Workers: Prior Earnings, Earnings Losses, and Unemployment Durations, 2001–03 Averagesa Units as noted
Worker type Trade displaced (TAA)b Total displaced Manufacturing displaced Services displaced Services potentially affected by offshoring Telecommunications ISP, data processing, and other information services Architectural, engineering, and related services Computer systems design and related services
Full-time workers displaced (thousands)
Average earnings on lost job (dollars)
Change in Average earnings number in new job of weeks (percent) without work
71,000 2,068,000 693,000 953,000
32,505 42,687 40,154 45,479
–21 –16 –20 –13
80c 11.9 14.1 10.5
205,000 77,674
60,535 52,830
–14 –26
13.1 14.7
9,000
62,366
–24
n.a.
41,000
61,058
–16
18.7
75,000
65,921
–6
14.5
Source: Department of Labor, Bureau of Labor Statistics. a. Table refers to full-time workers with at least two years’ tenure. b. TAA displaced worker estimate based on entire TAA population. Earnings estimates of TAA displaced workers based on those that completed the program. c. Authors’ calculations based on TAA data. n.a. = not available.
annual compensation cap would provide the greatest benefits to lower-income workers suffering a steep loss in earnings, while a lower replacement rate with a high annual cap would tilt compensation toward higher-income workers. Program costs are very sensitive to the choice of the replacement rate and the cap on insurance payments relative to the earnings losses of the eligible population. Table 4 compares average earnings, earnings losses, and unemployment durations for displaced workers certified by TAA with displaced workers in manufacturing overall, services overall, and the services activities that appear to be most vulnerable to offshoring. A few facts stand out. In those services sectors potentially most affected by offshoring, earnings before displacement are 51 percent higher than in the overall manufacturing sector, where trade adjustment programs have traditionally focused. And although the percentage earnings loss is more modest for offshoring-susceptible services than for manufacturing, the absolute earnings loss is greater. Trade-displaced workers certified under TAA stand out in table 4 as having substantially lower earnings than displaced workers in services and manufacturing overall, the greatest losses relative to prior earnings, and the longest duration of unemployment.
442
Brookings Trade Forum: 2005
Table 5 compares the costs and benefits for the average displaced worker of a wage insurance program that replaces 50 percent of earnings losses up to a maximum of $10,000 with UI, TAA, and various other proposals (a broader range of program specifications is assessed in our aggregate cost estimates in the next section). Even if the maximum benefit is doubled from the current allowable under ATAA, the cost of our wage insurance proposal compares very favorably with the cost of TAA, and indeed falls midway between the current unemployment and retraining benefits available under UI and WIA (Workforce Investment Act) and the all-in cost of TAA benefits. The Bush administration and some members of Congress have proposed Personal Reemployment Accounts (PRAs), which are essentially a cheaper alternative to the benefits currently available under WIA. The proposed PRAs would provide a voucher for $3,000 to workers who are likely to exhaust their UI benefits. The vouchers could be used to purchase reemployment services such as counseling and training or to extend unemployment benefits; or, for those who found employment within thirteen weeks, unused benefits could be distributed as income—with 60 percent (up to $1,800) distributed at the time of accepting a job and the remainder (up to $1,200) distributed following six months on the job. As Andrew Stettner and Amy Chasanov point out, however, as a training subsidy the PRA is starkly inferior to existing federally funded training vouchers, which amount to $10,000 per worker. And they cite Department of Labor research by Christopher O’Leary and Randall Eberts, which predicts that “PRA recipients might therefore reduce use of [training and counseling] services in hopes of receiving larger reemployment bonuses” and highlights “the incentive for some claimants to accept low-paying jobs simply to qualify for the first bonus paid upon reemployment,” which could result in short-duration matches and diminish the prospects for on-the-job training.17 Although the maximum compensation under our wage insurance proposal is much greater than the proposed PRA, it should also deliver much greater economic benefits. Those include more efficient and more durable job matches and incentives for companies to invest in the skills of new hires over a two-year period. One important question is whether wage insurance recipients would continue to have access to the HCTC during the two years they receive wage insurance, with the new employers possibly picking up the employee share of the premium. The advantage of such an approach would be to ensure continuous health coverage for workers as they move between jobs, while further lowering the initial 17. Stettner and Chasanov (2005); O’Leary and Eberts (2004, p. 4).
Lael Brainard, Robert E. Litan, and Nicholas Warren
443
Table 5. Costs per Worker under Alternative Adjustment Programs Benefit per worker in dollars
TAA a
Brainard/ Litan/ ATAAc UI and wage Warren Bush WIA b insurance proposal PRAsd
Skillimprovement tax credit
Maximum duration Benefit Maximum employersponsored training Total average income maintenance payments Maximum income subsidy Maximum one-time job search allowance Maximum one-time relocation allowance Maximum reemployment, training, or support services Training Average health coverage tax credit Total
1.5 years .5 year
2 years
2 years
1 year
5 years
…
…
…
…
…
5,000
19,300 …
6,800 …
… 10,000
… 20,000
… …
… …
1,250
…
…
…
…
…
1,250
…
1,250
…
…
…
… 4,800
… 5,000 – 9,000
… …
… …
3,000 …
… …
6,100
…
8,100
…
…
…
32,700
11,800 – 15,800
19,350
20,000
3,000
5,000
Sources: Authors’ calculations based on data from the Department of Labor and the Economic Policy Institute. a. Trade Adjustment Assistance b. Unemployment Insurance and Workforce Investment Act c. Alternative Trade Adjustment Assistance d. Personal Reemployment Accounts
costs of hiring to an employer who would otherwise provide health care coverage to its employees. The downside, of course, is an increase of roughly 25 percent to the per worker cost of wage insurance. While the overall cost of health-careenhanced wage insurance would still compare quite favorably with TAA, the gap with UI would widen considerably. Aggregate Cost Estimates The aggregate cost of wage insurance depends on several characteristics of the eligible population: the number of eligible displaced workers; the wage loss of those who accept work at lower pay; and the duration of unemployment before reemployment (if there is a program start trigger, as described below).
444
Brookings Trade Forum: 2005
Table 6. Reemployment Rate and Earnings Losses for Displaced Workers, 2000–03a Thousands of workers, unless otherwise noted
Unemployment rate (percent) Total displaced Total displaced from full-time jobb Total reemployedc Total reemployed full-timeb Total reemployed full-time at lower wagesd Average wage loss (dollars)
2000
2001
2002
2003
4.0 2,667 1,191 1,917 654 258 12,706
4.7 3,465 1,985 2,461 1,126 514 17,463
5.8 3,615 1,903 2,507 1,030 452 15,473
6.0 5,050 2,318 2,581 925 375 14,792
Source: Authors’calculations based on Bureau of Labor Statistics data. a. Workers are classified as displaced if they reported the reason for their job loss as one of the following: plant or company closed down or moved, insufficient work, position or shift abolished. b. Displaced from permanently lost full-time jobs with at least two years’ tenure. c. Displaced within last fifty-two weeks. d. Also restricted to those holding fewer than four jobs since reemployment.
Table 6 provides some of the key data drawn from the Department of Labor Displaced Worker Survey, which is necessary for constructing cost estimates. Among the key features of this table are national unemployment rates; the numbers of workers who would be eligible for wage insurance; and the mean annual earnings loss of eligible workers whose new jobs pay less than their old ones. We restrict eligibility to workers permanently displaced from full-time jobs with two or more years’ tenure who enter full-time reemployment within fifty-two weeks at lower pay; different requirements would change the size and characteristics of the eligible pool and thus the cost estimates. For 2003, table 4 further subdivides the data to compare the characteristics of displaced workers in overall manufacturing, overall services, and services potentially affected by offshoring. Interestingly, while displaced services workers earn roughly 10 percent more than displaced manufacturing workers on average, the absolute level of earnings losses is somewhat smaller, so wage insurance need not be any more costly for services workers on average than for manufacturing workers (although in aggregate the number of displaced workers is much larger in services than in manufacturing). Table 7 provides cost estimates for a wage insurance program under different assumptions about the duration of wage insurance payments (one and two years); the replacement rate (30, 50, and 70 percent); and the annual cap of compensation payments ($10,000 and $20,000) over the period 2000–03. One central takeaway from the table is that the estimated costs of the program have risen over time, especially since 2000. This is due to a combination of factors: higher unemployment and hence more displaced workers generally; a higher fraction of
Lael Brainard, Robert E. Litan, and Nicholas Warren
445
Table 7. Cost of General Wage Insurance Program, 2000–03 Total eligible workers (number)
Wage replacement rate (percent)
Cost with $10,000 cap (millions of dollars)
Cost with $20,000 cap (millions of dollars)
One-year program 2000
258,000
2001
514,000
2002
452,000
2003
375,000
30 50 70 30 50 70 30 50 70 30 50 70
864 1,249 1,529 1,984 2,606 3,033 1,776 2,469 2,907 1,381 1,803 2,104
968 1,496 1,945 2,462 3,535 4,340 2,001 3,080 3,995 6,106 2,436 3,026
Two-year program 2000
503,000
2001
772,000
2002
966,000
2003
827,000
30 50 70 30 50 70 30 50 70 30 50 70
1,854 2,592 3,112 2,848 3,854 4,562 3,760 5,074 5,939 3,158 4,272 5,011
2,087 3,219 4,135 3,430 5,030 6,286 4,463 6,615 8,335 3,607 5,517 7,021
Year
Source: Authors’ calculations based on Bureau of Labor Statistics data.
eligible workers who suffered a wage loss; and a higher average wage loss, which jumped especially between the 2001 and 2003 survey years. Accordingly, whereas in 2000 a wage insurance program with a two-year duration, a 50 percent replacement rate, and a $10,000 annual cap would have cost $2.6 billion, by 2003, when the national unemployment rate was substantially higher, the cost for the same program would have been $4.3 billion (in current dollars). It is important to put these cost estimates in perspective. In 2003, for example, state and federal governments paid out $42.4 billion in unemployment insurance benefits. This figure is more than ten times the estimated cost of a wage insurance program for that year (under the assumptions just laid out). Moreover,
446
Brookings Trade Forum: 2005
a substantial portion of the costs of a wage insurance program clearly would be tied to the economic cycle. The 2003-based cost estimate of $4.3 billion, for example, is not representative of the cost of the program over, say, a ten-year period. Based on the data shown in table 7, an annual average net cost of something like $3.5 billion for a two-year program with a 50 percent replacement rate and a $10,000 annual cap seems more realistic. And the costs of the program are substantially lower even in a year with high unemployment for more modest benefits. For 2003 alone, the costs of the program could range from a low of $1.4 billion for the most modest benefits (a one-year program with a 30 percent replacement rate and a $10,000 cap) to a high of $7.0 billion for the most generous package of benefits (a two-year program with a 70 percent replacement rate and a $20,000 annual cap). Cross-Over Political Appeal and Funding A more comprehensive, incentive-based safety net for displaced workers that encourages rapid reemployment and on-the-job training should be good politics across the political spectrum. Indeed, wage insurance was one of the few recommendations on which the bipartisan members of the U.S. Trade Deficit Review Commission were able to agree in their 2000 report. For those who support active labor market policies generally, the program we have outlined should have natural appeal. Wage insurance would supplement trade adjustment programs and unemployment insurance, so even more of many workers’ unemployment losses would be covered; wage insurance would also lower the duration of unemployment and provide potential employers with greater incentives to hire and train displaced workers. For those who are sympathetic to government programs that extend aid only for individuals who demonstrate responsibility, our proposal also should be appealing. The wage insurance subsidy would only kick in once a displaced worker had started a new job. As a result, government aid would provide incentives to actively seek reemployment. Names are everything in politics, and the wage insurance plan we have suggested here is no exception. Fundamentally, what we are proposing is insurance, and thus charges for it really are insurance premiums, not taxes (much as seniors paying for a portion of hospital insurance covered under Part B of Medicare are really paying a premium, albeit one that is subsidized). Indeed, the program could be implemented through an advanceable, refundable wage insurance tax credit.
