TAX POLICY
AND THE ECONOMY 16 edited by James M. Poterba
National Bureau of Economic Research The MIT Press, Cambridge, Massachusetts
Tax Policy and the Economy, Volume 16, 2002 ISSN: 9892-8649 E-ISSN: 1537-2650 ISBN: Hardcover 0-262-16210-5 Paperback 0-262-66129-2
Published annually by The MIT Press, Cambridge, Massachusetts 02142 An electronic, full-text version of Tax Policy and the Economy is available from MIT Press
Journals when purchasing a subscription. Subscription Rates
Hardcover/Print and Electronic: $58.00 Paperback/Print and Electronic: $25.00 Outside the U.S. and Canada add $10.00 for postage and handling. Canadians add 7% GST.
Subscription and address changes should be addressed to: MiT Press Journals, Five Cambridge Center, Cambridge, MA 02142-1407, phone 617253-2889; fax 617-577-1545; email
[email protected]. Claims wifi be honored free of charge if made within three months of the publication date of the issue. Claims may be submitted to
[email protected]. Prices are subject to change without notice. In the United Kingdom, continental Europe, and the Middle East and Africa, send back volume orders and business correspondence to:
The MIT Press, Ltd., Fitzroy House, 11 Chenies Street, London WC1E 7ET England, phone 44-020-7306-0603, fax 44-020-7306-0604, email
[email protected] In the United States and for all other countries, send single copy and back volume orders to:
The MIT Press, Five Cambridge Center, Cambridge, MA 02142, toll-free book orders 800-356-0343, fax 617-625-6660, email
[email protected] Copyright Information
Permission to photocopy articles for internal or personal use, or the internal or personal use of specific clients, is granted by the copyright owner for users registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided that the fee of $10.00 per copy is paid directly to CCC, 222 Rosewood Drive, Danvers, MA 01923. The fee code for users of the Transactional Reporting Service is: 9892-8649/02 $10.00. For those organizations that have been granted a photocopy license with CCC, a separate system of payment has been arranged. © 2002 by the National Bureau of Economic Research and the Massachusetts Institute of Technology.
NBER BOARD OF DIRECTORS BY AFFILIATION Officers: Carl F. Christ, Chairman Michael H. Moskow, Vice Chairman Martin Feldstein, President and Chief
Robert Mednick, Treasurer Kelly Horak, Controller and Assistant
Executive Officer
Gerardine Johnson, Assistant
Corporate Secretary
Susan Colligan, Vice President for Administration and Budget and
Corporate Secretary
Corporate Secretary
Directors at Large: Peter C. Aldrich Elizabeth E. Bailey John H. Biggs Andrew Brimmer Carl F. Christ John S. Clarkeson Don R. Conlan George C. Eads
Martin Feldstein Stephen Friedman Judith M. Gueron George Hatsopoulos Karen N. Horn Judy C. Lewent John Lipsky Michael H. Moskow
Alicia H. Munnell Rudolph A. Oswald Robert T. Parry Peter G. Peterson Richard N. Rosett Kathleen P. Utgoff Marina v. N. Whitman Martin B. Zimmerman
Directors by University Appointment: George Akerlof, Cahfornia, Berkeley Jagdish Bhagwati, Columbia William C. Brainard, Yale Michael J. Brennan, Cahfornia, Los Angeles
Glen G. Cain, Wisconsin Franklin Fisher, Massachusetts Institute of Technology
Saul H. Hymans, Michigan
Marjorie B. McElroy, Duke Joel Mokyr, Northwestern Andrew Postlewaite, Pennsylvania Nathan Rosenberg, Stanford Michael Rothschild, Princeton Craig Swan, Minnesota David B. Yoffie, Harvard Arnold Zellner, Chicago
Directors by Appointment of Other Organizations: Mark Drabenstott, American
Richard D. Rippe, National
Agricultural Economics Association Gail D. Fosler, The Conference Board
John J. Siegfried, American Economic
A. Ronald Gallant, American Statistical Association
Robert S. Hamada, American Finance Association
Robert Mednick, American Institute of Certfied Public Accountants Angelo Melino, Canadian Economics Association
Association for Business Economics Association
David A. Smith, American Federation of Labor and Congress of Industrial Organizations Josh S. Weston, Committee for Economic Development
Gavin Wright, Economic History Association
Directors Emeriti: Thomas D. Flynn Lawrence R. Klein
Franklin A. Lindsay Paul W. McCracken
Bert Seidman Eli Shapiro
Since this volume is a record of conference proceedings, it has been exempted from the rules governing critical review of manuscripts by the Board of Directors of the National Bureau (resolution adopted 8 June 1948, as revised 21 November 1949 and 20 April 1968).
CONTENTS Introduction: James M. Poterba vii
Acknowledgments xi ECONOMIC EFFECTS OF MEANS-TESTED TRANSFERS IN THE U.S. 1 Robert Moffitt
TAXES AND HEALTH INSURANCE 37 Jonathan Gruber
DEFINED CONTRIBUTION PENSIONS: PLAN RULES, PARTICIPANT CHOICES, AND THE PATH OF LEAST
RESISTANCE 67 James J. Choi, David Laibson, Brigitte C. Madrian, and Andrew Metrick
THINKING STRAIGHT ABOUT THE TAXATION OF ELECTRONIC COMMERCE: TAX PRINCIPLES, COMPLIANCE PROBLEMS, AND NEXUS 115 Charles E. McLure, Jr.
THE FISCAL EFFECTS OF POPULATION AGING IN THE U.S.: ASSESSING THE UNCERTAINTIES 141 Ronald Lee and Ryan Edwards
POTENTIAL PATHS OF SOCIAL SECURITY REFORM 181 Martin Feldstein and Andrew Samwick
INTRODUCTION James M. Poterba MIT and NBER
The annual NBER Tax Policy and the Economy conference began as a vehicle for communicating current findings from NBER research in the areas of taxation and government spending to policy analysts in both the government and the private sector. It has now been emulated in other
fields, with new research meetings on health policy and innovation policy, designed to achieve similar objectives. Tax Policy and the Economy
papers have addressed a wide range of issues. The six papers in this year's volume are no exception. They touch on both taxation and govern-
ment expenditure programs, and they include discussion of both very current policy topics and issues of long-standing and ongoing interest. The first paper, Robert Moffitt's "Economic Effects of Means-Tested Transfers in the U.S.," provides a wide-ranging analysis of government programs that provide income or in-kind assistance to individuals and families who satisfy particular income constraints. The paper summarizes the results of a recent NBER study on how these programs affect incentives for labor supply, for changes in family status, and for a range of other behaviors. This recent research concludes that these programs have important effects on their potential and actual recipients. In "Taxes and Health Insurance," Jonathan Gruber explores the many channels through which the income and payroll tax system affects the demand for health insurance. The paper begins by describing the key behavioral elasticities that are needed to evaluate potential tax reforms. These are the price sensitivity of insurance demand by individuals, the price sensitivity of employers' decisions about whether or not to offer health insurance, and the price sensitivity of worker's discrete decision whether or not to purchase health insurance conditional on it being
viü Introduction
offered by the employer. Gruber then surveys the existing empirical literature that bears on each of these issues, and he uses mid-range estimates from these studies to describe how potential changes in the
federal income tax might affect health insurance coverage. He concludes
that removing the current income tax subsidy to employer-provided health insurance, while leaving other tax subsidies, such as those through the payroll tax, in place, could reduce the number of persons with health insurance by nearly fourteen million. In "Defined Contribution Pensions: Plan Rules, Participant Choices, and the Path of Least Resistance," James Choi, David Laibson, Brigitte Madrian, and Andrew Metrick analyze the heterogeneity in the retire-
ment saving plans that firms offer their workers. They not only describe the plan-to-plan variation, but they also use this variation to investigate how various plan features affect voluntary enrollment and contribution decisions. They identify a number of plan features that are correlated
with high levels of participation in, and contributions to, retirement
saving plans. These include automatic enrollment programs and allowing a relatively large fraction of salary to be subject to an employer matching contribution. The paper concludes that a significant population of employees simply adopt the default option that their employer offers, with respect to both the fraction of their salary that they contribute to the plan and the asset allocation they choose within the plan. This suggests that it is important for firms to carefully evaluate the default options that they offer in their retirement plans. In the fourth paper, Charles McLure tackles the very current issue of taxing e-commerce. In his paper on "Thinking Straight about the Taxation of Electronic Commerce: Tax Principles, Compliance Problems, and Nexus," McLure places the Internet Tax Freedom Act in context by describing the general problem of taxing sales in an economy with multiple jurisdictions. The need to determine whether a seller has nexus in a given jurisdiction, which is central to the taxation of electronic commerce, arises in many other settings as well. The paper outlines general rules for designing a sales tax that minimizes economic distortions, and then explains the set of tax rules on Internet commerce that would be most consistent with these principles. The last two papers address issues related to social security policy. Ronald Lee and Ryan Edwards, in their paper on "The Fiscal Effects of Population Aging in the U.S.: Assessing the Uncertainties," present important evidence on the uncertainties associated with demographic projections. They note that most current projections of how aging will affect federal spending rely on scenario analyses that make a fixed set of assumptions about key parameter values, such as birth rates and migration
Introduction
ix
rates, and then evaluate the long-run fiscal consequences of various government policies. Lee and Edwards present an alternative methodology for addressing these issues, using stochastic simulation. Their results highlight the potential uncertainties associated with any single-scenario description of the long-run prospects for a given fiscal policy. They also suggest that it is risky to advertise policy reforms, such as an increase of a
given amount in the payroll tax, as "fixes" for problems such as the
potential insolvency of the Social Security Trust Fund. The uncertainty of future birth rates and death rates means that there is still a chance that such policies will result in an excess, or a shortfall, of resources for public programs. Finally, in the last paper, "Potential Paths of Social Security Reform," Martin Feldstein and Andrew Samwick consider a number of potential reforms of the current social security system. Each reform combines a
system of personal retirement accounts with a variant of the current defined benefit system. The paper evaluates how each of the various reforms would affect the benefits available to future retirees. It also considers how the reforms would affect national saving and the solvency of the social security trust fund. The paper addresses the risk of individual account investments and presents evidence on the range of possible benefit outcomes under different assumptions about the long-run average return on corporate equities. The papers in this volume illustrate the type of policy-relevant research that is carried out by the affiliates of the NBER Public Economics Program. Each of the papers provides important background informa-
tion for policy analysis, without making recommendations about the merits or demerits of particular policy options. I hope that each of these
papers wifi provide useful input to various participants in the policy process who are concerned with tax and expenditure program design.
ACKNOWLEDGMENTS In planning and organizing this year's Tax Policy and the Economy conference and the associated volume, I have incurred debts to many individuals. Martin Feldstein, President of the NBER, has been an active sup-
porter of this conference throughout its sixteen-year history. Jennifer Dayton did an outstanding job in updating the invitation list and in disseminating information about the conference to potential participants. The members of the NBER Conference Department, notably Lita Kimble, Rob Shannon, and Conference Department Director Carl Beck, handled conference logistics with efficiency and good cheer. Helena Fitz-Patrick oversaw the publication process with outstanding attention to detail and with exceptional speed and efficiency. Jam also grateful to R. Glenn Hubbard, the Chairman of the President's Council of Economic Advisers, for delivering a fascinating set of luncheon remarks at the conference at which these papers were presented. Glenn explained how the tragic terrorist attacks of September 11 had raised new questions about the appropriate role of government in the economy, and
he sketched many individual contexts in which there is ongoing debate about the government's role. Finally, I wish to thank each of the authors whose papers are included in this volume for striving to communicate their important research findings in a readable and clear fashion. I appreciate their efforts and willingness to participate in this very important opportunity for interchange between academics and policymakers.
ECONOMIC EFFECTS OF MEANS-TESTED TRANSFERS IN THE U.S. Robert Moffitt Johns Hopkins University and National Bureau of Economic Research
EXECUTIVE SUMMARY The system of means-tested transfers in the U.S. has evolved in important ways over the last decade, with significant expansions of Medic-
aid, the Earned Income Tax Credit, and the Supplemental Security Income program, and with significant contraction in Aid to Families with Dependent Children, now titled the Temporary Assistance for Needy Families program. To determine where we are in our understanding of each of these programs, as well as the other major programs in the system of means-tested transfers, a volume is under preparation by the National Bureau of Economic Research that surveys the current structure and historical evolution of each of these programs and
that synthesizes the results of the research that has been conducted on their economic effects. In addition to the AFDCTANF, Medicaid, EITC, and SSI programs, reviews have been conducted for the Food Stamp program and for housing, child care, job training, and child support programs. This paper summarizes the results of those reviews and highlights the large number of important findings from existing research. The author would like to thank the Smith-Richardson Foundation for support for the conference and NBER volume upon which this paper is based. Comments from Robert LaLonde, Edgar Olson, and Karl Scholz are appreciated.
2
Moffitt
Reform of the system of means-tested transfers in the U.S. continues to be an important topic for public policy as well as an area of continued
research by economists. Policy and research interest have been kept particularly high by significant transformations in the means-tested transfer system over the last decade. The most important structural changes have taken place in three programs. One is the Aid to Families with Dependent Children (AFDC) programnow named the Temporary Assistance for Needy Families (TANF) programwhose generosity has been significantly reduced and whose eligibility conditions have been restricted to those who can and are willing to comply with work requirements and other new rules. A second is the Medicaid program, which has been significantly expanded to cover more families and children off the AFDC-TANF program. The third is the Earned Income Tax Credit (EITC), whose benefits have greatly expanded and whose expenditures now exceed those in the traditional AFDC-TANF program. A
fourth program which has undergone significant expenditure and caseload expansion, although without major structural change, is the Supplemental Security Income (SSI) program.
To determine where we stand in our understanding of each of these programs, as well as the other major programs in the system of meanstested transfers, a volume is under preparation by the National Bureau of Economic Research that surveys the current structure and historical evolution of each of these programs and that synthesizes the results of the research that has been conducted on their economic effects (Moffitt, forthcoming). In addition to the AFDC-TANF, Medicaid, EITC, and SSI programs, reviews have been conducted for the Food Stamp program and for housing, child care, job training, and child support programs. This paper summarizes the results of those reviews. The paper first provides a brief background discussion of trends in expenditures on means-tested transfers as a whole. It then goes on to discuss each of the major programs individually.
1. OVERALL TRENDS IN EXPENDITURES IN MEANS-TESTED TRANSFER PROGRAMS Figure 1 shows trends since 1968 in per capita expenditures in the eighty
largest means-tested transfer programs in the country. The figure reveals that there have been four phases of spending growth: an expansionary phase beginning in the 1960s and running through the early or mid-1970s, a contractionary (or stationary) phase beginning in the mid1970s and running until the mid-1980s, another expansionary phase
Economic Effects of Means-Tested Transfers in the U.s.
3
FIGURE 1. Real Per Capita Expenditures on Means-Tested Transfers, 1968-1998
1965
1970
1975
1960
1955
1990
1995
2000
Year
Sources: Burke (1999, Tables 4 and 5), U.S. Dept. of Commerce (2000, Table 2, Population).
running from the late 1980s to the mid-1990s, and another contractionary (or stationary) phase beginning in the mid-1990s. The first phase saw an increase in AFDC benefits; enactment of a
major piece of welfare legislationthe 1967 Social Security Amendmentswhich raised earnings disregards in the program (i.e., lowered the tax rate on earnings); and the creation of the food stamp and Medicaid programs and, later in the period, the Supplemental Security Income
program. Caseloads grew rapidly in all four of these programs. This period was later termed the era of the "welfare explosion" and set the modern framework of means-tested transfers. The second phase saw a steady decline in real AFDC benefits; enactment of a major piece of AFDC legislationthe 1981 Omnibus Budget Reconciliation Actwhich effectively eliminated the earnings disregards enacted in 1967 and consequently cut thousands of families with earnings from the rolls; and witnessed an increasing interest in work requirements and mandatory training programs for welfare recipients among federal policymakers. Declining real AFDC benefits were accompanied by slow but steady growth in the number of single-mother families, and the offsetting effects of these two forces left AFDC expenditures more or less unchanged in real terms.
4 Moffitt
The third phasewhich is not always recognized, for it is often presumed that the system has been in steady contraction since the 1970s saw a dramatic expansion of the Earned Income Tax Credit (EITC); major
expansions of eligibility in the Medicaid program, primarily to nonAFDC families; and sizable expansions of the caseload in the SSI program, arising mostly from increased numbers of disabled adults and children. The Family Support Act of 1988, although occurring in the third phase and seemingly contractionaryit mandated work and training for AFDC recipients more heavily than in the pastis best viewed as neutral, for not only was it never effectively implemented, but it also could be interpreted as expansionary inasmuch as it required new expenditures on work programs for AFDC recipients. The runup of expendi-
tures in this period, although not quite as large in magnitude as that resulting from the welfare explosion of the late 1960s and early 1970s, occurred much more quicklyessentially occurring in a five-year period between 1990 and 1995.
The fourth phase, which is continuing at this writing, is a combined result of 1996 welfare legislation, which contracted the AFDC-TANF program, and a robust economy which has led to declining caseloads in many programs, thereby slowing expenditure growth. The Food Stamp and Medicaid programs, as well as AFDC-TANF, have seen declining caseloads.1
Table 1 shows, in more detail, the sources of expenditure growth in the third, expansionary period. AFDC expenditures actually declined, presaging the further decline which has occurred subsequent to the 1996 legislation. The Food Stamp program expanded by 42 percent, however, indicating robust growth. A very large percentage expansion occurred in the Medicaid program, which grew by 88 percent. As wifi be discussed further below, the Medicaid program covers different types of recipients, and the growth over this period came not only from expansions of expenditures for single mothers and their children, but also from increased expenditures on the disabled. While single mothers and their children represent the largest fraction of the Medicaid caseload, expenditures are greater for the disabled because of their greater medical needs. The largest percentage expansion in Table 1, however, occurred in the EITC program, whose expenditures almost tripled over the period. As wifi be discussed below, major expansions of the size of the credit re-
sulted in this growth. Housing programs grew modestly during the The unemployment rate appears to have started to increase in late 2000 or early 2001, possibly indicating the beginning of a recession. Whether this will signal the beginning of a fifth phase or a modification of the fourth remains to be seen, and wifi depend on legislative developments and on the course of expenditure growth over the next few years. 1
Economic Effects of Means-Tested Transfers in the U.S.
5
TABLE 1
Change in Real Expenditures' in Six Major Programs, FY1990 to FY1996 Expenditure (million 1996 dollars)
Program AFDC
Food Stamps Medicaid EITCb
Housingc SSI
1990
1996
24,758 20,654 84,658 8,092 16,922 20,125
23,677 27,344 159,357 24,088 19,877 32,065
Share of
Change
growth
(%)
(%)
4
1
42 88 198 17 59
7 60 13
4 10
Sources: Burke (1993, Table 15; 1999, Table 3, 12).
'Federal and state combined totals. blncludes reduction in tax liability, not just refundable portion. cSum of expenditures on public and Section 8 housing.
period, but the SSI program grew by a large amount, 59 percent, reflecting, as in Medicaid, increases in expenditures on the disabled.2 The last row of Table 1 shows the shares of total expenditure growth in the largest eighty means-tested transfers from 1990 to 1996 accounted for by each of these six programs. Medicaid expenditure growth, while not the largest in percentage terms, is the largest in dollar terms and accounts for the largest fraction, 60 percent. The EITC and SSI together account for another 23 percent. Altogether, these six programs accounted for 93 percent of the overall increase in means-tested expenditures in the 1990-1996 expansionary phase. Finally, Table 2 shows the expenditures and caseloads in the ten largest means-tested transfer programs in FY 1997. The largest is Medicaid,
as expected, and the next fiveSSI, EITC, Food Stamps, TANF, and subsidized housingare of the same general magnitude but at a large distance from Medicaid. The TANF program, which in the 1960s was the largest of the six, is now a distant fifth in rank. The evolution of means-tested transfers which has led to the ranking in Table 2 reflects several trends. One is the gradual decline of unrestricted cash transfers like AFDC relative to in-kind transfers like Medicaid, Food If medical care prices are used to deflate Medicaid expenditures instead of a general price index, Medicaid expenditure growth amounted to only 34 percent. Which index should be used depends on whether the goal is to value expenditures from the point of 2
view of the taxpayer or the recipient.
Moffitt
6
TABLE 2
Annual Expenditures and Caseloads in Ten Largest Transfer Programs, FY1997 Expenditures Program
($ million)
Medicaid
167,359 32,395 28,800 24,772 23,179 19,336 9,220
SSI
EITC FS
TANF
Subsidized housing Medical care for veterans without service disability Foster care Social service Federal Pell grants
6,794 6,400 5,660
Caseload (thousands)
Expenditures per recipient ($)
153
4,138 4,638 495 1,024 2,120 4,481 60,261
289
23,509
40,446 6,984 58,143 24,200 10,936 4,315
NA
NA
3,665
1,544
Source: Burke (1999, Table 12).
Stamps, and housing. Voters and legislators appear to prefer to make transfers tied to specific consumption items rather than open-ended cash
transfers. A second is the increasing narrowness of the targeting of transfers: the programs which have seen the largest growth in the last decade are tied to specific eligibffity groups. The EITC is specifically targeted to families with earnings, the SSI program is targeted to the disabled and elderly, and Medicaid is targeted to the disabled andin the expansions
that have occurredmainly to single mothers and their children off TANF. This development represents a continued, if not increased, cate-
gorization of the nation's welfare population into a system in which different demographic groups are judged to be needy not just on the
basis of income but on the basis of some other specific characteristic that leads them to be deserving in the eyes of the public. This also explains why the EITC and SSI programs, which provide tied cash transfers, have expanded while the AFDCTANF program has not. As a consequence of these developments, the great expenditure expansion of the late 1980s and early 1990s increased total transfers to the low-income population but also changed the distribution of those transfers. Families off welfare
with earnings and the disabled gained, for example, relative to lowincome single-mother families as a whole, particularly those on welfare or not working.
The transfer programs reviewed in the forthcoming NBER volume include the six largest programs shown in Table 2. In addition, several
Economic Effects of Means-Tested Transfers in the U.s.
7
smaller but important programs are covered. These include child care programs (approximate FY 1999 expenditures of $17 billion across all programs), programs for child-support enforcement ($3 billion in 1996), and job training programs for the disadvantaged (expenditures of $1 billion in 1998). These nine programs wifi be discussed in this chapter roughly in order
of their total expenditures. For each program, the discussion first covers the structure and rules of the program and its historical evolution, followed by a discussion of trends in expenditures and caseloads and recipient characteristics, then followed by a review of research findings.
2. MEDICAID The Medicaid program, as noted by Gruber (forthcoming), is really four separate programs rolled into one. One supports the medical expenses of low-income single mothers and their children. The other three provide public insurance for portions of medical expenditures not covered by Medicare for the low-income elderly, support medical expenses for the low-income disabled, and provide coverage of nursing-home expenditures of the institutionalized elderly. The first program has a majority of the recipients, but the other three programs are responsible for a majority of the expenditures. Medicaid was created in 1965 by the same legislation that created the Medicare program. It is administered by the states, which must operate under federal regulation, and the federal government pays a fixed share of state expenditures (the state share, determined by a formula involving various state characteristics including median income, is approximately 43 percent). The program was initially aimed at providing medical benefits to traditional welfare populationslow-income single mothers and children, and the aged, blind, and disabled. However, over time eligibility has been expanded to other groups. Early in the program some coverage was extended to low-income children in two-parent families, and a Medically Needy program was instituted which provided care for low-income families (usually single-mother families) with income too high for welfare eligibility, albeit with numerous restrictions on eligibility. Beginning in 1984, and accelerating after 1987, more significant expansions were first allowed, and then mandated, requiring states to cover children in families with incomes below 133 percent of the poverty line, or higher at state option. Pregnant women were also covered, but otherwise there was no expanded coverage for adults. These expansions are part of the reason for the expenditure increase discussed in the last section. A further major expansion took place in 1997 with the creation
8
Moffitt
of the Children's Health Insurance Program, which provides a capped federal match for state creation of programs to cover groups outside of existing Medicaid eligibility or with higher incomes. Some states have chosen to expand their Medicaid programs, while others have created wholly new programs to cover these additional groups. Medicaid mandates a specific list of services that states must provide to all "categorically needy" recipients. States may go beyond this at their option, but few do, and when they do, most states cover the same types of extra services; as a result, there is substantial uniformity in the service package across states. Reimbursement rates, on the other hand, are given much more leeway, and there is major cross-state variation. Reimbursement rates are generally quite low and below those of Medicare and private payers. States are allowed waivers to experiment with different options for care provision, and the major direction states have pursued is the use of managed care for their Medicaid caseloads. By 1998,54 percent of Medicaid recipients were in managed care plans. Gruber shows that both expenditures and enrollment in the Medicaid program have increased enormously over the last decade, as indicated in the background discussion above. The major enrollment growth has been among the disabled and among children under 21. Enrollment growth has slowed in recent years, possibly because of 1996 federal welfare reform legislation which contracted the AFDCTANF program. Calculations of participation rates in the Medicaid program have been computed only for children and pregnant women, because they are the only groups for which eligibility has been calculated; no estimates are
available for the elderly and disabled. Eligibility has expanded greatly, as already noted, but takeup has slipped behind significantly, resulting in declines in participation rates. Whereas participation rates among eligible children were close to 100 percent prior to 1989, by 1996 they had fallen to 73 percent. There has been a great deal of research by economists on the Medicaid
program. One issue concerns reasons for the declining participation rates just noted. Research has shown that much of the explanation is that the increased eligibility arising from coverage expansions was to groups with higher than usual incomes, groups with less need for insurance; to groups outside the AFDC program and who therefore do not have the relatively easy institutional access to the program that welfare recipients do; and to groups who already are covered by other forms of
insurance. This last finding is related to a significant area of research on Medicaid crowdout, which occurs when Medicaid expansions result in substitutions of Medicaid coverage for private insurance coverage. There are a variety of empirical estimates of the extent of crowdout, some
Economic Effects of Means-Tested Transfers in the U.s.
9
indicating relatively small and some quite large effects. For example, in the latter category estimates have indicated that approximately 50 percent of those who have taken up Medicaid would have been privately covered otherwise. Research continues in this area in an attempt to resolve the differences in the magnitude of the effect. Another area of research concerns the effect of the Medicaid expansions on health and health outcomes. The studies which use nationwide data rather than data from individual states typically show significant positive effects of the Medicaid expansions on infant mortality, prenatal care utilization, and child preventative care, and that they led to more hospitalizations (but fewer "avoidable" ones). Research indicates that the positive effects are larger for those in demographic groups with typically worse health, such as black families, immigrants, and those with low educational levels. Effects are also larger for targeted expansions which are aimed at low-income mothers and children than for broad expansions which reach further up the income distribution. The effect of the Medicaid program on the labor supply of recipients and on their AFDCTANF participation decisions has been another focus on research, concentrating on single mothers and their children rather than the elderly and disabled. There is a range of research using different methodologies, and virtually all of it shows that the close historical tie of Medicaid eligibility to AFDC receipt tended to increase AFDC participation rates, and that the Medicaid expansions which loosened that tie also tended to reduce AFDC participation rates. Also, because AFDC has some work disincentives, the historical link has tended to decrease labor supply, while the Medicaid expansions have tended to increase it. Although the magnitudes of these effects are not precisely estimated in the literature, their direction is supported by most studies. Related work on the effect of Medicaid expansions demonstrates that they lowered savings and increased consumption, consistent with the notion that welfare recipients engage in less precautionary savings when they know that the program wifi support them should their income decline. Finally, there has been research on the effects of reimbursement policy and long-term care provision in Medicaid. The literature on physician reimbursement rates mostly shows that higher reimbursement rates lead to somewhat increased participation by physicians in the program, increased access to care, and occasionally better health outcomes, although the linkage between reimbursement policy and utilization and health is far from simple. With regard to long-term care, research indicates that Medicaid recipients are often on long waiting lists and have less access to care than private pay patients, but also that increases in subsidies to nursing home care raise overall nursing home utilization.
10
Moffitt
Some other research raises the issue of whether nursing home quality might be reduced as nursing homes increase the percentage of their patients who are minimum-pay Medicaid recipients.
3. THE SUPPLEMENTAL SECURITY INCOME PROGRAM As described by Daly and Burkhauser (forthcoming), the SSI program is a federal program which pays cash benefits to low-income individuals who are 65 or older, or who are blind or disabled. It was enacted in 1972 and was to a large extent a product of the proposals by the Nixon administration for a negative income tax. Eligibility requires not only low income and assets but also, for the blind and disabled, a medical test. The test is most complex for disabled adults and involves a multistep process meant to ensure that the individual is incapable of working, for the goal of the program is to serve only those who are totally disabled. Assessments by medical examiners as well as more general determinations of the nature and severity of the disability and capacity for employment are conducted. An earnings test is also used, which requires that applicants earn less than a fixed dollar amount. All in all, about 63 percent of the applicants are denied by this process. Eligibffity determination for children is different because the employment test is inappropriate; it is instead based on the presence of a severe functional limitation. In 1990, a court decision (the Zebley decision) required that children also be given a particular additional functional assessment test which, when later implemented, effectively lowered eligibility standards by allowing children onto the rolls who did not pass the more formal medical tests. In the same 1996 legislation that restructured the AFDC program, Congress narrowed the basis for SSI child eligibility and moved it back towards the pre-199O standard in breadth. The legislation also denied SSI eligibility for noncitizens. The SSI program has work incentive features that reflect its origins in discussions of a negative income tax. After eligibility has been established and individuals begin receiving benefits, earnings (after disregards) are reduced by only 50 cents for every extra dollar of income, thus providing some incentives to work. However, despite these incentives, the percentage of SSI recipients with earnings has always been very low. Only 4.4 percent of the caseload had earnings in 1996, and the proportion had never exceeded 4.7 percent in the history of the program. In
addition, special incentives allowing working beneficiaries to retain Medicaid coverage after their incomes exceed normal eligibility levels
Economic Effects of Means-Tested Transfers in the U.s.
11
have been taken up by only 1.3 percent of the caseload. Consequently, work incentives are still a major issue in the program, and mechanisms for increasing work are still under active discussion. Caseload and expenditure growth in the program has been positive since its inception in 1974 but was exceptionally high in the early 1990s, as noted previously. This growth was disproportionately concentrated among the blind and disabled, children, and non-citizens (rather than the elderly). Growth in the number of recipients who qualified on the basis of mental impairments was particularly strong. Since the 1996 welfare legislation, the child and non-citizen caseloads have declined. Nevertheless, despite the strong caseload growth in the 1990s, there has been continued concern that many eligibles do not participate in the program. Estimated participation rates among the eligible elderly population, for example, range from 45 to 60 percent. Research on the SSI program has focused on a number of issues. One of the most heavily studied focuses on the reasons for the caseload growth. Much of this research has examined historical fluctuations in applications and awards for the nonelderly. These fluctuations, shown in Figure 2, have been very large. Increases in applications in the late 1970s, subsequent declines in the early 1980s, and revived growth in the mid-1980s can be explained largely by administrative changes in screening stringency over the period, as perceived by the eligible population. The rise in applications in the late 1980s and early 1990s and a portion of the decline after 1994 have been shown to be heavily affected by the business cycle. This is an important finding, because it establishes that labor-market participation is a realistic alternative for many disabled persons, contrary to the
notion that only those incapable of working are in the program. The business cycle is also responsible for much of the very large increase in applications in the early 1990s, but only for some eligibility groups. Caseload growth among the disabled with mental health and musculoskeletal (e.g., back pain) conditions, and among children, were equally the result of relaxation of screening and eligibility rules. Declines in applications subsequent to 1996 can also be partly attributed to the 1996 federal legislation as well as the business cycle.
Another factor in the growth of the child caseload identified in the research is the relationship between SSI and AFDC benefit levels. For AFDC families with children who can qualify for SSI, the greater benefit levels in SSI than in AFDC provide an incentive to move children from the latter program to the former. The gap between benefits has also been growing over time. Research has shown that this has made some contribution to the growth in the child SSI caseload. In addition, related research has shown that work disincentives for single mothers accompany
Moffitt
12
FIGURE 2. SSI Applications and Awards among Population Aged 18-64 Apptcations 10
9 8 .2
0.
C
-
Awards
0.
a a
2
0
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
Year
Source: Daly and Burkhauser (forthcoming).
this shift, as the availability of SSI benefits allows single mothers to participate less in the labor market than they otherwise would. Some research has also been conducted on exploring reasons for the low participation rates of SSI-eligible elderly. Lack of information about the program does not seem to be a factor, but financial need does, for many nonapplicants have alternative sources of income. Nevertheless, the research on this subject has failed to clarify sufficiently why so many eligibles fail to apply for the program. Research on the work incentives of the SSI program has yielded rather discouraging results to date for the disabled. Disabled SSI recipients appear to be relatively unresponsive to financial incentives, and experimental tests of programs which offer financial and other incentives to
undergo additional training or vocational rehabilitation have experienced very small takeup. In addition, as noted previously, very few recipients take advantage of the less than one-for-one benefit reduction rate in the program. A major and continuing policy challenge in the program is the search for mechanisms to encourage and allow disabled recipients to fulfill their employment potential.
4. THE EARNED INCOME TAX CREDIT The EITC, as noted by Hotz and Scholz (forthcoming), has been one of the fastest-growing means-tested tax programs in the country. Its popularity
Economic Effects of Means-Tested Transfers in the U.s.
13
stems from its emphasis on rewarding families that have significant levels
of employment and earnings. The program provides a refundable tax credit, which can be as high as $3,800 a year (1999), to families with earnings. A small credit to childless families is also available. The program was introduced into the tax code in 1975, but did not see significant expansion in generosity until the 1980s, when the size of the subsidy was increased and then indexed to inflation. Tax bills in 1990 and 1993 increased the amount of subsidy greatly and have led to the sizable growth in expenditures in the 1990s which was noted earlier. The subsidy is obtained by filing a tax return and reporting the number of qualifying children in the household and the earnings of the father and mother. The size of the tax credit is proportional to earnings up to some maximum level, and then is phased out as earnings increase. In 1999, for example, a two-child family could receive a credit equal to 40 percent of their earnings up to an annual earnings level of $9,540, and the credit was phased out at a 21-percent rate until the credit fell to zero, which occurred at an earnings level of $30,580. Thus families fairly high in the income distribution were eligible for benefits. Fourteen states and the District of Columbia have state EITCs which provide for further tax credits. An issue in the administration of the EITC has been overpayment of subsidies, which in 1995 were estimated to be 25 percent of tax expenditures. Most of these result from inaccuracies in the claim for qualifying
children. While the overpayments are high, it is often noted that
17
percent of taxes are not paid to the IRS overall and more than 25 percent of taxes are not paid for some forms of capital income and income from the informal sector. In addition, despite the overpayments in the EITC, participation rates of EITC eligibles appear to be less than 100 percent, sometimes much less so. For example, participation rates among eligible
single mothers, who historically have low tax filing rates, have been estimated to be in the range of 42 to 54 percent. Much of the research on the EITC has concerned its effects on work incentives, since this is one of the main appeals of the program. While the EITC should increase labor force participation, 77 percent of families eligible for the credit fall into the flat or phaseout region of the credit, where there are more likely to be work disincentives than work incentives. Most studies have indicated that there is a strong and significant
positive effect on the labor-force participation rates of single-mother households. The participation rates of such households have risen markedly over the last decade, and the FITC is one of the leading causes of that increase. At the same time, however, research has suggested that the program may have had a slight negative effect on the employment
14
Moffitt
rates of married women, for many women are married to men who earn sufficiently high wages that additional earnings from the wife fall into the phaseout region of the EITC (that is, the region where additional earnings actually reduce the amount of the credit). In addition, there is
some evidence that, while increasing employment rates overall, the EITC may have lessened the hours of work of men and women in twocareer families. There has also been some concern that the EITC may discourage marriage because men and women in certain earnings ranges can receive a greater EITC sum by not marrying and filing separate returns than by marrying and filing joint returns. The empirical evidence to date, however, suggests little effect of this incentive on actual patterns of marriage. There has also been some research on the advance payment option in the EITC by which recipients can receive their credit over the tax year in question, as they earn wages, rather than in a lump sum at the end of the year or in the following spring. Some observers believe that the work incentives of the EITC would be greater if recipients could see the link between their earnings and the credit more quickly and immediately. Hotz and Scholz point out the high administrative costs of making this option more widespread, however, and describe the potential for noncompliance and fraud which would make monitoring procedures necessary. The advance payment option is used in the United Kingdom in a somewhat different program, but it appears that little monitoring for noncompliance is conducted there.
5. FOOD AND NUTRITION PROGRAMS As discussed by Currie (forthcoming), the Food Stamp program (FSP) is
only one among several programs that support food expenditure and nutrition among low-income families. The FSP is the largest, but also important are the Special Supplemental Nutrition Program for Women,
Infants, and Children (WIC), the National School Lunch Program (NLSP), and the School Breakfast Program (SBP). Expenditures on the latter three
programs are over 50 percent of those of the FSP, thus constituting a
sizable additional amount of spending. All of these programs are federally financed and uniform across the states. Also, there are many other
smaller programs that support food and nutrition in the U.S., covering smaller and more specialized populations. The FSP provides food assistance to families and individuals, regardless of family structure, who meet income and asset conditions. Families on TANF, SSI, or general assistance are automatically eligible. The nominal tax rate on earnings in the programthe amount by which benefits
Economic Effects of Means-Tested Transfers in the U. S.
15
are reduced for each dollar increase in incomeis 30 percent, but this rate is affected by the presence of deductions and exemptions. Benefits have historically been paid by the issuance of paper coupons but recently have been increasingly paid by electronic transfers using debit cards. This change has been thought to reduce the incidence of fraudulent selling of paper coupons, although the extent of that activity has never been accurately determined. The FSP began as a small pilot program in 1961 and was gradually expanded over time, finally being mandated for all counties in 1974. It was indexed to inflation in the 1970s, thus preventing the decline in real benefit amounts experienced in the AFDC program. The program was largely untouched by the 1996 federal welfare legislation that restructured AFDC, although work requirements and eligibility for certain categories of single adults as well as immigrants were restricted. The WIC, NSLP, and SBP are quite different. WIC provides financial assistance for the purchase of nutritious foods, nutrition education, and access to health services for pregnant or lactating women and children under 5. It is thus aimed specifically at improving nutrition among women and young children. Eligibility requires not only low income and assets but also that the women and children be at "nutritional risk," such
as having inadequate or inappropriate nutritional intakes, specific nutrition-related health deficiencies, large weight-for-height, or a number of other measures that are set by the states. The NLSP and SBP allow children in low-income families to receive reduced-price or free school lunches or breakfasts. They are thus like the FSP in subsidizing food expenditure per se but like the WIC in having a specific target population. In addition, the meals provided to the children must meet USDA nutritional guidelines, although there has been some concern recently that the meals remain high in fat and low in certain nutrients. The NLSP
is the far larger program of the two, having almost five times larger expenditure than the SBP. Caseloads in the FSP rose in the late 1980s and early 1990s, but have fallen since the enactment of 1996 welfare reform legislation. The data show that this is partly the result of an improving economy as well, but partly a result of the decline in AFDC-TANF caseloads, for participation rates among eligibles have also declined. Expenditures and caseloads in the WIC and NSLP programs, on the other hand, having risen in the late 1980s and early 1990s, have continued to rise in the mid-1990s, albeit at a slower rate. There has been a considerable amount of research on the FSP, WIC, and NLSP programs. One area of research has focused on the effects of these programs on food expenditures, nutrient availability, and nutrient
16
Moffitt
intake. Research indicates unequivocally that the FSP increases food expenditures, although not dollar for dollar, implying that recipients re-
duce some food expenditure out of their own income and spend it on other goods. It also appears that the program increases nutrient availabilitythat is, the nutritional content of the foods purchased or brought into the homebut the evidence on nutritional intake (i.e., taking account of wastage and food eaten away from home) is much weaker. Evidence on the WIC program generally indicates favorable effects on child birth weight but also that the program tends to discourage breast feeding, which is generally preferable to using infant formula. The latter effect arises because the WIC program gives free formula to participating mothers. The effect of WIC on infant outcomes is more variable, but the evidence does indicate increases in nutrient consumption and reductions in the incidence of anemia. Research on the NLSP indicates that it improves nutrient intake.
There have also been a number of studies on the determinants of participation rates in the FSP, for such rates are generally in the range of 60 percent, and thus not all eligibles are in the program. The research indicates that three factors are important in explaining nonparticipation in the program: lack of information about eligibility for the program, transactions costs which make participation onerous, and the stigma of being a welfare recipient. Research on the WIC program indicates that administrative barriers to participation are an important factor in explain-
ing lack of takeup. In the NLSP, an additional factor is the nutrient content of the lunches offered, for it appears that the higher the nutrient content, the less likely students are to participate. Steps to make the school lunches both nutritious and appealing to students have been discussed. Two other areas of research on these programs concern whether the
FSP should be cashed outthat is, whether cash should be provided to recipients instead of food couponsand whether the programs have a negative effect on work incentives. The first of these issues is motivated in large part by the rather low levels of FSP coupons relative to private food expenditures of the poor, suggesting that the coupons simply sub-
stitute for private food spending and hence are no different than cash welfare to the recipient. Interestingly, both econometric evidence and evidence from cashout demonstrations indicate that Food Stamp coupons have a greater effect on food expenditures than does cash, creating a puzzle that has not been adequately explained. The second of these questions concerns the traditional issue of whether a welfare program such as the FSP, which provides assistance even to those who do not work, has work disincentives. The several studies on this issue show
Economic Effects of Means-Tested Transfers in the U.S.
17
relatively little labor-supply response to the program, perhaps because its benefits are small relative to other forms of income received by the household.
6. The AFDCTANF Program Despite its decline, in terms of caseloads and expenditures, to a point where it is only the fifth-largest means-tested program in the country, the AFDCTANF program continues to receive the most attention from policymakers, the general public, and researchers. In his review of past and current research developments in the program, Moffitt (forthcoming) charts its growth and decline over the last three decades and reviews the research conducted on it. The program was created by the 1935 Social Security Act, was targeted at low-income children living with only one biological parent, and was intended to support widows with children. The caseload grew slowly through the 1950s and then accelerated in the 1960s and early 1970s. Subsequent to the 1970s, benefit levels in the program declined in real terms and an emphasis on work requirements steadily grew. The 1988 Family Support Act mandated employment programs in all states but required a human-capital, education-and-training emphasis to be part of the program mix. But the 1996 Personal Responsibility and Reconciliation Act (PRWORA) changed the program in more
fundamental ways, by devolving the responsibility of major program design elements as well as financing to the individual states, imposing strict work requirements in order to qualify for federal aid, and imposing lifetime limits on the number of years of benefit receipt which could be paid to a parent out of federal funds.
