Tax Policy and the Economy, Volume 20 (NBER Tax Policy and the Economy)

50292Poterba 8/11/06 8:16 AM Page 1 and the Economy 20 NBER Edited by James M. Poterba National Bureau of Economi...

Author: James M. Poterba

30 downloads 1043 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

50292Poterba

8/11/06

8:16 AM

Page 1

and the Economy

20

NBER

Edited by James M. Poterba National Bureau of Economic Research

Tax Policy and the Economy

Tax Policy

20 2006

Tax Policy and the Economy

National Bureau of Economic Research Edited by James M. Poterba Who Bears the Corporate Tax? A Review of What We Know

Of related interest Tax Policy and the Economy, Volume 19 Edited by James M. Poterba

20

Tax Reform and Entrepreneurial Activity

Articles by Ann Dryden Witte with Marisol Trowbridge; Jonathan Gruber; James R. Hines Jr.; Michelle Hanlon and Terry Shevlin; Randall Morck

Behavioral Responses to Taxes: Lessons from the EITC and Labor Supply

Tax Policy and the Economy, Volume 18 Edited by James M. Poterba

Splitting Tax Refunds and Building Savings: An Empirical Test

Articles by Mihir A. Desai and William M. Gentry; Mark E. Doms,Wendy E. Dunn, Stephen D. Oliner and Daniel E. Sichel; Susan Dynarski; Emmanuel Saez;Todd Sinai and Joseph Gyourko

0-262-16240-7 978-0-262-16240-1

The MIT Press Massachusetts Institute of Technology Cambridge, Massachusetts 02142 http://mitpress.mit.edu

Household Ownership of Variable Annuities

NBER The MIT Press

Fiscal and Generational Imbalances: An Update

Tax Policy and the Economy 20

Tax Policy and the Economy 20

edited by James M. Poterba

National Bureau of Economic Research Cambridge, Massachusetts The MIT Press Cambridge, Massachusetts London, England

NBER/Tax Policy and the Economy, Volume 20, 2006 ISSN: 0892-8649 ISBN-13: 978-0-262-16240-1 (hc)—978-0-262-66198-0 (pb) ISBN-10: 0-262-16240-7 (hc), 0-262-66198-5 (pb) Published annually by The MIT Press, Cambridge, Massachusetts 02142 ( 2006 by the National Bureau of Economic Research and the Massachusetts Institute of Technology All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher. Standing orders/subscriptions are available. Inquiries, and changes to subscriptions and addresses should be addressed to Triliteral, Attention: Standing Orders, 100 Maple Ridge Drive, Cumberland, RI 02864, phone 1-800-366-6687 ext. 112 (U.S. and Canada), fax 1-800-406-9145 (U.S. and Canada). In the United Kingdom, continental Europe, and the Middle East and Africa, send single copy and back volume orders 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 info@ hup-MITpress.co.uk, website http://mitpress.mit.edu In the United States and for all other countries, send single copy and back volume orders to: The MIT Press c/o Triliteral, 100 Maple Ridge Drive, Cumberland, RI 02864, phone 1-800405-1619 (U.S. and Canada) or 401-658-4226, fax 1-800-406-9145 (U.S. and Canada) or 401-531-2801, email [email protected], website http://mitpress.mit.edu MIT Press books may be purchased at special quantity discounts for business or sales promotional use. For information, please email [email protected]. This book was set in Palatino on 3B2 by Asco Typesetters, Hong Kong. Printed and bound in the United States of America. 10 9 8

7 6 5

4 3 2 1

NBER Board of Directors

Ofﬁcers Elizabeth E. Bailey, Chairman John S. Clarkeson, Vice Chairman Martin Feldstein, President and Chief Executive Ofﬁcer Susan Colligan, Vice President for Administration and Budget and Corporate Secretary Robert Mednick, Treasurer Kelly Horak, Controller and Assistant Corporate Secretary Gerardine Johnson, Assistant Corporate Secretary

Directors at Large Peter C. Aldrich Elizabeth E. Bailey John H. Biggs Andrew Brimmer John S. Clarkeson Don R. Conlan Kathleen B. Cooper George C. Eads Jessica P. Einhorn Martin Feldstein Jacob A. Frenkel Judith M. Gueron Robert S. Hamada Karen N. Horn Judy C. Lewent

John Lipsky Laurence H. Meyer Michael H. Moskow Alicia H. Munnell Rudolph A. Oswald Robert T. Parry Richard N. Rosett Marina v. N. Whitman Martin B. Zimmerman

Directors by University Appointment George Akerlof, California, Berkeley Jagdish Bhagwati, Columbia Ray C. Fair, Yale Michael J. Brennan, California, 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 Uwe E. Reinhardt, Princeton Nathan Rosenberg, Stanford Craig Swan, Minnesota David B. Yofﬁe, Harvard Arnold Zellner (Director Emeritus), Chicago

vi

Directors by Appointment of Other Organizations Richard B. Berner, National Association For Business Economics Gail D. Fosler, The Conference Board Martin Gruber, American Finance Association Arthur B. Kennickell, American Statistical Association Thea Lee, American Federation of Labor and Congress of Industrial Organizations William W. Lewis, Committee for Economic Development Robert Mednick, American Institute of Certiﬁed Public Accountants Angelo Melino, Canadian Economics Association

NBER Board of Directors

Jeffrey M. Perloff, American Agricultural Economics Association John J. Siegfried, American Economic Association Gavin Wright, Economic History Association

Directors Emeriti Carl F. Christ George Hatsopoulos Lawrence R. Klein Franklin A. Lindsay Paul W. McCracken Peter G. Peterson Eli Shapiro Arnold Zellner

Relation of the Directors to the Work and Publications of the NBER

1. The object of the NBER is to ascertain and present to the economics profession, and to the public more generally, important economic facts and their interpretation in a scientiﬁc manner without policy recommendations. The Board of Directors is charged with the responsibility of ensuring that the work of the NBER is carried on in strict conformity with this object. 2. The President shall establish an internal review process to ensure that book manuscripts proposed for publication DO NOT contain policy recommendations. This shall apply both to the proceedings of conferences and to manuscripts by a single author or by one or more co-authors but shall not apply to authors of comments at NBER conferences who are not NBER afﬁliates. 3. No book manuscript reporting research shall be published by the NBER until the President has sent to each member of the Board a notice that a manuscript is recommended for publication and that in the President’s opinion it is suitable for publication in accordance with the above principles of the NBER. Such notiﬁcation will include a table of contents and an abstract or summary of the manuscript’s content, a list of contributors if applicable, and a response form for use by Directors who desire a copy of the manuscript for review. Each manuscript shall contain a summary drawing attention to the nature and treatment of the problem studied and the main conclusions reached. 4. No volume shall be published until forty-ﬁve days have elapsed from the above notiﬁcation of intention to publish it. During this period a copy shall be sent to any Director requesting it, and if any Director objects to publication on the grounds that the manuscript contains policy recommendations, the objection will be presented to the author(s) or editor(s). In case of dispute, all members of the Board shall be notiﬁed, and the President shall appoint an ad hoc committee of the Board to decide the matter; thirty days additional shall be granted for this purpose. 5. The President shall present annually to the Board a report describing the internal manuscript review process, any objections made by Directors before publication or by anyone after publication, any disputes about such matters, and how they were handled. 6. Publications of the NBER issued for informational purposes concerning the work of the Bureau, or issued to inform the public of the activities at the Bureau, including but not limited to the NBER Digest and Reporter, shall be consistent with the object stated in paragraph 1. They shall contain a speciﬁc disclaimer noting that they have not passed through the review procedures required in this resolution. The Executive Committee of the Board is charged with the review of all such publications from time to time.

viii

Relation of the Directors to the Work and Publications of the NBER

7. NBER working papers and manuscripts distributed on the Bureau’s web site are not deemed to be publications for the purpose of this resolution, but they shall be consistent with the object stated in paragraph 1. Working papers shall contain a speciﬁc disclaimer noting that they have not passed through the review procedures required in this resolution. The NBER’s web site shall contain a similar disclaimer. The President shall establish an internal review process to ensure that the working papers and the web site do not contain policy recommendations, and shall report annually to the Board on this process and any concerns raised in connection with it. 8. Unless otherwise determined by the Board or exempted by the terms of paragraphs 6 and 7, a copy of this resolution shall be printed in each NBER publication as described in paragraph 2 above.

Contents

Introduction

xi

James M. Poterba Acknowledgments

xv

1 Who Bears the Corporate Tax? A Review of What We Know

1

Alan J. Auerbach 2 Tax Reform and Entrepreneurial Activity

41

Julie Berry Cullen and Roger Gordon 3 Behavioral Responses to Taxes: Lessons from the EITC and Labor Supply 73 Nada Eissa and Hilary W. Hoynes 4 Splitting Tax Refunds and Building Savings: An Empirical Test 111 Sondra Beverly, Daniel Schneider, and Peter Tufano 5 Household Ownership of Variable Annuities

163

Jeffrey R. Brown and James M. Poterba 6 Fiscal and Generational Imbalances: An Update Jagadeesh Gokhale and Kent Smetters

193

Introduction James M. Poterba, MIT and NBER

The 2005 Tax Policy and the Economy conference marks the twentieth anniversary of this very popular and productive series. This conference was ﬁrst suggested by Martin Feldstein in the mid-1980s, and it was organized initially by Lawrence Summers. The late David Bradford subsequently organized several meetings. For the last ﬁfteen years, I have had the pleasure of arranging the programs for this meeting. The annual conference communicates current academic research ﬁndings in the areas of taxation and government spending to policy analysts in government and the private sector. The papers presented at this conference address issues with an immediate bearing on current policy debates as well as questions that are of longer-term interest. This conference has served as a model for researchers in other ﬁelds who are interested in bringing their applied research ﬁndings to the attention of policy analysts. A number of the papers that were written for this conference series have introduced important analytical tools or suggested durable empirical or conceptual insights about the economic effects of tax and expenditure programs. Most of the papers in this year’s volume focus on the economic effects of taxation, and the last paper presents important insights on the balance between expenditures and revenues. The ﬁrst paper is Alan J. Auerbach’s ‘‘Who Bears the Corporate Tax? A Review of What We Know.’’ Public ﬁnance economists have debated the incidence of the corporate income tax for many decades. Arnold Harberger’s celebrated analysis of this tax remains one of the best-known papers in public economics. Despite many years of research, however, there is still no consensus on who bears the burden of this tax. Different models suggest different results, and conclusive empirical tests of the alternative models have proven elusive. This comprehensive survey paper uses several different models of investment and ﬁnancing

xii

Introduction

behavior to evaluate the distribution of the burden of the corporate income tax. The paper catalogs what is known, and what remains unknown, about this tax. The paper is likely to become the starting point for almost all further discussions, in both policy analysis and in research, of the corporate income tax. The next paper is Julie Berry Cullen and Roger Gordon’s study of ‘‘Tax Reform and Entrepreneurial Activity.’’ There is broad agreement that entrepreneurial activity is an important determinant of long-run economic growth, but much less agreement exists on how such activity is affected by government policy. This paper systematically examines a number of provisions of the personal income tax, the corporate income tax, and the payroll tax that may affect an individual’s choice between working as a salaried employee or as a self-employed entrepreneur. The authors emphasize that the interaction between various tax provisions can be important for assessing the incentives for entrepreneurial activity. They demonstrate that the tax system’s net incentive to become an entrepreneur varied substantially between 1980 and 2000. The 1986 Tax Reform Act and the 1994 reduction in the capital gains tax rate on gains on stock in small businesses had particularly large effects on these tax incentives. The authors also present empirical results on the potential impact of several tax reforms on the level of entrepreneurial activity. Their results suggest that a shift to a singlerate income tax structure would encourage entrepreneurial activity at low-income levels while discouraging it at high-income levels. They also ﬁnd that changes in the tax rates at which losses can be deducted are at least as important as capital gains tax rates in determining incentives for entrepreneurship. The third paper is Nada Eissa and Hilary W. Hoynes’s study of ‘‘Behavioral Responses to Taxes: Lessons from the EITC and Labor Supply.’’ The EITC is the largest cash anti-poverty program in the United States. It affects nearly 20 million families and costs roughly $40 billion per year. Because the EITC is administered through the tax system and because it can substantially change the marginal tax rate that applies to a household’s labor income, a number of studies have tried to evaluate the effect of the EITC on labor supply. This paper provides new evidence on this issue and reviews the central ﬁndings from earlier studies. The paper points to two robust ﬁndings that emerge from the existing research and corroborates these ﬁndings with new data analysis. First, the labor supply subsidies that are provided to families with

Introduction

xiii

the lowest earning levels appear to encourage labor supply. This is consistent with the traditional theoretical analysis of how a wage subsidy should affect labor market activity. The second ﬁnding, however, is not as easy to reconcile with theoretical modeling of labor supply. The increases in marginal income tax rates on some families that fall into the phase-out region of the EITC do not appear to discourage labor supply. This is puzzling because the high marginal rates reduce the marginal return to an additional hour of work. The authors discuss several potential explanations for this puzzling but robust empirical ﬁnding. The next paper, by Sondra Beverly, Daniel Schneider, and Peter Tufano, is ‘‘Splitting Tax Refunds and Building Savings: An Empirical Test.’’ This paper examines the potential impact of changes in the administrative treatment of tax refunds on the saving behavior of lowincome households. A large fraction of the traditional policy debate surrounding tax incentives and saving focuses on the behavior of middle- and high-income households who might vary the share of their income that they save in response to tax incentives. This paper suggests that the rules surrounding the payment of income tax refunds may affect wealth accumulation by low-income households. Current tax administration rules require that a taxpayer’s refund be directed to a single recipient. That could be the taxpayer, who would receive the refund in the form of a check, or it could be a single ﬁnancial institution. This paper reports ﬁndings from a pilot experiment in Tulsa, Oklahoma, in which low-income households were allowed to split their tax refunds, asking in particular for part of the refund to be directed to a savings account at a ﬁnancial institution and part to be returned to them. The results suggest that the ﬂexibility provided by refund-splitting leads a substantial group of taxpayers to contribute some of their refund to a savings account. The empirical ﬁndings in this study suggest an intriguing and feasible way to try to increase saving and wealth holdings among low-income households. The ﬁfth paper, which Jeffrey R. Brown and I co-wrote, is ‘‘Household Ownership of Variable Annuities.’’ Variable annuities were one of the fastest-growing ﬁnancial products of the 1990s. They provide buyers with a wide range of investment options and with an opportunity to defer taxes on investment income through the ‘‘inside buildup’’ that is associated with investment products. This paper summarizes the tax treatment of variable annuities and explains how the after-tax

xiv

Introduction

returns on these products, relative to the returns on other ﬁnancial instruments such as mutual funds, are affected by marginal tax rates on dividends, capital gains, and ordinary income. The paper then examines data from the 2001 Survey of Consumer Finances to evaluate the concentration of variable annuity ownership among households in the highest income tax brackets. The paper ﬁnds that the probability of owning a variable annuity is higher for those in higher marginal tax brackets than for those in lower brackets, but the relationship is not monotonic. The ownership probability for those in the highest marginal tax bracket is lower than that for taxpayers in slightly lower tax brackets. Ownership of variable annuities is somewhat more diffuse than the ownership of a number of other ﬁnancial products. Households in the top 10 percent of the wealth distribution in 2001 owned 73 percent of all variable annuities, compared with 90 percent of corporate stock, 79 percent of mutual funds, and 90 percent of tax-exempt bonds. Finally, the last paper is Jagadeesh Gokhale and Kent Smetters’s analysis of ‘‘Fiscal and Generational Imbalances: An Update.’’ This paper presents new evidence on the long-run ﬁscal position of the federal government. The authors base their calculations on estimates of the rate of economic growth, population growth, the evolution of the population age structure, and the rate of growth of health care outlays relative to other components of gross domestic product. They evaluate the present discounted value of both tax revenues and federal expenditure commitments under current law. Their ﬁndings suggest a substantial imbalance between spending and revenues. Their best estimate of current ﬁscal stance suggests that the present discounted value of projected outlays exceeds the corresponding present discounted value of revenues by $63 trillion. This is substantially greater than the authors’ previous estimate of $44 trillion, which was based on policies in place and economic projections in 2003. The enactment of the Medicare prescription drug beneﬁt is the largest factor contributing to the worsening of the projected ﬁscal position. Each of these papers illustrates the type of policy-relevant research that is carried out by the afﬁliates of the NBER Public Economics Program. These studies provide important background information for policy analysis, without making recommendations about the merits or demerits of particular policy options. We hope they will provide a valuable basis for policy discussions both in Washington and in the broader policy community.

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. NBER President Martin Feldstein has been an active supporter of this conference throughout its history. Conference Department Director Carl Beck, Lita Kimble, and especially Rob Shannon have continually helped to keep our invitation list up to date and handled conference logistics with efﬁciency and good cheer. Helena Fitzpatrick has overseen the publication process with outstanding attention to detail and with exceptional speed and efﬁciency. I am grateful to Dr. Ben Bernanke, the Chair of the Council of Economic Advisers, for delivering a fascinating set of luncheon remarks at the conference at which these papers were presented. Ben’s remarks focused on the role of federal economic policy in contributing to economic recovery in the wake of the disaster caused by Hurricane Katrina. His remarks focused on broad macroeconomic themes, such as the importance of tax policy in contributing to long-run economic growth, as well as on particular policy design issues associated with disaster relief. Finally, I wish to thank the authors of this year’s conference papers. They have worked hard to communicate their important research ﬁndings in a readable and clear fashion. I appreciate their efforts and their enthusiasm for participating in this interchange between the research and policy communities.

1 Who Bears the Corporate Tax? A Review of What We Know Alan J. Auerbach, University of California, Berkeley and NBER

Executive Summary Who bears the corporate income tax? The answer to this question is important to our understanding of the distribution of tax burdens, but it has been elusive. Although the tax accounts for a small share of federal revenues, changes in the corporate income tax and its associated revenues have often been a signiﬁcant part of revenue legislation. Moreover, because its incidence is often perceived to fall on the afﬂuent, assignment of the corporate tax burden can have a signiﬁcant impact on the assessed progressivity of the tax system as a whole. This paper reviews what we know from economic theory and evidence about the burden of the corporate income tax. While the ultimate incidence of the tax remains somewhat unresolved, there have been many advances over the years in our thinking about how to assign the corporate tax burden. Among the lessons from the recent literature are the following: 1. For a variety of reasons, shareholders may bear a certain portion of the corporate tax burden. In the short run, they may be unable to shift taxes on corporate capital. Even in the long run, they may be unable to shift taxes attributable to a discount on ‘‘old’’ capital, taxes on rents, or taxes that simply reduce the advantages of corporate ownership. Thus, the distribution of share ownership remains empirically quite relevant to corporate tax incidence analysis, though attributing ownership is itself a challenging exercise. 2. One-dimensional incidence analysis—distributing the corporate tax burden over a representative cross-section of the population—can be relatively uninformative about who bears the corporate tax burden because it misses the element of timing.

2

Auerbach

3. It is more meaningful to analyze the incidence of corporate tax changes than that of the corporate tax in its entirety because different components of the tax have different incidence, and incidence relates to the path of the economy over time, not just in a single year. 1.

Introduction

Who bears the corporate income tax? The answer to this question is important to our understanding of the distribution of tax burdens, but it has been elusive. In his classic series of analyses of the incidence of the U.S. tax system, Pechman (e.g., 1985) provided alternative scenarios with different assumptions about the incidence of the corporate tax, reﬂecting his uncertainty about which assumption was best. (He did not do this for the individual income tax.) Distributional analyses provided by U.S. government agencies have, on most occasions, simply ignored the corporate tax. Thus, the Tax Reform Act of 1986, which was estimated to reduce individual income taxes and increase corporate income taxes, could illogically be characterized as being revenueneutral while providing a tax cut for each income class of a nine-class breakdown.1 This episode illustrates why it is important to understand the incidence of the corporate tax. Although the tax accounts for a small share of federal revenues, changes in the corporate income tax and its associated revenues have often been a signiﬁcant part of revenue legislation. Moreover, because its incidence is often perceived to fall on the afﬂuent, assignment of the corporate tax burden can have a signiﬁcant impact on the assessed progressivity of the tax system as a whole. The most evident difﬁculty in assigning the corporate tax burden is that, unlike most taxes, there is no guidance given by statutory incidence. While we may start with a working assumption that individual income taxes or sales taxes are borne by the people who are legally liable for them, for example, there is no comparable assumption for the corporation income tax, given the cardinal rule of incidence analysis that only individuals can bear the burden of taxation and that all tax burdens should be traced back to individuals. Thus, we must rely on deeper assumptions from the start, and with corporations increasingly becoming multinational enterprises, the individuals at risk of bearing the U.S. corporate tax burden clearly include those beyond our own national borders.

Who Bears the Corporate Tax?

3

This paper reviews what we know from economic theory and evidence about the burden of the corporate income tax. While the ultimate incidence of the tax remains somewhat unresolved, there have been many advances over the years in our thinking about how to assign the corporate tax burden; we don’t have all the answers, but we do have a much better idea where to look for them. I begin with some basic facts about the corporate income tax, and then move to the evolution of thought about its burden, starting from a fairly simple approach based on the ownership of corporate shares. After considering Harberger’s (1962) landmark contribution, I then discuss a variety of important issues absent from that analysis, including dynamics, investment incentives, corporate ﬁnancial policy, risk, imperfect competition, the choice of organizational form, international capital ﬂows, and managerial incentives. My focus is on the federal corporate income tax. States, too, impose corporate income taxes, but the incidence of these taxes is also inﬂuenced by additional factors, notably the degree of capital mobility across state boundaries and the formulas states use to apportion income according to the location of sales, assets, and employment, so that tracing the incidence would lead me too far astray from my main task.2 2.

Corporate Taxation in the United States

The U.S. corporate income tax, in ﬁscal year 2004, accounted for 10 percent of federal revenues, or 1.6 percent of gross domestic product (GDP). Figure 1.1 provides these two measures annually since 1962. From the ﬁgure, it is obvious that corporate revenues have declined over time as a share of revenues and of GDP. The ratios move closely together, consistent with the fact that revenues have been relatively stable as a share of GDP. The corporate tax today is far less important than in the 1960s, when it regularly accounted for more than 20 percent of revenue. Very recently, there has been concern that corporations have used increasingly aggressive strategies to limit tax liabilities. While these concerns may be valid, they are not responsible for the sharp decline in the importance of the corporate tax shown in ﬁgure 1.1. Discounting year-to-year movements and cyclical ﬂuctuations in this volatile stream of revenue induced by the volatility of corporate proﬁts themselves, there is little trend over the past two decades. Looking back over the decline that occurred between the 1960s and the 1980s, Auerbach and Poterba (1987) assigned a signiﬁcant share to

4

Auerbach

Figure 1.1 U.S. Federal Corporate Income Taxes, 1962–2004. Source: Congressional Budget Ofﬁce.

changes in tax policy but found other factors, such as changes in corporate ﬁnancial policy, to be important as well. Corporations vary enormously in size. While most corporations are relatively small, the preponderance of corporate income tax revenue comes from large corporations. In 2001, for example, 0.04 percent of all corporations, those with assets above $2.5 billion, accounted for 62 percent of all corporate income taxes (Treubert 2004, Table 1). Simple economic theories tend to distinguish between corporate and non-corporate enterprises, but there are many entity types with hybrid characteristics. Perhaps most relevant to the current discussion are S corporations, which share many of the legal attributes of traditional C corporations3 (perhaps the most important being limited liability) but have their income taxed directly to individual owners, as is the case for non-corporate ownership structures. When thinking of the incidence of the corporate tax, we treat S corporations as part of the non-corporate sector, although the ability of an entity to choose between C- and S-corporation forms has clear implications for the incidence of additional taxes on C corporations. The importance of S corporation status has grown steadily over the years. As of 1986, about one-fourth of all U.S. corporations were S corporations; by 1997, this

Who Bears the Corporate Tax?

5

share had risen to more than half (Luttrell 2005, Figure A).4 In 2001, S corporations accounted for almost a quarter of before-tax corporate proﬁts (Treubert 2004, Figure B). For various types of business, including sole proprietorships and partnerships as well as S corporations, income from the business is assigned to the business’s individual owners and then aggregated with the other incomes of these owners and subject to the individual income tax. The incomes of C corporations, by contrast, are subject to a distinct tax on corporate income that treats the corporation as an entity subject to taxation. Shareholder income from C corporations in the form of dividends and capital gains is then subject to additional taxation under the individual income tax. 3.

An Initial Approach to Corporate Tax Incidence

Perhaps the simplest and oldest theory of corporate tax incidence is that the tax falls on corporate shareholders in proportion to their ownership. This theory may be implicit in the minds of those who view the corporate tax as very progressive, for individual share ownership is highly concentrated among higher income individuals. In 2001, for example, 90 percent of families in the top income decile held stock (either directly, or indirectly through mutual funds or retirement accounts), with a median value among those holding stock of $248,000. For the population as a whole, 52 percent held stock, with a median holding of just $34,000 (Aizcorbe et al. 2003, Table 6), and with the fraction holding stock rising steadily with income. But even this simple method of assigning the burden of the corporate tax is not so simply applied. First, a corporation may have both preferred and common shares, and more than one class of common shares; each category of shares may confer different rights to the corporation’s income. If an increase in the rate of corporate taxation reduces a corporation’s after-tax income, it is not always clear how much of this reduction will be borne by different categories of shareholders. Indeed, this ambiguity is one of the reasons why S-corporation status is available only for corporations with one class of shares—to assign income to shareholders we must have a clearly deﬁned way of doing so. Second, even where the assignment of income is clear, not all shareholders are individuals. Table 1.1 provides a breakdown of the ownership of U.S. corporate equity at the end of 2004. Households owned less than half of all equity directly, with substantial fractions held by

6

Auerbach

Table 1.1 U.S. Corporate Equity Ownership, 2004 (End of Year, Billions of Dollars) Asset Holder Households Mutual funds Nonproﬁt organizationsa Retirement funds

Amounts 5,979 3,694 597 2,993

Private pension funds (DB)b

720

Private pension funds (DC)b

971

State and local government retirement funds

1,202

Federal government retirement funds Bank personal trusts and estates Life insurance companies Savings institutions State and local governments Rest of the worldc Market value of domestic corporations

Detail

99 221 1,065 28 89 467 14,198

Source: Board of Governors of the Federal Reserve System (2005), Table L.213. Amounts net out inter-corporate holdings. a Holdings of households and nonproﬁt organizations are grouped together for 2004; breakdown is based on assumption that proportion held by nonproﬁts is the same as in 2000, using Table L.100.a. b Detail from Tables L.119.b and L.119.c. c Rest of world equals holdings of U.S. issues by foreign residents less holdings of foreign issues by U.S. residents.

various institutions and ﬁnancial intermediaries. Ownership through mutual funds, the second-largest ownership category, is not a major issue because tax provisions allow the pass-through of income directly to individual owners of mutual fund shares. But the other major class of institutional owners, nonproﬁt institutions and retirement funds, poses a more difﬁcult problem.5 Among pension funds, we may distinguish one category, deﬁnedcontribution plans and other tax-sheltered vehicles, for which the accounts are owned by beneﬁciaries. For these accounts, it is natural to treat the individual beneﬁciaries as the ultimate owners of shares held in the funds. But for the remaining assets, held in the funds of deﬁnedbeneﬁt plans, the assignment of ownership is less obvious. For privatesector deﬁned-beneﬁt plans, the ﬁrst thought might be to assign the assets in these funds to the corporations that maintain them (and hence ultimately to the shareholders of those corporations) because the corporations are using the pension funds to meet pension liabilities.

Who Bears the Corporate Tax?

7

According to this line of reasoning, any ﬂuctuations in the fund balances attributable to changes in corporate income taxation will require offsetting contributions by the corporations; hence, the shareholders of these corporations bear the burden of these changes. But this reasoning breaks down if pension liabilities are responsive to the health of the pension fund, either because of inﬂuences on the relative bargaining power of employers and employees or because of the ability of employers to ‘‘put’’ pension liabilities to the Public Beneﬁt Guaranty Corporation (PBGC) at a cost lower than the actuarial pension liability.6 In this case, a portion of the pension fund really ‘‘belongs’’ to employees,7 but the breakdown between employers and employees is an empirical issue on which there is little evidence. A similar ambiguity arises with respect to public-sector deﬁned-beneﬁt plans, with taxpayers assuming the role taken by shareholders in the case of private-sector plans. For the remaining tax-exempt entities—nonproﬁt institutions such as universities and foundations—there are no owners to which incidence can be assigned. Presumably, the incidence of corporate taxes that reduce the income of such entities is borne in some measure by beneﬁciaries (through reduced services); donors (through increased contributions); employees (through reduced compensation); and perhaps others with more indirect connections, such as vendors. As with the division of ‘‘ownership’’ for deﬁned-beneﬁt pension plans, how the burden of reduced non-proﬁt funds would be borne is an issue on which there is little evidence. Another category of shareholders not represented in table 1.1 (because their holdings are netted out) is corporations themselves. The assignment issue is not a problem; if corporation A owns shares in corporation B, then the portion of corporation B’s corporate tax we allocate to corporation A can be attributed back to corporation A’s shareholders. But the tax burden will be different because of the additional level of corporate ownership. Corporations receive a deduction from taxable income of only 70 percent of dividends received, meaning that such dividends face an effective tax rate of 10.5 percent (30 percent of the current corporate tax rate of 35 percent); there is no deduction for inter-corporate capital gains. Thus, the corporate tax burden on shareholders’ income increases as that income passes through additional corporations. It would be an interesting exercise to confront each of the assignment problems just discussed and trace all corporate income taxes back

8

Auerbach

to individual taxpayers, to determine the incidence pattern implied by the ‘‘simple’’ approach of assigning corporate taxes to shareholders. That this has not been done probably reﬂects both the difﬁculty of the exercise and the fact that the shareholder-incidence method has been perceived to have little theoretical credibility. But as discussed below, incidence has a dynamic dimension that is often ignored. Even if shareholders eventually shift some, most, or even the entire corporate tax burden to others, this shifting need not occur immediately. To the extent that adjustment takes time, some of the corporate tax may indeed be borne by shareholders, and so the exercise just outlined would remain useful. Indeed, other considerations discussed below indicate that shareholders may be unable to shift the tax even in the long run. Once we move beyond the assignment of the corporate tax burden based on information about direct or indirect corporate ownership, an economic model is needed. Only with such a model can we estimate how the corporate tax affects the real incomes of different groups in the population through its impact on factor returns and product prices. In one of the most inﬂuential papers ever written in the ﬁeld of public ﬁnance, Harberger (1962) followed this strategy, analyzing the incidence of the corporation income tax using a two-sector general equilibrium model. Harberger’s contribution has had a lasting impact on incidence analysis and provides a useful benchmark against which to compare subsequent developments in the literature. 4.

The Harberger Model

Grouping all production in the U.S. economy into two sectors according to whether production was predominantly carried out by corporate or non-corporate businesses, Harberger characterized the corporate tax as an additional tax levied on capital income originating in the corporate sector, layered on top of the individual income tax collected on capital income from both sectors. He then estimated incidence through the changes in factor prices and product prices that would result from a small increase in the corporate tax. Harberger’s main conclusion is probably the most familiar aspect of the paper. In particular, under reasonable assumptions regarding the two sectors’ production elasticities of substitution and consumers’ elasticity of substitution between the two sectors’ products, Harberger showed that the corporate income tax was borne fully by owners of capital, economy-wide. This ﬁnding has two important elements. First,

Who Bears the Corporate Tax?

9

capital bears the entire tax; it is not shifted to labor or consumers, the other potential victims in the model. Second, it is all capital, not just corporate capital, that bears the tax. Thus, if corporate capital accounts for 25 percent of the economy’s capital, its individual owners will bear 25 percent of corporate taxes; the other 75 percent will be shifted to owners of non-corporate capital. Intuitively, the lower after-tax return that would be available in the corporate sector because of the higher tax burden drives capital into the non-corporate sector, pushing down the available non-corporate return and allowing the corporate return to recover. In equilibrium, the after-tax returns in the two sectors must be equal, and Harberger estimates that this new equilibrium level of aftertax returns will be lower by just the amount consistent with capital bearing the entire corporate tax. Harberger’s conclusion, which probably remains the most commonly held view on corporate tax incidence, indicated that the corporate tax was less progressive than under the shareholder-incidence assumption because shareholders as a group (at least in 1962, when pension funds accounted for a much smaller ownership share) were more afﬂuent than owners of capital as a whole, a large share of which is owner-occupied housing. But aggregate capital ownership is more concentrated among higher-income individuals than consumption or labor income, and so the corporate income tax could still be seen as contributing to tax progressivity. Another message of Harberger’s work, though, was that the corporate income tax distorted the allocation of capital between corporate and non-corporate uses in a way that an overall capital income tax did not. If the incidence of the two taxes were the same, then the only ‘‘contribution’’ of the corporate tax was gratuitous deadweight loss. Indeed, the subsequent optimal taxation literature supported the notion that taxes that distort production decisions are to be eschewed when sufﬁcient other tax instruments are available (Diamond and Mirrlees 1971). Thus, Harberger’s analysis has also lent support to the view that corporate tax is not a necessary or desirable component of an efﬁcient, progressive tax system. Harberger’s analysis spawned a vast literature over several years that extended and challenged his initial results. The simplicity of Harberger’s technique—comparative static analysis of small changes in a two-sector model—proved not to be a major source of concern given that similar ﬁndings resulted from analysis using a multisector computable general equilibrium model (Shoven 1976). But Harberger’s analysis also relied on several more important simplifying

10

Auerbach

assumptions. Two assumptions already mentioned are that (1) the corporate tax can be viewed as an add-on tax on capital income originating in the corporate sector, and (2) production in a particular sector must be exclusively either corporate or non-corporate. Other key assumptions include: (1) free mobility of factors across sectors; (2) ﬁxed economy-wide factor supplies; (3) competitive markets and constant returns to scale, implying that all corporate proﬁts represent normal returns to capital; (4) a closed economy; (5) no risk; and (6) no differences in spending patterns among individuals and between individuals and government. All of these assumptions have been examined in the literature. 5.

Dynamics

Even if the Harberger model paints an accurate picture of the longrun effects of the corporate tax, few would argue that these effects are observed immediately. Labor, and especially capital, cannot freely shift from one sector of production to another. While computers can be moved from one ofﬁce to another, it is considerably more difﬁcult to turn a nuclear power plant into a tractor. Thus, it is probably more reasonable to think of the shifts predicted by the Harberger model as occurring over time, with some capital moving right away and other capital shifting more gradually, for example, as capital in the corporate sector wears out and is replaced by different types of capital in the non-corporate sector. It is tempting to view this as simply a transition phenomenon, i.e., that the incidence is temporarily at variance with Harberger’s predictions but consistent with them in the long run. But the period of transition may be long, and its inﬂuence on incidence is immediate and quite important. Figure 1.2 illustrates the impact of gradual adjustment of capital to an increase in the corporate tax, under the assumptions that the economy is in long-run equilibrium at date 0 and there is an unexpected introduction of a corporate tax at date 1. Initially, the economy-wide rate of return, both before-tax and after-tax, is r0 . At date 1, the tax is imposed after capital allocation has been ﬁxed, so there can be no change in the before-tax returns in either sector—as capital is in ﬁxed supply in the corporate sector, it must absorb the entire tax through a lower after-tax return. Over time, as capital shifts from the corporate sector to the noncorporate sector, the before-tax return in the corporate sector rises and the after-tax return in the non-corporate sector falls, with the after-tax

Who Bears the Corporate Tax?

11

Figure 1.2 Adjustment to Equilibrium

returns in the corporate and non-corporate sectors, rnet and rnc , respectively, gradually coming together at a new equilibrium value of r2 . According to Harberger’s analysis, the economy-wide decline in the net return to capital from r0 to r2 , multiplied by the capital stock, will roughly equal the tax revenue collected in the long run. What is the incidence of the corporate tax in this instance? In terms of returns to capital, the impact is felt initially by corporate capital and then spreads to all capital. But in terms of capital owners, the answer is quite different with respect to timing. The distinction is due to capitalization—the reﬂection in asset values of anticipated differences in returns to capital. While capital and the returns to capital adjust slowly, asset values and asset returns adjust instantaneously. Because investors will demand the same after-tax rate of return on corporate and non-corporate assets, corporate assets must drop in value, relative to non-corporate assets, by an amount roughly equal in present value to the gap between the returns rnc and rnet . Thereafter, investors in corporate and non-corporate assets will receive equal rates of return at every point in time. What will this rate of return be? The answer depends on the technology of adjustment. Under the q-theory of investment envisioned by Tobin (1969) and developed by Hayashi (1982), Summers (1981), and others, the surge in demand for non-corporate capital will temporarily

12

Auerbach

increase the full cost of installed capital in that sector, driving up noncorporate asset prices relative to replacement cost. This, in turn, will reduce the asset-based return on non-corporate capital—and hence corporate capital as well—below rnc during the adjustment process. We have, then, a pattern of incidence that must be characterized not only in terms of rates of return to capital but also asset values. The corporate tax introduced at date 1 will be borne partially by current owners of corporate capital, through an initial drop in asset values, and partially by future investors in corporate and non-corporate capital, through lower rates of return. The total burden borne by these three groups as a whole will exceed the total burden of the tax because initial owners of non-corporate capital will gain from an increase in asset values.8 The allocation of the burden among these groups will depend on the adjustment technology. If adjustment is instantaneous, Harberger’s analysis applies, and the burden will fall entirely on future capital owners. If adjustment occurs at a glacial pace, then virtually the entire burden will be borne by existing corporate shareholders. The distinction between changes in asset values and changes in asset returns is important, even if all investors hold the same portfolio of corporate and non-corporate assets, because the timing of incidence differs. Whereas lower asset returns occur over time, changes in asset values occur immediately. This distinction can be best understood in a generational context. For older asset holders who have accumulated capital and have short planning horizons, the change in asset values will be most relevant. For younger individuals who have accumulated little wealth but have longer planning horizons, the change in the rate of return will matter more. Thus, we can think of the different components of the corporate income tax burden in generational terms: a reduction in asset values that primarily hits the old, and a reduction in rates of return that primarily hits the young. The pattern of incidence would be different for an anticipated increase in the corporate tax rate as adjustment would begin as soon as the future tax increase became known, leading to a smaller initial decline in corporate asset values and more of the burden being shifted to new investors. Thus, prior announcement of a corporate tax increase could be used to cushion the burden on existing asset owners, but if one moves beyond a world of ﬁxed factor supplies, it would affect economic efﬁciency as well: tax-induced reductions in rates of return distort saving and investment decisions, while unexpected drops in asset values do not.

Who Bears the Corporate Tax?

13

A ﬁnal caveat should be issued here regarding the distinction between share ownership and exposure to ﬂuctuations in share prices. With the growing use of stock options and other ﬁnancial derivatives, it has become easier and cheaper over time to hedge all or some risks associated with stock price ﬂuctuations. In theory, one could use derivatives to hedge the risks of tax changes, shifting the burden onto counterparties to the derivative transactions. While this is unlikely an important issue at present, the pace of ﬁnancial innovation suggests that it may become one in the future. Two important conclusions so far are that (1) it is misleading to allocate the burden of a corporate tax increase to all capital, even if that result holds in the long run after capital has completely adjusted, and (2) it is difﬁcult to convey the incidence story in a one-dimensional breakdown of households, say, by wealth, income, or asset ownership; the generational incidence pattern is extremely important as well. 6.

Investment Provisions

The corporate income tax is not simply a uniform tax on economic income originating in the corporate sector. The deviations in the tax base from economic income, in turn, affect the incidence of the corporate tax itself. One deviation relates to investment provisions. A second, discussed in the following section, involves the deductibility of corporate interest payments. As modeled by Harberger, the base of the corporate income tax equals income from all corporate capital. In particular, income from capital goods of different vintages is taxed at the same rate. In reality, capital goods of different ages receive different treatment, even though they are subject to the same statutory corporate tax rate, because of differences in depreciation provisions. This is true not only if the law has changed over time (in which case different vintages would be written off according to different schedules), but even if the law has remained constant, for depreciation allowances do not track the actual economic depreciation of assets. Given that an asset’s income equals its gross returns less depreciation, depreciation allowances that fall short of economic depreciation lead to a tax base greater than income, and allowances in excess of economic depreciation lead to a narrower tax base. Having depreciation allowances smaller or larger than economic depreciation simply leads to effective tax rates higher or lower than the

14

Auerbach

Figure 1.3 Depreciation Schedules

statutory corporate tax rate. But allowances that follow a different pattern over time than economic depreciation can also induce differences in the relative treatment of new and existing assets. This distinction is illustrated in ﬁgure 1.3, which depicts various potential depreciation schedules for an asset that decays at 10 percent per year. Economic depreciation for such an asset would follow the declining-balance method, starting with a 10 percent deduction in the ﬁrst year of ownership and following the pattern labeled ‘‘Economic.’’ A proportionate reduction in each year’s allowances would result in the pattern labeled ‘‘Reduced.’’ Following economic depreciation, but based on historic cost rather than current cost, would lead instead to the pattern labeled ‘‘Historic cost,’’ starting at the same point as the original schedule but falling faster as prices rise. An historic-cost schedule allowing faster write-off, perhaps to compensate for the erosion of allowances due to inﬂation,9 might look like the pattern labeled ‘‘Historic cost, accelerated.’’ Such a depreciation schedule is accelerated relative to economic depreciation both by historic cost accounting and explicit acceleration. With accelerated depreciation schedules, new assets are more attractive than old ones of the same productivity because they convey future depreciation allowances that are higher in present value. Prior to 1986, an additional distinction was provided by the investment tax credit, which was received upon an asset’s purchase but not available to capi-

Who Bears the Corporate Tax?

15

tal already owned. The overall impact of such investment provisions on the value of capital can be assessed using the expression: Vold ¼ Vnew ð1 k tznew þ tzold Þ

ð1:1Þ

where Vnew is the value of a new unit of capital, Vold is the value of an existing unit of equally productive capital, k is the investment tax credit, t is the corporate tax rate, Vnew znew is the present value of depreciation allowances for the unit of new capital,10 and Vnew zold is the present value of depreciation allowances for the unit of existing capital. For economic depreciation, k ¼ 0 and znew ¼ zold , so Vold ¼ Vnew . But typically Vold < Vnew . Calculations in my previous work (Auerbach 1983) found that the ratio Vold =Vnew fell to around 0.8 for corporate ﬁxed capital after the Economic Recovery Act of 1981 due to the combination of high inﬂation, accelerated depreciation, and the investment tax credit. The Tax Reform Act of 1986 reduced this discount substantially by lowering the corporate tax rate (which reduces the importance of differences in depreciation allowances), and eliminating the investment tax credit, with the drop in inﬂation over the same period working in the same direction. In later work (Auerbach 1996), I estimated a comparable value for the mid-1990s of greater than 0.9. What impact does this old-capital discount have on the incidence of the corporate tax? The discount: ðVnew Vold Þ=Vnew ¼ k þ tðznew zold Þ

ð1:2Þ

reﬂects the fact that old capital’s tax base is broader than new capital’s. An increase in the corporate tax rate, therefore, increases the discount by ðznew zold Þ per each unit tax increase. This differential increase represents a levy on existing capital—a portion of future corporate taxes that are immediately capitalized into the value of existing assets. The incidence of this capitalized portion should be on existing shareholders, with only the remaining future corporate taxes relevant for the incidence analysis already carried out.11 But recall that this previous analysis also called for a division of the corporate income tax into components, with some future corporate tax revenues capitalized into the value of existing corporate assets, and the remaining revenues spread among future capital owners as envisioned by the Harberger model. Thus, we now have layers of decomposition. Because capital is slow to adjust, a portion of any corporate income tax increase will be borne by existing shareholders. Of the remaining portion, an additional piece will also be borne by shareholders, in the

16

Auerbach

form of a ‘‘surcharge’’ on existing assets that does not affect the incentives to accumulate capital within the corporate sector. Using expression (1.2), one can also estimate the incidence of other changes in the tax structure. For example, an increase in corporate tax collections accomplished through a reduction in the generosity of depreciation allowances (a reduction in znew ) or a reduction in the investment tax credit, k, will reduce the old-capital discount. Thus, the effective increase in the corporate tax on new capital will be higher than is reﬂected in future corporate tax collections, for a portion of these collections will go to provide a windfall to existing capital. There have been few attempts in the literature to consider the combined capitalization effects of corporate tax changes attributable to gradual capital-stock adjustment and the distinction between old and new capital. One example is my earlier estimate (Auerbach 1989) of the impact of the Tax Reform Act of 1986 (TRA86) on the value of corporate equipment and structures. That analysis found that TRA86 increased corporate taxes at the margin of new investment, leading to a small decline in the value of existing assets following the logic of the previous section, but also provided substantial windfalls to existing capital through the corporate tax rate reduction and investment tax credit repeal. The net impact was a substantial increase in existing asset values, estimated at from 9 to 14 percent for equipment and from 4 to 14 percent for structures, with the results varying with assumptions about the speed of capital stock adjustment and expectations regarding the tax reform. Thus, changes that were estimated to have little net impact on corporate tax revenues12 nevertheless could have signiﬁcant incidence effects, the result of a combined increased burden on new capital and reduced burden on existing capital. These were predictions based on theory, of course, although contemporaneous empirical evidence provided some support (Cutler 1988).13 7.