Lael Brainard, Robert E. Litan, and Nicholas Warren
447
Financing Wage Insurance Admittedly, a key sticking point for many who are opposed to any spending increase on principle, and perhaps some who want limited resources directed elsewhere (such as toward more comprehensive health insurance), is how to finance the program. In part, we would expect some offsetting savings on unemployment and training programs from more rapid reemployment; the Canadian experience suggests savings of greater than $400 million. In addition, for workers displaced by trade who choose wage insurance rather than TAA, the per worker cost is likely to be roughly one-third less than the combined cost of the unemployment and training benefits of TAA, as shown in table 5. One relatively simple way to finance the uncovered costs of the program would be through a modest change in the current Federal Unemployment Tax (FUTA)—possibly split between employers and employees. Currently, employers pay a tax rate of 0.8 percent on workers’ wages up to $7,000 per year to cover the federal share of unemployment insurance. Under these conditions, FUTA raises approximately $7 billion per year. If, hypothetically, an additional $3 billion were needed (which is a high estimate considering the potential for other program offsets), then Congress could raise the FUTA premium rate on the first $7,000 of earnings or raise the earnings ceiling above $7,000, or some combination of the two measures (a smaller increase in the tax rate and the earnings ceiling). If the offsetting cost savings from wage insurance were greater, it might be necessary to raise even less than $3 billion per year (averaged over a ten-year period).
Conclusion Using a conservative estimate of offsetting savings in other unemployment and training programs, the net cost of $3.5 billion per year amounts to an insurance premium of roughly $25 per worker per year. That is a small price to pay for shorter periods out of work and more efficient retooling for workers. Wage insurance provides a critical tool to ease transitions in the face of accelerated job churning while preserving the benefits of an open and innovative economy. For the price of $25 per worker per year, the nation reaps economic benefits in the form of less job insecurity, more rapid returns to work, broader job searches, and more efficient reskilling through on-the-job training.
Comment and Discussion
Lawrence Mishel: Before addressing the specifics of the paper by Brainard, Litan, and Warren, it is important to address the wider context for their proposal and how the losses from globalization, including offshoring, are to be treated. This is the part of the conference where we address how to help those who are the losers in the offshoring/globalization process. It always comes after economists reaffirm that facilitating trade without any constraint is good, and in fact yields large gains for the nation, for the world, and for workers generally. Theoretical results, such as those in Markusen’s paper, showing possible losses from trade/offshoring, are duly noted. However, conference discussion usually, as in this one, is riddled with comments minimizing the role of trade in generating labor market problems or inequities as if it is possible for globalization to yield large gains with only minimal costs. By discussing policies that help dislocated workers, conferees are assured that they have appropriately addressed equity issues and, even more, have proposed policies that should remove any political obstacles to expanded trade. Any failure to win wide political acceptance for this approach is attributed to narrow, self-interested, special interests. I think a broader view would suggest the following.
There’s a Job Quality Problem It is not unreasonable at all for American workers to be skeptical about trade expansion given the persistent erosion of good jobs for non-college-educated workers, and even many college-educated workers, over the past twenty-five 448
Lael Brainard, Robert E. Litan, and Nicholas Warren
449
years or so. This is true whether trade has played a large or a small role. People rightly feel that good jobs are scarce, that if they lose their job they will likely obtain a far worse job, and that it is hard for young people to find good jobs. In particular, there is only limited access to jobs with good pensions and health care plans. And yet, as people face more risks there seems to be a corresponding effort to remove government assistance and to increase, rather than ameliorate, inequalities. Even though there has been very fast productivity growth, wages and family incomes have not advanced since 2000. Those interested in trade liberalization, in my view, need to address these larger concerns. These issues are easily illustrated by (1) the ongoing growth of the wage gap between high-wage (95th percentile) and middle-wage (median) workers (see table 1, columns 1 and 2); (2) the fact that the real hourly wages of middle- and low-wage men were no higher in 2004 than in 1979 (table 2) despite productivity growth of more than 80 percent; (3) the lower wages (table 1, columns 3 and 4) and the diminution of employer-provided benefit coverage (table 3, columns 1 and 2), particularly health care, in jobs obtained by new high school graduates; and (4) the modest growth in wage levels for new college graduates (table 1, columns 5 and 6), with declining wages in the last few years and a steadily declining ability to find jobs with employer-provided health care coverage (table 3 , columns 3 and 4).1
The Downsides of Trade Go Way beyond Dislocation Workers dislocated by trade who experience downward mobility are the most visible and acute losers from expanded trade. But the impact is much larger and includes the remaining workers in trade-impacted industries and workers with comparable skills throughout the economy. It includes young people who lose access to particular types of jobs. And it includes workers who take lower raises for fear of seeing their jobs moved overseas. Over the past ten years the impact has increasingly faced white-collar workers, and young college students are well aware of this. The data in table 2 show that the growing wage gap between the 95th and 50th percentiles among men over the 1979–2004 period yielded an $11.83 greater wage differential. If “trade” were responsible for “just” 20 percent of this inequality, then the median male lost about $2.30 an hour, or nearly $5,000 per 1. All of the data for these figures are drawn from Mishel, Bernstein, and Allegretto (2005).
450
Brookings Trade Forum: 2005
Table 1. Wage Trends, 1975–2004 Units as noted Wage inequality (ratio of 95th to 50th percentile)
Entry level hourly wage (2004 dollars) High school graduate
College graduate
Year
Men
Women
Men
Women
Men
Women
1975 1980 1985 1990 1995 2000 2004
2.16 2.13 2.52 2.53 2.68 2.77 2.99
2.16 2.18 2.37 2.47 2.67 2.69 2.75
12.29 12.50 10.89 10.36 9.82 10.74 10.76
9.19 9.39 8.75 8.68 8.36 9.18 9.02
16.32 17.23 17.95 17.79 16.41 19.84 18.58
13.55 13.32 14.65 15.61 15.08 16.84 16.54
Source: Author’s analysis of data from CPS ORG (Current Population Survey, Outgoing Rotation Group). See Mishel, Bernstein, and Allegretto (2005).
year for someone working full time year-round. It is hard to consider these losses inconsequential, and they go way beyond those who experience the direct, acute losses from trade: downwardly mobile displaced workers.
Policies to Address Concerns Are Limited Any survey of policies to help the losers from trade would suggest that we are not interested in compensating the losers, and we are not even all that interested in helping people adjust. This is especially the case with the current congressional majority, whose interest in expanded trade combines with a desire to
Table 2. Change in Male Hourly Wage, 1979–2001 Units as noted Percentile
Hourly wage (2004 dollars) 1979 2004 Change, 1979–2004 Dollars Percent
Gap
20th
50th
95th
95/50
9.97 9.35
15.97 15.26
34.49 45.61
18.52 30.35
–0.62 –6.2
–0.71 –4.4
11.12 32.2
11.83 …
Source: Author’s Analysis of CPS ORG data. See Mishel, Bernstein, and Allegretto (2005).
Lael Brainard, Robert E. Litan, and Nicholas Warren
451
Table 3. Employer-Provided Health Insurance and Pension Coverage, 1980–2003 Percent Recent high school graduates
Recent college graduates
Year
Health
Pension
Health
Pension
1980 1985 1990 1995 2000 2003
64.8 54.4 46.5 38.2 37.8 33.1
35.1 26.1 23.4 20.6 21.9 19.2
78.6 78.6 70.7 69.2 70.6 62.6
50.7 46.2 46.1 45.1 54.6 45.7
Source: Author’s analysis of March CPS data. See Mishel, Bernstein, and Allegretto (2005).
weaken worker assistance and protections. After the Gingrich revolution, I would remind everyone, there was an effort to eliminate Trade Adjustment Assistance (TAA). There has not been a policy choice available that expands trade and has the winners compensating the losers. The choice has been whether to have expanded trade where there is little, if any, compensation for the “losers” in a context where the overall economy is not working for working people. This is why many politicians will not vote for trade agreements that contain a modest adjustment or compensation plan. Sometimes, it is true, a politician who would like to vote for a trade agreement anyway likes the “cover” of an adjustment program, saying “I voted for ‘X’ treaty but I also voted to help the workers hurt by trade.”
What about Wage Insurance? I think the efforts by the authors and by others (Kletzer, Rosen, Jacobson) to develop better policies to assist dislocated workers facing difficult circumstances is commendable. I appreciate and support the effort to broaden the scope of who will be assisted to include all dislocated workers, not just those displaced by trade. But this reasonable idea can be counterproductive if wage insurance is substituted for TAA, as the current congressional majority prefers. In the introduction the authors do state that wage insurance should supplement existing efforts, but there is no further discussion in the paper of this issue. It is important that wage insurance be a supplement to the TAA (or better, an improved TAA program). If TAA were replaced, then instead of adding to workers’ options for dealing with their transition, trade-impacted workers would lose the option,
452
Brookings Trade Forum: 2005
as flawed as it is, to receive income support while undergoing retraining or to obtain job search and relocation assistance. In this respect, wage insurance represents a way to cushion the fall but offers no prospect of changing trajectories to end up in a better place. The possibility that TAA could be replaced by wage insurance is not an abstract concern; the current wage insurance experiment is labeled Alternative Trade Adjustment Assistance because important political actors want wage insurance to replace TAA. It is also interesting to note that the push for wage insurance arises from a handful of economists and not from any group that represents or speaks for workers who face layoffs and difficult transitions. Those groups, primarily unions and some advocacy groups, tend to want to improve and extend unemployment insurance, expand training options, and find ways to cover health insurance. My last comments involve some quibbles about the numbers used to estimate the costs of the program. These estimates clearly overstate the costs in one important way: the authors implicitly assume that 100 percent of those eligible will become beneficiaries of the program. There are also important ways that costs are understated. First, not all displaced workers are included in the counts because those displaced for “other” reasons are excluded. Farber has shown that this category, which is excluded from the official BLS definition of displacement, rose sharply in the late 1990s and comprises 30–40 percent of total displacement. Including this group would raise the costs by roughly 60 percent. The annual estimates of displacement are also too low for many of the years because of “recall bias,” reflected in less displacement being reported the longer the time lapse. Last, the estimates are based on reemployment rates even though many of the displacement spells are in progress when the surveys are done (especially since the most displacement reported is in the most recent twelve-month period). My guess is that the costs the authors present may be in the ballpark, but only because these biases are offsetting. General Discussion: The presentations prompted an animated discussion with considerable agreement on the critical need for better worker adjustment programs, but also considerable pessimism about the likelihood that such programs will be enacted in the current environment. Howard Rosen commented on the need for greater focus on domestic responses to trade policy. Part of the problem is fighting the conventional wisdom that perceives domestic adjustment programs as wasteful. As such, Rosen argued that those who focus too much attention on the effectiveness of trade adjustment programs tend to forget the political motivation behind them.
Lael Brainard, Robert E. Litan, and Nicholas Warren
453
Rosen stressed the need for more stringent eligibility criteria for trade adjustment programs in order to ensure that assistance is provided to those who really need it. He noted that this is an especially important issue for any wage insurance program. Rosen warned against placing too much emphasis on wage insurance, since only a minority of current TAA participants are eligible for this specific intervention. Finally, Rosen called on the authors to think more about ways to identify eligible workers and get them into the wage insurance program as soon as possible. Lori Kletzer countered that although trade is an effective hook for garnering support for worker assistance programs, it should be used appropriately. Trade has been overemphasized in the debates on the costs of job loss in manufacturing and services. The characteristics of workers and the job markets they are displaced in and live in are more significant. Kletzer pointed out that displaced workers suffer whether the job loss is due to trade or some other factor. Further, while linking job loss to trade provided an incentive to pass legislation for worker programs, such an approach might overshadow the persistent costs of displacement in terms of earnings losses following reemployment, which a wage insurance program addresses. In addition, a wage insurance program encourages workers to accept entry-level positions, providing them with job training in other sectors, another key benefit that should not be overlooked. Daniel Tarullo worried about the political feasibility of a wage insurance program, doubting any such program stands a chance of enactment in the near term. At the moment, Congress is gutting anything resembling a social insurance program in this country. Further, the existence of the deficit is going to be invoked over the next few years as a rationale for cutting existing programs in order to preserve the tax cuts and the repeal of the inheritance tax, as well as the spending in Iraq. The one group being protected in these spending cuts are people over 65, a group that votes in the highest proportions. Young blue-collar workers, who do not vote in particularly large numbers, are unlikely to be protected by social insurance programs. Catherine Mann commented that one has to keep in mind the macroeconomic benefits of having workers reemployed through a wage insurance program. Unemployment implies the economy is inside its production possibility frontier. The existence of unemployed workers also means there is skill depreciation. If there are market failures that result in firms not providing training and workers not taking up training, these can be ameliorated and at the same time move both the economy and the production possibility frontier outward. Lael Brainard agreed that the argument for wage insurance is not just a compensation argument but also an efficiency argument. With accelerated churning
454
Brookings Trade Forum: 2005
of skill demands, one can construct a model that yields individually rational but globally suboptimal underinvestment in specific skills. In this kind of a model, such as Brainard (1991), providing some social insurance for skills investments raises the level of human capital in the economy overall. She also agreed with both Lawrence Mishel and Howard Rosen that the compensation elements of TAA are very important for a certain segment of the displaced population and should be preserved in parallel with the establishment of a wage insurance program.