Table 3 shows the major elements of the 1996 Act and how they changed the program. The legislation converted the previous matching grant to a block grant and removed much of the federal regulatory authority over the design of the program, leaving the states free to set the benefit level, tax rate, income limits, asset requirements, and even the form of assistance (cash or in-kind services). In addition, no federal definition of who is to be included in the assistance unit is imposed; states can cover two-parent families at their own discretion, for example. The entitlement nature of the program is abolished, and states are not required to serve all eligibles. At the same time, however, the law imposed new federal authority in a few specified areas. Federal funds are not to be used to pay adults for more than 60 months of TANF benefits over their lifetimes (although states are allowed an exemption from this requirement for 20 percent of their caseloads), and new work requirements are imposed which require that states engage much greater
18
Moffitt
TABLE 3
Comparison of the AFDC and TANF Programs TANF
AFDC
Characteristic Financing
Matching grant
Block grant
Eligibffity
Children deprived of support of one parent or children in low-income twoparent families (AFDC-
Children in low-income families as designated by state; AFDC-UP abolished. Minor mothers must live with parents; minor mothers must also attend school
UP)
Immigrants
Illegal aliens ineligible
Aliens ineligible for five
Form of aid
Almost exclusively cash
payment
States free to use funds for services and noncash benefits
Benefit levels
At state option
Same
Entitlement status
Federal government required to pay matched share of all recipients
No individual entitlement
Income limits
Family income cannot exceed gross income limits
No provision
Asset limits
Federal limits
No provision
Treatment of earnings disregards
After 4 months of work, only a lump-sum $90 deduction plus child care expenses; nothing after 12 months
No provision
Time limits
None
Federal funds cannot be used for payments to adults for more than 60 months lifetime (20 percent of caseload exempt)
JOBS program
States must offer a pro-
JOBS program abolished
gram that meets federal law
years after entry and longer at state option
Economic Effects of Means-Tested Transfers in the U.s.
Work requirements
Parents without a child under 3 required to participate in JOBS
19
Exemptions from work requirements are narrowed, and types of qualified activities are narrowed and prespecified (generally excludes education and classroom training) and must be 20 h/week rise to 30 h/week for single mothers
Work-requirement participation requirements
JOBS participation requirements
Child care
Guaranteed for all JOBS participants
No guarantee, but states are given increased child care funds
Sanctions
General provisions
Specific provisions mandating sanctions for fail-
Participation for work requirements rises to 50% by FY 2002
ure to comply with work requirements, child support enforcement, schooling attendance, and other activities
Child support
States required to allow first $50 of child support received by mother to not reduce benefit
No provision
Source: Burke (1996).
fractions of the caseload and which exempt many fewer families (as many as 50 percent of single-mother recipients and 90 percent of twoparent families must comply). Recipients involved in general education and training cannot be counted toward these participation requirements; most activities require direct work. The most important new features are the time limits and work requirements. Lifetime time limits are a new concept in U.S. transfer programs and are based on a quite different philosophy of the aims of public assistance than has been the case heretofore, namely, that families are only entitled to temporary assistance. States have embraced time limits with vigor: half of them have chosen to adopt time limits even shorter than the federal five-year maximum. The work requirements in the new legislation are much stronger than in previous law and change its orientation
20 Moffitt
from education and training to work per Se. Indeed, most states have
adopted a work-first approach in which recipients and new applicants for benefits are moved as quickly as possible into work of any kind, with a deemphasis on education and training. The law also allows states to impose sanctions on recipients for failure to comply with the work requirements, sanctions which are much stronger than in past law and which have been actively enforced. With the aim of reinforcing the effect of these work requirements on employment, states have generally lowered their tax rates to encourage work as well, a feature that historically has been strongly supported by economists who believe they will provide work incentives. Another new goal of welfare programs in the 1990s has been to reduce the rate of nonmarital childbearing and to encourage marriage. Although there were few provisions of the PRWORA legislation that were directly aimed at these family-structure outcomes, the provisions aimed at reducing the amount of government assistance and encouraging women to sustain themselves off welfare were thought to implicitly encourage marriage and discourage nonmarital childbearing. As shown previously in Table 1 and as discussed earlier, caseloads in the AFDCTANF program have been falling for several years, and real expenditures have been declining. The per capita TANF caseload is now below what it was in 1970. The decline began prior to 1996 but accelerated thereafter. Expenditures have also changed in composition, as a smaller fraction is devoted to traditional cash expenditures and a larger fraction is devoted to noncash expenditures on services such as child care and other social services, reflecting a preference by states to support families in those ways. Real benefits also fell from the 1970s until the mid-1990s, when the decline abated and benefit amounts leveled off. It is important to note that the recent reforms have contracted the program in many ways, but reductions in benefit amount have not been one of them. The characteristics of AFDCTANF recipients have changed in some ways over time, but not dramatically. The one major change has been a shift in the types of single mothers on the rolls. Whereas the program began with a caseload composed mostly of widows, it shifted in the 1960s and 1970s to one composed mostly of divorced women. It then shifted again in the 1980s to one composed mostly of never married single mothers who have had children out of wedlock. These trends partially parallel larger trends in the society. They may also partially explain the decline in popularity of the program among voters. Economic research on the AFDC and TANF programs has been large in volume. The most heavily researched issue is the effect of the program on
Economic Effects of Means-Tested Transfers in the U. S.
21
labor supply and work effort. The research on this issue indicates that transfer programs like AFDC which provide open-ended support reduce work effort and that providing unrestricted benefits to those who do not work has work disincentives. At the same time, research on the effect of
reducing tax rates on recipientsby increasing earnings disregards
shows it to have a much smaller positive effect on overall labor supply than expected, because tax rate reduction has the offsetting effect of drawing additional women onto the rolls and inducing them to decrease their work effort. One way to reduce this offset is to provide additional work incentives to those off welfare as well as on, that is, to provide a more universal work subsidy (to low-income families) that is not tied to welfare.
The EITC is one program of this kind. Other programs that have been proposed allow women to "take" their subsidy off the rolls and to continue to obtain earnings supplements afterward. Under TANF, most states have increased their earnings disregards but the existing evidence therefore suggests that this will have little effect on overall work effort. However, the new work requirements are more likely to have a positive effect. Theoretical research on work requirements strongly suggest they wifi increase work effort, as should be expected, with a possible cost if some recipients who need assistance cannot comply and leave the welfare rolls. However, work requirements require that the welfare system be converted to a categorical program which divides recipients into those who can and cannot work, and imposes the work requirements only on the former. Dividing the caseload up in such a way is difficult and opens the door to possible inequitable treatment as borderline cases are assigned to one group or another and either gain or lose as a consequence. The theoretical literature in this area cannot answer the practical question of how well categorization can be implemented but it does highlight the tradeoff between better targeting of assistancethat is, providing work requirements to some recipients but benefits without requirements to othersand the costs of making that separation. Another area of economic research has focused on the relative merits of the work-first approach embodied in TANF and more education, human-capital oriented approaches. There have been a long series of evaluations of different types of employment and training programs in the AFDC program which, overall, show that modest positive effects on earnings can be achieved with relatively low-cost job search and job assistance programs even for the very unskilled population that the program covers. Typical gains, for example, are in the range of $300-600 per year. Research to date on work-first programs indicates that they have a more immediate effect on employment and earnings than human
22
Moffitt
capital programs, but one which fades out over time. Human capital programs appear to have effects that are more long-lasting. No clear winner emerges in this comparison, and many researchers have taken the rather different tack of investigating whether different programs might be tailored to different individuals, commonly called a mixed strategies approach, in which some recipients are deemed sufficiently
job-ready that a work-first approach is best while others are seen to be in more need of basic skills training. The other major feature of TANF, time limits, has also been the subject of considerable research attention. Researchers have noted that, while time limits should eventually force recipients off the welfare rolls with consequent increases in employment and earnings, recipients may also take action to leave the rolls early in order to bank their benefits for a later time when they are most in need of them. Indeed, the most important development thus far is how few recipients have hit time limits. The
massive reduction in the caseload, whether it has been the result of a good economy or of welfare reform, and regardless of whether some of it has been the result of banking behavior, has had the result that many families have not used up their years of eligibility even though five years have passed since 1996. However, this may change as the economy slows down and as states with five-year time limits experience more families hitting those limits. A large volume of research has been devoted to estimating the overall effects of the 1996 PRWORA legislation and of the creation of the TANF program. Descriptive evidence, for example, reveals that employment rates of single mothers have increased and that the incomes of all but a
small lower tail of the distribution have risen since the legislation. A sizable body of research has sought to disentangle the effects of the economy from those of welfare reform in explaining these trends. The majority of results from this literature indicate that the law has indeed had a significant and large additional effect beyond that of the strong economy. Prima facie evidence for this view is that, after 1996, the decline in the national unemployment rate slowed but the decline in the AFDCTANF caseload accelerated.
Another area of research has focused on the effects of welfare benefits on family structure and other demographic outcomes. Research on the
effect of AFDC on these outcomes is quite large in volume and has gradually moved, over the past decade, to a consensus that there are some non-zero effects of this kindthat is, that variation in AFDC benefits across states, received primarily by single mothers and not twoparent families, tends to be positively correlated with the rate of single motherhood. Research into the causes of the time-series increase in sin-
Economic Effects of Means-Tested Transfers in the U.S.
23
gle motherhood in the U.S., on the other hand, suggests that that increase is primarily the result of deeper economic and social forces such as the rise in female job opportunities and the decline in unskilled male wages. The TANF program, while having few direct provisions relating to marriage or fertility (aside from family caps), nevertheless was intended to have a positive effect on marriage and a negative effect on nonmarital fertility. The research evidence to date, however, is mixed at best in its results. There is little sign in the data of a strong effect of welfare reform per se on these outcomes, again, perhaps, because they are so driven by larger social, cultural, and economic forces. While the research evidence on the overall effects of welfare reform is
by now reasonably large in volume and has yielded important new findings, most of the research on the effects of detailed individual provi-
sions of TANFtime limits, work requirements, and so onhas unfortunately foundered on difficulties of evaluating their effects. By and large, researchers have not been successful in using the variation in programs across states to isolate the independent effects of these individ-
ual components of reform, and to estimate how much of the overall effect would have occurred if all elements of reform had been enacted except each of these components, in turn. The cross-state variation under TANF is sufficiently great, and the types of program variation so complex, that the effects of the components per se have not thus far been sufficiently isolated. Random-assignment evaluations could in principle do better, for they could be designed to alter each component while holding the others fixed, but they have not been designed in that way thus far. Finally, there has been considerable research attention paid to the effects of the block grant system put in place by PRWORA. The conventional view based on existing research on the effects of matching grants, which were used for AFDC, and block grants, which are in place for TANF, is that they have different price effects because the latter does not subsidize state expenditures above the block grant at all, whereas the former does. This should curtail spending over the block grant amount. Theoretical research has also shown the possibility of a "race to the bottom," as states facing a high price of expenditures lower their benefits to avoid immigration of the poor from other states, causing a cascading series of benefit reductions by all the states. However, to date none of these effects have occurred because the block grant allocations made to the states are generally much in excess of what states are spending, primarily because of the marked decline in the caseload and consequent reduction in state spending. Ascertaining whether the block grant structure wifi lead to restricted state spending or to benefit reductions around
24 Moffitt
the country will have to await a period when welfare spending rises up to the block grant level, where it wifi become binding.
7. HOUSING PROGRAMS The set of housing programs for low-income families in the U.S. consti-
tute a complex mix of programs with different features. As noted by Olsen (forthcoming), new programs have been added to the system over time and the rules of existing programs have changed frequently. These programs are much more expensive than commonly realized because they rely to a much greater extent than other welfare programs on indirect subsidies that do not appear in the records of the administering agency. Programs divide up into those that are project-based (owned either by
the government or by private contractors who are subsidized by the government) and those which are tenant-based, in which eligible families receive subsidies to defray the rent in private housing. The public housing program, begun in the 1930s, is the best-known project-based program; it offers rental units to low-income families in newly constructed projects owned and operated by the government. Beginning in 1954, the government began in addition to contract with private parties to construct low-income housing or to rehabilitate existing housing for this purpose, in most cases insuring the mortgages of the contractors. The Section 8 New Construction Program established in 1974 is the
largest program of this type. Under this program, the government subsidized the construction costs of privately built housing for low-income families and provided monthly rental payments. In 1983, Congress halted additional commitments under HUD's new construction programs except for small programs for the elderly and disabled. Today the largest housing subsidy program is tenant-based.3 The Housing Choice Voucher Program enacted in 1998 consolidated the two variants of the Section 8 Existing Housing Program that had operated simultaneously for fifteen years. This program pays a portion of the rent of eligible low-income
households that locate housing in the private market that meets the program's minimum housing standards. Although HUD or USDA programs have produced few new units in recent years, the IRS's Low Income Housing Tax Credit, enacted in 1986, will soon become the However, as Olsen discusses, the Low Income Housing Tax Credit, enacted in 1986, pays 70 percent of the development costs of projects for low-income families and has become the second-largest housing program in the country.
Economic Effects of Means-Tested Transfers in the U.s.
25
second-largest housing program in the country and it is growing much more rapidly than any other program. Eligibffity for these housing programs is based on a number of factors, the first being the requirement that adjusted family income fall below certain thresholds determined by family size and the median income in the locality. However, because fixed budgets are authorized for these programs and there is excess demand for subsidies, they must be rationed; that is, housing programs are not entitlements. Local housing authorities and owners of private projects, operating under general guidelines from Congress, determine their preferences in granting assistance to individual families by giving weight to characteristics of the families. Once assistance is granted, families in project-based programs are offered specific units and families in tenant-based programs are authorized to locate eligible units in the private market. A substantial majority of assisted families participate in programs that require them to pay 30 percent of their income toward rent. However, many assisted families pay rent that is independent of income. Research on housing programs has addressed a number of different topics. One concerns the cost-effectiveness of different program types. The studies are unanimous in finding that tenant-based assistance provides housing equal in quality to that of project-based assistance at a much lower total cost. Another key issue is whether the programs indeed increase the housing consumption of their recipientscertainly a main goal of the program, but not one guaranteed to occur, at least for projectbased housing. The literature indicates indeed that housing consumption is raisedthat is, that families occupy higher-quality housing than they would in the absence of the programboth in housing projects and housing occupied by voucher recipients. There appears to be some leakage in the subsidy, for consumption of nonhousing goods rises as well, although this should be expected if part of the goal is to enable families to reduce what are often very high housing expenditures. The housing programs also appear, according to the research, to increase housing consumption more than would a pure cash grant, consistent with the rationale for housing assistance. The Housing Allowance Supply Experiment conducted in the 1970s studied the market effects of an entitlement housing voucher program similar to the limited enrollment Section 8 Voucher Program that operated between 1983 and 1998. The Supply Experiment operated for ten years in two small metropolitan areas with very different initial vacancy rates and minority populations. About 20 percent of the families in the two counties were eligible to receive assistance. Participation rates of
26
Moffitt
eligibles never exceeded 50 percent, partly because subsidies for those with moderately high incomes were not large enough to outweigh the costs of moving and participating in the program. The results showed that an entitlement program of tenant-based assistance would produce a substantial increase in the supply of dwelling units meeting minimum housing standards but would have little effect on rent levels. There have been a few research studies on other topics as well. Some examine the work disincentives of housing programs, on the presumption that, like all welfare programs, the reduction in the subsidy with an increase in earnings wifi reduce the incentive to work. The results show that such work disincentives probably exist but that they are quite small. Another set of studies examine the relative effects of public housing and tenant-based housing on the choice of neighborhood, finding that public housing exacerbates economic and racial segregation while tenant-based subsidies ameliorate them to some extent. Findings from the recent Moving to Opportunity Experiment that offered randomly selected families in public housing vouchers on the condition that they move to neighborhoods with very low poverty rates indicate improvements in the educational attainment of the children involved and reductions in their violent criminal behavior. It also increases the earnings of adults in these families.
8. CHILD CARE PROGRAMS Blau (forthcoming) describes the structure of means-tested child care pro-
grams in the U.S. He notes that such programs have at least three
different goals, not always mutually compatible. One is to increase the rate of employment of low-income women, particularly when operated through the AFDC-TANF program or when aimed at assisting lowincome parents in general to work. A second is to increase the quality of child care for low-income children, and a third is to assist in the development of disadvantaged children through early education programs such as Head Start. These goals may conflict, as they do for programs which encourage low-income women to work through the provision of inexpensive child care.
A variety of programs serve one or more of these goals. One of the largest is the Child Care and Development Fund (CCDF), which provides funds to states to subsidize child care for low-income families and is intended to support employment of low-income parents. It was created in 1996 by Congress and consolidated a number of prior programs, some of which had served primarily the AFDC population and some of which had served the "working poor," meaning low-income families not on AFDC. When requiring such consolidation, Congress also required
Economic Effects of Means-Tested Transfers in the U.s.
27
that minimum percentages of the grant be spent on the AFDCTANFbased population (not only current recipients, but also families who had
recently left AFDC or are at risk of going on) and that minimum percentages be expended on the working poor. A second program, the Title XX Social Services Grant, provides states with funds to expend on a variety of social services for the poor, including child care; states spend approximately 15 percent of their funds on that service. The Dependent Care Tax Credit, a nonrefundable tax credit in the federal income tax, also
provides a subsidy for child care, which declines as income rises. Finally, three programs are intended for early education and child development, and are not tied to parental employment. These include the Head Start and Title I-A programs, which provide early education for disadvantaged children, and the Child Care and Adult Food Program, which provides subsidies for nutritious meals in child care settings for lowincome children. All the programs are federally financed and have uniform national rules, except the CCDF and Title XX Social Services Block
Grant, which give considerable discretion to the states on operation
within federal guidelines. In terms of expenditure on low-income families, the CCDF and Head Start programs are currently the largest at about $5 billion each, followed by the Dependent Care Tax Credit at approximately $3 billion. The number of children served is the largest in the Dependent Care Tax Credit, followed by the CCDF; Head Start is one of the smallest. Expenditures per child are essentially inversely related to size: the Dependent Care Tax Credit gives $720 per child, while Head Start gives $5,759 per child. Eligibility in the three federal early-education programs is related to various measures of low income and is nationwide, while eligibility in the CCDF and Title XX Grant is set by the states within federal guidelines. Title XX funds must be spent on children in families with income below entirely state-chosen limits, while CCDF funds must be spent on families with income no greater than 85 percent of state median income. States are free to set a wide variety of subsidy mechanisms in their CCDF-funded programs, with fees that have maximums and minimums or are waived
for certain groups, with vouchers or direct contracts with providers, and with flexibility in setting reimbursement rates for providers. Child
care facilities must meet state licensing and regulatory requirements. Research on the effects of child care programs has been concentrated on a few selected issues. One is whether child care subsidies in general increase the employment of mothers. Based on evidence from demonstrations and random-assignment trials as well as from nonexperimental studies which use variation in child care price to estimate employment responsiveness, the literature strongly indicates that child care subsidies
28
Moffitt
increase employment and hours of work. However, the magnitude of the effect is quite uncertain and varies considerably across studies. In his review, Blau finds that the studies with the best data and which account most realistically for the child-care market find relatively low price elasticities of employment response, but ones that are still statistically significant.
Research on several other issues has been conducted as well. One
study of the effect of price on the quality of care chosen by parents using formal day care centers found that child care subsidies led parents to use more care but care at lower-quality centers, as measured by childstaff ratios and staff training. Another study examined the effect of child-care
subsidies on the probability that a single mother would be on AFDC, and found that such subsidies lower that probability, presumably by allowing AFDC mothers to go off welfare and work. Blau also reviews the large literature on the effect of early childhood education on child outcomes, finding that the evidence supports an effect of such education on some outcomes for some programs. Whether the effects fade out over
time or persist is more controversial, although some studies do show persistent effects.
9. EMPLOYMENT AND TRAINING PROGRAMS As discussed by LaLonde (forthcoming), the main omnibus employment and training program in the U.S. at the present time is the Workforce Investment Act (WIA). Passed by Congress in 1998 and taking effect on July 1, 2000, WIA replaced the Job Training Partnership Act (JTPA),
which was the main program for employment and training in the U.S. from 1982 to 1998. WIA provides block grants to the states to fund employment and training programs for adults and youth. WIA contains several titles with different programs and different services. These indude Title I.B.5, which covers adults; Title 1 .B.4, which covers youth; and Title 1 .C, which covers the Job Corps, a high cost training program for disadvantaged youth. Except for the Job Corps, states have great freedom to design their own WIA-funded programs but must meet certain federal requirements. The adult programs are not restricted to low-income persons, but priority must be given to cash welfare recipients. Training programs for youth, on the other hand, both the Job Corps and other youth programs, do require that the recipient have low income and other measures of economic disadvantage. All adults are eligible for job search assistance, but more comprehensive services require that the recipient be unemployed and be unable to find a job or otherwise need intensive services
Economic Effects of Means-Tested Transfers in the U.S
29
to maintain employment. Training is primarily provided through individual training accounts, which allow the individual to choose from a list of
acceptable providers, and thus retains some features of a voucher. A system of "one-stop shopping" is required by the legislation, allowing WIA enrollees to go to only one agency, provider, or location to be directed to all services. There are three generic types of training programs typically provided. One general category is aimed at enhancing skill development, and in-
cludes both classroom training and on-the-job training. A second is "work experience," which involves temporary placement in an actual job. A third is employability development, which includes job search assistance and career counseling. The first is aimed at increasing the individual's long-run labor-market skill level, while the second and third are aimed more at encouraging immediate employment. Typically, an
individual's needs are first assessed, and then he or she is assigned to one of these types of programs. There is no research on the WIA program, because it has been put in place so recently. However, there is a large body of research on JTPA and related training programs, which should still be quite relevant to WIA, given that the basic types of programs are unlikely to change markedly. The majority of the research surveyed by LaLonde is from random-assignment evaluations, where the effects of the training program in question are measured as the difference in outcomesusually earnings and employmentbetween an experimental group and a control group. These experiments typically estimate training effects separately for adult women, adult men, and youth. Adult women are always separated because they include a high fraction of welfare recipients and,
indeed, many of the training programs are explicitly targeted at that group. The findings are quite different by group. For adult women, there is consistent evidence of positive effects of a variety of types of training programs on employment and earnings. The programs include weifareto-work programs tested by individual states on their welfare recipients, high-cost programs for disadvantaged women in general (such as Sup-
ported Work), and the JTPA program. The effects persist for several years and occur for all program types (job search assistance, work experi-
ence, and employability development). For the job search assistance programs, the magnitudes of the effects are modest in sizeranging up to $500 per year, typicallybut are also very modest in cost, leading to very favorable cost-effectiveness ratios. Programs that provide classroom instruction or which add work experience on top of job search assistance have somewhat greater effects. The high-cost programs, such
30
Moffitt
as Supported Work, have even greater effects, ranging up to $1,000 per year. Whether they are cost-effective depends crucially on whether these effects are permanent or fade over time. In the favorable event that they are permanent, even these programs have strong cost-effectiveness ratios (e.g., 15-percent rates of return).
The estimates for adult men and youth are more mixed. For men, most evaluations show little effect on employment and earnings overall. Some programs appear to have positive effects for certain subgroups of men, but the pattern does not have any clear explanation. For youth, effects estimated in the Supported Work experiment as well as JTPA and other training programs are typically very small. However, for youth,
the Job Corps has traditionally been thought of as the main program showing favorable results, based on past evaluations. A new experimental evaluation confirms that its effects on employment and earnings are positive. Four years after enrollment, annual earnings were on the order of 12 percent higher. The effectiveness of the Job Corps in comparison
with some of the other youth programs is thought to arise from the comprehensiveness of its training services as well as the relatively large expenditures on it. However, LaLonde also shows that subgroup analyses present a mixed picture of Job Corps effects, which vary markedly by age and ethnicity. Indeed, for some subgroups (e.g., 16-24 year old Hispanics) the program appears to have no effect. There also is some support in the data for effects being greater for young adults than for teenagers, but even this result is not completely uniform. LaLonde concludes his review by emphasizing the positive findings for adult women and youth. For adult women, low-cost training programs have a fairly large effect relative to cost and constitute what ap-
pears to be a worthwhile investment. He notes that the cost of these programs is far less than the cost of a year of formal schooling, for example, and should not be expected to have dramatic impact as a result. Higher-cost programs may be cost-effective as well, but this depends on the size of their long-term effect, about which little is known. For youth, it appears that only high-cost comprehensive training programs are likely to be productive social investments. The U.S. spends far
less than other countries on training programs, and this evidence suggests that a greater expenditure in that direction could increase the earnings of many groups in the disadvantaged population.
10. CHILD SUPPORT The child support system in the U.S., while not formally a means-tested program or a public transfer program at all, nevertheless plays an impor-
Economic Effects of Means-Tested Transfers in the U.S.
31
tant role in policy discussions on transfers to the low-income population and to single mothers in particular. Lerman and Sorensen (forthcoming) note that the Child Support Enforcement (CSE) system, the governmental program aimed at enforcing private child support obligations, is concentrated on the low-income population. In their chapter, Lerman and
Sorensen review the structure of the present system and the research that has been conducted on it. The CSE program was established by Congress in 1975 to provide matching funds to states to collect child-support obligations, establish paternity, and obtain support awards. States were required to provide child-support enforcement services to AFDC recipients and to any nonAFDC family that requested them. The statute also required that AFDC recipients assign their child-support rights to the statethat is, that any child support payments they received be taken by the state and used to compensate for the AFDC benefitand to cooperate in establishing paternity and securing support. Thus reducing welfare costs and increasing child support were both goals of the system, goals that have remained to the current time. Through legislation, Congress has steadily increased pressure on the states to strengthen the child-support enforcement system in many ways since 1975. In 1988 it set numerical goals for the states to establish paternity for children, and later required that states establish voluntary pater-
nity acknowledgement procedures in hospitals. In 1984 and again in 1988, Congress increased pressure on states to require judges to adhere to state child-support guidelines governing the setting of child-support awards, which are generally tied to the income of the noncustodial parent. This was aimed at preventing judges from setting child-support awards that were too low. Over the 1980s, Congress also increased requirements on states to use wage withholding to obtain payments from non-custodial parents, and in 1996 went further by requiring that every new hire be reported to the CSE agency in order to locate such noncustodial parents who were delinquent in their payments and had not
been locatable by the agency. The fraction of low-income custodial mothers who receive any child support at all was only 24 percent in 1997, and of those that receive child support, even fewer receive the full amount that has been awarded by the court. These low figures, despite the years of increased stringency of child-support enforcement, attest to the difficulty of the problem. The
fraction receiving any support is, however, larger than it was twenty years ago, when it was only 17 percent. The increase has arisen from a greater percentage of poor custodial mothers who actually have an award, which is no doubt partly a result of governmental efforts at
32
Moffitt
establishing paternity and encouraging awards. The increase would have been larger had it not been for a decline in the fraction of mothers who actually received anything even if they had an award. Part of the reason for this decline, though not all, has been a shift in the composition of poor custodial mothers from those who are divorced or separated to those who have never been married; the latter have always received less support than the former. The fraction of single mothers on AFDC receiving child support is approximately 17 percent, even lower than that of all poor custodial mothers. Research on child-support issues has had several purposes. One is determining the incomes of poor noncustodial fathers in order to determine how much they are capable of paying. This is a difficult task, because there is no ready data set to identify noncustodial fathers and their incomes, so most estimation is indirect. Estimates indicate that, overall, noncustodials fathers could pay 3 to 4 times more than they are actually paying, given their incomes and given customary guidelines for how child-support awards are based on income. However, no estimates are available for low-income fathers alone. Evidence from ethnographic
studies indicates that poor noncustodial fathers have high rates of
nonemployment, low levels of education, little work experience, and poor health, and often have criminal histories and unstable housing arrangements. Another area of research focuses on the effect of child-support collections on AFDC participation and on the work effort of welfare mothers. Because states collect most of the child support received by women on AFDC, an increase in child support paid by the noncustodial father has no impact on a woman's income while on welfare, but it increases income off welfare. This should therefore decrease AFDC participation and increase the labor supply. Although the evidence is not as strong as it could be, it does suggest that this is the case. Increases in CSE reduce AFDC caseloads, according to the evidence, and increases in child support reduce rates of AFDC participation and increase employment rates. However, there is also some evidence that increases in child support reduce the work effort of custodial mothers not on AFDC, for in this case the extra income allows them to reduce their hours of work or work effort overall.
Research in this area suggests that, in principle, child-support payments and CSE in particular might reduce the work effort of noncustodial fathers, as they are required to pay a percentage of their income toward support. However, the little empirical evidence available indicates little response of this kind. This may be because noncustodial
Economic Effects of Means-Tested Transfers in the U.S.
33
fathers have inelastic labor supply curves, but it may also be because only in a minority of cases do courts update award amounts as incomes of fathers change. Typically, award amounts are set in relation to income at the time of the initial court judgement, but no adjustments are made thereafter. Nevertheless, ethnographic evidence does suggest that child-support enforcement tends to drive many men into the underground economy, where income is not reported. Indeed, much of the research discussion of the incentives faced by noncustodial fathers focuses on the lack of incentives to pay child support given the fact that all payments go to the government instead of to the children if the mother is on welfare. An additional problem is that many men have accumulated large amounts of child-support debt, which are very difficult to work off. Attention has also been paid to the effects of child-support payments on marriage, divorce, remarriage, and nonmarital childbearing. The predictions of the effect are in most cases ambiguous, because, while increased child support gives men an incentive not to marry, remarry, or have children out of wedlock, it increases the custodial mother's incentives in the other direction by making single motherhood less financially onerous. The little evidence on the issue suggests that there are indeed effects in this direction, with child support appearing to reduce remarriage, nonmarital childbearing, and divorce, but their magnitudes are uncertain. Finally, there has been considerable research on the effectiveness of child-support enforcement policy itself on increasing paternity establishment, award rates, and payment of child support. The evidence sug-
gests that it has had an effect, particularly on the first of these. This is consistent with the time-series evidence mentioned earlier. Thus CSE policy has been shown to have an effect, and for this reason it continues to enjoy strong support as a public policy.
11. CONCLUSIONS Economic research on the effects of the nation's system of means-tested transfers has yielded a large volume of important findings. One of the most basic is the repeated finding that the programs are, by and large,
attaining their central goals of increasing the consumption of lowincome families of medical care, food, housing, child care, and other targeted goods. Another is that there has been an increased redirection of support toward the disabled, both adults and children, both for the receipt of cash support and for medical assistance, and toward needy
34
Moffitt
children off TANF, another worthy goal. A third is that the EITC has been successful in raising the employment rate of low-income single mothers, a long-sought goal of transfer policy in the U.S. At the same time, research has demonstrated that the attainment of other goals of these programs is still a challenge. Designing the transfer programs to provide strong work incentives which are acted on is still an issue in the SSI program, for example, and the EITC has some work disincentives for groups other than single mothers. The AFDC-TANF reforms have been successful in raising employment among single mothers, but the effects on their incomes are less unambiguously positive. The child-support system in the U.S. has made great improvements in support for low-income children, but too little support is still received by low-income mothers, yet the burden on low-income fathers is already onerous by many accounts. Effects of all transfer programs on family structure have become an important topic, but no program has been successful in making a major improvement. Designing reforms to address these and other issues wifi continue to make this a fruitful area of research.
REFERENCES Blau, D. "Child Care Subsidy Programs." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. ted Income: Burke, V. (1993). "Cash and Noncash Benefits for Persons with Eligibifity Rules, Recipient and Expenditure Data, FY 1990-92." Washington: Congressional Research Service. (1996). "New Welfare Law: Comparison of the New Block Grant Program with Aid to Families with Dependent Children." Report 96-72OEPW. Washington: Congressional Research Service.
(1999). "Cash and Noncash Benefits for Persons with Lted Income:
Eligibility Rules, Recipient and Expenditure Data, Fl 1996-Fl 1998." Washington: Congressional Research Services.
Currie, J. "U.S. Food and Nutrition Programs." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. Daly, M., and R. Burkhauser. "The Supplemental Security Income Program." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. Gruber, J. "Medicaid." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press.
Hotz, V. J., and J. K. Sholz. "The Earned Income Tax Credit." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. LaLonde, R. "Employment and Training Programs." Forthcoming in MeansTested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press.
Economic Effects of Means-Tested Transfers in the U.S.
35
Lerman, R., and E. Sorensen. "Child Support: Interactions Between Private and Public Transfers." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. Moffitt, R. (ed.) (forthcoming). Means-Tested Transfer Programs in the U.S. Chicago: University of Chicago Press. "The Temporary Assistance for Needy Families Program." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press.
Olsen, E. "Housing Programs for Low-Income Households." Forthcoming in Means-Tested Transfer Programs in the U.S., R. Moffitt (ed.). Chicago: University of Chicago Press. U.S. Department of Commerce, Bureau of the Census (2000). Statistical Abstract of the U.S.: 1996. Washington: Government Printing Office.
TAXES AND HEALTH INSURANCE Jonathan Gruber MIT and NBER
EXECUTIVE SUMMARY A common prescription for reducing the number of uninsured is to increase the tax subsidization of health insurance in the U.S. Yet, we already provide over $100 billion per year in tax subsidies to health insurance. This paper provides an assessment of the past and potential impacts of taxation on health insurance coverage and costs. I begin by reviewing the central facts on health insurance and taxation. I then provide a framework for assessing the impacts of tax policies on health insurance coverage and costs, and I review the existing empirical evidence on the key behavioral parameters required to model these impacts. I conclude with the policy implications of these findings for tax policies to expand insurance coverage.
1. INTRODUCTION Uninsurance is one of the worst social problems in the U.S., and it has continued to worsen as the economy has improved throughout the 1980s and the 1990s. In 1987, 14.8 percent of non-elderly Americans were without health insurance. Over the next decade, the non-elderly population without insurance coverage grew by nearly 25 percent, to 18 percent, before falling for the first time in two decades last year. Still, I am grateful to Jim Poterba for comments and to Robin McKnight, Tracey Seslen, and Avi Ebenstein for excellent research assistance.
38
Gruber
despite the recent good news, over 42 million Americans lack health insurance. The problem of the uninsured has been a major focus of policy debate throughout the 1990s. The most prominent example was the proposed national health insurance plan of the Clinton Administration, which was resoundingly defeated in 1994 (Cutler and Gruber, 2001). Tax incentives to expand health insurance coverage have also been considered continually through the last decade, and received particular attention during the presidency of George H. Bush and again in the presidential election campaign of 2000. In recent years, proposals have been made to offer tax
credits to individuals to buy insurance in the non-group market (by
President Bush); to expand those credits to cover the cost to individuals of their group insurance policies (by the Progressive Policy Institute); and to provide credits to firms to induce them to offer insurance (by a number of members of Congress). But the existence of these proposals should not be taken to imply that the U.S. doesn't already dramatically subsidize the provision of insurance in the workplace. In fact, the exclusion of employer (and some employee) health insurance expenditures from the income tax base costs the government over $100 billion in lost revenues annually (Shiels and Hogan, 1999). Indeed, some tax-based proposals would end or limit this exclusion of health insurance from the income tax base, and use the resulting funds to offer tax subsidies to individuals for insurance purchase. Disentangling the costs and benefits of these alternative approaches to tax subsidization of health insurance is difficult, and revolves centrally around a series of behavioral parameters that determine how individuals and firms will respond to changes in the tax treatment of health insurance. An incomplete list of such parameters includes: the price sensitivity of the decision of firms to offer health insurance; the price sensitivity of the takeup decision, conditional on offering, of employees; the price sensitivity of insurance demand among those not offered insurance; the influence of subsidies on the structure of employer-provided insurance plans, such as employee premium sharing. Despite the importance of these issues, however, we have remarkably little evidence on the key behavioral elasticities, and the evidence that does exist is often contradictory. The purpose of this paper is to lay out a framework for researchers and policymakers to think about how tax policies might affect the level and distribution of health insurance coverage in the U.S. I begin by reviewing the key facts that are relevant to thinking about health-insurance policy. I then turn to a discussion of the central parameters that we need to know to fully model both the effect of the existing tax subsidy and the effects of tax-based approaches to increasing insurance coverage in the
Taxes and Health Insurance 39
U.S. I then discuss what we know about these parameters. Finally, I discuss the implications of the facts and our existing knowledge for the design of tax policy toward health insurance in the U.S.
2. THE FACTS There are a number of key facts that must be considered when modeling the effect of taxes on health insurance in the U.S.
2.1 90% of Insurance Coverage Is Employer-Based The predominant source of insurance coverage in the U.S. is employerbased insurance. This is shown in Table 1, from EBRI (2000), which shows the sources of insurance coverage in the U.S. over time. A fairly constant feature of insurance coverage has been that more than 9 in 10 of
those who are privately insured derive their insurance from an employer, generally their own, their spouse's, or their parents'. Why is the employment setting the predominant source of insurance coverage? There are three potential reasons. First, there may be substantial economies of scale in administering insurance which increase the value of pooling mechanisms. Second, the major problem facing providers of insurance is adverse selection, so that insurers are constantly
searching for means of pooling large groups of individuals along dimensions exogenous to health in order to ensure a predictable distribution of medical costs. Workplaces provide just such a pooling mechanism. Finally, the U.S. tax code subsidizes health insurance purchase through the firm relative to the non-group market by excluding the value of that insurance from an individual's income, for both income and payroll tax purposes. This leads to a very large effective subsidy to the cost of health insurance for workers. The result of this subsidy is that there is a lower "tax price" of insurance:
TP -
1 - Tf - T
- T55 - Tmc
1 + T55 + Tmc
where Tf is the federal income tax marginal rate, ; is the state income tax marginal rate, ; is the marginal payroll tax rate for the OASDI program (the 6.2-percent tax rate that is levied equally on employees and employers); and Tmc is the marginal payroll tax rate for the Medicare HI pro-
gram.1 I differentiate the last two programs because, beginning in the
1 The reason that the payroll tax rate is additive in the denominator is that the employer is indifferent between purchasing one dollar of benefits and paying wages of 1/(1 + ;, + Tm,), since each dollar of wages requires a payroll tax payment as well.
TABLE 1
Dependent
35.4 35.0 34.7 33.8 33.5 32.9 30.7 30.9 31.1 31.2 31.5 31.7 32.4
Own name
33.8 33.9 33.9 33.1 32.8 31.8 32.9 32.7 32.7 32.9 32.8 33.1 33.4
Total
69.2 69.0 68.6 67.0 66.3 64.7 63.5 63.6 63.8 64.0 64.2 64.9 65.8
Year
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Employment-based
6.9 6.8 6.7 6.5 6.6
6.1 6.5 7.3 7.1
6.7 6.3 6.6 6.5
mdiv. purchased 13.3 13.3 13.2 14.5 15.5 16.0 16.7 16.9 16.6 16.0 14.8 14.3 14.2
Total
1.7 1.6 1.6 1.8 2.0 2.0 2.0 2.0
12.1 11.0 10.4 10.4
14.8 15.5 15.7 16.1 16.3 17.0 17.3 17.1 17.4 17.7 18.3 18.4 17.5
4.0 3.8
8.6 8.7 8.8 10.2 11.1 11.8 12.7 12.5 12.5 1.4 1.5 1.5 1.6 1.6
3.3 3.3 3.8 3.2 2.9 2.8 2.9 2.7
3.6 3.6 3.5
None
Tncare/ CHAMP VA
Medicaid
Public
Medicare
Proportion of population (%)
Insurance Coverage of Non-elderly Population over Time
Taxes and Health Insurance
41
early 1990s, the taxable maximum for the HI program was increased above that for the OASDI program (and was eventually removed altogether); the marginal rate is zero above the taxable maximum for payroll taxation. For a typical worker in the 15-percent tax bracket, facing a 5-percent state tax rate and a 15.3-percent combined payroll tax rate, this tax price is roughly 0.65; a dollar of health insurance costs 35 cents less than a dollar of other goods purchased with after-tax wages. In addition, there are also tax subsidies available to employees for their spending on employer-provided health insurance, under section 125 of the Internal Revenue Code. Section 125 generally provides that an employee in a cafeteria plan wifi not have an amount included in gross income solely because the employee may choose among two or more benefits consisting of cash and qualified benefits. A qualified benefit generally is any benefit that is excludable from gross income under the
Code, including health insurance, group-term life insurance, 401(k)
plans, child care, and adoption assistance. While employee contributions can therefore be excluded from taxation with a section 125 plan, protection of employee contributions is far from complete. The data on the prevalence of such arrangements are sketchy. The most recent available data, from Kaiser Family Foundation (2000), suggest that half of all workers are in firms that offered such flexible benefit plans. The reason for less than full coverage of this generous tax benefit is unclear, but some of it may have to do with extensive IRS regulation of these arrangements to ensure that they are not abused. For example, the regulations state that no more than 25% of the benefits of a plan can be attributed to any "highly compensated" employee, essentially ruling out the availability of section 125 plans for very small firms. Moreover, there are strict and complicated rules that limit the flexibility of employees to switch sources of insurance coverage during the year if they are paying their health insurance contributions on a pre-tax basis. Perhaps as a result of these inherent advantages of the group insurance market, the non-group insurance market has not provided a very hospitable environment for insurance purchase. Load factors in this market are high, and the generosity of the typical policy is much lower than in the group market (Gruber and Madrian, 1996). A recent study by the Kaiser Family Foundation found that, for those individuals in less than perfect health, it was often not possible to get coverage that was fully comprehensive, the particular illness of the individual often being underwritten out of the policy. Prices were also very variable in this market, making it difficult to effectively anticipate the cost of insuring oneself. There is an existing tax subsidy to the non-group market itself, for a particular group: the self-employed. Beginning in 1986, the self-employed
42
Gruber
were allowed to deduct 25% of their insurance premiums from their taxable income. This share has grown over time and is slated to reach 100% by 2003.
2.2 Employer-Based Coverage Has Been Eroding, with Public Insurance Picking lip Some of the Reduction The notable time trend in Table 1 has been the steady erosion of employer-provided insurance coverage in the post-1987 period. The share of the population with private coverage fell from 69.2 to 63.5 percent in 1993, before rising again slowly to 65.8 percent. This trend was partially offset in 1987-1993 by a sharp rise in the share of the population with Medicaid coverage, due to extensive expansions in that program, particularly for children. But the slow rise in employer coverage after 1993 has also been offset by a sharp decline in public coverage, particularly from 1996 to 1998, which may attribute to an unintended consequence of welfare reform (see Gruber, 2001, for a review of Medicaid program issues). A striking feature of the erosion in employer-provided insurance coverage is that it was not occurring through a decline in employer offering of health insurance, but rather through a decline in employee take-up of that insurance, conditional on it being offered (Cooper and Schone, 1997; Farber and Levy, 1999). A central, and unresolved, mystery is why we saw this decline in take-up. This period was marked by a rise in the
share of health insurance costs borne by employees (Gruber and
McKnight, 2001). But all available evidence, as I wifi review below, suggests that take-up of insurance by employees is not very sensitive to its
price. Indeed, takeup of employer-based insurance in general remains quite high, as I point out below.
2.3 Most of Those Offered Insurance Take It An key fact for designing tax policy towards the uninsured is that most of those who are offered insurance by their employers take it up. This is illustrated in Table 2, which represents tabulations of insurance take-up rates from a 1997 Robert Wood Johnson survey of employers. This table cross-tabulates firm size against average earnings in the firm, and in each cell lists the take-up rate of insurance, which is computed as the number of covered employees divided by the number of employees eligible for coverage. What is striking about this table (in particular in reference to the next table we wifi see) is the high and uniform rate of take-up across cells.
Taxes and Health Insurance 43 TABLE 2
Takeup Rates by Firm Size and Average Earnings Categories Takeup ratea
Average earnings ($) <15,000 15,000-30,000 30,000-50,000 >50,000
1-9 employees
10-24 employees
26-49 employees
50-99 employees
>100 employees
0.81 0.84 0.88 0.89
0.70 0.83 0.84 0.84
0.70 0.79 0.83 0.89
0.71 0.78
0.76 0.82 0.88 0.88
0.90 0.75
a From author's tabulations of 1997 Robert Wood Johnson Foundation survey of employers.