Corporate Financial Policy and Shareholder Taxes

As discussed above, one of Harberger’s assumptions was that the corporate tax rate was imposed as an increment, over and above the individual tax rate that applied in both corporate and non-corporate sectors. This would be a reasonable characterization of the situation in the United States prior to 2003 if all corporate-source income were paid out as dividends, for until 2003, dividends were taxed as ordinary income after the corporate tax had been applied. But only a share of cor-

Who Bears the Corporate Tax?

17

porate earnings are distributed as dividends, and only a share of the returns to corporate capital accrues as corporate earnings—a large portion passes out of the corporation as interest payments on corporate debt. With corporations having the option to issue debt (the interest payments on which are deductible at the corporate level) and to retain earnings (thereby trading off current dividends for capital gains on which taxes may be lower and can be deferred), how much ‘‘double taxation’’ does corporate capital actually face? In the extreme, if corporations ﬁnance all their investment by borrowing, there is no corporate tax imposed on investment; indeed, corporate tax liability is reduced because nominal interest payments—a portion of which simply compensates lenders for a loss in purchasing power—are tax deductible. Corporate capital structures are not exclusively debt, of course, but the presence of the debt-equity choice means that we must look more closely at the reasons for equity ﬁnance. While some theories argue simply that debt capacity is limited, and so that some corporate capital must be subject to double taxation, other theories suggest that the choice of equity capital indicates a tax preference for equity, i.e., that the overall tax on equity income is lower than that on debt income, so that the burden implied by debt ﬁnance represents an upper bound for the burden on corporate capital. Because the tax imposed on debt ﬁnance—with single taxation to the recipient—is similar to that on non-corporate capital, these theories, in turn, suggest that corporate capital may be favored by the tax system rather than being discriminated against. This implies, in turn, that the presence of the corporate income tax may not discourage corporate activity but also that increases in the corporate tax rate may simply be borne by shareholders who, as a result, derive lower beneﬁts from the corporate tax structure. Two such theories are those developed originally by Stiglitz (1973) and Miller (1977). Stiglitz argued that equity would be used by ﬁrms to ﬁnance only that portion of their value in excess of invested capital. For example, an entrepreneur already in possession of patents or other valuable intangible assets might be able to turn an investment of $1 million in plant and equipment into an enterprise worth $1.5 million. If the entrepreneur incorporated, he or she would wish to ﬁnance the $1 million investment using borrowed funds, so that the returns to capital could be sheltered from the corporate tax. But issuing any additional debt, up to the corporation’s full value of $1.5 million, would

18

Auerbach

require an immediate taxable distribution of funds to the owner/ entrepreneur.14 Thus, Stiglitz argued, corporate equity supported intangible assets within the corporation that had been accumulated without tax at either the corporate or individual level, and that would remain free of tax as long as the corporation could avoid paying dividends and the investor could avoid selling shares. Until then, only the additional income on these assets would be taxable annually, at the corporate level. Indeed, one can show that the decision of whether to cash in immediately or maintain ownership of the intangible assets in the form of equity depends only on whether the corporate tax rate is higher or lower than the individual tax rate, assuming that the tax treatment of the asset sale would be the same at different points in the future.15 That is, the effective tax rate on this component of equity is simply the corporate tax rate t—there is effectively no double taxation of corporate equity that arises in this manner. As it is empirically reasonable that the corporate tax rate will be less than or equal to the ordinary tax rate, tp , for well-to-do shareholders, this theory suggests that corporate equity may bear no higher burden than corporate debt, and hence that the corporate income tax imposes no additional burden. Further, given the same assumptions about the relationship between t and tp , the entrepreneur would gain nothing from eschewing the corporate form at the outset, for selling the intangible asset held in a non-corporate enterprise would still generate a tax liability at the capital gains tax rate, tg , and the annual returns on the asset held in non-corporate form would be taxed at rate tp ; the decision of whether to incorporate or not depends, again, on whether t is lower or higher than tp . Thus, a small increase in t that maintains the inequality t < tp would be borne by the entrepreneur-shareholders, at least to the extent that their original innovation activity was unaffected. While interesting and inﬂuential in its impact on subsequent research, Stiglitz’s theory fails to characterize most of the equity in the corporate sector. Recall that the theory predicts that debt will be used to ﬁnance capital expenditures. Hence, corporate debt should equal the corporate capital stock or, on a ﬂow basis, borrowing should equal capital expenditures. In fact, capital expenditures typically far exceed borrowing. In 2004, for example, U.S. non-farm, non-ﬁnancial corporations had $900 billion in capital expenditures and obtained $231 billion through credit markets (Board of Governors 2005, Table F.102). What can explain the remaining portion of equity? Here, the theory of Miller (1977) comes in. Miller focused on the heterogeneity of individual

Who Bears the Corporate Tax?

19

investors, arguing that, under a progressive tax system, there may be some investors in a high enough tax bracket that the extra taxation at the corporate level is more than offset by the preferential individual tax treatment of equity income. The after-tax return to equity holders from a dollar of pre-tax corporate returns is ð1 tÞð1 te Þ, where te is the effective individual tax rate on equity income. The return to debt holders is ð1 tp Þ. Hence, even though there is double taxation of equity income, equity faces a lower tax burden if: ð1 tÞð1 te Þ > ð1 tp Þ

or

tp te > tð1 te Þ

ð1:3Þ

That is, the taxes saved at the individual level exceed the net additional corporate taxes. According to Miller’s theory, investors with a tax preference for equity would hold equity, those with a tax preference for debt would hold debt, and corporations would be indifferent between the two, issuing enough of the two securities, in the aggregate, to satisfy the demands of investors. Assuming that the equity tax rate is some fraction of the ordinary tax rate, say, te ¼ ltp , the decision to hold equity, from the second equation in (1.3), becomes: tp ð1 lÞ > tð1 ltp Þ

or

tp >

t 1 tp 1 lð1 tÞ

ð1:4Þ

Expression (1.4) implies that investors will sort by personal tax rate: those with a personal tax rate below some critical level, tp , will hold only debt; those with a higher tax rate will hold only equity; and those at that critical tax rate will be indifferent. If the corporation’s before-tax rate of return equals r, then it will pay equity holders rð1 tÞ and debtholders r, reﬂecting the corporate-level tax differences. This sorting equilibrium is shown in ﬁgure 1.4. Even with investor heterogeneity, is it plausible that a signiﬁcant share of investors will have a tax preference for equity, based on expression (1.3)? Currently, the U.S. top rates of tax on corporations and individuals are 35 percent, so this would be impossible. Even before recent tax cuts, the top individual rate in recent years has not been substantially higher than the corporate rate since before 1981. Thus, a very low effective equity tax rate would be required, and this seems inconsistent with the fact that a substantial share of equity earnings come to investors as (until 2003) fully taxed dividends. However, according to the ‘‘new view’’ of dividend taxation (Auerbach 1979, Bradford 1981, King 1977), the effective rate of individual tax on equity may be the

20

Auerbach

Figure 1.4 The Miller Equilibrium

capital gains rate (adjusted for deferral—a very low rate—even if dividends are distributed) when retained earnings are the source of equity ﬁnance, as they are for most large corporations. Further, the relevant corporate tax rate in expression (1.3) may be below the statutory rate if corporations face limits on their ability to deduct additional interest payments, an issue that has been found to be relevant empirically in various studies.16 Thus, through potentially low values of te and t, expression (1.3) may be satisﬁed for a number of investors, and it is these investors’ portion of shareholder wealth, not of the shareholder population, that determines the extent of equity preference in the Miller model. The Miller model has interesting implications for the incidence of the corporate tax. Because investors holding equity are taxed at a lower rate than they would be holding debt, the corporate tax is a tax shelter—equity-holders would be worse off if corporate-source income were treated just like non-corporate income. Thus, an increase in the corporate tax rate will reduce the tax beneﬁt from holding equity, but unless the preference for equity over debt disappears, it will not affect portfolio choice. The dashed line in ﬁgure 1.4 illustrates the impact on equilibrium of an increase in the corporate tax rate from t to t 0 . Investors with personal tax rate above tp0 will continue to hold equity but will receive a

Who Bears the Corporate Tax?

21

lower return for doing so—they bear the full brunt of the corporate tax increase. Investors with tax rates between tp and tp0 will shift from equity to debt; doing so allows them to avoid a portion of the corporate tax increase. Investors with a personal tax rate below tp will not be affected at all. Unlike in the Harberger model, there is nowhere for those with tax rates above tp0 to go because equity is still their tax preferred asset—after-tax returns on different assets are not equal for them.17 Thus, there is no shift out of corporate equity for investors with wealth above tp0 and, because debt is always an option, no need for capital to shift out of the corporate sector, even for those investors with tax rates between tp and tp0 who now choose not to hold corporate equity.18 According to the Miller model then, an increase in the corporate tax is largely borne by shareholders—yet another way in which shareholders may bear the corporate tax. But leaving aside whether expression (1.3) is satisﬁed for an adequate portion of shareholder wealth, there is another serious challenge to the Miller model— investors clearly do not specialize. A large share of the portfolios of tax-exempt institutional investors takes the form of equity, and at least some corporate bonds are held in the portfolios of higher-income individuals. As discussed by Auerbach and King (1983), the Miller model breaks down when assets are risky and investors must balance the objectives of diversiﬁcation and tax minimization. High-bracket investors may wish not to hold only equity, and low-bracket investors may wish to hold a portion of their portfolio in higher-yielding risky assets such as equity. Tax preferences will inﬂuence portfolios—those in higher brackets will still gravitate toward assets, like equity, with more favorable individual tax treatment. This modiﬁcation of the model implies that the incidence conclusions based on the simple Miller model are overly strong; while high-bracket investors suffer more from an increase in the corporate tax because of their higher concentration in equity, even tax-exempt investors will bear some of the burden as well. A second implication is that corporate bonds and non-corporate equity are no longer perfect substitutes, tax considerations aside, so that investors ﬂeeing from corporate equity may need to look outside the corporate sector for their investments.19 Thus, the predictions of the Harberger model, that owners of corporate capital are hit initially by an increase in the corporate tax and that this leads to a shift of capital outside the corporate sector, are partially reestablished by modifying the Miller model to incorporate risk.

22

Auerbach

7.1 The Incidence of Corporate Integration Proposals The previous discussion shows that having two levels of tax on corporate-source income doesn’t necessarily imply double taxation of that income, in the sense of a cascade of corporate and individual rates. The structure of corporate and individual taxation may allow some investors to face marginal tax rates on corporate-source income that are a little higher or even lower than their tax rates on ordinary income. Just as having two levels of tax doesn’t equate simply to double taxation, reducing tax rates at one level doesn’t translate simply into marginal tax rate reductions. Proposals for the ‘‘integration’’ of corporate and individual income taxes typically do not involve full integration of the two taxes, in the sense of treating C corporations like S corporations or partnerships. As mentioned above, this would be difﬁcult, given the complexity of allocating income to different classes of shareholders in the modern C corporation. Rather, integration proposals and integration schemes in practice elsewhere in the world generally involve reduced taxation of dividends, reﬂecting the assumption that dividends face a higher individual tax burden and that ﬁrms can choose to pay earnings as dividends (or can be deemed to have done so) to qualify for the tax beneﬁt.20 What is the incidence of adopting such schemes, starting from the current U.S. system? A ﬁrst observation is that schemes can be made roughly equivalent regardless of whether they are imposed at the corporate or shareholder level. For example, a dividends-paid deduction for the corporation (also called a split-rate system) equates to a dividends-received credit for corporate taxes paid (also called an imputation system) as long as the latter is refundable to shareholders (such as tax-exempt investors) whose tax liability is insufﬁcient to cover the credit. Each of these schemes, the standard approaches to integration in practice, amounts to a reduction in the tax rate on dividends. This leads to a second observation: that the incidence of a corporate tax reduction depends on the manner in which corporate taxes are reduced. Reducing the corporate tax rate and reducing the tax rate on corporate dividends are not the same policy, even if the tax reductions are both implemented through a reduction in corporate tax payments and have the same revenue costs. Indeed, under the new view of equity taxation discussed above, the dividend tax does not impose a marginal tax rate on new corporate capital investment but is capitalized into the value of corporate shares. Thus, a reduction in that tax does not reduce the marginal

Who Bears the Corporate Tax?

23

tax rate on corporate capital but simply increases the value of corporate shares.21 This highlights yet another possible way, in addition to those already explored above, in which an increase in corporate taxation (in this instance, an increase in the rate of tax on dividends) would be borne by existing shareholders rather than being spread to other current and future owners of all capital. 8.

Risk

Since the work of Domar and Musgrave (1944), economists have noted that taxes on capital income provide insurance as well as imposing burdens. Consider an arms-length asset-market investment that yields a risky rate of return at rate r, which has an expected value greater than the safe rate of return, i. We may decompose the return on the risky asset into two components, the safe rate of return and the excess return: r ¼ i þ ðr iÞ

ð1:5Þ

As has been established in the literature,22 a proportional tax system that provides a full loss offset (that is, the same tax rate applies whether income is positive or negative) imposes a burden on investors only to the extent that the ﬁrst component on the right-hand side of equation (1.5), the risk-free return, is taxed. Put another way, for a hypothetical tax system that imposes a tax rate t on the safe return and a tax rate t on the excess return, leaving the investor with an after-tax return of: ið1 tÞ þ ðr iÞð1 t Þ

ð1:6Þ

the investor is indifferent to the value of t . The reason is that the investor can undo taxation of excess returns simply by holding more of the risky asset and less of the safe asset. This result, combined with the empirical observation that the real, safe rate of return is very close to zero, led Gordon (1985) to suggest that the corporate income tax imposes few economic distortions, although it collects tax revenue on average (i.e., in expected value). One could also express this argument as saying that the corporate tax has little incidence to attribute because it imposes little burden. What of the revenues the government collects? Under Gordon’s view, the revenues have positive expected value but have little market value to the investors who forgo them because of their risk. If capital markets

24

Auerbach

already spread risk efﬁciently, these revenues will be of no greater value to the government than to taxpayers because any pooling that could reduce aggregate risk would already have been done by the private sector. Hence, under this argument, a uniform corporate tax is of little consequence, and we need not devote much thought to concerns about its distortions and its incidence. This theory, however, must confront an obvious empirical contradiction. Since excess returns on the risky asset must sometimes be negative—otherwise, the risky asset would dominate the safe asset— corporate tax revenues should sometimes be negative as well, even if they are positive on average. There are basically three different explanations for this contradiction, with different implications for tax incidence. The ﬁrst explanation is that corporate earnings include more than a normal safe rate of return and an excess return to risk-taking, i.e., economic rents. If such rents are included in the corporate tax base, revenue will be positive. The incidence and distortions associated with this component of revenue depend on whether the rents respond to taxation. Pure economic rent in a competitive market will not respond, and hence a tax on it would be borne by shareholders. The second explanation is that the tax on excess returns isn’t symmetric, as characterized in expression (1.6). As the overall return to capital equals gross returns less actual depreciation, a tax on excess returns would include depreciation allowances that track actual depreciation. As depreciation allowances do not vary in this way, taxpayers face a higher burden than expression (1.6) implies (Bulow and Summers 1984). Also, the tax on excess returns should be negative when excess returns are negative, but loss offsets are limited; this, too, increases the prospective tax burden on investors. In both of these cases, the corporate tax would impose a net burden on investors even if the safe rate of return were zero, with corporate investment being discouraged and incidence analysis once again relevant. The third possible reason for corporate revenues being positive is that private capital markets may not be fully efﬁcient. If individual investors do not fully pool risks, then assets that are risky from the investor’s perspective, and hence yield excess returns, may not be as risky from the government’s perspective; only the risks common to all assets would remain once the government pooled its revenue from the assets. In this case, the revenue would have value to the government but not to the taxpayers, and it could be positive in all aggregate states

Who Bears the Corporate Tax?

25

of nature. Because the revenue, once pooled, could then be redistributed to taxpayers or spent by government, it ultimately would have value to the population of taxpayers. In this case, the corporate tax on excess returns would have negative incidence—it would impose no initial burdens but would make at least some individuals better off. However, this potential explanation for persistently positive corporate tax revenues would not seem particularly relevant given the very large share of tax revenues attributable to extremely large companies, the vast majority of which are easily traded on major stock exchanges. In summary, the fact that corporate revenues are risky reduces the burden of corporate taxation. Given that corporate tax revenues are always positive, though, the corporate tax cannot be seen simply as a symmetric tax on excess returns. The necessary modiﬁcation of theory could mean higher burdens on shareholders; higher burdens on capital-owners more generally; or, less plausibly, negative burdens, depending on why corporate revenues are positive. 9.

Imperfect Competition

We have evaluated the impact of a tax on the normal return to corporate capital and on the excess return to corporate capital that is attributable to risk. But are there other components of corporate proﬁts with which we must deal? The question of economic rents has already been discussed above. Once one subtracts the normal return to capital providers and the return to risk, any proﬁts that remain represent a rent received by the corporation’s owners. But this rent could come from many sources, with different consequences for incidence. Corporate rents could simply represent the earnings on ideas, as discussed above in relation to Stiglitz’s (1973) theory. In this case, the corporation tax might effectively be a tax on entrepreneurial labor, for it would reduce the present value of the efforts that lead to the development of intangible capital; that is, the garages of Silicon Valley might have been used to store cars if the corporate tax rate had been higher. Corporate rents can also arise in a competitive model if there are decreasing returns to scale in production. In this case, theory tells us that a tax on rents imposes no distortions and is borne by shareholders. Finally, corporate rents can arise from imperfect competition. In the simplest case of monopoly provision, the consequences are the same— a tax on corporate rent is not distortionary because a monopolist is already maximizing before-tax rent and cannot do better once the tax

26

Auerbach

is imposed. But under more complicated types of imperfect competition, before-tax rents due to imperfect competition could respond to taxation.23 In an oft-cited empirical study, Krzyzaniak and Musgrave (1963) examined the behavior of corporate taxes and corporate proﬁts over time and came to the startling conclusion that after-tax proﬁts rose in the short run in response to increases in the corporate tax rate: shareholders actually beneﬁted from corporate tax increases, even in the short run! This conclusion necessarily points to imperfect competition because theory under competitive markets predicts that a tax on corporate rents cannot be shifted at all, and (as discussed above in relation to ﬁgure 1.2) a tax on corporate capital will only gradually be shifted over time. The study’s methodology does not allow one to identify the nature of corporate responses, but presumably corporations in the world of Krzyzaniak and Musgrave raise proﬁts by restricting output and increasing product prices, thereby passing the corporate tax on to the consuming population. The presence of imperfect competition would also inﬂuence the corporate response to taxation of the normal return to capital. Noncompetitive rents occur in the ﬁrst place because producers restrict output below the competitive level. A further tax on one of these inputs, in this case, capital, would lead to further restriction of output. Starting from a point where output is already restricted, it is possible that producers will over-shift in response to this tax as well—prices could rise by more than the increase in costs.24 Thus, as corporations respond to the increase in the corporate tax rate, there could be an even greater shift of capital out of the corporate sector than the Harberger model predicts, although this is not an unambiguous prediction. One thing is sure, though—a tax on production in an industry in which output is already restricted by imperfect competition will be more distortionary than one in a competitive environment because it exacerbates an already existing distortion. 10.

The Structure of Production

A key assumption of the Harberger model is that corporate and non-corporate enterprises produce different commodities. This was obviously a simpliﬁcation given that Harberger divided industries into corporate and non-corporate sectors based on each industry’s predominant, not universal, organizational form. As a logical matter,

Who Bears the Corporate Tax?

27

though, having corporate and non-corporate producers of the same commodities poses a problem for incidence analysis, for if production methods and organizational form (for tax purposes) can be chosen separately, then the corporate form will be adopted if and only if its tax treatment is preferred. Thus, the coexistence of corporate and non-corporate entities producing the same commodity requires that either (1) organizational form and production techniques are not independent, or (2) producers have access to different technologies, some of which beneﬁt from corporate tax treatment and some that achieve a lower tax burden outside the corporate sector. Gravelle and Kotlikoff (1989) model the corporate–non-corporate distinction following the ﬁrst of these approaches, assuming that incorporation facilitates operation on a larger scale, while non-corporate operation facilitates the use of entrepreneurial ability. With a scarcity of entrepreneurs, production will balance between the corporate and non-corporate sectors at the point where the non-corporate advantage provided by entrepreneurial ability is just offset by the scale economies provided by incorporation. In this model, and maintaining Harberger’s other major assumptions, the corporate income tax is still bad for capital, which is driven out of the corporate sector, but it is good for entrepreneurs, whose services are in greater demand. The impact on labor is uncertain, depending on relative substitution elasticities in corporate and non-corporate production; if workers and entrepreneurs are sufﬁciently complementary, the increased demand for entrepreneurs will help workers as well. Although they offer an explanation of the coexistence of corporate and non-corporate ﬁrms in the same industry, Gravelle and Kotlikoff do not test this explanation. Subsequent empirical analyses by Gordon and Mackie-Mason (1997) and Goolsbee (1998) ﬁnd much smaller implied, within-industry responses to changes in the relative taxation of corporate and non-corporate income, and hence much smaller implied deadweight losses from differential taxation than Gravelle and Kotlikoff report. These and other contributions to the literature also emphasize, following the second approach listed above, why differences among ﬁrms (with respect to risk, for example) might lead some to opt for corporate taxation and others to prefer taxation as noncorporate entities. In one result of note, Gordon and Mackie-Mason ﬁnd that increases in the tax ‘‘price’’ of being in corporate form attract ﬁrms with negative taxable income but deter ﬁrms with positive taxable income. This result also highlights the dynamic nature of the

28

Auerbach

choice of organizational form. Although transition between corporate and non-corporate form is far from costless, the availability of the two forms may, over time, provide ﬁrms with a net tax beneﬁt, even if one form is usually preferred, by allowing them to switch when tax incentives dictate. This is unlikely to be a realistic option at an annual frequency, but it might be relevant over a ﬁrm’s life cycle and especially in the transition from (usually loss-making) start-up to (usually proﬁtmaking) established company (Cullen and Gordon 2002). While this more recent literature has made progress in understanding the impact of taxes, particularly the corporate income tax, on the choice of organizational form, and has provided estimates of the deadweight loss of this impact, there has been limited analysis of the implications for corporate tax incidence. 11.

International Issues

In recent decades, the U.S. economy has experienced a steady increase in the importance of international trade and capital ﬂows. Between 1987 and 2003, the net stock of private U.S. ﬁxed capital, valued at current cost, rose from $10.71 trillion to $24.82 trillion,25 at an annual growth rate of 5.4 percent. By comparison, U.S. privately owned assets abroad, at current cost, rose over the same period from $1.39 trillion to $7.37 trillion, at an annual growth rate of 11.0 percent, and foreignowned U.S. assets (excluding foreign ofﬁcial assets) rose from $1.44 trillion to $8.23 trillion, at an annual growth rate of 11.5 percent.26 The signiﬁcance of the international investment channel has immediate implications for incidence analysis because capital ﬂeeing the U.S. corporate income tax now has an alternative potential destination that is much bigger than the U.S. non-corporate sector and is therefore much more able to absorb the capital without driving down the pretax rate of return. Aside from the expanded size of the relevant capital market, however, considering the international capital market affects corporate tax incidence analysis in three other important ways. First, the corporate tax burden need not be borne fully by domestic residents but can potentially be partially ‘‘exported’’ abroad. Second, there is an added dimension of tax rules to analyze, governing how cross-border ﬂows are treated by different countries; one must know whether the corporate tax is essentially source-based or residence-based, for example. Finally, with other governments’ tax systems involved, their responses are relevant to analyzing the effects of U.S. tax changes.

Who Bears the Corporate Tax?

29

With capital mobile in a large world capital market, one’s intuition might be that capital owners should be more able to avoid the burden of U.S. corporate taxation. But this intuition is misleading. As discussed by Bradford (1978) and Kotlikoff and Summers (1987), the total burden on capital need not fall with the ability of the United States to affect the worldwide rate of return because, at the same time, the burden of U.S. taxation is being spread over a larger total, worldwide capital stock. What does fall as the United States becomes small relative to the world capital market is its ability to export the burden of a sourcebased capital income tax, even if some capital is foreign owned. How can this be so, if capital bears all of the tax? The answer, in this model, is that the tax is also being borne by other domestic factors (in this case, land), while comparable factors abroad gain.27 The tax treatment of cross-border ﬂows adds considerable complexity to the analysis of taxation and its effects, including incidence analysis. Unlike in the purely domestic context, there is a distinction between where income is earned and where its owner resides, and the concept of residence itself is applied not only to individuals but also to corporations. Countries may seek to tax corporate income on a source basis, a residence basis, or some combination of the two, and most countries follow this last approach, taxing at least some income at source at the corporate level, even if the corporation is owned abroad, and taxing at least some portfolio income of domestic residents on holdings of foreign assets. But the effects of a change in, say, the U.S. corporate tax rate cannot even be considered without ﬁrst characterizing the equilibrium that might result in this very complex tax environment. As in the analysis underlying the Miller model, an equilibrium with individuals possessing different relative tax preferences for different assets leads to specialization of the highest-bracket investors in the most tax-favored assets (Gordon 1986), but the number of possible allocations of assets among investors is increased by the fact that individuals may hold foreign assets in many countries and in a variety of ways (e.g., portfolio investment versus direct investment), and corporations (and, to a lesser extent, individuals) can change the location not only of their investments but also of their tax residence. To this complexity of individual and ﬁrm choices, one must add the strategic interactions of governments in their choice of tax systems. A thorough discussion of the effects of corporate taxation in this context is well beyond this paper’s scope; the reader is referred to the survey by Gordon and Hines (2002). But

30

Auerbach

some important threads of the literature as it relates to incidence can be highlighted. 11.1 Residence Versus Source Taxation Taxation of corporate capital income on a residence basis would seem to leave less scope for shifting than taxation on a source basis. In the former case, the U.S. tax rate would apply wherever the capital moved, so it would be harder to avoid. But residence is not immutable, particularly for corporations. Thus, a residence-based corporate income tax might induce less shifting of capital but more shifting of residence.28 It is also important to remember that the U.S. corporate income tax, although sometimes referred to as being a residence-based tax, has many features that cause it to resemble a source-based tax. First, by allowing tax credits for foreign income taxes, the U.S. tax collects little additional tax on foreign-source income of U.S. corporations. Second, income of foreign subsidiaries is taxed only upon repatriation to the United States. Thus, little foreign-source income is subject to tax, and at low net tax rates. 11.2 Tax Rate Versus Tax Base As already discussed, changes in the corporate tax burden effected through changes in the tax base, as through depreciation provisions, have a different impact than equal-revenue changes in the corporate tax rate because the two policies have different relative effects on old and new capital. In the international context, there is an added reason why these policies’ effects would differ—the ability of ﬁrms to locate corporate income independently from corporate capital through the mechanism of ‘‘transfer’’ pricing (or the prices assigned to transactions between related parts of the ﬁrm located in different tax jurisdictions). Variations in transfer prices can be used to shift income among jurisdictions but are responsive to the tax rates on additional income, not to investment-related provisions. Thus, transfer pricing reduces the relative efﬁciency of investment incentives. Simultaneous increases in the corporate tax rate and investment incentives that hold constant the marginal effective tax rate on new investment would no longer just impose a capital levy on existing corporate assets but would be partially shifted through behavioral responses. The corporate tax rate could have similar effects even if shifts in corporate income location were limited to actual changes in investment location rather than to transfer pricing. If the use of capital in production

Who Bears the Corporate Tax?

31

in a given location is ‘‘lumpy’’ and does not obey the assumption of constant returns to scale, then the investment decision also involves a discrete location decision; the decision will be not simply one of plant size in each location but also of where to locate the plants. Without constant returns, there may be economic rents associated with the location decision, and the tax on these rents will depend only on the corporate tax rate. Thus, while the investment decision, conditional upon location, may be analyzed as in the domestic context, the location decision will depend on the combined burden on capital and rents, strengthening the impact of the corporate tax rate (Devereux and Grifﬁth 2003). In this context, corporate shareholders can shift not only a tax on corporate capital but one on rents as well. 12.

Managerial Issues

Economists stress that only individuals and not entities can bear tax burdens. From this perspective, it is difﬁcult to see the logic of a separate tax on corporations. With no retreat from the position that only individuals bear taxes, though, there may be something to the view of the corporation as a separate entity to be taxed in the sense that corporate managers, as a group, may be affected by the corporate tax in ways that differ from the effects of other taxes and may in turn have an objective different than proﬁt maximization. This discussion will relate primarily to the corporate tax rate, rather than to investment-related tax provisions, because the issues all concern the extent to which behavior by corporate managers, holding investment ﬁxed, affects the reported corporate tax base. In one limited sense, the corporate income tax should have no impact on the behavior of managers. As employee compensation is tax deductible, it is still in the corporation’s interest to pay employees, including managers, their before-tax marginal products.29 But there are other respects in which the existence of the corporate tax may affect managerial behavior.30 First, the corporate income tax reduces the after-tax cost to shareholders of managerial underperformance. Thus, to the extent that the costs of monitoring and acquiring information about managerial performance are not deductible from the corporate tax, the effect of the tax may be to reduce efﬁciency in managerial performance. Even if managers receive lower compensation as a result, the incidence of the increased inefﬁciency is still to be considered; as with a decline in

32

Auerbach

managerial input, the impact on capital and other factors would depend on relative complementarities in the structure of production. Second, tax compliance is largely a managerial decision, and a quite substantive one given the great complexity of the corporate tax system. It is customary to distinguish between (legal) tax avoidance and (illegal) tax evasion, but the choices are better characterized as being along a continuum of legal probability. In the standard model of individual income tax evasion, the individual trades off tax savings from successful evasion against the penalty if caught. The impact of an increase in the tax rate depends on a number of factors, including the agent’s risk aversion and the penalty structure.31 In the corporate context, the situation is even less clear because it is difﬁcult to know what motivates managers to evade on behalf of shareholders. The incidence of corporate tax evasion depends upon the ‘‘technology’’ of tax evasion. Following Slemrod (2004), we can think of two questions that affect the outcome: (1) is evasion general? Or (2) is it limited to a few managers? In the former case, evasion will reduce taxes for all concerned and may reduce effective tax rates, driving down the before-tax rate of return to capital; the effect would be like a corporate tax cut, though one with a lesser reduction in deadweight loss because of the resources wasted through evasion arrangements. In the latter case, the taxes saved will not be eroded through a reduction in beforetax returns, so the reduced burden will beneﬁt the shareholders and possibly managers. The second question involves the relationship between evasion and the scale of operations. If evasion is not part of the ‘‘constant returns’’ technology, but rather more of an inframarginal activity, then it resembles a lump-sum transfer to shareholders, even if it is widely practiced. As in the discussion at various points above, there is an important distinction here between the treatment of old and new capital; in this case, the question is whether the ‘‘evasion tax cut’’ extends to new capital. But the corporate tax evasion game is different from the individual game in a very fundamental way because there are not just two players, but (at least) three: the government, the manager, and the shareholder. The manager decides not only what to report to the government but also what to report to the shareholder, and these decisions are as distinct as tax accounting and ﬁnancial accounting. This leads to interesting interactions that have only recently been explored. As discussed by Desai and Dharmapala (2004), one needs to consider the technology that governs the two processes of hiding resources from

Who Bears the Corporate Tax?

33

the government and hiding resources from shareholders. Quite plausibly, the two processes are complementary; in which case, managers who are aggressive with respect to corporate tax evasion may also engage in large-scale diversion of shareholder resources. If this is true, then the act of evasion may not actually beneﬁt shareholders (even if there are no competing evaders to reduce before-tax returns), and strengthened corporate tax enforcement need not make shareholders worse off.32 In summary, the corporate tax may affect managerial behavior in two ways: through the incentive effects of changes in after-tax payoffs, and by giving rise to decisions regarding tax avoidance and evasion. While some of the potential incidence effects have been considered, this is an area in which the literature is still developing. 13.

Conclusions

Our journey beyond the Harberger model through the more recent literature takes us both forward and backward: forward in considering issues not previously studied, but backward in reestablishing the relevance of the shareholder incidence approach. For a variety of reasons, shareholders may bear a certain portion of the corporate tax burden. They may be unable to shift taxes attributable to a discount on ‘‘old’’ capital, taxes on rents, or taxes that simply reduce the advantages of corporate ownership. In the short run, they may also be unable to shift taxes on corporate capital. Thus, the distribution of share ownership remains empirically quite relevant to corporate tax incidence analysis, though attributing ownership is itself a challenging exercise. Another of this paper’s lessons is that one-dimensional incidence analysis—distributing the corporate tax burden over a representative cross-section of the population—can be relatively uninformative about who bears the corporate tax burden because it misses the element of timing. For example, for a tax that is shifted over time from shareholders to all owners of capital, as depicted in ﬁgure 1.2, the part not shifted will fall entirely on initial shareholders, while the part that is shifted will fall on future capital owners. Collapsing the burdens on shareholders and capital owners into a single cross section completely misses this important distinction. A related point is that it is more meaningful to analyze the incidence of corporate tax changes than of the corporate tax in its entirety because (1) different components of the tax have different incidence (e.g.,

34

Auerbach

a change in the corporate tax rate versus a reduction in corporate tax payments through a dividends-paid deduction), and (2) incidence relates to the path of the economy over time, not just in a single year; for example, it would make little sense to consider the incidence of one-year’s depreciation deductions for a long-lived investment. A further point is that corporate tax collections don’t equate to corporate tax incidence, even in the aggregate. A well-known reason for this distinction is the deadweight loss of taxation, which makes burdens exceed revenue collections. But burdens may also fall short of corporate revenue collections if (1) distortions are reduced (as in the case of improved risk-sharing) or (2) other taxes (e.g. personal taxes) are avoided. Finally, while exploring many extensions of the Harberger model, I have devoted little attention to one of that model’s important omissions: the impact of corporate income taxes on capital accumulation. But the implications are clear. For taxes on capital income, in general, we would expect an increase in the effective tax rate on new saving and investment to reduce capital accumulation. The resulting decline in the capital-labor ratio would increase before-tax returns to capital and lead to a fall in wages, thus partially shifting the tax burden from capital to labor. This analysis would apply to the corporate tax as well but only to the extent that the corporate income tax represents a tax on new saving and investment. The shift in the corporate tax burden from capital to labor can proceed only if it is ﬁrst shifted from shareholders. Notes This paper was presented at the NBER’s Tax Policy and the Economy conference, held on September 15, 2005, in Washington. I am grateful to Jim Poterba, Mihir Desai, Dhammika Dharmapala, Joel Slemrod, and conference participants for comments on earlier drafts. 1. U.S. Joint Committee on Taxation (1986, Table 1). This approach was hardly without its critics. See, for example, Feldstein (1988), who also took issue with Pechman’s methodology and developed one particular way of allocating the corporate tax changes of TRA86 based on the distribution of real capital income. 2. For further discussion, see McLure (1980) and Gordon and Wilson (1986). 3. The letters S and C stand for the corresponding subchapters of Chapter 1 of the Internal Revenue Code. 4. An alternative to the S corporation that also provides limited liability and passthrough tax treatment is the limited liability company (LLC). LLCs offer more ﬂexibility than S corporations in some dimensions, notably in not restricting the number of investors. They, too, have also grown in importance in recent years, although as of 2002, S cor-

Who Bears the Corporate Tax?

35

porations accounted for considerably more net income ($183.5 billion) than did LLCs ($48.6 billion). See, respectively, Luttrell (2005, Figure C) and Wheeler and Parsons (2004, Figure I). 5. This problem applies not only to these institutions’ direct holdings of corporate equity but also to their indirect holdings via mutual funds. 6. The 2005 PBGC takeover of some of United Airlines’ pension plans is a recent example. 7. See Bulow and Scholes (1983) for further discussion. 8. Indeed, one can consider additional groups of winners and losers from gradual adjustment to changes in the corporate tax burden. For example, Goolsbee (2003) ﬁnds that workers in industries that produce capital goods experience an increase in wages in response to tax-induced increases in investment demand. 9. This, indeed, was the approach taken by the Tax Reform Act of 1986, which sought to provide depreciation allowances equal in present value to economic deprecation but based on historic cost. 10. The term znew is the present value of allowances per dollar of new capital. 11. Auerbach, Gokhale, and Kotlikoff (1991) provide estimates of the impact of this adjustment on the generational incidence of taxation. 12. The 1986 act contained a number of other corporate tax provisions that accounted for a projected overall increase in corporate tax revenues. See Auerbach and Slemrod (1997), Table 1. 13. One can apply the same methodology to the most recent changes in capital recovery provisions, the temporary ‘‘bonus depreciation’’ schemes of 2002 and 2003. The provisions (described in more detail by Desai and Goolsbee 2004) provided immediate writeoff rather than depreciation for 30 percent (under the 2002 legislation) or 50 percent (under the 2003 legislation) of qualifying investment purchases (equipment investment, plus special-purpose structures with tax lifetimes of twenty years or less); this accelerated write-off acted like a small, temporary investment tax credit. The reduction in market value due to the new–old capital distinction should have been approximately 1 to 2 percent of the affected capital stock, with a portion of this being offset by gradual adjustment to the increased incentive to invest. The size of this latter effect depends not only on adjustment costs but also on expectations regarding the permanence of the provisions; but under reasonable assumptions, the net overall impact predicted is a decline in value of less than 1 percent of the affected capital stock. 14. The ﬁrm could, of course, invest additional borrowed funds in interest-bearing assets, with no net tax impact, but this action would not represent any change in the value of its equity. 15. Cashing in immediately would yield ð1 tg Þ for each dollar inside the ﬁrm, assuming capital gains tax treatment at rate tg . If the assets remained in equity form, they would accumulate at the rate rð1 tÞ, where r is the before-tax rate of return and t is the corporate tax rate. Cashing in at some future date T would thus yield a net amount of ð1 þ rð1 tÞÞ T ð1 tg Þ, providing an annual return of rð1 tÞ per dollar of funds retained as equity. Had the funds been withdrawn, they would have earned an annual return rð1 tp Þ, where tp is the ordinary individual tax rate.

36

Auerbach

16. See Auerbach (2002) for further discussion. 17. This specialization highlights another requirement of the Miller model: that investors not be able to hold unlimited short positions in either debt or equity. 18. These conclusions, like those of the original Harberger model, are based on the assumption of ﬁxed wealth. If corporate shareholders’ wealth accumulation declined in the face of higher tax rates, this could drive up before-tax returns to all investors, partially offsetting the initial impact of the corporate tax. 19. Some equity investors would also ﬂee the corporate sector if the Miller model were extended to include an additional tax-favored asset, e.g., tax-exempt municipal debt. In that case, investors in the very highest individual tax brackets would choose to hold municipal bonds, and an increase in the corporate tax rate would make such bonds attractive to individuals near the equity–municipal bond margin. As with a shift of investors into the non-corporate sector, this would spread the incidence of the tax increase to the returns on other assets. 20. See Graetz and Warren (1998) for a detailed discussion of various integration mechanisms developed in two studies during the early 1990s, one by the U.S. Treasury and the other by the American Law Institute. 21. Under the same theory, a reduction in the corporate tax rate would lower the cost of corporate capital. 22. For a good exposition of this and related results, see Kaplow (1994). 23. For one such analysis of tax incidence in this environment that takes the Harberger model as its starting point, see Davidson and Martin (1985). 24. See the discussion in Auerbach and Hines (2003). 25. U.S. Bureau of Economic Analysis (www.bea.gov, Fixed Asset Table 2.1, March 8, 2005). 26. Nguyen (2005), Table 2. 27. Gravelle and Smetters (2001) argue that these domestic U.S. factors will bear little of the corporate tax in the long run as a result of the large size of the United States and the additional market power conveyed by imperfect substitutability of foreign and domestic capital and commodities. 28. An example is the corporate ‘‘inversions’’ of recent years, when U.S. companies relinquished ‘‘parent’’ status to foreign subsidiaries to become, for tax purposes, subsidiaries of foreign corporations. See Desai and Hines (2002). 29. This does not hold exactly for stock options, which are deductible only when exercised rather than when granted, but the offsetting deferral of individual income tax should roughly offset this delay. 30. There will also be ways in which the structure of corporate and shareholder taxes may affect managerial decisions, for example, the decision whether to distribute earnings in response to the taxation of dividends. 31. See Slemrod and Yitzhaki (2002) for further discussion. 32. See also Crocker and Slemrod (2005) on this topic.

Who Bears the Corporate Tax?