References AFL-CIO. 2004. “The Bush Record on Shipping Jobs Overseas.” AFL-CIO Issue Brief. Washington (August). Agrawal, Vivek, and Diana Farrell. 2003. “Who Wins in Offshoring.” McKinsey Quarterly, Special Edition: 36–41. Bloom, Harold, Saul Schwartz, Susanna Lui-Gurr, and Suk-Won Lee. 1999. Testing a Re-employment Incentive for Displaced Workers. Ottawa: Social Research and Demonstration Corporation. Bishop John H., and Suk Kang. 1991. “Applying for Entitlements: Employers and the Targeted Jobs Tax Credit.” Journal of Public Policy Analysis and Management 10 (1): 24–45. Brainard, Lael. 1991. “Protecting Losers: Optimal Diversification, Insurance, and Trade Policy.” Working Paper 3773. Cambridge, Mass.: National Bureau of Economic Research. Brainard, Lael, and Robert E. Litan. 2004. “‘Offshoring’ Service Jobs: Bane or Boon and What to Do?” Brookings Policy Brief 132 (April). Burtless, Gary, and Robert Litan. 2001. Globaphobia Revisited: Open Trade and Its Critics. Brookings. Committee for Economic Development. 2001. From Protest to Progress: Addressing Labor and Environmental Conditions through Freer Trade. Washington: Library of Congress. Congressional Budget Office. 2004. Family Income of Unemployment Insurance Recipients (March). Davidson, Carl, and Stephen A. Woodbury. 1995. “Wage-Rate Subsidies for Dislocated Workers.” Upjohn Institute Staff Working Paper 95-31. Kalamazoo, Mich: W. E. Upjohn Institute for Employment Research (January). Department of Labor, Employment and Training Administration. “Trade Adjustment Assistance (TAA) and Alternative Trade Adjustment Assistance (ATAA) Services and Benefits.” www.doleta.gov/tradeact/benefits.cfm [January 12, 2006]. Dickert-Conlin, Stacy, and Douglas Holtz-Eakin. 1999. “Employee-Based versus Employer-Based Subsidies to Low-Wage Workers: A Public Finance Perspective.”
Lael Brainard, Robert E. Litan, and Nicholas Warren
455
JCPR Working Paper 79. Joint Center for Poverty Research, Northwestern University/University of Chicago. Dickert, Stacy, Scott Hauser, and John Karl Scholz. 1995. “The Earned Income Tax Credit and Transfer Programs: A Study of Labor Market and Program Participation.” In Tax Policy and the Economy, vol. 9, edited by James M. Poterba. MIT Press. Eissa, Nada, and Hilary Hoynes. 1998. “The Earned Income Tax Credit and the Labor Supply of Married Couples.” Working Paper W6856. Cambridge, Mass.: National Bureau of Economic Research. Eissa, Nada, and Jeffrey B. Liebman. 1996. “Labor Supply Response to the Earned Income Tax Credit.” Quarterly Journal of Economics 111 (2): 605–37. Greenspan, Alan. 2004. Transcript from Federal Document Clearing House, as distributed by Bloomberg news service, of the question and answer session of the testimony of Alan Greenspan, Chairman of the Board of Governors of the Federal Reserve System, before the Joint Economic Committee, April 21. Groshen, Erica L., and Simon Potter. 2003. “Has Structural Change Contributed to a Jobless Recovery?” Current Issues in Economics and Finance 9 (8). Federal Reserve Bank of New York. Katz, Lawrence. 1996. “Wage Subsidies for the Disadvantaged.” Working Paper 5679. Cambridge, Mass.: National Bureau of Economic Research (July). Kletzer, Lori G., and Robert E. Litan. 2001. “A Prescription for Worker Anxiety.” Brookings Policy Brief 73 (March). Kletzer, Lori G., and Howard Rosen. 2005. “Easing the Adjustment Burden on U.S. Workers.” In The United States and the World Economy: Foreign Policy for the Next Decade, edited by C. F. Bergsten. Washington: Institute for International Economics. Levy, Frank, and Richard J. Murnane. 2004. The New Division of Labor: How Computers Are Creating the Next Job Market. Princeton University Press. Liebman, Jeffrey B. 1993. “The Impact of the Earned Income Tax Credit on Incentives and Income Distribution.” In Tax Policy and the Economy, vol. 12, edited by James M. Poterba. MIT Press. Meyers, Bruce, and Dan Rosenbaum. 1997. “Welfare, the EITC, and the Labor Supply of Single Mothers.” Working Paper. Northwestern University (November). Mishel, Lawrence, Jared Bernstein, and Sylvia Allegretto. 2005. The State of Working America, 2004/2005. Ithaca, N.Y.: ILR Press (imprint of Cornell University Press). Organization for Economic Cooperation and Development (OECD). 1994. The OECD Jobs Study: Facts, Analysis, Strategies. Paris. O’Leary, Christopher J., and Randall W. Eberts. 2004. “Personal Reemployment Accounts: Simulations for Planning Implementation.” Occasional Paper 2004-08. W. E. Upjohn Institute for Employment Research, U.S. Department of Labor ETA (May). Perloff, Jeffrey M., and Michael L. Wachter. 1979. “The New Jobs Tax Credit: An Evaluation of the 1977–78 Wage Subsidy Program.” American Economic Review 69 (2): 173–79. Schultze, Charles. 2004. “Offshoring, Import Competition, and the Jobless Recovery.” Brookings Policy Brief 136 (August).
456
Brookings Trade Forum: 2005
Stettner, Andrew, and Amy Chasanov. 2005. “Setting the Wrong Course: Personal Reemployment Accounts Fail to Address the Real Needs of Jobless Workers.” Economic Policy Institute (February). U.S. Government Accountability Office. 2004a. “Health Coverage Tax Credit: Simplified and More Timely Enrollment Process Could Increase Participation.” GAO-04-1029 (September). ———. 2004b. “Trade Adjustment Assistance: Reforms Have Accelerated Training Enrollment, but Implementation Challenges Remain.” GAO-04-1012 (September).
K I M B E R LY A . C L A U S I N G Reed College
The Role of U.S. Tax Policy in Offshoring
U
.S. corporate tax policy affects firms’ choices regarding offshoring in important ways. The decision about how to provide an intermediate good or service is influenced by the tax treatment afforded different modes of provision. In addition, when firms choose to provide an intermediate good or service via in-house offshoring (foreign direct investment), their choice of country as well as their subsequent financial decisions regarding these transactions are influenced by international tax incentives. This paper examines the role of U.S. corporate tax policy in influencing offshoring behavior. It addresses four related questions. First, how does the U.S. tax system operate? I address how the current U.S. tax system is designed and how recent legislative changes have affected that design. Second, what are the incentives provided by this system? I examine how the U.S. system of taxation affects the decision to offshore activities, the choice of foreign country for offshoring operations, and the nature of offshoring transactions. I also briefly describe the extent to which the U.S. system differs from those in other countries. Third, what should an international tax system do? I discuss four potential goals for an international tax system: enhancing efficiency, improving macroeconomic indicators, augmenting external effects associated with multinational activity, and generating government revenue. Finally, given the current U.S. system as well as these goals, what are the merits of suggested policy alternatives? I evaluate several major policy design changes with respect to these policy goals; I also consider pragmatic smaller changes that could improve the functioning of the U.S. tax system.
Research underlying this paper was supported by the U.S. National Science Foundation, research grant #0136293. I am grateful to the discussant, Kevin Hassett; the editors, Susan Collins and Lael Brainard; and the Brookings Trade Forum participants for their insightful comments and suggestions.
457
458
Brookings Trade Forum: 2005
Background There is substantial disagreement, in the literature and in the popular press, over the definition of offshoring. Some treat it as a synonym for outsourcing; those who distinguish between the two differ regarding the activities that fall under each heading.1 My preferred definition of offshoring follows Cronin, Catchpowle, and Hall (2004); they generate a handy 2 x 2 matrix similar to table 1. In this formulation, both foreign direct investment (FDI) and arm’slength offshoring are classified as offshoring. As the following sections of the paper make clear, U.S. international tax policy has two main effects on the decisions of firms in this context. First, tax incentives influence firms’ decisions about the optimal method of providing a particular good or service. Second, if a firm chooses in-house offshoring or foreign direct investment, U.S. international tax policy will influence the firm’s decisions regarding location choices, investment levels, and the prices and quantities of intrafirm transactions. But once a firm has decided to outsource part of its production process at arm’s length, either at home or abroad, tax rules will have relatively little impact on pricing and transactions. The U.S. International Tax System The U.S. government taxes U.S. multinational firms on a residence basis.2 Thus, U.S. multinational firms are taxed on income earned abroad as well as income earned in the United States. However, they receive a tax credit for taxes paid to foreign governments. This tax credit is limited to the firm’s U.S. tax liability, although firms may (in some cases) use excess credits from income earned in high-tax countries to offset U.S. tax due on income earned in low-tax countries. Only income that is repatriated is taxed; thus, firms have an incentive to report income in low-tax countries, where it can grow tax-free before it is repatriated. Firms also typically have an incentive to avoid reporting income in high-tax countries because the tax credit received by the U.S. firm is limited to the U.S. tax liability. As an example, consider a U.S.-based multinational firm that operates a subsidiary in Ireland. Assume that the U.S. corporate income tax rate is 35 percent 1. Bhagwati, Panagariya, and Srinivasan (2004) are careful to define outsourcing as solely arm’s-length trade in services across borders. This makes outsourcing analytically equivalent to conventional trade, and it limits their analysis accordingly. They do not discuss offshoring as a separate phenomenon. 2. Some of the following text that provides background information is also found in Clausing (2005a).
Kimberly A. Clausing
459
Table 1. Means of Intermediate Good or Service Provision At home Within the firm
In-house
Arm’s length
Outsourcing
Abroad In-house offshoring/ Foreign direct investment (FDI) Outsourced offshoring
and that the Irish corporate income tax rate is 12.5 percent. The Irish subsidiary earns 800 and decides to repatriate 70 of the profits to the United States. (Assume, for ease of computation only, a 1:1 exchange rate.) First, the Irish affiliate pays 100 to the Irish government on profits of 800. It then repatriates $70 to the United States, using the remaining profit ( 630) to reinvest in its Irish operations. The firm must pay U.S. tax on the repatriated income, but it is eligible for a tax credit of $100 (taxes paid) times 70/700 (the ratio of dividends to after-tax profits), or $10. This assumes that the U.S. multinational firm does not have excess foreign tax credits from its operations in high-tax countries; if it does, it can use those credits to offset taxes due on the repatriated Irish profits. The remaining profits ( 630) can grow abroad tax-free before repatriation; this process is referred to as deferral. However, U.S. tax law does contain some provisions aimed at discouraging firms from taking full advantage of deferral. Under the Subpart F provisions of U.S. tax law, certain foreign income of controlled foreign corporations is subject to immediate taxation.3 Most important, this includes both income from passive investments and foreign base company income.4 Some countries (such as the United Kingdom and Japan) use a tax credit system similar to that used by the United States. Others (such as France and the Netherlands) exempt foreign income from taxation; this is referred to as a territorial system of international taxation. In theory, multinational firms based in these countries have an even greater incentive to incur income in low-tax countries because such income will not typically be taxed upon repatriation. However, some scholars argue that excess foreign tax credits and deferral blur the distinction between these two systems.5 In addition, several countries have hybrid systems that lie in between these two systems; for instance, a firm’s foreign 3. Controlled foreign corporations are foreign corporations with over 50 percent American ownership, where each owner (an individual or corporation) owns at least a 10 percent stake. 4. Foreign base company income is derived from sales of goods between related parties where the goods are both manufactured outside the base country and sold for use outside the base country. 5. See, for example, Altshuler (2000). De Mooij and Ederveen (2003) find evidence in support of this view.