While take-up does rise somewhat with firm size and with earnings, it is quite high in every cell in this table. If employees are offered insurance, they appear to take it up uniformly at quite high rates.
2.4 Insurance Offering Is Highly Correlated with Firm Size and Average Wage The offering of insurance, on the other hand, does very quite significantly across firms, along both of these dimensions. Table 3 crosstabulates (using somewhat different categories for emphasis) firm size and average earnings again, this time summarizing in each cell the average rate of offering health insurance. Offering of insurance is in fact quite low for the smallest firms, even at high wage levels, and for larger firms at the very lowest wage levels. The correlation with firm size corresponds to the non-tax arguments made above for why firms would offer insurance: economies of scale and predictability of insurance expenditures both rise with firm size. Indeed, the Congressional Research Service (1988) reports that the loading factors on insurance are roughly 35% higher for the smallest than for the largest firms. The correlation with average wages may reflect preferences across firms for insurance offering; but it is notable that when insurance is offered in these low-age firms, employees then take it up at a high rate. This suggest some disconnect between the offering and takeup decisions, which I will return to in the evidence section below. Given these last two pieces of evidence, it should not be surprising the the primary correlate of being uninsured is not being offered insurance. Data from the Current Population Survey data show that over threequarters of the uninsured are not offered health insurance on their jobs.
44
Gruber TABLE 3
Insurance Offer Rates by Firm Size and Average Earnings Categories Offer Ratea
Average earnings ($) <10,000 10,000-15,000 15,000-20,000 20,000-25,000 25,000-30,000 >30,000+
1-9 employees
10-24 employees
25-49 employees
50-99 employees
>100 employees
0.24 0.32 0.43 0.50 0.55 0.61
0.45 0.55 0.70 0.77 0.83 0.88
0.63 0.76 0.83 0.86 0.92 0.94
0.81 0.88 0.93 0.95 0.91 0.95
0.95 0.93 0.98 0.97 0.98 0.98
From author's tabulations of 1997 Robert Wood Johnson Foundation survey of employers.
2.5 The Uninsured Are Quite Mixed with the Insured Finally, any solution to address the problem of the uninsured, tax-based or not, must recognize a fundamental conundrum: the uninsured are not
an isolated and easily identified sub-population. This is illustrated in Table 4, also from EBRI (2000), which shows the income distribution of the uninsured. The second column shows the number of uninsured in each income category listed in the first column; the third column shows the percentage of the uninsured in each income category; and the final colunm shows the percentage of that category that is uninsured. The last column is the least surprising: the share of any income group that are uninsured declines with increasing income. What is more striking is the second column: there are many uninsured who are not poor or even near poor. Indeed, almost a quarter of the uninsured live in families with incomes over $50,000 per year. This table highlights the difficulty of targeting programs to cover the majority of the uninsured. To cover the majority, you must go fairly high up in the income distribution. But, as you do so, you enter ranges where only a small fraction of the group is uninsured.
3. HOW DOES TAX POLICY AFFECT INSURANCE COVERAGE? To fully understand how alternative tax policies might affect insurance coverage requires knowledge of a wide variety of key behavioral parameters. In this section, I review the parameters that must be measured to fully assess the range of tax policy effects.
Taxes and Health Insurance 45 TABLE 4
Income Distribution of the Uninsureda Income category ($) <5000 5000-9999 10,000-14,999 15,000-19,999 20,000-29,999 30,000-39,999 40,000-49,999 >50,000 Total
Number of uninsured (in millions)
Percentage of uninsured in income category
Percentage of income category that is uninsured
4.8 3.3 4.5 4.5 6.8 5.3 3.3 9.7 42.1
11.4 7.8 10.6 10.7
44.3 31.0 34.6 32.0 24.6 19.4 13.9 8.5 17.5
16.1 12.5
7.9 22.9 100
a From EBRI (2000).
Figure 1 summarizes the channels through which tax policies might affect coverage, and the resulting behavioral responses that are key to assessing tax-policy effects. The leftmost part of this chart shows the three possible types of tax subsidies to insurance: subsidies to employers to offer coverage (either reforming the existing subsidy, or offering new subsidies); subsidies to employees to take up coverage (once again, ei-
ther reforming the existing section 125 option, or offering new subsidies); and subsidies to individuals for non-group insurance coverage. Each of these types of tax subsidies has effects on both firms and on individuals, as shown in the middle of the figure. Moreover, the decisions of both firms and individuals feed back to each other. Finally, for sizable tax interventions, there may be effects on the insurance market itself which wifi affect the decisions made by firms and workers. In terms of firm decisionmaking, the key element of response is the decision to offer insurance in response to an increased subsidy to employer coverage, or to drop insurance in response to a reduction in the net subsidy to employer coverage. For increases or reductions in the existing tax subsidy to employer-provided insurance, these responses are likely to be symmetric. But a key unresolved question is the extent to which changes in other subsidies would impact employer decisions. For example, would a 10% subsidy to employees for their expenditures on employer-provided insurance have the same impact on insurance offering as an additional 10% subsidy to employers for their insurance spending? In terms of simple economic theory, the answer is clearly yes. But there may be differences in practice that make these responses asymmetric.
46
Gruber
FIGURE 1. Channels for Effects of Tax Policies on Coverage Tax Subsidy to Employers
Offer/Drop Coverage
Firms
Raise/Lower Contributions Raise/Lower Plan Generosity
Tax Subsidy to Employees
Takeup/Drop Employer Coveragi
Takeup/Drop Nongroup Coverag Takeup/Drop Public Coverage Tax Subsidy to NonGroup
Takeup/Drop Spousal Coverage
Markets
For example, subsidies to employers may be targeted by firm size or some other firm characteristic, but are available to the firm as a whole. But subsidies to employees that are targeted will likely leave some employees in a firm eligible and others not. While the subsidy amounts may be the same on average (e.g., a 50-percent subsidy to the firm or a 100percent subsidy to 50 percent of the workers), in practice the impacts may be quite different, depending on how employee preferences are aggregated in the determination of benefits. Similarly, in terms of economic theory, subsidies to non-group and to group insurance should have similar and opposite effects. But, once again, in practice the reactions of firms may not be symmetric, since workers may not view the (currently) inhospitable non-group market as an effective substitute for the group market. Firms also have other margins of response besides the decision to offer (or not to offer) insurance. One important margin is the decision on how much to contribute to insurance costs. Currently, firms pay about 75 percent of the costs of insurance, but this has fallen dramatically over the past 15 years (Gruber and McKnight, 2001). If higher employee premiums lead employees not to take up their employer coverage, then this shift in costs from employee to employer could be significant. Similarly, employers may react to tax subsidies by reducing the generosity of their insurance plans along a variety of dimensions, such as shifting from fee-
Taxes and Health Insurance
47
for-service to managed care plans, raising patient copayments, or restricting benefit coverage. Indeed, it is this subsidy on the margin to insurance costs that led Martin Feldstein to criticize the tax subsidy to employer-
provided insurance in the early 1970s. He claimed that this subsidy would lead to overly generous insurance coverage, which through moral hazard would then raise spending on medical care, increasing further the demand for insurance coverage, and thereby leading to a spiral of rising medical costs.
Individuals can also respond directly to tax subsidies either to employee purchases of insurance or to non-group subsidies. There are four dimensions along which individuals can respond. The first is to change their take-up of employer-provided insurance, conditional on its being offered. The second is to move into or out of non-group insurance coverage. The third is to move into or out of public insurance coverage. This channel may seem more controversial, but the majority of those made eligible for public insurance over the past 15 years have been eligible as well for employer-provided insurance, so this is a margin of potentially active substitution. Finally, married couples can shift insurance coverage from one spouse's job to the other's as the relative subsidy to one spouse or the other changes. In addition, the decisions of employers and workers can have feedback effects on each other. As employers change their offering of insurance, this will affect employee take-up. Likewise, changes in employerprovided insurance generosity can affect take-up and decisions to move across spouses or into public or non-group coverage. Moving the other direction, when economists model firms' benefit decisions, they do not
think of a firm as a distinct entity, but rather as an aggregation of its workers. So individuals' responses to tax interventions can also feed back to firms' decisionmaking. An important research question, alluded to above, is how worker preferences are aggregated; once again, this aggregation may differ across types of tax subsidies and along the margin of employer response (e.g. offering vs. employee contributions). I discuss the scant evidence on this question below. Finally, tax policy can affect the insurance market directly. So long as the market supply for insurance is upward sloping, any major intervention that increases demand wifi lead to a partially offsetting pre-tax rise in insurance prices. On the other hand, many have argued that the high
and unstable prices in the non-group market reflect the "thinness" of this market, and that a major subsidy to non-group policies which led to more purchase could reduce inefficiency and lower prices in that market. These changes in pretax prices will obviously mitigate or exacerbate any direct response by firms and individuals to tax incentives.
48
Gruber
Thus, there are an enormous number of margins that must potentially be modeled to assess the effect of a full range of tax subsidy options on insurance coverage. In principle, there are many direct channels from tax policy to individual and firm behavior, as well as a large number of potential feedback effects from firms to individuals, from individuals to firms, and from the market to both. Given the impossibility of estimating all of these responses, a key question is the extent to which response symmetry can be called upon to apply behavior from one type of subsidy to another.
4. WHAT DO WE KNOW ABOUT KEY BEHAVIORAL PARAMETERS? There is a large literature devoted to estimating some of the behavioral parameters that were discussed in section 3. In this section, we review that literature. We begin by reviewing what is known about firm behavior, then turn to individual behavior.
4.1 Firm Offering Decisions The behavioral response on which there has been the most work is the elasticity of insurance offering by firms. There have been several approaches to estimating this elasticity. The first approach discussed here is
to use variation in the premiums faced by firms to identify the price sensitivity of their offering decision. Two examples of this work are Feldman, Dowd, Leitz, and Blewett (1997), who use information from 1993 for a sample of small firms in Minnesota to estimate price elasticities of 3.9 (single coverage) to 5.8 (family coverage); and Marquis and Long (1999),
who use data from 1993 for 10 states to estimate a much smaller price elasticity of only 0.14. A key problem with this approach, however, is that one only observes premiums for the firms that do offer insurance, and they must be imputed to firms that do not. Thus, instruments must be found that are correlated with the price of insurance but not firm demand, and previous articles have not used firm characteristics that are likely to meet this criterion (e.g. whether the firm is unionized). The third approach is to use variation in taxation to identify the price elasticity of offering, in essence asking whether those firms with higher tax-related subsidies to insurance purchase are more likely to offer insur-
ance. Leibowitz and Chernew (1992) use variation in tax rates across states to examine the effect of after-tax prices on insurance offering by small firms, as well as using variation in premium quotes across locations obtained from small-group insurers. They separately estimate the
Taxes and Health Insurance
49
response to premiums and subsidies, and obtain an elasticity between 0.8 (premiums) and 2.9 (subsidies). Royalty (1999) also uses crossstate variation in marginal tax rates to estimate an elasticity of firms' insurance offering across all employers at 0.63. Gentry and Peress (1994) study cross-city differences in the average share of workers offered health insurance benefits, as a function of cross-state differences in after-tax prices of insurance. They find that for each percentage-point increase in the price of health insurance, the fraction of blue-collar workers covered by employer-provided insurance declines by 1.8 percentage
point, implying an elasticity of 1.4; however, there is no statistically significant effect for white collar workers. These types of studies have the advantage that differences across cities and states in tax rates should be independent of insurance-offering deci-
sions. But they may not be entirely independent: cities and states with substantial taste for insurance may be the ones that offer the largest tax breaks, which would lead to a strong relationship between price and offering. This criticism is addressed in recent work by Finkelstein (1999), who studies the removal of the large (25%) tax subsidy to supplemental private health insurance in Quebec in 1993, and finds an elasticity of 0.42 to 0.54 for employer offering. But it is somewhat unclear how to apply the elasticity of offering of supplemental insurance for a national health insurance scheme to the decision of U.S. firms to offer full private health insurance plans. A third approach comes from running small-scale subsidy pilot programs for small businesses and evaluating the response of firms to subsi-
dized prices. These pilot programs have the advantage of essentially providing a randomized intervention. Two such pilot programs are evaluated in Helms, Gauthier, and Campion (1992) and in Thorpe et al. (1992). The former study finds a wide variety of price responsiveness across sites, with sites such as Utah offering 40% discounts and seeing only 4% enrollment among uninsured firms (an elasticity of only 0.1) and other sites such as Arizona offering 10% discounts and seeing 411% enrollment (an elasticity of 0.4 to 1.1). The latter finds very weak
response to a program that provided a 50% subsidy to the price of insurance for small firms in New York, with an elasticity of only 0.07 to 0.33. But it is unclear whether the small elasticities estimated here are because of the temporary experimental nature of these subsidies; firms may be reluctant to set up insurance plans based on subsidies that wifi only last for a short time. There could be much larger responses to more permanent changes in the after-tax price of insurance. A final approach is to use responses of firms to hypothetical questions about changes in the price of insurance. Morrisey, Jensen, and Merlock
50
Gruber
(1994) use the response to such hypothetical questions to estimate a price elasticity of insurance offering among small firms of 0.92. But it is unclear whether firms respond in the same way to hypothetical questions as they do when faced with an actual insurance purchase decisions. A final issue with this literature is that, with firm-level data, one does
not observe the characteristics of the employees to which the firm is responding in making its benefit decisions: who is the marginal worker whose preferences determine the firm's decisions? One article which attempts to address both the identification and marginal-worker issues is Gruber and Poterba (1994). They study how the self-employed responded to the Tax Reform Act of 1986, which introduced a subsidy to the insurance purchases of the self-employed. This natural experiment provides exogenous variation in the after-tax price of insurance. More-
over, for the self-employed, there is no issue of deciding who is the marginal worker. They find significant increases in the insurance coverage of the self-employed relative to the employed over this period, with an implied price elasticity as large as 1.8. Unfortunately, however, it is unclear how generalizable these results are to firms, which must aggregate the preferences of all their workers in making benefit decisions. A more recent approach to surmounting these problems is Gruber and
Lettau (2000). In that paper, the authors use data from the Employee Compensation Index (ECI) dataset, which collects information both on firm insurance provisions and on the characteristics of a sample of workers in the firm. The latter feature allows them to directly measure the
distribution of tax rates within the firm, and to assess which tax rate seems to be most central in driving benefit provision decisions. They also introduce a new identification strategy, extending the previous taxbased work to rely on state tax progressivity and changes in state taxes over time, while controlling for mean differences across states that are likely to be correlated with tastes for insurance. They estimate an elastic-
ity of insurance offering of 0.3 to 0.4, towards the lower end of the previous literature. Gruber (2001) recently applied the identification strat-
egy of Gruber and Lettau's paper to data from the Current Population Survey, which gathers data on a random sample of workers but not on the distribution of workers in a firm. He finds a higher elasticity of offering of 0.7 in these data. But the ECI estimates seem more reliable in view of the higher quality of the data.
4.2. Employer Insurance Spending There is also a sizable literature on the effect of after-tax prices on employer insurance spending. Estimates of this elasticity come from three types of studies. The first is time-series evidence on how total spending
Taxes and Health Insurance 51
on employer-provided health insurance responds to changes in federal tax rates, presented in Long and Scott (1982), Vroman and Anderson (1984), and Turner (1987). These studies typically yield estimates of the price elasticity of demand between 0 and 0.5. But the results are hard to interpret, as there are many things changing in the time-series data; for example, the fact that health insurance coverage fell in the 1980s may be the result of declining marginal tax rates, but it may also be the result
of a shift in the job base towards service jobs that are less likely to provide insurance. A second set of studies, including Taylor and Wilensky (1983), Holmer (1984), and Sloan and Adamache (1986), analyze cross-sectional data on individuals or firms and ask whether those with higher tax-related subsidies to insurance purchase spend more on insurance coverage. But a potential problem with these studies is that differences across individu-
als in their tax rates arise in part from differences in the underlying behavior of individuals or firms, such as differences in labor supply, family structure, or the nature of the work force. It is impossible to tell whether differences in observed insurance coverage are due to taxes or these behavioral differences. A wide range of estimates emerge from these studies; Pauly (1986) summarizes the consensus range as 0.2 to
more than 1.0.
The final approach that attempts to overcome the problems inherent in the previous cross-sectional literature examines how demand for insurance responds to plausibly independent legislated tax differences. Woodbury and Hammermesh (1992) analyze all fringe-benefit expenditures around the Tax Reform Act of 1986 in a panel data set of colleges and universities. They conclude that tax reform substantially reduced the demand for fringes, with an estimated elasticity in excess of 2. But this is not focused on health insurance spending per se, so it is difficult to disentangle the impact on health insurance. Gruber and Lettau (2000), in the study described above, also examine the impact of tax subsidies on employer-provided insurance spending, using similar variation across states in their tax systems to Woodbury
and Hamermesh. They also find a quite large elasticity of insurance spending: 0.94.
4.3 Employer Contributions to Health Insurance A particularly important margin of response to tax subsidies is how employers change their contributions to health insurance for employees.
This response would be subsumed in the spending elasticities cited above, but it is important to break it out distinctly due to the potential impact of changing contributions on employee take-up.
52
Gruber
There is only one study of which I am aware on this response. Gruber and McKnight (2001) use the Current Population Survey to study the impact of tax changes on the decisions of employers to pay all of the cost of their health insurance plans. The share of employers paying all those costs fell from 44 percent in 1982 to 28 percent by 1998. They model this variable as a function of a variety of actors, including the tax subsidy to employer-sponsored insurance. They find that the tax subsidy did have a very important impact on employer contribution policy. Their central findings suggest that for each 10-percentage-point reduction in the tax subsidy, the share of employers paying all of the costs of health insurance falls by 1.7 to 3.8 percent.
4.4 Employee Take-up of Employer-Provided Insurance A central parameter for evaluating the effects of tax subsidy policy is the
elasticity of employee take-up of employer-provided insurance if it is offered. As highlighted earlier, take-up among those offered is fairly high for all firm sizes and average firm wage levels. But does take-up respond to the prices charged to employees for insurance? The answer, to date, appears to be no. Two studies have examined how employee take-up responds to the price charged for insurance (Chernew, Frick, and McLaughlin, 1997; Blumberg, Nichols and Banthin 2001). Both papers find that firms that charge more for insurance have no lower take-up of their insurance policies. These papers do run into the possible problem that firm premium decisions are endogenous to employee tastes. The direction of bias here is unclear, and depends on whether firms set low employee premiums when there are tastes for insurance, or when there are not (because of paternalism or to satisfy insurance-company conditions for high employee take-up). Gruber (2001) also investigates this question by assessing whether the existing employer tax subsidy affects coverage, conditional on offering, as it should for those with a section 125 plan (roughly half of employees by the most recent estimates). But I find no such effect. The lack of elasticity of employee take-up is very striking in view of the time-series trends discussed earlier. Over the mid-1980s through the late 1990s, the trend was towards rising employer contributions and falling employee take-up. This is shown in Figure 2, from Gruber and McKnight (2001), which shows the share of workers who have employer-sponsored insurance over time, and the share of workers whose employers pay all of the costs of insurance over time. As those authors note, there is a striking correspondence between these series, and a price elasticity of employee takeup of 0.4 would explain the time-series trend in take-up. But this appears to be well above the best estimates to date of the take-up elastic-
Taxes and Health Insurance
53
FIGURE 2. Group Health Insurance vs. Employer Pays All 0 % with Group I-Il
% whose Employers Pay All I
I
.803
.44
I 0
0
a
.73 -
.28 I
1979
1983
I
1987 year
1991
1995
1998
ity. Thus, the cause of this trend towards declining take-up remains a mystery. On the other hand, these findings are consistent with a growing body of evidence which suggests that it is the decisions that employers make for their workers that are most important in determining worker benefit provision, not active decisions taken by those employees. The most striking example is Madrian and Shea (2001), who find that when a firm moved the default investment option and contribution level for its 401(k) plan, the vast majority of workers did not move from that default, despite it being a worse choice for many of them.
4.5 Substitution between Forms of Insurance Coverage As the lower part of Figure 1 illustrates, a key issue for modeling the
effect of tax policy on individual insurance coverage is the substitutability across different forms of insurance coverage. The margin of substitutability for which there is the best evidence is substitutions between private and public coverage. There is a large literature on crowdout over the past 5 years which examines the question of whether those made eligible for public insurance will drop their private insurance to take it up. The earliest estimates of crowdout suggested it was quite large, with one person losing private insurance for every two gaining public insurance in the late 1980s and early 1990s (Cutler and Gruber, 1996). But
54
Gruber
subsequent estimates suggest that the effect may be more modest, on the order of 10-20 percent; see Dubay (1999) for a review. There is less evidence on the other key margins of substitutability,
between own employer coverage, non-group coverage, and spousal coverage. Gruber (2001) did recently extend the analysis of Gruber and Lettau (2000) to consider the impact of changing the employer-provided insurance tax subsidy on all insurance margins, not just on the margin of insurance offering. We find that a larger subsidy to employerprovided insurance causes a reduction in public insurance, consistent with the crowdout literature. But we find little substitutability with nongroup coverage.
To date, there is little evidence on the substitutability of own and spousal insurance coverage, On net, I find that the effect on offering is almost directly translated to a bottom-line effect on being uninsured, suggesting that employer offering is the key margin of response to existing subsidies.
4.6 Substitution across Spouses There is enormous scope for substitution of health insurance coverage across spouses. For example, of the 28 million male married workers aged 21-64 who are insured on the job, 15 million have wives that are offered health insurance. More than 8 million report that their wives actually take up the coverage on their jobs. Of the 16 million married female workers aged 21-64 who are insured on the job, 12 million have husbands who are offered insurance, and almost 9 million of those husbands take up that offer. With such a large number of spouses who are jointly offered insurance, it would seem that substitution across spouses as policy changes would be a real possibility. There is only one article that investigates this issue (Monheit, Schone, and Tayor, 1999), using 1987 data, and their results suggest that the decision to take double coverage (health insurance coverage through both the husband and wife), conditional on both spouses being offered insurance, is very sensitive to incentives. For example, they find that the odds of having double coverage rise by more than a third if both plans are free to the husband and wife. They also find that double coverage take-up is higher among those families in ifi health, and that it is more common when it results in a more comprehensive set of benefit coverage than does take-up by one spouse alone. This set of findings suggest that substitution across spouses who are offered insurance may be quite fluid as tax policy changes. How much
substitution there would be depends on how universal the policy changes are, and how similar the jobs are that wives and husbands hold.
Taxes and Health Insurance
55
For an income-targeted subsidy, for example, it is possible that a husband who works with many low-income workers wifi lose his offer of health insurance, while his wife who works with many high-income workers wifi not.
4.7 The Big Mystery: Take-up among Those Not Offered At least three-quarters of the uninsured are not offered health insurance. Thus, in focusing on the effects of tax subsidies on non-group insurance, the key question is how price-sensitive this group will be in their take-up decisions. Unfortunately, we have essentially no evidence on this critical question. The one relevant paper is by Marquis and Long (1994), who estimated the demand for non-group insurance coverage among workers not offered employer-sponsored coverage as a function of the area-
specific price of non-group coverage. They estimate an elasticity of non-group coverage of 0.3. The problem with this approach is that the price of insurance reflects not only true price differences in insurance (differences in the load factor, or the premium cost relative to expected claim expenses), but also differences in medical costs and differences in underlying tastes for insurance. Both of these latter two factors will bias
downward any estimate of the effect of area insurance prices on demand. Thus, we are left with only a lower bound on the elasticity of insurance take-up among those not eligible for employer coverage.
4.8 How Are Employee Preferences Aggregated? As noted above, appropriate modeling of the implications of subsidies to employer-provided insurance requires an understanding of the mechanism for aggregating employee preferences in the firm's benefits decisions. The best discussion of this issue in the context of benefit provision
is in Goldstein and Pauly (1976). They conclude that the equilibrium benefit determination could arise in one of two ways. One is through the collective choice of the existing set of workers, through an insider-
outsider or union mechanism. In this case, through standard voting arguments, the benefits chosen wifi reflect the tastes of the median worker. The second is through the choices of employers, whose goal is to minimize their total labor costs, and wifi therefore design their benefits packages to reflect the average preferences of their workers. If there is a perfect Tiebout equilibrium across firms, with workers sorted completely by their tastes for insurance, then the average and median tax prices wifi be everywhere the same and the distinction between these models wifi not be important. However, as Gruber and Lettau (2000) discuss, there remains considerable dispersion between these measures within firms; almost 10 percent of firms have a median
56
Gruber
and a mean tax price which differ by 5 percentage points or more. If there is imperfect sorting, then these models can have very different implications, depending on the difference between the median and the mean. Further complications arise when considering the fact that both mobility and influence within the firm differ across workers. Firms may not consider the average of all workers' preferences in making benefits, but may weight more highly the preferences of either mobile workers or the "most influential" workers in the workplace. Assessing the appropriate model of within-workplace benefit determination is not purely an academic concern. Modeling the implications of tax reform may depend critically on appropriately capturing the structure of how tax prices throughout the firm affect benefit decisions. If, for
example, the median tax price is the only one that matters, then tax reforms such as that in 1993 which raised tax rates only at the top of the distribution wifi have essentially no impact on insurance decisions. But if other movements of the distribution matter, then such reforms may have larger impacts. Gruber and Lettau investigate this issue by drawing on the strength of their ECI data to include several moments of the distribution of tax prices in their insurance demand model. They find that the median tax price explains benefit determination significantly better than the average. But they also find that there is an important additional role for the highestpaid worker in the firm (the worker with the lowest tax price). So the
appropriate voting model in their context appears to be one with a decisive median voter but some extra influence for the highly paid worker. Unfortunately, however, it is not clear if these results, estimated in the context of an unlimited tax deduction, extend to other tax-subsidy structures, such as targeted tax credits. Consider a very generous non-group
credit that pays the full costs of insurance for 49 percent of a firm's
workers, but is zero for those at the median and above. It seems unlikely that the firm would not respond at all to such an outside option. In these types of cases, it may be the mean incentive across all workers which better captures how preferences are aggregated.
4.9 Market Responses Another mystery area is how markets wifi respond to tax interventions. As noted above, tax subsidies could lead to rises or falls in group or nongroup market insurance prices. A particularly critical question is whether, for non-group subsidies of a given value (such as the Bush plan), nongroup plans will emerge that are targeted to that dollar value. Advocates
Taxes and Health Insurance
57
of non-group subsidies point to the availability of very cheap insurance plans over the Internet. But, even if such plans are offered, there is the question ofwhether they wifi be demanded. After all, even in the face of existing low-cost policies, there remain over 42 million uninsured persons. An outstanding mystery in insurance markets is why there is not more demand for low-cost, catastrophic policies, given that they insure the substantial risks we should be most worried about but can reduce costly moral hazard. But given this lack of demand, even the offering of such low-cost policies tied to subsidy amounts may not significantly increase take-up.2
5. POLICY IMPLICATIONS The discussion of section 3 and the evidence presented in section 4 have several clear implications for tax policy towards insurance. Before presenting these implications, it is important to consider the goals of subsidizing insurance through the tax code. Presumably, the most important goal is to raise insurance coverage in the U.S., and in the most costefficient manner. Thus, the discussion below wifi focus on the efficiency of various alternative policies, measured as their cost per person newly insured. But a secondary goal of tax policy may be horizontal equity, or redistribution to those already paying for their insurance without the tax subsidies available to others. This is not a goal that will receive attention in the discussion below, but it is important to recognize that if it is the goal of policy, it may be well served by some of the alternatives I dismiss as "inefficient" below.
5.1 Reforming the Existing Tax Subsidy I first consider the implications of reform of the existing tax subsidy to employer-provided health insurance. One alternative here would be to end this subsidy, perhaps redistributing the dollars to other forms of insurance subsidization. Gruber and Lettau (2000) perform some simulations using their estimates to assess the implications for employer offering and insurance spending. The results are summarized in Table 5, which shows the impact of a several alternative reforms on the rate of insurance offering, the level of insurance spending conditional on offering, and the overall level of spending. The ranges of estimates reflect the 2
Moreover, the low prices of catastrophic policies in today's market may reflect positive selection: such policies wifi only be demanded by the healthy, so that prices can stay low. When subsidies are available that make such policies cheaper, then there may be more demand for them by the sick, which would lead to price rises.
58
Gruber
TABLE 5
Implications of Reform of Existing Tax Subsidy
Reform
Remove all tax subsidies Remove income tax subsidy 10% tax rate cut
Change in
Change in spending among
offeringa
Reduction in employerinsuredb
(%)
(million $)
(%)
14.1
22.3
35.4
50.3%
8.6 0.9
13.6
23.7
32.8%
1.4
2.8
3.7% [$95]
offereda
Total
change in spendinga
[$1231]
[$815]
'From Gruber and Lettau (2000). b Result of applying reductions in preceding column to national total for employer-insured.
alternative models estimated in this paper; to be conservative, in this discussion I focus on the lower-bound estimates.3 This paper finds that there would be very significant implications of reducing the existing tax subsidy. Completely removing the existing subsidy, with respect to all (federal and state) income taxes as well as with respect to the payroll tax, would lower the rate of employer insurance offering by 14%. Assuming (following the results in Gruber, 2001) that all of those dropped by their employer would become uninsured, this implies that 22 million Americans would become uninsured. This would represent a roughly 50% rise in the number of uninsured. Moreover, this paper also finds a very large reduction in spending on those who have insurance, with the net result being a 50% reduction in the dollars spend on employer-provided insurance.4
These are enormous impacts, and the potential for 22 million more uninsured Americans should lead anyone to pause before removing the existing employer subsidy. But it is also important to remember that the existing subsidy costs over $100 billion per year in forgone revenue. So this says that we are spending about $5000 per person to insure these 22 million persons, a quite high level. So the question becomes whether It is of course important to recognize that these projections are only as precise as the underlying estimates; the central estimates on which they are based are a probit coefficient on insurance offering of 5.424 (2.017), and an OLS coefficient on log spending of 1.465 (0.404).
The calculations in the final column are conjectural in that they assume that the firms that stop offering insurance are spending the average amount on insurance before dropping. It seems likely that those firms were spending less than average, so that there is a smaller reduction in overall spending than is implied in Table 5.
Taxes and Health Insurance 59
other alternatives are available at a more reasonable cost to insure large numbers of Americans. Table 5 also shows the impacts of less dramatic reforms to our existing system. The first is to remove the existing subsidy in the income tax, but in for the payroll tax. This would still lead to very large reductions in insurance offering and spending, with the potential for 13.6 million more uninsured. The final column illustrates that even very modest reforms to the tax code can have large effects on employer-provided insurance. This column shows the impact of cutting all tax rates by 10%, roughly akin to the original proposal of the Bush Administration. Even this very small change could lead to 1.4 million more uninsured, and a reduction in total spending on employer-provided insurance of almost 4%. One potential reform which would may less dramatic implications for coverage would be not to reduce the existing employer subsidy but to cap it, for example at the mean or median cost of a group insurance plan. As discussed in more length in Gruber and Poterba (1996), such a reform might temper the inflationary aspects of the tax subsidy highlighted by
Feldstein, and significantly raise government revenues, but without causing a major displacement of the employer-insured. In principle, administering such a cap would be straightforward: employers would simply be asked to report, for tax purposes (either income tax alone or payroll tax as well), any spending they make on an employee's behalf for insurance beyond some cap level. But, in practice, caps run into important administrative and political difficulties. First, there are very large regional disparities in the cost of health insurance, which would ideally be reflected in the cap level. But there has never been a regionally adjusted tax credit, and efforts in other arenas (e.g. poverty measurement) to have regional adjustment have run into daunting political difficulties. Second, a cap would penalize workers for having high-cost co-workers, since it would reflect average and not individual-specific insurance spending. This would lead to general redistributions from older, higher-cost industries to newer, lower-cost ones, raising further daunting difficulties.
5.2 New Subsidies to Employers A more likely direction for reform is to offer new subsidies to employers. But doing so immediately runs into the type of efficiency considerations highlighted above. The majority of employers already offer health insur-
ance. For this group, new subsidies are just redistribution, with no impact on insurance coverage, except through feedback effects on employees through reduced employee contributions or more generous insurance levels.
60
Gruber
Thus, to be cost-effective, new subsidies to employers must be well targeted. Table 2 illustrates how such targeting would be most effective:
if subsidies were largest for the smallest and lowest-wage firms. But there is a trade-off with targeting: the more targeted are subsidies, the more distortionary "cliffs" there may be to firms' decisions on pay and hiring. This was in fact an important criticism of the proposed subsidies to small firms embedded in the Clinton HSA plan. But such "cliffs" need not in fact arise, so long as there are smooth phaseouts with respect to wages and firm size. Consider for example the following subsidy structure:
There is a maximum subsidy rate of 0.4 for firms of 10 employees or fewer with average annual earnings of $10,000 or less. The subsidy is reduced by 0.01 for each extra employee above 10, so that it reaches zero at 50 employees. The subsidy is also reduced by the amount that earnings rise above $10,000 per year. This reduction factor is proportional to firm size that is, the bigger the firm, the faster the subsidy is reduced as wages rise. The formula is: Firms of 10 employees or fewer: subsidy reduced by 0.0222 for each $1000 rise in earnings above $10,000 per year.
Firms of more than 10 employees: subsidy reduced by (firm size)/ 450 for each $1000 rise in earnings in hourly wages above $10,000 per year. For example, for a firm of 25, the subsidy is reduced by 0.0555 for each $1000 rise in earnings above $10,000 per year.
I have simulated the effect of such a policy in the RWJF data used to compute Tables 2 and 3, drawing on the evidence presented above. The results suggest that for an annual cost of $3.8 billion, 2.2 million persons could be insured per year, for a cost of $1720 per newly insured. And there are very low implicit tax rates embedded in this gradual phaseout structure. On average, for every additional $100 paid out in wages, firms only lose $1.2 in subsidy; even at the maximum, the subsidy loss is only $15 per $100 paid out in wages, a modest distortion by the standards of our tax system. Similarly, for each worker hired, the subsidy loss is on average only $82; the maximum possible subsidy loss per hire is $720. So distortions need not be enormous to offer very targeted subsidies. But one interesting feature is that the efficiency of such firm subsidies diminishes as they get larger. For example, an expanded version of the
above subsidy plan that delivers a maximum subsidy of 50 percent
Taxes and Health Insurance 61
would cost 175 percent as much ($6.7 billion/year), but only cover 50 percent more persons (3.3 million), so that the cost per newly covered rises to $2070. This is a common finding in all simulation work on tax subsidies: their efficiency is inversely related to their scope. This is because as the subsidies get larger, they necessarily become less targeted and more attractive to those who are already providing insurance. Nevertheless, the efficiency of these firm subsidies seems quite high relative to the other policies considered below.
5.3 New Subsidies to Employees The second major alternative discussed earlier was new subsidies to employees to take up employer coverage. The first point to note about such subsidies is that, as a device for targeting take-up per se, they are likely to be very inefficient. This is because fewer than 10% of those offered insurance are actually uninsured. Moreover, the work reviewed above suggests that the take-up decision is not very elastic with respect to price. These two facts are a recipe for an inframarginal subsidy that wifi serve only as redistribution and not to increase insurance coverage.
But, in a general economic model, such subsidies should also increase employer insurance offering. Indeed, there is no economic rationale for not treating them symmetrically with a subsidy to employers:
both are subsidies to offering insurance through the workplace. In practice, however, their effects might differ somewhat. Employer subsidies are targeted to the characteristics of a firm, while employee subsidies are targeted to employees. This has the advantage that it may be possible to do better income targeting with employee subsidies, since even low-wage firms have high-wage workers. But it has the disadvan-
tage that it may be harder to target low-wage firms by simply giving subsidies to low-wage employees. Many low-wage employees work in high-wage firms that already offer insurance. So even a very tightly targeted subsidy to employees is likely to result in little new offering, as most of the dollars flow to those already in firms offering insurance.
5.4 Non-group Subsidies The final alternative is to subsidize non-group purchase of insurance. The prototypical proposal here, which is quite similar to that proposed by President Bush, would be: $1000 credit for individuals; $2000 for families
Usable for non-group insurance purchase only Refundable and advanceable Phased out for upper-middle- and upper-income families
62
Gruber
There are several difficulties faced by non-group credits. The first is a simple fact: roughly half of the uninsured do not pay taxes (Gruber and Levitt, 2000). This makes the tax code a problematic mechanism for
delivering subsidies to the uninsured. In principle, this can be addressed by making credits refundable. In practice, this raises two difficulties. First, refundability is a very difficult political goal in recent years, as many conservatives view refundable credits as akin to cash welfare. There were enormous battles over the refundability of the child credit in 1997 and again in 2001, with the result being only very partial refundability of this credit. Similar battles are likely to occur with these tax credits. Second, and more fundamentally, the usefulness of refundability is quite limited without effective advanceability. Insurance premiums are
due from the beginning of year t, but tax refunds do not arrive until spring of year t + 1. Thus, the uninsured must have sufficient resources to advance-fund their insurance purchases if credits are to be effective;
but most uninsured, indeed, most Americans, do not have sufficient liquid assets to do such advance funding. In principle, this could be surrounded by legislating an advanceable credit. But our one experience
with such a feature is not encouraging. Individuals can claim their Earned Income Tax Credit (EITC) throughout the year rather than the next spring, and it would be sensible for most claimants to do so; but fewer than 1% of claimants take advantage of this option (Liebman, 1998). The reasons for this low take-up are unclear, but the main conclusion is that low-income taxpayers appear to be reluctant to take any risk that they wifi end up facing a tax liability, rather than receiving a refund, on April 15. This will limit the effectiveness of advanceability for a nongroup credit as well. Finally, the major difficulty faced by such a credit is that non-group insurance is very expensive. Today, the typical non-group policy for a family costs $6,000 to $7,000 per year. Thus, even a sizable $2,000 credit leaves the family with $4,000 to $5,000 in costs to pay, which is enormous relative to the incomes of the working poor uninsured. This point ties to the earlier issue about market responses. If the insurance market can respond to the availability of this credit by delivering low-cost insurance products that are demanded by the public, then these affordability barriers may be overcome.
As a result of these limitations, my previous work suggests only modest impacts of nongroup credits on net insurance coverage (Gruber and Levitt, 2000; Gruber, 2000). This work is based on a major micro-
simulation model that takes as its base the 1997 Current Population
Taxes and Health Insurance
63
Survey, and imposes on that base a complicated set of behavioral equations that make assumptions on all of the relevant margins in Figure 1 (as detailed in the appendix to Gruber, 2000). The base case is a $1000 or $2000 refundable credit which is not effectively advanceable, due to the limitations noted above. I assume that this credit is available to singles with incomes up to $75,000 and families with incomes up to $100,000. It is worth noting that subsidies that are more tightly income-targeted are more efficient, as for incomes above the median there are relatively few
uninsured. But, once again, there is an important political question about whether it is feasible to have a truly targeted new credit in today's environment; all of the middle-class entitlements of the past 5 years have extended to income ranges similar to those used here (or higher). I find that for this base-case policy there is a cost of $13.3 billion per year, and net reduction in the uninsured of 4 million persons, for cost of $3300 per newly insured, well above the costs cited above for employer subsidies. The effectiveness of the credit rises significantly if it is advanceable, but the cost per newly insured remains above $2500. Thus, non-group credits do not appear nearly as effective as group credits, due to the problems noted above.
6. CONCLUSIONS Tax policy towards health insurance is likely to remain a topic of vigorous debate in the years to come. This paper has laid out some key facts, economic evidence, and policy simulations to help guide this debate. The key conclusion that I draw from existing facts and evidence is that policies targeted to firms are more likely to be effective than are policies targeted to individuals, as firms appear quite price-responsive in their insurance-offering decisions, and the actions of firms appear to be directly translated to individual coverage. But the prevalence of offering means that to be cost-effective such subsidies must be tightly targeted to the firms least likely to offer insurance: small and low-wage firms. The more important conclusion to be drawn from this paper is that we still know remarkably little about a number of key parameters that determine the effectiveness of tax policy towards insurance coverage. Most notable among these is the responsiveness, to new subsidies to buying insurance, of the existing uninsured who are not offered insurance. But there are a variety of other unanswered questions as well that must be addressed by future research if we are to draw fully informed conclusions as to the efficacy of tax policy.
64
Gruber
REFERENCES Blumberg, Linda, Len Nichols, and Jessica Banthin (2001). "Worker Decisions to Purchase Health Insurance," mimeo, Urban Institute. Chernew, Michael, Kevin Frick, and Catherine G. McLaughlin (1997). "The Demand for Health Insurance Coverage by Low-Income Workers: Can Reduced Premiums Achieve Full Coverage?" Health Services Research 32:453-470. Congressional Research Service. Costs and Effects of Extending Health Insurance Coverage. Washington, DC: U.S. Government Printing Office, 1988.
Cooper, Phillip, and Barbara Schone (1997). "More offers, fewer takers for employment-based health insurance: 1987 and 1996," Health Affairs, 16, 142149.
Cutler, David, and Jonathan Gruber (2001). "Health Policy in the Clinton Era: Once Bitten, Twice Shy," forthcoming in Jeffrey Frankel and Peter Drszag, eds., Economic Policy During the 1990s.
Cutler, David, and Jonathan Gruber (1996). "Does Public Insurance Crowd Out Private Insurance?," Quarterly Journal of Economics, 111(2), May 1996, 391-430.
Dubay, Lisa (1999). "Expanding Public Insurance Coverage and Crowd-Out: A Review of the Evidence," in Options for Expanding Health Insurance Coverage: l'Vliat Difference Do Different Approaches Make?, co-edited by Judith Feder and Sheila Burke (Washington DC, Henry J. Kaiser Family Foundation).
Employee Benefits Research Institute (2000). "Sources of Health Insurance and Characteristics of the Uninsured: Analysis of the March 2000 Current Population Survey." Washington, D.C.: EBRI. Farber, Henry, and Helen Levy (1999). "Recent Trends in Employer-Sponsored Health Insurance Coverage: Are Bad Jobs Getting Worse?" Journal of Health Economics. 19, 93-119.
Feldman, Roger, Bryan Dowd, Scott Leitz, and Lynn A. Blewett (1997). "The Effect of Premiums on the Small Firm's Decision to Offer Health Insurance," Journal of Human Resources 32:635-658.
Finkelstein, Amy (1999). "The Effect of Tax Subsidies to Employer-Provided Health Insurance on Workplace Pooling: New Evidence From Canada." Forthcoming in Journal of Public Economics.
Gentry, William and Eric Peress (1994). "Taxes and Fringe Benefits Offered by Employers." NBER Working Paper no. 4764. Goldstein, G. S.; Pauly, M. V. (1976). "Group Health Insurance as a Local Public Good" in The Role of Health Insurance in the Health Services Sector, Richard N. Rosett, ed. New York: Neale Watson Academic Publications, p. 73-114. Gruber, Jonathan (2000). "Medicaid," forthcoming in Means Tested Transfer Programs in the U.S., Robert Moffitt, (ed.) University of Chicago Press, (still forthcoming). Gruber, Jonathan (2001). "The Impact of the Tax System on Health Insurance Decisions." MIT. Mimeo. Gruber, Jonathan, and Michael Lettau (2000). "How Elastic Is the Firm's Demand for Health Insurance?" NBER Working Paper no. 8021. November. Gruber, Jonathan, and Larry Levitt (2000). "Tax Subsidies for Health Insurance: Costs and Benefits." Health Affairs 19:72-85. Gruber, Jonathan, and Brigitte Madrian (1996). "Health Insurance and Early Retirement: Evidence from the Availability of Continuation Coverage," in Ad-
Taxes and Health Insurance
65
vances in the Economics of Aging David Wise, ed. University of Chicago Press, pp. 115-143. Gruber, Jonathan, and Robin McKnight (2001). "Why Are Employee Health Insurance Contributions Rising?" MIT. Mimeo.