37

References Aizcorbe, Ana M., Arthur B. Kennickell, and Kevin B. Moore (2003). ‘‘Recent Changes in U.S. Family Finances: Evidence from the 1998 and 2001 Survey of Consumer Finances,’’ Federal Reserve Bulletin, 89(1):1–32. Auerbach, Alan J. (1979). ‘‘Share Valuation and Corporate Equity Policy,’’ Journal of Public Economics, 11(3):291–305. Auerbach, Alan J. (1983). ‘‘Corporate Taxation in the United States,’’ Brookings Papers on Economic Activity, 14(2):1451–1505. Auerbach, Alan J. (1989). ‘‘Tax Reform and Adjustment Costs: The Impact on Investment and Market Value,’’ International Economic Review, 30(4):939–962. Auerbach, Alan J. (1996). ‘‘Capital Allocation, Efﬁciency and Growth,’’ in H. Aaron and W. Gale (eds.), Economic Effects of Fundamental Tax Reform. Washington: Brookings, 29–81. Auerbach, Alan J. (2002). ‘‘Taxation and Corporate Financial Policy,’’ in A. Auerbach and M. Feldstein (eds.), Handbook of Public Economics 3. Amsterdam: North-Holland, 1251– 1292. Auerbach, Alan J., and James R. Hines, Jr. (2003). ‘‘Perfect Taxation with Imperfect Competition,’’ in S. Cnossen and H. Sinn (eds.), Public Finance and Public Policy in the New Century. Cambridge: MIT Press, 127–153. Auerbach, Alan J., and Mervyn A. King (1983). ‘‘Taxation, Portfolio Choice and DebtEquity Ratios: A General Equilibrium Model,’’ Quarterly Journal of Economics, 98(4):588– 609. Auerbach, Alan J., Jagadeesh Gokhale, and Laurence J. Kotlikoff (1991). ‘‘Generational Accounts: A Meaningful Alternative to Deﬁcit Accounting,’’ in D. Bradford (ed.), Tax Policy and the Economy, 5:55–110. Auerbach, Alan J., and James Poterba (1987). ‘‘Why Have Corporate Tax Revenues Declined?’’ in L. Summers (ed.), Tax Policy and the Economy 1:1–28. Auerbach, Alan J., and Joel Slemrod (1997). ‘‘The Economic Effects of the Tax Reform Act of 1986,’’ Journal of Economic Literature, 35(2):589–632. Board of Governors of the Federal Reserve System (2005). Flow of Funds Accounts of the United States: Flows and Outstandings, First Quarter 2005, June 9. Bradford, David F. (1978). ‘‘Factor Prices May Be Constant but Factor Returns Are Not,’’ Economics Letters, 1(3):199–203. Bradford, David F. (1981). ‘‘The Incidence and Allocation Effects of a Tax on Corporate Distributions,’’ Journal of Public Economics, 15(1):1–22. Bulow, Jeremy I., and Myron S. Scholes (1983). ‘‘Who Owns the Assets in a Deﬁned Beneﬁt Pension Plan?’’ in Z. Bodie and J. Shoven (eds.), Financial Aspects of the U.S. Pension System. Chicago, Ill.: University of Chicago Press, 17–36. Bulow, Jeremy I., and Lawrence H. Summers (1984). ‘‘The Taxation of Risky Assets,’’ Journal of Political Economy, 92(2):20–39.

38

Auerbach

Crocker, Keith J., and Joel Slemrod (2005). ‘‘Corporate Tax Evasion with Agency Costs,’’ Journal of Public Economics, 89(9–10):1593–1610. Cullen, Julie Berry, and Roger H. Gordon (2002). ‘‘Taxes and Entrepreneurial Activity: Theory and Evidence for the U.S.,’’ NBER working paper no. 9015, June. Cutler, David (1988). ‘‘Tax Reform and the Stock Market: An Asset Price Approach,’’ American Economic Review, 78(5):1107–1117. Davidson, Carl, and Lawrence W. Martin (1985). ‘‘General Equilibrium Incidence Under Imperfect Competition: A Quantity-Setting Supergame Analysis,’’ Journal of Political Economy, 93(6):1212–1223. Desai, Mihir A., and Dhammika Dharmapala (2004). ‘‘Corporate Tax Avoidance and High Powered Incentives,’’ NBER working paper no. 10471, May. Desai, Mihir A., and Austan D. Goolsbee (2004). ‘‘Investment, Overhang, and Tax Policy,’’ Brookings Papers on Economic Activity, 35(2):285–355. Desai, Mihir A., and James R. Hines, Jr. (2002). ‘‘Expectations and Expatriations: Tracing the Causes and Consequences of Corporate Inversions,’’ National Tax Journal, 55(3):409– 440. Devereux, Michael P., and Rachel Grifﬁth (2003). ‘‘Evaluating Tax Policy for Location Decisions,’’ International Tax and Public Finance, 10(2):107–126. Diamond, Peter A., and James Mirrlees (1971). ‘‘Optimal Taxation and Public Production, I: Production Efﬁciency,’’ American Economic Review, 61(1):8–27. Domar, Evsey D., and Richard A. Musgrave (1944). ‘‘Proportional Income Taxation and Risk-Taking,’’ Quarterly Journal of Economics, 58(2):388–422. Feldstein, Martin (1988). ‘‘Imputing Corporate Tax Liabilities to Individual Taxpayers,’’ National Tax Journal, 41(1):37–59. Goolsbee, Austan (1998). ‘‘Taxes, Organizational Form, and the Deadweight Loss of the Corporate Income Tax,’’ Journal of Public Economics, 69(1):143–152. Goolsbee, Austan (2003). ‘‘Investment Tax Subsidies and the Wages of Capital Goods Workers: To the Workers Go the Spoils?’’ National Tax Journal, 56(1, part 2):153–166. Gordon, Roger H. (1985). ‘‘Taxation of Corporate Capital Income: Tax Revenues versus Tax Distortions,’’ Quarterly Journal of Economics, 100(1):1–27. Gordon, Roger H. (1986). ‘‘Taxation of Investment and Savings in a World Economy,’’ American Economic Review, 76(5):1086–1102. Gordon, Roger H., and James R. Hines, Jr. (2002). ‘‘International Taxation,’’ in A. Auerbach and M. Feldstein (eds.), Handbook of Public Economics 4. Amsterdam: North-Holland, 1935–1995. Gordon, Roger H., and Jeffrey K. Mackie-Mason (1997). ‘‘How Much Do Taxes Discourage Incorporation?’’ Journal of Finance, 52(2):477–505. Gordon, Roger H., and John D. Wilson (1986). ‘‘An Examination of Multijurisdictional Corporate Income Taxation under Formula Apportionment,’’ Econometrica, 54(6):1357– 1373.

Who Bears the Corporate Tax?

39

Graetz, Michael J., and Alvin C. Warren, Jr. (1998). Integration of the U.S. Corporate and Individual Income Taxes: The Treasury Department and American Law Institute Reports. Arlington, Va.: Tax Analysts. Gravelle, Jane G., and Laurence J. Kotlikoff (1989). ‘‘The Incidence and Efﬁciency Costs of Corporate Taxation When Corporate and Noncorporate Firms Produce the Same Good,’’ Journal of Political Economy, 97(4):749–780. Gravelle, Jane G., and Kent Smetters (2001). ‘‘Who Bears the Burden of the Corporate Tax in the Open Economy?’’ NBER working paper no. 8280, May. Harberger, Arnold C. (1962). ‘‘The Incidence of the Corporation Income Tax,’’ Journal of Political Economy, 70(3):215–240. Hayashi, Fumio (1982). ‘‘Tobin’s Marginal and Average q: A Neoclassical Interpretation,’’ Econometrica, 50(1):213–224. Kaplow, Louis (1994). ‘‘Taxation and Risk-Taking: A General Equilibrium Perspective,’’ National Tax Journal, 47(4):789–798. King, Mervyn A. (1977). Public Policy and the Corporation. London: Chapman and Hall. Kotlikoff, Laurence J., and Lawrence H. Summers (1987). ‘‘Tax Incidence,’’ in A. Auerbach and M. Feldstein (eds.), Handbook of Public Economics 2. Amsterdam: North-Holland, 1043–1092. Krzyzaniak, Marian, and Richard A. Musgrave. (1963). The Shifting of the Corporation Income Tax. Baltimore, Md.: Johns Hopkins Press. Luttrell, Kelly (2005). ‘‘S Corporation Returns 2002,’’ Statistics of Income Bulletin 24(4):59– 113. McLure, Charles E. (1980). ‘‘The State Corporate Income Tax: Lambs in Wolves’ Clothing,’’ in H. Aaron and M. Boskin (eds.), The Economics of Taxation. Washington: Brookings. Miller, Merton (1977). ‘‘Debt and Taxes,’’ Journal of Finance, 32(2):261–276. Nguyen, Elena L. (2005). ‘‘The International Investment Position of the United States at Yearend 2004,’’ Survey of Current Business, 85(7):30–39. Pechman, Joseph A. (1985). Who Paid the Taxes, 1966–85? Washington: Brookings. Shoven, John B. (1976). ‘‘The Incidence and Efﬁciency Effects of Taxes on Income from Capital,’’ Journal of Political Economy, 84(6):1261–1283. Slemrod, Joel (2004). ‘‘The Economics of Corporate Tax Selﬁshness,’’ NBER working paper no. 10858, October. Slemrod, Joel, and Shlomo Yitzhaki (2002). ‘‘Tax Avoidance, Evasion, and Administration,’’ Handbook of Public Economics 3. Amsterdam: North-Holland, 1423–1470. Stiglitz, Joseph E. (1973). ‘‘Taxation, Corporate Financial Policy and the Cost of Capital,’’ Journal of Public Economics, 2(1):1–34. Summers, Lawrence H. (1981). ‘‘Taxation and Investment: A Q Theory Approach,’’ Brookings Papers on Economic Activity, 12(1):67–127. Tobin, James (1969). ‘‘A General Equilibrium Approach to Monetary Theory,’’ Journal of Money, Credit and Banking, 1(1):15–29.

40

Auerbach

Treubert, Patrice E. (2004). ‘‘Corporation Income Tax Returns, 2001,’’ Statistics of Income Bulletin, 24(1):118–140. U.S. Joint Committee on Taxation (1986). Data on Distribution by Income Class of Effects of the Tax Reform Act of 1986, JCX-28-86, October 1. Wheeler, Tim, and Maureen Parsons (2004). ‘‘Partnership Returns, 2002,’’ Statistics of Income Bulletin 24(2):46–125.

2 Tax Reform and Entrepreneurial Activity Julie Berry Cullen and Roger Gordon, University of California, San Diego

Executive Summary The objective of this paper is to forecast the effects of plausible tax reforms on the extent of entrepreneurial activity in the United States. To do so, we draw on recent work we have done assessing the many routes through which the tax structure affects the amount of entrepreneurial activity, and estimating the responsiveness of behavior to these incentives. Using these estimates, we forecast that the effect of tax reforms on entrepreneurial activity can be very sensitive to whether or not current tax provisions aimed to encourage risk-taking in small ﬁrms remain part of the tax code. If they are left in place, we forecast that a shift to a Hall-Rabushka ﬂat tax will leave the overall amount of entrepreneurial activity largely unaffected, although it will lead to a drop in activity among the highly skilled and an offsetting increase in activity among the less highly skilled. However, if in the process of fundamental tax reform, ‘‘net operating loss’’ carrybacks are disallowed, and section 1244 allowing capital losses on equity in small businesses to be reclassiﬁed as ordinary losses is repealed, then overall entrepreneurial activity could fall by more than half. The devil is in the details. Tax reform will soon be at the top of the policy agenda, following the release of the tax reform proposals from the president’s Tax Reform Commission. Tax reform has many possible objectives. The objective we focus on in this paper is to unleash the creative energies in the economy through making entrepreneurial activity more attractive. To what degree will any of the likely reforms that might be proposed help to encourage more entrepreneurial activity? Why a focus on entrepreneurial activity per se? Going back at least to Schumpeter (1976), entrepreneurial activity has been viewed as the

42

Cullen and Gordon

key source of economic growth. The presumption has been that individuals and ﬁrms, when they come up with new products, new processes, or new ways of organizing economic activity, generate far more beneﬁts to society than they receive personally. As a result, the incentives that an individual faces to engage in entrepreneurial activity are inadequate. The potential for such ‘‘externalities’’ is pervasive. The new ideas generated by an entrepreneur can often be copied by another ﬁrm. In many cases, the copying ﬁrm succeeds in capturing the market and the resulting proﬁts.1 With competition, beneﬁts are also shared with consumers through a lower price. New ideas can also make it easier for future entrepreneurs to identify proﬁtable opportunities in other markets, beneﬁts again not received by the initial entrepreneur.2 To the extent entrepreneurial activity generates such spillovers, there are economic grounds to try to intervene to encourage more such activity. The tax system can provide this kind of encouragement in a variety of ways. Differential tax treatment of business versus wage and salary income or of losses versus proﬁts can change the incentives to engage in a risky venture. In addition, entrepreneurial ﬁrms are normally closely held, in part due to lemons problems.3 Since an individual with a valuable idea for a new product and the skills needed to bring this idea to fruition may not be willing to impose the resulting risks on herself and her family, the misallocation of risks caused by these lemons problems implies less innovative activity. By absorbing a share of the losses as well as the proﬁts, taxes help share the risks faced by entrepreneurs with other taxpayers and can thereby increase the amount of entrepreneurial risk-taking. The focus of our recent research has been on assessing, both qualitatively and quantitatively, how the tax structure affects the amount of entrepreneurial activity. In this paper, we build on our past work to assess the degree to which past tax reforms have affected the aggregate amount of entrepreneurial activity. We then examine how possible tax reform proposals would likely affect the amount of entrepreneurial activity in the United States in coming years. 1.

What Is Entrepreneurial Activity?

In order to assess the effects of the tax structure on entrepreneurial activity, the ﬁrst task is to come up with some measure of the extent of entrepreneurial activity. The key concern is to measure behavior that

Tax Reform and Entrepreneurial Activity

43

generates externalities to others. Externalities presumably arise from new information about feasible technologies or organizational structures and their proﬁtability. Our presumption is that the key source of new ideas generating such spillovers is start-up ﬁrms. To begin with, any investment in a highly risky activity should initially be small to avoid putting too many funds at risk until more is known about the likely success of the project. It should also be much easier to produce an entirely new product or to produce a product in an entirely new way if one can design a ﬁrm from scratch around this concept rather than redesign an existing ﬁrm. Within an existing ﬁrm, employees have human capital invested in the production of existing products and existing internal forms of organization, while the shareholders have a ﬁnancial claim on proﬁts from existing products made obsolete by a new product, thus making both groups of stakeholders reluctant to shift course. Even if these stakeholders ultimately agree to try out a new product, the process of reaching this decision will take time, by which point a start-up will likely have seized the initiative.4 Not all start-up ﬁrms, of course, produce information of much value to others. For example, shifting from being an employee to being an independent contractor is simply an accounting change, with little or no change in real activity. To measure the extent of innovative activity in a start-up, we focus on the degree of risk-taking within such ﬁrms because producing new information is inherently risky. We then attempt to measure the degree to which the tax law affects risk-taking among start-up ﬁrms.5 2.

Taxes and Risk-Taking

The most obvious element of the tax law that potentially affects innovative activity is the research and development (R&D) tax credit. Hall and Van-Reenen (2000) document that the credit has been effective in increasing the amount of R&D activity in the United States. A key restriction with the existing credit, though, is that it is nonrefundable. This implies that it is of much less use to start-up ﬁrms since start-up ﬁrms have no current proﬁts and may not have proﬁts for many years on which tax liabilities can be reduced through use of the R&D credit. Even if the R&D credit is effective at encouraging R&D in existing profitable ﬁrms, it therefore provides little or no help in dealing with externalities generated by start-up ﬁrms. Yet according to the evidence

44

Cullen and Gordon

reported in Bound et al. (1984), R&D expenditures in small ﬁrms lead to patents at a much higher rate than do equivalent expenditures in larger ﬁrms, making it of particular value to encourage R&D in smaller ﬁrms. There are many other ways in which the tax law can affect the amount of risk-taking in addition to the R&D tax credit, however. To make clear these many other routes, we focus ﬁrst on effects assuming that the entrepreneur is risk-neutral. We then examine what additional effects arise when the entrepreneur is risk-averse. 2.1 Net Tax or Subsidy to Risk-Taking Assume then that the entrepreneur is risk-neutral. She currently earns a wage of, say, $100,000 as an employee, but is considering becoming an entrepreneur instead. New businesses are very risky. To capture this, assume that the ex post income the individual would get as an entrepreneur has an equal chance of being $100,000 or $300,000. In this simple example, mean income is unaffected by becoming an entrepreneur since .5($100,000 þ $300,000) ¼ $100,000, so that, ignoring taxes, this risk-neutral individual would be indifferent between the two options. How do taxes affect this choice? Assume to begin that the individual has other income (perhaps spouse’s income) of $150,000, yielding adjusted gross income (AGI) of $250,000 as an employee and either $50,000 or $450,000 as an entrepreneur. Consider ﬁrst a very simple tax structure that provides $30,000 in exemptions and deductions and that has a ﬂat tax rate of 25 percent on all further income. With such a constant marginal tax rate, the individual will continue to be indifferent to the choice between being an employee or an entrepreneur. The comparison is now between earning 250,000 .25 220,000 ¼ 195,000 as an employee or earning an expected income of :5½ð50;000 :25 20;000Þ þ ð450;000 :25 420;000Þ ¼ 195;000 as an entrepreneur. Compared to being an employee, an entrepreneur can end up with either $200,000 less income or $200,000 more income pretax, and G$150,000 after tax. The expected change in income from becoming an entrepreneur remains equal to zero. Ex ante, enough taxes are saved on any shortfall in income arising from self-employment to offset the extra taxes due when the business turns out to raise income on net. To see these effects of the tax structure visually, examine ﬁgure 2.1, which compares the tax payments made by an employee (labeled E)

Tax Reform and Entrepreneurial Activity

45

Figure 2.1 Tax Payments Under an Illustrative Proportional Income Tax

with the range of possible tax payments the individual owes if instead she becomes an entrepreneur (the line labeled N). If the individual is risk-neutral, then taxes distort the choice to take on entrepreneurial risk to the extent that the expected tax payments as an entrepreneur differ from those paid as an employee. In this example, if ex post AGI as an entrepreneur is always above $30,000, so that the individual always faces a 25 percent marginal tax rate, and if expected entrepreneurial income equals income as an employee, then expected tax payments are unaffected by the choice to become an entrepreneur, or even the choice concerning how risky a project to pursue. The particular outcomes described in the above example satisfy both restrictions. In general, though, any tax structure where the marginal tax rates vary over the relevant range of incomes can affect this individual’s career choice. Consider, for example, a tax schedule in which any income between the exempt amount of $30,000 and $140,000 is taxable at a tax rate of 15 percent, while any income above that is taxable at a rate of 35 percent. By construction, the individual’s after-tax income as an employee is still $195,000. However, the individual’s expected income after tax as an entrepreneur now drops to $186,000. A risk-neutral individual will no longer ﬁnd this project attractive enough to justify becoming an entrepreneur. Ex ante, any tax savings when the ﬁrm has losses no longer offset the additional taxes due if the ﬁrm turns out to

46

Cullen and Gordon

Figure 2.2 Tax Payments Under an Illustrative Progressive Income Tax

be successful. Given the less favorable tax treatment in this example, the proﬁts when the ﬁrm is successful would need to increase to at least $327,700 to be sufﬁcient to induce the individual to consider becoming an entrepreneur. Figure 2.2 compares the tax payments made by an employee versus an entrepreneur under this progressive tax schedule. With a progressive tax schedule, tax payments are a convex function of AGI. If the ~Þ > TðEY ~Þ ¼ tax function Tð:Þ were strictly convex, we know that ETðY ~ TðYE Þ, where Y denotes the random entrepreneurial income, and YE denotes income as an employee. Taxes then discourage entrepreneurial activity. This result continues to hold, even if the tax schedule is only weakly convex, as long as there is some chance that the individual would end up in a different tax bracket as an entrepreneur than she would as an employee. In the example, this occurs as long as there is some chance that the individual’s AGI will end up below $140,000. The higher the probability that the individual will end up in a different tax bracket than she would as an employee, and the greater the difference between the tax rates, the larger is the tax penalty for becoming an entrepreneur. The probability of ending up in a different tax bracket will vary depending on the initial income level of the entrepreneur and the amount of risk faced by the ﬁrm. If the ﬁrm had been less risky in the above example, yielding a pretax return that is always above

Tax Reform and Entrepreneurial Activity

47

$140,000, then the tax law is again neutral—labor income is simply taxed at a marginal rate of 35 percent, regardless of its form. Conversely, if the ﬁrm is even riskier, then the tax penalty arising from the progressive tax schedule becomes more important. Consider, for example, the choice whether to start up a business with payoffs of $100,000 and $300,000 or one with payoffs of $100,100 and $300,100. A riskneutral individual would be indifferent if taxes are not a factor. With a constant marginal tax rate, of say, 35 percent, the two choices remain equally attractive. But under the progressive tax schedule described above, the extra losses are ð1 :15Þ 100 while the extra proﬁts are only ð1 :35Þ 100, implying a net expected loss of $10 on the extra risk. Progressive taxes therefore discourage this extra risk-taking for individuals who have already decided to start a new business. Here the net tax penalty depends directly on the difference in the relevant marginal tax rates: per dollar of extra spread in outcomes, the net expected gain is :5 ðð1 :35Þ ð1 :15ÞÞ ¼ :5 ð:15 :35Þ. The tax penalty on risk-taking also varies depending on the income level of the individual. For an individual with higher outside income, pushing point E in ﬁgure 2.2 to the right, the probability of ending up with AGI in a different tax bracket drops; it rises if the individual has lower outside income. With a symmetric distribution of possible outcomes, the tax penalty on entrepreneurial activity would be greatest for an individual with income as an employee exactly at the kink point.6 The existing personal income tax includes many other provisions, in addition to the basic rate structure, that either exacerbate or lessen this tax distortion. The most important examples are probably the earned income tax credit (EITC), the payroll tax, and the option ﬁrms face to incorporate rather than to operate in noncorporate form. Consider ﬁrst the implications of the EITC. Due to the credit, an entrepreneur with a child faces more than a 30 percent subsidy to her net wage and/or self-employment income up to roughly $10,000, no taxes on subsequent income until $15,000, then a surtax of roughly 20 percent until the credit has been fully repaid (which with these ﬁgures occurs at an income of $30,000). Introducing the EITC causes a shift in the tax schedule from that appearing in ﬁgure 2.2 to that seen in ﬁgure 2.3. Two new convex kinks are introduced at incomes of $10,000 and $15,000. Each of these convex kinks creates a penalty to risk-taking for the same reasons as occurred when the convex kink was created at an income of $140,000

48

Cullen and Gordon

Figure 2.3 Tax Payments Under an Illustrative EITC

(when we shifted from a proportional to a progressive rate structure). For example, consider an individual with labor income of $15,000 (and no other sources of income), a point labeled E in ﬁgure 2.3, who might instead become an entrepreneur earning either $5,000 or $25,000 pretax. Taking advantage of the EITC, as an employee this individual would receive $18,000. As an entrepreneur, however, this individual’s expected income is only :5ð$6;500 þ $26;000Þ ¼ $16;250. Taxes also discourage extra risk-taking by this potential entrepreneur. If the ﬁrm could instead have payoffs of $4,900 or $25,100, the expected pretax income on the extra risk is $0, but the change in expected after-tax income from taking on the added risk is $25! The EITC strongly discourages risk-taking for such individuals. In addition, however, the EITC converts the convex kink at $30,000 in ﬁgures 2.1 and 2.2 into a slightly concave kink in ﬁgure 2.3 due to the phase-out of the recapture tax under the EITC. By making the tax schedule less convex in this range, the EITC can potentially encourage extra risk-taking if outcomes in this range are more relevant for any given individual. A second important complication is the payroll tax. The payroll tax adds both a new convex kink and a new concave kink to the overall tax schedule. Consider ﬁrst the convex kink. Under the payroll tax, self-employment proﬁts are taxable, but self-employment losses are

Tax Reform and Entrepreneurial Activity

49

Figure 2.4 Tax Payments Under an Illustrative Payroll Tax

not tax deductible. With an effective payroll tax rate of 9.1 percent, taking on extra risk to increase both potential business proﬁts and business losses by $100 implies a drop in expected after-tax income of $4.55.7 The concave kink occurs at the point where individuals earn the maximum income taxable under the Social Security and Disability Insurance (SSDI) components of the payroll tax (currently $90,000).8 For an individual with expected earnings as an entrepreneur just equal to this amount, above-expected earnings are free of most payroll taxes, while the individual saves on payroll taxes to the extent that earnings are less than expected. This phase-out of the payroll tax therefore creates in itself a subsidy to risk-taking. Figure 2.4 graphs the payroll tax as a function of AGI in the above example and shows both new kinks. A third important complication is the option the entrepreneur faces to check the box and pay taxes on this business income under the corporate tax rather than under the personal tax. When would this be attractive to do? The individual can make this choice ex post, knowing the taxable income that year, though must stick with this choice for at least ﬁve years. Based on tax incentives alone, entrepreneurs should ﬁnd it attractive to check the box if the business has (and should continue to have) net proﬁts, though they will normally prefer to remain noncorporate while the ﬁrm has net losses. Why is this?

50

Cullen and Gordon

When the ﬁrm has proﬁts, the key issue is the relative size of the personal tax rate versus the corporate tax rate on these proﬁts. The initial bracket under the corporate tax is 15 percent, though this applies by statute to only the ﬁrst $50,000 of corporate proﬁts. However, as described in detail in Sommerfeld and Jones (1991), a ﬁrm can be divided into multiple corporations, with each ﬁling corporate taxes separately. If there are no real costs of doing this, as they argue is the case for reasonably small ﬁrms, then the corporate proﬁts can all be subject to just this 15 percent corporate tax rate. Corporate proﬁts are also subject to further taxes when they are taken out of the ﬁrm as dividends or realized capital gains. We focus on capital gains realizations since this is the more tax efﬁcient way to extract money from a corporation. The conventional estimate of the effective capital gains tax rate, coming from Feldstein, Dicks-Mireaux, and Poterba (1983), is one-quarter of the statutory tax rate, reﬂecting tax savings from deferral and write-up of basis at death. With a statutory capital-gains tax rate of 15 percent and a 50 percent exclusion of capital gains from small business stock held for at least ﬁve years (under IRC section 1202), this implies an effective capital gains tax rate of just :25 :5 :15 ¼ :01875 on the proﬁts left net of the corporate tax. The overall tax rate on corporate proﬁts is then :15 þ :01875 :85 A :166. If the ﬁrm were instead to remain noncorporate, the marginal personal income tax rate on the income will be at least 15 percent in the above example if the individual’s AGI is above $30,000. Noncorporate proﬁts are also subject to statutory payroll taxes at an effective tax rate of approximately 9.1 percent.9 Given that half of payroll tax payments are deductible from taxable income under the personal income tax, the overall tax rate on noncorporate proﬁts for an individual with AGI of at least $30,000 is a minimum of :15 ð1 :5 :153Þ þ :091 ¼ :23. Clearly, almost all individuals do better by using the corporate tax schedule whenever the ﬁrm has proﬁts. What about when the ﬁrm has losses? If the ﬁrm remains noncorporate when it has losses, then these losses can be deducted against other personal income under the personal tax that year, reducing the entrepreneur’s personal taxes immediately in proportion to her marginal tax rate. When losses are large enough to make net income negative, then these ‘‘net operating losses’’ can be carried back (or carried forward) and used to generate an immediate refund of taxes paid in

Tax Reform and Entrepreneurial Activity

51

another tax year in proportion to the marginal tax rate faced in that year.10 For a corporation, in contrast, business losses can only be offset against proﬁts earned in past or future years for that particular business. For entrepreneurial start-ups, there are no past proﬁts and there is a reasonably high probability that the ﬁrm will not succeed, in which case it has no future proﬁts either. Even when the ﬁrm does have future proﬁts, the corporate taxes saved on tax-loss-carryforwards depends on the corporate marginal tax rate, which we argued above will likely be only 15 percent. Current business losses, however, also reduce the capital gain or increase the capital loss for the entrepreneur on the equity the entrepreneur has in the ﬁrm. The resulting tax savings on reduced capital gains would be trivial, as argued above. However, IRC section 1244 allows capital losses on small business stock to be treated as ordinary losses under the personal income tax, regardless of holding period, and imposes much higher limits on the amount of losses that can be taken compared with other capital losses.11 If the capital loss simply equals the accumulated annual losses until the business is shut down, then the deduction for capital losses under the personal income tax is the same as would have occurred if the ﬁrm remained noncorporate, except that the deductions are deferred in time.12 Whether it is advantageous on tax grounds to be corporate or noncorporate if the ﬁrm has losses then depends on the probability the ﬁrm with current losses ultimately has a capital gain or a capital loss, and on the personal marginal tax rate the individual faces now on noncorporate losses compared with the expected marginal tax rate the individual would face in the future on realized capital losses (correcting for the loss due to deferral of these tax savings). Regardless of all these complications, the option to be corporate or noncorporate in itself encourages more entrepreneurial risk-taking. Firms with proﬁts will choose whichever form faces the lowest tax rate, while ﬁrms with losses will choose the form generating the greatest tax savings from these losses. To the extent that the option to incorporate reduces the tax rate on proﬁts or increases the tax savings on losses, risk-taking is encouraged. Figure 2.5 captures the tax implications of the option the ﬁrm has to incorporate. Here, we assume the same progressive tax schedule under the personal tax as used in ﬁgure 2.2, but now we allow the ﬁrm to

52

Cullen and Gordon

Figure 2.5 Effects of Checking the Box

incorporate whenever it is advantageous on tax grounds. As described above, incorporation is advantageous whenever the ﬁrm has proﬁts, which occurs in our example for AGIs above $150,000. In our example, the ﬁrm will also choose to incorporate when it would otherwise be in a zero personal tax bracket since then the ability to deduct capital losses on the corporate equity from personal taxable income provides at least some tax savings. Comparing ﬁgure 2.5 with ﬁgure 2.2, we ﬁnd that the effective tax schedule has become less convex, so that the option to check the box encourages entrepreneurial risk-taking. Note that this option to check the box can make being an entrepreneur more attractive than being an employee even if the ﬁrm has no risk. An employee would end up at point E in ﬁgure 2.5 since wages and salaries are necessarily taxed under the personal income tax. Business income in contrast can be taxed as corporate income if the individual chooses to check the box. When the effective tax rate on corporate income is below the relevant personal tax rate, as is true in this example, then this option to incorporate encourages entrepreneurial activity even for a ﬁrm facing no risk. As a result, there are separate tax incentives affecting whether to be self-employed and then, if an individual is self-employed, how risky a project to pursue.

Tax Reform and Entrepreneurial Activity

53

2.2 Net Sharing of Risks with the Government When the entrepreneur is risk-averse, the tax law can potentially provide a further encouragement to entrepreneurial activity by sharing the resulting risks with the government. In order to characterize the behavior of a risk-averse individual, assume that the potential entrepreneur has constant relative risk aversion and maximizes EUðYÞ ¼ E ln Y.13 Given this degree of risk aversion and if we ignore taxes, an individual would be indifferent between being an employee earning $100,000, with outside income of $150,000, and being an entrepreneur if the potential payoffs from being an entrepreneur are equally likely to be either $300,000 or $11,100. How would taxes affect this choice? If we impose a ﬂat tax rate of 25 percent on all income, with no exemptions or deductions, then behavior does not change. This is a consequence of our assumption here that the utility function has constant relative risk aversion. With a ﬂat tax rate of 25 percent on all income above an exempt amount of $30,000, in contrast, being an entrepreneur becomes slightly more attractive. The individual would now be indifferent if the payoffs from being an entrepreneur were either $300,000 or $13,000 pretax. Now, shifting to a progressive tax structure can easily generate yet more entrepreneurial activity. For example, when the tax rate instead is 15 percent on income between $30,000 and $140,000, and 35 percent on any further income, the after-tax income of an employee earning $100,000 with $150,000 of outside income is unaffected. However, for an entrepreneur, after-tax income when the ﬁrm does badly is higher than when the individual faces a ﬂat 25 percent tax rate. Though aftertax income when the ﬁrm does well is much lower under this progressive schedule, raising net-of-tax income when the individual is badly off is more important to a risk-averse individual than an offsetting reduction in income when the ﬁrm succeeds. With the assumed degree of risk-aversion, the individual would be indifferent to becoming an entrepreneur if the payoffs from being an entrepreneur were either $300,000 or $17,800 pretax. As a result, more people will choose to become entrepreneurs under this progressive tax structure than under the proportional tax structure, and those who do become entrepreneurs would choose to take on more risk. To see more formally why shifting to a progressive rate schedule can encourage entrepreneurial risk-taking, compare the utility of the entre~ð1 tf ðY ~ÞÞÞ and EUðY ~ð1 preneur under the two tax systems: EUðY

Figure 2.6 Change in Tax Payments and Expected Utility from a Shift to a Progressive Tax

54 Cullen and Gordon

Tax Reform and Entrepreneurial Activity

55

~ÞÞÞ, where tf ðY ~Þ describes tax payments under a ﬂat tax on all intðY ~Þ describes a nonlinear tax come above an exempt amount, while tðY schedule on all income above this same exempt amount. Using a ﬁrstorder Taylor approximation, expected utility under the nonlinear tax structure can be approximated by: ~ð1 ~tÞÞ Q EUðY ~ð1 ~tf ÞÞ E½ð~t ~tf ÞY ~ U 0 ðY ~ð1 ~tf ÞÞ: EUðY

ð2:1Þ

Whether a progressive tax encourages or discourages entrepreneurial risk-taking, compared with a ﬂat tax, depends on the sign of the second term in equation (2.1). This term is a weighted sum of the changes in tax payments across income realizations, where the weights are given by the associated marginal utilities of income ðU 0 Þ. Since U 0 is a declining function of net-of-tax income, the tax savings at lower income levels under a progressive rate structure receive more weight, whereas the extra tax payments at higher income levels receive less weight. For a risk-averse individual, this greater weight on tax savings than on tax surcharges can easily make entrepreneurial risk-taking more attractive under a progressive tax than under a proportional tax. Figure 2.6 graphs these tradeoffs. We ﬁrst graph the change in overall tax payments at each income level (labeled DT) arising from the use ~. With a proof a progressive rather than a proportional tax: ð~t ~tf ÞY gressive schedule, the entrepreneur pays less taxes when the ﬁrm does badly and more taxes when the ﬁrm does well. As before, expected tax payments are higher under a progressive tax. We then graph the loss in utility, denoted DU, arising from use of a progressive tax under our previous assumption that UðYÞ ¼ lnðYÞ. The tax savings under a progressive rate schedule if the ﬁrm does badly now receive much more weight relative to the potential tax surtaxes if the ﬁrm does well. The expected utility loss can easily be negative, so that utility from entrepreneurship is higher under a progressive rate schedule than under a ﬂat tax. 2.3 Tax Evasion One omission from the preceding discussion is the greater ease of tax evasion on business income than on wage and salary income. Andreoni et al. (1998), quoting statistics from past Tax Compliance and Measurement Program (TCMP) audits in the United States, reports estimates of evasion rates among sole proprietorships ranging from 16 percent to 39 percent, depending on industry, with the highest evasion rates in retail trade and the lowest rates in wholesale trade, ﬁshing,

56

Cullen and Gordon

agriculture, real estate, and ﬁnance.14 This greater ease of tax evasion certainly makes self-employment per se more attractive. Yet greater opportunities for tax evasion can easily reduce the amount of risktaking among those who do become entrepreneurs, implying ambiguous net effects on overall entrepreneurial risk-taking. Why might this be the case? We expect that individuals will take advantage of any opportunities to avoid tax ex post whenever the probability of being caught and the resulting ﬁnes are low enough. Such evasion means that net-of-tax incomes become more risky ex ante. Subsidies to risk-taking and risk-sharing with the government both become less important. For both of these reasons, the optimal amount of risk-taking can drop when evasion is easier, even though the utility of the entrepreneur is certainly higher. So, though more people may become self-employed due to the greater opportunities to evade taxes on business than on wage income, those who choose selfemployment are likely to choose less risky projects. Given the ambiguity of the role of tax evasion, this is the one effect described above that we do not account for explicitly in the empirical work reported on below. 3. Empirical Evidence on the Net Effects of Taxes on Entrepreneurial Activity The above discussion lays out the range of complications in the tax law that potentially affects entrepreneurial risk-taking. This discussion, however, provides no information about the quantitative importance of any particular provision in the tax code for the incentives to engage in entrepreneurial risk-taking. To judge this, we must turn to the data. In doing so, we draw heavily on our past work, reported in Cullen and Gordon (2005). 3.1 Evidence Based on Past Data and Tax Law In Cullen and Gordon (2005), we begin by deriving an expression capturing the overall effect of the tax structure on entrepreneurial risktaking, capturing each of the effects described above. In particular, we start with an explicit measure of individual utility and calculate the optimal extent of entrepreneurial risk-taking for each individual given the range of possible outcomes that could result and the resulting tax treatment for each possible outcome. Individual choices include whether to become self-employed; if so, the size of the business (own labor input,

Tax Reform and Entrepreneurial Activity

57

number of employees, and capital investment); and then the riskiness of that business. In capturing tax incentives, the paper takes account of the progressivity of the personal tax schedule, complications in the personal tax law such as the EITC, the tax treatment of ‘‘net operating losses,’’ the speciﬁc treatment of capital gains on small business stock, the choice whether to incorporate some or all of the ﬁrm’s activities, and the effects of the payroll tax if the ﬁrm remains noncorporate. The resulting expression for tax incentives depends on several behavioral parameters, in addition to characteristics of the tax law. For one, the degree of risk aversion matters since this affects the importance of any risk-sharing with the government. In addition, the typical degree of riskiness of a start-up ﬁrm relative to an entrepreneur’s overall income matters. The importance of risk considerations relative to the tax savings that occur with a new business even without risk also needs to be inferred from the data. The next step in jointly estimating these parameters determining effective tax incentives along with the degree of responsiveness of entrepreneurial risk-taking to tax incentives is to construct a measure of entrepreneurial risk-taking. Our perception was that the ideal would be a measure of the riskiness of the individual’s business activity relative to that individual’s earnings ability. The extent of an individual’s risk-taking is inherently unobservable, however, depending on ex ante assessments. All we can potentially observe is ex post outcomes. We chose to focus on a simple indicator for the extent of entrepreneurial risk-taking equal to a dummy variable measuring whether or not the individual reported noncorporate losses exceeding 10 percent of reported wage and salary income.15 While proﬁts of any given size can arise even without risk, if the size of the ﬁrm is appropriate, losses should occur only if the ﬁrm has undertaken a risky project.16 The riskier the project or the larger the project for any given degree of riskiness, the higher is the individual’s overall risk-taking relative to his or her earnings ability, and the higher is the probability of having ex post losses above any given fraction of their reported labor income. In Cullen and Gordon (2005), we then estimate the various parameters involved in both the measurement of tax incentives and the responsiveness of individual behavior to these tax incentives to best explain variation in the extent of entrepreneurial risk-taking, using a data set consisting of individual tax returns for roughly 2 million single individuals spanning twenty-two years between 1964 and 1993.17 In estimating the responsiveness of behavior to tax incentives, we

58

Cullen and Gordon

calculated the average tax incentives within various quantiles of the potential earnings distribution18 and then used this information to forecast the fraction within each quantile that reported large noncorporate losses.19 Using a difference-in-difference estimation strategy, we then examined how this fraction changes over time for one quantile relative to another compared with how relative tax incentives change; formally, we did this by including quantile and year dummies in the speciﬁcation. The parameters affecting the measurement of tax incentives, and the responsiveness of behavior to these tax incentives, were then chosen to best explain the patterns observed in the data. We restricted the study to single individuals because we had too little information about the choices made by married couples to calculate tax incentives properly.20 Our forecasted sensitivity of behavior to tax incentives as a result is based solely on the behavior of single individuals. Any inference about the behavior of married couples is speculative. If married couples are equally responsive to tax incentives as single individuals, and if the weighted average change in tax incentives is the same for single ﬁlers as for all tax payers, then our estimates for the percentage changes in entrepreneurial risk-taking would still be appropriate. We presume that married individuals tend to have higher earnings ability than single individuals, though. Since the change in tax incentives varies by earnings level, our forecasts below for individuals earning higher incomes should get more weight when forecasting effects for the population as a whole. In presenting results based on our prior estimates, we ﬁrst describe how the weighted average net tax incentives to engage in entrepreneurial risk-taking have varied over time for selected years between 1964 and 2001.21 Results are reported graphically in ﬁgure 2.7. In making sense of the patterns in this ﬁgure, it helps to step back to think through the role of different elements in the tax structure. As described above, all but the lowest ability entrepreneurs have an incentive to incorporate their business whenever it has proﬁts but to remain noncorporate as long as the business has tax losses. Expected after-tax proﬁts then increase to the extent that personal tax rates are high relative to corporate tax rates, since to that extent the entrepreneur has to bear a smaller fraction of the losses net of tax than she gets to keep of any proﬁts. In addition, given risk-aversion, a higher overall marginal tax rate implies more risk-sharing with the government. Since the entrepreneur then has to bear less risk, the marginal cost of risk-bearing should be lower, thus encouraging more risk-taking.

Tax Reform and Entrepreneurial Activity

59

Figure 2.7 Weighted Average Net Tax Incentive by Year

The reported numbers in ﬁgure 2.7 reﬂect the interaction of many changing provisions in the tax code. For example, between 1968 and 1972 a key change discouraging entrepreneurial risk-taking was a general drop in marginal tax rates, implying less risk-sharing with the government.22 This effect was partially offset, though, by an increase in the payroll tax rate. During the rest of the 1970s, inﬂation caused bracket creep, pushing taxpayers into steadily higher personal tax brackets. With a higher personal tax rate, entrepreneurs bore a smaller fraction of any business losses, while business proﬁts largely were taxed at the corporate rate and so were unaffected by this bracket creep. This bracket creep therefore encouraged more entrepreneurial risk-taking. The tax reforms during 1981–1983 introduced a series of cuts in personal, corporate, and capital gains tax rates. Such a drop in tax rates discourages risk-taking through reducing the sharing of risk with the government, and in particular cutting the sharing of losses with the government. With the 1986 tax reform, personal tax rates fell substantially relative to the effective tax rate on corporate income, particularly given the jump in the effective capital-gains tax rate. As a result, entrepreneurs bore yet more of their losses and could keep less of any proﬁts, thus discouraging risk-taking. The other main change in tax incentives occurred in 1994. In this year, the capital gains tax rate on small business stock was cut in half, encouraging more risk-taking.

60

Cullen and Gordon

Figure 2.8 Net Tax Incentive by Potential Earnings Quantiles

The earnings limit for the Medicare payroll tax was eliminated in this year, penalizing wage and salary income relative to corporate business income. In addition, the EITC program was expanded dramatically, encouraging risk-taking for those eligible for the program primarily when their ﬁrm has losses. Figure 2.7 focuses on aggregate incentives only. This hides substantial heterogeneity within the population in the incentives faced. In ﬁgure 2.8, we graph how average tax incentives varied across different potential earnings quantiles over this same time period.23 Here, we ﬁnd that tax incentives to engage in entrepreneurial risk-taking are almost always higher when a person’s earnings ability is higher. Since proﬁts are taxed largely at the same corporate tax rate, regardless of quantile, those in higher tax brackets gain by having to absorb a smaller fraction of any business losses. Tax incentives are also more volatile for those in the top tax brackets. Tax incentives do not move in parallel across earnings levels; in fact this is the source of identiﬁcation in our estimation procedure. How important are these tax incentives for behavior? Figure 2.9 graphs the actual data on the fraction within each quantile that reports large entrepreneurial losses. Comparing ﬁgure 2.8 with ﬁgure 2.9

Tax Reform and Entrepreneurial Activity

61

Figure 2.9 Loss Rates by Ability Quantile

shows that tax incentives and behavior appear to be closely linked across ability groups, and there are some striking shared movements across years. Our estimates are not based on the overall correlation, however, but instead depend on differing changes in behavior across quantiles as a function of differing changes in tax incentives. Focusing on this more subtle and useful source of variation, we ﬁnd strong links between taxes and risk-taking. We see this in the results reported next, simulating the behavioral responses to possible tax reforms. 3.2 Forecasts Based on Possible Future Tax Laws We now turn to simulating the effects of possible tax reforms on the extent of entrepreneurial risk-taking. The latest year of data we have for actual behavior is 2001. We chose to use these data as our benchmark, simulating how behavior in 2001 would have differed had the tax law been different in 2001. Row (1) of table 2.1 reports data on the fraction of taxpayers within each potential earnings quantile predicted to have large entrepreneurial losses in 2001, given the 2001 sample of non-joint returns and the 2001 tax code.24 Since entrepreneurial risk-taking is heavily concentrated among the highest earning ability individuals, we again deﬁne

62

Cullen and Gordon

Table 2.1 Forecasted Rates of Active Losses Potential Earnings Quantile Tax Provisions

(0,70)

(70,80) (80,90) (90,95) (95,99) (99,100)

Aggregate

Actual tax law (1)

2001 tax law

0.010

0.023

0.020

0.022

0.044

0.124

0.023

(2)

2005 tax law

0.008

0.018

0.016

0.018

0.035

0.094

0.018

0.008

0.017

0.017

0.019

0.032

0.066

0.016

Successive reforms (3) Removing taxation of ﬁnancial income (4)

Removing special treatment of small business capital gains

0.006

0.013

0.012

0.013

0.022

0.046

0.012

(5) Flat personal tax rate (19%)

0.009

0.020

0.018

0.017

0.024

0.038

0.015

(6) Setting the corporate income tax rate to 19%

0.008

0.016

0.015

0.014

0.020

0.031

0.013

(7) Disallowing carry forward/backward of net operating losses

0.005

0.010

0.010

0.009

0.013

0.020

0.008

(8)

0.007

0.013

0.011

0.011

0.017

0.031

0.011

Removing the EITC

Successive reforms (9)

Flat personal tax rate (19%)

0.014

0.032

0.027

0.026

0.037

0.067

0.024

(10) Removing taxation of ﬁnancial income

0.014

0.029

0.027

0.025

0.035

0.049

0.022

(11) Setting the corporate income tax rate to 19%

0.011

0.024

0.022

0.020

0.028

0.041

0.018

Notes: Each column shows the predicted rate of active business losses for the ability quantile indicated for our sample of non-joint taxpayers in 2001. The prediction equation is based on our prior estimation results, with the relevant tax incentives calculated under the tax provision indicated in each row. Each taxpayer is assigned to a potential earnings quantile based on predicted earnings as an employee, as described in note 18 in this paper. To calculate the aggregate rate shown in the ﬁnal column, we weight the rate for each quantile by its share of overall predicted earnings. In the ﬁrst row, the tax measures are calculated using the actual tax provision in place in 2001. The second row shows the predicted rates had the 2005 provisions (suitably adjusted for inﬂation) been in place. The remaining rows predict loss rates under various hypothetical reforms.