460
Brookings Trade Forum: 2005
income may be exempt from taxation in its home country provided that the tax system in the foreign country where it operates is sufficiently similar to that in the home country. Recent Changes in the U.S. Tax System Shortly before the 2004 election, the U.S. Congress passed the American Jobs Creation Act; it was signed into law on October 22, 2004. The bill repeals an illegal trade subsidy—the Extraterritorial Income (ETI) exclusion—and creates a number of new tax breaks for business interests, including a deduction for U.S. “production” income. For the purpose of this analysis, the most noteworthy provisions were those enacting changes in the U.S. international tax system. These provisions represent a somewhat subtle shift toward a territorial system of taxing international income in the United States. Many of the changes are complex, given the arcane world of international taxation, but the overall effect is indisputable. The legislation clearly lessens the burden of the U.S. tax system on the foreign income of multinational firms. For example, the legislation contains a provision to allow a temporary tax holiday for dividend repatriations of 5.25 percent. U.S. firms may elect a one-year window during which they may deduct 85 percent of extraordinary cash dividends received from controlled foreign corporations. (This effectively taxes those dividends at 5.25 percent, or 35 percent of 15 percent.) Typically, under U.S. law, when a corporation repatriates income from a low-tax country, it must pay the difference between the U.S. tax rate (35 percent) and the foreign tax rate, although in many cases it can use excess foreign tax credits from affiliates based in high-tax countries to offset the taxes due.6 This temporary provision of the new law would provide a substantial tax advantage to repatriate funds from lowtax countries in the year of the tax break. Although the dividend tax holiday will increase U.S. government revenue temporarily, it will reduce revenue over a longer time frame because firms will have less accumulated income to repatriate at the normal rate. Further, such a provision sends a strange message to U.S. multinational firms, who will in coming years have an incentive to not repatriate income back to the United States in the hope of enjoying similar holidays in the future.
6. Careful tax planning likely explains why U.S. taxes paid on foreign income earned in lowtax countries are often quite small. In addition, firms have been clever at finding alternative methods of effectively repatriating funds without triggering U.S. taxation, such as the use of hybrid corporate structures.
Kimberly A. Clausing
461
The law’s provisions require firms to reinvest repatriated funds in the United States. Firms must make a domestic reinvestment plan (or “drip”) indicating how the funds will be used. The provision was open-ended, and Internal Revenue Service guidance has indicated a very flexible application of the provision. Given the fact that funds are fungible, it is likely that firms will be able to meet the requirements of the law without changing their underlying investment decisions. In addition, even if a substantial inflow of capital results, in our world of mobile capital markets these inflows would likely just displace other inflows of funds from abroad, leaving total investment and interest rates in the United States unchanged. On net, this holiday makes investments in low-tax countries more attractive relative to the prior status quo because there is now the promise of methods for repatriating profits without incurring large tax costs. Further, if such holidays are anticipated for the future, there is a smaller incentive to undercapitalize initial investments in low-tax countries in order to take full advantage of deferral. In addition, many of the other provisions in the legislation allow for a more generous tax treatment of foreign income. There are reductions in the number of foreign tax credit baskets, a change that enables a more efficient use of foreign tax credits to offset tax due on income earned in low-tax countries. Further, there is a more generous carry-forward of excess foreign tax credits, several provisions weaken Subpart F, and there are changes in interest expense allocation rules that should allow firms to use foreign tax credits more efficiently. On net, the Joint Committee on Taxation (2004) estimates that the international tax provisions of the legislation will reduce tax revenue by more than $40 billion over ten years; revenue losses would be larger in the absence of phase-ins. Tax writers admit that they view the bill as taking “baby steps to a territorial system.”7 George Yin, the chief of staff of the Joint Committee on Taxation, has concluded that the American Jobs Creation Act indeed takes the U.S. system of taxation closer to a territorial system, and has speculated that future tax policy could move further in that direction.8
What Are the Tax Incentives behind Offshoring? The effects of the U.S. international tax system that are related to offshoring can be divided into three categories (see figure 1). First, the U.S. international tax 7. Quoted in Glenn (2004a). 8. See Glenn (2004b).
462
Brookings Trade Forum: 2005
Figure 1. Points of Tax Influence on Decisions Regarding Market Provision Tier 1: Decision regarding means of provision
Option B Outsourcing with domestic supplier
Option A In-house domestic provision
Option C In-house offshoring/ foreign direct investment
Option D Outsourced offshoring
Tier 2: Decision regarding location and investment
Country 1
Country 2
Country 3
Tier 3: Financial decisions affecting reported profits in affiliate and parent firms, includng decisions regarding financing and intrafirm transactions
system affects the tier 1 decision regarding how to provide a good or service. Consider, for example, the choice between options A and C. As explained earlier, U.S. multinational firms face a lower tax burden on their foreign income earned in low-tax countries than on their domestic income. Further, operating abroad can also lower the taxes paid on domestic income if, for example, U.S. income is shifted abroad to low-tax destinations.9 Note, however, that the tax preference associated with operating abroad is contingent on picking a low-tax location. Investing in a high-tax location can increase a firm’s global tax burden because U.S. tax credits for foreign income tax paid are limited to the U.S. tax liability.10 Tax incentives also affect the choice between options C and D. The decision about whether to provide a good (or service) in-house via a foreign affiliate or to purchase the good (or service) from an arm’s-length provider has tax consequences. If the good or service is provided within the firm, the affiliate and the
9. Income shifting is discussed in more detail later in the paper. 10. Still, excess tax credits can be handy in offsetting tax due on income earned in low-tax countries; in addition, excess tax credits can be used to offset interest and royalty income from abroad.
Kimberly A. Clausing
463
parent firm can together make pricing and provision decisions in a way that maximizes the firm’s global after-tax profits. If the good or service is provided at arm’s length by an unaffiliated firm, such decisions are a result of two firms’ independent profit-maximizing decisions. The theory of multinational firms has long recognized that multinational activity is contingent on firms having both firm-specific assets and an incentive to internalize transactions (see, for example, Dunning 1974); tax incentives can provide a strong motive for internalization. Gordon and Hines (2002) note that theories of multinational activity based on firm-specific assets (intangible capital) and the facilitation of tax avoidance offer plausible explanations for the observed empirical patterns of multinational activity.11 Second, if the firm has decided to pursue in-house offshoring (FDI) as a mode of intermediate good or service provision, the U.S. international tax system affects the tier 2 decision regarding the location and scale of foreign direct investment. Low-tax countries are more attractive places to locate, ceteris paribus, and it is expected that low-tax locations will therefore attract disproportionate amounts of multinational activity. In reality, the distinction between tier 1 and tier 2 decisions is somewhat artificial since the decision about the location of foreign direct investment is likely made contemporaneously with the decision to pursue in-house offshoring (FDI). Still, tier 2 decisions are simpler to investigate empirically because one can focus on a simple testable hypothesis. As noted below, there is consequently a more extensive literature examining these effects. Finally, tier 3 financial decisions are also affected by tax influences. For example, decisions about the amount and sources of investment funds are likely influenced by tax factors. It may be advantageous for a multinational firm to undercapitalize an initial investment in a low-tax country in order to take maximum advantage of tax savings due to deferral. Undercapitalization provides firms with good uses for reinvested profits, thus allowing profits to grow abroad free of U.S. tax.12 In addition, it may also be advantageous for multinational firms to alter the debt/equity ratios of affiliated firms in high- and low-tax countries in order to 11. For example, models of multinational firms as financial intermediaries suggest that foreign investment should equalize after-tax rates of return across countries. This generates the prediction that before-tax profit rates should be higher in high-tax countries, to compensate for the tax disadvantage. In reality, however, before-tax profit rates in high-tax countries are found to be systematically lower in the literature, evidence of pervasive income shifting. 12. Sinn (1993) and Hines (1994) have demonstrated this result. If firms can reinvest earnings in passive investments that earn competitive returns, however, they will have no incentive to underinvest, as demonstrated by Weichenrieder (1996). Further, there may be plentiful alternative investment opportunities in other active investments as well.
464
Brookings Trade Forum: 2005
maximize interest deductions in high-tax countries and taxable profits in low-tax countries. Further, decisions regarding the prices and quantities of intrafirm transactions will also be affected by tax-minimizing incentives. For example, multinational firms have an incentive to distort the prices on intrafirm transactions in a way that shifts income to low-tax locations. For example, firms can follow a strategy of under- (over-)pricing intrafirm exports (imports) to (from) low-tax countries, following the opposite strategy with respect to high-tax countries. Further, multinational firms have an incentive to undertake more intrafirm transactions if such transactions generate tax savings. Firms are limited in their ability to engage in tax-motivated transfer pricing by fear of detection. Governments generally employ an arm’s-length standard, requiring multinational firms to price intrafirm transactions as if they were occurring at arm’s length. Nonetheless, there is agreement in the literature that this standard leaves substantial room for tax incentives to affect pricing, since arm’s-length prices are often difficult to establish for many intermediate goods. In the context of services offshoring, tax-motivated transfer pricing may be particularly hard to detect, given the intangible nature of many services. Beyond these three types of tax influences on offshoring behavior, there has been some recent concern over multinational firms’ incentive to change their ownership structure in order to minimize their global tax burden. For example, U.S.-based multinational firms may undertake inversions by moving their real or putative headquarters (and their corporate residence for tax purposes) to low-tax countries. Such a change inverts the corporate structure, as the parent firm becomes a subsidiary and the subsidiary becomes the parent.13 In addition, there is concern about an increasing use of “hybrid” entities that allows firms to repatriate earnings from low-tax countries without triggering U.S. tax. Such strategies are complex; some examples are given in Sheppard (2004). What Does the Evidence Suggest? Most of the literature has focused on the second and third tiers of tax influences described above, rather than considering the tier 1 choices among provision
13. Desai, Foley, and Hines (2002) note that inversions are capable of both reducing any U.S. tax due on foreign income (since income earned abroad by non-U.S. firms is not subject to U.S. taxation) and further facilitating tax avoidance on U.S.-source corporate income. Case study evidence regarding actual inversions indicates that market reactions to these events are too large to simply reflect reduced tax payments on foreign income. The authors also find that inversions are
Kimberly A. Clausing
465
modes. Still, decisions between provision options A and C can be considered conceptually equivalent to the decision about where to invest abroad; the only difference is that the home country is one possible location choice. Decisions between options C and D are more difficult to tackle, given data constraints. Several studies have considered the tax responsiveness of foreign direct investment. Altshuler, Grubert, and Newton (2001) find that the elasticity of real capital with respect to the host country after-tax rates of return was 1.5 in 1984, rising to almost 3 by 1992; they attribute the increase to greater globalization. Mutti (2003) also finds that the tax responsiveness of foreign investment increased over time; he too focuses on the period 1984 to 1992. De Mooij and Ederveen (2003) perform a meta-analysis of twenty-five studies that consider this question; they find that the median value of the tax rate semi-elasticity is –3.3, indicating that a 1 percentage point reduction in the host country tax rate raises foreign direct investment by 3.3 percent. They also explicitly note that the elasticities of the reviewed studies increase over time. Reported elasticities were centered at 2.4 in 1987, rising to 3.7 by 2002. Alas, the existing studies in this area typically do not distinguish services from other types of multinational activity, so there is little to report at present regarding whether services activity differs in terms of tax responsiveness.14 Low-tax locations are attractive places in which to locate new real investments, and the ability to reduce worldwide tax burdens by shifting income to lightly taxed locations enhances the attractiveness of such locations. There is substantial indirect evidence of tax-motivated income shifting in the literature. Hines (1997 and 1999a) provides thorough reviews. Owing to data limitations, most previous evidence is necessarily indirect, relying on statistical relationships between country tax rates and affiliate profitabilities or tax liabilities. For example, Hines and Rice (1994) find that 1 percent tax rate differences are associated with 2.3 percent differences in before-tax profitability. A few studies have directly considered the impact of transfer pricing incentives on trade prices. Swenson (2001) uses trade price data, together with variations in country-level tax rates as well as product-level tariff rates, to identify the more likely for large firms, firms with a larger share of foreign assets, highly indebted firms, and firms with affiliates in low-tax countries, findings consistent with theoretical predictions. 14. However, it is worth noting that for U.S.-based multinational firms, offshoring of services by parent firms is not particularly large at present. As Borga notes elsewhere in this volume, less than 1 percent of parent firm purchases of goods and services are imported from affiliates abroad, and this fraction did not increase over the period 1994–2002. Still, both van Welsum and Reif and Jensen and Kletzer, also in this volume, document reasons to suspect that services offshoring could become far more important quantitatively in the future.