Gruber, Jonathan, and James M. Poterba (1994). "Tax Incentives and the Demand for Health Insurance: Evidence from the Self-employed," Quarterly Journal of Economics 109:701-733.
Gruber, Jonathan, and James Poterba (1996). "Tax Subsidies to EmployerProvided Health Insurance." In Empirical Foundations of Household Taxation, Mar-
tin Feldstein and James Poterba (eds.). Chicago: University of Chicago Press. Helms, W. David, Anne K. Gauthier, and Daniel M. Campion (1992). "Mending the Flaws in the Small-Group Market." Health Affairs 11:8-27. Holmer, Martin (1984). "Tax Policy and the Demand for Health Insurance." Journal of Health Economics 3:203-221.
Kaiser Family Foundation (2000). Employer Health Benefits Survey. Menlo Park, CA: Kaiser Family Foundation.
Liebman, Jeffrey B. (1998). "The Impact of the Earned Income Tax Credit on Incentives and Income Distribution," in James Poterba, ed. Tax Policy and The Economy 12, 83-120.
Liebowitz, Arleen, and Michael Chernew (1992). "The Firm's Demand for
Health Insurance." In Health Benefits and the Workforce, U.S. Department of Labor. Washington: U.S. Government Printing Office. Long, James E., and Frank A. Scott (1982). "The Income Tax and Nonwage Compensation." Review of Economics and Statistics 64:211-219.
Madrian, Brigitte C., and Dennis F. Shea (2001). "The Power of Suggestion:
Inertia in 401(k) Participation and Savings Behavior." Quarterly Journal of Economics. Vol. 116, No. 4, pp. 1149-4187. Marquis, M. Susan, and Stephen H. Long (1994). "Worker Demand for Health Insurance in the Non-group Market." RAND Corp. Mimeo. Marquis, M. Susan, and Stephen H. Long (1999). "To Offer or Not to Offer: The Role of Price in Employer Demand for Insurance." RAND Corp. Mimeo. Marquis, M. Susan, and Charles E. Phelps (1987). "Price Elasticity and Adverse
Selection in the Demand for Supplementary Health Insurance." Economic
Inquiry 25:299-313. Monheit, Alan, Barbara Steinberg Schone, and Amy Taylor (1999). "Health Insur-
ance Choices in Two-Worker Households: Determinants of Double Cover-
age." Inquiry 26:12-29.
Morissey, Michael A., Gail A. Jensen, and Robert J. Morlock (1994). "Small Employers and the Health Insurance Market." Health Affairs, 13, 149-161. Pauly, Mark (1986). "Taxation, Health Insurance and Market Failure in the Medical Economy," Journal of Economic Literature, 24, 629-675.
Royalty, Anne Beason (1999). "Tax Preferences for Fringe Benefits and Workers' Eligibility for Employer Health Insurance." Journal of Public Economics. Vol. 75, issue 2, February 2000, pp. 209-227. Sheils, John, and Paul Hogan (1999). "Cost of Tax-Exempt Health Benefits in 1998." Health Affairs 18:176-181.
Sloan, Frank, and Killard Adamache (1986). "Taxation and the Growth of
Nonwage Benefits." Public Finance Quarterly 14:115-139. Taylor, Amy, and Gail Wilensky (1983). "The Effect of Tax Policies on Expenditures for Private Health Insurance." In Market Reforms in Health Care: Current
66
Gruber
Issues, New Directions, Strategic Decisions, Jack Meyer (ed.). Washington: Ameri-
can Enterprise Institute.
Thorpe, Kenneth L, et al. (1992). "Reducing the Number of Uninsured by Subsidizing Employment-Based Health Insurance: Results from a Pilot Study." Journal of the American Medical Association 267:945-948.
Turner, Robert (1987). "Are Taxes Responsible for the Growth of Fringe Benefits?" National Tax Journal 40:205-220.
Vroman, Susan, and G. Anderson (1984). "The Effect of Income Taxation on the Demand for Employer-Provided Health Insurance." Applied Economics 16:33-43.
Woodbury, Stephen A., and Daniel S. Hammermesh (1992). "Taxes, Fringe Benefits, and Faculty." Review of Economics and Statistics 73:287-296.
DEFINED CONTRIBUTION PENSIONS: PLAN RULES, PARTICIPANT CHOICES, AND THE PATH OF LEAST RESISTANCE James J. Choi Harvard University
David Laibson Harvard University and NBER
Brigitte C. Madrian University of Chicago and NBER
Andrew Metrick University of Pennsylvania and NBER
EXECUTIVE SUMMARY We assess the effect on savings behavior of several different 401(k) plan features, including automatic enrollment, automatic cash distributions, We thank Hewitt Associates for their help in providing the data. We are particularly grateful to Lori Lucas and Jim McGhee, two of our many contacts at Hewitt. We also thank James Poterba and Olivia Mitchell for comments. Choi acknowledges financial support from a National Science Foundation Graduate Research Fellowship. Laibson and Madrian acknowledge financial support from the National Institute on Aging (ROl-AG-16605 and R29-AG-013020 respectively). Laibson also acknowledges financial support from the MacArthur Foundation and the Sloan Foundation.
68
Choi, Laibson, Madrian & Metrick
employer matching provisions, eligibility requirements, investment options, and financial education. We also present new survey evidence on individual savings adequacy. Many of our conclusions are based on an analysis of micro-level administrative data on the 401(k) savings behavior of employees in several large corporations that implemented changes in their 401(k) plan design. Our analysis identifies a key behavioral principle that should partially guide the design of 401(k) plans: employees often follow the path of least resistance. For better or for worse, plan administrators can manipulate the path of least resistance to powerfully influence the savings and investment choices of their employees.
1. INTRODUCTION Over the last 20 years, defined-contribution pension plans have gradually replaced defined benefit pension plans as the primary privately-
sponsored vehicle to provide retirement income. At year-end 2000, employers sponsored over 325,000 401(k) plans with more than 42 million active participants and $1.8 trillion in assets.1 The growth of 401(k)-type savings plans and the associated displacement of defined benefit plans have generated new concerns about the
adequacy of employee savings. Defined contribution pension plans place the burden of ensuring adequate retirement savings square on the backs of individual employees. However, employers make many decisions about the design of 401(k) plans that can either facilitate or hinder the employees' retirement savings prospects. Although the government places some limits on how companies can structure their 401(k) plans, employers nonetheless have broad discretion in their design. Making good plan design decisions requires an understanding of the relationship between plan rules and participant choices. In this paper, we
analyze a new data set that enables us to carefully assess many such relationships. The data set is compiled from anonymous administrative records of several large firms that collectively employ almost 200,000 individuals. Many of these companies implemented changes in the design of their 401(k) plans. These plan changes enable us to evaluate the impact on individual savings behavior of institutional variation in 401(k) plan rules. A list of the companies studied in this paper, along with the plan changes or other interventions that we analyze, appears in Table 1 Appendix A gives a brief description of the data analyzed for each company. 1
See EBRI Databook on Employee Benefits at http://www.ebri.org/facts/12OOfact.htm.
To maintain the anonymity of the companies described in this paper, we refer to them with letters. 2
TABLE 1.
30,000
20,000 10,000
40,000
50,000
Office equipment
Insurance
Food
Utility
Consumer packaged goods
Insurance
B
C
D
E
F
G
Change in eligibility
Change in eligibility Instituted employer match
Increased match threshold
Automatic enrollment
Automatic enrollment Financial education seminars
Automatic enrollment
Savings survey
Plan change or intervention
January 1997
July 1998 October 2000
January 1997
January 1998
January-December 2000
April 1998
January 1997
January 2001
Date of Change or Intervention
'Number of employees (rounded to the nearest 10,000) on December 31, 2000 (Companies A, B, D, 5, F), June 30, 2000 (Company C), or December 31, 1999 (Companies G, H).
30,000
10,000
Food
A
Sizea
Industry
Company
Companies and Their 401(k) Plan Changes or Other Interventions
70
Choi, Laibson, Madrian & Metrick
Because low employee savings rates have motivated plan administrators to adopt many of the 401(k)-plan changes that we discuss in the rest of the paper, we start off in Section 2 with a discussion of savings adequacy. Using new data from a survey that we designed, we find that two-thirds of employees believe that they are saving too little and that one-third of these self-reported undersavers intend to raise their saving rate in the next two months. By matching survey responses to administrative records, we show that employees who report that they save too little actually do have low 401(k) saving rates. However, almost none of the employees who report that they intend to raise their saving rate in the next two months actually do so. This finding introduces a theme that we return to throughout the paper. Specifically, at any point in time employees are likely to do whatever requires the least current effort: employees often follow the path of least resistance. Almost always, the easiest thing to do is nothing whatsoever, a phenomenon that we call passive decision. Such passive decisionmaking implies that employers have a great deal of influence over the savings outcomes of their employees. For example, employer choices of default saving rates and default investment funds strongly influence employee savings levels. Even though employees have the opportunity to opt out of such defaults, few actually do so. In section 3, the heart of our paper, we discuss the impact of changes in seven different types of plan rules. In section 3.1, we show that automatic enrollment in a 401(k) plan dramatically raises participation rates, but that the vast majority of employees accept the automatic-enrollment default contribution rate and investment allocation. By contrast, before automatic enrollment was instituted, few employees chose to invest at these defaults. In section 3.2, we discuss the effects of automatic cash distributions for terminated employees. We argue that automatic cash distributions, which are given to terminated employees with balances below $5,000, undercut retirement wealth accumulation. Most employees with balances below $5,000 who receive such automatic distributions consume the proceeds. By contrast, most employees with balances above $5,000 leave their money in the 401(k) plan. Hence, the automatic cash distributions seem to play a critical causal role in the consumption of these lowbalance 401(k) accounts. In section 3.3, we discuss different interventions designed to raise employee contribution rates. Benartzi and Thaler (2001b) have shown that employees are willing to commit to automatic schedules of slow 401(k) contribution rate increases, and that committing to such a sched-
Defined Contribution Pensions
71
ule wifi result in substantially higher 401(k) savings rates after only a few
years. We report an experiment of our own that shows that a savings intervention that does not include such an automatic commitment component is not successful. In section 3.4, we discuss the effects of the employer match rate and the employer match threshold (the maximum employee contribution that the employer matches) on savings outcomes. We show that adopting an employer match can increase 401(k) participation, and that the match threshold is an important focal point in the selection of employee contribution rates. We also show that increasing the match threshold can raise the contribution rates of individuals with low saving rates. In section 3.5, we discuss the impact of changes in eligibility waiting periods on the 401(k) participation profile (i.e. participation rates plotted against tenure at the job). We show that an increase in the length of time before 401(k) eligibility truncates, but does not shift, the participation profile.
In section 3.6, we discuss mutual-fund menus and the role of employer, or "company," stock. We argue that the menu of asset allocation options and the choice of the default asset allocation influence actual asset allocation decisions and portfolio diversification. Finally, in section 3.7 we discuss the role of financial education in the workplace. Using data that link employees' receipt of financial education to their actual saving behavior, we show that although many seminar attendees plan to make 401(k) savings changes, very few actually do so. Thus, while financial education does improve savings outcomes, its effects are modest at best. We see passive decisionmaking in many of the behavioral patterns described above. Passive decisionmaking partially explains the powerful influence of defaults, the anchoring effects of the match threshold, the remarkable success of automatic schedules of slowly increasing contribution rates, and the influence of mutual fund menus on asset allocation decisions.
We conclude the paper by encouraging employers to implement 401(k) plans that work well for decisionmakers who often use passive strategies like those described above. Employers and policymakers need to recognize that it is difficult to present a neutral menu of options for a 401(k) plan. Framing effects will influence employee choices, and passive employee decisionmaking implies that the default options will often carry the day. Sophisticated employers will choose these defaults carefully, keeping the interests of both employees and shareholders in mind.
72
Choi, Laibson, Madrian & Metrick
2. SAVINGS ADEQUACY In January 2001, we administered a savings adequacy survey to a random sample of employees at a large U.S. food corporation (Company A) with approximately 10,000 employees. Of these employees, 1,202 were sent an e-mail soliciting their participation in a Web-based survey on satisfaction with various aspects of the company-sponsored 401(k) plan.3 Because participation in the survey was solicited by e-mail and the survey itself was conducted on the Web, the universe of potential respondents is restricted to those with Internet access at work.4 Our survey had two different versions. In this section, we discuss the savings adequacy version that was sent to 590 of the employees with computers. From this sample we received 195 usable responses. A copy of the complete survey is reproduced in Appendix B; we discuss only a subset of the questions in the analysis below. In addition to the survey
responses, we also have administrative data on the 401(k) savings choices of survey respondents both before and after the survey. This includes participation decisions, contribution rates, and asset allocation choices from January 1996 through April 2001. We first asked respondents to report how much they should ideally be saving for retirement.5 The average response is 13.9 percent of income. We than asked respondents to evaluate their actual saving rate. Twothirds (67.7 percent) of the respondents report that their current saving rate is "too low" relative to their ideal saving rate.6 One-third (30.8
percent) of the respondents report that their current saving rate is "about right." Only 1 out of 195 respondents (0.5 percent) reports that his or her current saving rate is "too high." To evaluate how well individual perceptions of savings adequacy correlate with actual savings behavior, we report in Table 2 the distribution
The solicitation included an inducement to actually complete the survey: two respondents were randomly selected to receive gift checks of $250, and one respondent was selected to receive a gift check of $500.
" Naturally, restricting our sample to Internet users biases our sample toward employees with greater financial sophistication. Our survey reveals that an employee's level of Internet experience correlates with his self-reported financial knowledge. Likewise, home Internet access also correlates with financial knowledge. See question 10 from the survey (Appendix B).
See question 11 from the survey (Appendix B). For our empirical analysis we aggregate the categories "far too low" and "a little too low" into one category ("too low"). Likewise, we aggregate the categories "far too high" and "a little too high" into one category ("too high"). 6
Defined Contribution Pensions
73
TABLE 2.
Self-reported Retirement Savings Adequacy and the Distribution of Actual 401(k) Contribution Rates (Company A) Distribution of 401(k) Contribution Rates as a Fraction of Income
Respondents who describe their savings rate as "too low" Respondents who describe their savings rate as "about right"
5%-8%
9%-12%
36%
36%
27%
12%
15%
73%
See question 11 from the survey in Appendix B. We aggregate the categories "far too low" and "a little too low" into one category ("too low").
of actual pre-tax 401(k) saving rates conditional on respondents' answers
to the savings adequacy questions discussed above. Since we use the plan's administrative records, our analysis of actual 401(k) saving rates does not suffer from reporting biases. We divide the actual pre-tax 401(k) saving rates into three categories: 0 to 4 percent of income, 5 to 8 percent of income, and 9 to 12 percent of income. Our scale tops out at 12 percent
because this is the maximum pre-tax 401(k) contribution rate in Company A. Among the respondents who said that their current saving rate is "too low," 36 percent had an actual 401(k) saving rate of 0 to 4 percent,
another 36 percent had a saving rate of 5 to 8 percent, and 27 percent had a saving rate of 9 to 12 percent. In contrast, among those who said that their current savings rate is "about right," 12 percent had a 401(k)
saving rate of 0 to 4 percent, 15 percent had a saving rate of 5 to 8 percent, and 73 percent had a saving rate of 9 to 12 percent. These comparisons reveal that respondents who report that their saving rate is too low do have lower actual saving rates than respondents who report that their saving rate is about right. In the former group the average pre-tax 401(k) contribution rate is 5.8 percent of income, in contrast to 9.0 percent in the latter group. We also asked respondents to describe their plans for the future. None of our respondents expressed an intention to lower their contribution rate. But 35 percent of the respondents who said that their saving rate was too low intended to increase their contribution rate over the next few months. By contrast, 11 percent of respondents who said their saving rate was about right intended to increase their contribution rate over the next few months. Among those who planned to raise their contribution rate, over half (53 percent) said that they planned to do so in the
74
Choi, Laibson, Madrian & Metrick
next month. Another quarter (23 percent) planned to make the change within two months. So far our data shows a familiar pattern. Respondents report that they save too little and that they intend to raise their saving rate in the future. Other savings adequacy surveys reach similar conclusions (Bernheim, 1995; Farkas and Johnson, 1997). Our survey is distinguished by our ability to cross-check responses against 401(k) records. We have shown that respondents who say that their saving rate is too low actually do have substantially lower pretax 401(k) contribution rates. So their retrospective reports are accurate. We have also checked to see whether their forward-looking plans are consistent with their actual subsequent behavior. Of those respondents who report that their saving rate is too low and that they plan to increase their contribution rate in the next few months, only 14 percent actually do
increase their contribution rate in the four months after the survey. Hence, we find that respondents overwhelmingly do not follow through on their good intentions. In summary, out of every 100 respondents, 68 report that their saving rate is too low; 24 of those 68 plan to increase their 401(k) contribution rate in the next few months; but only 3 of those 24 actually do so. Hence, even though most employees describe themselves as undersavers and many report that they plan to rectify this situation in the next few months, few follow through on this plan.
Needless to say, these data are hard to interpret. It's not clear what subjects mean when they say that they save too little. It's also not clear what subjects mean when they say that they intend to raise their contribution rate in the next few months. However, this evidence is at least consistent with the idea that employees have a hard time carrying out the actions that they themselves say they wish to take. Employers seem to be concerned about such failures. Many of the institutional changes discussed below in Section 3 were initiated by plan administrators in an effort to raise employee saving rates.
3. Seven Institutional Features of 401(k) Plans In this section, we turn to an analysis of how several features of 401(k) plans affect employee 401(k) saving behavior.
3.1 Automatic Enrollment The typical 401(k) plan requires an active election on the part of employees to initiate participation. A growing number of companies, however, have started automatically enrolling employees into the 401(k) plan unless the employee actively opts out. While automatic enrollment is still
Defined Contribution Pensions
75
uncommon, a recent survey indicates that its adoption has increased quite rapidly over the past few years.7 The interest of many companies in automatic enrollment has stemmed from their persistent failure to pass the IRS nondiscrimination rules that apply to pension-plan provision. As a result of failing these tests, many companies have had to make either ex post 401(k) contribution refunds to highly compensated employees or retroactive company contributions on behalf of non-highly compensated employees in order to come into compliance. In addition, many companies have tried to reduce the possibility of non-discrimination testing problems by ex ante limiting the contributions that highly compensated employees can make. The hope of many
companies adopting automatic enrollment has been that participation among the non-highly compensated employees at the firm will increase sufficiently that non-discrimination testing is no longer a concern. While some companies have been concerned about the potential legal repercussions of automatically enrolling employees in the 401(k) plan, the U.S. Treasury Department has issued several opinions that support employer use of automatic enrollment. The first Treasury Department opinion on this subject, issued in 1998, sanctioned the use of automatic enrollment for newly hired employees.8 A second ruling, issued in 2000, further validated the use of automatic enrollment for previously hired employees not yet participating in their employer's 401(k) plan.9 In addition, during his tenure as Treasury Secretary, Lawrence H. Summers publicly advocated employer adoption of automatic enrollment.'0
A growing body of evidence suggests that automatic enrollmenta simple change from a default of non-participation to a default of participationsubstantially increases 401(k) participation rates." To assess the impact of automatic enrollment on savings behavior, we examine the experience of three large companies analyzed in Choi, Laibson, Madrian, and Metrick (2001) that implemented automatic enrollment between January 1997 and April 1998. Companies B and C implemented In a recent survey, Hewitt Associates (2001) reports that 14 percent of companies utilized automatic enrollment in 2001, up from 7 percent in 1999. 8 See IRS Revenue Ruling 98-30 (Internal Revenue Service, 1998).
See IRS Revenue Ruling 2000-8 (Internal Revenue Service, 2000a). See also Revenue Rulings 2000-33 and 2000-35 (Internal Revenue Service, 2000b).
See "Remarks of Treasury Secretary Lawrence H. Summers at the Department of Labor Retirement Savings Education Campaign Fifth Anniversary Event" at http://www.ustreas .gov/press/releases/ps785.htm along with related supporting documents. " See Madrian and Shea (2001a), Choi, Laibson, Madrian, and Metrick (2001), Fidelity (2001), and Vanguard (2001).
76
Choi, Laibson, Madrian & Metrick
automatic enrollment for new hires. Company D also implemented automatic enrollment for new hires, but in addition subsequently applied it to non-participating employees who were 401(k)-eligible at the time when automatic enrollment was initially adopted.12 Table 3 illustrates the difference in 401(k) participation rates by tenure before and after automatic enrollment. For each company, we report three columns of figures. The first and second columns contain the fraction of employees hired before and after automatic enrollment was implemented who are 401(k) plan participants at six-month increments of tenure.'3 The third colunm differences these participation rates, yielding the incremental impact of automatic enrollment on plan participation. In all three companies, 401(k) participation for employees hired before automatic enrollment starts out low and increases quite substantially
with tenure. At six months of tenure, 401(k) participation rates range from 26 to 43 percent at these three companies. Participation rates increase to 50 to 62 percent at 24 months of tenure, and to 65 to 69 percent at 36 months of tenure. The profile of 401(k) participation for employees hired under automatic enrollment is quite different. For these employees, the 401(k) participation rate starts out high and remains high. At six months of tenure, 401(k) participation ranges from 86 to 96 percent at these three companies, an increase of 50 to 67 percentage points relative to 401(k) participation rates prior to automatic enrollment. Because 401(k) participation increases with tenure in the absence of automatic
enrollment, the incremental effect of automatic enrollment on 401(k) participation declines over time. Nonetheless, at 36 months of tenure, 401(k) participation is still a sizable 31 to 34 percentage points higher under automatic enrollment. While most companies that implement automatic enrollment do so only for newly hired employees, some companies have applied automatic enrollment to previously hired employees who have not yet initiated participation in the 401(k) plan. Choi, Laibson, Madrian, and Metrick (2001) show that for previously hired employees at Company D, automatic enrollment also substantially increases the 401(k) participation rate, Because of concurrent changes in eligibility for employees under the age of 40 at Company D, we restrict the sample of employees in the analysis at the company to those aged 40 or over at the time of hire. These employees were immediately eligible to participate in the 401(k) plan, both before and after the switch to automatic enrollment. 13 Because of differences in the available data from these companies, the numbers are not directly comparable across companies. For Company C, we have data on 401(k) participation on the data collection dates, and thus the numbers in columns 1 and 2 for Company C represent contemporaneous 401(k) participation rates. For Companies B and D, we have data on the date of initial 401(k) participation, and thus the numbers in columns 1 and 2 for Companies B and D represent the fraction of employees who have ever participated in the 12
401(k) plan.
26.4 37.8 47.7 54.1 60.0 64.7
Before AE
93.4 95.7 97.0 97.6 97.7 98.8
After AE
Hire date
Company B
67.0 57.9 49.3 43.5 37.7 34.1
before
After-
--
35.7 40.2 44.3 49.8
-
85.9 85.3 86.0 85.7
-
45.1 41.7 35.9
50.2
Participation (%) Company C Hire date AfterBefore AE After AE before 42.5 49.6 56.6 61.7 65.6 69.0
100.0
96.0 96.6 97.2 99.1 98.8
Hire date Before AE After AE
Company D
53.5 47.0 40.6 37.4 33.3 31.0
before
After-
The sample for Companies B and C is all 401(k)-eligible employees. The sample for Company D is 401(k)-eligible employees aged 40+ at the time of hire. For Company D, the data in the "Before AE" column include only employees not yet subject to automatic enrollment when it was applied to previously hired non-participants. For Companies B and D, the first two columns of numbers give the fraction of employees who have ever participated in the 401(k) plan. For Company C, the first two columns give the fraction of employees contemporaneously participating in the 401(k) plan.
24 30 36
12 18
6
Tenure (months)
TABLE 3.
401(k) Participation by Tenure Before and After Automatic Enrollment
78
Choi, Laibson, Madrian & Metrick
although the increase in participation is slightly smaller than that seen for newly hired employees. Madrian and Shea (2001a) and Choi, Laibson, Madrian, and Metrick (2001) also discuss how the effects of automatic enrollment vary across various demographic groups. While automatic en-
rollment increases 401(k) participation for virtually all demographic groups, its effects are largest for those individuals least likely to participate in the first place: younger employees, lower-paid employees, and Blacks and Hispanics. One might conclude that since 401(k) participation under automatic
enrollment is so much higher than when employees must choose to initiate plan participation, automatic enrollment "coerces" employees into participating in the 401(k) plan. However, if this were the case, we should expect to see participation rates under automatic enrollment declining with tenure as employees veto their "coerced" participation and opt out. But remarkably few 401(k) participants at these companies, whether hired before automatic enrollment or hired after, reverse their participation status and opt out of the plan. In our three companies, the fraction of 401(k) participants hired before automatic enrollment who drop out in a 12-month period ranges from 1.9 to 2.6 percent, while the fraction of participants subject to automatic enrollment who drop out is only 0.3 to 0.6 percentage points higher. To us, this evidence suggests that most employees do not object to saving for retirement. In the absence of automatic enrollment, however, many employees tend to delay taking action. Thus, automatic enrollment appears to be a very effective tool for helping employees begin to save for their retirement. While automatic enrollment is effective in getting employees to participate in their company-sponsored 401(k) plan, it is less effective at motivating them to make well-planned decisions about how much to save for retirement or how to invest their retirement savings. Because companies cannot ensure that employees will cho9se a contribution rate or an asset allocation before the automatic enrollment deadline, the company must establish a default contribution rate and a default asset allocation. Most employees follow the path of least resistance and passively accept these defaults.
Figure 1 shows the distribution of 401(k) contribution rates at our three companies for employees hired before and after automatic enrollment. Because contribution rates may change with tenure, for all three companies we have restricted the sample to employees hired before and after automatic enrollment with equivalent levels of tenure.14 All three 14 In Company B, the sample is restricted to employees with 24-35 months of tenure; in Company D to those with 0-23 months of tenure; and in Company D to those with 12-35 months of tenure.
Defined Contribution Pensions
79
FIGURE 1. The Distribution of Contribution Rates of 401(k) Participants Hired Before and After Automatic Enrollment 80 70
J
5
6050
.2
71
64
42
40 30 20
1
U-
Iii 1214
C)
0 Company B Company B Company C Pre-AE
Post-AE
Pre-AE
AE default and <Match Threshold
42
3 27
22.-
12
Company C Company D Company D Post-AE Pre-AE Post-AE
AE Default Match Threshold
D>Match Threshold
companies match employee contributions up to 6 percent of compensation, the match threshold in Figure 1. But the default contribution under automatic enrollment is much lower than this-2 percent in company B and 3 percent in companies C and D. Before automatic enrollment, 63 to 79 percent of plan participants at these companies contribute at or above the match threshold. Only 11 to 20 percent voluntarily choose the contribution rate specified by their employers as the default under automatic
enrollment. In contrast, 42 to 71 percent of participants hired under automatic enrollment contribute at the default rate, while only 26 to 49 percent contribute at or above the match threshold. Automatic enrollment has similar effects on the asset allocation of plan participants. Figure 2 shows the allocation of 401(k) balances between stocks, bonds and the combination of stable value and money market funds. Once again, because asset allocation may change with tenure, we have restricted the sample to employees with equivalent levels of tenure.15 In two of the three companies, the default fund under automatic enrollment is a stable-value fund, while in the third it is a money market fund. As Figure 2 shows, employees hired before automatic enrollment have the majority of their plan assets (53 to 81 percent) allocated to the stock market, and only a small fraction of their assets (10 to 18 percent) allocated to stable-value or money market funds. These percentages are 15
See footnote 14.
80
Choi, Laibson, Madrian & Metrick
FIGURE 2. Asset Allocation of 401(k) Participants Hired Before and After Automatic Enrollment 100%
<
80%
o e
60/a
0
;0 0 U-
40% 20% 0%
Company B Pre-AE
Company B Post-AE
Company C Pre-AE
Company C Post-AE
Company 0 Pre-AE
Company D Post-AE
0% Stocks 0% Bonds % Stable Value/MM
effectively reversed for employees hired under automatic enrollment. For this group of participants, 48 to 81 percent of assets are invested in stablevalue or money market funds, a group that includes the automatic enrollment default at all three companies, and only 16 to 51 percent of assets are invested in the stock market. Overall, the fraction of assets allocated to the
stock market falls by 22 to 53 percentage points, while the fraction of assets allocated to stable value funds or the money market increases by 31 to 71 percentage points. Choi, Laib son, Madrian, and Metrick (2001) show
that these effects are driven both by the conversion of would-be nonparticipants to the defaults and by employees who would have participated in the absence of automatic enrollment but with different elections.
Given the evidence of delay in the election of 401(k) participation before automatic enrollment shown in Table 3, one might speculate that there is the same type of delay in the movement away from the default contribution rate and asset allocation under automatic enrollment. Table 4 suggests that this is indeed the case. At six months of tenure, between 55 and 73 percent of participants contribute at the default and have their assets invested wholly in the default fund. At 24 months of tenure, the fraction of participants at the default falls to 40 to 51 percent, and at 36 months of tenure to 44 to 48 percent. So, with time, employees do move away from the automatic enrollment defaults. Nonetheless, after three years, almost half of participants are still "stuck" at the default.16 16
Choi, Laibson, Madrian, and Metrick (2001) show that compensation is the strongest
determinant of how quickly employees move away from the automatic enrollment
Defined Contribution Pensions 81
TABLE 4.
Fraction of 401(k) Participants at the Automatic Enrollment Default Tenure (months) 6 12 18
24 30 36
Company B 67.2 61.2 61.4 51.4 53.9 43.6
Fraction (%) Company C 72.6 59.3 47.6 39.6
-
Company D 54.5 50.9 43.7 39.5 39.4 48.2
The sample for Companies B and C is 401(k) participants hired after automatic enrollment. The sample for Company D is further restricted to participants aged 40+ at the time of hire.
Taken as a whole, the evidence in this section indicates that defaults can have a powerful effect on the nature of individual saving for retirement. In terms of promoting overall savings for retirement, automatic enrollment as structured by most employers is a mixed bag. Clearly automatic enrollment is very effective at promoting one important aspect of savings behavior, 401(k) participation. This simple change in the default from non-participation to participation results in much higher 401(k) participation rates. But, like companies B, C, and D, most employers that have adopted automatic enrollment have chosen very low default contribution rates and very conservative default funds (Profit Sharing/401(k) Council of America 2001; Vanguard 2001). These default choices are incon-
sistent with the retirement savings goals of most employees. This evidence does not argue against automatic enrollment as a tool for promoting retirement saving; rather, it argues against the specific automatic enrollment defaults chosen by most employers. Employers who seek to facilitate the retirement savings of their employees need to respond to the tendency of employees to stick with the default. Employ-
ers should choose defaults that foster successful retirement saving when the defaults are passively accepted in their entirety. Automatic enrollment coupled with higher default contribution rates and more aggressive default funds would greatly increase wealth accumulation for retirement.17 The results in this section also suggest an important caveat in thinking about the design of personal accounts in a reformed defaulthighly compensated employees tend to move away from the default more rapidly than those with lower pay. 17
See section 3.3 for another alternative to a higher initial default contribution rate.
82
Choi, Laibson, Madrian & Metrick
Social Security systemwhatever defaults are chosen will need to be chosen carefully.
3.2 Automatic Cash Distributions for Terminated Employees with Low Account Balances Another aspect of 401(k) plan design that highlights the importance of defaults on 401(k) savings outcomes is the treatment of the 401(k) balances of former employees. When an employee leaves a firm, the employee may explicitly request a cash distribution, a direct rollover of 401(k) balances to an IRA, or a rollover to another employer's 401(k) plan. If the terminated employee does not make an explicit request, the balances typically remain in the 401(k) plan. Under current law, however, if the plan balances are less than $5,000 and the former employee has not elected some sort of rollover, the employer has the option of compelling a cash distribution.
To document the importance of this mandatory cash distribution threshold, figure 3a and b plot the relationship between the size of 401(k) balances and the likelihood that a terminated employee receives a distribution from the 401(k) plan at Companies B and D. We consider the experience of 401(k) participants whose employment terminated any time during 1999 or January through August of 2000.18 We order the employees according to the size of their 401(k) balances and then divide them into groups of 100. We then calculate the average balance size for each group (the x axis, plotted on a log scale) and the average fraction of employees who receive a distribution from the plan by December 31, 2000 (the y axis). The measure of 401(k) balances used on the x axis is the average participant balance as of December 31 of the year prior to the year in which the termination occurred.19 This measure of balances is likely to understate the actual balances of plan participants at the time of termination because the incremental contributions made to an individual's account between December 31 of the previous year and the date of
termination are excluded (as are any capital gains or losses over this period). In both companies, around 90 percent of terminated participants with prior year-end balances less than $1,000 receive a distribution subsequent to termination. In contrast, in Company D, a rather constant one-third of terminated participants with year-end balances greater than $5,000 reThis includes both voluntary and involuntary terminations. That is, employees terminated in 2000 have a balance measure from December 31, 1999, while employees terminated in 1999 have a balance measure from December 31, 1998. We use this measure of balances because it is the only measure that we have in our data. 18 19
Defined Contribution Pensions
83
FIGURE 3. Balance Size and the Likelihood of a 401(k) Distribution for Terminated Employees: (a) Company B; (b) Company D
.75
.5 -
.25 -
0
I
I
I
I
25000 250000 401(k) balance at year-end Drior to termination (dollars) 1000
100
5000
(a)
1
.75 -
.5 -
.25 -
0 I
I
I
I
100 1000 5000 25000 250000 401(k) balance at year-end prior to termination (dollars)
(b)
84
Choi, Laibson, Mcdrian & Metrick
ceive a distribution. In Company B, this fraction is even lower, at about 18
percent, and there is some additional slight decline in the likelihood of receiving a distribution with respect to balance size beyond the $5,000 threshold. Between $1,000 and $5,000 in year-end balances, the fraction of
terminated participants receiving a distribution falls rather steadily and quite significantly at both companies. This reflects the decreasing likelihood that terminated participants will have a final balance of less than $5,000 that is subject to an involuntary cash distribution. For example, consider an employee at Company D making $40,000 per year who is contributing 6 percent of pay to the 401(k) plan with a 50percent employer match that is vested. If this individual leaves his job at the end of August, the additional employer plus employee contributions to the 401(k) plan wifi amount to $2,400. Assuming no net capital gains or losses, this individual will face a mandatory cash distribution if his prior year-end balances were less than $2,600 (because $2,400 plus anything less than $2,600 will fall under the $5,000 distribution threshold). If his prior year-end balances were higher than $2,600, however, the company would not be able to compel a cash distribution, because his total balances subsequent to termination would exceed $5,000. Thus, employees with higher prior-year-end balances will be less likely to face an automatic distribution upon termination, because they are more likely to have had balance increases that bring them above the $5,000 threshold.
Of course, even in the case of an automatic cash distribution, the former employee does have the option to roll the account balance over into an IRA or the 401(k) plan of another employer, regardless of the size of the account balance. But previous research suggests that the probability of receiving a cash distribution and rolling it over into an IRA or another 401(k) plan is very low when the size of the distribution is small. Instead, these small distributions tend to be consumed.2° When employers compel a cash distribution and employees receive an unexpected check in the mail, it is much easier to consume the distribution than to figure out how to roll it over into an IRA or another employer's 401(k) plan. This default treatment of the account balances of terminated employees provides another example of how many individuals follow the path of least resistance. When balances exceed $5,000, the vast majority of employees leave their balances with their former employer, the leastPoterba, Venti, and Wise (1998a) report that the probability that a cash distribution is rolled over into an IRA or another employer's pian is only 5 to 16 percent for distributions of less than $5000. The overall probability that a cash distribution is rolled over into an IRA or another employer's plan or invested in some other savings vehicle is slightly higher at 14 to 33 percent. 20
Defined Contribution Pensions
85
effort option. When balances are below $5,000 and are subject to a mandatory cash distribution unless the employee elects otherwise, most individuals receive an unsolicited check in the mail and then consume the money rather than rolling it over into another type of saving plan also the least-effort option.
This analysis suggests that the rollover provisions of the recently passed Economic Growth and Tax Relief Reconciliation Act of 2001 (EGTRRA) wifi indeed increase retirement savings. Under the new law, if the account balance is between $1,000 and $5,000, employers will no longer be able to compel a cash distribution if a former employee does not elect a rollover; rather, employers will be required to establish an IRA on behalf of participants if they choose not to maintain these accounts (Watson Wyatt, 2001). Although this provision of the law does not take effect until the Department of Labor issues final regulations regarding implementation (something that is not required to happen until 2004), firms need not wait until then to voluntarily adopt similar measures. As with automatic enrollment in 401(k) plans, default rollovers have also been sanctioned by the IRS.21 Such a change in the default treatment of the small balances of terminated employees is a simple step that would further enhance the retirement savings plans of many individuals.22
3.3 Automatic Contribution Rate Increases One 401(k) plan feature designed to capitalize on the inertia described in sections 3.1 and 3.2 is the Save More Tomorrow (SMT) plan developed by Shiomo Benartzi and Richard Thaler (Benartzi and Thaler 2001b).
Under this plan, participants agree in advance that in the absence of explicit action on their part, their 401(k) contribution rate wifi be increased by a certain amount each time they receive a nominal pay raise until it achieves a preset maximum. For example, suppose that a worker agrees to have his contribution rate increased by 2 percentage points each time he gets a raise. If the worker receives one raise in each of the following three years, then his contribution rate would rise a total of 6
percentage points over this time period. This plan is carefully constructed to make use of several themes in behavioral economics. By 22
See IRS Revenue Ruling 2000-36 (Internal Revenue Service, 2000b).
We should note, however, that previous research also suggests that although small distributions tend to be consumed rather than rolled over into other retirement savings vehicles, these small distributions represent a small fraction of total retirement savings (Poterba, Venti, and Wise, 1998a). Thus, while automatically rolling such distributions over into an IRA will undoubtedly increase retirement saving, its impact on aggregate retirement saving is likely to be modest.
86
Choi, Laibson, Madrian & Metrick
requiring a present commitment for future actions, the SMT plan alleviates problems of self-control and procrastination. And by increasing contributions on dates of future salary increases, the effects of loss aversion are mitigated, because workers will see little or no reduction in their nominal take-home pay. (This presumes that participants are subject to money illusion.)
The striking results of the first experiment with the SMT plan are
reported in Benartzi and Thaler (2001b). This first experiment was conducted at a mid-size manufacturing company. This company was experiencing problems in getting low-salary workers to participate and contribute at high levels to the 401(k) plan. To combat these problems, the company hired an investment consultant to meet with employees and help them plan their retirement savings. After an initial interview with each employee, the consultant would gauge the employee's wifiingness to increase his savings rate. Employees judged to have a high willingness to save more would receive an immediate recommendation for a large increase in their savings rate. 79 workers fell into this group. Employees judged to be reluctant to save more would be offered the option of enrolling in the SMT plan. 207 workers fell into this group. The version of the SMT plan that was implemented set up a schedule of annual contribution rate increases of 3 percentage points. This is a relatively aggressive implementation, as the annual nominal salary increases at this company were only a little bit higher than 3 percent. The results of the experiment show that the SMT plan can have an enormous impact on contribution rates. Of the 207 participants offered the SMT plan as an option, 162 chose to enroll. Furthermore, 129 of these
162 (80 percent) stayed with the plan through three consecutive pay raises. At the beginning of the SMT plan, these 162 workers had an average contribution rate of 3.5 percent; by the time of their third pay raise, these workers (including those that eventually dropped out) had an average contribution rate of 11.6 percent. Recall that these original 207 participants were selected from a larger sample based on their relative reluctance to increase their savings rates. In comparison, 79 workers had indicated a willingness to increase their contributions immediately and were never enrolled in the SMT plan; these workers increased their average contribution rate from 4.4 percent to 8.7 percent over the same
time period. Since it is reasonable to assume that this latter group of workers represents a more highly motivated group of savers than the SMT-plan participants, the increases by the SMT-plan participants are very striking. As a further comparison, consider that the median 401(k) contribution rate of participants in 401(k) plans in general is approximately 7 percent of pay (Investment Company Institute, 2000). Thus, the
Defined Contribution Pensions
87
SMT-plan participants went from half of this median contribution rate before signing up for the SMT plan to a contribution rate 50 percent higher three years later. Despite the clear success of the SMT plan in increasing contribution rates, there remain several important caveats. First, the plan is not guided by any well-specified model of what ideal savings should be. Even if we accept that cleverly designed commitment devices can enable workers to break from suboptimal behavior patterns, these same devices may overshoot the optimal targets. Second, the increases in 401(k) contribution rates may be offset by dissaving elsewhere.23 Although 401(k) saving has many advantages, it may still be inefficient if it leads participants to increase high-interest credit-card debt. Also, we do not know how much of the additional contributions were later reduced by plan loans or hardship withdrawals. In a plan that does not have an employer matchunlike the one used in the original SMT experimentit is not clear that increasing 401(k) contributions is always a good idea. Notwithstanding these caveats, the SMT plan is certainly a provocative attempt
to use behavioral economics to increase savings rates, and the early results are highly encouraging and deserve further study. Our 401(k) survey (discussed in section 2) sheds light on the mechanisms that make the SMT plan work. We generated two versions of our survey. One version (already discussed above) asked questions about both savings adequacy and intentions regarding planned future investment changes (e.g. plans to change the contribution rate and the asset mix). We call this the savings-adequacy version. We also generated a pared down version of the survey that contained no questions about either savings adequacy or intentions. We call this the control version. We randomly assigned the two different versions of the survey to employees, and we checked to see whether the savings-adequacy questionnaire had an impact on subsequent 401(k) investment choices. In other words, we looked to see whether the process of thinking about savings adequacy and formulating one's future savings plans actually led to a greater propensity to subsequently increase (or decrease) one's saving rate.
It turns out that this attention manipulation had no impact. In other words, getting someone to think about his or her own savings adequacy did not lead to any differential future behavior. This result sheds some light on the success of the SMT plan. The SMT plan has many different effects. It encourages employees to think about their savings adequacy. See Engen, Gale, and Scholz (1994, 1996) for a discussion of asset shifting and its consequences for measuring 401(k) effectiveness. See Poterba, Venti, and Wise (1996, 1998b) for evidence that asset-shifting effects are not large.
88
Choi, Laibson, Mad nan & Metrick
It also sets in motion a series of automatic contribution rate increases. Our survey experiment demonstrates that getting employees to think about savings inadequacy is not enough. Employees also need a loweffort mechanism to help them carry out their plans to increase their contribution rate. The SMT plan provides exactly such a tool.