Tax Reform and Entrepreneurial Activity

63

the quantiles to focus on the behavior among those in this upper tail of the distribution of potential earnings. As seen in the table, the fraction reporting large losses in the top 1 percent of the earnings distribution is over twelve times as high as the overall fraction in the bottom 70 percent of the earnings distribution, and even almost three times as large as the fraction among the rest of the top 5 percent of the earnings distribution. The starting point for tax reform, of course, is not the tax law as of 2001 but the tax law in place currently. Row (2) of table 2.1 simulates how these reported fractions would have differed had the 2005 law been in place in 2001. The forecast is that entrepreneurial risk-taking has fallen by over 14 percent due to the tax reforms between 2001 and 2005. One aspect of the reform, the cut in the capital gains tax rate, in itself should have increased the amount of entrepreneurial risk-taking. Since small businesses already face such a low effective capital gains tax rate due to IRC section 1202, however, the effects of this change on small business activity would be muted. In contrast, the drop in personal tax rates, holding corporate tax rates ﬁxed, implies less sharing of losses with the government, making risk-taking more costly to an entrepreneur. Taxes on potential proﬁts, in contrast, largely remain unaffected by the tax reforms since they should largely be subject to the unchanged corporate tax rate. We then compare the forecasted rates of entrepreneurial risk-taking that would have occurred in 2001 under any particular proposal with those that we forecast would have prevailed in 2001 under the actual 2005 tax law. Since the president’s tax reform commission has not yet submitted its recommendations, we can only speculate about the types of reform proposals that it will come up with.25 Our presumption is that these tax reform proposals will include sharp cuts in the taxes due on ﬁnancial income from savings. The ﬁrst tax reform we then simulate is one that eliminates any taxes on interest, dividend, and capital gains income, though we leave in place the deductibility of mortgage interest payments. The results are shown in row (3) of table 2.1. As with the tax reforms between 2001 and 2005, there should be two offsetting effects on entrepreneurial risk-taking. First, the elimination of any taxes on capital gains income in itself will encourage more entrepreneurial risk-taking, though this effect will be small since the effective capital gains tax rate is already very low. In addition, the drop in taxable income due to the exemption of ﬁnancial income from savings will cause entrepreneurs to fall into a lower or

64

Cullen and Gordon

zero personal tax bracket more quickly, for any degree of business risk. Facing lower potential personal tax rates, a smaller fraction of business losses will be shared with the government, making risk-taking more costly. As with the 2005 reforms, this latter effect dominates overall— the forecasted rate of entrepreneurial risk-taking falls by 11 percent relative to the 2005 benchmark. Note, however, that effects are not uniform across quantiles since the relative importance of these two offsetting effects differs substantially across taxpayers. In the above simulation, even though capital gains were made tax exempt, we left in place a special provision (IRC section 1244) that allows owners of equity in small businesses to deduct any capital losses (up to $50,000 per year for a single individual) as ordinary losses rather than capital losses. If this provision were repealed as well,26 eliminating the tax deductibility of capital losses on small business stock, then entrepreneurial risk-taking would certainly fall further. The estimated effects of repealing section 1244, in addition to repealing the taxation of other forms of ﬁnancial income from savings, appear in row (4) of table 2.1. Overall entrepreneurial risk-taking is predicted to fall by a further 22 percent relative to the 2005 benchmark. This fall occurs roughly proportionately in all quantiles. We presume that another likely proposal for tax reform is a shift to a ﬂat tax rate. Here, we made an attempt to choose a ﬂat tax rate that would leave overall tax revenue unaffected,27 otherwise leaving the tax base unchanged.28 The rate we found was 19 percent. Shifting to such a ﬂat tax rate implies an increase in marginal tax rates for lower-earning individuals, and a drop in marginal tax rates particularly for the highest-earning individuals. As a result, lower earners who become entrepreneurs need to absorb a smaller fraction of any losses from their businesses, but higher-ability entrepreneurs have to absorb a larger fraction. Consistent with this, forecasted entrepreneurial risk-taking goes up among all quantiles other than the very top one, for an overall increase in risk-taking of 17 percent, judged from the 2005 benchmark. A ﬂat rate structure under the personal tax still leaves a difference in tax rates between noncorporate and corporate businesses. We next simulate the impact on entrepreneurial activity of increasing the corporate tax rate on small businesses from 15 percent to the presumed ﬂat tax rate of 19 percent. By increasing the tax rate on business proﬁts while leaving unaffected the tax rate on (noncorporate) business losses, risk-taking should be discouraged. Forecasted risk-taking in fact drops

Tax Reform and Entrepreneurial Activity

65

among all quantiles, for an overall drop of 11 percent relative to the 2005 benchmark. One other likely objective of tax reform is tax simpliﬁcation. The tax law includes a huge number of special provisions affecting virtually all aspects of the economy. With a major tax reform, one aim could be a wholesale repeal of these special provisions in order to generate a simpler and more transparent tax code. One provision of particular importance for noncorporate businesses is IRC section 172, which allows a carryback or carryforward of net operating losses into other tax years.29 A parallel provision, allowing income averaging, was repealed with the 1986 reform. What would be the implications for entrepreneurial risk-taking if section 172 were repealed in the process of tax simpliﬁcation? With such a restriction on the deductibility of large business losses, risk-taking would clearly be discouraged. The forecasted effects are reported in row (7) of table 2.1. Here, we start from the tax provisions in row (6). The ﬁgures show a sizeable drop in risktaking in all quantiles, for a further overall drop of 28 percent relative to the 2005 benchmark. Note that the tax law simulated in row (7) should largely correspond to the ﬂat tax proposed in Hall and Rabushka (1995), with all income (corporate and noncorporate) above an exempt amount subject to one tax rate, with other tax provisions complicating the current tax law (including sections 172 and 1244 that speciﬁcally affect risk-taking in small businesses) repealed. Therefore, our forecast is that such a fundamental tax reform would reduce entrepreneurial activity by over half. One further change that in principle could occur through fundamental tax reform is a repeal of the EITC. Certainly, if the ﬂat tax were implemented through a tax just at the ﬁrm level, with no remaining personal tax ﬁling, then the EITC would be difﬁcult to maintain. Row (8) simulates the impact of such a repeal, starting with the tax system in row (7). Here, we ﬁnd that entrepreneurial activity increases for all groups, implying that the EITC discourages entrepreneurial risktaking, at least starting from a Hall-Rabushka ﬂat tax.30 Among the various tax reforms we have considered, the key reform that in itself resulted in an increase in forecasted entrepreneurial activity is the shift to a ﬂat tax rate under the personal income tax. What would happen if this were the only change implemented? Results appear in row (9) of table 2.1. With a ﬂat tax, marginal tax rates go up for lower-income taxpayers, providing more sharing of business losses with the government. In contrast, marginal tax rates fall sharply for

66

Cullen and Gordon

those with the highest earnings. Consistent with this, we forecast an increase in entrepreneurial risk-taking among those in all but the highest ability group, for an overall increase in risk-taking by one-third. Starting from a ﬂat tax, if we then eliminated any taxes on interest, dividends, or capital gains, entrepreneurial activity again falls, and almost entirely among the highest ability group, as seen in row (10) of table 2.1. While the drop in taxable personal income cannot affect an entrepreneur’s marginal tax rate as long as she continues to have positive taxable income, the probability of being in a zero tax bracket increases. Again, there is less risk-sharing with the government. This reform affects primarily the top quantile simply because the ﬁnancial income from assets is much larger relative to potential labor income in this quantile. If we then proceeded as before to tax corporate income at this same ﬂat personal tax rate, forecasted entrepreneurial risk-taking again falls, as seen in row (11) of table 2.1, to a point effectively equal to that under current law. On net, as seen comparing row (11) with row (2), shifting from current law to such a Hall-Rabushka ﬂat tax, being careful to leave in place the special provisions encouraging entrepreneurial activity, has no net effect on overall entrepreneurial activity. However, such a tax change reduces sharply the amount of entrepreneurial risk-taking by the highest ability taxpayers, while encouraging more such risktaking by the rest of the population. Recall, however, that all of these ﬁgures reﬂect forecasts for entrepreneurial risk-taking among single individuals. Married individuals tend to be in higher tax brackets than single individuals. Assuming they are as responsive to tax incentives as single individuals, and that their tax incentives change to the same extent controlling for their initial tax bracket, we can forecast changes for the overall economy given these ﬁgures by putting more weight on changes in the higher tax brackets.31 As a result, we forecast some fall in entrepreneurial activity from fundamental tax reform, even under the optimistic assumption that the special provisions that encourage entrepreneurial risk-taking (IRC sections 172 and 1244) remain in place. 4.

Summary

Fundamental tax reform necessarily will generate complicated and farreaching changes in the U.S. economy. We focus on only one: entrepre-

Tax Reform and Entrepreneurial Activity

67

neurial risk-taking. To what degree would plausible directions for tax reform induce an increase in entrepreneurial activity? The ﬁndings are discouraging. Under the most dramatic change considered, a shift to a Hall-Rabushka ﬂat tax,32 forecasted entrepreneurial activity falls by half. The devil is in the details, however. Two special provisions under the current tax code (IRC sections 172 and 1244) are forecasted to have large effects on entrepreneurial risk-taking. Section 172 allows ‘‘net operating losses’’ in a noncorporate ﬁrm to be carried back or carried forward into other tax years. Section 1244 allows realized capital losses on shares in a small business to be deducted as ordinary rather than capital losses (up to $50,000 per year for a single individual). The above forecast assumes that both provisions would be repealed in the process of fundamental tax reform. If these provisions were left in place in the process of adopting a Hall-Rabushka ﬂat tax, for example, then we forecast that overall entrepreneurial activity would remain effectively unchanged. These are two small sections in the Internal Revenue Code among many others that could well disappear in the process of fundamental tax reform. We therefore close on a note of caution. The current complications in the law developed over many years, each in response to particular economic (and political) pressures. The result, of course, is a very complicated tax code, encouraging discussion of an abandonment of the existing income tax and replacing it with an entirely different tax structure. In the process, tax reform discussion focuses on the most basic aspects of the tax code, particularly the overall rate structure, and pays much less attention to these countless details in the tax code. Yet these details may have large economic effects that need to be given careful thought during discussion of fundamental tax reform. Certainly these two detailed provisions in the tax code appear to be the dominant consideration when forecasting the effects of fundamental tax reform on entrepreneurial risk-taking. Notes This paper was written for the NBER Conference on Tax Policy and the Economy, held in Washington, D.C., on September 15, 2005. 1. As an example, Sony was the key initial innovator in creating the video-cassette recorder, yet the market was ultimately captured by those using an imitating (VHS) technology. As another example, the Star desktop computer invented by Xerox included

68

Cullen and Gordon

many of the key innovations still in use in PCs. Yet Apple, then IBM, and later other imitators captured the market for PCs instead. 2. Of course, the proﬁts received by the initial entrepreneur may in part be at the expense of reduced proﬁts going to a prior entrepreneur, imposing negative externalities. We presume in this discussion that the positive externalities from entrepreneurial activity far outweigh these negative spillovers. 3. Lemons problems can be particularly severe for entrepreneurial ﬁrms, given that the value of the ﬁrm is heavily dependent on the likelihood, timing, and quality of a possible innovation, information that would be particularly hard for an outside investor to monitor. 4. Existing ﬁrms may still be effective at taking the innovative ideas produced within start-up ﬁrms and then manufacturing and marketing the resulting products. These later stages in the process are not the likely sources of any externalities, however. 5. Tax structures that encourage risk-taking encourage all forms of risk-taking, and not just those that generate important externalities to other ﬁrms and to customers of the ﬁrm, so that they are not ﬁnely targeted on entrepreneurial activity. However, entrepreneurial activity certainly generates risks and so would be encouraged by any tax structure that encourages risk-taking in start-up ﬁrms. 6. Gentry and Hubbard (2004) focus on the effects of such progressivity in the personal tax schedule by comparing the expected marginal tax rate an entrepreneur would face at particular outcomes for proﬁts versus losses, assuming risk-neutrality. The analysis we describe below includes many other features of the tax code, including the choice whether or not to incorporate, distortions created by the payroll tax, a variety of additional complications in the personal tax law such as the earned income tax credit and the treatment of ‘‘net operating losses,’’ as well as possible risk aversion. 7. The effective payroll tax is lower than the statutory rate since additional taxes paid are partly offset by higher future Social Security beneﬁts due to the extra covered earnings. Diamond and Gruber (1997) ﬁnd that the present value of extra beneﬁts should offset roughly half of the extra payroll taxes due. Applying this ﬁnding, we set the effective payroll tax rate to one-half the 12.4 percent statutory rate used to ﬁnance Social Security and Disability Insurance, plus the 2.9 percent statutory Medicare rate. 8. The payroll tax used for Medicare is not subject to an earnings ceiling and does not phase out. 9. We continue to follow Diamond and Gruber (1997) in assuming that half of the (nonMedicare) payroll taxes due on extra income are offset by extra future beneﬁts. 10. By statute, ‘‘net operating losses’’ are losses in excess of positive wage and capital income net of standard/itemized deductions. 11. Under current tax law, a couple can convert $100,000 in capital losses on small business stock into ordinary losses in each tax year. 12. The individual’s marginal tax rate can also change over time, however, for example, if the individual realizes the capital loss after returning to being an employee. 13. This utility function in fact embodies less risk aversion than existing evidence suggests is plausible. See Barsky et al. (1997) for further discussion.

Tax Reform and Entrepreneurial Activity

69

14. A natural explanation for these differences in evasion rates among industries is the greater potential for cross-checking of claimed items when transactions occur with another ﬁrm than when they occur with an individual. The ﬁrm with a potential tax deduction arising from a transaction has an incentive to report it, giving the IRS an independent source of information about this source of taxable income of the seller. There is no such independent source of information for sales to individual customers. 15. Speciﬁcally, we deﬁned noncorporate losses to be those reported on Schedule C or on Schedule E as either partnership losses or subchapter S corporate losses. 16. This is not entirely true. A riskless ﬁrm can still have losses in initial years and proﬁts later. Such ﬁrms would be picked up by our indicator of entrepreneurial risk-taking if the initial losses are large enough. 17. For this current paper, we brought in comparable tax return data for another eight years (through 2001). We reestimated the responsiveness of behavior to tax incentives in this expanded sample and found similar results. However, given the complexity (nonlinear estimation on a huge sample), we did not attempt to reestimate the parameters (such as the degree of risk aversion) involved in the measurement of tax incentives. We therefore use our original estimates in the simulations reported below. 18. In order to use a consistent measure of potential earnings for each individual, regardless of whether the individual was self-employed, we forecast the labor income an individual would earn as an employee based on other information reported on the tax return—principally demographic information, income from ﬁnancial assets, and certain itemized deductions (if present). 19. Speciﬁcally, we assumed a logit speciﬁcation. If Pit denotes the fraction of quantile i in year t that has large losses, then a logit speciﬁcation implies that lnðPit =ð1 Pit ÞÞ ¼ f ðXit ; bÞ þ eit , where Xit includes tax incentives and various control variables, and b is a vector of parameters to be estimated. Given the complicated ways in which risk affects tax incentives, there are unavoidable nonlinearities in the function f ð:Þ. 20. Both tax incentives and the nontax proclivity to engage in entrepreneurial risk-taking depend on an individual’s earnings level. Yet we observe only a couple’s combined wage and salary income and combined self-employment income. 21. The latest year in which a random sample of tax returns is now available is 2001. We weight the tax incentives faced by any quantile by their potential earnings (deﬁned as described in note 19). 22. In particular, the maximum tax rate on labor income fell from 75.3 percent to 50 percent, while the corporate tax rate fell from 24.2 percent to 22 percent. The maximum capital-gains tax rate also increased, raising the tax rate on proﬁts while not affecting the tax rate on losses. 23. Since entrepreneurial activity is heavily concentrated among the highest earning individuals, we focus more heavily on the incentives faced at the upper end of the earnings distribution. The six series graphed represent the following percentile ranges of the potential earnings distribution: (0 percent–70 percent), (70 percent–80 percent), (80 percent–90 percent), (90 percent–95 percent), (95 percent–99 percent), and (99 percent– 100 percent). See note 19 for information about the deﬁnition of potential earnings. 24. The reported fraction of tax payers with large entrepreneurial losses certainly understates the fraction of taxpayers engaged in entrepreneurial risk-taking since the ﬁgures

70

Cullen and Gordon

include only those with large ex post losses. Given our estimate for the riskiness of entrepreneurial activity, roughly 30 percent of businesses should have tax losses, while fewer will have losses exceeding 10 percent of reported wage and salary income. The fraction of the population taking large entrepreneurial risks therefore exceeds 1/.3 times the fraction with large ex post losses. 25. One item on the reform agenda we did not focus on is the alternative minimum tax (AMT). As of 2001, very few people faced the AMT, so it would be of little importance using data from 2001 as a benchmark. 26. Section 1244 is one of the detailed provisions listed in the overall legislation dealing with capital gains taxation, sections 1201–1291, so it would be repealed if this entire section of the tax code were eliminated. 27. Speciﬁcally, we calculated the revenue generated by any given personal income tax rate for all taxpayers, married as well as single, and chose the rate generating the same revenue as under the current rate structure. 28. In doing so, we left in place the EITC, under the presumption that this poverty alleviation program would be viewed as a separate issue from the overall rate structure. 29. Such a carryback can be of particular value for a small start-up since these losses can be carried back to a tax year in which the individual was still a full-time employee and would be in a much higher tax bracket. 30. Starting from the 2005 law, we found in another simulation, in contrast, that the EITC encourages entrepreneurial risk-taking among those in the upper tax brackets, for whom the new concave kink in ﬁgure 2.3 dominates, but discourages risk-taking overall because of more than offsetting effects among those in lower tax brackets, where the effects of the two convex kinks in ﬁgure 2.3 dominate. 31. For example, in our data for 2001, 80 percent of single ﬁlers are either in a 0 percent or a 15 percent tax bracket, whereas this is true for only 60 percent of married couples. In contrast, 2.9 percent of single ﬁlers are in the top three tax brackets in that year, whereas 8.3 percent of married couples are in these brackets. 32. Under a Hall-Rabushka tax, taxes on ﬁnancial income from capital would be eliminated, corporate as well as personal income would face this one ﬂat tax rate, while many of the complicated special provisions in the current law would be repealed.

References Andreoni, James, Brian Erard, and Jonathan Feinstein (1998). ‘‘Tax Compliance,’’ Journal of Economic Literature, 36:818–860. Barsky, Robert B., Miles S. Kimball, F. Thomas Juster, and Matthew D. Shaprio (1997). ‘‘Preference Parameters and Behavioral Heterogeneity: An Experimental Approach in the Health and Retirement Survey,’’ Quarterly Journal of Economics, 112:537–579. Bound, John, Clint Cummins, Zvi Griliches, Bronwyn H. Hall, and Adam B. Jaffe (1984). ‘‘Who Does R&D and Who Patents?’’ in Zvi Griliches (ed.), R&D, Patents and Productivity, Chicago, Ill.: University of Chicago Press, pp. 21–54. Cullen, Julie Berry, and Roger H. Gordon (2005). ‘‘Taxes and Entrepreneurial Activity: Theory and Evidence for the U.S.,’’ mimeo.

Tax Reform and Entrepreneurial Activity

71

Diamond, Peter, and Jonathan Gruber (1997). ‘‘Social Security and Retirement in the U.S.,’’ NBER working paper no. 6097. Feldstein, Martin, Louis Dicks-Mireaux, and James Poterba (1983). ‘‘The Effective Tax Rate and the Pretax Rate of Return,’’ Journal of Public Economics, 21:129–158. Gentry, William M., and R. Glenn Hubbard (2004). ‘‘‘Success Taxes,’ Entrepreneurial Entry, and Innovation,’’ NBER working paper no. 10551. Hall, Bronwyn, and John Van-Reenen (2000). ‘‘How Effective Are Fiscal Incentives for R&D? A Review of the Evidence,’’ Research Policy, 29:449–469. Hall, Robert E., and Alvin Rabushka (1995). The Flat Tax, 2d ed. Stanford, Calif.: Hoover Institution Press. Schumpeter, Joseph (1976). Capitalism, Socialism, and Democracy, 5th ed. London: Allen and Unwin. Sommerfeld, Ray M., and Sally M. Jones (1991). Federal Taxes and Management Decisions, 2d ed. Homewood, Ill.: Richard D. Irwin, Inc.

3 Behavioral Responses to Taxes: Lessons from the EITC and Labor Supply Nada Eissa, Georgetown University and NBER Hilary W. Hoynes, University of California and NBER

Executive Summary Twenty-two million families currently receive a total of $34 billion in beneﬁts from the earned income tax credit (EITC). In fact, the EITC is the largest cash-transfer program for lower-income families at the federal level. An unusual feature of the credit is its explicit goal to use the tax system to encourage and support those who choose to work. A large body of work has evaluated the labor supply effects of the EITC and has generated several important ﬁndings regarding the behavioral response to taxes. Perhaps the main lesson learned from the evidence is the conﬁrmation that real responses to taxes are important; labor supply does respond to the EITC. The second major lesson is related to the nature of the labor supply response. A consistent ﬁnding is that labor supply responses are concentrated along the extensive (entry) margin, rather than the intensive (hours worked) margin. This distinction has important implications for the design of tax-transfer programs and for the welfare evaluation of tax reforms. 1.

Introduction and Overview

The past two decades in the United States have witnessed an increasing reliance on the tax system as a means of providing cash assistance to lower-income families with children. A series of tax acts starting with the 1986 Tax Reform Act have increased assistance to the working poor through expansions of the earned income tax credit (EITC), making it the largest federal cash-transfer program for lower-income families with children. Over 20 million families are projected to have beneﬁted from the tax credit in 2003, at a total cost to the federal

74

Eissa and Hoynes

government of more than $34 billion.1 These transfers have been especially successful at reducing poverty in the United States. In fact, census data suggest that the federal credit now lifts more children out of poverty than any other government program. ‘‘Some 4.9 million people, including 2.7 million children, were removed from poverty in 2002 as a result of the federal EITC’’ (Llobrera and Zahradnik 2004). These estimates are consistent with the fact that the 1993 credit expansion was designed such that a full-time worker earning the minimum wage would not live in poverty. This shift in the structure of public assistance is the outcome of a long-standing criticism that traditional welfare programs generate adverse incentives for work and family. Advocates of the earned income tax credit have argued that the credit helps ‘‘promote both the values of family and work.’’ Indeed, economic theory suggests that the EITC should be especially successful at promoting entry into the labor force among eligible, single taxpayers. The recent expansions have offered researchers an excellent opportunity to evaluate the impacts of the EITC on behavior. Hotz and Scholz’s (2003) review of the literature on the EITC shows that researchers have taken full advantage of this opportunity, examining outcomes such as work behavior, marriage, fertility, consumption expenditures, and overall family well-being. Here, we concentrate on the impacts of the EITC on labor supply and in particular on the lessons learned about the labor supply responses to tax-transfer programs. We argue that the evaluation of the EITC has been especially useful for understanding the size and the composition of labor supply responses to tax incentives and has informed the literature on the welfare evaluation of tax reforms as well as the optimal design of tax-transfer schemes. In this review, we ﬁrst provide a brief history of the EITC, along with a description of the current policies and recent tax expansions. In Section 3, we provide a descriptive analysis of the EITC population. In particular, we use IRS data to examine the characteristics of EITC recipients, including number of children, ﬁling status, and income. In Section 4, we use economic theory to discuss the expected impact of the EITC on labor supply. In short, among eligible unmarried women with children, the EITC is expected to increase employment but reduce hours of work for those already in the labor market. To target beneﬁts to lower-income families, however, the EITC is based on family income, leading to a very different set of incentives for married taxpayers. Among secondary earners in married couples, the EITC is

Behavioral Responses to Taxes

75

expected to reduce labor force participation and hours worked for those already in the labor force. Section 5 summarizes what is known about the impacts of the EITC on labor supply. The nature of the expected labor supply effects of the EITC leads us to structure our summary in the following manner. First, we examine the impacts on single and married taxpayers separately, focusing exclusively on women.2 Second, we examine separately the two different margins of labor supply responses—the extensive (employment) and intensive (hours worked by working individuals) margins. The overwhelming ﬁnding of the empirical literature is that EITC has been especially successful at encouraging the employment of single parents, especially mothers. There is little evidence, however, that the EITC has reduced the hours worked by those already in the labor force. The empirical literature on married women is somewhat smaller but again consistent in its ﬁndings. The studies show that the EITC leads to modest reductions in the employment and hours worked of married women. In Section 6, we discuss the possible explanations for the empirical ﬁnding that the EITC affects the extensive but not the intensive margin. We also discuss the general equilibrium implications of the EITC—in particular whether the EITC affects gross wages. Finally, we discuss the implications of this work for the optimal design of the EITC. Recent work has emphasized the role of the extensive margin in the design of the tax-transfer schemes (Saez 2002) and the welfare evaluation of tax reform (Eissa, Kleven, and Kreiner 2004). We ﬁrst review this work and then discuss whether there are potential efﬁciency gains associated with modifying the EITC. Section 7 presents our conclusions. 2.

Operation and Brief History of the EITC

The Earned Income Tax Credit began in 1975 as a modest program aimed at offsetting the Social Security payroll tax for low-income families with children. It was the outcome of a vigorous policy debate surrounding the efﬁcacy of a negative income tax (NIT) as a means of reducing poverty. The concern was that the NIT—which guarantees a minimum standard of living to everyone—would discourage labor market activity as it is gradually phased out. Ultimately the EITC was born out of a desire to reward work. The EITC is refundable, so that a taxpayer with no federal tax liability, for example, would receive a tax refund from the government for

[0.000; 0.500]

[0.000; 0.500]

1984

1985

1986

[0.150; 0.396]

1993

[0.150; 0.396]

[0.150; 0.396]

[0.150; 0.396]

1994

1995

1996

OBRA1993d

[0.150; 0.310]

[0.150; 0.330]

1990

[0.150; 0.310]

[0.150; 0.330]

1989

1992

[0.150; 0.330]

1988

OBRA90 1991

[0.110; 0.390]

1987

c

[0.000; 0.500]

Year

TRA86

[Lowest Marginal Tax Rate; Highest Marginal Tax Rate]

$2,550; $5,900

$2,500; $5,750

$2,450; $5,600

$2,350; $5,450

$2,300; $5,250

$2,150; $5,000

$2,050; $4,750

$2,000; $4,550

$1,950; $4,400

$1,900; $2,540

$1,080; $0

$1,040; $0

$1,000; $0

Personal Exemption, Standard Deductiona,b

Federal Income Tax Parameters

Table 3.1 Federal Income Tax and EITC Parameters, 1984–2005

0.340; 0.400

0.340; 0.360

0.263; 0.300

0.185; 0.195

0.176; 0.184

0.167; 0.173

0.140

0.140

0.140

0.140

0.110

0.110

0.100

Phase-In Rate

$2,152; $3,556

$2,094; $3,110

$2,038; $2,528

$1,434; $1,511

$1,324; $1,384

$1,192; $1,235

$ 953

$ 910

$ 874

$ 851

$ 550

$ 550

$ 500

Maximum Credit

0.160; 0.211

0.160; 0.202

0.160; 0.177

0.132; 0.139

0.126; 0.131

0.119; 0.124

0.100

0.100

0.100

0.100

0.122

0.122

0.125

Phase-Out Rate

$25,078; $28,495

$24,396; $26,673

$23,755; $25,296

$23,050

$22,370

$21,250

$20,264

$19,340

$18,576

$15,432

$11,000

$11,000

$10,000

Maximum Earnings

EITC Parameters (Family with One Child; Family Two or More Children)

76 Eissa and Hoynes

[0.150; 0.396]

[0.150; 0.396]

Year

1997

2000

[0.100; 0.35]

[0.100; 0.35]

2004

2005

$3,200; $7,300

$3,100; $7,150

$3,050; $7,000

$2,900; $6,650

$2,650; $6,050

Personal Exemption, Standard Deductiona,b

0.340; 0.400

0.340; 0.400

0.340; 0.400

0.340; 0.400

0.340; 0.400

Phase-In Rate

$2,662; $4,400

$2,604; $4,300

$2,547; $4,204

$2,353; $3,888

$2,210; $3,656

Maximum Credit

0.160; 0.211

0.160; 0.211

0.160; 0.211

0.160; 0.211

0.160; 0.211

Phase-Out Rate

$31,030; $35,263

$30,338; $34,458

$29,201; $33,178

$27,413; $31,152

$25,750; $29,290

Maximum Earnings

EITC Parameters (Family with One Child; Family Two or More Children)

Source: U.S. House of Representatives [The Green Book] (various years) and authors’ calculations from Internal Revenue Service (IRS) forms and publications. a The standard deductions are given for head of household tax return. b In 1984–1986, there were no standard deductions because of the zero bracket. The ﬁfteen brackets include the zero bracket. c Basic EITC only. Does not include supplemental young child credit or health insurance credit. d Introduced a small beneﬁt for taxpayers with no qualifying children, phased-in at 0.0765 up to a maximum credit of $306. e The ﬂat and phase-out regions are expanded by $2,000 ($1,000) in 2005 (2002–2004) for married couples ﬁling jointly.

[0.100; 0.386]

2002

EGTRRA2001e

[Lowest Marginal Tax Rate; Highest Marginal Tax Rate]

Federal Income Tax Parameters

Behavioral Responses to Taxes 77

78

Eissa and Hoynes

the full amount of the credit. Taxpayers may also receive the credit throughout the year with their paychecks; but in 2000, less than 5 percent of EITC recipients availed themselves of this early payment option (Friedman 2000). A taxpayer’s eligibility for the earned income tax credit depends on the taxpayer’s earned income (or in some cases, adjusted gross income) and the number of qualifying children who meet certain age, relationship, and residency tests. First, the taxpayer must have positive earned income, deﬁned as wage and salary income, business self-employment income, and farm self-employment income. Also, the taxpayer must have adjusted gross income and earned income below a speciﬁed amount (in 2004, the maximum allowable income for a taxpayer with two or more children was $34,458). Second, a taxpayer must have a qualifying child, who must be under age 19 (or age 24 if a full-time student) or permanently disabled and residing with the taxpayer for more than half the year.3 Until 1991, the rules for EITC eligibility were more complicated and depended on the taxpayer’s ﬁling status.4 The amount of the credit to which a taxpayer is entitled depends on the taxpayer’s earned income; adjusted gross income; and, since 1991, the number of EITC-eligible children in the household. There are three regions in the credit schedule. The initial phase-in region transfers an amount equal to the subsidy rate times their earnings. In tax year 2004, the subsidy rate of the EITC was 34 percent for taxpayers with one child and 40 percent for taxpayers with two or more children. In the ﬂat region, the family receives the maximum credit ($2,604 and $4,300, respectively), while in the phase-out region, the credit is phased out at the phase-out rate (16 and 21 percent). Table 3.1 summarizes the parameters of the EITC over the history of the program. The program grew slowly from its introduction in 1975 until 1986, and in fact shrank in real terms due to inﬂation. The 1987 expansion of the EITC, passed as part of the Tax Reform Act of 1986 (TRA86), increased the generosity of the credit for the lowest-income workers and extended its beneﬁts beyond the poorest (see Eissa and Liebman 1996).5 By 1988, taxpayers with incomes between $11,000 and $18,576 became eligible for the credit and faced its phase-out marginal tax rate for the ﬁrst time. The 1987 EITC expansion interacted with other tax changes implemented after TRA86—including greater personal exemption and standard deduction—to help eliminate taxes on poor families. The largest single expansion over this period was contained in the Omnibus Reconciliation Act of 1993 (OBRA93). The

Behavioral Responses to Taxes

79

Figure 3.1 Earned Income Tax Credit, 1984–2003 (Current year dollars). (a) Schedule for Family with One Child. (b) Schedule for Family with Two or More Children. Notes: Authors’ tabulations using tax parameters in table 3.1.

80

Eissa and Hoynes

Figure 3.2 Real Spending on the EITC (Billions of 2003 Dollars). Source: U.S. House of Representatives (2004), Table 13–14.

1993 expansion of the EITC, phased in between 1994 and 1996, led to an increase in the subsidy rate, from 19.5 percent to 40 percent (18.5 to 34 percent), and an increase in the maximum credit, from $1,511 to $3,556 ($1,434 to $2,152), for taxpayers with two or more children (taxpayers with one child). This expansion was substantially larger for those with two or more children. The phase-out rate was also raised, from 14 percent to 21 percent (13 to 16 percent), for taxpayers with two or more children (taxpayers with one child). Overall, the range of the phase-out was expanded dramatically so that, by 1996, a ﬁling unit with two children would still be eligible with income levels of almost $30,000. Figure 3.1 illustrates the nominal value of the credit from 1984 to 2003 and highlights the dramatic expansion of the credit over time, as well as its effects on the families of different sizes. These expansions have led to a dramatic and recent increase in the total cost of the EITC. Figure 3.2 shows the total cost (in real 2003 dollars) of the EITC from 1974–2003. The ﬁgure clearly shows the rising expenditures associated with the 1986, 1990, and 1993 tax acts. Between 1990 and 1996, the program more than doubled in real terms. The popularity and perceived success of the EITC can also be seen in the increasing number of states that offer state credits. As of 2004, a total of eighteen states have introduced state EITCs that supplement

Behavioral Responses to Taxes

81

Table 3.2 State Earned Income Tax Credits, 2004 State (year started)

Percentage of Federal Credit (2004)

Refundable credits Colorado (1999)

10% (suspended until 2006)

District of Columbia (2000)

25%

Illinois (2000)

5%

Indiana (1999)

6% (starting in 2003)

Kansas (1998)

15%

Marylanda (1987)

20%

Massachusetts (1997)

15%

Minnesota (1991) New Jersey (2000)

Varies with earnings; averages 33% 20% (for families with earnings below $20,000)

New York (1994)

30%

Oklahoma (2002)

5%

Vermont (1988)

32%

Wisconsin (1989)

4% (one child); 14% (two children); 43% (three children)

Non-refundable credits Iowa (1990)

6.5%

Maineb (2000)

4.92%

Oregon (1997) Rhode Islandc (1975)

5% 25%

Virginia (2004)

20% (effective 2006)

Source: Llobrera and Zahradnik (2004) and www.stateEITC.info (State EITC Online Info Center). a Maryland also offers a non-refundable EITC set at 50% of the credit. Taxpayers may claim either the refundable credit or the non-refundable credit, but not both. b Maine’s non-refundable EITC was reduced from 5% to 4.92% effective in tax years 2003, 2004, and 2005. The credit will return to 5% in 2006. c Rhode Island made a very small portion of its EITC refundable effective beginning in tax year 2003.

the federal credit.6 Almost all states structure their credits as a share of the federal credit, varying between 5 percent (Illinois) to more than 40 percent (Minnesota and Wisconsin), and almost all make the credit refundable like the federal credit. Table 3.2 presents the characteristics of state EITCs along with the date of their introduction. It is notable that several states either introduced or expanded credits since the economic downturn starting in 2000, including Virginia, Indiana, Illinois, and Kansas. This continuing growth of state EITCs has been attributed to a recognition among state policymakers that ‘‘state EITC’s can help working families stay aﬂoat’’ in difﬁcult labor markets (Llobrera and

82

Eissa and Hoynes

Figure 3.3 EITC Beneﬁt for Selected Hourly Wage Levels, by Annual Hours Worked (2004 Tax Year). (a) Schedule for Family with One Child. (b) Schedule for Family with Two or More Children. Source: Eissa, Rothstein, and Nichols (2005). Figure shows the value of the EITC by wage level and annual hours worked assuming that there is one worker in the family and no other sources of income.

Behavioral Responses to Taxes

83

Zahradnik 2004). The extent to which this is true will depend on the EITC’s incidence, however. An important observation about the EITC is that the credit is based on annual earnings. As a result, its relationship to a worker’s hourly wage is indirect. Figure 3.3 shows exposure to the credit at different annual hours worked and hourly wages. Among full-time workers, those earning the minimum wage receive the maximum credit, while those earning $15 per hour would just miss eligibility. Moreover, the greater generosity of the program for two-child families implies that slightly higher-wage families would be eligible for the credit. This indirect relationship has implications for the interpretation of the empirical evidence. We return to this issue in Section 6. 3.

The EITC Population

IRS Statistics of Income (SOI) reports can be used to provide a proﬁle of the EITC recipient population. Here, we present tabulations of the 2001 Statistics of Income Public Use Tax File, which is a nationally representative sample of all individual tax returns ﬁled in tax year 2001. The ﬁle consists of a total of 143,221 records (Internal Revenue Service 2004). We use the tax ﬁle data to present tabulations of EITC recipients and costs by ﬁling status, number of children, and earned income. In 2001, there were a total of 20 million EITC recipients costing a total of $33.3 billion. In table 3.3, we present tabulations of these recipients and claims according to number of qualifying children and ﬁling status. The results show that the number of EITC returns is about evenly split between those with one child versus two or more children. Owing to the more generous credit for larger families, however, ﬁlers with two or more children receive 61 percent of total EITC dollars (refunds and tax expenditures), while those with one child receive 37 percent of expenditures (the remaining 2 percent funds the small credit for childless ﬁlers). The ﬁrst panel of table 3.3 also shows that single and head of household ﬁlers represent fully three-quarters of EITC returns and expenditures, while married couples ﬁling jointly make up the remaining quarter. This disproportionate share of unmarried ﬁlers among the EITC population reﬂects the higher eligibility rates—due to lower earnings and income—of single women with children. The third panel of table 3.3 presents the average EITC credit amount by ﬁling status and number of children. The average credit amount rises sharply with number of children—$2,459 for those with two or more

84

Eissa and Hoynes

Table 3.3 Percentage Distribution of EITC Returns and Claims by Filing Status and Number of Qualifying Children, 2001 Number of Qualitfying Children None

One

Two or More

Percentage distribution of EITC returns Head of household ﬁlersa

16%

33%

28%

Married ﬁling joint ﬁlers

2%

7%

14%

18%

40%

42%

Head of household ﬁlersa

2%

32%

43%

Married ﬁling joint ﬁlers

0%

6%

17%

All ﬁlers

2%

37%

61%

Head of household ﬁlersa

$205

$1,615

$2,612

Married ﬁling joint ﬁlers

$205

$1,403

$2,144

All ﬁlers

$205

$1,578

$2,459

All ﬁlers Percentage distribution of total EITC claims

Average EITC claim

Source: Tabulations from Tax Policy Center, Urban Institute, using 2001 Statistics of Income Public Use Tax File. a Includes individuals who ﬁle as single.

children compared to $1,578 for those with one child—reﬂecting the more generous schedule for families with two or more children. In addition, average EITC claims for single parents exceed average payments to married couples ﬁling jointly. As above, this reﬂects the higher incomes of married couples. Table 3.4 presents the percentage distribution of returns and expenditures for tax ﬁlers with children by EITC bracket, using 2001 SOI data. This distribution is the critical determinant of the net labor supply effect of the EITC and its expansion (see discussion in the next section). The table shows that about one-quarter of EITC head of household returns and expenditures go to ﬁlers in the subsidy (phasein) region of the credit. Far fewer married couples (about 15 percent) have income that place them in that region of the credit, however. It is useful to separate single and married ﬁlers because the design of the EITC can generate very different labor supply incentives for these two populations. In general, most EITC tax returns are located in the phase-out region of the credit. This is especially true for married couples where about half of head of household and more than two-thirds of married couples are in the phase-out region. EITC expenditures

Behavioral Responses to Taxes

85

Table 3.4 Distribution of EITC Recipients and Claims by EITC Range, 2001 EITC Range Phase-in

Flat

Phase-out

All

Percentage distribution of EITC recipients by EITC range Head of household ﬁlers One child Two or more children

23% 28%

25% 15%

52% 57%

100% 100%

One child

14%

22%

64%

100%

Two or more children

15%

10%

75%

100%

Married ﬁling joint ﬁlers

Percentage distribution of EITC claims by EITC range Head of household ﬁlers One child

21%

37%

42%

100%

Two or more children

27%

23%

50%

100%

One child

14%

37%

49%

100%

Two or more children

17%

19%

64%

100%

Married ﬁling joint ﬁlers

Notes: Tabulations from Tax Policy Center, Urban Institute, using 2001 Statistics of Income Public Use Tax File.

show broadly the same pattern, though the distribution is slightly more evenly spread across the regions. About one-quarter of EITC beneﬁts for head of household ﬁlers go to those in the subsidy region, while about 40–50 percent of beneﬁts go to those in the phase-out region. Among married couples, the data show only 15 percent of dollars beneﬁt those with the lowest income, while more than half of the dollars go those in the phase-out region. The SOI reports include a very limited set of observable characteristics of EITC recipients. We expand our discussion by presenting a picture of the EITC population using survey data, which represents the data used in essentially all evaluations of the labor supply effects of the credit. In particular, we provide demographic characteristics of single mothers—who represent the largest group of EITC recipients. Eissa, Kleven, and Kreiner (EKK) (2004) provide summary statistics for a sample of single mothers age 18–49 from the 1986, 1991, 1994, and 2001 March Current Population Surveys (CPS), covering income and earnings data for the prior calendar year—1985, 1990, 1993, and 2000. The results (presented in table 3.5) show that the typical single mother during this period is in her early thirties, with a high school

1142 (961)

$7,922 (9,210)

$2,756 (4,605)

$6.53 (4.43)

Number of children Labor force participation

Annual hours worked

Wage and salary income

Non-labor income

Gross hourly wage

OBRA1990 0.017 (0.051) 0.053 (0.016)

0.145 (0.105)

TRA1986

0.059 (0.095) 0.076 (0.029)

4,498

Effect of tax act

Marginal tax rate Average tax rate

Observations

5,011

0.033 (0.142) 0.128 (0.076)

OBRA1993

0.042 (0.010)

0.185 (0.212)

$8.68 (7.00)

$3,553 (5,912)

$10,572 (12,740)

1148 (970)

1.78 (1.0) 0.697 (0.459)

0.031 (0.172)

0.338 (0.473)

0.793 (0.406)

32.98 (7.79)

1992

4,072

0.034 (0.071) 0.023 (0.014)

EGGTRA2001

0.041 (0.180)

0.231 (0.288)

$11.24 (10.68)

$3,961 (7,740)

$16,430 (19,526)

1426 (903)

1.74 (0.93) 0.830 (0.376)

0.037 (0.188)

0.315 (0.464)

0.787 (0.409)

33.44 (8.15)

2000

Source: Eissa, Kleven, and Kreiner (2004). Calculations based on March CPS and include single mothers age 18–49. Non-labor income is calculated as the difference between total income and earnings, and therefore includes income for various sources such as welfare assistance, capital income, Social Security income and worker’s compensation. The wage is deﬁned for workers only and is topcoded at $200 in 2000. Tax rates are calculated using NBER’s TAXSIM model and include all federal income taxes and the payroll tax (assuming individuals bear the entire tax). Calculations for ‘‘Effect of Tax Act’’ use the sample for a ﬁxed year and include tax rates under the old law and the new law—the differences are reported. Standard errors are in parentheses. All monetary amounts are in nominal dollars.