466
Brookings Trade Forum: 2005
Figure 2. Where Were the Jobs in 2002? a Percent
14.0 12.0 10.0 8.0 6.0 4.0 2.0 United Canada Mexico Germany France Brazil Kingdom
China Australia Japan
Italy
Source: Data are from the Bureau of Economic Analysis (BEA) web page. See BEA annual surveys, Operations of U.S. Parent Companies and Their Foreign Affiliates (www.bea.doc.gov/bea/ai/iidguide.htm#link12b [March 10, 2005]). a. In 2002, majority-owned affiliates of U.S. multinational firms employed 8.2 million people. This figure shows percentages of the worldwide (non-U.S.) total employment occurring in each of the top-ten employment countries. Thus, each percentage point translates into approximately 82,000 people.
incentives to manipulate transfer prices; results indicate that trade prices are responsive to these tax incentives.15 Clausing (2003) finds evidence of taxmotivated transfer pricing using intrafirm price data from the Bureau of Labor Statistics. Given the breadth and size of the aforementioned tax influences, it is worth putting this information in perspective with a quick overview of where U.S. multinational firms are operating. If one considers real measures of multinational activity, such as employment, assets, or sales, it appears that a primary driving force behind multinational firm location decisions is market access, since most multinational activity occurs in countries with big markets. For example, figure 2 shows the top ten host countries in terms of affiliate employment for U.S. multinational firms. In total, majority-owned affiliates of U.S. multinational firms employed 8.2 million people in 2002. The countries with large employment are the usual suspects in terms of large markets with close ties to the United States. As table 2 indicates, the average effective tax rate paid 15. Effects are statistically significant but quantitatively small. One difficulty with the analysis is the lack of trade price data that separate intrafirm from arm’s-length transactions.
Kimberly A. Clausing
467
Table 2. Top Employment and Income Countries in 2002 Top employment countries United Kingdom Canada Mexico Germany France Brazila China Australia Japan Italy Averagea
Effective tax rate (percent) 31 26 37 27 34 >100 17 29 39 41 31
Top income countries Ireland Bermuda Netherlands United Kingdom Canada Luxembourg Switzerland Japan Mexico Singapore Average
Effective tax rate (percent) 8 2 9 31 26 1 4 39 37 11 17
Source: Data are from the Bureau of Economic Analysis (BEA) web page. See BEA annual surveys, Operations of U.S. Parent Companies and Their Foreign Affiliates (www.bea.doc.gov/bea/ai/iidguide.htm#link12b [March 10, 2005]). Effective tax rates are calculated as foreign taxes paid by U.S. affiliate firms in a given country relative to net (pre-tax) income. a. Brazil’s effective tax rate exceeds 100 percent owing to negative net income and positive foreign tax payments. The average is calculated without this figure.
by U.S. affiliates in these countries (31 percent) is similar to the U.S. statutory tax rate (35 percent). The situation changes if one considers instead the distribution of total profits of U.S. multinational firms across locations, as shown in figure 3. In total, majority-owned affiliates of U.S. multinational firms earned $205 billion of net income in 2002. Six of ten top profit locations (Ireland, Bermuda, the Netherlands, Luxembourg, Switzerland, and Singapore) are not particularly large economies, but they are nonetheless attractive places to earn profits because of their low effective tax rates (averaging 6 percent); none of these countries are top-ten employment countries. Table 2 indicates that the average effective tax rate paid by U.S. affiliate firms for all of the top-ten profit countries is 17 percent. The U.S. System in an International Context Before turning to policy options, it may be useful to consider the U.S. corporate income tax system in the context of the policies of other member countries of the Organization for Economic Cooperation and Development (OECD). Among OECD countries, approximately one-third employ a credit system like that used in the United States, approximately one-third employ a territorial system (exempting foreign income from taxation), and approximately one-third employ a hybrid system with elements of both territorial and credit systems. As figure 4 indicates, the average statutory tax rate of OECD member countries has
468
Brookings Trade Forum: 2005
Figure 3. Where Were the Profits in 2002? a Percent
10.0 8.0 6.0 4.0 2.0
Ireland Bermuda Nether- United Canada Luxem- Switzer- Japan bourg lands Kingdom land
Mexico Singapore
Source: Data are from the Bureau of Economic Analysis (BEA) web page. See BEA annual surveys, Operations of U.S. Parent Companies and Their Foreign Affiliates (www.bea.doc.gov/bea/ai/iidguide.htm#link12b[March 10, 2005]). a. In 2002, majority-owned affiliates of U.S. multinational firms earned $205 billion of net income. This figure shows percentages of the worldwide (non-U.S.) total net income occurring in each of the top-ten income countries. Thus, each percentage point translates into approximately $2 billion of net income.
declined over the previous quarter century, from 43 percent in 1979 to 29 percent in 2002.16 By 2002 the U.S. corporate tax rate was nearly one standard deviation above the OECD average; since 1988 the U.S. rate has been gradually increasing relative to rates in other OECD countries. At the same time, the United States collects relatively little revenue from the corporate income tax, as figure 5 demonstrates. The U.S. corporate tax generates about half as much revenue (1.43 percent of GDP in 2002) as that of an average OECD country (2.9 percent of GDP in 2002).
Policy Goals for the U.S. International Tax System The current system of U.S. international taxation favors foreign income earned in low-tax countries. Multinational firms can lower their worldwide tax burden by undertaking real operations in low-tax countries and by employing
16. Hines (2005) indicates that both small and large countries have lowered their tax rates over this period. Further, the tendency for small countries to have lower corporate income tax rates than large countries at the beginning of this period (1982) had disappeared by the end of this period (1999).
Kimberly A. Clausing
469
Figure 4. OECD Statutory Corporate Income Tax Rates, 1979–2002 a Percent
50
United States
Plus one standard deviation
40 Average 30 Minus one standard deviation 20
10
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
Source: Tax rate data come from annual publications of PricewaterhouseCoopers, Corporate Taxes: Worldwide Summaries. a. Averages, +/– one standard deviation, and U.S. rates.
financial strategies that allow them to shift income toward low-tax countries. These incentives likely influence both the extent and nature of offshoring activity. Before considering policy alternatives in detail, one should ponder the ideal objectives of the U.S. international tax system. Four possible policy goals are discussed below: enhancing efficiency, improving U.S. macroeconomic indicators, garnering more beneficial external effects associated with multinational activity, and enhancing government revenue.17 Enhancing Efficiency Traditional tax theory (starting with Richman 1963) suggests that the optimal allocation of worldwide investment will result when multinational firms allocate investment irrespective of tax treatment, a goal referred to as capital 17. Other important policy goals include equity (both horizontal and vertical), administrative simplicity and ease of compliance, enforceability, and conformity with international norms.
470
Brookings Trade Forum: 2005
Figure 5. OECD Corporate Income Tax Revenue relative to GDP, 1979–2002a Percent
4.5 4.0 3.5 Plus one standard deviation 3.0 Average 2.5 2.0 1.5 United States 1.0 Minus one standard deviation 0.5
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
Source: Corporate tax revenue data are from the OECD revenue statistics. GDP data are from the World Bank’s World Development Indicators database (www.devdata.worldbank.org/dataonline). a. Averages, +/– one standard deviation, and U.S. data.
export neutrality. Under typical assumptions, this would require that governments tax foreign income as it is earned, providing unlimited tax credits for tax payments to foreign governments.18 Such a tax system would ensure that investors did not take tax considerations into account when making their international location decisions, as the resulting income would be taxed at the same rate irrespective of their decision. Thus, decisions would be made based on a comparison of before-tax rates of return, and investments would flow toward those locations with the highest return, thus providing the greatest economic gain from a worldwide efficiency perspective. Further, there would be no incentive to shift income to more lightly taxed locations, as all locations would be taxed equally from an investor’s perspective. 18. Several researchers have offered extensions to the classic capital export neutrality result, such as recognizing that countries’ tax-setting policies may be interdependent, allowing savings to be endogenous, and allowing positive economic spillovers from foreign investment. The first and third of these extensions are discussed later in this paper; Hines (1999b) reviews all of these extensions in greater detail.
Kimberly A. Clausing
471
Of course, neither the U.S. credit system nor a traditional exemption system meets this description. Under the U.S. system, tax credits are not unlimited, and foreign income is not taxed as earned, but upon repatriation. These two features make investors responsive to tax rate differences across countries in both their real and financial decisions. Under an exemption system, foreign income is not taxed at all, further heightening the incentive to respond to tax differences across countries. Desai and Hines (2004) advocate focusing instead on the concept of capital ownership neutrality, which would require that tax rules not distort ownership patterns. They argue that such a goal is consistent with theories of multinational activity that focus on ownership advantages as a key impetus for foreign direct investment. Conformity among tax systems would promote capital ownership neutrality, although conformity could take several forms. Improving U.S. Macroeconomic Indicators Some observers maintain that the U.S. international tax system could better promote the health of U.S. macroeconomic indicators, including the level of output and jobs in the economy as well as U.S. international financial balances. Basic macroeconomic theory suggests that the level of output and jobs fundamentally depends on such variables as the capital stock, the labor force, and the technological possibilities of the economy. Thus, if the international tax system is to affect the amount of output in the economy, it likely does so by enhancing the capital stock. A focus on enhancing the domestic capital stock might suggest that tax policy should favor U.S. investments relative to those in other locations. The national neutrality doctrine implies that in order to maximize the national gain from the worldwide investment of U.S. capital, the tax system should favor U.S. investments by treating foreign tax payments as a (deductible) expense associated with doing business abroad. Such a goal would require current taxation of foreign income, with only a deduction allowed for taxes paid to foreign governments.19 Such a policy also has two rather obvious drawbacks. First, under this system, double taxation of income in two countries is likely, as (for example) U.S.
19. The analog to capital ownership neutrality is here national ownership neutrality, which implies, according to Desai and Hines (2004, p. 26), “that countries should want to exempt foreign income from taxation . . . [because] countries have incentives to select tax rules that maximize the productivity of foreign and domestic investment.”
472
Brookings Trade Forum: 2005
multinationals would pay income tax twice on income earned abroad. From a worldwide perspective, this system would lead to an inefficiently low level of foreign investment, as investments abroad would be tax disadvantaged relative to investments at home. Second, such beggar-thy-neighbor tax policy could encourage similar policies by other governments, reducing foreign investment in the United States and further harming both world and national welfare. Even if one rejects the doctrine of national neutrality, as most economists and policymakers do, that still leaves many analysts with the desire to enhance the U.S. “competitive” position via international tax policy. For example, some have argued that the U.S. tax system puts U.S. goods at a disadvantage relative to goods from countries that rely more heavily on value-added taxes (see, for example, Hufbauer 2002). Value-added taxes (VATs) are typically charged on a country’s imports, while exports receive VAT rebates. Still, while it would appear that a VAT system would encourage exports relative to imports, there are both theoretical and empirical reasons to doubt this conclusion. First, economists have long recognized that exchange rate changes should offset these VAT effects; in particular, currency appreciation will undo any export advantage provided by a VAT.20 Further, recent empirical research by Desai and Hines (2003) indicates that countries that have VATs or that rely on VATs for more revenue actually have lower export performance than other countries. This finding holds even if one controls for country specific attributes such as GDP, income per capita, geography, and tariffs. The United States is faced with persistent trade and current account deficits, so perceived tax solutions to this problem will continue to be attractive. However, it is quite unlikely that border tax adjustments, a move toward a VAT, or export tax incentives such as the ETI provisions will have an effect on U.S. financial balances. If U.S. policymakers want to take serious steps to address trade imbalances, the logical first step would be to address government fiscal imbalances, as basic national income accounting analysis demonstrates.21 Garnering Beneficial External Effects Associated with Multinational Activity Those who are concerned with watching over the competitiveness of U.S. multinational firms often assume that U.S.-based multinational firms generate 20. See, for example, the lengthy list of citations to such studies in Viard (2004). 21. See Clausing (2005b) for a more detailed demonstration of this result.