3.4 Matching Although automatic enrollment and the SMT plan provide lots of food for thought, they are still relatively new 401(k)-plan features that have yet to be adopted on a widespread scale. A more common feature of 401(k) plans is the employer match. For each dollar contributed by the employee to the plan, the employer contributes a matching amount up to a certain threshold (e.g. 50 percent of the employee contribution up to 6 percent of compensation). Although the effects of employer matching on 401(k) participation and contribution rates have been widely studied, the conclusions from this research are decidedly mixed. This derives in part from the inherent difficulties associated with identifying the influence of matching on 401(k) saving behavior. In theory, introducing an employer match should increase participation in the 401(k) plan. In practice, however, it is difficult to disentangle this effect from the potential correlation between the savings preferences of employees and the employer match. For example, companies that offer a generous 401(k) match may attract employees who like to save, biasing upward the estimated effect of an employer match on 401(k) participation. Using cross-sectional data, Andrews (1992), Bassett, Fleming, and Rodrigues (1998), Papke and Poterba (1995), Papke (1995), and Even and Macpherson (1997) all find a positive correlation between the availability of an employer match and 401(k) participation. The results are more varied, however, in studies that attempt to control for the correlation between the employer match and other unobserved factors that affect 401(k) saving behavior. Even and Macpherson (1997) use an instrumental-variables approach to allow for the endogeneity of the employer match and still find a large positive effect of matching on 401(k) participa-
tion. However, it is not clear that the firm characteristics they use as instrumental variables are in fact uncorrelated with unobservable employee savings preferences. Because she uses longitudinal data on firms, Papke (1995) is able to include employer fixed effects to allow for the correlation between the employer match and other factors that affect saving behavior. With the addition of these fixed effects, the relationship
between the employer match and 401(k) participation goes away, but these results are difficult to interpret because Papke only observes aver-
Defined Contribution Pensions
89
age match rates, not marginal rates. Kusko, Poterba, and Wilcox (1998)
examine several years of individual-level data in a company whose match rate varied from year to year with the company's prior-year profitability. They also find no relationship between the match rate and 401(k) participation. However, the transient nature of the match-rate changes at this company make it difficult to extrapolate these results to the permanent types of match changes that most companies are likely to consider. The empirical evidence on matching and 401(k) contribution rates is even less decisive than that on 401(k) participation, although in theory the effects here are less straightforward as well. Whereas introducing an employer match where there wasn't one before should lower the contribution rates of employees who were already contributing in excess of the match threshold (an income effect), its impact on those previously contributing at or below the match threshold is ambiguous (opposing income and substitution effects). The effects would be similar for increasing the match rate while maintaining the same match threshold. Increasing an existing non-zero match threshold while keeping the match rate constant
should have no effect on people contributing below the old threshold, increase contributions for people at the old threshold (a substitution effect), have an ambiguous effect for people above the old threshold but at or below the new threshold (opposing income and substitution effects), and decrease contribution rates for people above the new threshold (an income effect).
The actual empirical research on matching and 401(k) contribution rates has focused largely on the relationship between the match rate and average 401(k) contribution rates. Andrews (1992) finds that a higher employer match rate reduces the average 401(k) contribution rate; Bassett, Fleming, and Rodrigues (1998) find no effect; Papke and Poterba (1995) and Even and Macpherson (1997) find a positive relationship; and Kusko, Poterba, and Wilcox (1998) find a small but positive effect of the match rate on average 401(k) contribution rates. Papke (1995) finds a positive effect of the match rate on total employee contributions at low match rates, but a negative effect at higher match rates. These disparate results are perhaps not so surprising, given that theory has little to say about the effect of the match rate per se on the average 401(k) contribution rate. In this paper, we are able to avoid some of the confounds of previous matching studies by examining the individual behavior of participants before and after permanent changes in the 401(k) match structure at two companies. In these natural experiments, participant behavior before the changes serves as a control for participant behavior after the changes. We also examine the effect of matching on the distribution of 401(k)
90
Choi, Lciibson, Madrian & Metrick
contribution rates rather than on the average 401(k) contribution rate and show the importance of considering the match threshold, a facet of employer matching largely ignored in previous research, as well as the match rate. The first company that we consider, Company B, increased its match
threshold on January 1, 1997, while keeping its match rate constant. Before that time, union workers received a 50-percent match on the first 5 percent of income contributed to the 401(k) plan, while management employees received a 50 percent match on the first 6 percent of income. On January 1, 1997, the match threshold for union employees increased to 7 percent, while that for management employees increased to 8 percent. Contributions up to the new threshold were still matched at 50
percent, although the match on the incremental 2 percent of the new threshold was invested in employer stock, whereas the match up to the old threshold had been, and continued to be, invested at the discretion of the employee. To examine the impact of this change in the match structure on 401(k)
savings behavior, we utilize a combination of both longitudinal and cross-sectional data. We have longitudinal data on the 401(k) contribution rate in effect on each day from March 31, 1996 to February 28, 2000 for every worker who was enrolled in the 401(k) plan during that time.
We also have cross sections of all active employees at Company B at year-end 1998, 1999, and 2000 that contain information on participation status, original enrollment date, original hire date, and demographics. In order to assess the effect of the threshold change on participation, we estimate a Cox proportional-hazard model of time from hire until the date of initial participation in the 401(k) plan. We control for gender and age (with both linear and quadratic terms), and also include a dummy variable that equals 1 after the new threshold took effect (January 1, 1997). We exclude all employees hired before January 1, 1996, because the company eliminated its length-of-service requirement for 401(k) participation on that date. We also exclude employees hired after December 31, 1997, because the company switched from a traditional definedbenefit to a cash-balance pension plan at that time for newly hired employees. The first colunm of Table 5 presents the estimated hazard ratios
associated with each independent variable. As one might expect for a change that does not affect the marginal incentives to participate in the 401(k) plan, we find that this increase in the match threshold has no significant effect on 401(k) participation. We next look at the impact of the threshold change on 401(k) contribution rates. Figure 4 plots the distribution of contribution rates over time
for all workers who were contributing to the 401(k) plan on March 31,
Defined Contribution Pensions
91
TABLE 5.
Employer Matching and 401(k) Participation Hazard ratio
Independent variable Female
Company E
Company F
0.8964
1.0237 (0.45)
(-1.21) Age Age2
1.1376** (3.54)
0.9985**
(-3.25) Threshold change
1.1480** (6.58)
0.9984**
(-5.88)
0.7976
(-1.69) Match introduction
1 .4642**
(6.84) Coefficients estimated from a Cox proportional-hazard model of 401(k) participation with time-varying covariates. For Company F, the sample is employees hired during 1996 or 1997 and still employed at year-end 1998, 1999, or 2000. For Company F, the sample is employees hired on or after January 1, 1998 and still employed at year-end 1998, 1999, or 2000. In Company E, the variable threshold change is
a dummy variable that equals 1 after the match threshold was raised in Company G (on January 1, 1997). In Company F, match introduction is a dummy variable that equals 1 after the company match was armounced to employees (on July 1, 2000). The reported coefficients are hazard ratios, with corresponding z statistics in parentheses. 'indicates that the coefficient is significantly different from unity at the 1-percent level.
1996. As workers leave the firm, they are dropped from the sample. The switch from the old threshold to the new threshold is clearly apparent. There is an immediate jump from the old threshold to the new threshold
when the change occurred in January 1997, and a continued slower adjustment over the next three years as more and more people shift from the old to the new threshold. This suggests that there is a strong substitution effect for contributors at the old threshold. In contrast, the fraction
of participants at the other contribution rates is fairly stable over this entire period, implying only a very small income effect for contributors above the old threshold. The shift in contribution rates from the old to the new match threshold may also reflect an "anchoring effect" of the match threshold. Specifically,
the match threshold serves as a salient starting point in the decision of which contribution rate to select. Numerous studies have shown that final decisions tend to be anchored by such starting points (Kabneman and Tversky, 1974).
92
Choi, Laibson, Madrian & Metrick
FIGURE 4. The Evolution of the 401(k) Contribution Rate Distribution over Time (Company E) 50%
40% C.)
0% Mar-96
Oct-96
May-97
Jul-98
Dec-97
Feb-99
Sep-99
Time 1-4%
5-6%
7-8%
9-10%
11-15%
16-25%
The second company that we consider is Company F, which introduced a 25-percent match on contributions up to 4 percent of income on October 1, 2000. We suspect that this was adopted as a response to the fact that at year-end 1999, only 34 percent of its active employees had ever participated in its 401(k) plan.24 Communication about the change
started at the beginning of July 2000. Prior to this date, there was no employer match offered in the plan.25 Our data include cross sections of all active employees at Company F
at year-end 1998, 1999, and 2000. These data contain information on participation status, original enrollment date, effective year-end contribution rate, original hire date, and demographics. We exclude all employees hired before July 1, 1998, because on that date the company eliminated a one-year length-of-service requirement for 401(k) eligibility. To assess the impact of the employer match on 401(k) participation,
we again estimate a Cox proportional-hazard model of time from hire until the date of initial participation in the 401(k) plan. As with company 24 We should note that Company F has a primary defined-benefit pension plan for its employees. The company did have three acquired divisions that had employer matches previously and were not affected by this change. These divisions, as well as three divisions that were acquired after 1998, are excluded from our analysis.
Defined Contribution Pensions 93
FIGURE 5. Employer Matching and 401(k) Participation (Company F) 30% S
25% 20%
5% 0% 0
2
4
6
10
12
14
16
18
20
22
24
26
Tenure (months) No match (predicted)
Match (predicted)
we control for gender and age, and we include a dummy variable that equals 1 after the match was announced to employees (July 2000). Results are presented in colun-m 2 of Table 5. We find that introducing the match has a positive and highly significant effect on participation, with a z statistic of 6.84. In order to assess the economic significance of
the results, we plot in Figure 5 the predicted participation rate by tenure for a hypothetical population of 40-year-old males. At three to four months of tenure, the model predicts a 10.9-percent participation rate when there is an employer match, which is 3.4 percentage points higher than would be the case without an employer match. Results at longer tenure levels are more speculative, because we don't actually observe employees with more than three months of tenure who have had the match in place since hire. Keeping this caveat in mind, we see that the model predicts 17.8-percent participation at one year after hire with an employer match (a 5.3-percentage-point increase) and 24.2-percent participation at two years of tenure (a 7.0-percentage-point increase). Although these numbers may seem small, note that this company had unusually low participation rates to start with. When compared against the baseline, the employer match appears to have increased 401(k) partici-
pation by over 40 percent. Furthermore, relative to the match structure in other 401(k) plans, this employer match is not particularly generous.26 26
The modal employer match is 50 percent of employee contributions up to 6 percent of compensation (Bureau of Labor Statistics, 1998).
94
Choi, Laibson, Madrian & Metrick
FIGURE 6. Employer Matching and the Distribution of 401(k) Contribution Rates (Company F) 2.0% w
1.5% E
0 0
1.0%
0.5% -
U.
0.0% 1%
2%
3%
4%
5%
6%
7-9%
10%
11-15%
Contribution Rate DHired Jul-Dec 1998,12131/98 rate Hired Jul-Dec 1999, 12/31/99 rate DHired Jul-Dec 2000, 12/31/00 rate
A higher match rate might be expected to have a larger effect on participation.27
The introduction of a match seems to have had a meaningful effect on the distribution of contribution rates as well. Figure 6 is a histogram of contribution rates by hire cohort at the end of the calendar year in which the cohort was hired.28 Before the employer match, the most frequently chosen contribution rates of plan participants are 5, 10, and 15 percent (which is lumped together with the 11- to 14-percent rates in the graph). After the employer match, we see a large increase in the fraction of employees with a 4-percent contribution rate (the new match threshold) relative to previous cohorts with the same tenure at the company. This is consistent with our previous observation that the match threshold may serve as a powerful focal point in employees' choice of a contribution rate. In sum, our limited evidence suggests that employer matching does have a significant impact on both 401(k) participation and contribution rates. Company F demonstrates that implementing an employer match However, Bassett, Fleming, and Rodrigues (1998) conclude that the mere presence of a match increases participation, with no marginal effect from increasing the match rate. We cannot test this hypothesis with our data. While the distribution of employees at the various contribution rates is based on the full sample of employees, not just plan participants, we have excluded the non-contributors from the graph, because they constitute over 90 percent of the sample, and including them makes variation in contribution rates across the contributing population difficult to see. 27
Defined Contribution Pensions
95
can increase 401(k) participation. Company E demonstrates that increas-
ing the match threshold can increase 401(k) contribution rates. Both Company E and Company F show that the match threshold has an important effect on the distribution of 401(k) contribution rates, with many participants clustering at the threshold.
3.5 Eligibility Another common 401(k)-plan feature is a waiting period before employees become eligible to participate in the plan. Employers adopt eligibility requirements for a variety of reasons, including the fixed costs of administering accounts for newly hired workers with high turnover rates, and the possibility that low participation rates of newly hired employees wifi
adversely affect an employer's non-discrimination testing. This latter explanation, however, is less relevant as recent legislative changes have made it easier for companies to institute shorter length-of-service requirements for 401(k) participation without substantially increasing the company's risk of failing non-discrimination tests.
Earlier eligibility is valuable for employees, since a shorter waiting period increases their tax-deferred savings opportunities. The extent of this benefit, however, depends on how waiting periods affect the participation profilethe relation between 401(k) participation and tenure. For example, waiting periods may merely truncate the participation profile, so that upon eligibility, employee participation quicidy catches up to the participation rate that would arise without a waiting period. Alter-
natively, waiting periods may shift the participation profile, so that employees who face a waiting period have permanently lower 401(k) participation rates than those who do not. In this subsection, we examine the effect of eligibility requirements on
401(k) participation in two companies that eliminated their eligibility requirements. Both Company F and Company G went from a one-year eligibility period to immediate eligibilityCompany F on July 1, 1998, and Company G on January 1, 1997.29 To illustrate the impact of waiting periods on 401(k) participation, we plot in Figure 7a and b the 401(k) participation profiles of employees who
faced either a one-year or no eligibility requirement. For Company F (Figure 7a), the two groups are employees hired between July 1, 1996 and July 1, 1997 with a full one-year waiting period, and employees Company C also subsequently changed the windows in which participants could enroll in the plan. Prior to September 1, 1997, participants could enroll only once a year. Beginning on November 22, 1997, however, new enrollments were allowed on a daily basis. To the extent that these deadline changes affect the time path of participation, Company G's results could be biased.
96
Choi, Laibson, Madrian & Metrick
FIGURE 7. Waiting Periods and 401(k) Participation: (a) Company F; (b) Company G
0
6
3
12
9
21
18
15
24
Tenure (months) EIigibIe immediately
EligibIe after 1 year
90%
-
80% (U
o
60%
0. 50% 40% 30% (U
20%
I-
U. 10% 0% 0
6
24
18
12
30
36
Tenure (months) EIigibIe after 1 year
4EIigible immediately
hired between July 1, 1998 and December 31, 2000, who faced no waiting period. For Company G (Figure 7b), the two groups are employees hired between January 1, 1995 and January 1, 1996 with a full one-year waiting period, and employees hired between January 1, 1997 and December 31, 1999 with no waiting period.
Defined Contribution Pensions
97
At both companies, the employees with a one-year waiting period do not immediately attain the 401(k) participation achieved at equal tenure by employees with shorter waiting periods, but this gap closes fairly quickly over time. If we assume that the participation series are drawn independently, the differences between the two groups are no longer statistically significant at 18 months of tenure in Company F and at 22 months of tenure in Company G. Another way to look at these participation profiles is to consider participation rates by the time since 401(k) eligibility. Doing so, we see that, conditional on time since becoming eligible, employees with a one-year eligibility requirement actually have a higher 401(k) participation rate than employees who were immediately eligible. The difference in participation rates is between 2.5 and 4.6 percentage points for Company F and is always significant at the 1-percent level for the first twelve months after eligibility. At Company G the difference is approximately 7 percentage points and is almost always significant at the 1-percent level for the first 24 months after eligibility. These findings are inconsistent with the notion that eligibility requirements simply shift the 401(k) participation profile without affecting its shape.
Overall, the evidence from these two companies suggests that the 401(k) participation rates of employees who face eligibility requirements catch up fairly quickly (within a matter of months) to levels that would occur without waiting periods. While this is certainly better for retirement wealth accumulation than would be the case if eligibility requirements resulted in permanently lower 401(k) participation rates, we do
not take this as evidence to suggest that waiting periods are "not that bad." Nobody seems to lose when shorter waiting periods are adopted, so we see no reason why companies should not be encouraged to allow immediate eligibility for participation 401(k) savings plans.
3.6 Asset Allocation Choices The bulk of this paper is focused on the 401(k) participation and contribution decisions of employees. If we are concerned about savings adequacy at retirement, the questions of whether to participate in a savings program and how much to save conditional on participation are of primary impor-
tance. After these two questions have been answered, the next most important question is how to allocate savings among different asset classes. A
small but growing literature has addressed these questions in recent years; not surprisingly, many of the same behavioral issues present in the participation and contribution decisions also play a role in participants' asset allocation choices. As discussed earlier, Madrian and Shea (2001a) and Choi, Laibson, Madrian, and Metrick (2001) show that automatic
98
Choi, Liibson, Madrian & Metrick
enrollment results in many participants remaining at the employer-
specified default for both the contribution rate and asset allocation. Such passive decisionmaking in asset allocation choices is also present in many other guises. In a series of papers, Shlomo Benartzi and Richard Thaler demonstrate several related behavioral regularities in asset allocation decisions. Benartzi and Thaler (2001a) study the relationship between the menu of investment choices and the eventual pattern of asset holdings across different classes. They suggest that participants use naive diversification strategies that are heavily influenced by the menu offered by their plan; a plan sponsor that offers ten equity options and five non-equity options may be subtly influencing its employees to put two-thirds of their money into equities. Using a database of 170 retirement savings plans, Benartzi and Thaler (2001a) find that approximately 62 percent of the funds offered in these plans are equity investments; the fraction of total assets held in equities by the participants in these 170 plans is remarkably close to 62 percent as well. Furthermore, they find a
positive relationship at the plan level between the fraction of equity funds offered by the plan and the fraction of individual portfolios invested in equities. These findings are further reinforced by experimental data and by evidence on individual decisions made by TWA pilots in their corporate plan.
In another study, Benartzi and Thaler (2001c) gave participants a choice between the distribution of retirement outcomes implied by the actual asset allocation in their 401(k) plan and the distribution implied by the average allocation among all participants in the same plan. Most participants preferred the average distribution to the one based on their own allocation. Since most participants have portfolios that are, almost by definition, more extreme than the average allocation, Benartzi and Thaler characterize this result as an example of an aversion to "extremeness." Such results call into question whether most participants are choosing an allocation that could be called optimal in an economic framework. Perhaps the most disturbing aspect of 401(k) participants' asset allocation choices is the large fraction of balances invested in employer stock. About half of all 401(k) plans (by assets) offer participants the opportunity to invest in company stock. Some plans even require that all matching contributions be held in company stock, at least for some period of time. Because this asset class is both very volatile (since it consists of only a single stock) and highly correlated with the labor earnings of employees, holding company stock is certainly a poor diversification strategy for participants. Nevertheless, a significant fraction of plan assets are held in
Defined Contribution Pensions
99
company stock. For firms that offer company stock in their plans, Holden, VanDerHei, and Quick (2000) find that about 33 percent of plan assets are held in this asset class. Among all firms, including those that do not offer company stock, this fraction is 18 percent. While this level of holdings itself seems high, the manner in which participants decide to invest in company stock is also troubling. Benartzi (2001) finds that current contributions to company stock are heavily influenced by the returns earned by that stock over the preceding ten years. It seems that naive diversification is combined with naive extrapolation of past returns and an apparent lack of concern for the risk consequences of company-stock investment. Indeed, a first-order improvement in diversification could be gained by the simple elimination of company stock from 401(k) plans. Interestingly, ERISA restricts the investments of defined benefit pen-
sion plans in the stock or real estate of the employer to 10% of total assets. 401(k) plans, however, are exempt from this rule. The recent collapse of Enron has publicly highlighted the diversification danger of company stock in 401(k) plans. With several bills now pending before Congress to address the diversification problems created by 401(k) investments in company stock, the current policy question has changed
from one of whether restrictions on 401(k) investments in company stock are warranted to what type of restrictions are warranted (and are politically feasible). While Enron is at the center of the public controversy,
class-action lawsuits have been filed by 401(k) participants at several companies in the wake of dramatic declines in the value of company stock. Even absent legislative reforms, these lawsuits are prompting many other companies to reconsider the emphasis on company stock in their 401(k) plans. While some companies are reevaluating whether they should match in company stock, others are addressing whether to eliminate company stock as an investment option altogether.
3.7 Financial Education at the Workplace Recognizing that many employees are ifi equipped to make wellinformed retirement savings decisions, particularly with respect to asset allocation, many employers have turned to various forms of financialeducation provision to help their employees meet the challenges of planning for an economically secure retirement. These efforts, which vary widely across employers, run the gamut and include paycheck stuffers, newsletters, summary plan descriptions, seminars, individual consultations with financial planners, and, more recently, access to Internetbased education and planning tools.
100
Choi, Laibson, Madrian & Metrick
The previous literature on the effects of financial education on saving behavior has found rather consistent evidence that financial education increases savings, although the inadequacy of the data in many of these studies makes their conclusions somewhat speculative. There are two broad strands in the literature. The first is case studies of the impact of financial education at specific companies or organizations. These studies typically evaluate the effect of a particular financial education initiative, often financial education seminars, on either saving behavior or measures of financial well-being (Kratzer et al., 1998; HR Focus, 2000; DeVaney et al., 1995; McCarthy and McWhirter, 2000; Jacobius, 2000). While all of these studies conclude that financial education motivates improvements in saving behavior, these conclusions are often based on dramatic changes in what participants plan to do with respect to retirement saving without actually verifying that the prophesied changes eventually do take place. Unfortunately, a growing body of both theoreti-
cal and empirical evidence, including the survey results reported in section 2 of this paper, suggests that despite the best intentions of employees, retirement saving is one area in which individuals excel at delay (Madrian and Shea, 2001a; O'Donoghue and Rabin, 1998; Diamond and Koszegi, 2000; Laibson, Repetto, and Tobacman, 1998). Thus, measures of intended behavior are likely to dramatically overstate the actual effects of financial education.
The second broad category of analyses in the previous literature on financial education has utilized cross-sectional surveys of individuals from across the population, not just from a single company or organization (Bernheim and Garrett, 1996; Bernheim, Garrett, and Maki 1997; Milne, Vanderei, and Yakaboski, 1995), or data from surveys of multiple employers (Bayer, Bernheim, and Scholz, 1996; Milne, Vanderei, and Yakaboski, 1996; Murray, 1999). This category of studies has the advantages of applying to a general population and utilizing actual saving choices instead of saving intentions. However, the cross-sectional data sets also pose numerous problems. Their greatest drawback is that financial-education provision and/or utilization may be correlated with other factors that have a strong influence on saving behavior across individuals or organizations (e.g. the structure of the 401(k) plan, the availability of other types of savings and/or pension programs, the level and structure of employee compensation, the corporate culture). To the extent that these confounding factors are not completely observed and controlled for, the measured effects of financial education could be quite biased. The definition of what constitutes "financial education" is also subject to interpretation and is likely to vary from one respondent to another.
Defined Contribution Pensions
101
The household surveys have the additional disadvantage that survey answers to questions about financial education are likely to be subject to recall bias. This could result, for example, if individuals who participate in and benefit the most from employer-sponsored savings programs find financial education more salient and are thus more likely to remember
that such programs were offered. This type of non-random measurement error in the availability of financial education wifi lead to estimates
of the effects of financial education that are too large. The employerbased surveys have the additional disadvantage that response rates tend to be quite low, and it is unlikely that the non-response is random. Moreover, it is almost impossible to determine how the selection of the firms into the sample is likely to affect the results. A recent study by Madrian and Shea (2001b) examines the impact of financial education seminars on saving behavior in Company C, one of the companies discussed in section 3.1. Company C enlisted a financialeducation provider to give one-hour seminars at its various locations throughout the country during 2000. The curriculum at these seminars was general and covered topics directly related to retirement savings, such as setting savings goals to meet retirement income targets and the fundamentals of investing (asset classes, risk, diversification, etc.), in addition to more general financial issues such as managing credit and debt and using insurance to minimize exposure to financial risks. The financial-education data from this company are unique in that seminar attendance was tracked in a way that made it possible to match seminar attendance to administrative data on both previous and subsequent saving behavior. We have data on the individuals who attended financial-education seminars between January 1 and June 30, 2000, and
on the 401(k) saving choices of all employees at this company on December 31, 1999, before any of the seminars were offered, and on June 30, 2000, by which time the seminars had been offered at 42 different loca-
tions. One-third of the employees at the company work at these 42
locations, and about 17 percent of employees at these locations attended the financial-education seminars. Table 6 presents some very basic statistics on the planned changes in saving behavior that attendees of the financial education seminars reported, along with the actual changes in saving behavior that were made subsequent to the seminars. The statistics in Table 6 paint a somewhat more muted picture of the impact of financial education on saving behavior than has been estimated in the previous literature. In an evaluation of the financial-education seminars given to attendees at the conclusion of the seminar, attendees were asked, "After attending today's presenta-
tion, what, if any, action do you plan on taking toward your personal
102
Choi, Laibson, Madrian & Metrick
TABLE 6.
Financial Education and Actual vs. Planned Savings Changes (Company C) Seminar attendees
Planned action Non-participants: Enroll in 401(k) plan 401(k) participants: Increase contribution rate Change fund selection Change fund allocation
Non-attendees
Actual change
Planned change
Actual change
100%
14%
7%
28% 47% 36%
8% 15% 10%
5% 10% 6%
The sample is active 401(k)-eligible employees at company locations that offered financial-education seminars from January to June 2000. Actual changes in saving behavior are measured over the period
from December 31, 1999 through June 30, 2000. Planned changes are those reported by seminar attendees in an evaluation of the financial-education seminars at the conclusion of the seminar. The planned changes from survey responses of attendees have been scaled to reflect the 401(k) participation rate of seminar attendees.
financial affairs?" followed by a list of choices (with multiple responses
allowed). 71 percent of those attending the seminars filled out and turned in these evaluation forms.3° Of those who filled out the evaluation, 12 percent reported that they intended to start contributing to the 401(k) savings plan. But 88 percent of seminar attendees were already participating in the 401(k) plan, so virtually all of the non-participating seminar attendees planned to enroll in the 401(k) plan. By June 30, 2000, however, only 14 percent of the non-participating seminar attendees had actually joined the plan, and some of these would likely have enrolled in the 401(k) plan without the availability of a financial-education seminar (as did 7 percent of the employees who did not attend the seminars).
Of those seminar attendees who were already participating in the
plan, 28 percent reported plans to increase their 401(k) contribution rate,
41 percent reported plans to make changes in the selection of their
investment choices within the 401(k) plan, and 36 percent reported plans
to change the fraction of their money allocated to the various 401(k)
investment choices. By June 30, 2000, however, only 8 percent of 401(k) ° The evaluation responses that we have are from all locations offering financial-education seminars during 2000, not just those offering the seminars during the JanuaryJune 2000 period for which we have savings data. Unfortunately, we do not have the evaluation responses on an individual basis, only the aggregated responses for all attendees. Thus, we cannot ascertain on an individual basis how many seminar attendees actually followed through on the planned behaviors listed on the evaluation form.
Defined Contribution Pensions
103
participants attending the seminars had increased their contribution rate, 15 percent had made changes to their investment choices, and 10 percent had changed their fund allocations. While the fraction of seminar attendees making such changes is slightly higher than the fraction of non-seminar-attendees, it is substantially below what the attendees reported they planned on doing. One could certainly argue that the low rate of actual changes relative to planned changes results from the fact that the data used to observe the plan changes are, for employees at some locations, for times not long after the actual financial-education seminars. However, there is relatively little correlation between the fraction of seminar attendees making changes to their 401(k) saving behavior
and the length of time between their seminar and June 30, 2000. It appears that seminar attendees make changes either almost immediately or not at all. Madrian and Shea (2001b) draw similar conclusions when they try to control for differences in the underlying saving propensities of employees who do and do not attend financial-education seminars. Their final assessment is that financial education increases savings-plan participation and results in greater portfolio diversification, particularly among employees hired under automatic enrollment, but the estimated magni-
tudes are not particularly large. Overall, while financial education is important, it does not appear to be a powerful mechanism for encouraging 401(k) retirement savings.
4. CONCLUSIONS The evidence discussed above provides an incomplete sketch of the retire-
ment preparation process. Our analysis only covers 401(k) savings and necessarily misses other important types of wealth like home equity, IRAs, and defined benefit pensions. However, even our incomplete evidence provides intriguing hints about the economic and psychological forces that drive financial planning. Most of our evidence highlights the importance of passive decisionmaking. For better or for worse, many households appear to passively accept the status quo. For example, in companies without automatic enrollment, the typical employee takes over a year to enroll in his or her company-sponsored 401(k) retirement plan. In companies with automatic enrollment, employees overwhelmingly accept the automaticenrollment defaults, including default savings rates and default funds. For terminated employees, the key determinant of whether they consume or save their 401(k) balances is whether that balance is above or below the automatic cash distribution threshold of $5,000. Many plan
104
Choi, Laibson, Madrian & Metrick
participants allow the menu of investment funds to drive their asset allocation decisions. Most employees feel that they save too little, and many plan to raise their contribution rate in the near future, but few act on these good intentions. By contrast, employees do succeed in raising their contribution rates if they are given a low-effort opportunity to sign up for an automatic schedule of increases in their contribution rate.
All of these examples have a common theme: employees often take the path of least resistance. As a result, employers have a large measure of control over the savings choices that their employees make. Most savings plans an employer creates wifi advantage certain passive or nearly passive choices over other active choices. Sophisticated employers should choose their plan defaults carefully, since these defaults will strongly influence the retirement preparation of their employees. Policymakers should also recognize the role of defaults, since policymakers can facilitate, with laws and regulations, the socially optimal use of defaults. For example, default contributions to company stock may
lead to insufficient diversification. Policymakers could legally cap default investments in such problematic asset categories. Likelihood, policymak-
ers could facilitate default contributions to more appropriate investments, like the S&P 500, by giving corporations legal protections for picking such risky but highly diversified default funds. It is easy to identify dozens of ways that thoughtful regulations can influence passive decisionmakers without encroaching on the freedom of active decisionmakers to opt out of the defaults and choose in their own (perceived) best interest. However, regulating defaults is a two-
edged sword. If one has confidence in the government, then such regulations wifi serve the common good. If one does not have such confidence, then regulating defaults will open up one more avenue for the misuse of governmental power. Our analysis demonstrates that defaults matter, but our evidence does not reveal who should control them.
APPENDIX A: DATA This appendix describes the data for each of the companies analyzed in this paper:
Company A. (1) Cross-sectional survey data from January 2001 for a random sample of employees; (2) longitudinal 401(k) savings data from January 1996 through April 2001 for all 401(k) participants. Company B. Cross-sectional 401(k) savings data from December 31 of 1998, 1999, and 2000 for all active employees (both 401(k) participants and non-participants) and non-employee 401(k)-plan participants.
Defined Contribution Pensions
105
Company C. (1) Cross-sectional 401(k) savings data from June 1, 1997; December 31, 1997; June 30, 1998; December 30, 1998; March 31, 1999; June 30, 1999; September 30, 1999; December 31, 1999; March 31, 2000; and June 30, 2000 for all active employees; (2) financial-education seminar attendees from January 1, 2000 through June 30, 2000. Company D. Cross-sectional 401(k) savings data from December 31 of 1998 and 1999 for all 401(k)-plan participants (employee and nonemployee), and from December 31, 2000 for all active employees (both
401(k) participants and non-participants) and non-employee 401(k)plan participants. Company E. (1) Cross-sectional 401(k) savings data from December 31 of 1998, 1999, and 2000 for all active employees (both 401(k) participants and non-participants) and non-employee 401(k)-plan participants; (2) longitudinal 401(k) savings data from March 1996 through March 2000. Company F. Cross-sectional 401(k) savings data from December 31 of 1998, 1999, and 2000 for all active employees (both 401(k) participants and non-participants) and non-employee 401(k)-plan participants.
Company G. Cross-sectional 401(k) savings data from December 31, 1999 for all active employees (both 401(k) participants and nonparticipants) and non-employee 401(k)-plan participants. The cross-sectional data available for these various companies include basic demographic information (age, hire date, gender, income), as well as point-in-time information on 401(k) saving such as participation status, contribution rate, account balances, and asset allocation. The longitudinal data include daily information on the 401(k) contribution rate, account balances, and asset allocation of 401(k)-plan participants. They do not include demographic information or information on non-participating employees.
APPENDIX B: 401(K)-PLAN PARTICIPANT SATISFACTION SURVEY QUESTIONS
Section I 1. Which of the following statements describes your current participation in the XXX Company, Inc. 401(k) Plan?
106
Choi, Laibson, Madrian & Metrick
D I am currently contributing to the plan D I am not currently contributing to the plan, but I have previously contributed to the plan D I am not currently contributing to the plan, and I have never contributed to the plan
2. For each of the following questions, please indicate how strongly you agree or disagree with respect to the XXX Company, Inc. 401(k) plan. To indicate your level of agreement, please use the following scale (if you have no experience with a given item, please respond with "have no opinion"). Strongly agree Somewhat agree Neither agree nor disagree Somewhat disagree Strongly disagree Have no opinion
I have a good understanding of the 401(k) savings plan overall I have a good understanding of the 401(k) savings plan investment fund choices I think the 401(k) plan meets my needs The )OO( Company, Inc. 401(k) plan is better than plans offered by other companies
3. For each of the following questions, please indicate how satisfied you are with that aspect of the XXX Company, Inc. 401(k) plan. To indicate your level of satisfaction, please use the following scale (if you have no experience with a given item, please respond with "have no opinion"). Very satisfied Somewhat satisfied Neither satisfied nor dissatisfied Somewhat dissatisfied Very dissatisfied Have no opinion
Convenience of payroll deductions for savings Number of investment options Variety of investment options
Defined Contribution Pensions
107
Account Statements Internet access to your 401(k) plan Loans
Please use the space provided to fill in your response to the following question:
4. What, if anything, could your company do differently in terms of the XXX Company, Inc. 401(k) plan that would increase your satisfac-
tion level, relating to any of the items listed above?
Section II Please check the appropriate box for each of the following questions: 5. How would you describe yourself as an Internet user? E Very experienced E Somewhat experienced E Not too experienced E Not at all experienced
6. Do you have access to the Internet at home? L Yes
LNo 7. How would you describe your level of financial knowledge? LI Very knowledgeable LI Somewhat knowledgeable LI Not too knowledgeable LI Not at all knowledgeable
8. Which of the following best describes your job? LI Management
108
Choi, Laibson, Madrian & Metrick
U Other salaried position U Hourly
U Other
Which of the following best describes your level of education? U High school or less U Some college U College graduate U Graduate school
Section III These next few questions discuss retirement savings. Please check the appropriate box(es) for each of the following questions, and/or fill in the blanks, as appropriate: First, based on anything you may have heard or read, what percent of your income do you think you should ideally be saving for retirement? U 5 percent of income or less U Between 5 percent and 9 percent of income U Between 10 percent and 14 percent of income U Between 15 percent and 19 percent of income U Between 20 percent and 24 percent of income U At least 25 percent of income
Think about how much you are actually currently saving for retirement. Compare your actual saving rate to your ideal saving rate. Right now, your actual retirement saving rate is: U Far too low U A little too low U About right U A little too high U Far too high
IF YOU ARE CURRENTLY CONTRIBUTING TO YOUR COMPANY 401(K) PLAN, PLEASE ANSWER QUESTIONS 12 THROUGH 17.
Defined Contribution Pensions
109
IF YOU ARE NOT CURRENTLY CONTRIBUTING TO YOUR COMPANY 401(K) PLAN, PLEASE SKIP TO QUESTION 18.
Are you contributing currently at the maximum 401(k) savings rate? Yes
LINo Which of the following statements best describes your 401(k) contribution plans over the next few months? LI I plan to raise my contribution rate. LI I plan to lower my contribution rate. LI I don't plan to make any changes.
IF YOU ARE NOT PLANNING TO MAKE ANY CONTRIBUTION CHANGES, PLEASE SKIP TO QUESTION 15.
What percent of your salary are you planning to contribute? Which of the following statements best describes your 401(k) fund allocation plans over the next few months? LI I am considering selecting different funds. LI I am considering rebalancing among the funds I currently have. LI I am not planning to make any changes in regard to my fund allocations. LI I am considering both selecting different funds and rebalancing among the funds I currently have
When do you next plan to make changes in your 401(k) plan? LI In the next few days LI In the next week LI In the next two weeks LI In the next three weeks LI Sometime in the next month LI Sometime in the next two months
LI Other
What company resources will you use to make changes to your 401(k) plan? Check all that apply. LI Speak to benefit center representative
110
Choi, Lciibson, Madricin & Metrick
El Use the 401(k) web site El Consult the new hire kit (given to all new employees) El Other: Please specify
IF YOU ARE CURRENTLY CONTRIBUTING TO YOUR COMPANY 401(K) PLAN, PLEASE SKIP TO QUESTION 21. IF YOU ARE NOT CURRENTLY CONTRIBUTING TO YOUR COMPANY 401(K) PLAN, PLEASE ANSWER QUESTIONS 18 TO 20.
When you enroll/re-enroll in the XXX Company, Inc. 401(k) plan, what percent of your salary do you expect to contribute to the plan? El Between 0 percent and 3 percent of income El Between 4 percent and 6 percent of income El Between 7 percent and 9 percent of income El Between 10 percent and 12 percent of income El Between 13 percent and 15 percent of income
When do you plan to enroll/re-enroll in the 401(k) plan? El In the next few days El In the next week El In the next two weeks El In the next three weeks El Sometime in the next month El Sometime in the next two months El Other
What company resources will you use to enroll in the 401(k) plan? Check all that apply. El Speak to benefits center representative El Use the 401(k) web site El Consult the new hire kit (given to all new employees) El Other. Please specify
Thank you for your participation in this survey. For more information on the )OO( Company, Inc. 401(k) plan, click here: URL.
Defined Contribution Pensions
111
REFERENCES Andrews, Emily S. (1992). "The Growth and Distribution of 401(k) Plans." In Trends in Pensions 1992, John Turner and Daniel Belier (eds.), Washington: U.S. Department of Labor, Pension and Welfare Benefits Administration. Bassett, Wffliam F , Michael J. Fleming, and Anthony P. Rodrigues (1998). "How
Workers Use 401(k) Plans: The Participation, Contribution and Withdrawal Decisions." National Tax Journal 51(no. 2):263-289. Bayer, Patrick J., B. Douglas Bernheim, and J. Karl Scholz (1996). "The Effects of Financial Education in the Workplace: Evidence from a Survey of Employers." NBER Working Paper no. 5655. Benartzi, Shlomo (2001). "Excessive Extrapolation and the Allocation of 401(k) Accounts to Company Stock." Journal of Finance 56(no. 5):1747-1764. Benartzi, Shlomo, and Richard Thaler (2001a). "Naive Diversification Strategies
in Defined Contribution Savings Plans." American Economic Review 91(no. 1):79-98.
Benartzi, Shlomo, and Richard Thaler (2001b). "Save More Tomorrow: Using Behavioral Economics to Increase Employee Saving." University of CaliforniaLos Angeles. Working Paper. Benartzi, Shiomo, and Richard Thaler (2001c). "How Much is Investor Autonomy Worth?" University of California-Los Angeles. Working Paper. Bernheim, B. Douglas (1995). "Do Households Appreciate Their Financial Vulnerabiiities? An Analysis of Actions, Perceptions, and Public Policy." In Tax Policy for Economic Growth in the 1990s. Washington: American Council for
Capital Formation. Bernheim, B. Douglas, and Daniel M. Garret (1996). "The Determinants and Consequences of Financial Education in the Workplace: Evidence from a Survey of Households." NBER Working Paper no. 5667. Bernheim, B. Douglas, Daniel M. Garrett, and Dean M. Main (2001). "Education and Saving: The Long-Term Effects of High School Financial Curriculum Mandates." Journal of Public Economics 80(3), 435-465. Bureau of Labor Statistics (1998). Employee Benefits in Medium and Large Private Establishments. Washington: U.S. Department of Labor, Bureau of Labor Statistics.
Choi, James, David Laibson, Brigitte Madrian, and Andrew Metrick, (2001). "For Better or for Worse: Default Effects and 401(k) Savings Behavior." NBER Working Paper no. 8651.
DeVaney, Sharon A., et al. (1995). "Saving and Investing for Retirement: The Effect of a Financial Education Program." Family Economics and Resource Management Biennial, 153-158.
Diamond, Peter, and Botand Koszegi (2000). "Quasi-Hyperbolic Discounting and Retirement." Department of Economics, Massachusetts Institute of Technology. Working Paper no. 00-03. Engen, Eric M., William G. Gale, and John Karl Schoiz (1994). "Do Saving Incentives Work?" Brookings Papers on Economic Activity 1994 (no. 1):85-180.
Engen, Eric M., William G. Gale, and John Karl Scholz (1996). "The Illusory Effects of Saving Incentives on Saving." Journal of Economic Perspectives 10(no. 4) : 113-138.
Even, William E ,and David A. Macpherson (1997). "Factors Influencing Partici-
pation and Contribution Levels in 401(k) Plans." Florida State University. Working Paper.
112
Choi, Laibson, Madrian & Metrick
Farkas, Steve, and Jean Johnson (1997). Miles to Go: A Status Report on Americans' Plans for Retirement. New York: Public Agenda. Fidelity Investments (2001). Building Futures: A Report on Corporate Defined Contribution Plans, Volume II. Boston: Fidelity Investments. Hewitt Associates (2001). Trends and Experience in 401(k) Plans. Lincoinshire, IL: Hewitt Associates.
Holden, Sarah, Jack VanDerHei, and Carol Quick (2000). "401(k) Plan Asset Allocation, Account Balances, and Loan Activity in 1998." Investment Company Institute Perspective 6(no. 1). HR Focus (2000). "What is the Value of Financial Education Workshops?" HR Focus, February, p. 15. Internal Revenue Service (1998). Internal Revenue Bulletin 98-25 (June 22), p. 8. http:/ /ftp.fedworld.gov/pub/irs-irbs/irb98-25.Pdf. Internal Revenue Service (2000a). Internal Revenue Bulletin 2000-7 (February 14), p. 617. http:/ /ftp.fedworld.gov/pub/irs-irbs/irb00-7.pdf. Internal Revenue Service (2000b). Internal Revenue Bulletin 2000-31 (July 31), pp.
138-142. http://ftp.fedworld.gov/pub/irs-irbs/irb003l.pdf.
Investment Company Institute (2000). 401(k) Plan Participants: Characteristics, Con-
tributions, and Account Activity. Washington: Investment Company Institute. Jacobius, Arleen (2000). "Top-Notch Education Campaigns Honored." Pensions & Investments 28(3):29.
Kahneman, Daniel, and Amos Tversky (1974). "Judgment under Uncertainty: Heuristics and Biases." Science 185 (4157):1124-1131.
Kratzer, Constance Y., et al. (1998). "Financial Education in the Workplace:
Results of a Research Study." Journal of Compensation and Benefits 14(3):24-27. Kusko, Andrea, James Poterba, and David Wilcox (1998). "Employee Decisions with Respect to 401(k) Plans." In Living with Defined Contribution Pensions: Remaking Responsibility for Retirement, Olivia Mitchell and Sylvester Schieber (eds.). Philadelphia: University of Pennsylvania Press. Laibson, David I., Andrea Repetto, and Jeremy Tobacman (1998). "Self-Control and Saving for Retirement." Brookings Papers on Economic Activity 1998(no. 1):91-196.