4,850

0.088 (0.089)

0.259 (0.166)

Average tax rate

0.216 (0.175)

$7.85 (5.49)

$3,277 (5,742)

$10,390 (11,642)

1219 (969)

1.77 (0.99) 0.72 (0.447)

0.023 (0.151)

0.345 (0.475)

0.773 (0.418)

32.91 (7.77)

1989

Marginal tax rate

Tax rates

0.021 (0.142)

1.94 (1.10) 0.707 (0.455)

Non-white

0.784 (0.411)

0.321 (0.467)

Black

32.04 (7.75)

Education—fraction 12 or more years of schooling

Age

Demographic characteristics

1985

Year

Table 3.5 Demographic Characteristics and Tax Rates for Single Mothers, 1985–2000

86 Eissa and Hoynes

Behavioral Responses to Taxes

87

diploma, and fewer than two children. In 2000, 83 percent of single mothers between the ages of 18 and 49 in the CPS are working, on average, 1,426 hours per year at $11.24 per hour. The data also show that the demographic characteristics of the typical eligible single mother have changed little over the period that the EITC has been expanded. Between 1985 and 2000, the typical single mother is aging somewhat over the 15-year period—by just under a year and a half—and has fewer children and more education. The most striking change for single mothers during this period is clearly the increase in labor supply, observed along both the margin of participation and hours worked. 4.

Expected Effects of the EITC

In this section, we examine the impact of the credit on labor supply and discuss the expected impacts on the extensive (employment) and intensive (hours worked by working individuals) margins.7 The participation and hours incentives are modeled assuming that taxpayers face no labor market restrictions in their choice of hours, given the (ﬁxed) offered wage.8 For simplicity, we consider ﬁrst the case of an unmarried taxpayer and then extend the analysis to examine labor supply incentives for married couples. 4.1 Unmarried Taxpayers Because the EITC is available only to taxpayers with earned income, standard labor supply theory predicts that the EITC will encourage labor force participation among single parents. The stylized budget constraint, in ﬁgure 3.4, shows this impact graphically. The ﬁgure plots hours worked (horizontal axis) against income (vertical axis), ignoring for simplicity all other features of the taxtransfer system that affect single parents.9 In the absence of the EITC, the taxpayer earns a gross wage w for each hour worked—hence, the no-EITC budget constraint is given by segment AD, with slope w. The EITC alters the budget constraint to ABCD. In the phase-in region (AB), the EITC acts as a pure wage subsidy and increases the net wage from w to wð1 þ ts Þ where ts is the subsidy rate (40 percent for taxpayers with two or more children and 34 percent for taxpayers with one child in 2004). In the ﬂat region of the credit (BC), the taxpayer’s budget constraint is shifted out an amount equal to the maximum credit ($2,604 for taxpayers with one child and $4,300 for taxpayers with two or more children in 2004) and her gross (and net of tax)

88

Eissa and Hoynes

Figure 3.4 Stylized Budget Constraint for the EITC

hourly wage is w. Each dollar earned in the phaseout region of the EITC (CD) reduces the credit by a phaseout rate of tp (about 21 percent). The net wage earned by the taxpayer in this region, wð1 tp Þ, may be reduced even further once the taxpayer starts paying federal tax. This ﬁgure is drawn to reﬂect the fact that the phase-out region of the credit is much larger than the phase-in region. For example, for a single woman with two children in 2004, the phase-in region spans about $10,000 (from $0 to $10,750) while the phase-out region spans about $20,000 (from $14,040 to $34,458). The ﬁgure shows that the well-being of a taxpayer who is not working is not affected by the EITC. Any taxpayer who preferred working before will still prefer working, and some taxpayers may ﬁnd that the additional after-tax income from the EITC makes it worth entering the labor force. Therefore, the impact of introducing or expanding the EITC on the labor force participation of unmarried taxpayers is unambiguously positive. The impact of the EITC on the hours worked by a single working taxpayer, however, depends on which region of the credit the taxpayer is in before the credit is expanded or introduced. If she is in the phase-in region, the EITC leads to an ambiguous impact on hours worked due

Behavioral Responses to Taxes

89

Table 3.6 Average Tax Rates for Hypothetical Single Women with Children (2004 Tax Law): Taxes Calculated with and Without the EITC Average Tax Rates with the EITC

Average Tax Rates Without the EITC

Entry to PT Work

Entry to PT Work

Entry to FT Work

PT to FT Work

Entry to FT Work

PT to FT Work

A. Single taxpayer with one child (by hourly wage level) $5.15/hour

19%

9%

1%

15%

15%

15%

$7.50/hour

18%

3%

12%

15%

12%

12%

$10/hour $12/hour

10% 7%

7% 13%

23% 33%

15% 14%

14% 16%

14% 16%

$15/hour

3%

19%

41%

12%

19%

19%

B. Single taxpayer with two or more children (by hourly wage level) $5.15/hour

25%

25%

25%

15%

15%

15%

$7.50/hour

25%

13%

2%

15%

12%

12%

$10/hour

25%

3%

18%

15%

10%

10%

$12/hour

21%

3%

26%

14%

11%

11%

$15/hour

13%

12%

38%

12%

14%

14%

Notes: Calculated using TAXSIM. Numbers represent the average tax rate associated with increasing labor supply (either from no work to work or from part-time (PT) to fulltime (FT) work). Taxes include all federal income taxes plus the payroll tax (assuming the individual bears the entire 15.3% rate). Part-time work is assumed to be 20 hours per week and 52 weeks per year, and full-time work is assumed to be 40 hours per week and 52 weeks per year.

to the negative income effect and positive substitution effect. In the ﬂat region, however, the EITC produces a negative income effect, leading to an unambiguous reduction in hours worked. In the phase-out region, the EITC produces a negative income and negative substitution effect, leading again to an unambiguous reduction in hours worked. Moreover, the phase-out of the credit alters the budget set in such a way that some taxpayers with incomes beyond the phase-out region may choose to reduce their hours of work and take advantage of the credit. Recall from table 3.4 above that about three-quarters of single EITC recipients have earnings in the ﬂat and phase-out region of the credit—thus, the expectation is that the EITC will reduce the number of hours worked by most eligible single taxpayers already in the labor force. To illustrate how the EITC affects the incentives faced by single taxpayers, table 3.6 presents average tax rates faced by hypothetical single

90

Eissa and Hoynes

taxpayers earning different wage rates, with different family sizes. These calculations are based on the National Bureau of Economic Research (NBER) TAXSIM model and cover all federal income taxes and payroll taxes under 2004 law.10 We calculate average tax rates for three cases: (1) moving from no work to part-time work (20 hours per week, 52 weeks per year); (2) moving from no work to full-time work (40 hours per week, 52 weeks per year); and (3) moving from parttime to full-time work. The average tax rates are calculated with the EITC (columns 1–3) and without the EITC (columns 4–6). The tax rates are calculated for women with one child and women with two or more children to reﬂect the different EITC schedules and for women over a range of wage rates from the federal minimum ($5.15) to $15 per hour. These calculations show that the EITC reduces average tax rates of entering (part- and full-time) work across all family sizes and all wage levels, thereby illustrating the prediction that participation should unambiguously increase for single parents. The impact of the EITC on these tax rates for entering work can be sizable, even for higher-wage women. For example, in tax year 2004, a woman earning $10 per hour with two children faces a 25 percent average tax rate (subsidy) for entering work at part-time levels and a 3 percent average tax rate (subsidy) for entering work at full-time levels. In the absence of an EITC, the same woman would face average tax rates of 15 percent and 10 percent, respectively. The results show that the impacts of the EITC are greatest for lower-wage women (with more earnings in the phasein region), for part-time entrants, and for women with two or more children. The table also illustrates how the EITC penalizes increasing work for those already in the work force. If a woman with two children, earning $10 per hour, were to increase work from part- to full-time, however, she faces an average tax rate of 18 percent with the EITC (due to the phase-out rate) compared to 10 percent without the EITC. This work disincentive on the intensive (hours) margin exists for all but the lowest-wage women (who even at full-time work are not yet in the phase-out region of the credit). In addition to these tax rates for hypothetical taxpayers, Eissa, Kleven, and Kreiner (2004) use their CPS sample of single mothers for 1985, 1990, 1993, and 2000 and calculate average and marginal tax rates (using TAXSIM) at observed earnings. The results are presented in the bottom of table 3.5 (presented earlier)—and their results show that the average tax rates declined consistently over the period, from

Behavioral Responses to Taxes

91

14.5 percent in 1985 to a negative 4.1 percent in 2000. It is notable that the time series change in the marginal tax rate is more moderate and less systematic than that of the average tax rate. Note that these tax rates will vary over time due to changes in tax policy (the EITC and all other income tax changes), and also with changes in the earnings and family size of their sample of single women. EKK also isolate the impact of the federal tax reforms (again EITC and non-EITC tax changes) on tax rates by imputing a post-reform tax rate at given pre-reform characteristics. These results (labeled ‘‘Effect of tax act’’ in table 3.5) show that all tax acts since 1986 have reduced both the average and marginal tax rates faced by single taxpayers and have generally reduced tax liability (average tax rates) more signiﬁcantly than marginal tax rates. The largest reduction in tax rates is due to OBRA93. 4.2 Married Taxpayers In contrast to the predictions for single women, the EITC is expected to reduce the participation and hours worked of most eligible married women. This occurs also because the credit is based on family earnings and income. Suppose family labor supply follows a sequential model, with the husband as the primary earner and the wife as the secondary earner. In this model, the effect of the credit on the labor supply of primary earners is the same as that of single taxpayers—incentives to participate in the labor market strengthen unambiguously, while hours of work incentives are ambiguous, though they are likely negative given the distribution of taxpayer income along the EITC schedule. The impact of the EITC on secondary earners is more complicated because it depends on the earnings of the primary earner. Assume for simplicity that the husband is the primary earner and further that his earnings place the family in the phase-out range.11 The family receives the credit if the wife remains out of the labor force, but the credit amount will decrease with each dollar earned if she enters the labor market. In the phase-out range, therefore, the EITC unambiguously reduces hours worked and participation by raising family income and reducing the net-of-tax wage. If the husband’s earnings place the family in the ﬂat region, the credit unambiguously lowers labor-market participation and hours worked by married women (through a pure income effect). Of course, it is also possible for the wife’s work effort to increase the family’s credit if the husband’s earnings are in the subsidy region (up to $7,660 or $10,750 in 2004, depending on the number of

92

Eissa and Hoynes

Table 3.7 Average Tax Rates for Hypothetical Married Women with Children (2004 Tax Law): Taxes Calculated with and Without the EITC Average Tax Rates with the EITC

Average Tax Rates Without the EITC

No Work to PT Work

No Work to PT Work

No Work to FT Work

PT to FT Work

No Work to FT Work

PT to FT Work

A. Married women with one child (by hourly wage) $5.15/hour $7.50/hour

8% 30%

18% 36%

27% 41%

5% 14%

8% 20%

11% 25%

$10/hour

41%

35%

30%

25%

27%

29%

$12/hour

35%

33%

30%

27%

29%

30%

$15/hour

30%

30%

30%

30%

30%

30%

B. Married women with two or more children (by hourly wage) $5.15/hour

9%

18%

26%

5%

5%

5%

$7.50/hour

26%

32%

39%

5%

11%

18%

$10/hour

36%

36%

36%

14%

21%

28%

$12/hour $15/hour

41% 34%

36% 32%

30% 30%

23% 29%

27% 29%

30% 30%

Note: Calculated using TAXSIM. Numbers represent the average tax rate associated with increasing labor supply (either from no work to work or from part-time to full-time work). For each calculation, the husband is assumed to be working full-time and full-year at the same hourly wage as the wife. Taxes include all federal income taxes plus the payroll tax (assuming the individual bears the entire 15.3% rate). Part-time work is assumed to be 20 hours per week and 52 weeks per year, and full-time work is assumed to be 40 hours per week and 52 weeks per year.

children), but very few married couples are likely to have incomes in this range. The impact of the EITC on the labor supply of married couples therefore depends on the distribution of family income along the schedule, the distribution of earnings within the household, and the relative elasticities of labor supply with respect to taxes. Secondary earners whose spouses have earnings in the ﬂat and phase-out regions should be less likely to work and should work fewer hours, while those whose spouses have income in the phase-in region should be more likely to work but can increase or reduce their hours worked. Given the distribution of income, it is unlikely the EITC will have any positive effect on the labor supply of married women. To illustrate the work incentives for married women, table 3.7 presents average tax rates (using TAXSIM) similar to those presented

Behavioral Responses to Taxes

93

above for single women. Again the table presents average tax rates for entering and increasing work, including and excluding the EITC, and for a range of wages and family sizes. These calculations also require an assumption about the earnings of the husband. We assume that the husband’s labor supply is ﬁxed at full-time work and that he earns the same hourly wage as his wife. There are two striking observations from the table. First, average tax rates are everywhere higher for married women with the EITC compared to without the EITC. For example, a woman earning $10 per hour with two children who works part-time faces a 36 percent average tax rate with the EITC and a 14 percent tax rate without the EITC. The work disincentive impacts of the EITC extend quite high into the wage distribution, reﬂecting the high breakeven earnings point in the EITC ($31,338 for those with one child and $35,458 for those with two or more children in 2004). Our assumption of full-time work by the husband implies that our calculations do not show any positive participation incentives even for women with very low-earning husbands. With employment rates for low-educated married men approaching 95 percent or higher, however, those facing positive incentives are likely to be a small group of women (Eissa and Hoynes 2004). Second, married women with children face much higher marginal tax rates compared to single women with children. This occurs in part because the federal tax schedule is progressive, although calculations (in tables 3.6 and 3.7) show that the EITC plays an important role. The role of increasing marginal tax rates has been discussed in the taxes and labor supply literature more broadly (e.g., Eissa 1995), but there is less discussion of the EITC in this setting. 5.

What Have We Learned About Labor Supply and the EITC?

A large body of work has examined both the distributional impacts and the behavioral responses to the earned income tax credit, including labor supply, family formation, and consumption. Perhaps because of its explicit goal of encouraging work, the bulk of this evidence has been on labor supply. Our goal in this discussion is to summarize the major ﬁndings in the literature and to discuss their implications for the design of tax-transfer programs. Readers interested in a more exhaustive summary should see the comprehensive review by Hotz and Scholz (2003).

94

Eissa and Hoynes

To presage the main ﬁndings, there is overwhelming evidence that the EITC encourages work among single mothers but little evidence that eligible-working women adjust their hours of work in response to the EITC. Perhaps most striking about these ﬁndings is their consistency across different empirical methods—including both quasiexperimental methods (Eissa and Liebman 1996; Eissa and Hoynes 2004; Hotz, Mullin, and Scholz 2002) and more structural methods (Dickert, Houser, and Scholz 1995; Meyer and Rosenbaum 2001)—as well as different EITC expansions. To interpret and compare the empirical results, we ﬁrst clarify the different sources of identifying variation. While all individuals face the same tax schedule at any point in time, they face different tax rates based on their ﬁling (i.e., family) status, family size (number of children), non-labor income, and earned income (wages and hours worked). Additionally, tax schedules, and therefore rates, vary over time with policy reforms. The main difference between the quasiexperimental approach and more structural methods is the use of group versus individual-level variation in taxes. The ﬁrst approach assumes that all relevant wage and income changes are captured by group level variation in family type and size (presence and number of children) and time. The EITC effect is the change in employment (relative to that of childless women) of women with children after the EITC expansion. To the extent that tax rates, wages, and incomes are measured with error, this grouping approach reduces any measurement error bias in the coefﬁcients of interest. Most studies take advantage of the large federal expansions in the credit in 1986, 1990, and 1993 that affected incentives for single and married women with children. To control for other factors that may be contemporaneous with the policy changes, many studies utilize control groups and estimate difference-in-difference models. The EITC design and expansions suggest a number of possible strategies—such as comparing women with different family sizes (presence and number of children), marital status, earnings, or education levels—for identifying labor supply responses. These models seem to work well and provide robust estimates for the impact of the EITC on participation, but they may be less well suited for estimating the impacts on hours worked. Analyzing the determinants of hours worked is more complicated due to the changes in the composition of the working sample and the endogeneity of work more generally. We return to this ﬁnal point in Section

Behavioral Responses to Taxes

95

6. First, however, we review the empirical evidence on different margins of labor supply and for different groups. 5.1

The EITC and the Labor Supply of Single Taxpayers

5.1.1 Labor Force Participation The discussion of labor supply above shows that the EITC is expected to increase the participation of single women with children and reduce the hours worked for the majority of those already in the labor force. Several papers have estimated the impact of the EITC on employment decisions of single women (Dickert, Houser, and Scholz 1995; Eissa and Liebman 1996; Ellwood 2000; Grogger 2003; Hotz et al. 2002; Keane and Mofﬁtt 1998; Meyer and Rosenbaum 2000, 2001; Rothstein 2005), all consistently ﬁnding that the EITC increased their labor force participation. Several papers estimate difference-in-difference models applied to a single expansion or multiple expansions of the EITC. In these models, changes in the employment rate of the treatment or eligible group (single women with children) is compared to the change in the employment rate of a control or unaffected group. This approach is used by Eissa and Liebman (EL 1996), Ellwood (2000), Hotz et al. (2002), Meyer and Rosenbaum (MR 2000), and Rothstein (2005). The most common control group is childless single women (EL 1996, MR 2000, Rothstein 2005). Other comparisons exploit different features of the design of the EITC. Ellwood (2000) and Rothstein (2005) use the fact that the shape of the credit implies different incentives at different hourly wages and compare single mothers at different wage or skill levels. EL (1996), Hotz et al. (2002), and Meyer and Rosenbaum (2000) exploit the second child marginal credit and compare single women with one versus those with two or more children. To illustrate the ﬁndings from the difference-in-difference literature, ﬁgure 3.5 presents annual employment rates for women over the period 1984–2003. We show the employment rates for four groups: single women with children, single women without children, married women with children, and married women without children. We use the March CPS, where employment is deﬁned by any work over the calendar year, for these calculations.12 The ﬁgure shows the dramatic increase in employment rates for single women with children compared to single women without children. For example, between 1984 and 2003, employment rates of single mothers increased by 12 percentage

96

Eissa and Hoynes

Figure 3.5 Annual Employment Rates for Women by Marital Status and Presence of Children, 1984–2003. Notes: Calculations are based on the 1985–2004 March CPS and include women age 19– 44, not in school and not disabled. It also excludes women with positive earnings but zero hours and those with positive hours and zero earnings. Married women include those with spouse present, and single women include divorced, widowed, and never married women. Figure shows the fraction of women who worked at all in the calendar year.

points—from 73 percent in 1984 to 85 percent in 2003. Most of this change occurred between 1992 and 1999, when employment rates increased by 16 percentage points. This is during the period of the largest expansion in the EITC due to OBRA93. Over this same period, there was little change in employment rates of single women without children. Several authors (Ellwood 2000, Rothstein 2005, Meyer and Rosenbaum 2000) ﬁnd that the group with the most to gain from EITC expansions (e.g., women with lower wages, lower education levels, more children, and single women) experienced larger gains in employment rates. To illustrate these ﬁndings, the next two ﬁgures reproduce results from Meyer and Rosenbaum (2000) and Rothstein (2005). Figure 3.6 reproduces Meyer and Rosenbaum’s (2000) difference-indifference estimates of employment rates for three comparisons— single mothers and single women without children, single mothers and black men, and single mothers and married mothers—where

Behavioral Responses to Taxes

97

Figure 3.6 Difference-in-Difference Estimates of Annual Employment Rates, Single Mothers Versus Control Groups, 1984–1996. Source: Reproduced from Meyer and Rosenbaum (2000), Figure 9. Treatment group includes single women age 19–44 from the March CPS for 1985–1997. Each line represents the difference in employment rates between the treatment group (single mothers) and one of three control groups. The estimates are from a probit model that controls for demographic variables, residential location, unearned income, state unemployment rate (and its interactions with single mother and education level), state and year ﬁxed effects.

the difference is normalized to 0 in 1984.13 The ﬁgure shows that single mothers’ employment increased relative to each of these control groups, and the gains are well timed to the expansions of the EITC in 1986, 1990, and 1993. Figure 3.7, reproduced from Rothstein (2005), extends the single-mother-versus-childless-single-woman comparison and estimates difference-in-difference models of the 1993 expansion by wage level. The ﬁgure shows that employment increased with OBRA93 for all single mothers relative to childless women and, as expected, the impacts decrease with the woman’s wage. Another approach recognizes that tax-transfer policy has nonneutral effects within groups and uses individual variation in net wages and net non-labor income. Consequently, identiﬁcation is based on cross-sectional and time variation in the EITC and federal taxes. For example, Meyer and Rosenbaum (2001) use instead state-time variation in labor supply incentives and measure the gains to work from wages,

98

Eissa and Hoynes

Figure 3.7 Difference-in-Difference Estimates of OBRA93 on Annual Employment Rates, by Wage Level, Single Mothers Versus Single Women without Children. Source: Reproduced from Rothstein (2005), Figure 10. The sample includes women age 19–64 from the March CPS. The pre-OBRA93 period is 1992–1993, and the post-OBRA93 period is 1996–1997. Each line represents the difference in employment rates between the treatment group (single mothers with one or two or more children) and the control group (single women without children). The estimates are from a semi-parametric model that controls for education, potential experience, race/ ethnicity, number and presence of children, and residence.

net of federal and state taxes, and transfer beneﬁts (AFDC/TANF, food stamps, and Medicaid).14 Overall these studies suggest a strong positive relationship between the EITC and employment rates of single women with children. Further, the results are remarkably consistent across different policy expansions, different control groups, and different methodologies. The estimated size of the labor supply effect differs depending on the particular expansion considered. Eissa and Liebman (1996) ﬁnd that the 1986 expansion of the EITC led to a 2.8 percentage point increase in participation (out of a base of 74.2) for single mothers. Meyer and Rosenbaum (2001) ﬁnd that 60 percent of the 8.7 percentage point increase in annual employment of single mothers between 1984 and 1996 is due to the EITC. They ﬁnd that a smaller amount, 35 percent of the increase in participation between 1992 and 1996, is due to the EITC (with the remainder due to welfare reform and other changes). The range of the implied labor force participation elasticity with respect to net income across all studies is quite narrow—between 0.69 and 1.16 (Hotz and Scholz 2003).15

Behavioral Responses to Taxes

99

Figure 3.8 Average Annual Hours Worked for Working Single Women by Presence of Children, 1984–2003. Notes: Calculations based on the 1985–2004 March CPS and include single women age 19–44, not in school, not disabled, and who worked at all during the calendar year. It also excludes women with positive earnings but zero hours and those with positive hours and zero earnings. Single women include divorced, widowed, and never married women. Figure shows average annual hours worked conditional on working.

5.1.2 Hours Worked The EITC is expected to unambiguously reduce hours worked by the vast majority of eligible workers. In sharp contrast to the ﬁndings above, however, there is little evidence consistent with this prediction. A limited set of papers have examined the impact of the EITC on the hours worked by single mothers. This is in part because estimating the hours worked response of workers to the EITC budget constraint is fundamentally a harder empirical problem. In particular, one has to deal with the selection of individuals into the labor force. Eissa and Liebman (1996) apply their difference-in-difference model to annual hours worked (conditional on working) and ﬁnd a small positive (and marginally signiﬁcant) impact on all single mothers and a zero impact on low-educated single mothers. Meyer and Rosenbaum (1999) ﬁnd mixed (positive and negative) but insigniﬁcant impacts of the EITC on hours worked (conditional on working). Rothstein (2005) ﬁnds no difference between single mothers and childless single women in weekly hours worked (conditional on working) across the wage distribution. Keane and Mofﬁtt (1998) estimate a structural model of labor supply choice and simulate the effect of the EITC on total hours but do not present a separate estimate for those already working.

100

Eissa and Hoynes

Figure 3.9 Average Annual Hours Worked for Single Women in the Phase-out Region of the EITC, 1984–2003, by Presence of Children. Notes: Calculations based on the 1985–2004 March CPS and include single women age 19–44, not in school, and not disabled. The ﬁgure also excludes women with positive earnings but zero hours and those with positive hours and zero earnings. Single women include divorced, widowed, and never married women. The ﬁgure shows average annual hours worked for the sample of single women who have earnings within $1,000 of the 1996 phase-out region of the EITC (in real terms).

To illustrate the ﬁndings from the literature, ﬁgure 3.8 presents average hours worked (conditional on working) for working single women with and without children over 1984–2003 using our CPS sample described above. Consistent with the studies surveyed, the ﬁgure provides no evidence that hours of work decreased for single mothers relative to single women without children as the EITC expanded. Figure 3.9 reﬁnes this analysis by focusing on the phase-out region of the credit where the marginal tax rates are highest. Speciﬁcally, the ﬁgure plots average earnings for single women with earnings within $1,000 of the 1996 phase-out region of the credit. Consider a single woman with two children who, in 1984, earned $15,000 per year. At that time, she was above the EITC breakeven earnings point. After the 1986 expansions, the same woman (at the same real earnings level) faced a 10 percent EITC phase-out rate. After the 1990 EITC expansions, the phase-out rate she faced was about 14 percent, and ﬁnally after the 1993 expansions the phase-out was up to about 21 percent. Single mothers with earnings in this range (which we adjusted for a change in prices each year) are in the range where the tax rates have increased the most. Strikingly, ﬁgure 3.9 shows no pattern of a reduc-

Behavioral Responses to Taxes

101

Figure 3.10 Difference-in-Difference Estimates of Annual Employment Rates, 1984–1996, Married Mothers Versus Married Women Without Children. Source: Reproduced from Eissa and Hoynes (2004), Figure 5. The sample includes married women age 25–54 with less than a high school education from the March CPS for 1985–1997. The sample excludes couples where one spouse was ill or disabled, in the military, or in school full-time during the previous year. The line with square markers gives the change (relative to 1984) in average annual employment rates for married women with children compared to married women without children. The estimates are from a probit model that controls for demographic variables, state dummies, state labor market variables, and year effects for both groups. The line with circle markers gives the change (relative to 1984) in the average credit calculated for a secondary earner (the wife) for a ﬁxed (1996) distribution of the husband’s earnings.

tion in hours worked for single women with children relative to single women without children. Another source of evidence builds on the prediction from labor supply theory that taxpayers should be bunched at the kinks in the EITC schedule (and should be less present at the end of the EITC schedule). Liebman (1998) and Saez (2002) use tax return data and ﬁnd no evidence consistent with these predictions. At the end of this section, we return to these results and discuss why the literature ﬁnds a participation effect but no hours worked effect. 5.2 The EITC and Labor Supply of Married Taxpayers While the literature on married women and the EITC is limited, the results are quite consistent with the theoretical expectations. In our earlier work (Eissa and Hoynes 2004), we estimate a differencein-difference model comparing married mothers to married women

102

Eissa and Hoynes

without children and ﬁnd that the 1993 EITC expansion led to a one percentage point reduction in the participation rate of married mothers. These results are illustrated in ﬁgure 3.10, reproduced from Eissa and Hoynes (2004), which presents estimates of the change (relative to 1984) in average employment rates for married women with children compared to married women without children (controlling for demographics, macroeconomic cycles, and state and year ﬁxed effects). For comparison, the ﬁgure also shows the change (relative to 1984) in the average EITC for their sample. This shows that employment by married mothers declined relative to that of childless married women following the 1990 and 1993 expansions (though not the 1986 expansion). This ﬁnding is consistent with Ellwood (2000) who compares married mothers in high- and low-wage quartiles and ﬁnds that expansions of the EITC are associated with a reduction in labor market participation by married mothers. We also estimate (in Eissa and Hoynes 2004) a reduced form employment model for married women (as secondary earners) and parameterize the impact of the EITC and federal taxes on after-tax gains to entering work. We ﬁnd that the expansions in the EITC between 1984 and 1996 led to about a one percentage point reduction in the employment of married mothers. These estimates are similar to but somewhat smaller than our difference-in-difference estimates. In Eissa and Hoynes (forthcoming), we ﬁnd that expansions in the EITC are also associated with reductions in hours worked for married women who are working. We estimate hours of work as a function of after-tax wages and unearned income, again assuming women are secondary earners. We instrument for the after-tax wage using tax parameters. We ﬁnd that expansions in the EITC from 1984 to 1996 led to a small, 1 to 4 percent, decrease in annual hours for married women with children. Heim (2005) estimates a structural model of family labor supply and ﬁnds similar impacts on hours worked of married women (yet he ﬁnds no impact of the EITC on the employment of married women). 6.

Discussion

6.1 Why Do We Observe a Participation Effect but No Hours Effect? A consistent and somewhat puzzling ﬁnding in the empirical literature on the EITC and labor supply is the large response of the participation decision and the lack of any response in the reported hours worked by

Behavioral Responses to Taxes

103

taxpayers in the labor force. Here, we focus on single mothers who, as noted above, represent over three-quarters of EITC recipients. The puzzle is why no impact on hours worked is found in any evaluation of the EITC. Theory suggests that we should observe a decline in the hours worked by all taxpayers beyond the phase-in of the credit. While the income effect operates to reduce hours worked in the ﬂat region, both income and substitution effects operate to reduce hours worked in the phase-out region of the credit. We review several explanations that have been offered. The ﬁrst explanation is based on a standard ﬁnding in the empirical labor supply literature: the labor supply elasticity falls when estimated using a sample of working (rather than all) women (Mroz 1987, Triest 1990). The implication of this ﬁnding is that the participation (extensive) margin is more responsive than the hours (intensive) margin. It may well be that the hours worked elasticity is too small to be estimated precisely using quasi-experimental approaches. If, for example, the elasticity of hours worked by single women with children is 0.3, then the 1993 EITC expansion would reduce hours of work by 8 percent.16 It is not unreasonable to consider this reduction too small to be identiﬁed empirically, especially given the somewhat crude comparison-of-means approach typically used. If this is correct, then the next question is, Why is the elasticity on the intensive margin consistently lower than the extensive margin? One reason why there may be no effect of the EITC on hours worked is that workers are unable to choose continuous hours of work. This would occur if, for example, workers are bound by institutional restrictions or by norms for part- and full-time work.17 To examine the validity of this explanation, ﬁgure 3.11 presents the distribution of annual hours worked for our CPS sample of single mothers in 1986, 1990, 1993, and 2000. (Note this includes all single mothers and, in particular, is not limited to those with earnings in the EITC eligible range.) Two observations are worth noting. First, annual hours are highly concentrated at full-time work. While some single mothers work part-time (primarily part-year), the majority of single mothers work full-time (full year). Second, the increase in employment mirrors the increase in full-time, full-year work. One (testable) interpretation of this trend is that new workers enter the labor market at full-time work—consistent with a labor supply model with large ﬁxed costs of work. Another reason why no hours-worked effect has been estimated is that reported hours in the survey data are measured with error (Bound and Krueger 1991; Bound, Brown, and Mathiowetz 2001).

104

Eissa and Hoynes

Figure 3.11 Kernel Density Estimate of Annual Hours, Single Mothers. Source: Reproduced from Eissa, Kleven, and Kreiner (2004). The ﬁgure uses the March CPS sample of single women with children, age 19–44, not in school, and not disabled. It also excludes women with positive earnings but zero hours and those with positive hours and zero earnings. Single women include divorced, widowed, and never married women. The ﬁgure provides a smoothed estimate of the distribution of annual hours worked (nonworkers have annual hours equal to 0).

Annual hours of work are typically constructed as the product of weeks worked last year times usual hours worked per week last year. Other than standard recall error, individuals with varying weekly hours may make errors in averaging. Finally, taxpayers may not be fully aware of the structure of the EITC schedule. Almost all taxpayers receive the EITC as a lump sum payment with their annual tax return (Friedman 2000), and therefore have little opportunity to learn about the features of the credit. This may be especially true for the phase-out region of the credit, in part because of the confounding effects of non-EITC federal income taxes. This EITC structure and its delivery are in contrast to the monthly reporting period in welfare programs and to the weekly or biweekly paycheck and its opportunities for learning about FICA or ordinary income taxes. Further, informal and formal surveys of people eligible for the EITC suggest that knowledge about the credit is relatively high,

Behavioral Responses to Taxes

105

though by no means universal (Liebman 1998, Phillips 2001, Smeeding et al. 2000, Romich and Weisner 2000). Phillips, for example, ﬁnds that about 66 percent of families nationally had heard of the EITC. There is scant evidence, however, that those likely eligible for the credit understand its structure or the different incentives implied. 6.2 Implications for Optimal Tax and Welfare The ﬁnding that labor force participation responses are more signiﬁcant than hours worked responses has several important implications for the design of tax-transfer programs and the welfare evaluation of taxation. Recent work has shown that accounting for labor force participation responses can change the optimal transfer program (Saez 2002). More precisely, this work has shown that with sufﬁciently high participation elasticities, the optimal tax-transfer scheme can be similar to the EITC—with negative marginal tax rates at the bottom of the earnings distribution. On the other hand, an EITC would be inefﬁcient in a standard model with only intensive (hours worked) responses.18 Liebman (2002) extends this work by examining more closely the optimal design of the EITC. He uses a micro-simulation model calibrated to 1999 CPS data to illustrate the trade-offs in the design of an EITC—including the optimal maximum credit, phase-in and phaseout rates—with ﬁxed costs and participation effects. Liebman ﬁnds that the efﬁciency cost of transferring income through the EITC is substantially lower than previous studies have found, in large part because of the participation response of single mothers and the associated reduced welfare spending. His simulations suggest a cost of less than $2 to provide a transfer worth $1 to EITC recipients. Eissa, Kleven, and Kreiner (2005) take a different approach and examine the impact of participation responses on the welfare evaluation of actual tax reforms. They extend the standard framework for welfare evaluation of tax reforms to account for discrete labor market entry by way of non-convexities in preferences and budget sets. Such nonconvexities are signiﬁcant because they allow ﬁrst-order welfare effects along the extensive (participation) margin. EKK simulate the effects of the 1986, 1990, 1993, and 2001 tax acts in the United States and show that each had different effects on tax rates along the intensive and extensive margins. The 1993 EITC expansion, for example, reduced the tax rates on labor force participation but increased the marginal tax rates on hours worked for most workers. The authors show that conﬂating these two tax rates in welfare analysis can be fundamentally

106

Eissa and Hoynes

misleading. For tax reforms that change average tax rates differently than marginal tax rates (such as the 1993 expansion of the EITC), ignoring the participation margin can lead to even the wrong sign of the welfare effect. 6.3 Incidence Understanding the effects of the EITC on wages is directly relevant to evaluating the effectiveness of the EITC and its longer-run effects of labor supply. If the EITC reduces gross wages below what they would have been otherwise, the credit may do little, if anything, to encourage labor supply, and may operate to the primary beneﬁt of low-wage employers. Recent work by Leigh (2004) and Rothstein (2005) directly examine the effect of the EITC on pre-tax wages, using very different methods. Leigh uses variation across states in the presence and generosity of state EITC add-ons and ﬁnds a very strong negative effect of the credit on wages, implying very little beneﬁt of the credit for its recipients. One drawback to Leigh’s approach is that state EITCs are small relative to the federal program, and many recipients may not be aware of their existence. As a consequence, behavioral responses may be muted, making it difﬁcult to identify the incidence of the credit. Rothstein uses variation from the 1993 federal credit expansion along the wage distribution and ﬁnds that low-skill women’s wages actually increased slightly even as their labor force participation increased. These results imply an upward-sloping demand curve, though they are by no means sufﬁciently precise to reject a small downward slope. One possible explanation for the conﬂicting ﬁndings of the Leigh and Rothstein studies is the different sources of identiﬁcation. Clearly more work needs to done to increase our understanding of the broader labor market effects of the EITC. 7.

Conclusions

An unusual feature of the EITC is its explicit goal to use the tax system to encourage and support those who choose to work. A large body of work has evaluated the labor supply effects of the EITC and has generated several important ﬁndings regarding behavioral response to taxes. This paper reviews this empirical literature and focuses on two main lessons. It argues that perhaps the main lesson learned from the evidence is the conﬁrmation that real responses to taxes are important;

Behavioral Responses to Taxes

107

labor supply does respond to the EITC. The second major lesson is related to the nature of the labor supply response, namely, that the response is concentrated along the extensive (entry) margin rather than the intensive (hours worked) margin. We discuss different explanations of this pattern of labor supply responses and its implications for the optimal design of tax-transfer programs and for the welfare evaluation of tax reforms. Empirical evidence suggests that transferring dollars to the needy using an EITC—rather than a negative income tax—may be optimal. This evidence cannot yet support more precise judgments about the optimal design of the EITC, however. That would require—among other things—identifying the hours of work response of workers to the EITC. Notes This paper was prepared for the Tax Policy and the Economy conference held on September 15, 2005, in Washington, D.C. We wish to thank Jim Poterba for very useful comments, Bruce Meyer and Jesse Rothstein for providing us with data, Dan Feenberg for special tabulations from TAXSIM, Jeff Rohaly of the Urban Institute for tabulations of SOI data, and Alan Barreca for research assistance. 1. By comparison, federal spending on Temporary Assistance to Needy Families (TANF), which provides block grants for Aid to Families with Dependent Children (AFDC), stands at about $18.6 billion (Hotz and Scholz 2003). 2. Our analysis of married couples focuses on married women because they have been found to be considerably more responsive to changes in wages and income than married men. 3. Beginning in 1994, a small credit is available to low-income workers without children. 4. See Eissa and Liebman (1996) for a more extensive discussion of EITC rules. 5. TRA86 returned the real maximum credit to its 1975 level and indexed the EITC for inﬂation. 6. Legislation adopting a state EITC in Virginia was passed in 2004 but will not take effect until January 2006. In addition, local governments in Montgomery County, Maryland, and Denver, Colorado, offer their own version of EITCs. 7. Another dimension of labor supply which we do not discuss here is human capital accumulation. This is examined in Heckman, Lochner, and Cossa (2002) and reviewed in Hotz and Scholz (2003). 8. We discuss later the potential impact of the EITC on wages earned by less-skilled workers. 9. An important omission here are means-tested transfers such as AFDC/TANF and food stamps. These programs have large implicit tax rates and create large disincentives

108

Eissa and Hoynes

to enter and increase work (Mofﬁtt 1992). Ignoring transfer programs does not affect the qualitative conclusions in this section. 10. These calculations assume that the individual bears the burden of the entire FICA tax, with a combined marginal tax rate of 15.3 percent. For these calculations, total taxes paid equal federal income tax plus the payroll tax less the (nonrefundable and refundable) child tax credit and the EITC. 11. The ﬁgures in Eissa and Hoynes (2004) suggest that—measured by relative earnings—women are predominantly secondary earners. In a sample of couples with a high school education or less, about 90 percent of all wives, and 85 percent of working wives, earn less than their husbands. In this discussion and in the literature more broadly, women are assumed to be secondary earners. 12. We limit the analysis to women age 19–44 who are not in school or disabled. We also exclude women with positive earnings but zero hours and women with positive hours and zero earnings. Single women are divorced, separated, widowed, or never married and married women include those with a spouse present. The results are based on the CPS survey years 1985–2004. 13. Meyer and Rosenbaum’s (2000) estimates are based on a sample from the March CPS for 1984–1996. The estimates are from a probit model that includes controls for demographic variables, residential location, unearned income, state unemployment rate (and its interactions with single mother and education level), and state and year ﬁxed effects. 14. Meyer and Rosenbaum (2001) also control for welfare reform variables and state spending on educational training, job search assistance, and child care assistance for AFDC recipients. The gains to work are obtained by integrating over the empirical distribution of wages and hours worked for working women and calculating net income at each wage-hours possibility. 15. This range excludes the estimates from Hotz et al. (2002). Their studies differ from the others in that their sample is limited to welfare recipients (or prior recipients). 16. The 1993 expansion in the EITC increased the phase-out rate to 21.06 (15.98) percent from 13.93 (13.21) percent for families with one child (two or more children). Assuming the woman faces a 15 percent marginal tax rate and the full 15.30 percent FICA rate, this implies a 13 percent reduction in the net of tax wage. 17. Alternatively, one can argue that a worker can change jobs as an additional way to change hours. 18. Saez shows that the optimal program is instead a classical negative income tax program, with a substantial income guarantee that is phased out at a high rate.

References Bound, John, Charles Brown, and Nancy Mathiowetz (2001). ‘‘Measurement Error in Survey Data,’’ in James J. Heckman and Edward Leamer (eds.), Handbook of Econometrics, Volume 5. Amsterdam: North Holland. Bound, John, and Alan B. Krueger (1991). ‘‘The Extent of Measurement Error in Longitudinal Earnings Data: Do Two Wrongs Make a Right?’’ Journal of Labor Economics, 9:1–24.