Kimberly A. Clausing
473
external benefits for the U.S. economy. If such external effects are important, it may make sense to favor foreign income in order to ensure that U.S.-based multinational firms are not at a disadvantage relative to competitor firms from other countries that tax foreign income more lightly or not at all. An exemption system may be suitable for meeting such goals. Such a system meets the standards of capital import neutrality, a doctrine that focuses on ensuring that all firms in a given location are treated equally for tax purposes. If home countries do not tax the foreign income of their multinational affiliate firms, then affiliate firms will not be disadvantaged when competing with other firms in low-tax countries. Thus, capital import neutrality generally puts the international competitiveness of a country’s multinational firms ahead of considerations regarding optimal investment location or government revenue. For example, capital may be allocated inefficiently toward low-tax locations because after-tax rates of return in such locations are higher. This doctrine raises the question of why governments should care about the international competitiveness of domestically owned multinational firms. Rationales usually include beneficial spillover effects due to the domestic location of headquarters or research and development activities. Also, some evidence suggests that foreign multinational activity may encourage domestic activity. Still, leaving such activity untaxed in the hope that there will be beneficial external effects begs a comparison with other activities in the economy that provide beneficial external effects. Further, usually there are more direct methods of encouraging the external effects in question, such as subsidization of research and development. In addition, it is important to keep in mind that the effective tax burden on the foreign income of U.S. multinational firms is likely to be low. For example, Grubert and Mutti (1995) undertake a series of calculations that indicate that the effective U.S. tax rate on active foreign income in 1990 is quite low, approximately 2.7 percent, or less under some assumptions.22 Such low tax rates imply that the competitiveness cost of the U.S. tax system may be far less than one might infer from a simple interpretation of the parameters of U.S. law. 22. For example, lower tax rates result if one accounts for unrepatriated foreign income in the denominator of the tax payments/income fraction. Other adjustments allow royalty and sales source income to count as domestically sourced income rather than foreign source income, adjust for artificial income allocation rules, and allow for worldwide fungibility. Grubert (2001) and Altshuler and Grubert (2001) support similar conclusions using 1996 data. Desai and Hines (2004) infer that the tax burden on foreign income is rather higher; their calculations are based on far less detailed data.
474
Brookings Trade Forum: 2005
Enhancing Government Revenue As noted above, the current U.S. corporate tax system, including the American Jobs Creation Act, generates a paradox. The U.S. statutory corporate income tax rate (35 percent) is now well above the OECD average (29 percent). Still, despite this relatively high statutory tax rate, the U.S. corporate tax generates about half as much revenue (1.43 percent of GDP in 2002) as the average OECD corporate tax (2.9 percent).23 Figure 5 indicates that the United States experienced a reduction in corporate income tax revenue relative to GDP over a period (1979–2002) when the average OECD country experienced an increase in corporate income tax revenues. There is very little evidence regarding how the tax system itself affects government revenue. However, Clausing (2004) suggests that countries with a tax credit system raise more revenue than those with a territorial system, controlling for other factors that are likely to influence corporate tax revenues (such as corporate profitability, the corporate share of the economy, and the statutory tax rate). Tax credit countries receive approximately one percentage point more corporate income tax revenue relative to GDP than territorial countries, a substantial difference. Still, with any tax system design, the devil is in the details, and a move to a territorial system in the United States could presumably be designed in a way that ultimately raised revenue, as Grubert and Mutti (2001) postulate. Whither Corporate Taxation? Some economists view the corporate income tax itself as an anachronism. It has long been recognized that corporate taxation ultimately results in the taxation of individuals, and thus corporate taxation may lead to the double taxation of corporate profits, as individuals are also taxed at the personal level on dividends and capital gains. Still, the corporate income tax also acts as a backstop for the personal income tax, particularly for high-income individuals. Dividends and capital gains are often taxed preferentially (in the United States, both are currently taxed at a rate of 15 percent or below, while the top income tax bracket is 35 percent), and capital gains income is taxed only upon realization, thus allowing income to grow tax-free in the interim.24 Income shifting between the
23. In 2003, U.S. corporate income tax revenues fell further, to 1.2 percent of GDP. Numbers for 2002 are used in the text to facilitate comparison with other OECD countries. 24. Short-term capital gains, or gains on assets held less than a year, are not taxed preferentially.
Kimberly A. Clausing
475
personal income tax base and the corporate income tax base is likely important, as Gordon and Slemrod (2000) demonstrate. In this context, eliminating the tax on corporate income could have serious negative consequences for the broader integrity of the U.S. tax system. Many high-income individuals have substantial flexibility in how they receive their income, and such individuals would be able to shift wage income toward the corporate sector, enabling a substantial erosion of the personal income tax base. Gordon and Makie-Mason (1995) note that despite the theoretical prediction that small open economies should not tax corporate income, income shifting between the personal and corporate income tax bases generates an important role for corporate income taxation. In fact, in order to avoid such domestic income shifting, theory suggests that the top personal and corporate tax rates should be the same. Income shifting between domestic and foreign corporate tax bases also presents a problem that increases with tax rate differences between countries; such considerations may provide for a lower corporate tax rate. Other arguments for the corporate income tax include the presence of administrative difficulties associated with taxing capital income on an individual level, the existence of pure profits or rents associated with corporate activity, and the benefit principle. Further, there may be a role for corporate income tax coordination among countries. The theoretical literature regarding tax competition among countries indicates that competition to attract mobile corporate income tax bases can lead to suboptimal outcomes. Evaluating Policy Criteria in the Context of Offshoring The decision of a U.S. multinational firm to offshore a portion of its activities can be considered in the context of these policy objectives. One is more likely to object to offshoring activities, or the nature of such activities, to the extent that such activities are viewed as (a) occurring for reasons other than economic efficiency, (b) harming U.S. macroeconomic indicators, (c) reducing external gains associated with U.S. multinational production, and (d) undermining the U.S. government’s tax base. However, the mere existence of offshoring can work in either direction with respect to these policy objectives. Changes in the U.S. tax system that work toward enhancing the potential for offshoring to be compatible with the four policy objectives of efficiency, macroeconomic activity, beneficial external effects, and government revenue are particularly desirable. The following section reviews possible policy changes in this context.
476
Brookings Trade Forum: 2005
Policy Options for Taxing U.S. Multinational Firms Several factors suggest a need for changes to the U.S. system of taxing multinational firms. This section first reviews the rationale for change and then discusses several policy options, including (a) continued small steps toward a territorial system of taxing foreign income, (b) wholesale adoption of a territorial system, (c) proposals to eliminate deferral of U.S. taxation on foreign income earned in low-tax countries, and (d) increased international coordination of corporate tax policy. Finally, I consider pragmatic policy changes that could improve the functioning of the current system without changing it in a fundamental manner. Any of these changes is likely to affect firms’ decisions regarding offshoring. Not only will changes affect the relative attractiveness of choosing offshoring relative to alternative modes of intermediate good or service provision, but tax policy changes will also affect the character of offshoring decisions once they are made, including the location choice among countries as well as the structuring of intrafirm transactions. Why Change? There are several problematic aspects of the current system of taxing U.S. multinational firms. First, the U.S. statutory tax rate is nearly one standard deviation higher than the average rate of OECD countries, as documented in figure 4. This tax differential provides a large incentive for U.S. multinational firms to locate real economic activity in other countries, as well as a large incentive for U.S. multinational firms to shift income toward more lightly taxed locations. The strength of these incentives likely increases the extent of in-house offshoring activities. To the extent that these activities occur for tax purposes, the efficiency of such activities is in doubt. There may also be concerns associated with a potential loss of U.S. economic activity, although the nature of such effects is ambiguous. Second, the U.S. corporate tax system collects relatively little revenue in comparison with that collected by comparable OECD countries. Low revenues from this source imply some combination of higher deficits, lower public good provision, and higher required revenues from other sources. Low corporate tax revenues are likely due to a combination of factors, including an uncompetitive tax rate, an increasingly narrow tax base, and increasing opportunities for tax avoidance. Tax avoidance opportunities are increasing owing to the increasing globalization of U.S. business and constant innovations in the provision of tax
Kimberly A. Clausing
477
Figure 6. FDI Stocks/GDP for the United States, 1980–2002 a Percent
30
25
20
15
10
5
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
Source: Data are from United Nations Conference on Trade and Development (www.unctad.org). a. Data on FDI include both inward and outward stocks.
shelters. Figure 6 provides one indication of increasing globalization; foreign direct investment stocks have increased dramatically as a share of GDP since 1980. This trend accompanies the increased offshoring activity documented elsewhere in this volume. Third, the U.S. international tax system would benefit from dramatic reductions in complexity. The Tax Reform Act of 1986 greatly simplified the U.S. tax system while reducing tax rates and broadening the tax base. However, the U.S. corporate tax system has grown increasingly complex and loophole ridden. One stated objective of the American Jobs Creation Act was to simplify the taxation of international income, yet many practitioners argue that the resulting system is even more complex.25 Finally, a host of other concerns have been raised. Some observers worry that the U.S. tax system undermines the competitiveness of U.S. multinational firms. 25. See, for example, Taylor (2005).
478
Brookings Trade Forum: 2005
For reasons explained in the previous section, these concerns may be overstated, but they nonetheless drive many of the policy prescriptions below. Some observers additionally feel that the very existence of a corporate income tax is anachronistic and advocate a greater reliance on other sources of government finance, such as consumption taxes. As discussed above, if such a move were contemplated, it would be important to consider how such a change would affect the integrity of the broader tax system. More Steps toward a Territorial System? The American Jobs Creation Act of 2004 (AJCA) took the U.S. tax system closer to a territorial tax system by providing a temporary tax holiday for repatriating dividends from low-tax countries and by taking several steps to lighten the taxation of foreign income of U.S. multinational firms. Some have suggested that U.S. policy may continue to move in this direction (see, for example, Glenn 2004b). By lightening the taxation of foreign income, the AJCA and similar steps in that direction increase the incentives of multinational firms to offshore activities to low-tax countries and to shift their worldwide income to low-tax destinations. Further, it is not clear that the current taxation of foreign income of multinational firms is unduly burdensome. As the calculations of Grubert and Mutti (1995) and Altshuler and Grubert (2001) demonstrate, the effective tax burden on the foreign income of U.S. multinational firms is already quite low, implying that further steps to lighten the tax burden on foreign income may be unwarranted. In addition, one might be concerned about the incentives created by temporary tax measures such as the one in the AJCA. Temporary tax breaks for dividend repatriation are likely to boost repatriations in the year in question, but they will likely lessen repatriations in other years. In fact, relative to the status quo absent temporary tax breaks, firms have a lower incentive to return funds to the United States in future years, as they may prefer to postpone repatriation until a more favorable tax environment arises. Complete Adoption of a Territorial System Relative to the more incremental nature of the AJCA, a complete adoption of a territorial system would have both advantages and disadvantages. In its favor, such a change could simplify the U.S. system of taxing international income: exempting foreign income from taxation would reduce the need for a great deal
Kimberly A. Clausing
479
of tax planning effort. However, many aspects of the current complexity of the international tax system would be retained, including the need to distinguish foreign and domestic income, to distinguish passive income, to allocate expenses appropriately, and to determine transfer prices. Such a system would also likely enhance the competitiveness of U.S. firms in low-tax countries, potentially increasing external benefits associated with multinational activity in the United States. In addition, depending on how such a system was designed, it need not lose revenue relative to the status quo. For example, as Grubert and Mutti (2001) point out, such a change could be accompanied by the increased taxation of interest and royalty income from abroad, as well as changes in interest allocation rules that could reduce tax deductions on income earned in the United States. However, it is important to point out that a move to a territorial system could also undoubtedly be designed in a manner that loses government revenue, and there is no reason a priori to make the assumption that revenue would be gained. Further, evidence in Clausing (2004) suggests that, for the OECD countries as a group, exemption countries raise less revenue from their corporate income tax than countries with a tax credit system, controlling for other factors that are likely to affect revenues. An additional argument against adoption of a territorial system is that exempting foreign income from taxation strengthens the tax preference favoring offshoring and other economic activity in low-tax countries. In addition, such a change would strengthen the incentive to shift income toward low-tax countries. Such tax-motivated changes in behavior are unlikely to be consistent with economic efficiency, and there is the potential for such behavior to further erode the U.S. corporate income tax base.26 Eliminating Deferral Various observers have suggested eliminating deferral of U.S. taxation on unrepatriated income earned in low-tax countries. One such proposal was made 26. Previous literature is equivocal on the question of whether investors from territorial (or exemption) systems are more sensitive to taxation than those from tax credit systems such as the United States. In their meta-analysis of the previous studies on taxation and foreign direct investment, de Mooij and Ederveen (2003) find little difference in the tax sensitivity of FDI under the two systems. However, Hines (1996) finds that investment across U.S. states is sensitive to whether the home country taxes foreign income on a territorial or tax-credit basis. Gropp and Kostial (2000) also find that tax sensitivity is larger for investments sourced in territorial countries.