Madrian, Brigitte C., and Dennis F. Shea (2001a). "The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior." Quarterly Journal of Economics. 116(4), 1149-1187.
Madrian, Brigitte C., and Dennis F. Shea (2001b). "Preaching to the Converted and Converting Those Taught: Financial Education in the Workplace." University of Chicago. Working Paper. McCarthy, Mike, and Liz McWhirter (2000). "Are Employees Missing the Big Picture? Study Shows Need for Ongoing Financial Education." Benefits Quarterly 16(no. 1):25-31.
Milne, Deborah, Jack Vanderhei, and Paul Yakoboski (1995). "Can We Save Enough to Retire? Participant Education in Defined Contribution Plans," Washington: Employee Benefit Research Institute. EBRI Issue Brief no. 160. Milne, Deborah, Jack Vanderhei, and Paul Yakoboski (1996). "Participant Education: Actions and Outcomes." Washington: Employee Benefit Research Institute. EBRI Issue Brief no. 169. Murray, M. Christian (1999). "401(k) Plan Sponsors Find Education Pays." National Underwriter 103(no. 19):36-38.
Defined Contribution Pensions
113
O'Donoghue, Ted, and Matthew Rabin (1998). "Procrastination in Preparing for Retirement." University of CaliforniaBerkeley. Working Paper. Papke, Leslie E. (1995). "Participation in and Contributions to 401(k) Pension Plans." Journal of Human Resources 30(no. 2):311-325.
Papke, Leslie E., and James M. Poterba (1995). "Survey Evidence on Employer Match Rates and Employee Saving Behavior in 401(k) Plans." Economics Letters 49(3):313-317. Poterba, James M., Steven F. Venti, and David A. Wise (1996). "How Retirement Saving Programs Increase Saving." Journal of Economic Perspectives 10(no. 4):91112.
Poterba, James M., Steven F. Venti, and David A. Wise (1998a). "Lump Sum Distributions from Retirement Savings Plans: Receipt and Utilization." In Inqui-
ries in the Economics of Aging, David A. Wise (ed.). Chicago: University of Chicago Press. Poterba, James M., Steven F. Venti, and David A. Wise (1998b). "Personal Retirement Saving Programs and Asset Accumulation." In Frontiers in the Economics of Aging, David Wise (ed.). Chicago: University of Chicago Press. Profit Sharing/401(k) Council of America (2001). "Automatic Enrollment 2001: A Study of Automatic Enrollment Practices in 401(k) Plans." http://www.pcsa .org/data/autoenroll200l.asp, April 19. Vanguard Center for Retirement Research (2001). "Automatic Enrollment: Vanguard Client Experience." http://institutional.vanguard.com/pdf/automatic. enrollmenLclientexp.pdf, September 6. Watson Wyatt Worldwide (2001). "Retirement Plan Provisions: What, When and How Much? Economic Growth and Tax Relief Reconciliation Act of 2001." http://www.watsonwyatt.com/homepage/services/retirement/ resrender.asp?id=W-471&page= 1, November 5.
THINKING STRAIGHT ABOUT THE TAXATION OF ELECTRONIC COMMERCE: TAX PRINCIPLES, COMPLIANCE PROBLEMS, AND NEXUS Charles E. McLure, Jr. Hoover Institution, Stanford University
EXECUTIVE SUMMARY The Internet Tax Freedom Act (ITFA), which imposes a moratorium on
state and local taxes on Internet access and prohibits "multiple and discriminatory" state and local taxes on electronic commerce, has been extended until November 1, 2003, but the debate on whether and how to tax electronic commerce has not ended. This paper is intended to assist clear thinking about the taxation of electronic commerce. Under an economically neutral sales tax, all sales to consumers would be taxed, all sales
to business would be exempt, and sales by local merchants and by remote (out-of-state) vendors would be taxed equally. A compliancefriendly sales tax would exhibit substantial simplicity and interstate uniformity in the tax base, legal framework, and administrative procedures. Existing sales taxes exhibit none of these characteristics. Many sales to The author wishes to thank James Poterba and Walter Hellerstein for comments on an earlier draft.
116 McLure
consumers are exempt, many sales to business are taxed, and, because of the Supreme Court decision in Quill, which is based on the complexity of the system, many sales by remote vendors are not taxed. The system is extremely complex, in large part because there is essentially no unifor-
mity from state to state. Under the Streamlined Sales Tax Project the states have recently begun serious efforts to simplify their sales taxes, by making them more nearly uniform, in hopes of gaining Congressional or judicial reversal of Quill. These efforts would substantially simplify the
system, but would not achieve economic neutrality, as many sales to consumers would remain exempt and many sales to business would still be taxed. Moreover, differences in tax bases, legal frameworks, and administrative procedures would remain. Technology (lookup tables that categorize products as taxable or exempt in each state) might be able to handle differences in tax bases, but not those in legal structures and administrative procedures.
1. INTRODUCTION The Internet Tax Freedom Act (ITFA) imposes a moratorium on state and
local taxes on Internet access and prohibits "multiple and discriminatory" state and local taxes on electronic commerce. Initially enacted in 1998, the ITFA has been extended until November 1, 2003.1 Unlike the "naked" extension of the ITFA that was enacted, some proposals for extension would have significantly restricted the taxing power of state and local governments, while others would potentially have expanded that power. The temporary extension of the ITFA did not definitively resolve the choice between these options, which wifi continue to be debated until at least the end of October 2003. This paper is intended to assist clear thinking about the taxation of electronic commerce.2 The ITFA (which also created the Advisory Commission on Electronic Commerce, to be considered in section 5), was part of the Omnibus Appropriations bill. It expired briefly, its timely extension a victim of the terrorist attack on the World Trade Center and the anthrax scare that followed. The Act's pre-emptive effect is more symbolic than real, except in the case of taxes on Internet access, since (1) the Act has virtually no effect on sales taxes that do not discriminate against e-commerce and (2) decisions of the U.S. Supreme Court (to be considered in section 4) already effectively eliminate taxation of many e-commerce sales to consumers. The symbolism has at least three facets: the emergence of e-commerce as a politically favored sector, the victory of anti-tax forces, and Congressional intrusion into fiscal decisions of state and local governments. 1
2
Electronic commerce is "the use of computer networks to facilitate transactions involving
the production, distribution, and sale and delivery of goods and services in the marketplace." This definition, from Abrams and Doernberg (1997), is more useful than that in U.S. Treasury Department (1996, ¶3.2.1), the crucial part of which is". . . the exchange of goods or services . . . using electronic tools and techniques." The Treasury definition does not
Thinking Straight about the Taxation of Electronic Commerce
117
The current debate over the taxation of electronic commerce in the U.S. reflects basic defects in the way state and local governments tax sales and corporate income more than the inherent difficulty of taxing most electronic commerce. For the most part, the advent of electronic commerce has highlighted problems that were there all along; it did not create them. Most of these problemsbut not allstem from the lack of uniformity of the taxes imposed by the states (and the District of Columbia) and their numerous political subdivisions. Pointing to that lack of uniformity, the U.S. Supreme Court has said, most recently in the 1992 Quill decision (504 U.S. 298), that a state cannot require a remote (out-of-
state) vendor to collect its sales tax unless the vendor has a physical presence in the state.3 Since the states have not solved these problems cooperatively by increasing uniformity, despite having had ample opportunity and incentive to do so, some believe that it would be appropriate for the Congress to mandate a solution. Others warn that this is a step that should not be taken lightly, lest the fiscal sovereignty of the states be unduly compromised. They fear that Congress might legislate solutions to state problems where none are needed or impose inappropriate solutions; in any event, there would be unprecedented federal intervention in the fiscal affairs of state and local governments. Given recent state efforts to find a solution, they counsel a wait-and-see attitude, to allow the states time to simplify their systems. Some believe that the Congress should have signaled that it would relax the physical-presence rule if the states significantly simplify their sales tax systems. Section 2 describes how a state sales tax that is economically neutral
and relatively easy to comply with (hereafter the economically neutral and compliance-friendly system) would be structured, indicates how existing sales taxes deviate from that norm, and discusses the effects of those deviations. E-commerce, per se, plays a minor role in the discussion, which is equally applicable to sales by other types of remote vendors. It is perhaps best seen, alternatively, as a catalyst for longoverdue action by state and local governments to simplify their taxes or as a stalking horse for those who would undermine the taxing power of the states. clearly differentiate what is commonly known as electronic commerce from such activities as
telemarketing and television shopping, which are excluded by the AbramsDoernberg definition. This paper, does, however, exclude financial services from its purview. Strictly speaking, remote vendors that have nexus in a state are required to collect the use tax, which is legally imposed on the in-state purchaser's use of the purchased item, rather than the sales tax, which is imposed on in-state sales and cannot be collected on sales originating outside the state. Except where required for clarity, this distinction is generally ignored in favor of the generic term "sales tax."
118 McLure Section 3 summarizes arguments that have been made for and against exempting e-commerce from taxation and suggests that little revenue would be affected by the choice of whether to tax e-commerce. Section 4 describes key elements of recent proposals to accompany extension of the Internet Tax Freedom Act. Section 5 describes three prior attempts to find a solution to the problem of whether and how to tax e-commerce. Section 6 compares three approachesthe naked extension that was enacted and two alternativesto reforms required to achieve the economically neutral and compliance-friendly system. The Appendix discusses application of business activity taxes, which include corporate income taxes and franchise taxes measured by income, to remote commerce. It discusses design issues and describes defects of the current system. It is substantially less detailed than the discussion of sales and use taxes, in part because the latter is where most of the current debate has focused.
2. AN ECONOMICALLY NEUTRAL AND COMPLIANCE-FRIENDLY SALES TAX4 It is generally agreed that taxes should, to the extent possible and consistent with other goals, be economically neutral and compliance-friendly.5
2.1 A Neutral Sales Tax A state sales tax that was economically neutral would tax all consumption occurring in a state equally. A tax on all consumption is economically neutral in ways specified below. Moreover, to the extent private consumption and consumption of public services are correlated, such taxes can be seen as consistent with the benefit principle of taxation, which requires that taxes be related to benefits received from public services. Finally, taxing all consumption is simpler than taxing (or exempting) only selected products; this is explained further below. A state sales tax that satisfied this basic design objective would tax all consumption in the state; tax no sales to business; and tax sales by local and remote vendors equally. I have described this system and its rationale in greater detail in various publications, including McLure (1997), (1998a), (1998b), (2000a), (2000d), and (2001b).
Some may object that the rules of optimal taxation, rather than economic neutrality, should guide tax policy. Those rules, which call for differential taxation of products,
depending on the elasticities of demand and supply for the products, generally ignore the administrative difficulty of implementation, as well as the fact that a vast amount of information is required to put them into practice. See Slenirod (1990). On the other hand, not taxing business inputs, which is generally consistent with the conclusions of Diamond and Mirrlees (1971), is relatively easy to implement.
Thinking Straight about the Taxation of Electronic Commerce
119
Economists emphasize the neutrality of such a system, which would
not discriminate between types of consumption, would not distort choices of techniques of production and distribution, would treat local
and remote vendors the same, and would not distort the location of economic activity. The increased efficiency of markets that is generated by e-commerce arguably makes achievement of the conditions for economic neutrality even more important than before.
2.2 Administrative Tractability If the requirement that remote vendors collect tax is not to be onerous, and thus an unconstitutional burden on interstate commerce, the sales and use tax must be made "elegantly simple."6 Simplicity, in turn, requires substantial uniformity of the sales and use taxes of all states. Moreover, clear de minimis nexus rules and realistic vendor discounts (rebate of taxes intended to cover part of the costs of compliance) would eliminate or substantially reduce compliance burdens on remote vendors making only small amounts of sales into a state.
2.2.1 Interstate Uniformity A uniform tax base, a uniform legal structure (i.e., uniform statutes, regulations, and interpretations), and simplified administrative procedures (e.g., "one-stop" registration, filing, etc. for all states) would simplify compliance. Uniformity of the tax base has many dimensions, including uniform definitions of products; uniform classifications of products as taxable or exempt; uniform treatment of tax-exempt sales to business, to non-profit organizations, and to governments; uniform tax-exemption certificates; and conformity of state and local tax bases in a given state. Uniform tax rates would not be required; states would retain complete control over tax rates, which is the key to fiscal sovereignty and not a source of complexity, at least for state taxes. Under this system a vendor in San Jose, California (or in Tallahassee, Austin, or anywhere else), being familiar with the tax law of its own state, could easily calculate the tax due on sales to customers in any other state, simply by knowing three things: (1) Is it a taxable sale to a consumer or an exempt sale to business (or to a non-profit organization or a government)? (2) Where is the purchaser located? (3) What is the applicable tax rate? The vendor would not need to contend with interstate differences in such matters as the definitions of taxable and exempt products, exempt sales to business, exemption certificates for sales to 6
This is the term Governor Mike Leavitt of Utah used in his address to the Inaugural
Meeting of the Implementing States of the Simplified Sales Tax Project, held in Salt Lake City on November 28-29, 2001.
120 McLure
business, non-profits, and governments, or details of tax laws; by assumption these would be the same in all states. 2.2.2 The Simplicity of Taxing All Consumption and Only That Little has been said in the current debate about the need to tax all consumption and exempt all sales to business. (What has been said has usually focused on the political difficulty of taxing services and the reduction in
the tax base and thus revenuesand the need to increase ratesthat
would result from exempting sales to business.) Since these two reforms do not figure in the Quill decision, why emphasize them? Is it only a question of economic neutrality?
Consider the statement abovethat "a vendor .
.
.
, being familiar with
the tax law of its own state, could easily calculate the tax due on sales to customers in any other state. . . ." The condition stated in italics would be adequate to guarantee simplicity of compliance with the laws of other states only if the tax bases (and the tax laws) of all states were identical. There are, of course, an infinite number of ways to define the identical tax base. But none (at least none that make sense) are as simple as the
rule under the economically neutral and compliance-friendly system, which can be summarized in two rules.7 First, if a product is sold to a household, tax it. Second, if it is sold to a business, a government, or a tax-exempt organization, exempt it. Certainly this approach is far simpler than basing the tax of each state on uniform definitions of products and of types of sales that could be either taxed or exempt8 or even on a uniform base that differed from that under the economically neutral and compliance-friendly system.
2.2.3 Nexus and Vendor Discounts Remote vendors that make only small amounts of taxable sales into a state could face a significant burden if required to collect the state's use tax. This burden could easily be avoided by exempting vendors from the duty to collect tax in states where they do not have a substantial physical presence and their taxable sales fall below a de minimis level.9 (Stated differently, states could assert nexus only if a remote vendor had either a substantial physical presence or non-de-minimis taxable sales in the state.) This dual nexus rule for remote vendors could be supplemented by realistic vendor disIt would, of course, be simpler to tax all sales, whether to businesses or to others. Such turnover taxes have long been known to create unacceptable inequities and distortions. 8
This essentially describes the approach of the Simplified Sales Tax Project, to be de-
scribed below. States where such vendors are located could be allowed to tax such sales.
Thinking Straight about the Taxation of Electronic Commerce 121
counts that compensate for the costs of tax collection, which are disproportionately high for those making small amounts of sales in a state. In applying the physical-presence prong of the dual nexus rule, the physical presence of a dependent agent in a state could be considered evidence of physical presence of the principal, as now. But, contrary to some state court decisions, so could the in-state presence of corporate affiliates offering essentially the same products as the out-of-state entity. (The presence of affiliates providing services such as delivery, acceptance of returns, and repairs would presumably be evidence of an agency relationship and thus a physical presence.'°)
Sales of agents and affiliates would presumably be combined with those of the principal or parent for purposes of the sales prong of the dual nexus rule. Thus a corporation or corporate group could not avoid nexus by employing agents or legally separate affiliates to make sales. The physical-presence prong of the dual nexus rule just described contains a word not found in the nexus test the Supreme Court has providedthe word "substantial." This higher threshold for the physicalpresence test would not have the same negative implications for revenues and the competitive position of local vendors as it would if merely added to the present nexus rule, as it would be supplemented by the sales test. Remote vendors that made a non-de-minimjs amount of sales in the state would be found to have nexus, even if they had only an insubstantial physical presence in a state.
2.3 Inevitable Administrative Problems Several administrative problems would exist even under the economically neutral and compliance-friendly system.
2.3.1 Digitized Content The discussion to this point has focused implicitly on commerce in tangible products. Aside from the difficulty of sourcing sales to local jurisdictions (assigning them to jurisdictions of destination), to be discussed below, there is relatively little difficulty in sourcing
sales of tangible products, since the vendor generally knows where the products are shipped or delivered.1' The situation can be quite different in 10 The California Board of Equalization recently decided that Borders.com could be required to collect use tax, although it has no physical presence in the state, because its parent (Borders) acts as its agent in accepting returns in exchange for cash refunds. For further discussion of "entity isolation," see McIntyre (1997).
Gifts of tangible products constitute an exception to this generalization. But the basic questionwhether to assign gifts to the jurisdiction of the donor or that of the recipient is, in the first instance, more a philosophical issue than an administrative one. Assignment to the jurisdiction of the donor is probably simpler.
122
McLure
the case of digitized content, which is directed to an e-mail address. In this
case the vendor may not knowand may have no reliable way to learn
where the buyer is located. It may be possible to use the billing address as a proxy for the "ship to" address, but at some risk of manipulation.12
It is worth noting that this is the first time this discussion has touched upon an issue that is unique to electronic commercethat is, one that is not also present in traditional mail-order business.
2.3.2 Local Taxes The discussion to this point has also focused on state sales taxes. In fact, local governments also levy sales taxes. This
creates the need to determine the local tax rate(s) that should be applied to sales by remote vendors and the locality (or localities) that should receive the revenue from tax on such sales.13 Provided an exact match of addresses and taxing jurisdictions is not required, it should be possible to overcome this problem in one of several ways, for example, by relying on nine-digit ZIP codes. Alternatively, at the cost of even less precision, remote vendors might be allowed to employ a blended rate that reflects the average of all local rates in a given state, relying on the state to divide revenues among local jurisdictions. (This approach would not survive judicial scrutiny under current law, because it would impose an unconstitutional burden on interstate shipments to local jurisdictions with tax rates below the blended rate. The Congress, acting pursuant to the Commerce Clause, could relax this constraint.) Of course, in either event it would be essential that the tax base and the legal and administrative framework for local taxes be identical to those for state taxes.'4
2.4 Existing Deviations from the Economically Neutral and Compliance-Friendly System The existing state sales taxes violate all of the principles of economic
neutrality stated above and are extremely complex. First, the taxes do not 12 For a more detailed discussion, see Eads et al. (1997).
13 Varian (1999) has suggested that, because of the difficulties of implementing local sales taxes, local governments should abandon the sales tax in favor of the local income tax. McLure (2000b) argues that while Varian might be correct if the nation were starting de novo to create a system of tax assignment, the costs of transition to such a system seem too great to make it a viable alternative. Because of commuting between taxing jurisdictions, a local income tax, which could be imposed on the basis of residence, would probably track benefits of public services more closely than a local payroll tax, which would ordinarily be based on employment. 14 A vendor that had nexus in a state would presumably have nexus for all local use taxes in the state.
Thinking Straight about the Taxation of Electronic Commerce
123
apply to all consumption; most services are exempt, as are a variety of tangible products (with the exemptions varying from state to state).15 Second, the taxes apply to a wide range of sales to business; it has been
estimated that, depending on the state, as much as 20-70 percent of taxable sales are not made to consumers.'6 Third, there is essentially no uniformity in any aspect of state sales taxes.'7 Moreover, some 2,300 local jurisdictions levy sales taxes, not all of which conform to the tax base or
other provisions of the state tax of the state where they are located and some of which administer their own taxes. States provide few de minimis rules, and vendor discounts generally do not offset compliance costs, especially for small vendors. Fourth, because of this complexity, a state cannot compel a vendor to collect its use tax unless the vendor has a physical presence in the taxing state. The existing system creates economic distortions in all the dimensions identified above. It favors the consumption of untaxed products; since many of these are services, the exemptions favor the more affluent, who consume disproportionately large amounts of services. Because many business inputs are taxed, it distorts choices of production and distribution techniques. It discriminates against local vendors. It distorts the location of economic activity (favoring remote vendors and producers who are not subject to tax on their inputs). This distortion is aggravated by entity isolation, the use of legally separate entities to avoid having nexus in a state.'8 Moreover, the de facto inability to tax remote sales to consumers implies that, for a given tax rate, revenues will be lower than if the tax applied to all purchases from remote sellers. Unfair competition from remote vendors and loss of tax base have received the greatest attention in the recent debate.
3. IMPLICATIONS OF EXEMPTING ELECTRONIC COMMERCE Over the past several years many arguments have been heard for and against taxing electronic commerce. These are reviewed here, followed by a review of estimates of the revenue implications of the choice. 15 See Due and Mikesell (1994). 16 See Ring (1999).
' For a more complete description of interstate differences in tax bases, see Due and Mikesell (1994). Cline and Neubig (2000) tell horror stories involving interstate differences. McLure (2001b) calls the current system a "Great Swamp." 18 On entity isolation, see McIntyre (1997).
124 McLure
3.1 Arguments against Taxing Electronic Commerce Arguments for not taxing e-commerce take several forms, some of which are intertwined with arguments for exempting all remote sellers from a duty to collect use tax. One line of reasoning might be characterized as an "infant industry" argument: that e-commerce should experience a period of tax exemption
in order to allow it to "get on its feet." Several years ago, when it appeared that e-commerce would swallow the entire economy, the case for a tax subsidy on these grounds was not persuasive.19 Now that ecommerce has "cratered," despite the de facto existence of that subsidy for many sales to consumers, the wisdom of the policy is even more suspect. According to the "digital divide" argument, electronic commerce in general, and Internet access in particular, should be tax-exempt in order to avoid burdening low-income families, for whose children Internet access may represent an important way out of poverty. The general argument for exempting e-commerce is not persuasive, as affluent families spend far more on electronic commerce than do poor ones. The more limited case for exempting all charges for Internet access is also not compelling, since the ostensible objective could be achieved by exempting only basic service. Besides favoring affluent households, exempting all Internet access invites Internet service providers to "bundle" content with basic service. The difficulty of distinguishing between basic and Advocates of exempting e-commerce have badly misrepresented the implications of the work of Austan Goolsbee. Goolsbee's conclusion that taxing e-commerce would cause a reduction in the amount of e-comrnerce is not reason enough to exempt e-commerce; a similar conclusion could be reached for a tax on almost any product or form of commerce. 19
What is required is evidence that there is some form of external benefit (e.g., network externalities) that should be subsidized by exemption. Given current levels of usage of the Internet (or even the levels of a few years ago), it is hard to believe that significant network externalities remain unrealized. Thus, Goolsbee and Zittrain (1999, p. 424), state, "The major network externalities are likely to exhaust or at least diminish once the Internet achieves major scale. Too often, infant industry protection turns into established industry protection. Further, we expect that eventually there wifi be an important negative network externality. . . increasing Internet congestion. . . . The congestion problem is likely to get worse as the Internet grows and it argues against subsidizing the growth rate through tax policies." Similarly, testifying on behalf of the Congressional Budget Office, Thomas Woodward (2001) has said, "Network externalities arising from additional users, however, occur primarily in the early stages of a network's growth. At this point
in the Internet's development, there appears to be few external network benefits to be garnered from additional users. . . Effectively exempting remote purchases from sales taxes is an indirect and unevenly focused means of promoting the Internet's growth that is unlikely to bring significant benefits in terms of additional users or uses." See also .
Zodrow (2000, 2002).
Thinking Straight about the Taxation of Electronic Commerce
125
enhanced services could be sidestepped by limiting the exemption to a given dollar amount per month, say $25. Advocates of exempting electronic commerce have adopted a theory
advanced by the mail-order industrythat remote vendors should not be required to collect tax because they do not benefit from services provided by the states where their customers are located. This argumentand counter-arguments that implicitly accept the validity of its basic premiseconfuse the issue by focusing on services provided to remote vendors, which should be essentially irrelevant. The point is that purchasers pay the sales tax and it is to them that states provide services; the remote vendor would merely collect the tax. There is no reason to believe that a consumer of a given product consumes fewer state services simply because the product is bought from a remote vendor. Some cite the threat of competition from foreign vendors as a reason to exempt e-commerce. This threat is clearly far greater for digitized content than for tangible products. First, the cost of international shipping would limit the threat in the latter case.20 Second, sales tax on tangible products could, in principle, be collected at customs or the postoffice. (In fact, the lack of incentive for the federal government or the Postal Services to collect the tax and the complexity of state and local sales taxes makes this solution problematic.) By comparison, digital content can be "shipped" without significant cost, and it does not stop at the customhouse or the postoffice. This highlights the need for international cooperation in the taxation of e-commerce.21 At the very least, cooperation must encompass all members of the Organisation for Economic Cooperation and Development (OECD), and it may need to extend well beyond that. Yet another argument against taxing e-commerce involves holding Main Street merchants hostage in order to gain lower taxes. The reasoning is that if e-commerce (or all remote commerce) is not taxed, representatives of Main Street will pressure state and local governments to lower taxes, so that they wifi not be at so great a competitive disadvantage.
The basis for one argument for not extending the present sales tax system to electronic commerce (or to any form of remote commerce)20 The cost of shipping cannot, however, justify not taxing imported products, as has been argued in the case of interstate mail order. To the extent that the tax exemption subsidizes shipping that would not otherwise occur, it causes resources to be wasted. 21 On the taxation of electronic commerce in the European Union, see McLure (2001a). On the way globalization limits national sovereignty in taxation, see McLure (2001c).
126
McLure
namely, the complexity of the systemis indisputable. This does not, of course, mean that e-commerce vendors should not collect a tax that is dramatically simplified.
3.2 Arguments for Taxing Electronic Commerce Arguments for taxing electronic commerce are based on equity, economic neutrality, revenue (or lower tax rates), and simplicity of compliance and administration. They are, however, predicated on the assumption that
the system is simplified substantially. Neither equity nor neutrality would be furthered by requiring remote vendors to collect tax in the absence of simplification.
There are several strands to the equity argument for taxing ecommerce. Perhaps most important, it is unfair to exempt remote sellers, including those involved in electronic commerce, from the duty to collect
a tax that local merchants must collect. Also, it is unfair to exempt
e-commerce purchases, which are made disproportionately by the relatively affluent, while taxing purchases from local vendors, made disproportionately by the less affluent. The neutrality argument for taxing e-commerce is implicit in the earlier description of an economically neutral system and the distortions of the existing system. Two aspects of the additional distortions that would be created by exempting all e-commerce deserve special mention. First, many products can be delivered in either a tangible or an intangible (digitized) form. Exempting the latter would tilt choices toward that form of delivery. Second, given the "footloose" nature of many aspects Woodward (2001) makes most of these points. The objective of taxing electronic commerce like traditional commerce has been endorsed by the U.S. Treasury Department (1996), the Organisation for Economic Co-operation and Development (2001), and Senator Wyden, one of the original sponsors of the Internet Tax Freedom Act, who has included these words in 5. 288:
As a matter of economic policy and basic fairness, similar sales transactions should be treated equally, without regard to the manner in which sales are transacted, whether in person, through the mails, over the telephone, on the Internet, or by other means. More than 100 academic tax specialists endorsed the "Appeal for Fair and Equal Taxation of Electronic Commerce," which the author submitted to the Advisory Commission on Electronic Commerce at its meeting in San Francisco in November 1999. The Appeal, which is reproduced in McLure (2000a) and (2000d), contained the following language:
Electronic commerce should not permanently be treated differently from other commerce. There is no principled reason for a permanent exemption for electronic commerce. Electronic commerce should be taxed neither more nor less heavily than other commerce.
Thinking Straight about the Taxation of Electronic Commerce
127
of electronic commerce, exempting sales made by remote vendors would aggravate distortions of locational decisions. Given the complexity of the present system, it may come as a surprise that simplicity of compliance and administration is cited as an advantage of taxing electronic commerce. Exemption of e-commerce in the content of an otherwise radically simplified system would place a premium on
the definition of e-commerce and spur efforts to "shoe-horn" various forms of traditional commerce into the exempt categoryand also efforts to prevent thisthereby creating complexity.
3.3 How Much Revenue Is at Stake? A few years ago state and local officials were issuing dire warnings that their tax base would vanish into cyberspace. Even before the end of the dotcom boom it came to be realized that these fears were vastly overstated.23
First, remote sales to business represent a large fraction of
e-
commerce. Some of these (e.g., sales for resale) are exempt, and tax on much of the rest can be collected directly from the buyer, which risks being audited, if the vendor does not remit the tax. Second, some e-commerce transactions would not be taxable in any event, because the products are exempt (e.g., food in many states and services in most). Third, some e-commerce sales represent a shift from traditional re-
mote transactions that would effectively go untaxed because of the physical-presence nexus rule of Quill. Finally, some e-commerce sales are by vendors who have nexus and thus collect tax. Chine and Neubig (1999) found revenue losses in 1998 from the failure to tax electronic commerce to be only one-tenth of one percent of total sales tax revenue. Goolsbee and Zittrain (1999) estimated that revenue
losses in 1998 were less than one-quarter of one percent of sales tax revenues and that in 2003 losses would be less than 2 percent of total sales tax revenues. Bruce and Fox (2000) estimated losses in 2003 to be
about 1.5 percent of total state and local tax revenues; they did not translate this figure into percent of sales tax revenues. Emphasizing the uncertainty of any such estimates, the General Accounting Office (2000) estimated that revenue losses for 2000 would be less than 2 percent of total sales tax revenues and that in 2003 revenue losses would fall within the range of 1 to 5 percent of total sales tax revenues. See Cline and Neubig (1999), Coolsbee and Zittrain (1999), Bruce and Fox (2000), and U.S. General Accounting Office (2000).
128 McLure
4. EXTENSION OF THE INTERNET TAX FREEDOM ACT To understand the debate over provisions that might accompany extension of the ITFA, it is necessary to know a bit about nexus rules, the judicial and statutory standards that determine whether a state can tax income or require remote vendors to collect sales tax or pay business activity taxes (BATs).24
4.1 Nexus for Use Tax Collection Even before enactment of the ITFA in 1998, many remote e-commerce sales were already effectively exempt from taxation. In 1967 (in National Belas Hess, 386 U.S. 753) and again in 1992 (in Quill), the U.S. Supreme
Court ruled that the sales and use taxes imposed by the states are so complicated that requiring remote vendors to collect the tax would impose an unconstitutional burden on interstate commerce. Only if the vendor has a physical presence in the state can it be required to collect tax. Although the buyer is legally liable to remit the tax on purchases from remote vendors who lack nexus, few non-business purchasers actually do so.25 Thus the tax on sales made to consumers by many remote vendors is, in effect, a voluntary tax that few pay. Proposals for changes in the nexus rules affecting e-commerce fall into
three groups. Industry representatives favored a proposal that would elevate the test of nexus for duty to collect use tax on remote sales from "physical presence" to "substantial physical presence" and provide a list of in-state activities that would not be deemed to constitute nexus. These
changes in the nexus standard would significantly restrict the taxing power of state and local governments.
In response to the projected growth of e-commerceand thus of effectively exempt salesmost of the states are participating in the Streamlined Sales Tax Project (SSTP), which is intended to simplify and 24 For a much more detailed discussion, see Hellerstein (1997).
Since the test of nexus under the Due Process Clause is not as high as that under the Commerce Clause, states might be able to require remote vendors to provide information on sales to their residents and use that information to collect tax from purchasers. Or states might agree to collect this information for each other. Compared to collection of tax at the
time of sale, such an approach would be extremely inefficient and costly and would probably not be worthwhile, except for big-ticket items. States where mail-order houses are concentrated would face pressure not to participate in a scheme for exchange of information. If such a scheme were put in place, it can be expected that states would compete for the headquarters of businesses by not participating in it.
Thinking Straight about the Taxation of Electronic Commerce
129
modernize sales and use tax collection and administration; the next section describes the SSTP in greater detail. State and local governments favored a legislative proposal that would, in effect, override Quill for states that adopt the recommendations of the SSTP.26
The legislation actually enacted simply extends the ITFA for two years. It thus leaves the physical-presence nexus rule of. Quill intact, without assuring the states that simplification would be rewarded by relaxation of that rule.
4.2 Nexus for Business Activity Taxes The Supreme Court has not applied to BATs the same physical presence test of nexus it applies to sales taxes. But in 1959 the Congress limited state assertion of nexus by enacting P.L. 86-272, which prohibits taxation of the income of a seller whose only business activity in the state is solicitation of orders (including solicitation by agents) for sales of tangible personal property to be filled by shipment from outside the state. P.L. 86-272 provides no protection for a corporation selling intangible property in a state. The Supreme Court of South Carolina has thus ruled (in Geoffrey, 437 S.E. 2d 13, cert. den. 114 5. Ct. 550, 1993) that the presence of intangibles in the state creates nexus. Since the U.S. Supreme Court refused to grant certiorari in the case, other states have sought to assert "Geoffrey nexus" in similar cases. There is considerable concern in the business community that this legal doctrine could be used to justify income taxes on out-of-state e-commerce firms selling intangibles. To prevent this from happening, one proposal would extend the protection of P.L. 86-272 to sellers of intangible products and apply to business activity taxes a newly enacted "substantial physical presence" test of nexus (which would clarify that the presence of intangible assets in a state would not constitute nexus). 26
On August 17, 2001 the governors of 42 states sent the following letter to all members of Congress: August 8, 2001 TO ALL MEMBERS OF THE UNITED STATES CONGRESS: If you care about a level playing field for main street retail businesses and local control of states, local governments, and schools, extend the moratorium on taxing Internet access ONLY with authorization for the states to streamline and simplify the existing sales tax
system. To do otherwise perpetuates a fundamental inequity and ignores a growing problem.
The governors of California, Colorado, Delaware, Georgia, Massachusetts, New Hampshire, New York, and Virginia did not sign the letter, which is available at
http://www.nga.org/nga/legislafiveUpdate/1,1169,cLEnER-D46600htJ
130
McLure
5. PRIOR EFFORTS TO FIND A SOLUTION Over the past several years several groups have attempted to find a solution to the problems posed by the advent of electronic commerce, especially in the sales tax area; one of these efforts continues and seems to offer the best hope for a solution.
5.1 TheNTA Project In 1997 the National Tax Association convened a large group representing business, state and local government, and "others." Virtually all of
the business representatives were from the e-commerce and high-tech sectors; other, "traditional" business interests were essentially unrepresented. (There seems to be a presumption that it is representatives of the industries that would be affected who should be asked about tax policy in this area. Not surprisingly, they have responded that they would rather not pay income taxes or collect use taxes.) The NTA Project met periodically for two years. From the outset the Project focused on a possible "compromise" involving greater simplification of the sales and use tax in exchange for an expanded duty to collect tax on remote sales. The Project considered only "broad brush" reforms, such as uniform definitions of potential elements of the tax base (that is, uniform definitions of products that might be taxed or exempt in a given state), the sourcing of e-commerce transactions, and technological fixes such as processing by credit-card companies; it never considered the many details of compliance and administration the SSTP has addressed, much less more radical reforms, such as the economically neutral and compliance-friendly system described earlier, which were deemed to be beyond its frame of reference. Because of strict rules that demanded a substantial qualified majority to make a decision, the Project was unable to reach a consensus on recommendations. Even so, the Project served the useful purpose of identifying issues and increasing mutual understanding among the various parties represented.27 One sticking point that deserves notice was concern that government representatives might attempt to parlay business agreement to an expanded duty to collect use taxes into a lower nexus threshold for business activity taxes. (See also the Appendix.)
27 The Final Report of the NTA Project is available at http://ntanet.org/. It does not seem unreasonable to attribute part of the states' continued interest in simplifying their systems to greater appreciation that the potential compliance problems cited by business are real.
Thinking Straight about the Taxation of Electronic Commerce
131
5.2 The Advisory Commission on Electronic Commerce The Internet Tax Freedom Act created the Advisory Commission on Electronic Commerce, which was charged with making "a thorough study of Federal, State and local, and international taxation and tariff treatment of transactions using the Internet and Internet access and other comparable intrastate, interstate or international sales activities." While members of the Advisory Commission ostensibly represented state and local governments, traditional business, and consumers, as well as the e-commerce and high-tech sectors and the federal government, in fact the membership was packed with members who opposed taxation of e-commerce.28
At the end of the day it was only the combination of the requirement for a two-thirds majority for the adoption of a "finding or recommendation" and the unwillingness of the federal representatives to go along with the opponents of taxation that prevented the Advisory Commission from forwarding to the Congress proposals that, if adopted, would have drastically reduced the taxing powers of the states. Even so, the Advisory Commission forwarded the following "majority policy proposals."29 First, a five-year exemption for digitized content downloaded from the Internet and "their non-digitized counterparts." This proposal would effectively exempt recorded music, videos, books and magazines, games, and software from taxation. Second, codification of the Quill decision regarding nexus for use tax
purposes and provision of safe harbors that would prevent corporate affiliation, repairs, and returns from being construed as evidence of a physical presence in the state. Third, application of the physical-presence test of Quill, extended as described above, to nexus for business activities taxes. Fourth, a suggestion that the National Conference of Commissioners
on Uniform State Laws (NCCUSL) be asked to draft a uniform sales and use tax act that would include: (1) uniform tax base definitions; (2) uniform vendor discount; (3) uniform and simple sourcing rules; (4) one sales and use tax rate per state and uniform limitations on state rate changes; (5) uniform audit procedures; (6) uniform tax returns! forms; (7) uniform electronic filing and remittance methods; (8) uniform exemption administration rules (including a database of all exempt entities); (9) a methodology for approving software that sellers may rely On this, see McLure (1999).
Advisory Commission on Electronic Commerce (2000).
132 McLure
on to determine state sales tax rates; (10) a methodology for maintaining revenue neutrality in overall sales and use tax collections within each state.
5.3 The Streamlined Sales Tax Project Forty of the 45 sales tax states are currently involved in the Streamlined Sales Tax Project (SSTP), 35 as voting participants and 5 as non-voting observers. The SSTP, in the words of its Executive Summary: is an effort created by state governments, with input from local governments and the private sector, to simplify and modernize sales and use tax collection and administration. The Project's proposals will incorporate uniform definitions within tax bases, simplified audit and administrative procedures, and emerging technologies to substantially reduce the burdens of tax collection. The Streamlined Sales Tax System is focused on improving sales and use tax administration systems for both Main Street and remote sellers for all types of commerce.3°
On December 22, 2000 state representatives to the SSTP voted unanimously to approve the Uniform Sales and Use Tax Administration Act and the Streamlined Sales and Use Tax Agreement. The Uniform Act
would authorize the taxing authority of the state "to enter into the Streamlined Sales and Use Tax Agreement with one or more states to simplify and modernize sales and use tax administration in order to substantially reduce the burden of tax compliance for all sellers and for all types of commerce." The Agreement contains the many details of simplification and uniformity. The National Conference of State Legislatures (NCSL) endorsed the Act and Agreement on January 24, 2001,
but only after making modifications that significantly reduce the implied simplification. (Among the items deleted pending further review are uniform definitions of items in the tax base and limitations on tax rate caps, thresholds, and sales tax holidaysall provisions that are important for simplification.) To date 22 states have approved some form of the legislation, and legislation has been introduced in 7 more states.31 3°
This description is from the Executive Summary of the SSTP, available at http://
208.237.129.206/sline/execsum.pdf.
Links to the Uniform Sales and Use Tax Administration Act and the Streamlined Sales and Use Tax Agreement, issue papers that explain the SSTP proposals, and a map showing the status of legislation are available at http://www.geocities.com/stream1ined2000/ and at http://www.nga.org/nga/salestax/1,1169,,00.html. McLure (forthcoming) examines the SSTP proposals. 3°
Thinking Straight about the Taxation of Electronic Commerce
133
The key features of the Streamlined Sales Tax System developed by the SSTP include32:
Uniform definitions of products. Legislatures will still be able to choose what is taxable and exempt, but wifi use common definitions for broad groups of products (and a few more narrowly defined products, such as soft drinks). State and local governments in a given state wifi use the same tax base. Simplified exemption administration. Proposals for simplification cover both
use- and entity-based exemptions. Uniform codes wifi be used for exempt products. Rate simplification. States will be encouraged to simplify state and local tax rates. States will accept responsibility for notice of rate and boundary changes. State administration of local taxes. States wifi be responsible for the adminis-
tration of all state and local taxes and the distribution of revenue from local taxes to local governments. Uniform sourcing rules. The states will have uniform sourcing rules for all property and services. Simplified"one-stop" online registration and uniform returns. Vendors would
be able to register simultaneously in all participating states and file uniform returns. Uniform audit procedures. Sellers who use one of the certified technology models described below either will not be audited or wifi have an audit
that is limited in scope, depending on the technology model used. Reliance on technology. The states would jointly certify software to be used
to calculate the tax imposed by each sales tax jurisdiction. A vendor could use such software to calculate and remit its own tax, or it could rely on a certified service provider (CSP, called a "trusted third party" in earlier descriptions of this approach) to calculate and remit its tax. Alternatively, vendors could rely on proprietary software that had been certified by the states. Paying for the system. To reduce the financial burdens on sellers, states will assume much of the financial burden of implementing the Streamlined Sales Tax System.
Other simplifications would include standardized geographic coding, limited frequency and required advance notice for changes in tax rates and jurisdictional boundaries, limitations on the use of caps and thresholds in defining tax bases, requirement that tax holidays employ standard definitions of products, uniform provisions for bad debts, and uniform rounding rules. Where there are differences between the original SSTP proposals and those approved by the NCSL, the former are described. 32
134
McLure
The system being developed by the SSTP is a political compromise between simplicity and state sovereignty. It would substantially reduce complexity. But it falls short of the uniformity and simplicity of the economically neutral and compliance-friendly system described earlier, largely due to the desire to retain state control over many of the parameters of sales tax policy. Rather than proposing a uniform tax base, the SSTP proposes uniform definitions of what might be taxed or exempt, leaving to each state the
choice of whether or not to tax each identified item. Moreover, it achieves relatively little uniformity in the treatment of sales to business. To handle interstate differences in the tax base, the Project would rely heavily on technology, essentially comprehensive lookup tables that categorize products as taxable or exempt in each state. Rather than proposing a legal and administrative framework that is uniform from state to state, the SSTP would leave the existing structure unchanged, except in particular ways. Thus vendors and CSPs will still
face interstate differences in statutes, regulations, and interpretations and in administrative procedures.
6. SUMMARY AND ANALYSIS 6.1 Summary The economically efficient sales tax system described earlier can be summarized in three rules: First, all sales to consumers would be taxed. Second,
all sales to business would be exempt. Third, sales by local and remote vendors would be taxed equally. A compliance-friendly sales tax would exhibit substantial simplicity and uniformity. Uniformity would extend to the tax base, to statutes, regulations, and their interpretation, and to administrative procedures, which would be drastically simplified. Remaining compliance costs would be mitigated by de minimis rules and vendor discounts. Existing sales taxes exhibit none of these characteristics. At present many sales to consumers are exempt, many sales to business are taxed,
and many sales by remote vendors are not taxed. The system is extremely complex, in large part because there is essentially no uniformity from state to state. States make little use of de minimis rules, and vendor discounts are inadequate to compensate for compliance costs. In recent years the states have begun serious efforts to simplify their sales taxes and make them more nearly uniform. These efforts would
substantially simplify the system, but do not go as far as the economically neutral and compliance-friendly system. Many sales to con-
Thinking Straight about the Taxation of Electronic Commerce
135
sumers would remain exempt, and many sales to business would still be taxed.