Behavioral Responses to Taxes

109

Dickert, Stacy, Scott Houser, and John Karl Scholz (1995). ‘‘The Earned Income Tax Credit and Transfer Programs: A Study of Labor Market and Program Participation,’’ in James Poterba (ed.), Tax Policy and the Economy. Cambridge, Mass.: MIT Press. Eissa, Nada (1995). ‘‘Taxation and Labor Supply of Married Women: The Tax Reform Act of 1986 as a Natural Experiment,’’ NBER working paper no. 5023. Eissa, Nada, and Hilary Hoynes (2004). ‘‘Taxes and the Labor Market Participation of Married Couples: The Earned Income Tax Credit,’’ Journal of Public Economics, 88(9– 10):1931–1958. Eissa, Nada, and Hilary Hoynes (forthcoming). ‘‘The Hours of Work Response of Married Couples: Taxes and the Earned Income Tax Credit,’’ in Peter Birch Sorensen (ed.), Tax Policy and Labor Market Performance, CESifo/MIT Press conference volume. Eissa, Nada, Henrik Kleven, and Claus Kreiner (2004). ‘‘Evaluation of Four Tax Reforms in the United States: Labor Supply and Welfare Effects for Single Mothers’’ NBER working paper no. 10935. Eissa, Nada, and Jeffrey Liebman (1996). ‘‘Labor Supply Response to the Earned Income Tax Credit,’’ Quarterly Journal of Economics, CXI:605–637. Eissa, Nada, Jesse Rothstein, and Austin Nichols (2005). ‘‘Tax-Transfer Schemes and Wages: The Earned Income Tax Credit,’’ paper presented at AEA/NEA session ‘‘Skills, Policy, and Labor Market Outcomes Across Demographic Groups.’’ Ellwood, David (2000). ‘‘The Impact of the Earned Income Tax Credit and Social Policy Reforms on Work, Marriage and Living Arrangements,’’ National Tax Journal, 53(4):Part 2. Friedman, Pamela (2000). ‘‘The Earned Income Tax Credit,’’ Welfare Information Network, issue notes. Grogger, Jeffrey (2003). ‘‘The Effects of Time Limits, the EITC, and Other Policy Changes on Welfare Use, Work, and Income Among Female-Headed Families,’’ Review of Economics and Statistics, 85(2):394–408. Heckman, James, Lance Lochner, and Ricardo Cossa (2002). ‘‘Learning-By-Doing vs. OnThe-Job-Training: Using Variation Induced by the EITC to Distinguish Between Models of Skill Formation,’’ NBER working paper no. 9083. Heim, Bradley (2005). ‘‘The Impact of the Earned Income Tax Credit on the Labor Supply of Married Couples: Structural Estimation and Business Cycle Interactions,’’ Duke University, mimeo, May. Hotz, V. Joseph, Charles Mullin, and John Karl Scholz (2002). ‘‘The Earned Income Tax Credit and the Labor Market Participation of Families on Welfare,’’ UCLA, mimeo. Hotz, V. Joseph, and John Karl Scholz (2003). ‘‘The Earned Income Tax Credit,’’ in Robert Mofﬁtt (ed.), Means-Tested Transfer Programs in the United States. Chicago, Ill.: University of Chicago Press. Internal Revenue Service (2004). General Description Booklet for the 2001 Public Use Tax File, Statistics of Income Division. Keane, Michael, and Robert Mofﬁtt (1998). ‘‘A Structural Model of Multiple Welfare Program Participation and Labor Supply,’’ International Economic Review, 39(3):553–589.

110

Eissa and Hoynes

Leigh, Andrew (2004). ‘‘Who Beneﬁts from the Earned Income Tax Credit? Incidence Among Recipients, Coworkers and Firms.’’ Australian National University, mimeo. Liebman, Jeffrey (1998). ‘‘The Impact of the Earned Income Tax Credit on Incentives and the Income Distribution,’’ in James Poterba (ed.), Tax Policy and the Economy, Volume 12. Cambridge, Mass.: MIT Press. Liebman, Jeffrey (2002). ‘‘The Optimal Design of the Earned Income Tax Credit,’’ in Bruce D. Meyer and Douglas Holtz-Eakin (eds.), Making Work Pay: The Earned Income Tax Credit and Its Impact on American Families. New York, NY, Russell Sage Foundation. Llobrera, Joseph, and Robert Zahradnik (2004). ‘‘A Hand Up: How State Earned Income Tax Credits Help Working Families Escape Poverty in 2004.’’ Washington, D.C.: Center for Budget and Policy Priorities. Meyer, Bruce, and Dan Rosenbaum (1999). ‘‘Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers,’’ NBER working paper no. 7363. Meyer, Bruce, and Dan Rosenbaum (2000). ‘‘Making Single Mothers Work: Recent Tax and Welfare Policy and Its Effects,’’ National Tax Journal, 53(4):Part 2. Meyer, Bruce, and Dan Rosenbaum (2001). ‘‘Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers,’’ Quarterly Journal of Economics, 116(3):1063–1113. Mofﬁtt, Robert (1992). ‘‘Incentive Effects of the U.S. Welfare System: A Review,’’ Journal of Economic Literature, 15(1):1–61. Mroz, Thomas (1987). ‘‘The Sensitivity of an Empirical Model of Married Women’s Hours of Work to Economic and Statistical Assumptions,’’ Econometrica, 55:765–799. Phillips, Katherin Ross (2001). ‘‘Who Knows About the Earned Income Tax Credit?’’ The Urban Institute, New Federalism, National Survey of America’s Families, Series B, No. B-27. Romich, Jennifer, and Thomas Weisner (2000). ‘‘How Families View and Use the EITC: Advance Payment Versus Lump Sum Delivery,’’ National Tax Journal, 53(4):Part 2. Rothstein, Jesse (2005). ‘‘The Mid-1990s EITC Expansion: Aggregate Labor Supply Effects and Economic Incidence,’’ Princeton Unversity, mimeo. Saez, Emmanuel (2002). ‘‘Do Taxpayers Bunch at Kink Points?’’ University of California, Berkeley, mimeo. Smeeding, Timothy, Katherin Ross Phillips, and Michael O’Connor (2000). ‘‘The EITC: Expectation, Knowledge, Use, and Economic and Social Mobility,’’ National Tax Journal, 53(4):Part 2. Triest, Robert (1990). ‘‘The Effect of Income Taxation on Labor Supply in the United States,’’ Journal of Human Resources, XXV:491–516. U.S. House of Representatives (2004). ‘‘Background Materials and Data on Programs within the Jurisdiction of the Committee on Ways and Means.’’ Washington, D.C.: Government Printing Ofﬁce.

4 Splitting Tax Refunds and Building Savings: An Empirical Test Sondra Beverly, University of Kansas Daniel Schneider, Harvard Business School Peter Tufano, Harvard Business School, NBER, and D2D Fund

Executive Summary Families are more likely to save if they can commit to savings before funds are in-hand (and subject to spending temptations). For low- and moderate-income U.S. families, an important savings opportunity arises annually, during income tax season. We study a group of lowincome individuals in Tulsa, Oklahoma, who were encouraged to save parts of their federal refunds at the time of tax ﬁling. Those who agreed to save directed a portion of their refund to a savings account and arranged to have the rest sent to them in the form of a check. Eligible individuals could also open low-cost savings accounts. We document the demand for these services, the characteristics of those who sought to participate, the savings goals of those who participated, the immediate savings generated by the program, and the disposition of savings a few months after receipt. This pilot study suggests that there may be demand among low-income families for a refund-splitting program that supports emergency needs as well as asset building, especially if a basic savings product is available to all at the time of tax ﬁling. 1.

Introduction

Saving is hard work. A saver must defer gratiﬁcation today for the sake of a better life tomorrow, much like a dieter, a recovering alcoholic, or an ex-smoker must resist the nearby chocolate cake, beer, or cigarette. In each case, removing the source of temptation is one prudent strategy. In this spirit, would-be savers often direct money to savings before they are tempted to spend it. Financial advisors tell people to ‘‘pay themselves ﬁrst,’’1 and this advice is embodied in a number of institutional programs. In the workplace, automatic investment plans

112

Beverly, Schneider, and Tufano

enable workers to deduct monies from their paychecks to fund savings programs like 401(k) and 403(b) plans. Workers will sometimes force savings by increasing the amount of withholding on their W2 forms. Also in the workplace, the novel SMarT program, proposed by Thaler and Benartzi (2004), enables workers to precommit to save their raises. Outside the workplace, automatic investment plans, like those offered by mutual funds, allow investors to regularly sweep money from transactional accounts into long-term investment accounts. Nearly all of the 500 largest mutual funds (deﬁned by assets under management) allow investors to set up automatic monthly contributions. All of these programs share a common trait: they tap into potential savings sources before a person has the opportunity to spend. For lowand moderate-income (LMI)2 U.S. families, annual federal and state tax refunds are an important source of savable funds. In 2001 (the most recent year for which data are available), LMI tax-ﬁlers received more than $78 billion in total federal refund payments, including the earned income tax credit (EITC), the child tax credit (CTC), other refundable credits, and refunds from over-withholding (Internal Revenue Service, Statistics of Income 2001).3 This massive ﬂow of funds, which takes place primarily during a few weeks in the early part of the tax season, represents a substantial portion of the inﬂows of an LMI family. With an average value of $1,415 in 2001, a federal refund payment is often more than the amount of a low-income family’s regular bi-weekly paycheck, perhaps hundreds of dollars more.4 For a family with income of less than $30,000 a year, the federal tax refund would likely be the single largest payment received all year. In addition to being large, tax refunds are perhaps the most savable of funds, out of reach for most of the year and hence beyond temptation. Yet just as most workers cannot direct all of their salary into an automatic investment program, most refund recipients cannot direct all of their tax refunds to savings. In both cases, some funds are needed for current living expenses and perhaps to pay past-due bills. Ideally, a refund recipient could save part of her refund at the source, rather than receiving all of the money and then having to decide how much of it to spend and how much to save. This precommitment could be physical in nature (investing in a savings product with limited liquidity), mostly psychological in nature (mentally segregating funds for savings), or a combination of the two. However, neither the Internal Revenue Service (IRS) nor the ﬁnancial service sector has moved quickly to enable LMI families to split their refunds and simultaneously precom-

Splitting Tax Refunds and Building Savings

113

mit to savings. It is unclear whether this reluctance is motivated by a lack of facts (e.g., what fraction of families would use refund splitting if it were offered and how much money would be saved out of refunds), a lack of focus on low-income families by ﬁnancial service ﬁrms, operational barriers to implementation, or perhaps other factors. To gather some tentative empirical evidence about refund splitting, we entered into a collaborative research project in the spring of 2004.5 As described in more detail below, for four weeks during the tax season, the team offered about 516 LMI ﬁlers at one of the nation’s largest free tax preparation sites the ability to split their refunds into saving and spending components and, if needed, to open new savings accounts to house their new savings. A participant could also deposit his entire refund into a newly established account. The collaborative research team included the Community Action Project of Tulsa County (CAPTC), Doorways to Dreams (D2D) Fund, the Bank of Oklahoma (BOk), the Annie E. Casey Foundation (AECF), and the coauthors of this study. The program was called Refunds to Assets (R2A), and the participants—whose adjusted gross incomes averaged $12,300—were drawn from CAPTC’s existing tax preparation sites, with staff from CAPTC working on-site to enroll clients in the program. D2D Fund oversaw the legal and administrative apparatus of splitting. BOk opened new accounts for participants who desired them.6 AECF and D2D provided funding for the experiment. We, the academic research team, developed the research design, created and administered a series of surveys, and analyzed data from the experiment. While our results are only suggestive, they support the intriguing notion that refund splitting might increase savings for low- and moderate-income families or help them better manage their spending. In short, we ﬁnd that:

Over 20 percent of refund recipients studied sought to participate in R2A by splitting their refunds, opening new savings accounts, or both. Fifteen percent of refund recipients were able to participate in R2A. Of these 79 participants, 56 percent opened new accounts and split their refunds, 27 percent split their refunds and used existing accounts, and 17 percent chose not to split and instead deposited their entire refunds into new accounts. The average participant saved $606 or about 47 percent of her refund. One-fourth of participants had existing savings. These participants sent

114

Beverly, Schneider, and Tufano

an average of $924 to savings; by their reports, this represented a 90 percent increase in savings. Three-fourths of participants reported no prior saving, and these participants sent an average of $479 to savings. Participants had many savings goals, with the most common being general precautionary savings, car purchase or repair, and home purchase or repair. While we sometimes equate savings with long-term goals, many of the saving goals mentioned by these low-income families were short-term. About four months after receiving refunds, bank account level data shows that the average balance had dropped 83 percent. However, 78 percent of participants said that they were still saving or had met at least one of their savings goals, and a larger fraction of R2A participants were still saving a portion of the refund or had met a savings goal, when compared with a sample of refund recipients at a comparison site without access to the R2A service. Follow-up survey data show that an overwhelming majority of participants were pleased with the service, planned to use it again, would recommend it to their friends, and were willing to pay a nominal fee to split. We are not alone in advancing the idea that tax refunds could constitute an important pool of potential savings. A refund-based savings plan has been tested in a study by a team of researchers associated with the Retirement Security Project. Working with H&R Block, the researchers carried out an experiment in St. Louis in which they tested whether they could increase saving out of refunds by providing matchfunding to savers opening individual retirement accounts (IRAs). The preliminary results conﬁrm evidence from other studies that matching programs can augment savings;7 in this case, IRA contributions increase with the level of match funding offered (Duﬂo, Gale, Liebman, Orszag, and Saez 2005). While our studies differ along some key dimensions, they independently provide suggestive evidence of the power of refund-based savings programs. The remainder of this paper provides the motivation for this research project (Section 2), describes the experiment (Section 3), presents our ﬁndings regarding take-up and initial saving (Section 4), presents follow-up results related to saving outcomes and perceptions of the program (Section 5), and offers some tentative conclusions and suggestions for future research (Section 6).

Splitting Tax Refunds and Building Savings

2.

115

Why Test Refund-Splitting for LMI Tax Filers?

2.1 Public Policy and Asset Building U.S. public policy enthusiastically encourages asset building through a host of programs, estimated by some to cost the government $335 billion a year (Woo, Schweke, and Buchholz 2004). Calls for an ownership society underscore the belief that asset building can strengthen families and societies. By building ﬁnancial assets, families can weather temporary income shocks, demonstrate credit-worthiness, purchase non-ﬁnancial assets (like homes, durables, or education), earn income, and prepare for retirement. Evidence suggests that asset holding can lead to increased residential stability, higher levels of social and political engagement, and better life outcomes for children (Sherraden 1991, Page-Adams and Scanlon 2001). Despite incentives to save and clear beneﬁts from saving, Americans save little of their annual earnings and have little in ﬁnancial assets. Data from the National Income and Products Accounts show that the personal saving rate in the United States is low by historical standards. In 2003, Americans saved just 1.4 percent of disposable income, sharply down from the 1950s through the 1990s when the saving rate was between 7 percent and 10 percent (Bureau of Economic Analysis 2004).8 In addition to having low saving rates, Americans generally have small stocks of assets. Haveman and Wolff (2001) have deﬁned asset poverty as lacking enough assets to live at the poverty level for three months. Using this deﬁnition Caner and Wolff (2002) found that, in 1999, 40 percent of Americans did not have enough liquid assets to live at the poverty level for three months. In a later paper, Wolff and Caner (2004) ﬁnd that asset poverty rates have been largely static from 1984 to 1999. More recent data from the Survey of Consumer Finances emphasizes the degree to which the poor have very low levels of assets; in 2001, 25 percent of households in the bottom income quintile had no ﬁnancial assets at all (Aizcorbe, Kennickell, and Moore 2003). Public policy supporting asset building largely targets those who have higher incomes and more assets. A recent report estimates that the bottom 60 percent of tax ﬁlers by income received less than 5 percent of the approximately $335 billion that the federal government spent to encourage asset building through homeownership, retirement savings, small business development, and investing (Woo, Schweke, and Buchholz 2004). This bias is due in no small part to the provision

116

Beverly, Schneider, and Tufano

of asset-building incentives through the income tax system, thereby delivering the highest beneﬁts to those who pay the highest taxes. Reasonable people might disagree about where incentives for asset building should be concentrated on the basis of equity or macro-economic effect. However, in terms of strengthening families and civic values, adding some savings to those with the least probably has the largest marginal effect. Creating incentives to add $10,000 in assets to a family with $1 million in assets will surely have less impact than adding the same amount to a family with no assets. 2.2 Refunds and Asset Building The current system encourages saving by reducing tax payments made by the well-to-do. These tax-payer based incentives are ineffective for low-income families who pay little tax, but the federal tax refund system is a potentially powerful way to facilitate asset building among LMI families. As discussed below, the federal tax system delivers sizeable refunds that can be practically and mentally separated from regular employment earnings used to support everyday expenses. Many LMI households receive large tax refunds as a result of two federal tax credits: the fully refundable earned income tax credit (EITC)9 and the partially refundable child tax credit (CTC).10 The average refundable EITC beneﬁt in 2001 was $1,840 for the 10 million EITC claimants with adjusted gross income (AGI) of less than $15,000, and $1,640 for the 6.5 million families with AGI between $15,000 and $30,000 (Internal Revenue Service 2001). Data on the size of the refundable portion of the CTC are not available by income, but outlays on the refundable portion for all income groups were approximately $980 million in 2001 and $5.8 billion in 2003 (Carasso and Steuerle 2003). When all refund sources are considered (i.e., EITC, CTC, other refundable tax credits, and over-withholding), in 2001, more than 30 million ﬁlers with AGIs below $15,000 received refunds worth $1,176 on average, and 24.5 million ﬁlers with AGIs between $15,000 and $30,000 received $1,710 on average. In the aggregate, these LMI families received more than $78 billion in federal refunds (Internal Revenue Service 2001). Just as the tax system subsidizes the saving of wealthy Americans through preferential tax treatment for investment income, large tax refunds can help other Americans to save. Not only does the tax system distribute substantial funds, but it does so in a way that—in part—facilitates savings. Neoclassical economic theory assumes that funds are fungible, but behavioral economists have documented the tendency for people to mentally group different

Splitting Tax Refunds and Building Savings

117

monies into different pots. For example, one might write checks for day-to-day expenses from one account and for big ticket items from another. Or one might designate income from a primary job for living expenses and income from a secondary job as savings. This notion of mental accounting relates to the behavioral hypothesis that people tend to view ‘‘irregular’’ income differently than wage and salary income, especially when the irregular in-ﬂows are large (see, Thaler and Shefrin 1988). One speciﬁc prediction derived from behavioral economic theory is that people who receive sizeable tax refunds will consider saving at least some portion of their refunds or using refunds to make special purchases rather than ﬁnancing routine expenses. Some evidence supports this hypothesis. Smeeding et al. (2000) found that 33 percent of a sample of 650 EITC recipients planned to save at least a portion of their tax refunds; 16 percent planned to purchase a car, repair a car, or make car payments; 13 percent planned to purchase furniture or household appliances; 10 percent planned to pay educational expenses; and 5 percent planned to purchase homes or move. Consumer Expenditure Survey data suggest that families often use tax refunds to purchase consumer durables, such as furniture and vehicles (Barrow and McGranahan 2000). Thaler (1994) recognized the savings opportunity presented by tax refunds and argued for an additional credit for ﬁlers directly depositing refunds into individual retirement accounts (IRAs). Evidence regarding the use of tax refunds has also led volunteer and for-proﬁt tax preparers to encourage refund recipients to save their refunds. Several free Volunteer Income Tax Assistance (VITA) tax preparation sites, locations that provide free income tax preparation services to low-income ﬁlers, have allowed tax ﬁlers to open savings accounts on-site and directly deposit their tax refunds into these accounts.11 These programs usually involve a partnership between a tax preparer (commercial or volunteer) and a ﬁnancial institution. Table 4.1 lists some of these programs. The most complete evaluation to date of this type of service comes from a 2000 study of a program sponsored by Shorebank and the Center for Economic Progress. Over 400 Chicago tax ﬁlers were invited to open low-cost Shorebank savings accounts with their tax refunds, and about 20 percent did so. Twoﬁfths of account-openers depleted their accounts very rapidly, but 14 percent maintained balances and perhaps added additional savings to the accounts (Beverly, Tescher, Romich, and Marzahl 2002). In the private sector, some H&R Block ofﬁces have piloted a program to encouraged clients to open savings accounts on-site as a means

Offered accounts on site (waived ChexSystems, no fee)

Offered accounts on site (waived ChexSystems, no fee for 1 year) Offered accounts

Offered accounts (did not waive ChexSystems, no fee) Offered checking accounts on site

Offered savings accounts through First Accounts program

Ithaca, NY Alternatives Credit Unionb

Wilmington, DE Nehemiah Gateway CDC Riverside, CA Riverside Family Asset Building Program

Seattle, WA United Way of King County Denver, CO The Piton Foundation

New Orleans, LA Central City Asset Building Coalition

Boston, MA EITC Coalition

New York, NY CFRC and ORCA

State Wide, IL Tax Counseling Project

Offered accounts on site (did not waive ChexSystems, no fee) Hosted large general ‘‘asset building’’ event at start of season

Baltimore, MD Baltimore CASH Campaign

Offered bank accounts

Offered bank accounts

Offered savings accounts

Opening Savings Accounts

Table 4.1 Free Tax Preparation Sites and Asset Buildinga

Provided information on IDAs

Provided information on IDAs

Planning to link tax clients with ‘‘Prosperity Club’’ program (ﬁnancial education and IDAs)

Provided information and program applications

Some sites offered IDAs

Opening IDAs

Referred to ﬁnancial literacy classes

Planning to link tax clients with ‘‘Prosperity Club’’ program (ﬁnancial education and IDAs)

Provided referrals to ﬁnancial education

Planning to expand ﬁnancial education on site

Financial Education

118 Beverly, Schneider, and Tufano

Offered savings accounts

Offered savings accounts (waived ChexSystems, no fees for 1 year)

Provided referrals to partner agencies

Provided information and referrals on IDAs

Opening IDAs

Referred to ﬁnancial education classes at partner agencies

Enrolled individuals in ‘‘Get Checking’’ classes

Financial Education

Phone interviews were conducted with practitioners at community organizations, foundations, and ﬁnancial institutions involved in linking asset building and free tax preparation. Respondents were identiﬁed through a listing of Annie E. Casey grant recipients and through mutual referral. The interviews took place in April and May 2004. Information is presented for each community organization and is organized into three general program categories. b Additionally, Alternatives offered refund loans designed to help clients to build credit. A client could arrange to have her refund sent to the credit union and in exchange received a loan in the amount of her anticipated refund one or two days after ﬁling. Alternatives then held the refund as collateral and allowed the client to pay back the loan in regular payments, thus building credit. There was very little risk for Alternatives because it held refunds as collateral.

a

Milwaukee, WI Milwaukee EITC/ Asset Building Project Washington, DC DC Cash Coalition

Opening Savings Accounts

Splitting Tax Refunds and Building Savings 119

120

Beverly, Schneider, and Tufano

of accumulating emergency savings (Brookings Institution 2004). In the 2003 tax season, Block opened approximately 400 accounts with an average opening balance of $870. Only 2 percent of Block clients at the ofﬁces offering the service opened accounts, but this take-up rate was depressed by rigorous account opening requirements based on a credit check (Tufano and Schneider 2004). Block also offered refund recipients on-site opening of IRAs. Over the three tax seasons from 2001 to 2004, the company opened more than 440,000 Express IRA accounts. In 2004, the average balance on these accounts was $575 (Tufano and Schneider 2004). As we mentioned earlier, a recent study conducted in St. Louis in conjunction with H&R Block used the Express IRA product to test whether it would be possible to increase retirement savings at tax time (Duﬂo et al. 2005). Comparing this experiment with ours suggests some interesting hypotheses, which we will discuss later. Behavioral economists posit that people often make irrational decisions about ﬁnancial matters, due in large part to spending temptations (e.g., Shefrin and Thaler 1992; Thaler 1994, 2000). However, it is possible to deliberately modify incentives and constraints to avoid or overcome these temptations. Theory and empirical evidence suggests that helping people enforce self-discipline by precommiting money to savings is effective. For example, some people arrange to have retirement contributions automatically deducted from their paychecks. One novel program has increased savings by having workers precommit to save future raises.12 Another recent experiment in the Philippines allowed participants to make deposits into a restricted withdrawal account, not in exchange for higher interest rates, but merely as a device to commit to savings. Participants had signiﬁcantly higher balances over time than members of a control (Ashraf, Karlan, and Yin 2006). Some people use mental or even physical accounting strategies to separate spending money from savings, and evidence suggests that these psychological and behavioral strategies help people resist the urge to withdraw savings (Beverly, Moore, and Schreiner 2003). 2.3 Splitting Refunds and Building Assets If people undersave because they have trouble resisting spending temptations and if mental accounting helps people set aside money for savings, then refund splitting—physically separating funds designated for saving from funds designated for spending—may be a valuable tool to encourage saving. Currently, the IRS will send a refund to only one account, and most poor families cannot save all of their refund. Thus, whether the money is sent via a single paper check or directly

Splitting Tax Refunds and Building Savings

121

deposited to a single account, the would-be saver must receive the entire refund, make a decision about saving, and then execute the saving plan while spending part of the money. Any program that deposits all money in one place and requires recipients to save some leaves them open to spending temptations. Refund recipients may mentally earmark part of their refunds for savings, like a worker planning to save part of her paycheck. But unlike the worker who can use a variety of programs to precommit some earnings to savings, the refund recipient can only rely on her best intentions, saving once the refund check or deposit has been received. If it were possible to split the refund and send it to multiple destinations at the time of ﬁling a tax return, then a family could mentally and physically separate saving and spending money and make saving automatic, thereby reducing the mental energy required to save. The intention to save would be further reinforced if the savings portion were directed to a savings-oriented ﬁnancial product, especially one that restricts withdrawals to some degree (although this may discourage some from saving at all). Some higher-income families have access to refund splitting. Vanguard, the second largest mutual fund company in the United States, allows investors to split their refunds among as many as four existing Vanguard accounts (including money market accounts with checkwriting privileges that can be used as transaction accounts).13 In 2004, Vanguard was the only company out of the ten largest banks and ten largest mutual funds with a formal splitting service,14 but some customers with some brokerage accounts in other ﬁrms may be able to have their brokers automatically split refunds across multiple investments. LMI families, however, are not likely customers of Vanguard (whose funds typically require a minimum investment of at least $3,000)15 nor of brokerage ﬁrms. LMI families do have access to some refund splitting, albeit of a different form. Clients who take out refund anticipation loans (RALs) split their refunds, sending some to pay the tax preparer and receiving the remainder. RAL splitting does not build assets, however, but instead satisﬁes the need for very immediate spending. In addition, clients opting to purchase H&R Block’s Express IRA product at the time of tax preparation have the option of splitting their refunds, directing some to the newly opened IRA and receiving some as a direct deposit or as a paper check (Tufano and Schneider 2005). However, splitting tied explicitly to retirement savings may be less appealing to LMI families with shorter-term savings goals.

122

Beverly, Schneider, and Tufano

On logical grounds, we hypothesized that a simple splitting program, with an option to open a new savings account, offered to LMI families at the time of tax ﬁling would facilitate asset building among LMI families. We hypothesized that (1) some LMI refund recipients would choose to precommit part of their refunds; (2) splitting and precommitting would increase immediate savings; and (3) precommitting and splitting, by leveraging mental accounting, would perhaps even help people resist temptations to deviate from planned savings goals. Our collaborative research team was formed to gather evidence to support or reject these hypotheses. 3.

The 2003 Refunds to Assets Program

During the 2003 tax season, the Community Action Project of Tulsa County (CAPTC) partnered with D2D Fund (D2D) to deliver the Refunds to Assets (R2A) program. CAPTC is a large community action agency based in Tulsa, Oklahoma, providing a variety of services, including housing assistance, child care, health services, emergency food and ﬁnancial assistance, and individual development accounts (IDAs).16 CAPTC also provides free tax preparation to lowincome households. In 2002, volunteers completed about 14,300 federal returns. D2D Fund (www.d2dfund.org) is a non-proﬁt organization dedicated to improving the delivery of ﬁnancial services to low-income families and focused on the development and testing of innovative ﬁnancial products and services.17 Clients at two of CAPTC’s free tax preparation sites were invited to split their refunds, sending part to a savings account via direct deposit and arranging to receive the rest by mail as a check. We enlisted Bank of Oklahoma (BOk) as a partner so that participants could open new BOk savings accounts without leaving the tax preparation site.18 To be eligible, clients had to be at least 18 years old, have an expected federal tax refund, and agree to a set of legal authorizations.19 In addition, clients wishing to open a savings account with BOk had to pass a credit-check system used by the bank (ChexSystems) and have a valid driver’s license or passport. The pilot program was offered in two time periods during the tax season, during the ﬁrst two weeks of February (the peak season for ﬁling by those expecting refunds) and for two weeks in March. Recruitment and enrollment occurred on-site, while individuals were waiting to meet with tax preparers.20 CAPTC staff and/or a member of our re-

Splitting Tax Refunds and Building Savings

123

search team approached groups of CAPTC clients who were waiting to meet with tax preparers. They used a prepared script to describe R2A, emphasizing the program as a tool to split refunds and to encourage saving. Next, in one-on-one conversations, staff explained legal and research authorizations and collected preliminary information from those who did and did not want to participate. Individuals wanting to open BOk accounts completed a short form. Generally, at this point, potential participants began their tax preparation appointments while program staff opened accounts for those who requested them. Near the end of the tax preparation appointment, interested clients completed the R2A enrollment process and gave detailed written instructions for how their refunds should be divided. Although the script encouraged people to split their refunds, individuals who wanted to open new BOk accounts and deposit their entire refunds into these accounts were allowed to do so. Depending on ﬁling date and on IRS processing times, the savings portion was directly deposited into the new BOk account or the participant’s existing account about ten to fourteen days after the date of tax ﬁling, and the participant received the remainder as a check shortly thereafter. We collected data from three different groups: 1. Takers—137 individuals at the two CAPTC test sites who expected to receive refunds and who attempted to sign up for the program. These were subsequently divided into two subgroups: seventy-nine who ended up participating (participants), and ﬁfty-eight who sought to participate but were unable to usually because they did not pass the BOk credit screen (foiled participants). 2. Decliners—Approximately 379 individuals at the two test sites who expected refunds but were not interested in participating in R2A. 3. Comparison group—ﬁfty-three individuals at a CAPTC tax preparation site where R2A was not available who said that they would have participated in R2A had the program been offered. By design, this group had similar motives as the R2A participants, although they did not have the beneﬁt of the program. We collected several different sources of information (see table 4.2). Takers, decliners, and members of the comparison group completed a short written baseline survey during the enrollment process. For takers, this survey included questions about demographic characteristics, current savings, and planned refund uses. For decliners, this survey asked about demographic characteristics and reasons for

124

Beverly, Schneider, and Tufano

Table 4.2 Response Rates per Data Source by Research Samplea

Data Source Takers Participantsb Foiled

Baseline Survey

Taxwise Summary Page

Splitting Instructions

Follow-up Survey

99%

95%

NA

NA

100%

99%

83%

55%

97%

89%

NA

NA

Decliners

63%

65%

NA

NA

Comparison group

87%

NA

NA

49%

a

Information was collected from a number of sources for individuals approached on site by R2A program staff. All individuals were asked to complete a baseline survey, and information from a Taxwise summary sheet was collected for all decliners and takers. Information was also collected from the splitting instructions issued by those participants opting to split their refunds. Follow-up telephone surveys were conducted 3–5 months after refund receipt with participants and members of the comparison group. Finally, BOk provided account data for select participants. Data collected 2/9–2/21 and 3/15–3/ 26 of 2004 in Tulsa, Oklahoma, and 5/30–8/12 by telephone from Boston, Massachusetts. b Calculated as the number of participants with data divided by the total number of participants signing research consent forms (n ¼ 75). Four participants who did not complete research consent forms are excluded.

declining. For the comparison group, the baseline survey asked about demographic and ﬁnancial characteristics as well as planned refund uses.21 Data on anticipated refund amount and adjusted gross income (AGI) were collected for takers and decliners from a summary information page generated by the Taxwise tax preparation software. Data on how R2A participants allocated their refunds between savings deposits and checks were taken from the refund-splitting instructions issued on site. In addition to these baseline data, R2A participants and members of the comparison group were asked to complete a ﬁfteen-minute phone survey three to ﬁve months after they received their refunds. Fifty-ﬁve percent of participants and 49 percent of comparison group members completed this follow-up survey. In total sixty-three respondents were successfully contacted, forty-one participants and twenty-two comparison group members. The survey included questions on actual refund use, satisfaction with R2A (for participants), and interest in similar services and products. Before we discuss the research ﬁndings, we acknowledge some important research limitations: we studied a relatively small number of individuals in a single city under particular research conditions,

Splitting Tax Refunds and Building Savings

125

and therefore we cannot be conﬁdent that our results are generalizable. However, given the paucity of research on this topic, even this small study may be useful in understanding the potential of refund splitting. One can think of our results as akin to a test market approach to research, as used by business marketers, where consumers are exposed to a product or service to gauge their likely reception. We organize our discussion of the research ﬁndings in the next section around ﬁve topics: 1. Is there demand for refund splitting? What is the apparent take-up rate of refund splitting, and what are structural impediments to increasing take-up? 2. Who participated? What are the characteristics of participants? 3. Why did people choose to participate? What goals were participants trying to reach? 4. What was the immediate impact of the program on savings? 5. After a few months, did participants have better saving outcomes than the comparison group? We defer a discussion of the sixth set of ﬁndings, regarding participants’ thoughts about alternative splitting options, to the ﬁnal section of the paper, where we discuss the implications of our research. 4.

Baseline Research Findings from the R2A Program

4.1 Take-Up and Participation Rates Table 4.3 summarizes information regarding the take-up rate for the R2A program at the two Tulsa sites. In brief, about 27 percent of refund recipients sought to participate in the R2A program (takers); 15 percent were able to participate (participants) and the remainder were ‘‘foiled’’ because they did not meet various eligibility standards, as described below. These ﬁgures represent our estimates of the demand for the R2A version of an asset-building strategy leveraging tax refunds. As we describe below, a more conservative estimate of the potential and actual demand for splitting alone would be 21 percent and 12 percent, respectively. To determine potential demand, we divided the number of people who enrolled or attempted to enroll by the number of refund recipients approached by the staff. R2A program staff approached 556 individuals about enrolling in R2A during the four weeks the service was offered. Of these, we estimate that 516 anticipated a refund; this

126

Beverly, Schneider, and Tufano

Table 4.3 Enrollment Outcomesa Outcome

Number

Percentage

Take-up rate—programb Take-up rate—splittingc

137 110

27 21

98

19

79

15

Participation rate—splittingf

62

12

Participation rate—new accountg

55

11

58

11

379

73

Take-up rate—new accountd Participation rate—programe

Disqualiﬁcation rate—programh Declinedi

a Individuals who were approached at the tax preparation site (n ¼ 516) have been grouped into several categories: those who enrolled in the program, those who wanted to enroll in the program but could not, and those who declined to enroll. The ﬁrst two of these groups are combined to form the group ‘‘Interested in Program.’’ These individuals are referred to as ‘‘takers.’’ The take-up rate, deﬁned as the percentage of those approached on site expecting a refund who were interested in the program, is also presented. Take-up rates are estimated for the program as a whole (splitting and account opening), for splitting, and for account opening. Participation rate is deﬁned as the percentage of those approached on site expecting a refund who successfully enrolled in the program. Participation rates are estimated for the program as a while, for splitting, and for account opening. Data collected 2/9–2/21 and 3/15–3/26 of 2004 in Tulsa, Oklahoma. b This group includes those who actually enrolled in the program (participants) and those who were interested in enrolling but were ineligible (foiled participants). c Take-up rate for splitters is calculated by multiplying the percentage of participants splitting their refunds (83%) by the number of foiled participants (n ¼ 58), adding that number to the number of successful splitters (n ¼ 62), and then dividing that total by the number of individuals approached who were expecting a refund (n ¼ 512 and excludes four participants who did not sign research consent forms). d Take-up rate for new accounts is calculated following the same method as described above but for new accounts instead of splitters. e This group includes individuals who successfully enrolled in the program. f Participation rate for splitting is calculated as the number of splitters (n ¼ 62) divided by the number of individuals approached expecting refunds (n ¼ 512). g Participation rate for new accounts is calculated as the number of individuals opening new accounts (n ¼ 55) divided by the number of individuals approached expecting refunds (n ¼ 512). h This group includes individuals who wanted to enroll in the program but were not eligible. See table 4.4 in this paper. i This group includes those who were approached by CAPTC staff about enrolling, but declined to participate. See table 4.6 in this paper for a more detailed explanation.

Splitting Tax Refunds and Building Savings

127

Table 4.4 Enrollment Outcomes for Takersa Outcome

Number

Percentage of Takers

Enrolled

79

58

Wanted to enroll

58

42

ChexSystemsb

36

26

Declined to waitc

6

4

Refund too small in the endd

6

4

Lacked proper identiﬁcatione

4

3

Underagef Did not have existing account number g

3 2

2 1

Paper return error h

1

1

a

Some individuals who expressed interest in the program were unable to enroll. The percentage of all takers who either enrolled or were disqualiﬁed for speciﬁc reasons is shown in the table. Data collected 2/9–2/21 and 3/15–3/26 of 2004 in Tulsa, Oklahoma. b This category includes the individuals denied accounts, and thus participation in the program, by ChexSystems. c Waiting time to complete enrollment in the program during very busy periods could be as long as 15–20 minutes; some clients declined to wait. d Interested individuals began the enrollment process before learning the value of their refund; some individuals ultimately decided that their refunds were too small to make splitting worthwhile. e In order to open an account, BOk required a valid driver’s license issued at least 3 months prior to the account-opening date. f In order to open a non-custodial account, BOk required that the account-holder be at least 18 years old. g In order to split a refund between a check and an existing account, individuals needed to know the account number of their existing savings account. h Nearly all individuals enrolling in the program ﬁled electronically. Although it would have been possible to enroll in R2A even if paper-ﬁling, one individual was mistakenly turned away for this reason.

number serves as the denominator for take-up and participation rates and equals the takers plus decliners. Of these 516, 137 wanted to enroll in the program, producing an estimated take-up rate of 27 percent. Fifty-eight of the would-be enrollees (11 percent of the 516 eligible individuals) were turned away. Excluding these individuals from the numerator yields a participation rate of 15 percent, considerably above the 2 percent rate experienced by H&R Block, mentioned above.22 Table 4.4 gives reasons for disqualiﬁcation. While some takers were foiled by lack of identiﬁcation or age requirements, the most severe constraint was that individuals who wanted to open a BOk account had to pass BOk’s ChexSystems standard. ChexSystems is a credit-

128

Beverly, Schneider, and Tufano

screening device used by approximately 90 percent of bank branches nationwide (Quinn 2001). Subscribing banks use the product in two ways: as a reporting mechanism for current clients who have had problems using their accounts, often in the form of overdrafts or bounced checks, and as a data source to check the ﬁnancial histories of potential new account-holders. Our team failed to ask BOk to waive ChexSystems requirements, and as a result, BOk’s default rules were in place. The bank would not open a savings account for any individual with a ChexSystems record.23 This policy disqualiﬁed 26 percent of the individuals who wanted to enroll in the program. Finally, we consider the actual demand for splitting alone. The R2A program allowed participants to split their refunds and to open new accounts. Among participants, 56 percent split to a new account, 27 percent split to an existing account, and 17 percent simply opened a new account and deposited all of their federal refund in it (even though this was not the intent of the program). If we exclude participants who did not use splitting services, then the potential and actual take-up rates are reduced to 21 percent and 12 percent, respectively. All of the take-up and participation rates discussed thus far assume that participants split their refunds because they perceived value in dividing their refunds into portions for spending and saving and in being able to precommit to saving. However, it is possible that participants split for other reasons, and if this is the case, the take-up and participation rates above may overestimate the demand for splitting. First, participants may have split because they believed it was a prerequisite for account opening. The R2A outreach emphasized splitting, and some may have split simply to gain access to the BOk account. This interpretation has some support: in response to an open-ended question, 15 percent of the thirty-three splitters who completed the follow-up survey reported that opening an account was their primary reason for splitting. (Follow-up methods, response rate, and data are discussed in detail below.) On the other hand, the 17 percent who chose not to split their refunds recognized that splitting was not required. Second, participants who opened new accounts and split their refunds may have been interested in both program features but may have been more interested in account opening. In the follow-up survey, more than 70 percent of those who split into new accounts ðn ¼ 22Þ reported that opening a new account was more important to them than splitting their refund. However, the fact these participants valued

Splitting Tax Refunds and Building Savings

129

account opening over splitting does not necessarily mean that their interest in splitting was insigniﬁcant. In fact, in the follow-up survey, 94 percent of participants who said that account opening was more important than splitting said that they planned to split again next year. Third, when asked why they split, a substantial share of splitters gave reasons that did not obviously and directly relate to splitting or physical/mental accounting. Fourteen percent of splitters said that they did so because of an interest in trying something new, 6 percent said they split to receive their refund faster, and 14 percent gave miscellaneous reasons (e.g., ‘‘because of the business’’ or ‘‘it just sounded like a good thing to do’’). Another 18 percent of respondents gave a response that spoke broadly about saving without speciﬁcally referring to splitting (e.g., ‘‘to save for car insurance’’ or ‘‘just to have extra cash in case I ever run out’’). These responses may indicate that people split because they believed splitting would help them save. However, if a participant believed that having a new account would help him save and that individual split simply to open an account, then this type of response could reveal perceived beneﬁts of account-opening rather than perceived beneﬁts of splitting. To the extent that R2A participants split simply to gain other beneﬁts (most notably access to a savings account), even our estimates of 21 percent (potential take-up) and 12 percent (actual take-up) for splitting alone are biased upward. 4.2 Characteristics of Subgroups Table 4.5 reports demographic characteristics of the various subgroups, including participants, foiled participants, takers, and decliners. Where possible, it also describes residents of Tulsa County. We begin by summarizing characteristics of participants. Participants had an average age of 35, were predominantly female, and were predominantly African-American or Caucasian. Their federal tax refunds averaged $1,381. Over half of participants held full-time jobs, and more than three-quarters worked at least part-time. However, 12 percent of respondents were unemployed and currently looking for work, a ﬁgure that was twice the national unemployment rate The average adjusted gross income (AGI) was $12,297. Overall, participants might be described as the working poor. The last four columns of table 4.5 indicate whether participants are different from foiled participants and whether takers are different from decliners.24 Participants differ from foiled participants in three key

$9,264

$10,526

Standard deviation

Median

$1,446

$648 52%

27%

45%

31%

Standard deviation

Median Children in householdc

Receives public assistanced

Has health insurance

Owns home

57%

43%

Female

Male

Gender

$1,381

Mean

Federal refund

$12,297

Mean

32

13

Standard deviation

Median Incomeb

35

All Participants (n ¼ 75)

Mean

Age

Demographics

30%

70%

23%

36%

33%

$456 55%

$1,040

$835

$5,602

$11,529

$9,614

26

13

30

All Foiled (n ¼ 58)

38%

62%

28%

42%

29%

$518 53%

$1,316

$1,153

$7,850

$10,312

$11,177

30

13

33

All Takers (n ¼ 133)

35%

65%

33%

57%

30%

$746 44%

$1,412

$1,261

$12,328

$10,136

$13,927

39

16

40

All Decliners (n ¼ 303)

35

48%

52%

62%

86%

—

— 35%

—

—

$38,213

—

$51,756

—

36

Tulsa County (n ¼ 563,299)

Table 4.5 Demographic Characteristics of R2A Participants, Foiled Participants, Takers, Decliners, and Residents of Tulsa Countya

0.15

0.42

0.37

0.47

0.70

0.02

0.15

0.04

P0F (p-Value)

0.62

0.32

<0.01

0.89

0.07

0.47

0.01

<0.01

D0T (p-Value)

130 Beverly, Schneider, and Tufano

16% 12%

8%

4%

Working part-time Not working, looking for work

Not working, student

Not working, otherf

8% 8%

Some college

Associate’s degree Bachelor’s degree

50%

39%

3%

5%

3%

African-American

Caucasian

Native American

Hispanic

Other h

Race/ethnicity

37%

33%

High school diploma or GED

14%

Less than high school diploma

Educationg

60%

All Participants (n ¼ 75)

Working full-time

Employmente

Demographics

3%

18%

0%

21%

59%

3% 5%

32%

50%

11%

13%

5%

21% 21%

39%

All Foiled (n ¼ 58)

3%

10%

2%

33%

53%

6% 7%

32%

41%

13%

7%

7%

18% 15%

53%

All Takers (n ¼ 133)

4%

9%

2%

32%

52%

10% 12%

28%

41%

10%

13%

4%

19% 15%

49%

All Decliners (n ¼ 303)

6%

6%

5%

73%

11%

7% 27%

25%

27%

15%

33%

64% 3%

Tulsa County (n ¼ 563,299)

0.08

0.60

0.14

P0F (p-Value)

(continued)

0.95

0.50

0.35

D0T (p-Value)

Splitting Tax Refunds and Building Savings 131

4%

Widowed

3%

15%

13%

70%

All Foiled (n ¼ 58)

4%

22%

16%

59%

All Takers (n ¼ 133)

9%

22%

22%

47%

All Decliners (n ¼ 303)

6%

55%

14%

25%

Tulsa County (n ¼ 563,299) 0.26

P0F (p-Value)

0.06

D0T (p-Value)

Note: Sample sizes vary across variables due to missing data. a Individuals approached at the R2A test site who expected to receive a refund have been classiﬁed into two groups. ‘‘Takers’’ includes those who either enrolled in the program or expressed interest but were blocked from participation. ‘‘Decliners’’ includes those who declined to participate. The group ‘‘takers’’ is divided into two sub-groups: ‘‘participants,’’ those who successfully enrolled in the program, and ‘‘foiled,’’ those who were interested in the program but were unable to enroll. Data for residents of Tulsa County are also presented. The table reports several ﬁnancial and demographic characteristics for each group. The results of signiﬁcance tests comparing the characteristics of participants and foiled participants and takers and decliners are shown in the right-hand columns. Data collected 2/9–2/21 and 3/15–3/26 of 2004 in Tulsa, Oklahoma. Data on Tulsa County are drawn from the US Census (2000). b Annual gross income (AGI) as reported on summary TaxWise page. c Percentage of participants in each group reporting children under the age of 18 in household. d Percentage of participants in each group reporting currently receiving TANF, SSI, or food stamps. e Employment data for Tulsa County residents are available only for the general categories of ‘‘employed,’’ ‘‘unemployed,’’ and ‘‘not in the labor force.’’ f Percentage of participants in each group reporting being retired, a homemaker, or other non-student categories. g For Tulsa County residents, education statistics include those over the age of 25 (U.S. Census Bureau 2000). h Percentage of participants in each group reporting their ethnicity as Asian, Hawaiian, or Paciﬁc Islander, or ‘‘other.’’ i For Tulsa County residents, statistics on marital status include those over the age of 15.