480
Brookings Trade Forum: 2005
by presidential candidate John Kerry in 2004.27 A complete elimination of deferral would bring the U.S. tax system closer to the ideal of capital export neutrality because there would no longer be an incentive to earn income in low-tax countries or to shift profits to such locations.28 These actions would therefore be attractive to those who are concerned that the favorable tax treatment of foreign income encourages offshoring and an inefficiently large presence in low-tax countries. Such a change would also likely increase government revenues. However, such a change would exacerbate concerns about the international competitiveness of U.S.-based multinational firms, as U.S. firms would face a tax disadvantage relative to firms based in other countries when operating in low-tax markets. Further, such a change could encourage corporate inversions, although legislative actions could be taken to prevent this response. Eliminating deferral thus raises the possibility of reduced external benefits associated with U.S. multinational activity abroad. Still, if the elimination of deferral were accompanied by a large reduction in the U.S. corporate income tax rate, competitiveness concerns would be ameliorated. This combination of policies would also have the advantage of providing a more even tax treatment for foreign and domestic activity, and it would thus reduce the incentive to undertake tax-motivated real or financial decisions. In addition, this combination of policies could be designed to be revenue neutral. International Coordination Efforts to share information across countries or to reduce harmful tax competition from tax haven countries could prove successful in protecting the U.S. tax base and reducing inefficient tax-motivated activities. As one more extreme example, some policymakers in the European Union have advocated corporate income tax harmonization among European Union countries. A harmonization of tax rates as well as tax base definitions would ensure both capital export neutrality and capital import neutrality among European location options (though not worldwide), eliminating within-Europe concerns regarding tax competition, efficiency distortions, and international competitiveness. 27. The Kerry proposal would have limited deferral unless the income earned abroad was generated from production and sale in the country where the affiliate was based. This proposed change would therefore have applied to a minority of foreign income. Further, such a partial limit on deferral would create some problematic incentives to relabel income earned from out-of-country sales, or alter prices on within-firm transactions in order to make out-of-country operations appear unprofitable. 28. There would still be an incentive to avoid earning income in high-tax countries, assuming that U.S. tax credits for foreign income taxes paid were still limited to the U.S. tax liability.
Kimberly A. Clausing
481
Still, such a policy change is politically extremely difficult even among European Union countries, and likely near impossible among OECD countries, much less worldwide. In addition, there may be good economic reasons for countries to choose different corporate tax rates, including variations in preferences, revenue needs, openness, and country size. Further, under some tax competition models there may be gains associated with tax competition (see Wilson 1999 for one review). Pragmatic Improvements to the Current System There are several steps that would likely improve the functioning of the current U.S. system of taxing international income without entailing a fundamental shift in policy. For example, the statutory corporate income tax rate could be lowered. This change might not reduce government revenue—even absent corresponding base broadening—because the United States is likely to the right of the revenue-maximizing point on the corporate income tax Laffer curve.29 In addition, any further reduction in the corporate income tax rate could likely be financed in a revenue-neutral fashion by broadening the tax base. This would have the additional benefit of reducing distortions to activity brought about by the narrow tax provisions in the current U.S. tax code.30 For example, the most noteworthy effort to broaden the U.S. corporate income tax base, the Tax Reform Act of 1986, was designed to increase revenues, even accompanied by a reduction in the tax rate from 46 percent to 34 percent.31 These two changes in tandem would achieve many useful objectives. First, economists generally agree that a tax with a broad base and a low tax rate is preferable (in terms of efficiency cost per dollar raised) to a tax with a narrow base and a high tax rate. Second, it would reduce the tax distortions associated with investment choices between the United States and other countries (as the U.S. tax rate would be closer to that of other countries), thus likely increasing the 29. Clausing (2004) charts such curves for OECD countries over the previous quarter-century. Corporate tax revenues are related to the tax rate, the square of the tax rate, corporate profitability, the share of the corporate sector in the economy, and other control variables. It appears that the revenue-maximizing corporate tax rate is approximately 30 percent over the later part of the sample period. 30. For example, the domestic production deduction in the recent American Jobs Creation Act of 2004 favors domestic production activity. This measure likely encourages such activity as well as an ample amount of accounting effort directed at generating more tax-favored activities on firms’ books. 31. See Auerbach and Slemrod (1997) for a detailed description of the economic effects of this tax reform.
482
Brookings Trade Forum: 2005
efficiency of worldwide investment allocation and offshoring decisions. Third, U.S. investments would be more attractive than they were previously, and there would be a reduced incentive to shift income to other countries. This should increase economic activity and profits in the United States. Fourth, it would increase the competitiveness of U.S.-based multinational firms because our tax burden on foreign income would be lower owing to a lower U.S. tax rate. This has the potential to increase any external benefits associated with U.S. multinational firms’ activities, as well as to reduce the incentive to undertake corporate inversions. With respect to offshoring, these tax changes would help ensure that offshoring occurs for reasons that are consistent with economic efficiency and the U.S. national interest, rather than being motivated by large tax rate differentials.
Conclusion U.S. corporate tax policy taxes the worldwide income of U.S. multinational firms, granting tax credits for taxes paid to foreign governments. Because profits are taxed only upon repatriation to the United States, the tax system provides an incentive to locate real economic activity as well as profits in low-tax countries. Further, recent changes in tax law under the American Jobs Creation Act of 2004 strengthen these incentives. In total, the U.S. tax system provides firms with incentives to undertake in-house offshoring of activities in low-tax countries, and it likely affects the nature of offshoring transactions in important ways. Previous empirical work provides abundant evidence that U.S. multinational firms are responsive to tax influences when choosing investment locations and when structuring intrafirm transactions. Any system of taxing multinational firms will reflect a compromise among the often competing goals of ensuring efficient worldwide capital allocation, protecting the competitiveness of U.S. multinational firms, and seeking government revenue. Current U.S. tax laws do not satisfy these criteria particularly well and are also extraordinarily complex. While dramatic changes to the structure of U.S. tax laws should be made with caution, proposals to lower the corporate tax rate, broaden the tax base, strengthen enforcement, and simplify the tax system deserve close attention, since such changes would improve the performance of the U.S. tax system with respect to all of these criteria. Further, such changes can help ensure that offshoring activities occur in a manner that enhances efficiency and is consistent with the national interest.
Comment and Discussion
Kevin A. Hassett: This is a terrific and concise review of the literature about the impacts of U.S. tax policy on multinationals’ offshoring decisions. The international tax literature is notably arcane and difficult to follow, and this effort is sure to become a must-read (and a must-assign) for tax and trade economists. My role as a discussant is unusually difficult because the author has done such an excellent job reviewing the current state of play in the relevant literature. Often, however, once one pauses to observe a complete body of work, one perceives shortcomings that were less visible before an able review effort. This paper by Kimberly Clausing provides the necessary ammunition to take a critical look at the current international tax literature that relates to offshoring. I will focus my critical remarks on the shortcomings and omissions of the literature, rather than those of the author, except to say that a more skeptical view of the literature than has been adopted by the author is probably justified. I focus on two major problems in my remarks. First, much of the international tax literature has, since the seminal work of Richman (1963), emphasized the important role that capital export (or import) neutrality, an international tax design in which firms make their investment location decisions without consideration of the international tax consequences, has in the design of an optimal tax. Reform proposals are often weighed against the ideals of neutrality, as is done in this paper, and policy prescriptions made based on this analysis. The problem with this approach is that the intuitive neutrality conditions that are so ubiquitous in the international tax literature are not actually justified by results derived from modern and well-developed models. If intuition were 483
484
Brookings Trade Forum: 2005
enough, economists would not need models. Sometimes, intuition can lead one astray, as is most likely the case in this application. The study of optimal taxation in dynamic economies has made enormous gains in recent years, following the pioneering work of Judd (1985) and Chamley (1986). The international tax literature has, for the most part, fallen far behind the curve. This optimal tax literature has almost uniformly produced the result that the optimal marginal tax on capital income is zero. Now consider a many-country version of the basic Judd model. It would certainly (absent possibly realistic complications) also have the characteristic that capital taxes should be zero, and this should be true across all countries. This world would be optimal, it would be harmonized, and it would be neutral with respect to crossborder flows. But now suppose that the social planner does not control the policy variables of each country. At that point, the social planner is in a dynamic game. Given the tax policies of others, she must evaluate her optimal reaction function. It is wholly possible, indeed even likely, that in this second-best world, an individual country could increase its welfare by abandoning capital export neutrality. Suppose, for example, that a country has a high relative corporate tax rate, but it enacts a law allowing firms to defer taxation by leaving their profits overseas. Companies could then lower the effective rate on domestic activity through transfer pricing by setting up a foreign subsidiary in a low-tax jurisdiction. The reduced capital tax could well improve domestic welfare, depending on the complementarity of foreign and domestic activity. Today the United States functions in a world filled with nonoptimal taxes, and in addition, the United States is approximately the high-tax country. In a recent review of the literature on the benefits of consumption taxation, Auerbach and I (2005) concluded that the literature suggests that very large first-order growth effects could be achieved through a consumption tax reform here in the United States. To the extent that international considerations reduce the suboptimal U.S. tax, they could well provide large first-order benefits. The empirical literature described by Clausing suggests that this is the case. Indeed, there is a strong negative correlation between corporate tax rates and revenues across countries. In a world with mobile capital, apparently, the gains to having rates below those of your neighbor are enormous. It would be an error to ignore such gains when designing U.S. policy because of capital export neutrality. To repeat, it is incorrect to assert that capital export neutrality is advisable today in the United States. Such a prescription has not been demonstrated in a formal model, nor is it likely to be. It is also incorrect to argue, as the author does, that harmonization is proscribed by economic theory. Again, countries
Kimberly A. Clausing
485
may be using the excuse of tax competition to reduce capital taxes to their previously politically infeasible Judd-Chamley levels. The author appears to discount this possibility because low capital taxes may inevitably lead to avoidance activity, even asserting that corporate taxes should be as high as wage taxes. If one considers the simple design of, say, the HallRabushka flat tax, however, then one can easily visualize a consumption tax system that is as enforceable as the current one (see Hall and Rabushka 1985). One key issue raised by this (and one that has been addressed in a valuable recent paper by the author, Clausing 2003) is transfer pricing. There is likely enormous variation in the effectiveness with which countries enforce arm’slength pricing. This makes it very difficult to calculate effective tax rates. It may well be that the United States is a low-tax country because transfer pricing to tax havens is so easy, and that this increases domestic welfare. Given the important issues raised here, a top priority for the empirical literature should be an effort to identify cross-country variation in effective capital taxes after transfer pricing is accounted for. This is particularly important for white-collar workers, as tax evasion schemes typically involve locating intellectual property overseas. I have a second theoretical quibble with the literature. The production theory is poorly developed. The author cites several papers that estimate the elasticity of capital with respect to the host-country corporate tax rate. Much of this literature, however, applies the Allen two-variable elasticity of substitution. But Blackorby and Russell (1989) have demonstrated that the Allen elasticity of substitution is a flawed concept. It provides no information on the ease of substitution, curvature of the isoquant, or relative factor shares for cases in which there are more than two inputs. The examples where the Allen elasticity leads one astray tend to be exactly the type that might apply to multinationals—for example, functions with separable subfunctions and more than two inputs. The literature needs to get its production theory straight and use the Morishima (1967) elasticity of substitution, which provides a better measure of the ease of substitution. Until it does, we will not have enough empirical information to assess the essential substitution issues. In conclusion, the international tax literature has provided significant reducedform information about the impact of taxes. This literature suggests that taxes are very important to the offshoring decisions of multinationals. It also shows that many countries may be on the wrong side of the Laffer curve. The connections between these observations and optimal tax design in a dynamic open economy need to be worked out more fully than has been done in the existing literature. Until that is done, it will be difficult to evaluate the impact of taxation on incentives to locate activity abroad, and on the welfare of people at home.