6.2 Analysis of ITFA Options As noted earlier, proposals for extension of the ITFA fell into three groups:
Extension only. The legislation that was enacted leaves many remote purchases by consumers effectively untaxed, with obvious implications for equity, efficiency, and revenues. It maintains pressure on states to simplify their systems, in hopes of gaining approval (from the
Supreme Court or the Congress) of an expanded duty to collect use tax, but does not provide any assurance that the requisite approval will be forthcoming if simplification is achieved. Government position. State and local governments favored a proposal that would authorize interstate cooperation to simplify sales and use taxes,
promising that if such cooperation is forthcoming, the physical presence test of nexus would be replaced (for cooperating states) with one
based on the volume of sales a remote vendor makes in a state. It would have been more conducive to economic neutrality, equity, and simplicity than the legislation actually enacted. Business position. Business favored a proposal that would prohibit state assertion of nexus unless a remote vendor has a "substantial physical presence" in the taxing state and lists activities that would not constitute a substantial physical presence. This proposal would have eliminated tax on many sales; to the extent it achieved simplicity it would do so at the cost of economic neutrality, equity, and revenue. Some industry proposals also would have extended the expanded nexus test to BATs, further undermining state revenues, especially where states have adopted the sales-only apportionment factor or do not require unitary combination. (On this, see the Appendix.)
6.3 Postscript On November 28-29, 2001, after the conference where this paper was presented, the "implementing states" of the SSTP held their inaugural meeting to chart the way forward. The proceedings of that meeting suggest that the states take seriously the need to simplify their sales and use taxes, in part because they realize that, if they fail to achieve real simplification by the time the ITFA expires again on November 1, 2003, the Congress may simply extend the ITFA permanently, without considering elimination of the physical-presence test of Quill.
136
McLure
APPENDIX: BUSINESS ACTIVITY TAXES Whereas the debate on sales and use taxes has, from the outset, involved an attempt to gain agreement on a compromise that would combine simplification and an expanded duty to collect use tax, what little debate has occurred on business activity taxes has been limited to the question
of nexus. (Business representatives would like to see the physicalpresence nexus rule of Quill, perhaps expended as described in the text, applied to BATs; they fear that if they agree to a compromise on nexus for use taxes, states wifi attempt to stretch the parameters of that compromise to assert nexus for BATs.) But this is an extremely narrow view of the problem; nexus must be considered simultaneously with other aspects of BATs if a reasonable result is to be reached.33
AJ Issues in the Design of State Business Activity Taxes Unlike the state sales tax, the state corporate income tax and related forms of BATs cannot easily be justified under any accepted principle of taxation. The benefit principle does not provide a fully satisfactory justifi-
cation, as it is hard to argue that the only businesses that benefit from public services are those that are incorporated, that only profitable corporations receive such benefits, or that benefits are proportionate to taxable profits. The "squishy" view that states are "entitled" to tax income that has its source within their boundaries fares only a little better. Perhaps
the best approach is to be pragmatic, recognizing that the purpose of state corporate income taxes is to tax income that originates in the state.34 Attempting to implement this "standard" encounters several obstacles and raises a number of issues that must be considered along with nexus rules.
A.1.1 Formula Apportionment Corporate taxpayers do not employ geographically separate accounting which attempts to measure the income that originates in each state, and requiring them to do so, besides being impractical, would be conceptually suspect and subject to abuse. First, because of the economic interdependence between parts of a corporaAs Dan Bucks, Executive Director of the Multistate Tax Commission, has said, "One cannot study business activity nexus separate from the rest of the structure of corporate tax or franchise taxes. Nexus standards interact with apportionment formulas and with report-
ing methodsand by reporting methods, I'm talking generally about combined reporting versus separate-entity corporate reporting. And the overall issue is very complex, and it really involves looking not just at nexus, but looking at an entire structure of the corporate taxes." Public testimony before the NCSL, reported in Sheppard (2001). The U.S. Supreme Court has applied a looser test: tax must be "reasonably related" to the taxpayer's activities in the state.
Thinking Straight about the Taxation of Electronic Commerce
137
tion operating in various states, it is conceptually impossible to isolate the income originating in each state. Second, corporations could manipulate transfer prices for transactions between parts of the corporation operating in different states to shift net taxable income to low-tax states. To overcome these problems states use formulas to apportion the total income of a multistate corporation among the states where it operates. For many years most of the states used a standard formula that assigned equal weights to three apportionment factors: payroll, property, and sales (at destination). More recently there has been a decided shift toward using only sales to apportion income, so that now "sales only" is the most common formula. It appears that this shift has occurred as part of an attempt to attract economic activity, not because it is thought to produce a better measurement of income originating in the state.
A.1.2 Unitary Combination If each member of a corporate group is taxed as a distinct entity, manipulation of transfer prices and economic interdependence make isolation of the income of the various entities problematic. To overcome these problems some states combine the activi-
ties of related corporations deemed to be engaged in a "unitary business" in order to determine the income of corporations doing business within the state. Under unitary combination transactions between members of the unitary group (e.g., sales, interest payments, and dividends) are ignored and the total domestic income of the group is apportioned among the states according to the apportionment factors of the entire group.35 Only a minority of states employ combination, despite its manifest advantages, and those do so do not all employ the same definition of a unitary business.
A.2 Defects of the Present System The present system exhibits the following characteristics.
A.2.1 Inconsistent Nexus Rules and Apportionment Formulas It is readily apparent that the nexus rule of P.L. 86-272 and the sales-only apportionment formula are logically inconsistent; if merely having sales
in a state does not create nexus and only sales are used to apportion income, substantial amounts of income may escape taxation. The inconsistency of the two rules can result in substantial loss of revenue, especially when a state does not combine the activities of corporate affiliates engaged in a unitary business; an out-of-state corporation could have During the 1980s some states' application of unitary combination on a worldwide basis created considerable international consternation and controversy. That practice has now ended and is not considered here.
138
McLure
significant sales in a stateincluding one where it has affiliateswithout having nexus.36 Enactment of the physical-presence test of Quill would exacerbate this problem.
To some degree individual states suffer from a problem they have created and could correct acting alone. That is, a state could minimize the damage to its revenues by requiring unitary combination and avoiding the sales-only apportionment formula. (By comparison, the inability to assert nexus for use tax is the result of the collective inability of the states to simplify the system and thus cannot be overcome by any individual state acting alone.) But this would not eliminate the loss of tax base that occurs when a corporation meets the standards of P.L. 86-272 (or that would occur under the "substantial physical presence" test of nexus). This could be avoided by replacing P.L. 86-272 with a de minimis
nexus rule based on the in-state presence of significant amounts of the apportionment factors (as well as the existence of a significant amount of apportionable income) A.2.2 The Lack of Uniformit,' Such a reform would expose many more corporations to liability for BAT in states where they have no physical presence. This would accentuate compliance problems caused by the lack
of uniformity of such taxes unless the relaxation of nexus rules were accompanied by simplification. A simplified system might exhibit the following forms of uniformity, some of which are analogous to those of the economically neutral and compliance-friendly sales tax system: first, a uniform definition of apportionable income (presumably based on the federal definition of taxable incomep with such adjustments as are re-
quired by combination); second, application of unitary combination, based on a uniform definition of a unitary business; third, a uniform apportionment formula, based on uniform definitions of apportionment factors; fourth, uniform statutes, regulations, and interpretations; and fifth, simplified administrative procedures ("one-stop" registration, filing, etc.). As under the economically neutral and compliance-friendly sales tax system, states would retain complete control over tax rates. There is currently a lack of uniformity especially in the application of
unitary combination and the choice of apportionment formulas. This creates both complexity and the possibilities of over- and undertaxation. Because of reliance on federal concepts, there is somewhat more uniformity of legal and administrative standards than in the sales tax field. For further discussion of inconsistencies between these provisions, see Mazerov (2001). This argument is developed more fully in McLure (2000c).
Thinking Straight about the Taxation of Electronic Commerce
139
REFERENCES Abrams, Howard E., and Richard L. Doernberg (1997). "How Electronic Commerce Works." State Tax Notes 13(July 14):121-136. Advisory Commission on Electronic Commerce (2000). Report to Congress. Bruce, Donald, and William F Fox (2000). "E-Commerce in the Context of Declining Sales Tax Bases." National Tax Journal 52 (no. 4, part 3, December):13731388.
Cline, Robert J., and Thomas S. Neubig (1999). "The Sky Is Not Falling: Why State and Local Revenues Were Not Significantly Impacted by the Internet in 1998." State Tax Notes 17(July 5):43-51.
Cline, Robert J., and Thomas S. Neubig (2000). "Masters of Complexity and Bearers of Great Burden: The Sales Tax System and Compliance Costs for Multistate Retailers." State Tax Notes 18(January 24):297-313. Diamond, Peter A., and James A. Mirrlees (1971). "Optimal Taxation and Public Production II: Tax Rules." American Economic Review 61 (no. 2, June):261-278. Due, John F., and John L. Mikesell (1994). Sales Taxation: State and Local Structure
and Administration (2nd edition). Washington: Urban Institute Press. Eads, James, et al. (1997). "National Tax Association Communications and Electronic Commerce Tax Project Report No. 1 of the Drafting Committee." State Tax Notes 13(November 17):1255-1270.
Goolsbee, Austan, and Jonathan Zittrain (1999). "Evaluating the Costs and Benefits of Taxing Internet Commerce." National Tax Journal 52(September):413428.
Hellerstein, Walter (1997). "State Taxation of Electronic Commerce." Tax Law Review 52(no. 3, spring):425-506. Mazerov, Michael (2001). "The Single-Sales-Factor Formula: A Boon to Economic Development or a Costly Giveaway?" Center on Budget and Policy Priorities.
McIntyre, Michael J. (1997). "Taxing Electronic Commerce Fairly and Efficiently." Tax Law Review 52:625-654.
McLure, Charles E., Jr. (1997). "Taxation of Electronic Commerce: Economic Objectives, Technological Constraints, and Tax Laws." Tax Law Review 52 (no. 3, spring):269-424. McLure, Charles E., Jr. (1998a). "Electronic Commerce and the Tax Assignment
Problem: Preserving State Sovereignty in a Digital World." State Tax Notes 14(April 13):1169-1381.
McLure, Charles E., Jr. (1998b). "Achieving a Level Playing Field for Electronic Commerce." State Tax Notes 14(June 1):1767-1783. McLure, Charles E., Jr. (1999). "Electronic Commerce and the State Retail Sales Tax: A Challenge to American Federalism." International Tax and Public Finance 6(May):193-224.
McLure, Charles E., Jr. (2000a). "The Taxation of Electronic Commerce: Background and Proposal." In Public Policy and the Internet Privacy, Taxes and Contracts, Nicholas Imparato (ed.). Stanford, CA: Hoover Institution Press. McLure, Charles E., Jr. (2000b). Rethinking State and Local Reliance on the Retail Sales Tax: Should We Fix the State Sales Tax or Discard It? Brigham Young University Law Review 2000:77-137.
McLure, Charles E., Jr. (2000c). "Implementing State Corporate Income Taxes in the Digital Age." National Tax Journal 53(December):1287-1305.
140 McLure McLure, Charles E., Jr. (2000d). "Howand Whythe States Should Tax Electronic Commerce: Explanation of My ACEC Proposal for Radical Simplification." State Tax Notes (January 10):129-140. McLure, Charles E., Jr. (2001a). "Taxation of Electronic Commerce in the European Union." Prepared for presentation at a conference on "Tax Policy in the European Union" held at the Research Center for Economic Policy, Erasmus University, Rotterdam, October 17-19. McLure, Charles E., Jr. (2001b). "SSTP: Out of the Great Swamp, but Whither? A Plea to Rationalize the State Sales Tax." 22 State Tax Notes 14 (December 31):1077-1086.
McLure, Charles E., Jr. (2001c). "Globalization, Tax Rules and National Sovereignty." Bulletin for International Fiscal Documentation 55(August):328-341.
McLure, Charles E., Jr. (forthcoming). "Taxing Electronic Commerce: Legal, Economic, Administrative and Political Issues." Prepared for presentation at the Annual Meeting of the Association of American Law Schools, San Francisco, January 5, 2001. Forthcoming in Urban Lawyer.
Ring, Raymond J., Jr. (1999). "Consumers' Share and Producers' Share in the General Sales Tax." National Tax Journal 52(March):79-90. Organisation for Economic Co-operation and Development (2001). "The Ottawa Taxation Framework Conditions." In Taxation and Electronic Commerce: Implementing the Ottawa Framework Conditions. Paris: OECD.
Sheppard, Doug (2001). "NCSL Approves Internet Tax Resolutions." State Tax Notes 21(August 20):570.
Slemrod, Joel (1990). "Optimal Taxation and Optimal Tax Systems." Journal of Economic Perspectives 4(no. 1, winter):157-178.
U.S. Department of the Treasury, Office of Tax Policy (1996). "Selected Policy Implications of Global Electronic Commerce." Available at http:// www.ustreas.gov/taxpolicy/intemet.html, visited August 22, 2001. U.S. General Accounting Office (2000). Sales Taxes: Electronic Commerce Growth Presents Challenges; Revenue Losses Are Uncertain.
Varian, Hal (1999). "A Proposal to Eliminate Sales and Use Taxes," a submission
to the Advisory Commission on Electronic Commerce. Available at http:// www.ecommercecommission.org/library.htm. Woodward, G. Thomas (2001). "Economic Analysis of Taxing Internet and Other
Remote Sales." Testimony before the Committee on Finance, U.S. Senate, August 1. Zodrow, George R. (2000). "Evaluating the Case for a Temporary Moratorium on the Taxation of Electronic Commerce." State Tax Notes 19(December 4):15151521.
Zodrow, George R. (2002). "Network Externalities and Indirect Tax Preferences for Electronic Commerce." Forthcoming in International Tax and Public Finance.
THE FISCAL EFFECTS OF
POPULATION AGING IN THE U.S.: ASSESSING THE UNCERTAINTIES Ronald Lee and Ryan Edwards University of California
EXECUTIVE SUMMARY Population aging, accelerating as the Baby Boom generations age, will have important fiscal consequences because expenditures on social security, Medicare, and institutional Medicaid make up more than a third of the federal budget. However, the projected fiscal pressures are far in the future, and long-term projections are very unreliable. Our analysis here has two goals: to examine the fiscal impact of population aging, and to do this in a probabilistic setting. We find that the old age dependency ratio is virtually certain to rise by more than 50% through the 2030s, and will probably continue to increase after 2050, possibly by a great deal. Under current program structures, population aging would be virtually certain to increase the costliness of Federal programs as a share of GDP by 35 percent (±2 percent) by the 2030s, and by 60 percent (± 15 percent) in the second half of the century. We project Federal expenditures (excluding interest payments and pre-funded programs) to rise from 16 percent of GDP in 2000 to 30 percent in 2075, almost doubling, while state and local Research for this paper was funded by a grant from NIA, R37-AC11761. We thank Timothy Miller for his help with various parts of the analysis, and James Poterba for helpful comments.
142
Lee & Edwards
expenditures rise only modestly relative to GDP. Almost all of this increase is for programs going primarily to the elderly, which rise from 8 percent of GDP in 1999 to 21 percent of GDP in 2075, due mainly to costs of health care for the elderly, with pensions a distant second. We expect that
governments wifi respond to these aging-induced cost changes by altering program structures, so that these conditional projections will not be realized. Looking at social security, we find that raising the payroll tax rate by 1.89 percent would have relatively little effect on the probabilities of early exhaustion, raising the 2.5 percent bound for the exhaustion date from 2024 to 2036, but raising the median date of exhaustion from 2036 to
2070, and with a 55 percent chance of insolvency within the 75 year horizon. Looking at Medicare, which now costs 2.2 percent of GDP, we project a median share in 2075 of 11 percent, five times as great, with a 95 percent probability interval at 5 percent to 26 percent of GDP. Thus there is a 97.5 percent chance that the ratio wifi at least double, and a 2.5 percent chance that it wifi increase at least twelve-fold. Although the future is highly uncertain in many respects, unforeseen demographic or economic change wifi almost certainly not avert the long-run fiscal crunch. Changing demographic realities will require some combination of substantial tax increases or substantial benefit cuts, or other forms of restructuring.
1. INTRODUCTION Expenditures on social security, Medicare, and Medicaid for nursinghome care together make up more than a third of the federal budget.1 These are all programs for the elderly, paid for largely out of taxes on the working-age population. Around 2010, the Baby Boom generations will begin to turn 65 and to draw on these programs, and by 2040, the ratio of
elderly people to those in the current working ages wifi have doubled from about 0.2 to about 0.4. Clearly, population aging will exert heavy pressure on the federal budget in the coming decades. Nonetheless, there are many questions about this process. The projected fiscal pressures are far in the future, and long-term projections are very unreliable; might not the whole problem evaporate if we just wait? For example, wifi increased costs in programs for the elderly be offset by decreased expendi-
tures for children elsewhere in the federal budget and in the state and local budgets? Wifi the passing of the Baby Boom generations bring fiscal
relief? Wifi high-fertility immigrants raise national fertility, as Census projects? Or will fertility in the U.S. move towards low European levels, Institutional Medicaid is the part of the Medicaid program that pays for nursing-home care, primarily for the indigent elderly. 1
The Fiscal Effects of Population Aging in the U.S.
143
closer to one child per woman, greatly intensifying population aging? In
addition to these demographic sources of uncertainty, there are very substantial economic uncertainties. These are well illustrated by the rapidity with which projections of fiscal doom in the mid-1990s were replaced
in the late 1990s by projections of burgeoning surpluses, which have recently been corrected to show declining surpluses over the short run. To formulate fiscally responsible policies, policymakers must have not only best-guess forecasts of the future, but also a measure of the uncertainty surrounding these forecasts. Before proceeding, some background will be helpful. First, it is important to put population trends in the U.S. into international perspective. Among the industrial nations, the U.S. has high fertility and a stingy public pension system. Combined with middling mortality and immigration, this adds up to relatively mild population aging and to relatively modest fiscal pressure, compared to most European nations or Japan. An OECD study (Roseveare, Leibfritz, Fore, and Wurzel, 1996) found that the fiscal imbalance in the U.S. public pension system over the next 75 years was among the smallest of any of the OECD countries, relative to GDP. The fiscal consequences of population aging for the U.S. wifi
indeed be severe, but in many other industrial nations they wifi be simply staggering.
Second, it is important to distinguish between the fundamental resource problem posed by population aging, which would exist whatever our institutional structures, and the particular problems that arise specifically because of our institutional arrangements for supporting the elderly, including government programs. Population aging raises the number of elderly people relative to the number of working-age people, when we hold the age boundary constant at some level such as age 65. In this sense, population aging occurs partly because individuals live longer, and partly because birth rates are
lower, so that younger generations are smaller at birth than the older generations. Longer life, by contrast, results at least in part from better health, so that elderly people at any specific age are more vigorous and less likely to be disabled. For many years, it was feared that people whose lives were saved by declining mortality might be functionally impaired by their close brush with death or by weaker constitutions, so that disability rates at older ages would rise. In the U.S., at least, this has not happened, and indeed rates of disability at older ages are declining at roughly the same rate as mortality itself. Older people wifi increas-
ingly be functionally able to prolong their working lives (Manton, Corder, and Stallard, 1997; Freedman and Martin, 1999; Crimmins, Saito, and Ingegneri, 1997). People may choose to take their additional
144
Lee & Edwards
years of life as leisure years rather than as working years, and consequently they may need to save more or pay higher taxes to support those additional years of leisure. That is a matter of choice, and not caused by
aging itself. In this sense, longer life does not cause a fundamental resource problem. By contrast, lower fertility means there are fewer working-age people in the population relative to the elderly, without altering the health or functional status of the elderly. Population aging due to low fertility, unlike that due to low mortality, does fundamentally alter the resource constraints facing society. Population aging occurs in the context of a particular set of institutions and traditions. In the U.S., the median age at retirement, far from rising with increasing longevity, has declined by five years since 1950. To some degree, this decline reflects a choice for more leisure at the end of life, influenced by higher lifetime incomes and the public and private pensions which have made it easier for people to realize what would anyway have been their preferred life-cycle plans. That is the positive side of the story. On the negative side, however, both public and private pension programs have incorporated incentives for earlier retirement, whether by design or by accident. In Europe, the easy availability of government-
provided disability and unemployment benefits for older people has added to these incentives, and extended them to younger ages (Gruber and Wise, 1999). These pension, disability, and unemployment programs create an implicit tax on continuation of work, inducing many people to
retire early. When population aging occurs in the context of rigid and distortionary institutions, particularly severe problems may arise. This is the case for most public pension programs throughout the OECD countries, and to a lesser extent for the U.S. Institutional arrangements surrounding the provision of health care, particularly for the elderly, shape the impact of population aging on public health care costs. Once again, we must distinguish between longer life and lower fertility as causes of population aging. Longer life goes with improved health in old age, and the net effect on health costs appears to be slight (Lubitz and Prihoba, 1984; Lubitz, Beebe, and Baker,
1995). Nor have the costs of typical medical procedures risen in real terms, on the contrary, they may have fallen (Cutler, McClellan, Newhouse, and Remler, 1998; Cutler and Sheiner, 2000). Why, then, are government health care expenditures projected to rise so strongly relative to GDP over the twenty-first century? First, because lower fertility wifi mean fewer workers to bear the cost of health care for the elderly (that is, future GDP will be lower than otherwise). Second, because populations of industrial nations have a great appetite for the costly new procedures made available by striking technological advances in medi-
The Fiscal Effects of Population Aging in the U.S.
145
cine over recent decades, advances which are expected to continue in the
future. In the U.S., these two factorspopulation aging due to low fertility and the purchase of costly new biomedical technologiesare expected to account for roughly equal shares of the projected increased expenditure on health care relative to GDP (Lee and Miller, 2001b). The portion of the increase due to low fertility and population aging cannot be avoided, except possibly by pro-natalist and pro-immigration policies. It is important to make sure that the other part, due to increased quantity and quality of health care services consumed, is growing in a way consistent with individual and social preferences, not simply be-
cause of distortions arising from the structures of institutions formed decades ago. It is easy to exaggerate the fiscal pressures generated by population aging. In practice, the structure and generosity of programs do not remain fixed as population changes, and changing population age distributions have only modest power to explain government expenditures on social welfare programs in the past. The structure and generosity of programs may be rigid in the short term, but in the longer term these too adjust. We should keep in mind, however, that even if the fiscal pressures are mitigated by program changes, these mitigating changes may simply pass costs on to the beneficiaries, in the form of later retirement or reduced medical benefits. This paper will begin by discussing demographic change in the U.S. Next, it wifi discuss approaches to assessing the uncertainty of projections, which is then followed by an overview of our own probabilistic projections of population aging. Then it wifi consider how population aging alters the budgetary trade-offs that constrain government programs in the aggregate, contingent on the continuation of current program structure. An overview of stochastic budget projection techniques follows, motivated by a summary of deterministic predictions. The pa-
per will then discuss in more detail the uncertainty of projections for social security and for Medicare. In the final section, we discuss our results.
2. POPULATION CHANGE IN THE U.S. 2.1 Fertility Over the past two centuries in the U.S., the economic roles of women and
children have changed, incomes have risen dramatically, mortality has declined, the frontier has been settled, and contraceptive technology has advanced. Consequently, fertility has declined steadily from 7 or 8
births per woman in 1800 to 2.0 births per woman today, a decline
146
Lee & Edwards
interrupted by the Baby Boom between 1946 and 1965. Similar factors have led to a fertility decline in other industrial countries as well. However, fertility in the U.S. has always been high compared to levels in other industrial populations. The average level for European populations is 1.4 births per woman, with some countries like Spain, Italy, and Germany close to 1.2, and a few smaller populations close to 1.0. Will fertility in the U.S. decline toward European levels in the coming decades? Two considerations reduce the likelihood. First, European women typically report on
surveys that they would like to have 2 children on average, suggesting that fertility may rise in the future. Second, the average age of childbearing has been rising steadily in most European countries in recent decades, as women postpone giving birth. This trend distorts the standard fertility measure downward by 0.2 to 0.4 births, relative to fertility levels over the life cycle of these women, again suggesting that European fertility may rise in the future. Fertility is high in the U.S. in part due to the higher fertility of minority groups such as African Americans, Latinos, and Asians. Projections indicate that the population share of non-Hispanic Whites wifi decline from around 70 percent today to around 50 percent by 2050. Will growing population shares of these minority groups lead to higher aggregate fertility? The U.S. Census Bureau has assumed in its recent projections that it will. We are skeptical. It is mainly first-generation immigrants who have high fertility; by the third generation fertility has historically converged to the levels of the general population (Smith and Edmonston, 1997). For example, first-generation Latino women have more than three births on average, whereas third-generation Latino women have only two. Furthermore, fertility in the sending countries in Latin America and East Asia is rapidly falling In Mexico, fertility is now down to 2.6
births per woman and falling rapidly, while many populations in East Asia, including China, have fertility below replacement level. This suggests that future immigrants will not have fertility that is much higher than the rest of the U.S. population. Overall, therefore, we believe it is reasonable to project, as a point estimate, that fertility levels continue at the current level of about 2.0, while noting that there is a great deal of uncertainty about this central forecast.
2.2 Mortalitj
Improvements in nutrition, public sanitation and hygiene, personal habits, biomedical technology, and health service delivery have caused mortality declines throughout the world. Dramatic progress first against infectious disease, and then against chronic and degenerative disease,
The Fiscal Effects of Population Aging in the U.S.
147
has brought life expectancy in most industrial nations from 30 to 40 years at birth in the early nineteenth century to 75 to 80 years at birth today.
The acceleration of the rate of decline in death rates at the older ages, even above 100, in recent decades has been particularly striking (Kannisto, Lauritsen, Thatcher, and Vaupel, 1994). Mortality is expected to continue to decline in the twenty-first century, but how far and how fast is open to question and controversy. The pessimists believe that it will be difficult to raise life expectancy above 85 years without truly revolutionary medical advances to slow the progress of aging itself (Fries, 1980, 1984; Olshansky, Carnes, and Desesquelles, 2001; Board of Trustees, 2001). Optimists believe that advances in stem-cell and genetic therapies may raise life expectancy as high as 150 years in this century. Forecasting methods based on long-run trend extrapolation suggest that life expectancy in the U.S. wifi rise from its current 76.7 years to around 86 years by 2075, plus or minus 4 years (Lee and Carter, 1992; Lee and Miller, 2001a; Tuljapurkar, Li, and Boe, 2000). Our forecasts in this paper wifi be based on this latter approach.
2.3 Immigration The annual number of net immigrants to the U.S. has risen linearly since 1950, and shows no signs of decelerating in recent years. Nonetheless, because immigration is a policy variable, we have elected to follow the Social Security Actuaries in assuming that the net annual number will remain at 900,000. Certainly a case could be made for forecasting the linear trend to continue.
2.4 Projections by the U.S. Bureau of the Census and the Social Security Administration While the assumptions about net immigration are very similar between Census and SSA, the assumptions for fertility and mortality are quite different. The SSA assumes that fertility will decline from about 2.05 children per woman today to 1.95 by 2025, remaining constant thereafter. The USBC assumes that fertility will instead increase from its current level to 2.22 in 2050, then declining slightly to 2.20 in 2075. The difference of about 0.25 in fertility implies an eventual difference of about 0.5 percent per year in the population growth rate, which is substantial. Further-
more, the SSA assumes a slower increase in life expectancy, to 81.7 in 2050 and 83.0 in 2075, vs. 84.0 and 87.1 for the USBC. Thus by 2075, Census has life expectancy higher by four years than the Actuaries. These differences in fertility and assumed mortality decline lead Census to predict more rapid population growth, and indeed, while the SSA projects growth at 0.3 percent per year in 2050 and 2075, Census projects 0.7
148
Lee & Edwards
percent per year in both years, more than twice as rapid. By 2075, the population projected by Census is 80 million greater than that by the SSA. Higher fertility eventually makes a population younger, while higher life
expectancy makes it older, so the differences in the projected old age dependency ratio (OADR) are not as great as otherwise. In 2050, the SSA OADR is lower, and in 2075, the Census OADR is lower.
2.5 Population Projections Based on this discussion of fertility, mortality, and immigration, it is straightforward to carry out the arithmetic of a population projection, generating forecasts for total population size and age distributions from which measures such as the OADR could be calculated. Before doing this, however, we wifi pause to consider how much confidence should be put in results of this sort.
3. THE UNCERTAINTY OF PROJECTIONS The public and the government have become well aware of the impending fiscal pressures that wifi be caused by population aging as the Baby Boom grows older. Major changes in social security and Medicare, the programs expected to be most heavily affected, are currently under consideration. Indeed, important changes in social security were already
made in the early 1980s, with the expectation that they would help restore long-run balance to the system far into the future. Those expectations now appear to have been too optimistic, for a variety of reasons. Might our current expectations about population aging and its consequences again turn out to be incorrect? Certainly they will. The projec-
tion horizon for social security is 75 years, but projection only a few years into the future is fraught with error. The question is not whether there will be errors, but rather how large and how important the errors wifi be. For this reason, it is very important that forecasts present not only the best guess about future outcomes, but also an indication of the uncertainty surrounding them. The typical projection results from assumptions about trajectories for several input variables, such as fertility, mortality, immigration, productivity growth, inflation, interest rates, and so on. The traditional way of assessing and conveying the uncertainty of a long-run projection begins by developing high, medium, and low trajectories for each of the input variables. It then bundles combinations of these trajectories together to calculate high, medium, and low projection scenarios. The way the bun-
dling is done depends on the purpose of the projection, and has an important influence on the results. Consider, for example, population
The Fiscal Effects of Population Aging in the U.S.
149
forecasts. The Social Security Actuaries bundle together high fertility, high mortality, and high immigration, because all of these conduce to a lower OADR and to lower projected costs per worker. Census bundles together high fertility, low mortality, and high immigration, because all of these conduce to more rapid population growth. The resulting high low bounds for projected outcomes wifi differ. For example, the Social Security projections of population size in 2075 range from 344 million to 486 million, whereas those from Census range from 304 to 809 million. At the same time, the Social Security projections for the OADR in 2075 range from 0.314 to 0.563, whereas those from Census range from 0.343 to 0.494. For population size, the highlow range of Census is 3.5 times as great as that of the Social Security projections. For the OADR, the highlow range for Social Security is 1.6 times as great as that of Census. Since they give very different indications of the uncertainty associated with their forecasts of different items, they cannot both be right. There are four ways in which the scenario-based approach to assessing uncertainty of forecasts is seriously flawed.2 First, by its very nature,
it is forced to make patently false assumptions about the correlation structure of forecast errors in the input variablesspecifically, that all the cross-correlations in errors are either +1.0 or 1.0. This problem results from the bundling just described. When Social Security bundles high fertility with high mortality, it assumes that a large positive forecast
error for fertility always goes with a large positive forecast error for mortality. Census assumes the opposite. The second problem is similar, but applies across time rather than across variables. The scenario method must assume that the input variables will either always follow the highest plausible trajectory or always follow the lowest one, thereby ruling out the possibility of long-run fluctuations like the Baby Boom, which could produce greater variations in some outputs such as the OADR. Here, it is assumed that the correlation of errors across time is always + 1.0. Third, the indications of uncertainty attached by the scenario method to differing outcome variables such as population size, 2
Although flawed for assessing uncertainty, scenario-based projections can be very useful for analytic purposes and for sensitivity tests.
In principle, one could calculate the error from each of the past Census or SSA forecasts of fertility and mortality, by comparing the forecasts with subsequent realized outcomes, and then these errors could be used to find the actual ex post correlation of errors in government agency forecasts of fertility and mortality. Unfortunately, this has not yet been done. Alternatively, we could fit time-series models to fertility and mortality, and examine the correlation of the residuals. This results in correlations that are insignificantly different than 0. However, the time-series models fit short-term movements in the series, whereas forecast errors arise most dramatically from errors in forecasting long-term levels and trends.
150
Lee & Edwcirds
births, fertility, life expectancy, and OADRs are inconsistent. This happens because each of these outcomes will reflect differing ways of averaging out the errors in forecasting inputs, with differing degrees of cancellation of errors, but the method is unable to take this into account, due to the assumed rigid covariance structure for errors. Fourth, the scenario
method is intrinsically unable to assign probabilities to its high-low ranges. (See Lee, 1999, and Lee and Tuijapurkar, 2000, for an extensive discussion of the problems with scenario-based forecasting.) The kinds of inconsistencies that result from scenario-based forecasts are illustrated in Table 1, which contrasts uncertainty ranges for different items in forecasts by Census and Social Security, and probabilistic forecasts by Lee and Tuljapurkar, to be discussed in a moment. We see that the range for children in 2050 is larger for Census than for Social Security; for workers, the Census range is twice as wide; for the elderly, it is three times as wide; but for the OADR it is only one-seventh as wide, while for the total dependency ratio it is very much wider. For Census, the working-age population is supposedly known within plus or minus 26 percent, and the elderly population within plus or minus 27 percent, yet their ratio, the OADR, is supposedly known within 3 percent! Clearly these indications of uncertainty are inconsistent. Similarly, for
Social Security, the total dependency ratio supposedly has only tiny uncertainty, which is again inconsistent. The probabilistic forecasts TABLE 1
High-Low Ranges for Forecasts of Selected Items to 2050, as Percentage of Middle Forecast R2nge (%)
Item
Children Working age Elderly OADR
Soc. Sec. Actuaries
Lee and Tuljapurkar
Census (1992)
(1992)
(1994)
±44 ±26 ±27 ±3
±31
±49
± 13
n.a.
±9 ±21
±10 ±35
±10
±0
±24
65+ /20-64
Total Dep. Ratio (<20 + 65+)/20-64
Calculated as (highlow)/(2Xmiddle). For Census, high minus low; for Social Security Actuaries, high cost minus low cost; for Lee and Tuljapurkar, upper 95-percent bound minus lower 95-percent bound.
The date of publication of the forecast is indicated; all are for the year 2050, which is the latest published by the Census Bureau. For Census, children are <18; for the others, <20. Elderly are always 65+. Lee and Tuljapurkar (1994) did not publish a probability bound for the working-age population, so none is shown.
The Fiscal Effects of Population Aging in the U.s.
151
shown in the last column have fully consistent indications of uncertainty, taking into account all covariances. Although the traditional and widely used scenario method for assess-
ing uncertainty of forecasts is seriously flawed, there are two other general approaches that are more useful. The first is analysis of the performance of past forecasts, and from that analysis, development of probability distributions for current forecasts on the assumption that the methods used and other circumstances are sufficiently similar in the past and future to make this useful. This approach is illustrated by the probability distributions provided for Congressional Budget Office (CBO) fore-
casts of the federal surplus (CBO, 2001). The CBO had only a short historical record of forecast performance to analyze, so its probability distributions were provided for only five years ahead, a serious limitation. Another difficulty is that a separate historical analysis of forecast errors must be conducted for each variable of interest. For an application of this approach in demography, see National Research Council (2000, chapter 7). For an application to the Social Security Actuaries' forecasting record, see Lee and Tuljapurkar (2000). The second approach is to develop stochastic forecasts that incorporate errors in the forecast of each input, and reflect their propagation through the forecast process. Here we will follow a variant of this second approach, in which timeseries methods are used to fit stochastic models for each input variable, and the propagation of errors is tracked through stochastic simulation. With this approach, a probability distribution can be calculated for any outcome of interest, including joint probability distributions for multiple
outcomes. In most cases, we constrain the central trajectory for each input (that is, the long-run mean) to match an assumption by the Social Security Actuaries, the Centers for Medicare and Medicaid Services (previously the Health Care Financing Administration), or the Congressional
Budget Office, but this is not the case for mortality. In all cases, the variances and covariances are estimated from the historical data series.
4. STOCHASTIC POPULATION FORECASTS The population forecasts we report below are distinctive in two respects. First, they reflect the choices for central trends in fertility, mortality, and immigration that we have just discussed, which differ from those in the projections by Social Security (in having lower mortality) and the Census Bureau (in having lower fertility). Second, they are probabilistic forecasts based on a new method (Lee and Tuljapurkar, 1994). Figure 1 shows the forecast for the OADR (defined as population 65 +
divided by the population 20 to 64). Focus first on the central Lee-
Lee & Edwards
152
FIGURE 1. Old-Age Dependency-Ratio Forecasts: 1999 to 2080 LT:97.5%
Lee-Tuijapurkar: median with probability intervals SSA: low, middle, and high USBC: low, middle, and high
0 (0
0
- -SSA LT:75%
0
.--USBC
0
LT:50% USBC
0
LT:25% USBC SSA
co
0
LT:25%
2000
2020
2040
2060
2080
Date
Tuijapurkar forecast, labeled "LT." Over the next decade, there wifi be a slight decline in the OADR, while the smaller generations born when fertility was low during the 1930s and early 1940s move into old age. The larger Baby Boom generations begin to reach 65 in 2010, and we see rapid aging for the next twenty years, slowing or slightly reversing as the Baby Boomers die off, and the smaller Baby Bust generations born since the mid 1960s enter old age. Then the OADR resumes its upward trend. Note that the Baby Boom generations are not the cause of popula-
tion aging; they merely usher it in. It is caused by low fertility and falling mortality, and unlike the Baby Boom, these are expected to continue indefinitely. The figure also plots quantiles of the probability distribution. There should be a 50-percent chance that the true future value falls between the 25- and 75-percent bounds, and a 95-percent chance that it falls between the 2.5- and 97.5-percent bounds. Before 2020, all the uncertainty in the ratio is due to uncertain survival of people through the working ages and of the elderly who are already born. After 2020, the uncertain births due to uncertain fertility in 2000 and thereafter begin to enter the working ages, adding increasing uncertainty to the ratio as time passes. After
2040, uncertain fertility is applied to uncertain numbers of young
The Fiscal Effects of Population Aging in the U.S.
153
women, compounding the uncertainty in the ratio. Finally, after 2065 the highly uncertain size of birth cohorts begins to affect the projected numbers of elderly in the numerator as well as workers in the denominator. Note also that there is nearly twice as much uncertainty in the upward direction as in the downward direction.4 By 2075, there is a 2.5-percent chance that the increase in the OADR wifi be twice as large as the central forecast, and a 2.5-percent chance that it wifi be only one-fourth as large as forecast. In any case, however, it is virtually certain that substantial population aging wifi occur over the next forty years. For comparison, Figure 1 also plots the central projections by Social Security (SSA) and Census (USBC), along with their non-probabilistic
high and low projection variants. We note that there are not major differences in the central forecasts, but that after 2040, the Census and Social Security projection ranges have much less than 95-percent probability coverage. The upward range for both Census and Social Security is close to the LeeTuljapurkar 75-percent bound, meaning that the true value would be expected to exceed the high bound for these projections about 25 percent of the time, if the LeeTuijapurkar probability distribution is correct.
5. HOW POPULATION AGING AFFECTS GOVERNMENT BUDGETS It is straightforward and natural to use population projections to project the future costs of benefits, on the assumption that program structures will remain as they are now. Such projections are useful for tracing out the implications of current policies, and thereby informing decisions about changing those policies. These exercises should be viewed only as conditional forecasts, however. Studies of the effect of population aging in the past on government budgets show much smaller effects, because in practice programs are adjusted. For example, Gruber and Wise (2001)
examined data for OECD countries over time, and found that a 10percent increase in the proportion of elderly in the population led to a 5percent increase in expenditures on the elderly, so that expenditures per individual old person declined while the aggregate expenditure on the elderly increased (that is, they found an expenditure elasticity of 0.5, measured relative to GDP). They also found that spending in other areas of the budget was reduced, so that total government expenditures as a share of GDP did not change with population aging. This is typical of the probability distributions for population forecasts. Population growth is multiplicative, so uncertainty is lognormally distributed.
Lee & Edwards
154
FIGURE 2. Benefits bj Program and Age
00 0 0 CU
Medicare
0 0 U)
0 0
Social Security
0
0 0 0 U) Other benefits such as public assistance and congestibles.
00
20
60
40
80
Age
Evidently population change does not dictate outcomes, but rather alters the trade-offs and constraints faced by policymakers. In the rest of this section, we will consider how this works. We begin by presenting the current cost of benefits received by age in the U.S., 13(x), and tax payments by age, T(x). Figure 2 plots 13(x), broken down by broad category of expenditure, but originally based on 25 individual or household benefit programs (school lunches, TANF, energy assistance, SSI) plus additional non-individual programs (roads, police, etc.) (see the Appendix for explanation of how these were estimated). The data refer to average amounts
per surviving individual at each age, so keep in mind that there are relatively few survivors to very old ages. They include all government expenditures at the federal, state, and local levels, except for expenditures on public goods (mostly defense spending). Expenditures which do not accrue to individuals or households are assigned on a per capita basis. The concentration of expenditures on children and on the elderly is apparent. The average elderly person receives over $20,000, which is about four times as much as the average child. Note that Medicaid expenditures for elderly people are primarily for nursing-home care.
Figure 3 plots 7(x), again per surviving individual at each age, and broken down by kind of tax (see the Appendix for details of construc-
The Fiscal Effects of Population Aging in the U.S.
155
FIGURE 3. Taxes by Program and Age
c
0 0
-
0 0 0
Other taxes including federal corporate tax and charges/fees.
0
20
I
I
I
40
60
80
Age
tion). Note that for some kinds of taxes, the elderly pay about the same amount as prime-age adultsnotably corporate tax (inferred from divi-
dend income), property tax, and sales tax. However, because they don't have much labor income, they pay far less payroll tax and income tax, and in total pay much lower taxes than prime-age adults. To see how population aging will affect the costliness of our current
agebenefit structure, we can calculate how the changing population age distribution would alter the ratio of total taxes to total benefit expenditures. We will call this ratio the fiscal support ratio. We could imagine an individual or a planner weighing the utility of receiving the benefit schedule 13(x) over the life cycle, vs. receiving the after-tax income that would be released by reducing or eliminating the programs that 13(x) comprises. While individual utility from the stream of benefits is distributed over future years of the life cycle, the cost in taxes is determined by the crosssectional balanced-budget constraint in each year, which is in turn deter-
mined by the population age distribution. This interplay between the individual life cycle and the cross-sectional population age distribution generates the fiscal effects of population aging. In an important sense, the population age distribution determines the price of the vector of lifecycle benefits, 13(x). This price is the ratio of aggregate taxes to benefits,
156
Lee & Edwards
FIGURE 4. Projected Fiscal Support Ratio b!,' Level of Government, 2000 to 2100
1.1
State and Local
0 0
0.8
Federal
0.7 I
2000
2010
2020
I
I
2030
2040
I
2050 Year
P
2060
2070
2080
2090
2100
Note: The fiscal support ratio is calculated as the ratio of tax revenues to government expenditures, based on the age-specific tax and expenditure profiles for 2000, applied to the projected age distribution for each period.
evaluated for the changing population age distributions in the future, using the current age profiles of taxes and benefits, 13(x) and 7(x). A similar calculation could be made using the projected profiles of taxes and benefits for some later year, and that would give somewhat different results. Figure 4 plots the changing ratio of taxes to benefits over the next century, based on the central population forecast. It can be seen that there is hardly any effect at the state and local level; the ratio is quite constant over the century. At the federal level, however, population aging leads to a far bigger increase in benefit costs than in tax revenues. The same level of taxes represented by 7(x) would buy a level of benefits, 13(x), only 64 percent as high in 2075 as in 2000. Put differently, we might say that population aging wifi raise the price of this benefit bundle 13(x) by 56
The Fiscal Effects of Population Aging in the U.S.