27%

16%

Married

53%

Marital statusi Single, never married

Separated or divorced

All Participants (n ¼ 75)

Demographics

Table 4.5 (continued)

132 Beverly, Schneider, and Tufano

Splitting Tax Refunds and Building Savings

133

ways. Foiled participants were younger on average and had smaller average expected refunds than participants. They were also less likely to be white and more likely to be Hispanic and African-American. Compared to decliners, takers (foiled and actual participants) tended to be younger and to have less income. They were more likely to have children, more likely to have never been married, and less likely to have health insurance. The mean AGI for takers was $11,177, compared to $13,927 for decliners. In short, those who wanted to participate were probably among the least ﬁnancially stable households. This evidence—especially regarding income—is broadly consistent with the early evidence on IDAs, where one study has found that saving rates were highest among the lowest-income participants.25 If the ability to open a bank account was part of R2A’s appeal (discussed in detail below), then this ﬁnding may reveal unmet demand for low-fee bank accounts by low-income working households. 4.3 Stated Reasons for Declining and Participating In our baseline survey, we asked decliners why they were not interested in R2A. The question was closed-ended with an open-ended ‘‘other’’ option. Explanations for disinterest in R2A fall into two broad categories: ‘‘I don’t need the service’’ or ‘‘I don’t like the service’’ (see table 4.6). Ninety-four percent of respondents gave an answer that we classiﬁed in the ﬁrst category. Forty-four percent of decliners said that they had already decided how to use their refund, and another 24 percent said that they planned to spend all of their refunds. Looking forward, we believe that some of those who said that they did not need the service might use it under other circumstances. Tulsa’s economy was weak in early 2004,26 and our conversations with tax ﬁlers revealed many cases of unemployment and unusual ﬁnancial difﬁculties that prevented them from saving at all. In better economic times, some of these decliners might choose to split. A much smaller percentage (9 percent) of decliners gave responses that we classiﬁed as ‘‘don’t like the service.’’ Some (4 percent) sought to have matching funds linked with the service or to deposit in an investment product dedicated to a speciﬁc purpose,27 and others (2 percent) expressed concerns about the security and reliability of the processing system used to split refunds. Both of these concerns could be addressed in the future by offering additional savings choices or by implementing the system more simply, as would happen if the IRS were to enable refund recipients to split their refunds. Finally, 2

134

Beverly, Schneider, and Tufano

Table 4.6 Number and Percentage of Decliners Naming Reasons for Declininga

‘‘Don’t need the service’’ Already decided how to use refund

Percentage

285 132

94 44

Plan to spend entire refund

74

24

Refund too small to bother

37

12

Don’t need help saving or managing money

28

9

Already have accountb

11

4

3

1

Began process but sent all to savingsc ‘‘Don’t like the service’’ Interested if there was a match or dedicated used Moving awaye Trust/privacy issuesf Other a

Number

26

9

12 7

4 2

5

2

34

11

Individuals who declined to participate in R2A, despite expecting a refund, gave a number of reasons for choosing not to enroll. These speciﬁc reasons have been grouped into the broad categories of ‘‘don’t need the service’’ and ‘‘don’t like the service.’’ Individuals were permitted to give more than one response, so percentages sum to more than one 100%. Data collected 2/9–2/21 and 3/15–3/26 of 2004 in Tulsa, Oklahoma. b This group includes individuals who stated that they had an existing savings account with another bank and thus were not interested in opening a new savings account with BOk. Presumably, these individuals did not see a value in the splitting service alone. c Several individuals began the enrollment process, but after talking with CAPTC staff, they decided to direct deposit their entire refunds into existing savings accounts. d Individuals in this group stated that they would have been interested in the splitting program if additional incentives, such as a match or additional saving options, were available. e A number of individuals said that they did not want to participate because they were moving away from the Tulsa area. Several of these individuals did not want to open a new bank account with BOk, since they were leaving the area, and several noted that they were planning to close existing accounts and did not want to make additional deposits to those accounts. These responses essentially address portability concerns. f A few individuals decided not to participate based on a distrust of banking institutions generally or a desire to limit the number of parties involved with the processing of their refund to ensure their privacy.

Splitting Tax Refunds and Building Savings

135

percent said that they did not want to participate because they were moving away. Therefore, they did not want to open new accounts or make additional deposits into soon-to-be-closed accounts. In our follow-up survey, we asked participants why they chose to participate in R2A. As noted above, the most common reason respondents gave for splitting their refunds was to separate their refunds into spending and saving portions (27 percent). Participants also reported that they chose to split in order to save (18 percent), to avoid spending all of the refund (12 percent), to try something new (15 percent), and ﬁnally because they wanted to open an account (15 percent). Respondents who opened new accounts listed a desire to save (29 percent) as the leading reason for account opening. One-quarter of respondents reported that they had wanted a savings account, and 16 percent said that they opened an account simply because they were offered one. The BOk brand had some currency among this population, with 19 percent citing it as the primary reason they opened an account, the same share that chose to open an account because of the terms offered. When all account openers were asked about these speciﬁc account features, the largest share of respondents reported that the convenience of opening an account on site was most important to them (44 percent). The waiver of minimum balance requirements (13 percent) and fees (13 percent), as well as the opportunity to earn interest on savings (13 percent) also attracted many participants. Fewer participants were drawn in by the faster processing time for refunds deposited into bank accounts (9 percent) or using their tax refund as the opening balance (6 percent). The motives of participants may also be understood through the lens of their planned refund uses and, more concretely, their immediate savings goals. In the baseline survey, we asked participants if they planned to spend, save, or repay debt with their refunds and then used open-ended questions to probe for more speciﬁc plans in each of these categories. We grouped these open-ended comments into various categories. For example, people could report that they planned to use refunds to pay rent (spending), to make mortgage payments (debt repayment), or to save for a future home (savings). We classiﬁed all three as housing-related refund uses, but only the last would be classiﬁed as a savings goal. The ﬁrst three columns of table 4.7 summarize planned refund uses for participants. Savings uses made up the largest single grouping of planned uses. Eighty-one percent of participants planned to save part

136

Beverly, Schneider, and Tufano

Table 4.7 Percentage of Participants and Comparison Group Members Reporting Types of Planned Refund Use by Method of Use a Participants (n ¼ 69)

Comparison (n ¼ 36)

Save

Spend

Repay

Save

Spend

Repay

Any

81%

57%

45%

56%

58%

56%

General saving

16%

0%

0%

0%

0%

0%

Emergency saving

13%

0%

0%

6%

0%

0%

Housingb

15%

9%

1%

13%

6%

0%

Carc

13%

12%

4%

8%

11%

6%

4%

1%

1%

0%

8%

0%

Educationd Business

1%

1%

0%

0%

0%

0%

Retirement Children

5% 4%

0% 4%

0% 1%

2% 0%

0% 2%

0% 0%

Special eventse

4%

1%

0%

6%

0%

0%

Bills

3%

19%

13%

0%

2%

2%

Medical

1%

1%

3%

0%

0%

9%

Durables

1%

4%

1%

0%

4%

2%

Moving expenses

1%

4%

0%

0%

2%

0%

Living expensesf

1%

0%

1%

0%

8%

0%

Miscellaneous Credit cards

7% 0%

1% 0%

3% 12%

0% 0%

2% 0%

4% 13%

Utilities

0%

0%

5%

0%

0%

2%

Loans

0%

0%

3%

0%

0%

2%

Clothes

0%

7%

0%

0%

0%

0%

Taxes

0%

0%

0%

0%

0%

6%

Family transfersg

0%

0%

0%

0%

0%

2%

a

Participants and comparison group members reported a wide variety of planned refund uses. Respondents were asked if they planned to save part of their refund, spend part, and/or repay debt with part. The ﬁrst row of the table shows the percentage of participants naming any planned refund uses in each of these broad categories. The percentages sum to more than 100% because participants could name multiple methods of use. Respondents also provided information on speciﬁc planned uses. The percentage of participants reporting each speciﬁc refund use by method is presented in rows 2 to 22. Data collected 2/9–2/21 and 3/15–3/26 of 2004 in Tulsa, Oklahoma. b Includes home purchase, home improvement, and rental housing uses. c Includes car purchase, car repair, and other car-related uses (such as insurance). d Includes tuition and school supplies. e Includes trips, gifts, and events such as weddings and funerals. f Includes food purchases and other general household expenses. g Includes ﬁnancial transactions between family members, including repaying of debt and making loans and gifts.

Splitting Tax Refunds and Building Savings

137

of their refund, with 57 percent and 45 percent planning spending or debt repayment, respectively. As expected in a splitting program, participants planned multiple and varied uses for their refunds. Specifically, 41 percent of participants planned both a savings and a spending use; 32 percent planned a savings and a debt-repayment use; 17 percent planned a spending and a debt-repayment use; and 16 percent planned a savings, a spending, and a debt-repayment use (results not shown). Because R2A aimed to encourage saving, we paid special attention to planned savings uses. The most common savings goal, named by 16 percent of participants, was ‘‘general’’ saving, including responses like ‘‘to build savings,’’ or ‘‘just to have money.’’ The next most common savings goals were saving for housing-related uses (15 percent, mostly for home purchase but also for home improvement), saving for emergencies (13 percent), and vehicle-related saving (13 percent, mostly car purchase and car repair). Retirement saving, a major focus of public policy, was mentioned by only 5 percent of participants. This may indicate that participants have more immediate savings needs; however, these same people might become interested in saving for retirement in future years. It is also possible that the product choice available in R2A—a simple savings account best suited to short-term savings rather than an IRA—did not appeal to ﬁlers planning to save for retirement. 4.4 Initial Savings Amounts The initial savings that the R2A program generated can be benchmarked in three ways: in dollars per participant, as a percentage of refund amount, and as a fraction of pre-program savings. By any of these measures, R2A participants initially saved a substantial amount. See table 4.8 for data on initial deposits to savings. The mean (median) participant directed $606 ($203) into savings. This represented 47 percent (39 percent) of her refund. Combining baseline survey data on existing savings with data on amount of refund sent to savings shows the immediate effect of R2A on household savings. On a per capita basis, savings increased from $276 before the receipt of refunds to $863 after refund receipt. This change represented a $587 per capita increase, 213 percent growth.28 Participants contain two interesting subgroups: one subgroup reporting some savings before the program and a second reporting no prior savings. Individuals with existing savings had larger refunds and

138

Beverly, Schneider, and Tufano

Table 4.8 Aggregate Initial Saving out of Refunda All Participants (n ¼ 75)

All Splitters (n ¼ 62)

All New Account (n ¼ 58)

Amount sent to savings Mean

$606

$579

$490

Standard deviation

$905

$883

$1,011

Median

$203

$200

$174

$3 $4,000

$15 $4,000

$3 $4,000

Minimum Maximum Amount sent to savings as a percentage of anticipated refund Meanb

47%

38%

46%

Standard deviation

33%

27%

36%

Median

39%

33%

35%

2% 100%

2% 98%

2% 100%

Minimum Maximum a

Data on total and individual measures of amount of refund sent to savings are presented below for all participants, all participants who split their refunds, and all participants who opened new accounts. Data collected 2/9–2/21 and 3/15–3/26 of 2004 in Tulsa, Oklahoma. b Calculated as the mean of individual savings as a percentage of refund.

directed a larger portion (mean ¼ 60 percent) of their refunds to savings accounts. In contrast, greenﬁeld savers—individuals without existing savings—received smaller refunds and directed less to savings (mean ¼ 42 percent). With smaller refunds and smaller shares directed to savings, greenﬁeld savers sent an average of $479 to savings versus $924 by existing savers. These values represent an inﬁnite and a 90 percent increase in savings for the two groups, respectively. In both cases, the increase in savings was large, at least initially. 5.

Follow-Up Results

In this section, we use available data to examine whether R2A seemed to help participants save part of their refunds and meet speciﬁc savings goals. As noted above, we conducted a follow-up telephone interview of R2A participants and comparison group members three to ﬁve months after refund receipt. Forty-one (55 percent) of the seventy-ﬁve participants, and twenty-two (42 percent) of the ﬁfty-three comparison

Splitting Tax Refunds and Building Savings

139

group members completed this interview.29 To assess comparability between these samples, panel A of table 4.9 shows demographic characteristics of the baseline and follow-up samples, for participants and comparison group members. Panel B of table 4.9 shows planned refund uses for these four samples. The follow-up participant sample is quite similar demographically to the baseline participant sample. The most notable differences are related to home ownership, employment, and race. Members of the follow-up sample were less likely than those in the baseline sample to own homes (25 percent versus 31 percent). They were more likely to work full-time (67 percent versus 60 percent) and to be white (44 percent versus 39 percent). Planned refund uses are also quite similar for these two groups, with one exception: the follow-up sample was less likely to name a planned spending use than the baseline sample (49 percent versus 57 percent). Differences between the follow-up sample of participants and the follow-up comparison sample are more pronounced. Comparison group members were much more likely to own homes (56 percent versus 25 percent) and to be female (88 percent versus 59 percent). Comparison group members also had higher incomes. All of these differences are statistically signiﬁcant. Other differences approach statistical signiﬁcance: comparison group members were more likely to be white (65 percent versus 44 percent), they had less education, and they had smaller anticipated refunds.30 (Actual refund amounts were quite similar for the two groups.) With the exception of education, the comparison sample appears to be more advantaged than the participant sample. If one assumes that the advantaged sample is more likely to save part of their refunds and meet speciﬁc savings goals, then demographic characteristics suggest that comparisons between the followup participant and comparison samples represent a conservative test of the impact of R2A. Data on planned refund use complicate matters, however. The follow-up comparison sample was less likely than the follow-up participant sample to name a planned savings use (63 percent versus 82 percent) and more likely to name spending and debt repayment uses. Differences in savings and repayment goals approach statistical signiﬁcance. If these observed differences are not due to chance, they may represent individual differences not captured by the demographic variables. Or they may suggest that R2A had an immediate impact on planned refund uses. The latter is consistent with the fact

140

Beverly, Schneider, and Tufano

Table 4.9 Demographic Characteristics and Planned Refund Uses for Participants and Comparison Group Members in the Baseline and Follow-up Samplesa Participants Baseline

Comparison Follow-up

Follow-up Baseline

PF 0 CF p-Value

0.62

Panel A: demographic characteristics Ageb Mean

35

36

37

No data

Standard deviation

13

12

17

No data

Median

32

34

41

No data

$0–$10,000

50%

51%

15%

30%

$10,001–$20,000 $20,001þ

24% 26%

22% 27%

50% 35%

32% 39%

Mean

$1,381

$1,362

$775

$958

Standard deviation

$1,446

$1,567

$769

$1,080

$648

$624

$500

$500

Income rangesc 0.02

Refund anticipatedd

Median

0.15

Refund received Mean

No data

$1,377

$1,299

No data

Standard deviation Median

No data No data

$1,536 $704

$1,208 $700

No data No data

0.84

Children in householde

52%

51%

50%

40%

0.86

Receives public assistancef

27%

24%

17%

29%

0.58

Has health insurance

45%

45%

56%

64%

0.51

Owns home

31%

25%

56%

53%

0.05 0.03

Gender Female

57%

59%

88%

76%

Male

43%

41%

12%

24%

16% 84%

15% 85%

10% 90%

20% 80%

0.64

Working full-time

60%

67%

58%

57%

0.51

Not working fulltime

40%

33%

42%

43%

Marital status Currently married Not married Employment status

(continued)

Splitting Tax Refunds and Building Savings

141

Table 4.9 (continued) Participants

Comparison

Baseline

Follow-up

Follow-up Baseline

PF 0 CF p-Value

49%

48%

25%

51%

0.09

51%

53%

75%

49%

61% 39%

56% 44%

35% 65%

42% 58%

0.12

Education Some college or more High school diploma or less Race Asian, Black, Hispanic, Native American, Hawaiian, or other White

Panel B: planned uses of refund Savings

81%

82%

63%

56%

0.12

Debt repayment

45%

45%

68%

56%

0.09

Spending

57%

49%

58%

58%

0.51

n

75

41

22

53

Note: Sample sizes varies across variables due to missing data. a Data on participants and comparison group members are presented for all individuals in each group with baseline data and for individuals in each group who completed a follow-up interview. The table reports demographic and ﬁnancial information for members of each subgroup and compares the characteristics of baseline and follow-up participants and the characteristics of baseline and follow-up comparison group members. In addition, the last two columns give the results of statistical tests (including chi-square, Fisher’s exact, and t-tests) comparing follow-up comparison group members to followup participants. Fisher’s exact tests are two-sided. b For participants, data on age are taken from Taxwise summary sheets. Comparison group data on age come from follow-up surveys and so are available only for comparison group members contacted on follow-up. c Data on comparison group members’ incomes were reported on the baseline survey in ranges. For participants, income equals AGI as reported on summary Taxwise page and then converted to ranges. d Anticipated refund for participants is taken from Taxwise summary sheets. Anticipated refund data for comparison group members is from baseline surveys completed by respondents prior to having their taxes prepared. e Percentage of respondents in each group reporting children under the age of 18 in household. f Percentage of respondents in each group reporting currently receiving TANF, SSI, or food stamps.

142

Beverly, Schneider, and Tufano

that the program was presented as a tool to encourage saving and the assumption that participants enrolled in the program because they believed it would in fact facilitate saving. 5.1 Saving Outcomes for Participant and Comparison Groups Our ﬁrst evidence on the short-term impact of R2A comes from comparing saving outcomes for the participant and comparison groups. We present data separately for planned savers (i.e., those who named a savings goal at baseline) and for all respondents. Limiting the sample to planned savers may reduce bias caused by differences in planned refund use across the two samples. In table 4.10, the outcome of interest is money from refunds still held in savings. Among planned savers, 72 percent of participants said that they were still saving part of their refunds, compared to 42 percent of the comparison group. This difference is statistically signiﬁcant at the .10 level. Among all follow-up respondents, 66 percent of participants were still saving, compared to 36 percent of the comparison group. This difference is statistically signiﬁcant at the .05 level. Mean differences in amount of refund still saved, either in dollar terms or as a percentage of the total refund amount, are not signiﬁcant. Next, we broaden our outcome measure to include using the refund on at least one stated savings goal, in addition to continuing to save a portion of the refund. A person who planned to save to purchase a car and, at follow-up, had done so would meet this broader deﬁnition of success. Someone who planned to save for the future and used part of her refund for an emergency also meets this deﬁnition. Table 4.11 shows that 78 percent of planned savers in the participant group were still saving some portion of their refund or had spent it on a stated savings goal. In the comparison group, 42 percent of planned savers met this criterion. Among all follow-up respondents (not just planned savers), 71 percent of participants met this criterion, compared to 36 percent of the comparison group. Both differences are signiﬁcant at the .05 level. Again, mean differences in amount of refund still saved or spent on a savings goal are not signiﬁcant. In sum, evidence regarding the short-term impact of R2A on saving part of a refund and achieving savings goals is positive but somewhat mixed: participants were more likely than comparison group members to save and spend on savings goals, but amounts saved and spent on savings goals did not differ signiﬁcantly.

Splitting Tax Refunds and Building Savings

143

Table 4.10 Saving out of Refund at Follow-up, for Participant and Comparison Groupsa Planned Saversb Participants Percentage still saving portion of refund

72%

All Respondentsc

Comparison

P0C p-Value

Participants

Comparison

P0C p-Value

42%

0.06

66%

36%

0.03

0.75

$338

$278

Amount saved Mean

$423

$318

$1,041

$596

$931

$550

$50

$0

$35

$0

Minimum

$0

$0

$0

$0

Maximum

$4,000

$2,000

$4,000

$2,000

Percentage of refund savedd Mean

24%

21%

Standard deviation

31%

Median

11%

Minimum

0%

Maximum

100%

Standard deviation Median

n a

32

21%

19%

29%

29%

28%

0%

8%

0%

0%

0%

0%

83%

100%

12

0.72

41

0.78

0.74

83% 22

Saving outcomes for participants and comparison group members are compared. The percentage of participants who planned to save and are still saving three to ﬁve months later is compared to that share of comparison group members doing so. The percentage of all participants (not just planned savers) who are still saving is also compared to the share of comparison group members doing so. Data on the amount of refund still saved and the percentage of the refund received still saved are presented for both planned savers and all follow-up respondents. Chi-square, Fisher’s exact, and t-tests are used to test for signiﬁcant differences among the groups. Fisher’s exact tests are one-sided. b Planned savers are those who named a planned savings use of their refunds on the baseline survey. c All respondents include those who completed a follow-up interview, regardless of whether they listed a planned savings use on the baseline survey. d Refund equals actual, not anticipated, refund amount.

144

Beverly, Schneider, and Tufano

Table 4.11 Saving out of Refund or Spending on Savings Goals at Follow-up, for Participant and Comparison Groupsa Planned Saversb Participants Percentage still saving portion or having spent on savings goal

78%

All Respondentsc

Comparison

P0C p-Value

42%

0.02

Amount saved or spent on savings goals Mean Standard deviation

Participants

71%

Comparison

P0C p-Value

36%

0.01

0.80

0.88

$472

$568

$376

$415

$1,065

$1,192

$956

$943

$53

$0

$50

$0

Minimum

$0

$0

$0

$0

Maximum

$4,000

$3,900

$4,000

$3,900

Median

Percentage of refund saved or spent on savings goalsd

n a

Mean

32%

27%

Standard deviation

36%

Median

13%

Minimum Maximum

27%

22%

37%

34%

32%

0%

10%

0%

0%

0%

0%

0%

100%

98%

100%

98%

32

12

0.66

41

0.56

22

Saving outcomes for participant and comparison groups are compared, ﬁrst for planned savers, then for all respondents. Chi-square, Fisher’s exact, and t-tests are used to test for signiﬁcant differences between the groups. Fisher’s exact tests are one-sided. b Planned savers are deﬁned as those who named a planned savings use of their refunds on the baseline survey. c All respondents include all individuals who completed a follow-up interview, regardless of whether they listed a planned savings use on the baseline survey. d Refund equals actual, not anticipated, refund amount.

Splitting Tax Refunds and Building Savings

145

5.2 Bank-Level Account Data In addition to self-reported data, we obtained bank balance and transaction data for participants from the Bank of Oklahoma. BOk provided individual data for participants who either opened new savings accounts with BOk or who split their refunds into existing BOk accounts. The data include account status, monthly account balance information for August through November 2004, limited account-level data on transactions, and household-level data on the ownership of BOk products. These data were not available for comparison group members. BOk located account data for ﬁfty-eight out of the sixty-six individuals who either opened an account with BOk or split to an existing account. (These ﬁfty-eight individuals represent 77 percent of all R2A participants.) Of these ﬁfty-eight, 62 percent still had active BOk accounts by mid-August 2004, but 38 percent had closed their accounts, either at their request or due to bank policy, which closes zero-balance accounts. (The bank provided no additional information on these closed accounts and account-holders.) The 62 percent with still-open accounts kept their accounts open at least through November 2004, when BOk provided data to the research team. Participants with open accounts generally received larger refunds and had higher incomes than participants with closed accounts. Apart from these differences, there were no signiﬁcant differences between the two groups on other demographic variables. Participants with open accounts appeared to have greater familiarity with savings accounts (33 percent had a savings account at another ﬁnancial institution, compared to 22 percent of those with closed accounts), but the difference is not statistically signiﬁcant. The two groups appear almost identical in terms of R2A services used (splitting, account opening, or both) and checking account ownership. For participants who still had accounts open in November 2004, BOk provided household-level data on other relationships these individuals had with the bank. As we would expect, all of these households owned some kind of savings vehicle at BOk. (Ninety-seven percent had savings accounts, 8 percent had money market accounts, and 3 percent held IRAs at BOk.) Only 16 percent had ATM cards. Half of the households had a checking account of some kind, including a mix of standard checking accounts, free or ‘‘second chance’’ checking accounts, and even high-end products with high balance requirements (though the balances on these accounts were quite small and so the account

$4,018

$14

Standard deviation

Median

$15

$192

$91

$62

$248

$15

$142

$67

$60

$228

$155

$2,276

$133 $8

$63

$9

$65

$40

$61

$214

$147

$987

$54 $1

$27

October 31, November 30, 2004 2004 (n ¼ 36) (n ¼ 36)

$51

$181

$130

$106

$270

$207

$3,705

$405 $51

$185

August 31, 2004 (n ¼ 20)

$58

$240

$150

$116

$245

$199

$2,708

$206 $56

$135

September 30, 2004 (n ¼ 20)

$65

$175

$119

$142

$220

$189

$2,284

$162 $56

$114

$49

$74

$72

$135

$205

$180

$987

$66 $24

$49

October 31, November 30, 2004 2004 (n ¼ 20) (n ¼ 20)

Accounts Still Active as of November 2004a

b

Accounts are considered to be inactive if the balance is less than $10 and there is no account activity for more than 2 months. For any given date, the time elapsed since refund receipt varied somewhat across sample members. For the end of August, the average and median number of weeks since refund receipt among open accounts was 26, the minimum was 20, and the maximum was 27.

a

$82

$149

Mean

$56

$275

Standard deviation

Median Month-to-date average balance

$176

Mean

Year-to-date average balance

$166

$2,894

$312 $11

Total deposits

$167 $13

$112

Standard deviation Median

$80

September 30, 2004 (n ¼ 36)

Mean

Balance

August 31, 2004b (n ¼ 36)

Accounts Still Open as of November 2004

Table 4.12 Account Balances for Open and Active Bank of Oklahoma Savings Accounts, August–November 2004

146 Beverly, Schneider, and Tufano

Splitting Tax Refunds and Building Savings

147

holders were probably paying account fees). Only 8 percent of these households held any debt with BOk. (Three percent had installment loans, and 5 percent had mortgages.) Some of these checking accounts, debt products, and investment accounts appear to pre-date the tax season. However, approximately 56 percent of this subset of households did not have a relationship with BOk prior to the tax season, so the initial savings accounts generated new customers for the bank. Furthermore, one-quarter of these new R2A-BOk customers opened additional accounts (beyond the savings account) following the tax season, suggesting meaningful crossselling opportunities within the R2A participant sample.31 BOk provided end-of-month data on savings account balances for August, September, October, and November 2004 for accounts remaining open in mid-August. These data show that participants rapidly withdrew money from their accounts. This subsample of participants initially deposited an average of $644 into their BOk accounts. The average balance dropped to $112 by August (twenty-six weeks after refund receipt for most participants) and to $27 by November (thirtyeight weeks after refund receipt for most participants). The median balance of $203 at enrollment dropped to $11 in August and $1 in November. Although thirty-six accounts were still open in November, at least sixteen of those might be considered inactive: they had balances of less than $10 and no activity for at least two months. As detailed in table 4.12, excluding these inactive accounts from the balance calculations yields higher average and median balances but a substantially higher account closure rate. 5.3 Saving Outcomes for Participants With and Without R2A Further evidence on the short-term impact of R2A comes from data on saving part of refunds by participants in 2003, the year before R2A was available. In the baseline survey, we asked R2A participants whether they had received a federal refund in the previous year. We asked those who did receive a refund in 2003 whether they were still saving a portion three months after they received their refunds. We compared these data with data on savings at follow-up in 2004, when participants used R2A services. We use a McNemar chi-square to test for signiﬁcant differences across time among the participants responding to both the baseline question and the follow-up survey ðn ¼ 31Þ. Participants report a higher propensity to save with R2A relative to the prior year, when R2A was not available, a relationship that is highly signiﬁcant ðp < :01Þ.32

148

Beverly, Schneider, and Tufano

Table 4.13 Responses to Follow-up Survey Attitudinal Questions for Participants, by Participation Statusa Split Refund to New or Existing Account

Did Not Split (New Account Only)

All Participants

Service helped to save more of refund

89%

63%

83%

Service helped to spend refund more slowly

82%

88%

83%

Service helped to resist spending temptations

76%

75%

76%

Would split next yearb

97%

75%

92%

Would still split even if had to wait longer for refundc

87%

100%

89%

100%

NA

NA

$6.60

NA

NA

Would recommend service to a friend Average amount willing to pay to split refund n

33

8

41

a

Participants were asked a series of questions designed to gauge their satisfaction with R2A and their feelings about the program’s effectiveness. Participants were also asked several questions about possible future use of a splitting service. Responses are presented for participants who split their refunds, for participants who did not split their refunds, and for all participants. b Splitters were asked if they planned to split their refunds again next year. For nonsplitters, the concept of the splitting service was explained, and they were then asked if they would split their refunds next year. c Respondents (both splitters and non-splitters) who stated that they would split their refunds again next year were asked if they would still plan to split even if their refunds arrived more slowly as a result. Respondents who did not plan to split next year were not asked this question.

5.4 Participant Perceptions of R2A A ﬁnal source of evidence regarding the impact of R2A comes from responses to attitudinal questions included in the follow-up survey, displayed in table 4.13. The survey data suggest that R2A facilitated saving and thoughtful spending. Eighty-three percent of respondents said that the service helped them save more, 83 percent said that the service helped them spend their refunds more slowly, and 76 percent said that the service helped them resist spending temptations. In addition, all respondents said that they would recommend R2A to a friend, and 97 percent planned to split their refunds again next year. On aver-

Splitting Tax Refunds and Building Savings

149

age, participants were willing to pay $6.60 to split their refunds, and 89 percent were willing to split even if they had to wait longer to receive their refunds. These results suggest that participants perceived material beneﬁts from the service, albeit not necessarily related to longterm savings goals. 6.

Discussion

We willingly concede that this research has important limitations. This is a small study in a single location with a set of taxpayers that may not be representative of the LMI population in the United States. In addition, ﬁndings related to the impact of the program must be interpreted very cautiously because individuals were not randomly assigned to treatment and control groups, and the comparison group examined here probably differs from the treatment group in important (and unmeasured) ways. Nevertheless, we offer some tentative conclusions. Most are best viewed as hypotheses for future study rather than as conclusions in their own right. We also suggest some directions for future research. Refund splitting seems to appeal to LMI families. When we began this experiment, the team was prepared for the possibility that no one would want to use this new, untested service. We offered people a service that involved their money, where few had advance notice of the service, where there was no word of mouth, and where refund recipients had to ﬁll out additional paperwork in order to participate. We were surprised that demand was as strong as it was: 27 percent of refund recipients wanted to use the service. Afterwards, almost all participants said they were satisﬁed with the service and would use it again, and they were willing to pay an average of $6 for the splitting service. While it will take other experiments at other sites to conﬁrm these results, our data suggest that refund splitting could meet with very strong consumer demand among LMI families. LMI families are most interested in splitting into basic ﬁnancial products, but other products have some appeal. In our follow-up survey, we asked participants which ﬁnancial products they would like to split into (see table 4.14). There was greater demand for simpler ﬁnancial products, with 92 percent very likely or likely to select a savings account; 88 percent, a checking account; and 65 percent, a paper check. There was also interest in retirement savings products (53 percent); college savings products (51 percent); and, to a lesser extent, certiﬁcates of deposit,

150

Beverly, Schneider, and Tufano

Table 4.14 New Splitting Options, Participant Interest in Additional Productsa Very Likely

Somewhat Likely

Combined

Savings account

66%

26%

92%

Retirement savings

14%

39%

53%

College savings

27%

24%

51%

Certiﬁcate of deposit

12%

17%

29%

7%

17%

24%

9%

12%

21%

Savings

US savings bond Mutual fund Transaction Checking account

63%

25%

88%

Paper check

25%

40%

65%

Stored value cardb

15%

25%

40%

8%

14%

22%

Car loan

20%

25%

45%

Credit card Mortgage

32% 25%

12% 17%

44% 42%

International remittance Debt

a Participants and comparison group members (n ¼ 60) were asked how likely they would be to split their refunds into a number of different ﬁnancial products. Respondents were given four response options: ‘‘very likely,’’ ‘‘somewhat likely,’’ ‘‘somewhat unlikely,’’ and ‘‘very unlikely.’’ This table aggregates participant and comparison group member responses and presents the results for the ‘‘very likely’’ and ‘‘somewhat likely’’ responses, and combines them in the right-hand column. b The stored value card was described to respondents as ‘‘a card that you could use at an ATM/Transfund machine, like a debit card.’’

savings bonds, and mutual funds.33 Families also expressed a demand for a splitting service that directly paid down various debts (car loan— 45 percent, credit card—44 percent, and mortgage—42 percent). Finally, more than one-fourth of respondents expressed interest in a splitting service that would allow them to send international remittances.34 While some of these remittances would likely be spent by recipients, it appears that some remitted funds are used for asset building, including home construction and improvement (Thompson 2002). Lack of access to low-cost savings vehicles could substantially limit actual rates of refund splitting by LMI families. In our experiment, participants either split into existing accounts, split into new accounts, or opened new accounts without splitting. If we had limited the service to those who could split into existing accounts, our participation rate would

Splitting Tax Refunds and Building Savings

151

have been considerably lower. In the R2A pilot, only 27 percent of the participants split into existing accounts, with the remainder opening accounts. Slicing the data another way shows that the demand for splitting alone (4 percent) was less than one-half the demand for splitting coupled with account opening (8 percent). The ChexSystems screen appears to have been a signiﬁcant barrier to participation. In our experiment, potential participants foiled by ChexSystems had completed almost all of the required paperwork and clearly had strong demand for the service. Had we been able to offer them a savings vehicle, we could have increased our splitting participation rate from 12 percent to 19 percent.35 While some tax preparation sites have succeeded in convincing bank partners to waive ChexSystems requirements for refund depositors, these individual agreements have affected a relatively small number of ﬁlers. Although BOk did not waive the ChexSystems screen, the bank did forgo the $100 initial-deposit requirement and the $300 minimumbalance requirement for participants. This encouraged participation in R2A, but it will not be possible to take splitting to scale if program administrators must depend on the goodwill of individual ﬁnancial institutions to waive requirements of this sort. To implement splitting, one must address material operational concerns, which might suggest that this process be administered centrally by the IRS. Our paper primarily deals with the participant experience with the splitting service, not the operational complexities of offering it. Yet R2A changed the ﬂow of the tax-preparation site because account opening required an additional step with different personnel. It also required monitoring refund transmissions from the IRS, transferring funds according to splitting instructions, and responding to user inquiries. While a large commercial preparer could accommodate an R2A-like program,36 our approach would not work for self-preparers or thinly staffed tax-preparation sites. As much as one-third of all LMI returns are self-prepared, and we suspect that a meaningful fraction of the volunteer tax-preparation sites (which process 1 to 3 percent of LMI returns) might be operationally challenged by the requirements to open accounts on site.37 If splitting is sensible, then from an operational perspective, it can be accomplished either at the destination or at the source. Destinationbased splitting—in the R2A fashion or as currently offered to higher net worth customers—would require the participant to instruct her ﬁnancial institution or agent to disburse her refund in a certain fashion

152

Beverly, Schneider, and Tufano

once it was received from the IRS. Source-based splitting would provide instructions to the sender of the refund, the IRS, to split it to multiple destinations. Both of these approaches could coexist, but there are operational considerations that make source-based splitting appealing. Splitting instructions could be integrated into the tax preparation process seamlessly, as just another destination for refunds. There would be no need for private parties to coordinate with the IRS to track refunds. Perhaps more important for LMI ﬁlers, source-based splitting would ensure that the service was available to all, regardless of the ﬁnancial institution with which they do business. Apparently, current policy makers have reached this conclusion. The Bush administration has included a splitting proposal in its budget for the last two years. On January 31, 2005, twelve members of Congress wrote to IRS Commissioner Everson, requesting that the IRS adopt technical changes that would enable splitting. On March 25, 2005, Commissioner Everson replied that the IRS was ‘‘working toward making this program available as quickly as possible’’ and set a deadline of the 2007 ﬁling season. This model of refund splitting may provide business opportunities for private tax preparers. Currently, many private tax preparers, including large companies such as H&R Block and Jackson Hewitt, but also smaller businesses, derive substantial revenue from the sales of refund-related ﬁnancial products. To date, these have mostly included short-term loans, commonly called refund anticipation loans (RALs), which allow refund recipients to access their funds several days faster than would be possible with standard direct deposit. While H&R Block was the ﬁrst company to offer RALs, they are now widely available, and several of the larger tax preparation ﬁrms have begun to introduce stored value card (SVC) products that allow refund recipients to receive their funds on a plastic card, similar to a debit card. Refund splitting could encourage the sale of more than one refundrelated product to a single consumer. For example, a refund recipient might purchase an SVC and also open a savings account, splitting her refund between the two. Multiple-product purchase could both increase revenue and build the relationship between tax preparer and client, reducing costly client acquisition activity (Tufano and Schneider 2005). A well-executed splitting program could also serve as a source of differentiation in the highly competitive tax preparation sector. Rather than build a business about accelerating spending, a preparer could capitalize on its customers desire to slow down spending by sav-

Splitting Tax Refunds and Building Savings

153

ing. One might fear that preparers might offer inferior products to lowincome savers. In part, this concern would be addressed by rules that would subject the preparers or their business partners to NASD suitability rules or ‘‘know your customer’’ regulations in banking, depending on the type of product they offered to consumers. Competition could also address this concern in the long run if preparers competed on the basis of offering attractive savings products. However, savers’ interests could be protected by offering all refund savers a baseline product that could serve as a ﬂoor or default, yet encourage private preparers to offer products with terms at least as favorable or better. In particular, Tufano and Schneider (2005) discuss permitting refund recipients to purchase U.S. savings bonds with their refunds. Twentyfour percent of follow-up survey participants said they would be likely to purchase savings bonds with part of their refund (table 4.14). Another survey question described savings bonds (rather than simply naming them), and 76 percent of respondents said they would be likely to purchase bonds.38 While the data suggests that refund recipients might like to direct their money to many alternatives, this would ensure that there was at least one savings alternative available to all. No customer would be captive to just the products that their preparer offered. In addition, the savings bond’s terms would serve as a lower bound for any private product. Whether this would spur private savings products or crowd them out is an empirical question. There are many opportunities for related research and evaluation. Our study is admittedly a small-scale experiment, yet it can offer some hypotheses for future study. In particular, we are intrigued by the comparison of our results with those obtained in the St. Louis study (Duﬂo et al. 2005), cited earlier, which differed from ours in a few key dimensions. Our experiment uses a simple savings vehicle, whereas the St. Louis team directed savings into an IRA product. Our experiment provided greater liquidity, while theirs provided less liquidity and greater precommitment. The two products seem to have appealed to individuals with differing (though complementary) savings goals. Many R2A participants reported planning short-term savings for emergencies, while the Express IRA participants generally focused on longer-term retirement savings. Both are legitimate savings goals. Our savings program offered little explicit ﬁnancial inducement to participate, apart from waiving minimum balance requirements and monthly fees. The St. Louis experiment provided substantial ﬁnancial

154

Beverly, Schneider, and Tufano

incentives for participation, in the form of matching funds: Participants were randomly selected to be offered a 50 percent match, a 20 percent match, or no match. Those who were not offered a match had the lowest participation rates (3.3 percent) and contributed the smallest amount on average ($28). Interestingly, St. Louis ﬁlers who were offered the 50 percent match on the less-liquid IRA product had a very similar take-up rate (17 percent) compared to participants in the R2A study (15 percent) who were offered liquid accounts with no match (Duﬂo, Gale, Liebman, Orszag, and Saez 2005). Comparing the two results may give a sense about the size of the compensation required to induce long-term savings. While it may be appropriate to encourage long-term savings, this cannot come at the expense of acknowledging other shorter-term savings goals. If we examine the R2A program from the perspective of consumer satisfaction or self-reported ability to better manage ﬁnances, it looks attractive. Measured narrowly using data on account balances, the results are far less impressive because balances were rapidly drawn down. Our small experiment suggests that this service cannot be evaluated simply by looking at the size of a participant’s bank balance. While savings account balances dropped signiﬁcantly over our nine-month study, participants saved portions of their refunds for longer periods of time than individuals not offered the service. Perhaps the relevant measure of success is not account balance, but whether the participant is making progress toward—or has achieved—her ﬁnancial goals. Emergency saving is a legitimate savings goal for families, as is saving for auto repair, if that car helps an individual keep his job. If the program helps participants to weather emergencies or keep their cars functioning and hence maintain employment, then the program cannot be judged a failure because savings balances are depleted. Although accumulating and protecting long-term savings (say, for retirement) are important goals, there are other critical savings goals for LMI families. Any program must be measured against the participants’ goals. Our results suggest that a program of splitting with account opening might generate strong interest among LMI families. By forcing families to think about their refunds before they receive them, they may ﬁnd it easier to save and to resist spending temptations. Participants seem to have a somewhat better—or no worse—post-refund track record than other families. Future studies should further test the impact of programs like R2A, further unbundling the effects of account opening