486
Brookings Trade Forum: 2005
General Discussion: The two presentations sparked a debate among the conference participants on whether international corporate tax differentials are important drivers of real multinational activity. There was also great interest in how the tax implications might differ for services and manufacturing, where most of the empirical research has so far focused. James Markusen noted that recent empirical research on the subject of multinational enterprises concludes that taxes are a second-order determinant of investment location. Local-market sales account for the overwhelming proportion—approaching 80 percent—of sales by U.S.-owned multinational affiliates. It therefore appears that U.S. multinationals determine location primarily on the basis of their sales. Indeed, outflows from the United States appear highly inelastic with regard to the U.S. tax rate. Markusen noted that taxes likely do nonetheless affect location decisions, but that this substitution occurs within regional blocks. Thus, for example, if Ireland lowered its tax rate, the change would affect product location among European countries but have a negligible effect on the United States. Dalia Marin agreed with Markusen that taxes do not appear to be the critical driver of location decisions by multinationals but suggested that taxes are very important in determining whether the offshored activity is done in-house or outside the firm. Marin noted that, to the extent that the form of foreign involvement has different labor market outcomes, tax policies may influence domestic unemployment rates. Empirical evidence for German firms investing in eastern Europe suggests that offshoring investments in the “New Europe” (the former accession countries) tend to improve the competitiveness of German firms worldwide and thus tend to increase their labor market. Outsourcing, in turn, does not appear to have the same positive labor market effect. Kevin Hassett challenged Markusen’s notion of regional separation. In the past, he argued, a company would establish itself in the United States to secure access to the enormous domestic market, which conferred market power. In essence, this provided the United States with a property right, allowing it to collect higher corporate taxes than other countries. In such a model, big countries are able to maintain high tax rates, while the smaller countries are not. As a result, a significant dispersion in tax rates among countries could exist. Although this largecountry advantage was operative as recently as ten years ago, Hassett argued that it has now largely disappeared. In the dispersion of tax rates among nations, large countries have begun to look more like small countries. Catherine Mann noted that, under particular circumstances, there appears to be an important interaction between taxes and the data that are classified as crossborder transactions in services. When corporations move their headquarters to
Kimberly A. Clausing
487
low-tax jurisdictions, such as Bermuda or the Cayman Islands, service transactions that used to be counted as exports and imports associated with the headquarters will suddenly appear to shift location, even though the actual production location has not changed. It may appear, for example, that the United States is importing a tremendous number of services from Bermuda, when only the headquarters has moved to Bermuda. In some cases, the shifts are sufficiently large to stand out in the data. A similar problem arises when a corporation shifts the location where an intellectual property asset is considered to be owned. This might lead to serious misinterpretations of research findings in cases where such data are used as a proxy for offshoring Rafiq Dossani asked whether the issue of taxes had any bearing at all on services offshoring. To the extent that India and China merely export certain services that have been offshored there, the transactions entail only costs and no revenues or profits—potentially making the tax implications of the offshoring choice irrelevant. Lael Brainard noted that the difference between services and manufacturing in the ability to precisely measure output has important implications for how corporations make multinational location decisions on the basis of international corporate tax differentials. One of the most salient features of offshoring in services as compared with manufacturing is that the data are much less “hard.” By their nature, the manner and location in which services transactions are recorded allow for significantly greater discretion than in manufacturing, particularly when the choice is internal to the firm, and such discretion is particularly valuable for tax reporting purposes. Thus, there can be an even greater divergence between where profits are reported for tax purposes and where the underlying real activity takes place. Kent Hughes wondered whether the system of government restraints on transfer pricing would be completely broken with this enormous amount of discretion and whether there is a more effective approach. Kimberly Clausing agreed that services provided an even greater role in many cases for intangibles and thus for firm discretion in pricing because of the absence of a well-defined arm’s-length market. And the ability to shift profits to avoid taxation goes well beyond transfer pricing to include triangulating strategies, how affiliates are financed, and where deductions and earnings are booked. For tax purposes, where a firm books its profits will always be more important than where it undertakes its real activities.
488
Brookings Trade Forum: 2005
References Altshuler, Rosanne. 2000. “Recent Developments in the Debate on Deferral.” Tax Notes (April 10): 255–68. Altshuler, Rosanne, and Harry Grubert. 2001. “Where Will They Go If We Go Territorial? Dividend Exemption and the Location Decisions of U.S. Multinational Corporations.” National Tax Journal 54 (December): 787–809. Altshuler, Rosanne, Harry Grubert, and T. Scott Newton. 2001. “Has U.S. Investment Abroad Become More Sensitive to Tax Rates?” In International Taxation and Multinational Activity, edited by James R. Hines Jr., pp. 9–32. University of Chicago Press. Auerbach, Alan, and Kevin Hassett, eds. 2005. Toward Fundamental Tax Reform. Washington: AEI Press. Auerbach, Alan J., and Joel Slemrod. 1997. “The Economic Effects of the Tax Reform Act of 1986.” Journal of Economic Literature 35 (June): 589–632. Bhagwati, Jagdish, Arvind Panagariya, and T. N. Srinivasan. 2004. “The Muddles over Outsourcing.” Journal of Economic Perspectives 18 (Fall): 93–114. Blackorby, Charles, and R. Robert Russell. 1989. “Will the Real Elasticity of Substitution Please Stand Up? (A Comparison of the Allen/Uzawa and Morishima Elasticities).” American Economic Review 79 (September): 882–88. Chamley, Christophe. 1986. “Optimal Taxation of Capital Income in General Equilibrium with Infinite Lives.” Econometrica 54 (May): 607–22. Clausing, Kimberly. 2003. “Tax-Motivated Transfer Pricing and U.S. Intrafirm Trade Prices.” Journal of Public Economics 87 (September): 2207–23. ———. 2004. “Taxing Corporations in OECD Countries.” Reed College. ———. 2005a. “Tax Holidays (and Other Escapes) in the American Jobs Creation Act.” National Tax Journal 58 (September): 331–46. ———. 2005b. “U.S. Tax Legislation and the Activities of Multinational Firms in Europe.” Working Paper. Dublin: Institute of European Affairs. Cronin, Bruce, Leslie Catchpowle, and David Hall. 2004. “Outsourcing and Offshoring.” CESifo Forum 5 (Summer): 17–21. De Mooij, Ruud A., and Sjef Ederveen. 2003. “Taxation and Foreign Direct Investment: A Synthesis of Empirical Research.” International Tax and Public Finance 10 (November): 673–93. Desai, Mihir A., and James R. Hines Jr. 2003. “Value Added Taxes and International Trade: The Evidence.” Harvard Business School/University of Michigan. ———. 2004. “Old Rules and New Realities: Corporate Tax Policy in a Global Setting.” National Tax Journal 57 (December): 937–60. Desai, Mihir A., C. Fritz Foley, and James R. Hines Jr. 2002. “Chains of Ownership, Regional Tax Competition, and Foreign Direct Investment.” Working Paper 9224. Cambridge, Mass.: National Bureau of Economic Research (September). Dunning, J. H. 1974. Economic Analysis and Multinational Enterprise. New York: Praeger.
Kimberly A. Clausing
489
Glenn, Heidi. 2004a. “House-Passed ETI Bill Sets Stage for Difficult Conference.” Tax Notes (June 21): 1439. ———. 2004b. “Yin Wonders Whether Territorial System Shift Is Next.” Tax Notes (December 13): 1487–88. Gordon, Roger H. and James R. Hines Jr. 2002. “International Taxation.” In Handbook of Public Economics. Vol. 4, edited by Alan J. Auerbach and Martin Feldstein, pp. 1935–95. London: North Holland. Gordon, Roger H., and Jeffrey K. Makie-Mason. 1995. “Why Is There Corporate Taxation in a Small Open Economy? The Role of Transfer Pricing and Income Shifting.” In The Effects of Taxation on Multinational Corporations, edited by Martin Feldstein, James R. Hines, and R. Glenn Hubbard, pp. 67–91. University of Chicago Press. Gordon, Roger H., and Joel Slemrod. 2000. “Are “Real” Responses to Taxes Simply Income Shifting Between Corporate and Personal Tax Bases?” In Does Atlas Shrug? edited by Joel Slemrod. Harvard University Press. Gropp, Reint, and Kristina Kostial. 2000. The Disappearing Tax Base: Is Foreign Direct Investment Eroding Corporate Income Taxes? Working Paper 31. Frankfurt: European Central Bank (September). Grubert, Harry. 2001. “Enacting Dividend Exemption and Tax Revenue.” National Tax Journal 54 (December): 811–27. Grubert, Harry, and John Mutti. 1995. “Taxing Multinationals in a World with Portfolio Flows and R&D: Is Capital Export Neutrality Obsolete?” International Tax and Public Finance 2 (November): 439–57. ———. 2001. Taxing International Business Income: Dividend Exemption versus the Current System. Washington: American Enterprise Institute Press. Hall, Robert, and Alvin Rabushka. 1985. The Flat Tax. Stanford, Calif.: Hoover Institution Press. Hines, James R. Jr. 1994. “Credit and Deferral as International Investment Incentives.” Journal of Public Economics 55 (October): 323–47. ———. 1996. “Altered States: Taxes and the Location of Foreign Direct Investment in America.” American Economic Review 86 (December): 1076–94. ———. 1997. “Tax Policy and the Activities of Multinational Corporations.” In Fiscal Policy: Lessons from Economic Research, edited by Alan J. Auerbach, pp. 401–45. MIT Press. ———. 1999a. “Lessons from Behavioral Responses to International Taxation.” National Tax Journal 52 (June): 305–22. ———. 1999b. “The Case against Deferral: A Deferential Reconsideration.” National Tax Journal 52 (September): 385–404. ———. 2005. “Corporate Taxation and International Competition.” University of Michigan. Hines, James R., and Eric M. Rice. 1994. “Fiscal Paradise: Foreign Tax Havens and American Business.” Quarterly Journal of Economics 109 (1): 149–82.
490
Brookings Trade Forum: 2005
Hufbauer, Gary C. 2002. “The Foreign Sales Corporation Drama: Reaching the Last Act?” International Economic Policy Briefs PB02-10 (November). Washington: Institute for International Economics. Joint Committee on Taxation. 2004. Estimated Budget Effects of the Conference Agreement for H.R. 4520, The ‘American Jobs Creation Act of 2004.’” JCX-69-04. 2004 (Government Printing Office). Judd, Kenneth L. 1985. “Redistributive Taxation in a Simple Perfect Foresight Model.” Journal of Public Economics 28 (October): 59–83. Morishima, Michio. 1967. “A Few Suggestions on the Theory of Elasticity.” Keizai Hyoron 16: 144–50. Mutti, John. 2003. Foreign Direct Investment and Tax Competition. Washington: Institute of International Economics Press. Richman, Peggy B. 1963. Taxation of Foreign Investment Income: An Economic Analysis. Johns Hopkins University Press. Sheppard, Lee A. 2004. “Do-It-Yourself Repatriation.” Tax Notes 18 (October): 282–87. Sinn, Hans W. 1993. “Taxation and the Birth of Foreign Subsidiaries.” In Trade, Welfare, and Economic Policies, edited by Horst Herberg and Ngo Van Long, pp. 325–52. University of Michigan Press. Swenson, Deborah L. 2001. “Tax Reforms and Evidence of Transfer Pricing.” National Tax Journal 54 (March): 7–25. Taylor, Willard. 2005. “Presentation before the President’s Advisory Panel on Federal Tax Reform,” March 31. Tax Notes 4 (April). Viard, Alan D. 2004. “Border Adjustments Won’t Promote Competitiveness.” Tax Notes 4 (December): 122–24. Weichenrieder, Alphons J. 1996. “Anti-tax-voidance Provisions and the Size of Foreign Direct Investment.” International Tax and Public Finance 3 (January): 67–81. Wilson, John D. 1999. “Theories of Tax Competition.” National Tax Journal 52 (June): 269–304.