157
percent (0.64 = 1/1.56) over this century, in terms of after-tax income.5 As population aging alters this price, we might expect voters and policy makers to choose a lower level of the benefit age profile (x), and correspondingly more after-tax income for taxpayers.6 The net effect on aggregate benefit expenditures would be ambiguous. Although there will be a
greater number of elderly people, each of them would receive lower benefits, and consequently the net effect of population aging on both aggregate benefit expenditures and on aggregate taxes and tax rates would be ambiguous. This interpretation, although ignoring the costs of transitions between program regimes, is broadly consistent with Gruber and Wise's (2001) results described above. Note also that Figure 4 is based on the current program structure, and so it does not reflect the large expenditure increases per beneficiary that are projected for Medicare and Medicaid over the course of the twentyfirst century, due to projected increases in the quality and quantity of services consumed (Lee and Miller, 2001b). Figure 4 showed a single forecast of the support ratio, as if we actually knew what the future would bring. Figure 5 presents probabilistic forecasts of the support ratio, showing the median value (which was plotted in Figure 4) along with 95-percent probability intervals. In early years, uncertainty results largely from uncertainty about fertility; the effects of uncertain mortality emerge only over the longer run. Variations in fertility have a strong effect on state and local finance once they affect the
number of children of school age, that is, at age 5 or older. Thus the probability band for the state and local support ratio is very narrow for the first five years of the forecast, and opens up rapidly thereafter. The number of children has relatively little effect on the federal budget until they grow old enough to enter the work force and begin paying taxes, beginning around age 20. Even then, the steep slope of 7(x) implies that uncertain fertility does not have a large effect on taxes for a number of
years after that. Thus, the probability interval for the federal support ratio is very narrow for the first thirty years or so, and then widens as uncertainty about the size of the labor force grows. Uncertain mortality Strictly speaking, this interpretation makes sense only when all difference in population age distributions is due to change in fertility, not mortality, and the system is unchanging over time. When mortality is declining, then the expected value of the benefit package over the life cycle wifi rise, since the expected duration of receiving benefits in old age increases. When the program system is changing over time, then the link between individual benefits over the life cycle and the current benefit package is not tight. In reality, the shape of /3(x) could also be changed, for example by favoring programs for children at the expense of programs for the elderly. Population aging also alters the cost of providing benefits to a child relative to the cost of providing benefits to an elderly person.
158
Lee & Edwards
FIGURE 5. Projected Fiscal Support Ratio by Level of Government, 2000 to 2100 (Median and 95% Probability Interval) Co
.4-a
1
ci)
0
4-a
Co
C" 1
ci)
State and local
><
Cj
I0 0 0 .4-a
0 cc
0
00
Federal
C,)
0U)
0
Note: The Fiscal Support Ratio is calculated as the ratio of tax revenues to government expenditures, based on the age specitic tax and expenditure profiles for 2000, applied to the projected age distribution for each period. Stochastic population projections were bused on the methods in Lee and Tuljapurkar, 1994.
LL I
I
I
I
2000
2020
2040
2060
I
2080
2100
Year Note: The fiscal support ratio is calculated as the ratio of tax revenues to government expenditures, based on the age-specific tax and expenditure profiles for 2000, applied to the projected age distribution for each period. Stochastic population projections were based on the methods in Lee and Tuijapurkar (1994).
contributes a small amount of uncertainty to the support ratio, but not much. It is clear that demographic change is almost certain to cause serious pressures on the federal budget as the Baby Boom generations enter old age. Through 2040 or so, budgetary pressures can be projected with great confidence. After this it is not so clear whether pressures wifi continue to mount, or somewhat abate. At the state and local level, the median support ratio shows no trend, but there is a great deal of uncertainty. It is not clear that there would be any advantage to planning for a growing school-age population, when that population is just as likely to decline, relative to taxpaying workers.
The Fiscal Effects of Population Aging in the U.S.
159
In the long run, there is a negative correlation between errors in forecasting the state and local fiscal support ratio and the federal one. High fertility is costly for the state and local entities providing public education, but it allows lower taxes at the federal level, since it generates more
workers to support the elderly. Thus, there is less uncertainty in the total fiscal support ratio than one would expect from looking at its constituent parts.
6. CONSTRUCTING STOCHASTIC BUDGETARY FORECASTS We can build on these stochastic population forecasts to develop stochastic projections of government expenditures, assuming that the basic structure of programs is unchanged. To do so we need first to develop the linkage of population forecasts to costs of benefits, and second to incorporate some sources of economic uncertainty.7 Population is linked to benefit costs by the age schedule of costs of benefits currently received by a person in age group x. This average benefit profile is the 13(x) presented earlier in Figure 2. This schedule cannot be expected to remain fixed in the future, however, even under the assumption that program structure remains fixed. Benefits for most programs can be expected to rise as productivity increases. We will follow CBO in assuming that most benefits rise in real cost at the same rate as productivity growth, which raises per capita incomes and labor costs. Some programs, notably social security, Medicare, and Medicaid, require special treatment, however, as we now discuss. For social security, we take into account the legislated change in the normal retirement age from 65 to 67 in the coming decades. Our projec-
tions of benefits are based on the actual rules governing benefits in relation to prior earnings (see Lee and Tuljapurkar, 1998a and 1998b, for
details), and indirectly take into account such particulars as the notch generation, the selective effect of mortality at older ages, and the effects of loss of spouse on benefit levels. Benefit costs for Medicare have typically been rising much more rapidly than productivity growth, and are expected to do so for the foreseeable future. We constrain our median projection for health care costs per enrollee at each age to follow the CBO (2000) assumptions (which are very similar to the 2001 HCFA/CMMS projection assumptions in Board of Trustees of the Federal Hospital Insurance Trust Fund, 2001) in which the rate For a number of years, CEO published stochastic long-term forecasts based on these LeeTuljapurkar stochastic population forecasts, with deterministic economic variables.
160
Lee & Edwards
of increase per enrollee declines to an eventual level 1 percent per year more rapid than the growth rate of productivity. We differ, however, in taking into account the distribution of the population at each age by time until death. The Medicare costs of individuals have been shown to be closely associated with their proximity to death (Lubitz and Prihoba, 1984; Lubitz, Beebe, and Baker, 1995; Miller, 2001). In a projection, we know for
each year what proportion of people at a given age will die within one year, one to two years, ten years, and so on, and can allocate health costs accordingly. Time until death thus serves as a kind of index of health status. We apply the rate of increase of per enrollee cost to each category of time until death separately (see Lee and Miller, 2001b, for details). For Medicaid, we note that the proportion of the elderly population in long-term care facilities at each age has been declining for some time, presumably due to the improving health of the elderly population. We project this decline to continue, which partially offsets the increasing costs of care for those in institutions.
7. DETERMINISTIC FORECASTS OF THE FISCAL EFFECTS OF AGING We will begin by considering some deterministic projections, then turn to stochastic ones. Figure 6 plots projected government expenditures as shares of GDP for the federal government, and for state and local governments grouped together, as well as their sum. Excluded from these totals are interest payments on the federal debt, and benefits paid for pre-funded programs such as most state and local pensions and some insurance funds. Total expenditures are initially 25 percent of GDP, but are projected to rise above 40 percent of CDP by 2075 and to continue climbing thereafter. For
the federal budget and overall, there is an acceleration in the rate of increase between 2010 and 2030 when the Baby Boom is reaching old age, but clearly that is only a part of the story, since the trend continues rapidly
upward after 2040. At the state and local level, expenditures rise only mildly relative to GDP. Almost all the increase in the total is due to increases at the federal level, which is not surprising given the importance of federal transfers to the elderly. Federal expenses increase from 16 percent of GDP in 2000 to 30 percent iii 2075, almost a doubling, and by 2100 they are approaching 40 percent. It is also interesting to separate these expenditures by age group of the recipients. We define three categories: spending on the elderly, spend-
ing on children, and programs that are age-neutral. We have assigned
The Fiscal Effects of Population Aging in the U.S.
161
FIGURE 6. Government Expenditures as Shares of GDP
0.5
0.4
0.2
State and Local 0.1
0
2000
2010
2020
2030
2040
2050 Year
2060
2070
2080
2090
2100
each program to one of these three categories, based either on the nature of the program or on some criterion such as the average dollar-weighted age of the recipient. Figure 7 shows the result for all levels of government combined. Expenditures for children are flat over the next 100 years, relative to GDP. Age-neutral expenditures show some growth, but only to the extent that they include the non-institutional component of Medicaid, which grows faster than GDP due to excess growth in per capita health care costs. Almost all the projected increase in government spending over the next 75 years and beyond is due to increased expenditures on programs for the elderly. These rise from about 8 percent of GDP in 1999 to 21 percent of GDP in 2075, and they more than triple their share by 2100. It is also illuminating to look at the growth in expenditures by kind of program, rather than by age of recipient. Figure 8 shows the growth in projected expenditures for retirement programs (OASDI, federal employ-
ees, and railroad workers), health programs for the elderly (Medicare Parts A and B and institutional Medicaid), other expenditures for the
Lee & Edwards
162
FIGURE 7. Government Expenditures per GDP b Age Group
0.3
0.25
Elderly 0.2 0
0 0 0.15
AgeNeutral
0.1
Children 0.05
200
2010
2020
2030
2040
2050 Year
2060
2070
2080
2090
2100
elderly, and all other federal expenditures. It is striking that the growth in expenditures for retirement programs, including social security, is such a small part of the projected growth in federal spending, contrary to the attention allocated to retirement programs in public discussions. Retirement accounts for only one-eighth of the total growth, with most of the rest due to growth in health care for the elderly. This very low
share is in part due to the assumption of a 2.3-percent mean rate of productivity growth, which is 1.0 percent higher than the Social Security Actuaries' assumption, and which in itself would improve the summary
actuarial balance measure from 1.89 percent of the present value of payroll to only 0.89 percent, making more than half of the projected imbalance disappear. Without this assumption, retirement programs would account for about a quarter of the projected expenditure increase. The projected increases in health care for the elderly are roughly half
due to population aging (reflecting low fertility rather than mortality decline) and half to increases in costs per enrollee in excess of productivity growth (Lee and Miller, 2001b). The new assumptions by CBO and
The Fiscal Effects of Population Aging in the U.s.
163
FIGURE 8. Federal Expenditures per GDP by Type of Spending 0.4
0.3
0 0
+ NonElderly
0, ci) ci)
+Otherci)
E
0 + Health Care 0.1
Retirement
2000
2010
2020
2030
2040
2050 Year
2060
2070
2080
2090
2100
HCFA/CMIvIS on this excess rate of cost growth have a powerful influence on these projections.
8. STOCHASTIC BUDGETARY PROJECTIONS So far we have not discussed how economic uncertainty is incorporated in our projections. We treat productivity growth and (where relevant)
real interest rates and stock market returns as stochastic, following a modeling strategy similar to that used for fertility in the demographic projections (see the Appendix for details). That is, we model these as stochastic time series, and fit the models on historical data. The models we fit are constrained to have mean values that are consistent either with comparable official projections or the historical record, depending on the purpose of the forecast. Matching social security, our real interest rate
averages 3 percent per year; we set labor productivity growth at 2.3 percent per year, roughly its postwar average; and real stock market
164
Lee & Edwards
returns are 7 percent per year, reflecting historical trends in the S&P 500. Thus for the most part, our fitted models are providing the structure of errors for our forecasts of economic inputs, but not their mean or median values, which are rather imposed. For the productivity growth rate, the
standard error of the one-step forecast is 1.78 percent, so a 95-percent
probability interval has a width of 7 percent, wide indeed. For the interest-rate model, the one-step forecast has a standard error of 2.04 percent, so the width of a 95-percent interval is over 8 percent. These intervals are very much wider than the highlow assumption ranges of Social Security, which have a width of 1 percent for productivity growth and 1.5 percent for real interest rates. However, it must be borne in mind
that the stochastic interval refers to realized values in a single year, whereas the Actuaries' assumptions can best be thought of as referring to a long-run average. The actual stochastic forecast is then carried out through stochastic simulation. A single stochastic trajectory is calculated by drawing random numbers to determine the forecast errors for the first year, which are then inserted in the appropriate equation for each input, along with
the previous years' values, leading to a one-step forecast. Then the forecasts of population and benefit costs are derived mechanically from these forecasts of inputs. Then a second round of random numbers is
drawn to generate the second year of the forecast, and so on. In this way one stochastic trajectory is forecast. We generate many such trajectories, generally at least a thousand, and then use the frequency distribution for outcomes of interest to estimate the probability distribution of the forecast. Outcomes include total expenditures on benefits, expendi-
tures for a particular program, the date of Trust Fund exhaustion for social security, or the Trust Fund ratio, and so on. If desired, we can also project tax revenues in a similar way, and we can constrain tax rates
to be adjusted so as to maintain some target such as a pre-specified debt-to-GDP ratio.
9. STOCHASTIC PROJECTIONS FOR SOCIAL SECURITY We now turn to stochastic long-term projections of the finances of the social security system, drawing on Lee and Tuljapurkar (1998a, 1998b). We have already described the methods we have used, so we can move directly to results. Perhaps the most basic statistic is the cost rate, that is, the costs of benefits in a given year as a percentage of payroll in that year. In a pure pay-as-you-go system, with no accumulated trust fund,
The Fiscal Effects of Population Aging in the U.S.
165
FIGURE 9. Cost Rate (Outgo as a Percentage of Taxable Payroll) 35
20
15
2000
2010
2020
2030
2040
2050
20&3
2070
Year
this would be the payroll tax rate for each year. Figure 9 plots various probability quantiles and the mean of the cost rate for each year through 2075, with a projection base year of 2000. These runs are based on the productivity growth-rate assumption of Social Security, so aside from our forecast of more rapid mortality improvement, our central forecasts should match closely those of the Social Security Actuaries for the same base year (Board of Trustees, 2000). The results reported here were generated by a stochastic simulation program written by Michael Anderson and Shripad Tuljapurkar,8 which can be accessed free via the Internet at http: / /simsoc.demog.berkeley.edu. Users can modify many aspects of the policy environment, including plans for investing a portion of the Trust Fund in equities, raising the age at retirement, and raising the payroll tax rate. 8 The results presented in Figures 9-12 are based on the output of this program, which is currently in beta testing and not guaranteed to be bug-free.
166
Lee & Edwards
We see that by 2075, the median cost rate is 21.2 percent. There is a 2.5percent probability that the cost rate wifi be only 14.6 percent, but also a 2.5-percent probability that it wifibe at least as high as 36.5 percent. These figures can be compared with the Social Security projections (Board of Trustees, 2000), which give 19.5 percent for the intermediate trajectory and 13.9 to 28.3 percent for the range. Our central forecast is about 2 percent higher, due to the more rapid decline in mortality that we project. Our lower 2.5-percent bound is similar to the SSA low cost scenario, but our high 2.5-percent bound is almost 7 percent higherconsistent with Figure 1, which showed much larger uncertainty in the upward direction for the OADR than indicated by either Social Security or Census. By construction (see the Appendix) the long-run means of our forecasts for fertility, for the productivity growth rate, and for the real interest rate are identical to those assumed by the Social Security Actuaries, while our projected life expectancy for 2075 is about three years higher. We have compared our outcomes with those of the Actuaries, but it is important to note that ours are probabilistic whereas theirs are deterministic ranges and have no probabilistic meaning. It appears that the Actuaries' low cost scenario matches our lower 2.5-percent bound for the cost rate, while their high cost scenario corresponds roughly to our upper 86percent bound. That is, the chances that the cost rate wifi exceed the Actuaries' high cost boundary are more than five times greater than the chances of its failing to reach the low cost boundary. There is relatively little uncertainty in the income rate, that is, tax income as a proportion of payroll, since the only uncertainty comes from revenues from taxes on benefits. However, the highly uncertain cost rate leads to large uncertainty in the various measures of net outcome. For
example, our forecasts find a median date of fund exhaustion under current policy of 2038, very close to the intermediate projection of the Actuaries, 2037. We find a 2.5-percent chance of exhaustion by 2024, versus 2027 for the high cost projection of the Actuaries. We also find roughly a 4-percent probability of exhaustion after 2075, compared to no exhaustion by 2075 for the Actuaries' low cost projection, which also has a healthy Trust Fund ratio at that point. The Actuaries finds that an immediate rise in the payroll tax of 1.89 percentage points should restore actuarial balance over the 75-year horizon. We have also simulated the outcome assuming taxes are raised in
this manner, from the current rate of 12.4 percent to 13.29 percent. Figure 10 depicts a histogram of the 1,000 probabilistic dates of exhaustion generated by our model under such a policy. Our method makes explicit what may be fairly intuitive: An immediate rise in payroll taxes designed to restore actuarial balance wifi only prevent Trust Fund bank-
The Fiscal Effects of Population Aging in the U.s.
167
FIGURE 10. Histogram of 1,000 Dates of Exhaustion with Immediate Payroll Tax Increase of 1.89%
00
(Y)
0 C\J
0 00 U)
2.5% by 2036 16.7% by 2048 50.0% by 2070 83.3% after 2099+ 97.5% after 2099+
1
0 0 1
0 U) 0 2030
2040 2050 2060 2070 2080 2090 2100+ Year
ruptcy roughly 50 percent of the time. More strikingly, the lognormal dispersion of exhaustion dates in Figure 10 implies that the mode of the distribution actually occurs much earlier than 2075. Even a painfully large hike in the payroll tax today does not move the most frequently realized future date of bankruptcy past about 2055. Figure 11 displays a histogram showing 1,000 realizations of the 75year actuarial balance dating from 2000, under the prescribed 1.89percentage-point rise in the payroll tax. The long left tail indicates that the chances of undershooting actuarial balance, denoted by 0 on the horizontal axis of the graph, are more widely dispersed than the chances of overshooting. That is, although the risks are roughly balanced under an immediate payroll tax hike, the downside risks are more costly. Our stochastic framework lends itself particularly well to analyses involving social security's finances and risk. During the Clinton administration,
168
Lee & Edwards
FIGURE 11. Histogram of Actuarial Balances with Horizon to 2074
z.
-
-i
Actuarial balance
policymakers considered plans to invest part or all of the Trust Fund in equities, and currently the Bush administration is said to be weighing the option to replace part of the system with private accounts. An assessment of the riskiness of such plans is important in light of the uncertainty of stock market returns. Figure 12 presents a histogram of Trust Fund exhaustion dates under a particular investment plan: immediately placing 20 percent of the entire Fund balance in the S&P 500 in 2000, and increasing that share to 50 percent by 2010. The most striking characteristic of Figure 12 is that the distribution peaks soon around 2030-2035 and
then tapers off very rapidly, even though the median date of Fund exhaustion is 2044. This dynamic is due to the risky nature of stock returns, which may potentially help social security's finances considerably, but at the same time wifi not change the expected date of bankruptcy very much. We have also simulated the effects of investing 75 percent of the Trust Fund in equities immediately. This leads to a mean fund balance equal to three times GDP in 2074. However, the median
The Fiscal Effects of Population Aging in the U.S.
169
FIGURE 12. Histogram of 1,000 Dates of Exhaustion: Investing 20% in 2000, Rising to 50% by 2010
00
C\i
2.5% by 2025 16.7% by 2032 50.0% by 2044 83.3% after 2099+ 97.5% after 2099+
0 IA)-
-
1
0 0
0 LU
0 2020
2040
2060
2080
21 00+
Year
fund balance is minus half of GDP, and there is still a 63-percent chance of insolvency within 75 years. The mean outcome is a misleading measure because it is strongly affected by the long upward tail of the distribution of favorable stock market outcomes.
10. STOCHASTIC PROJECTIONS OF MEDICARE COSTS Detailed projections of Medicare costs as a share of projected GDP are developed in Lee and Miller (2001b). We wifi not describe the methods or results in detail here, but rather wifi summarize some of the main points. Using the methods described earlier, the LeeMiller median projection is that Medicare expenditures wifi rise from 2.2 percent of GDP now to 11 percent by 2075, quintupling. These dynamics are shown in
170
Lee & Edwards
FIGURE 13. Medicare as a Share of GDP: Median with 95-percent Probability Interval
0 C"
2000
2020
2040
2060
Year
Figure 13, where the middle line represents the median trajectory of Medicare's share of output. The LeeMiller forecast stays close to the CBO forecast until it terminates in 2040. More rapid mortality declines projected by Lee and Miller have offsetting effects on projected health costs, since, as discussed above, lengthening life spans generally coincide with lengthening healthy life spans. The projected increase in Medicare costs is huge, more than twice as big, as a share of GDP, as are total expenditures on social security today. Not surprisingly, however, the projected increase is highly uncertain. Before turning to the probability distribution for the forecast, however, we should consider the potential sources of error. First, there is demo-
graphic uncertainty. However, note that uncertainty arising through mortality is largely canceled by the parallel uncertainty in the health status of the population: If people live longer, it wifi be because they are
The Fiscal Effects of Population Aging in the U.S.
171
in better health, or so the time-until-death approach assumes. Second, note that uncertainty about the rate of productivity increase is also filtered out, once we express the costs relative to GDP. If productivity growth is 1 percent per year more rapid than expected, then by assumption health costs will also grow 1 percent per year more rapidly than expected, as will GDP. The ratio of total costs to GDP is unaffected. Therefore the most important sources of uncertainty in the ratio of costs to GDP will be fertility and the size of the gap between the rates of increase in per-enrollee costs and productivity growth. We have fitted a time-series model to this gap over the past 50 years and then used it to assess the variance structure of the gap.9 With this background, we found that the 95-percent probability interval for the cost-to-GDP ratio in 2075 is 5 to 26 percent, as shown in Figure
13. That is, there is a 97.5-percent chance that the ratio will at least double, and a 2.5-percent chance that it wifi increase at least twelvefold. The uncertainty in the upward direction is more than twice as great as in
the downward direction, similar to results we have seen before. This range of uncertainty reflects the U.S. experience with cost containment in the 1990s as well as earlier periods of more rapid growth. It is indeed difficult to plan for the future when there is so much uncertainty. Comparing our 95-percent intervals with the CBO projections, we find that while in later years of their forecasts their highlow range is
similar to our 95-percent range, for earlier years their range greatly understates the uncertainty. This is a common problem with the scenario approach for assessing uncertainty in projections.
11. DISCUSSION Population aging is virtually certain to occur in the coming decades, and
it wifi have a serious impact on the costliness of many government programs. We have assessed the fiscal pressures of population aging by examining its impact on many age-assignable government programs, as well as on tax receipts. However, recent economic change has underlined the dangers of ignoring the role of chance in formulating our plans. Many projections simply assume that the short-run or long-run future wifi unfold according to the pattern of the past few years, which is a risky practice. Good forecasts ought to provide some measure of this risk. Yet the scenario method, which is most widely used to incorporate We have followed the suggestion of the Technical Advisory Panel for HCFA/CMMS in using a more general measure of health costs to calculate this gap, rather than specifically Medicare costs, and therefore we are able to go back in time before Medicare was launched in 1965.
172
Lee & Edwards
uncertainty in government forecasts, is seriously flawed. We, together with collaborators, have developed new and explicitly probabilistic methods for forecasting population and government expenditures, based on
analysis of historical variability combined in many cases with expert judgement about central trends. Thus our analysis has had two goals: to examine the fiscal effects of population aging, and to do this in a probabilistic setting using stochastic simulation.
Beginning with the demography, we find that the OADR is virtually certain to increase by more than 50 percent in the 2030s. While it is possible that it wifi decline a bit thereafter as the Baby Boom generations die, more likely it will continue to increase after 2050, possibly by a great
deal. The chance of very high ratios is substantially greater than indicated by Census or Social Security projections. Population aging raises the cost of the current structure of government programs (including those for children) relative to tax revenues, and makes a given package of life-cycle benefits more costly relative to the life-cycle tax payments necessary to fund it. We find that population aging is virtually certain to increase the costliness of current federal programs by 35 percent (±2 percent) by the 2030s, and with less certainty by 60 percent (± 15 percent)
in the second half of the century. Although population aging will not affect the costliness of average state and local programs in the mean or median forecast, there is considerable uncertainty about this (±20 percent or so) after 2020. We expect that governments wifi respond to these aging-induced cost changes by altering program structures, as they have in the past. Although it is unlikely that the current program structure wifi remain unchanged, it is nonetheless useful to project the consequences of maintaining it. Under this assumption (while the retirement age rises as cur-
rently legislated and health care costs per enrollee rise as projected), federal expenditures are projected to rise dramatically relative to GDP, from 16 percent of GDP in 2000 to 30 percent in 2075, almost a doubling,
and by 2100 they are approaching 40 percent (these figures exclude interest payments on the debt and payments into pre-funded programs). State and local expenditures rise only modestly relative to GDP. Almost all of this increase is for programs going primarily to the elderly, which rise from 8 percent of GDP in 1999 to 21 percent of GDP in 2075 and which more than triple their share by 2100. Programs for health care for the elderly account for the greatest part of this increase, with pensions a distant second. Looking specifically at the social security system, although we believe the Actuaries underproject future mortality improvements, we are im-
The Fiscal Effects of Population Aging in the U.S.
173
pressed by the quality of their projections. However, we find that they underestimate the risk of very costly outcomes. According to our probabilistic projections, the chances that the cost rate will exceed the Actu-
ary's high cost boundary are more than five times greater than the chances of its failing to reach the low cost boundary. Raising the payroll tax rate by 1.89 percent, which according to the Trustees Report of 2000 would have put the system into 75-year actuarial balance, has relatively
little effect on the probabilities of early exhaustion, raising the 2.5percent bound of exhaustion from 2024 to 2036, while raising the median date of exhaustion from 2036 to 2070; there would still be a 55-percent
chance of insolvency within the 75-year horizon, with a median Trust Fund balance after 75 years of 6 percent of GDP. Investing some or all of the Trust Fund in equities may help solve the long-run problem in terms of average outcomes, but not in terms of more important measures such as the median outcome or the probability of insolvency. Looking specifically at Medicare, which now costs 2.2 percent of GDP, we found a median share in 2075 of 11 percent, five times as great. The 95-percent probability interval for 2075 is 5 to 26 percent of GDP, so that there is a 97.5-percent chance that the ratio wifi at least double, and a 2.5-percent chance that it wifi increase at least twelvefold. The uncer-
tainty in the upward direction is more than twice as great as in the downward direction, reflecting lognormality, as we have seen before. Because probabilistic forecasts have only recently become available, research on their uses and implications has barely begun. The immedi-
ate impulse is to treat these forecasts as if they simply provided an improved highlow range. In fact, they contain much more information than that, and they can support more powerful uses and analyses. One key question is how uncertainty should affect our planning. Should the possibility of worse outcomes lead us to take additional precautionary measures today, or should the possibility of better outcomes
lead us to postpone action until we are sure action wifi be necessary? (See, for example, Auerbach and Hassett, 2001). Another important question is how different kinds of policies perform in the context of uncertainty. Do some reduce the uncertainty and others amplify it? For example, indexing retirement benefits to life expectancy at retirement (as has been done in Sweden) wifi reduce uncertainty for the pension system arising from future mortality, by passing on the consequences of the uncertainty from the taxpayers to the beneficiaries. Medicare costs turn out to be only slightly affected by uncertainty in future mortality, because of offsetting effects of health improvement on numbers of enrollees and costs per enrollee. Using the stochastic simulations as a kind
174
Lee & Edwards
of experimental laboratory, various policies can be assessed in terms of criteria such as intergenerational equity, rapidity of changes in taxes, rates of return, and so on.
APPENDIX: METHODS USED FOR STOCHASTIC PROJECTIONS A.1 Demographic Projections
A.li Mortality Let m(x,t) be a central death rate for age [x,x+5) and time [t,t+1). Suppose we have a matrix of X age-specific death rates over T years. The LeeCarter method estimates the model: In
= a + b1k
+ 8x,t
(A.1)
using a singular-value decomposition (SVD) or some other appropriate method. This yields estimates of a, b, and k. A second-stage procedure adjusts k to match exactly the life expectancy at birth implied by the for each year t. We now have a time series of k over T years (for most purposes, we have used data from 1950 to 1999; for some purposes, we start in 1900). This time series is modeled using standard BoxJenkins methods. (Tests for covariance with the residuals from the fertility model described below showed no association, so they were modeled independently). In most applications, it is well fitted by a random walk with drift. The fitted
model for k can then be used to forecast k over the desired horizon, together with a probability distribution for each forecast year:
= k1 1.029 +,
s.e.e. = 1.366.
(A.2)
(0.195)
From the forecasts of k, using equation (A. 1), probability distributions and
mean or median values of 1n and the implied life expectancies can be calculated, along with probability distributions. These probability distributions wifi typically reflect the innovation error in k, along with the uncertainty of the estimate of the drift in the k process. They typically will not include the 8xt terms, nor the uncertainty in the estimates of the a and b, which do not add much to the uncertainty after the first decade or two. On all of this, see Lee and Carter (1992) and Lee and Miller (2001a).
A.1.2 Fertility A similar approach is followed, but the fertility rates themselves, rather than their logs, are modeled. The model for agespecific fertility g is
The Fiscal Effects of Population Aging in the U.s.
g
= c +
df, +
175
(A.3)
which is again estimated using a SVD. Time-series models applied to the history of fertility in the U.S. do not provide a plausible model or forecast for fertility, for various reasons, so the mean of the forecast is constrained to equal a level specified ex ante, and in practice is taken to equal the ultimate level of fertility assumed by the Social Security Actuaries,
currently 1.95 children per woman. The fitted time-series model then provides crucial information about the variability and autocovariance of fertility. See Lee (1993) for a discussion of all these issues, and exploration of some alternative modeling strategies. The fitted fertility timeseries model is = 0.96f - 0.0037 +
+ 0.52v,
(A.4)
where the standard deviation of v is 0.11.
A.1.3 Immigration Immigration was projected deterministically following the assumption of the Social Security Actuaries, since it was thought better to treat it as a policy instrument than to attempt to forecast future policy.
A.t4 Population Forecasts Initial conditions for the forecast come from the base-period population age distribution, taken from social security data. A single stochastic sample path is generated by drawing random numbers for the errors in the fertility and mortality equations, and thereby generating a trajectory of age-specific fertility and mortality rates over the desired horizon, say 100 years. Sample paths containing a total fertility rate below 0 or greater than 4 are discarded. In remaining paths,
any negative age-specific birth rates are set to 0. These are combined with the deterministic immigration rates. Using well-known accounting identities, the population forecast by age group is then calculated for this single sample path. The procedure is then repeated many times, sometimes 1,000 times and sometimes 10,000 times. The frequency distributions of outcomes of interest then provide estimates of the probability distributions for these outcomes, and joint distributions can be provided in a similar way.
A.2 Economic Projections A.2.1 Productivit, A demographically adjusted productivity growth series was constructed. First, an average wage profile by age and sex was calculated from the 1997 March Current Population Survey (CPS). Data
176
Lee & Edwards
on the agesex composition of the labor force were also taken from the CPS, from 1948 to the present. The effect of the changing agesex composition of the labor force, based on these agesex weights for wages, was then calculated for each year since 1948 and used to adjust the official measure of productivity growth in the private non-farm business sector to remove the effect of the changing demographic structure of the labor force. The adjustment made little difference in general.
Next, a constrained-mean time-series model was fitted to the adjusted productivity growth series. As with fertility, the time-series model provides information about the variance, autocovariance, and cross-covariance of the series, but not about the long-run mean, which is imposed. An autoregressive model of order one was found to fit the data best:
g-
= /3:g-1 - aLtg) + 8gt,
(A.5)
where /3 is estimated at 0.1640 with a standard error of 0.1408. The standard deviation of 8g,t is 0.0178. (The model was run on data in whole units rather than percentage points.) For general budget forecasts, including social security and other individual programs, before estimating the AR(1) we constrain the long-run mean gtO the historical post-W.W. II (arithmetic) average labor productivity growth rate, 2.3 percent. This is nearly 1 percent per year higher than the SSA assumption on labor productivity growth, but a bit below recent Congressional
Budget Office assumptions. For some runs in which we are looking solely at social security, and wish to contrast the stochastic forecasts with deterministic ones, we constrain productivity growth to match that assumed by the Social Security Actuaries. Note that the rate of growth of covered wages in the social security system, which is a central component of the Actuaries' projections, is less rapid than labor productivity growth, since the former is affected by changing hours of work and by the changing share of total compensation that takes the form of untaxed benefits. We assume the wedge is 0.3 percentage point. A.2.2 Interest Rates The bonds held in the Social Security Trust Fund are a special Treasury issue with a rate of return equal to an average of
rates on longer term Treasury bonds. We use this special issue rate, minus the rate of inflation as measured by the CPI-U, as our baseline real interest rate. All balances in government trust funds and govern-
ment debt held by the public earn interest at either this rate or a moving average of it, the latter of which is intended to capture the effect of the broader array of maturities held outside the Social Security
The Fiscal Effects of Population Aging in the U.s.
177
Trust Fund. To project the real interest rate, we fit a VAR of order three
that recognizes the conjoined behavior of bond returns (re) and real stock returns (se):
r = a1r_1 + a2r_2 + a3rt*_3 + Yi-i + Y2-2 + y3s3 + 8r,t' (A.6) where asterisks denote detrended variables. (The analogous equation for stock returns in the VAR is not shown.) Point estimates of the a's and '/s are, in order, 1.1785, 0.8029, 0.4826, 0.0065, 0.0232, and 0.0052. The standard deviation of e, is 0.0204. As with fertility and productivity growth, we constrain the long-run means of both series. Real stock returns are assumed to average 7.0 percent, while the long-run real interest rate is assumed to be 3.0 percent, mirroring the SSA assumption.
A.2.3 Economic Projections We begin with a stochastic sample path for population. From a recent March CPS, we calculate age-specific aver-
age labor earnings over all people in an age group (sometimes also by sex). We adjust this age profile multiplicatively so that in conjunction with the initial population age distribution, it implies aggregate labor earnings equal to a control total from the National Income and Product Accounts (NIPAs). Then a stochastic sample path for adjusted productivity growth is generated. For each forecast year, this sample path is used
to shift the ageearnings profile multiplicatively. Together with the population sample path, this results in an implied aggregate level of labor earnings for each forecast year. If needed, a GDP forecast is also generated by assuming that labor earnings are a constant share of GDP, as is implied by the behavior of an economy with CobbDouglas production that is in a steady state.
A.3 Fiscal Impact Projections A.3.1 Age Profiles Cross-sectional age profiles of per capita benefit costs (per member of the population, not just recipients) are estimated for the base period, using March CPS data for some programs, using adminis-
trative data for some others such as social security and Medicare, and sometimes combining data from surveys and administrative records. All profiles are adjusted to yield totals consistent with aggregate control totals when multiplied by the base-period population age distribution.
A.3.2 Social Security Benefits For these, a cross-sectional profile is not appropriate, because each generation's benefits are pegged to a proportion of its earnings level just before it retires, and the real benefit level is not subject to economic uncertainty thereafter. However, a generation's
178
Lee & Edwards
benefits do vary on a per capita basis as it ages, due to selective mortality, to changing proportions of widow(er)s and married couples, and to the continuing labor supply of some members combined with greater deferred benefits. We estimate this age-specific benefit profile for the older ages, and peg its initial value based on the stochastic productivity level just prior to retirement for each generation.
A.3.3 Health Costs The basic strategy was described in the text. Per enrollee health costs for each closeness-to-death category are projected according to a time-series model fitted to the gap between the growth rate of health cost data and productivity growth, with the mean gap constrained in future years to match the CBO projection assumptions, which are also similar to the recently revised HCFA/CMIMS assumptions. CBO and HCFA/CMMS assume that the gap will continue indefinitely, but that it will decline to about 1 percent in a few decades (Lee and Miller, 2001).
A.4 Budget Projections A.4.1 General Strategy Following CBO, we assume that most benefit age schedules rise with productivity growth; social security and health programs are exceptions, as discussed above. Each sample path wifi have different benefit cost profiles, due to stochastic productivity growth. The age profiles are then multiplied by the stochastic population forecasts to generate stochastic program cost estimates. Tax payments can be calculated using tax payment age profiles for various kinds of taxes, estimated cross-sectionally from a recent March CPS, and adjusted to match the NIPA control totals, or inferred from payroll tax rules and earnings. Taxes on capital are assumed to grow with GDP, since rents are assumed to be a constant share of income in the steady state. Trust funds are projected
along sample paths using accounting identities and the stochastic
interest-rate projections. Likewise, debt can be projected using accounting identities and a moving average of the same real interest rate. In such runs, Trust Fund balances and debt levels will be among the outputs. In some other runs, tax rates are set endogenously to comply with stipulated assumptions about limits on debt-to-GDP ratios, or on the Social Security Trust Fund ratio to costs. It is also possible to adjust benefit levels in addition to, or instead of, tax rates, to meet such imposed budget balance constraints.
A.4.2 Age-Neutral Budget Components Some important parts of the budget are not age-targeted. These include public goods, mainly expenditures on defense and research. We follow CBO in projecting these as a
The Fiscal Effects of Population Aging in the U. S
179
constant share of GDP. They also include private goods or quasi-public goods, such as expenditures for roads, sewers, police, fire, and so on. These are also assumed to be a fixed proportion of GDP, and therefore grow stochastically with population and productivity growth. Finally, there is national debt. The balance of flows into and out of national debt is determined endogenously by the budget balance each year.
REFERENCES Auerbach, A. J., and K. A. Hassett (2001). "Uncertainty and Design of Long-Run Fiscal Policy." In Demographic Change and Fiscal Policy, Alan J. Auerbach and Ronald D. Lee (eds.). Cambridge, U.K.: Cambridge University Press. Board of Trustees, Federal Hospital Insurance Trust Fund (2001). 2001 Annual Report of the Board of Trustees of the Federal Hospital Insurance Trust Fund. Wash-
ington: U.S. Government Printing Office. Board of Trustees, Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds (2000). 2000 Annual Report of the Board of Trustees of the Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds. Washing-
ton: U.S. Government Printing Office. Board of Trustees, Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds (2001). 2001 Annual Report of the Board of Trustees of the Federal Old-Age and Survivors Insurance and Disability Insurance Trust Funds. Washing-
ton: U.S. Government Printing Office. Congressional Budget Office (2000). "The Long-Term Budget Outlook" (October), www.cbo.gov. Congressional Budget Office (2001). "The Budget and Economic Outlook: Fiscal Years 2002-2011" (January), www.cbo.gov. Crimmins, E. M., Y. Saito, and D. Ingegneri (1997). "Trends in Disability-Free Life Expectancy in the United States, 1970-90." Population and Development Review 23(no. 3, September):555-572. Cutler, D., M. McClellan, J. Newhouse, and D. Remler (1998). "Are Medical Prices Declining? Evidence from Heart Attack Treatments." Quarterly Journal of Economics 113(no. 4, November):991-1024.
Cutler, D., and L. Sheiner (2000). "Demographic and Medical Care Spending: Standard and Non-standard Effects." In Demography and Fiscal Policy, A. Auerbach and R. Lee (eds.). Cambridge, UK: Cambridge University Press. Freedman, V. A., and L. G. Martin (1999). "The Role of Education in Explaining and Forecasting Trends in Functional Limitations among Older Americans." Demography 36(no. 4, November): 461-473. Fries, J. F. (1980). "Aging, Natural Death, and the Compression of Morbidity." New England Journal of Medicine 303:130-135.
Fries, J. F. (1984). "The Compression of Morbidity: Miscellaneous Comments about a Theme." The Gerontologist 24:354-359. Gruber, J., and D. A. Wise (1999). Social Security and Retirement around the World.
NBER Conference Report. Chicago: University of Chicago Press. Gruber, J., and D. A. Wise (2001). "An International Perspective on Policies for an Aging Society." NBER Working Paper no. 8103. Kannisto, V., J. Lauritsen, A. R. Thatcher, and J. Vaupel (1994). "Reductions in
180
Lee & Edwards
Mortality at Advanced Ages: Several Decades of Evidence from 27 Countries." Population and Development Review 20(no. 4):793-810.
Lee, R. (1993). "Modeling and Forecasting the Time Series of US Fertility: Age Patterns, Range, and Ultimate Level." International Journal of Forecasting 9:187202.
Lee, R. (1999). "Probabilitistic Approaches to Population Forecasting." In Rethinking Population Projections, supplement to volume 24 of Population and Development Review, W. Lutz, J. Vaupel, and D. Ahlburg (eds.). Lee, R. and L. Carter (1992). "Modeling and Forecasting U.S. Mortality." Journal of the American Statistical Association 87(419, September):659-671; "Rejoinder," 674-675. Lee, R., and T. Mifier (2001a). "Evaluating the Performance of Lee-Carter Mortality Forecasts." Demography, v. 38, n. 4 (Nov.) pp. 537-549. Lee, R., and T. Miller (2001b). "An Approach to Forecasting Health Expenditures, with Application to the U.S. Medicare System," under review. Lee, R., and S. Tuljapurkar (1994). "Stochastic Population Forecasts for the U.S.: Beyond High, Medium and Low." Journal of the American Statistical Association 89(no. 428, December):1175-1189. Lee, R., and S. Tuijapurkar (1998a). "Stochastic Forecasts for Social Security." In Frontiers in the Economics of Aging, David Wise (ed.). (Chicago: University of Chicago Press). Lee, R., and S. Tuljapurkar (1998b). "Uncertain Demographic Futures and Social Security Finances." American Economic Review (May):237-241. Lee, R., and S. Tuljapurkar (2000). "Population Forecasting for Fiscal Planning: Issues and Innovations." In Demography and Fiscal Policy, A. Auerbach and R. Lee (eds.). Cambridge, UK: Cambridge University Press. Lubitz, J., and R. Prihoba (1984). "The Use of Medicare Services in the Last Two Years of Life." Health Care Financing Review 5:117-131. Lubitz, J., J. Beebe, and C. Baker (1995). "Longevity and Medicare Expenses." New England Journal of Medicine 332:999-1003.
Manton, K., L. Corder, and E. Stallard (1997). "Chronic Disability Trends in Elderly United States Populations: 1982-1994." Proceedings of the National Academy of Science 94(March):2593-2598. Miller, T. (2001): "Increasing Longevity and Medicare Expenditures." Demography 38:215-226. National Research Council (2000). Beyond Six Billion: Forecasting the World's Population. Panel on Population Projections, John Bongaarts and Rodolfo A. Bulatao
(eds.). Washington: National Academy Press. Olshansky, S. J., B. A. Carnes, and A. Desesquelles (2001). "Demography. Prospects for Human Longevity." Science 291(23 February, no. 5508):1491-1492. Roseveare, D., W. Leibfritz, D. Fore, and E. Wurzel (1996). "Ageing Populations, Pension Systems and Government Budgets: Simulations for 20 OECD Countries." Economics Department Working Paper No. 168. Paris: Organization for Economic Co-operation and Development. Smith, J. P., and B. Edmonston, eds. (1997). The New Americans: Economic, Demographic, and Fiscal Effects of Immigration. Washington: National Academy Press.
Tuljapurkar, S., N. Li, and C. Boe (2000). "A Universal Pattern of Mortality Decline in the G7 Countries." Nature 405(June 15):789-792.