Splitting Tax Refunds and Building Savings

155

from splitting. In addition, while we have examined a limited set of account data, future work should look more closely at account-level transaction and balance data to understand how low-income ﬁlers use their accounts and the consequences of those use patterns for bank proﬁtability. More generally, all of these ﬁrst ﬁndings should be replicated with larger samples, where it would be possible to go further, to understand the product mix, features, and marketing activities that would best support savings by low- and moderate-income families. Notes We would like to thank the Annie E. Casey Foundation; the Community Action Project of Tulsa County; Doorways to Dreams Fund (D2D Fund); the Bank of Oklahoma; and the participants from the Pine and Lewis, Houston Center, and 21st Street tax preparation sites in Tulsa, Oklahoma, for allowing us to conduct this study. In particular, we would like to thank the team consisting of Monica Armstrong, Kim Cowden, Steven Dow, Brandy Holleyman, Tim Flacke, and Jeff Zinsmeyer, who helped plan and carry out this experiment. We would also like to thank Lily Bachelder, Michael Barr, Colleen Dailey, Stephen Dow, Fred Goldberg, John Gourville, Derek Martin, David Rose, Reggie Stanley, Mike Stegman, Anne Stuhldreher, Jen Tescher, Heather Tyler, John Zinman, as well as participants at the NBER Tax Policy and the Economy Conference, the National Community Tax Coalition/Annie E. Casey 3rd Annual Conference, and the 2004 IDA Learning Conference for their comments on this work. Toni Wegner and David Hann assisted us in the IRB approval processes. Financial support for this research project was provided by the Division of Research of the Harvard Business School, the Annie E. Casey Foundation, and D2D Fund. We thank them for their support but acknowledge that the ﬁndings and conclusions presented in this report are those of the authors alone and do not necessarily reﬂect the opinions of the Annie E. Casey Foundation, Harvard Business School, or D2D Fund. 1. In fall 2004, all three of the top-selling books on personal ﬁnance at Amazon.com emphasized the concept of ‘‘paying yourself ﬁrst.’’ See Kiyosaki and Lechter (2000), Bach (2003), and Clason (1988). 2. We deﬁne LMI as having adjusted gross income (AGI) of less than $30,000. 3. Authors’ calculations based on published Internal Revenue Service (IRS) data. The 2001 ﬁgures are derived from estimates prepared by the Statistics of Income Division of the IRS and are based on calendar year. Total refunds is the sum of all refunds to ﬁlers with AGI of less than $30,000. 4. Authors’ calculations based on data from Internal Revenue Service, Statistics of Income (2001). 5. Our research project roughly followed the action research cycle of problem deﬁnition, pilot program implementation, evaluation, and re-test. See Lewin (1948) and Kemmis and McTaggert (1988). 6. The Bank of Oklahoma is a subsidiary of the Bank of Oklahoma Financial Corporation (BOKF). BOKF operates banks in Oklahoma, Texas, New Mexico, Arkansas, and

156

Beverly, Schneider, and Tufano

Missouri. It is the largest bank in Oklahoma, with $5.9 billion in deposits, more than twice that of BancFirst, its closest competitor in the market. See www.bokf.com. 7. See, for example, Schreiner, Clancy, and Sherraden (2002); Sherraden and Barr (2004); and Hubbard and Skinner (1996). 8. This decline is less dramatic, though generally still evident, when alternative approaches to measuring the saving rate are used, such as those employed by Parker (1999) and Gale, Sabelhaus, and Hall (1999). Borsch-Supan and Lusardi (2003) use data from the Organisation for Economic Co-operation and Development (OECD) to show that the saving rate in the United States is signiﬁcantly below that of most European countries and Japan. 9. The EITC is available to low-income tax ﬁlers (individuals or families) with earnings. It is intended primarily for parents of children under age 19, but low-income workers without children may receive a small credit. The amount of the credit initially rises with earnings, then reaches a plateau, and ﬁnally decreases with each additional dollar earned. See Center on Budget and Policy Priorities (2004) for details. The maximum credit for tax year 2003 was $4,204. 10. The CTC is a federal tax credit for each dependent under age 17. For tax year 2003, the credit was worth a maximum of $1,000 per child. Filers with taxable earned income above $10,500 were eligible for a refundable credit. See Lee and Greenstein (2003) for details. When a credit is refundable, any portion of the credit that exceeds tax liability is transferred to the tax ﬁler as an income tax refund. 11. At VITA sites, savings accounts have generally been offered through partnerships between the community organizations running the tax-preparation sites and banks or credit unions. The terms of these accounts vary, with ﬁnancial institutions occasionally waiving fees, minimum balances, or ChexSystems requirements (discussed in more detail in section 4 of this paper). Product marketing differs across programs as well. At some sites, bank representatives are on-site and actively promote the accounts. At others, VITA staff may be responsible for selling the accounts and are required to call in to a partner bank to set up the account remotely. There also appears to be considerable variability in the success of these programs. Some practitioners have found that few participants are interested in opening accounts and that those who do open accounts deplete these accounts quickly; other programs have seen greater success. This assessment is based on phone interviews with staff from nine free tax-preparation sites and reports from two other taxpreparation sites. Interviews were conducted during April 2004 with staff at The Baltimore CASH Campaign (Baltimore, Md.), the Piton Foundation (Denver, Colo.), The Milwaukee EITC/Asset Building Project (Milwaukee, Wis.), Alternatives Federal Credit Union (Ithaca, N.Y.), The Central City Asset Building Coalition (New Orleans, La.), The Riverside Family Asset Building Program (Riverside, Calif.), The United Way of King County (Seattle, Wash.), The Center for Economic Progress (Chicago, Ill.), The Nehemiah Gateway CDC (Wilmington, Del.). Reports from the Boston EITC coalition (Boston, Mass.) and the Community Food Resource Center (New York, N.Y.) were also used. 12. In 1998, Richard Thaler and Shlomo Benartzi implemented the ﬁrst test of the Save More Tomorrow plan (SMarT) (Thaler and Benartzi 2004). The SMarT plan allowed workers to precommit to automatically save their 3 percent annual raises in the company pension plan. Of 286 employees who met with an investment advisor, 162 enrolled in the program. After twenty-eight months, these employees had increased their saving rates from an average of 3.5 percent to 11.6 percent of income. A comparison group of employees who did not precommit to save their raises, but who did agree to try to increase their

Splitting Tax Refunds and Building Savings

157

savings without an automatic feature, increased their saving rates from only 4.4 percent to 8.7 percent. 13. Prior to tax ﬁling, would-be splitters ﬁle a form with Vanguard outlining the allocation of their refunds. These ﬁlers then submit a general Vanguard account number and routing number on their tax return, and Vanguard automatically splits the refund deposited into this account according to the previously issued instructions. Additional information on Vanguard’s splitting policy is available at http://ﬂagship2.vanguard.com/ VGApp/hnw/TcDirectDepositController?cbd (accessed March 3, 2004). 14. This statement is based on phone interviews conducted in March 2004 with representatives from the nation’s ten largest banks and largest mutual funds by assets. 15. Seventy-two of the 500 largest mutual funds by assets as reported by Morningstar were Vanguard funds. Of these, only one fund had a minimum initial investment of less than $1,000. 16. Individual development accounts were ﬁrst proposed by Sherraden (1991). They are matched savings accounts for low-income people and are designed to encourage asset building. 17. In the interest of full disclosure, we note that one of the researchers is the founder and chair of D2D Fund. 18. These accounts earned the same interest rate as other basic BOk savings accounts (.5 percent per annum at the time of the study). However, BOk waived the $100 openingdeposit requirement. The bank also waived minimum-balance requirements and associated fees for accounts in this program. Had these requirements not been waived, account holders would have been charged $6 per quarter on accounts with balances of less than $300. 19. Clients were also asked, but not required, to sign a research consent form. 20. Postcards announcing the program were mailed in January 2003 to 3,200 clients of CAPTC, but this mailing was ineffective and generated virtually no detectable response. 21. Copies of survey instruments are available from the authors. 22. The take-up rate may have been higher in R2A because participants who have used other CAPTC services in the past and have a favorable opinion of the organization might be more likely to use another CAPTC service. However, CAPTC reports that tax clients do not generally identify the free tax preparation service with CAPTC. 23. While research on the impact of ChexSystems (and similar) credit scoring services on banking participation by the poor is limited, BOk’s practices are probably the norm among banks. However, some depository institutions waive ChexSystems either as a general rule or in the speciﬁc instance of targeted saving programs. 24. We ran several logistic regression models to identify predictors of take-up. Findings were not robust, so we cannot say whether differences that were signiﬁcant in a bivariate framework would remain signiﬁcant when other differences were controlled. 25. The American Dream Demonstration (ADD) project, a national study of IDA programs, found that participants with the lowest incomes had the highest average saving rates. However, regression analysis on ADD participants did not ﬁnd that income had a signiﬁcant effect on the likelihood of being a saver (deﬁned as saving a speciﬁed minimum over the life of the program). The ADD research also found that having health

158

Beverly, Schneider, and Tufano

insurance was positively related to being a saver (Schreiner, Clancy, and Sherraden 2002), but our ﬁndings suggest that health insurance is negatively related to R2A enrollment. 26. In 2003, the unemployment rate in Tulsa was 6.5 percent, the highest annual rate in ﬁfteen years (Bureau of Labor Statistics, U.S. Department of Labor 2004). 27. These decliners chose the following response option: ‘‘I’d be interested if there was a match or if there was an option to save for a speciﬁc purpose.’’ It is unlikely that individuals would have given this response to an open-ended question. Still, these responses suggest that the service, as delivered, was not attractive enough to elicit participation. 28. The per capita data is restricted to those participants who responded to the survey item on existing savings and for whom there is data on the amount of the refund sent to savings ðn ¼ 68Þ. 29. We attempted to reach participants and comparison group members multiple times over the follow-up period. Only two of those contacted refused to complete the followup survey. It seems unlikely that participants who did not save disproportionately avoided the follow-up interview. 30. Data on anticipated refund amounts come from different sources for the two groups. Comparison group members were asked to estimate refund amounts before tax preparation began. For participants, data were taken from the Taxwise software after tax preparation was completed. 31. Data are not available on the types of accounts opened or how many accounts each household opened. Among all participant households that kept their accounts open through November (not just new BOk customers), 19 percent (7) purchased new products after the tax season. 32. There are three important data quality issues in addition to the small sample. First, respondents may have been unable to accurately recall refund uses for 2003. Since the R2A outreach emphasized the value of saving, we assume that over-reporting of saving in 2003 was more common than under-reporting. If this is true, our test is conservative, all else equal. Second, we cannot account for changes in economic or personal circumstances that would affect the use of a refund. The ﬁnal issue involves mismatch in timing. For saving in 2003, each person was asked to report whether she was saving three months after refund receipt. For saving in 2004, each follow-up survey respondent was asked whether she was currently saving some portion of the refund, and the timing of followup surveys varied from three months to ﬁve months after refund receipt. Again, this data ﬂaw is likely to make our test more conservative because people are less likely to have some of the refund in savings as time passes. 33. Our experiment allowed refund recipients to invest only in a completely liquid vehicle—a standard savings account. Participants who wanted a less liquid product, such as a certiﬁcate of deposit or a savings bond, may have chosen not to split. In future research, it would be useful to examine the demand for less liquid products, which may encourage and facilitate longer-term saving and ﬁnancial planning. 34. It is possible that our survey underestimated demand for this service in the wider market. A large portion of the remittances sent abroad from the United States are directed to Latin America. The most likely senders of these remittances, Hispanics, were under-represented in our sample because we did not recruit participants at CAPTC’s Spanish-language tax-preparation site.

Splitting Tax Refunds and Building Savings

159

35. The splitting participation rate is calculated as the number of splitters (62) divided by the number of individuals approached expecting a refund (516). The participation rate had ChexSystems not been used is calculated as the number of participants splitting (62) and the number denied by ChexSystems (36) divided by the number of individuals approached expecting a refund (516). 36. H&R Block has offered two programs in which a participant can open either an IRA or a savings account at the time of ﬁling (Tufano and Schneider 2005) and effectively offers splitting services. 37. Wojciech Kopczuk and Cristian Pop-Eleches use 1999 SOI data to report that 67 percent of EITC eligible families used a professional tax preparer. See Kopczuk and PopEleches, ‘‘Electronic Filing, Tax Preparers, and Participation in the Earned Income Tax Credit,’’ Columbia University working paper, March 2005, http://www.columbia.edu/ ~wk2110/bin/eﬁle.pdf (last accessed May 30, 2005). Data on the size of the VITA market is drawn from ‘‘Tax Administration: IRS’s 2003 Filing Season Performance Showed Improvements,’’ 2003, United States General Accountability Ofﬁce (GAO); National Taxpayer Advocate 2004 Annual Report to Congress, 2004, Taxpayer Advocate Services, Internal Revenue Service; Internal Revenue Service, ‘‘Tax Stats,’’ http://www.irs.gov/taxstats/; Internal Revenue Service, Statistics of Income, 2001, Individual income tax statistics— 2001, Table 3.3—2001: Individual income tax, all returns: Tax liability, tax credits, tax payments, by size of adjusted gross income, http://www.irs.gov/pub/irs-soi/01in33ar .xls. 38. Speciﬁcally 76 percent of those surveyed answered yes to the following: ‘‘Most savings accounts in banks currently pay about 1 percent interest. Suppose you could send part of your refund to save in a savings bond that paid 3.4 percent but wouldn’t allow you to withdraw the money for several months. Would you consider sending part of your refund to save in a savings bond?’’ The rates we quoted in the survey question were representative of those offered at the time through bank accounts and savings bonds.

References Aizcorbe, Ana M., Arthur B. Kennickell, and Kevin B. Moore (2003). ‘‘Recent Changes in U.S. Family Finances: Evidence from the 1998 and 2001 Survey of Consumer Finances,’’ Federal Reserve Bulletin, 89:1–32. Ashraf, Nava, Dean S. Karlan, and Wesley Yin (2006). ‘‘Tying Odysseus to the Mast: Evidence from a Commitment Savings Product in the Philippines,’’ Quarterly Journal of Economics. Bach, David (2003). The Automatic Millionaire: A Powerful One-Step Plan to Live and Finish Rich. New York: Broadway. Barrow, Lisa, and Leslie McGranahan (2000). ‘‘The Effects of the Earned Income Credit on the Seasonality of Household Expenditures,’’ National Tax Journal, 53(4):1211–1243. Beverly, Sondra G., Amanda Moore, and Mark Schreiner (2003). ‘‘A Framework of AssetAccumulation Stages and Strategies,’’ Journal of Family and Economic Issues, 24(2):143–156. Beverly, Sondra G., Jennifer Tescher, Jennifer L. Romich, and David Marzahl (2002). ‘‘Linking Tax Refunds and Low-Cost Bank Accounts: Findings from the Extra Credit Savings Program,’’ Joint Center for Poverty Research, Chicago, Ill., working paper no. 277.

160

Beverly, Schneider, and Tufano

Borsch-Supan, Axel, and Annamaria Lusardi (2003). ‘‘Saving: A Cross-National Perspective,’’ in Axel Borsch-Supan (ed.), Life-Cycle Savings and Public Policy: A Cross-National Study of Six Countries. San Diego, Calif.: Elsevier Science. Brookings Institution (2004). ‘‘Proceedings of Filling the Savings Gap: How to Get Moderate-Income Households to Save for Retirement,’’ http://www.brook.edu/comm/ events/20040517.pdf. Bureau of Economic Analysis, U.S Department of Commerce (2004). ‘‘National Income and Products Accounts,’’ Table 2.9: Personal Income and Its Disposition by Households and by Nonproﬁt Institutions Serving Households, U.S Department of Commerce website, http://www.bea.gov/bea/dn/nipaweb/TableView.asp?SelectedTable=294& FirstYear=2001&LastYear=2002&Freq=Year (accessed July 16, 2004). Bureau of Labor Statistics, U.S Department of Labor (2004). ‘‘Local Area Unemployment Statistics,’’ Bureau of Labor Statistics web site, http://data.bls.gov/servlet/ SurveyOutputServlet (accessed July 14, 2004). Carasso, Adam, and Eugene Steuerle (2003). ‘‘Growth in the Earned Income and Child Tax Credits,’’ Tax Facts column, Tax Notes, 98(3):401. Clason, George S. (1988). The Richest Man in Babylon. New York: New American Library. Center on Budget and Policy Priorities (2004). ‘‘Facts about the Earned Income Credit,’’ Washington, D.C. http://www.cbpp.org/eic2003/fuctseic-2003.pdf. Duﬂo, Esther, William Gale, Jeffrey Liebman, Peter Orszag, and Emmanuel Saez (2005). ‘‘Saving Incentives for Low- and Middle-Income Families: Evidence from a Field Experiment with H&R Block,’’ Retirement Security Project Policy Brief no. 2005-5, May. Gale, William G., John Sabelhaus, and Robert E. Hall (1999). ‘‘Perspectives on the Household Saving Rate,’’ Brookings Papers on Economic Activity, 19(1):181–224. Haveman, Robert, and Edward Wolff (2001). ‘‘Who Are the Asset Poor? Levels, Trends, and Composition, 1983–1998,’’ University of Wisconsin, discussion paper no. 1227-01. Hubbard, Glen, and Jonathan Skinner (1996). ‘‘Assessing the Effectiveness of Saving Incentives,’’ Journal of Economic Perspectives, 10(4):73–90. Internal Revenue Service, Statistics of Income (2001). ‘‘Individual income tax statistics— 2001, Table 3.3—2001 Individual income tax, all returns: Tax liability, tax credits, tax payments, by size of adjusted gross income,’’ http://www.irs.gov/pub/irs-soi/01in33ar.xls. Kemmis, Stephen, and Robin McTaggart (1988). The Action Research Planner, Geelong, Victoria: Deakin University Press. Kiyosaki, Robert T., and Sharon L. Lechter (2000). Rich Dad, Poor Dad: What the Rich Teach Their Kids About Money—That the Poor and Middle Class Do Not! New York: Warner Business Books. Lee, Andrew, and Robert Greenstein (2003). ‘‘How the New Tax Law Alters the Child Tax Credit and How Low-Income Families Are Affected,’’ Washington, D.C.: Center on Budget and Policy Priorities. Lewin, Gertrude W. (1948). Resolving Social Conﬂicts: Selected Papers on Group Dynamics. New York: Harper & Row.

Splitting Tax Refunds and Building Savings

161

Page-Adams, Deborah, and Ed Scanlon (2001). ‘‘Assets, Health, and Well-Being: Neighborhoods, Families, Children, and Youth,’’ Washington University in St. Louis, research background paper no. CYSAPD 01-9. Parker, Jonathan (1999). ‘‘Spendthrift in America? On Two Decades of Decline in the U.S. Saving Rate,’’ in B. Bernanke and J. Rotemberg (eds.), NBER Macroeconomics Annual 1999, 14(2000):317–370. Quinn, Jane B. (2001). ‘‘Checking Error Could Land You on Blacklist,’’ Washington Post, September 30, 2001, http://global.factiva.com/en/eSrch/ss_hl.asp (accessed March 4, 2004). Schreiner, Mark, Margaret Clancy, and Michael Sherraden (2002). ‘‘Final Report: Saving Performance in the American Dream Demonstration, a National Demonstration of Individual Development Accounts 9,’’ Washington University in St. Louis. Shefrin, Hersh M., and Richard H. Thaler (1992). ‘‘Mental Accounting, Saving, and SelfControl,’’ in G. Loewenstein and J. Elster (eds.), Choice Over Time. New York: Sage Foundation. Sherraden, Michael (1991). Assets and the Poor: A New American Welfare Policy. Armonk, N.Y.: M.E. Sharpe. Sherraden, Michael, and Michael Barr (2004). ‘‘Institutions and Inclusion in Saving Policy,’’ working paper no. BABC 04-15, http://www.jchs.harvard.edu/publications/ ﬁnance/babc/babc_04-15.pdf. Smeeding, Timothy M., Katherin R. Phillips, and Michael O’Conner (2000). ‘‘The Earned Income Tax Credit: Expectation, Knowledge, Use, and Economic and Social Mobility,’’ National Tax Journal, 53(4):1187–1209. Thaler, Richard H. (1994). ‘‘Psychology and Savings Policies,’’ American Economic Review, 84(2):186–192. Thaler, Richard H. (2000). ‘‘From Homo Economicus to Homo Sapiens,’’ Journal of Economic Perspectives, 14(1):133–141. Thaler, Richard H., and Shlomo Benartzi (2004). ‘‘Save More Tomorrow TM : Using Behavioral Economics to Increase Employee Saving,’’ Journal of Political Economy, 112(1):164– 187. Thaler, Richard H., and H. M. Shefrin (1988). ‘‘The Behavioral Life-Cycle Hypothesis,’’ Economic Inquiry, 26(4):609–643. Thompson, Ginger (2002). ‘‘Big Mexican Breadwinner: The Migrant Worker,’’ New York Times, 25 March, p. A3. Tufano, Peter, and Daniel Schneider (2005). ‘‘H&R Block and Everyday Financial Services, HBS Case No. 205-013.’’ Boston, Mass.: Harvard Business School Publishing. Tufano, Peter, and Daniel Schneider (2005). ‘‘Reinventing Savings Bonds.’’ Special Report, Tax Notes, 109(5):637–656. U.S. Census Bureau (2000). ‘‘Census 2000 Population, Demographic, and Housing Information, Tulsa, Oklahoma,’’ http://quickfacts.census.gov/qfd/states/40/40143lk.html (accessed September 30, 2004).

162

Beverly, Schneider, and Tufano

Caner, Asena, and Edward Wolff (2002). ‘‘Asset Poverty in the United States, 1984–1999: Evidence from the Panel Survey of Income Dynamics.’’ Working paper no. 356. N.Y.: Levy Economics Institute, Annandale-on-Hudson. Wolff, Edward, and Asena Caner (2004). ‘‘Asset Poverty in the United States, Its Persistence in an Expansionary Economy,’’ public policy brief 76. Annandale-on-Hudson, N.Y.: Levy Economics Institute. Woo, Lillian, F., William Schweke, and David E. Buchholz (2004). ‘‘Hidden in Plain Sight: A Look at the $335 Billion Federal Asset-Building Budget.’’ Washington, D.C.: Corporation for Enterprise Development.

5 Household Ownership of Variable Annuities Jeffrey R. Brown, University of Illinois and NBER James M. Poterba, MIT and NBER

Executive Summary Variable annuities have been one of the most rapidly growing ﬁnancial products of the last two decades. Between 1996 and 2004, nominal sales of variable annuities in the United States more than doubled, from $51 billion to $130 billion. Variable annuities now account for almost two-thirds of annuity sales. The investment returns associated with variable annuities resemble those from mutual funds, and variable annuity buyers can select among a range of asset allocation options. Variable annuities are considered insurance products under the tax law, so buyers are not taxed on their investment returns until they make withdrawals from their variable annuity accounts. This paper describes the tax treatment of variable annuities, presents summary information on their ownership patterns, and explores the importance of several distinct motives for household purchase of variable annuities. The discussion of tax treatment examines the impact of the 2001 and 2003 tax bills on the relative tax treatment of variable annuities and other ﬁnancial products. Household data from the 1998 and 2001 Survey of Consumer Finances show that variable annuity ownership is highly concentrated among high-income and high net wealth sub-groups of the population. Variable annuity ownership is less concentrated, however, than ownership of several other types of ﬁnancial assets. Evidence on the role of tax incentives in encouraging ownership of variable annuities is mixed. The probability of owning a variable annuity rises with the marginal tax rate throughout most of the income distribution, but it is lower for households in the top tax bracket than for those with slightly lower tax rates.

164

1.

Brown and Poterba

Introduction

The shift from deﬁned beneﬁt pension plans to self-directed deﬁned contribution plans, the possibility of reforming Social Security so that it includes personal accounts, and the growth in individual retirement accounts (IRAs) are examples of a broad shift toward greater selfreliance in the provision of retirement income in the United States. In policy and academic discussions of individual retirement security in this new environment, two issues are particularly prominent. One is the role of equity markets in providing for future retirement income. There has been a steady rise in the extent of equity market participation over the past two decades, largely as a result of the growth of mutual funds and the expansion of IRAs and 401(k) plans. The second issue is the decline in life annuitization in retirement, arising primarily from the shift away from automatically annuitized deﬁned beneﬁt (DB) plans and towards deﬁned contribution (DC) plans. Many deﬁned contribution plans do not offer life annuities as a payout option. Despite the signiﬁcance of both equity ownership and annuitization, very little research has focused on the rapid growth during the last two decades of a class of products known as variable annuities, which in principle combine equity ownership and an option to annuitize. Variable annuities were introduced in the mid-1950s to compete with mutual funds. The College Retirement Equity Fund (CREF) offered the ﬁrst variable annuity product. The market for variable annuities remained quite small for several subsequent decades. In the early 1990s, however, the market began to grow rapidly. The American Council of Life Insurers (1999) reports that between 1990 and 1999, gross sales of individual (non-group) variable annuities rose from $3.5 billion to $63 billion. More recent data from the National Association for Variable Annuities (2005) suggest that variable annuity sales have declined since the late 1990s. The National Association for Variable Annuities (NAVA) data show 1999 sales of variable annuities outside pension accounts, almost exclusively individual annuities, of $60 billion, with a decline to $51.3 billion in 2004. Total variable annuity sales in 2004, combining qualiﬁed account and non-qualiﬁed account sales, totaled $129.7 billion. Sales to qualiﬁed accounts were slightly less than 40 percent of the market. Sales of ﬁxed annuities, the annuity products that are most often the subject of economic analysis, totaled $89 billion in 2004. Total assets invested in variable annuity products amounted to

Household Ownership of Variable Annuities

165

$1.12 trillion at the end of 2004, compared with $532 billion in ﬁxed annuities. Individuals may demand variable annuities for at least three, not necessarily exclusive reasons. The ﬁrst is a desire to accumulate wealth at favorable after-tax rates of return. Interest, dividends, and capital gains that accrue on assets held in variable annuity accounts are not taxed until the policyholder receives variable annuity payouts. This provides policyholders with the tax beneﬁts of ‘‘inside buildup,’’ just as in IRAs and 401(k) plans. Gentry and Milano (1998) use both crossstate variation in income tax rates and time series variation in federal rates between 1984 and 1993 to study how taxes affect variable annuity demand. They ﬁnd that aggregate sales of variable annuities are positively correlated with state marginal income tax rates, suggesting that variable annuities are purchased in part to avoid the tax burden on investments in traditional taxable accounts. At the time of their study, there was no publicly available household survey data on variable annuity ownership. A second potential attraction of variable annuities is their insurance component. Variable annuity contracts offer various forms of insurance. A common provision speciﬁes that if the policyholder dies before retirement, heirs receive at least the nominal value of the policy contributions. Milevsky and Posner (2001) use risk-neutral option pricing to value the guaranteed minimum death beneﬁt in variable annuities, and they conclude that in most cases the value of this insurance is quite small. However, many variable annuity contracts offer additional insurance features. We are not aware of any study that examines whether investor characteristics, such as self-reported risk aversion, can explain variable annuity demand. The third motivation for holding variable annuities is the option to convert the contract at some future date to a life annuity that provides an annuitized income stream, with the payouts indexed to the performance of a diversiﬁed investment portfolio. Brown, Mitchell, and Poterba (2001) explore this aspect of annuity demand in a stylized life-cycle model. They conclude that most consumers would ﬁnd it welfare-enhancing to hold at least a portion of their retirement portfolio in an equity-linked annuity product, but they do not examine actual patterns of annuity demand. Historically, very few variable annuity products have been converted into life annuities that pay beneﬁts during retirement.

166

Brown and Poterba

The limited body of research on household demand for variable annuities is explained largely by the lack of data. The 1998 Survey of Consumer Finances was the ﬁrst nationally representative household survey to ask detailed questions about the ownership of variable annuity products. We use both this survey and the following wave, the 2001 Survey of Consumer Finances (SCF), to study the ownership of variable annuity products. This paper is divided into four sections. The ﬁrst describes how variable annuities work, focusing particularly on the tax incentives, the insurance features, and the payout options at retirement. It also provides data on the size of the U.S. variable annuity market. Section 2 explains how the 1998 and 2001 SCF data can be used to analyze the crosssectional determinants of variable annuity ownership. It presents summary statistics on variable annuity ownership in the SCF surveys. The third section compares variable annuity ownership patterns to those of other ﬁnancial assets and also reports on the inter-relationships between ownership of variable annuities and these other assets. The fourth section concludes by sketching several research issues about the demand for variable annuities that our analysis raises but cannot resolve due to data limitations. 2.

The U.S. Variable Annuity Market

Variable annuities combine features of insurance products and mutual fund-style investment accounts. The funds invested in a variable annuity are held in designated subaccounts that are kept separate from the insurance company’s other assets. As a result, and unlike most life insurance or ﬁxed annuity products, the assets are not subject to claims by the insurance company’s creditors should the insurance company become insolvent. Income earned on the annuity investments is taxdeferred until the individual begins making withdrawals. The preferential tax treatment of variable annuities derives from the inclusion of life insurance elements in the contract. Because individuals who hold whole life insurance policies are not taxed on their accruing income, excluding income on annuity policies from taxation preserves comparable treatment of these two asset categories. Variable annuity sales in the United States exploded during the 1990s, and they have remained stable at a high level for the last half decade. Data from the American Council of Life Insurers (1999) suggest that individual (non-group) variable annuity considerations grew at a

Household Ownership of Variable Annuities

167

nominal annual rate of 38 percent between 1990 and 1999. One limitation of most data on sales of variable annuities is that the statistics refer to sales of new variable annuity policies rather than the net purchases of variable annuities. Net purchases are smaller than sales of new policies because of surrenders, withdrawals, and beneﬁt payments from existing policies, and because of ‘‘section 1035 exchanges.’’ This name refers to a provision in the tax code that allows a policyholder to make a direct transfer of accumulated funds in one annuity policy into another annuity policy without creating a taxable event. Within qualiﬁed plans, both mutual funds and variable annuities can be exchanged between vendors without triggering a tax liability. If an individual sells stocks in a taxable account in order to purchase shares of a different company, this exchange would trigger capital gains taxation. With annuities, however, there is no tax consequence. An individual can exchange one company’s product for another’s and the earnings from the original investment will remain tax deferred until the annuity owner withdraws money from the variable annuity contract. There is a substantial divergence between gross and net sales of variable annuity policies. Data for 2004 from NAVA (2005) suggest gross sales of $129.7 billion, and net sales of $40.2 billion. Cerulli Associates, Inc. (2001) reports that net purchases represented more than half of total variable annuity sales between 1995 and 1997; they declined to only 20 percent of total sales in 2001. Because 1035 exchanges represent a substantial part of the divergence between gross and net sales, to some extent insurance companies are competing for existing, rather than new, variable annuity business. 2.1 The Structure of Variable Annuity Products Variable annuities can be purchased in retirement accounts and outside these accounts. Qualiﬁed annuities are purchased using assets from qualiﬁed retirement plans, such as 401(k) plans. In many cases, such as university employees purchasing annuities through TIAA-CREF, qualiﬁed annuities may be purchased through an employer. Our analysis focuses on annuities purchased outside retirement plans; these are non-qualiﬁed annuities. Most variable annuity providers offer a broad range of sub-accounts in which the assets may be invested. Equity and bond portfolios are the most common options. A buyer may purchase the variable annuity with a single initial premium payment or with a sequence of premium payments over time. Most insurance companies selling variable

168

Brown and Poterba

Table 5.1 Expenses and Insurance Costs, Variable Annuities and Mutual Funds, Weighted by Assets Under Management, 2002 Variable Annuities Investment Objective

Management Expense

Insurance Charge

Total Expenses

Mutual Funds

All

0.57

1.09

1.65

0.92

Balanced

0.61

1.20

1.80

0.79

Corporate bond

0.55

1.13

1.68

0.72

Government bond

0.61

1.30

1.91

0.90

Growth

0.70

1.23

1.93

1.04

Growth and income High yield bond

0.34 0.70

0.81 1.29

1.15 1.99

0.66 1.09

International bond

1.01

1.33

2.34

1.05

International stock

0.84

1.17

2.01

1.11

Source: Authors’ tabulations from 2002 Morningstar Variable Annuities and Mutual Funds databases. Costs and expenses are measured in hundreds of basis points per year.

annuities collect two fees: an investment management fee and an insurance charge. The insurance charge covers insurance beneﬁts associated with the variable annuity. Many variable annuities have front-end retail loads, and there are often surrender penalties that apply if funds are withdrawn before a pre-speciﬁed time period, often seven years. These penalties, known as contingent deferred surrender charges, can be several percentage points of the annuity’s value. Historically, with the notable exceptions of TIAA-CREF and Vanguard, there were very few no-load variable annuities. Cerulli Associates, Inc. (2001) reports, however, that the no-load segment of this market has expanded in recent years. There has also been a shift toward unbundled variable annuities that offer buyers a minimal level of insurance, perhaps only a death beneﬁt, along with the option to purchase additional insurance on an a` la carte basis. The combination of investment management expenses and insurance charges substantially reduces the returns available to variable annuity investors. In 2002, Morningstar reported that the average total expense for variable annuities investing in diversiﬁed portfolios of domestic equities with a growth and income focus was 115 basis points, while that for variable annuities investing in government bonds was 191 basis points. These expenses are substantially larger than those on open-end mutual funds holding similar assets.

Household Ownership of Variable Annuities

169

Table 5.1 presents information on the average expenses and insurance costs for variable annuities by various categories. The data for this table are drawn from Morningstar databases for both variable annuities and mutual funds. The table shows that the asset-weighted average management expense for variable annuities in 2002 was 57 basis points, compared with 92 basis points for all mutual funds. The average variable annuity insurance charge was 109 basis points, however, making total expenses 165 basis points. Only 5 percent of variable annuity contracts have insurance expenses under 75 basis points, whereas 12 percent charge more than 140 basis points. The entries in table 5.1 are asset-weighted so they are somewhat different than other tabulations, such as those in National Association for Variable Annuities (2005), that weight all variable annuities equally. Table 5.1 shows that the management expenses vary by the variable annuity’s investment objective, with the highest charges on international bond and international stock funds. For mutual funds, the funds with these investment objectives also have two of the three highest average expense ratios. There is some variation across investment objectives in the variable annuities’ insurance charge. The cost of the insurance should depend on the investment portfolio, since the value of an option to repay the annuity principal or the highest value of the annuity assets on any policy anniversary depends on portfolio parameters such as the volatility of the underlying assets. The high insurance charge of 130 basis points per year for variable annuities invested in government bonds is puzzling, given that government bonds are a low-risk investment. However, variable annuity contracts are complex and they vary in the precise nature of their insurance component. It is possible that the insurance contracts typically associated with variable annuities that invest in government bonds are more generous than those associated with other asset allocations. The management costs associated with investments in mutual funds or variable annuities can have an important effect on long-run wealth accumulation. To illustrate this, assume that an individual contributes $1,000 to a qualiﬁed account at age 30 and allows the account to grow for 30 years at an average annual nominal return of 10 percent before administrative costs. At age 60, the value of the account will have grown to $13,563 if the expense charge is 92 basis points per year, but to $11,400 if the charge is 165 basis points. Thus, an increase in the expense ratio equal to the difference between the average expense ratio for mutual funds and that for variable annuities reduces wealth at age

170

Brown and Poterba

60 by roughly 16 percent. This calculation assumes that the insurance component of the variable annuity does not affect the purchaser’s wealth at age 60; there are some scenarios in which the insurance would affect the terminal value of the annuity contract. 2.2 The Tax Treatment of Variable Annuities: Accumulation and Payout Phases The opportunity for assets held in variable annuities to grow at the pre-tax rate of return offers investors the potential to generate higher after-tax wealth from variable annuity investments than from traditional taxable investments. The complex tax treatment of withdrawals from variable annuities, however, makes the after-tax return advantage sensitive to the annuity buyer’s payout decisions. If the payout takes place before the annuitant is 59 12 , unless the distribution takes the form of a life annuity, the distribution is subject to income tax on the difference between the payout and the premium, plus a 10 percent penalty tax. Thus, if an individual owned a single-premium variable annuity that was purchased at age 35 for $10,000, and he or she decided to withdraw the total value of the account at age 55, the tax on the proceeds would equal the individual’s marginal federal income tax, plus 10 percent, times the difference between the account value and $10,000. There is no reliable, publicly available data on withdrawals from variable annuities. Limited evidence suggests, however, that funds accumulated in variable annuity accounts are rarely converted to life contingent annuities at retirement. Brown and Warshawsky (2004) report that only about 1 percent of the individuals covered by variable annuity products are receiving payments from these accounts—the rest are still in the accumulation phase. Of course, these statistics do not imply that annuities currently in the accumulation phase will never be annuitized, but they underscore the importance of tracking the behavior of variable annuity owners over time. There are several different ways to receive distributions from a variable annuity, and they are subject to different tax rules. First, the policyholder could choose a lump-sum distribution. In this case, the tax due at the time of the distribution is t ðV PÞ, where t denotes the policyholder’s ordinary income tax rate, V denotes the value of the variable annuity at the time of the distribution, and P denotes the annuity premium. The premium, P, functions just like the purchase price for an asset that is subject to capital gains tax. Note that in this case, there is no annuitization associated with the variable annuity. If the

Household Ownership of Variable Annuities

171

policyholder chooses to take several distributions from the policy, the distributions are fully taxable as ordinary income until the policy’s remaining value falls below P. The early payouts from the policy are assumed to be income, while the later payouts are returns of principal. Second, the policyholder could choose to make periodic withdrawals from the variable annuity account. Such withdrawals are taxed according to an earnings ﬁrst, principal last rule. If the value of the variable annuity account exceeds the annuity’s purchase price at the time of the withdrawal, the withdrawal is fully taxable as ordinary income until the withdrawal reduces the value of the variable annuity contract to less than the purchase price. Withdrawals from an annuity with a value below the purchase price are treated as returns of principal and are not included in taxable income. A third payout structure the policyholder might choose is a stream of variable payouts for a pre-speciﬁed length of time, such as 10 years. In this case, the insurance company ﬁnds the value A0 that satisﬁes the equation: V¼

T X

A0

t¼1

ð1 þ RÞ t

where V is the value of the accumulation; R is the variable annuity’s assumed interest rate, as in Bodie and Pesando (1983); and T is the number of periods over which the annuitant chooses to receive payouts. Variable annuity payouts depend on the returns on the assets that underlie the annuity. A variable annuity is deﬁned by an initial annuity payment A0 , and an updating rule that relates the annuity payout in future periods to the previous payout and the intervening portfolio returns. If the return in each period is denoted by zt , then the updating rule for the annuity payout At is: Atþ1 ¼

At ð1 þ zt Þ ð1 þ RÞ

With a ﬁxed number ðTÞ of variable payouts, the annuitant’s tax in period t is: P TAXt ¼ t At T This formula distributes the premium amount equally across all annuity payouts.

172

Brown and Poterba

Finally, the policyholder could choose a life contingent annuity, which therefore has an unknown number of payments and uncertainty about the payout size. For a life annuity, A0 is determined by solving: V¼

T X A0 St t¼1

ð1 þ RÞ t

where St is the probability that the individual will live to period t, and T is chosen to represent the maximum number of periods over which the annuitant might live. The tax treatment of life-contingent payouts differs from that of certain payouts that are paid over a ﬁxed time period. For life annuities, the IRS speciﬁes an inclusion ratio ðlÞ, which determines the share of annuity payments in each period that must be included in the recipient’s taxable income. The inclusion ratio is designed to measure the fraction of each annuity payout that reﬂects the capital income on the accumulating value of the annuity premium. The inclusion ratio is calculated by ﬁnding the expected number of years over which the annuitant can expect to receive beneﬁts. This period, T 0 , is determined by the Internal Revenue Service (IRS) using the Uniform Life Expectancy Table and the individual annuitant’s age at the time when payouts begin. The inclusion ratio is: l¼1

P T 0 A0

Until T 0 years after the annuity payout begins, the tax payment on each annuity payment is given by TAXt ¼ t l At . After T 0 years, all payouts from the annuity policy are considered taxable income. This tax rule causes a discrete increase in the annuitant’s tax burden, often at an advanced age. Payouts from variable annuities are taxed as ordinary income. Investors who hold variable annuities that invest in corporate equities or other assets that may generate substantial capital gains are therefore giving up the opportunity to receive capital gains tax treatment on the value of their appreciating assets. The difference between the capital gains tax rate and the ordinary income tax rate is therefore a critical determinant of the tax advantage of investing in variable annuities. Consider a simple example of an equity index fund that earns an 8 percent return each year, net of expenses, with 2 percent from dividends and the remaining 6 percent from capital gains. Assume that the tax regime is similar to the one that applied during our sample pe-

Household Ownership of Variable Annuities

173

riod, and that it assigned a notably higher tax rate to equity income than the current U.S. income tax does. In particular, assume that the investor is in a 33 percent marginal tax bracket for ordinary income, and that the statutory long-term capital gains tax rate is 20 percent. Further assume that capital gains are taxed as they accrue. The annual return on this fund is therefore 6.14 percent (6:14 ¼ :67 2 þ :80 6 percent). Now imagine that the investor held the same investments in a variable annuity so that all taxes are deferred until the assets are withdrawn. The investor beneﬁts from tax deferral but loses because the withdrawals are taxed at ordinary income tax rates rather than capital gains rates. Assume that all of the assets in the variable annuity account are withdrawn at once; this is the lump sum distribution option described above. If the net-of-expense return on the variable annuity is the same as that on the mutual fund, then the value of a $1 investment in the variable annuity, after K years, is: VðKÞ ¼ e 0:08K t ðe 0:08K 1Þ ¼ ð1 tÞ e 0:08K þ t With t ¼ :33, this reduces to ð:67Þ e 0:08K þ :33. For the variable annuity to outperform the open-end mutual fund, we need ð:67Þ e 0:08K þ :33 > e 0:0614K , which depends on K. When K is low, the mutual fund results in a larger terminal wealth than the variable annuity. When K ¼ 5, for example, the after-tax value of the taxable mutual fund is 1.36, while the value of the variable annuity is 1.33. At an investment horizon ðKÞ of thirteen years, the advantage switches to the variable annuity. Indeed, if the horizon is forty years, the value of the taxable equity index fund is 11.66, while the after-tax value of the variable annuity is 16.77. This simple analysis may overstate the advantage of a variable annuity because it ignores the ability to use realized capital losses on taxable mutual funds to offset taxable gains, or up to $3,000 each year in ordinary income. Table 5.1 suggests that the expenses associated with a variable annuity will typically exceed those on the equity mutual fund. We can compute the greatest possible amount by which the expense ratio on the variable annuity can exceed that on the mutual fund, such that the after-tax terminal wealth from the variable annuity will exceed that from the mutual fund. These calculations are in the spirit of earlier studies, such as Milevsky and Panyagometh (2001), PricewaterhouseCoopers (2000), Reichenstein (2000), and Toolson (1991), which have compared the after-tax investment returns available in variable annuities and in mutual funds.

174

Brown and Poterba

Table 5.2 Expense Differential (Variable Annuity Expenses Mutual Fund Expenses) Such That Investor Would Accumulate Equal Wealth, by Holding Period (in Basis Points per Year, Assuming 8 Percent Rate of Return) Holding Period (Years)

Pre-2003 Tax Rates

Post-2003 Tax Rates

5

59

145

10

18

98

15

13

63

20

37

36

25 30

55 70

15 1

35

83

14

40

93

25

Source: Authors’ calculations assuming 8% return (2% from dividends, 6% from capital gains). Assumed tax rates for pre-2003 period are 20% for capital gains, 33% for dividends and ordinary income. For post-2003, rates are 15% for capital gain and dividends, and 33% for ordinary income.

Table 5.2 presents our calculations of the relative attractiveness of mutual funds and variable annuities. Under pre-2003 tax rates and assuming that the taxable mutual fund has a post-expense rate of return of 8 percent, at a horizon of twenty years, for example, expense differentials of less than 37 basis points will result in a higher terminal value with the variable annuity. Given an average expense differential between mutual funds and variable annuities of 73 basis points, one would have to hold the investment for thirty-one years in order for the tax advantage of variable annuities to offset the expense differential. The 2003 Jobs and Growth Tax Relief Reconciliation Act reduced the maximum tax rates on both dividends and capital gains to 15 percent. In this case, the value of the maximum expense differential falls substantially and indeed becomes negative for holding periods as long as twenty-nine years, meaning that for shorter holding periods, a taxable account will yield higher account balances even with an identical cost structure. Even with a horizon of forty years, under the new tax rates, variable annuities provide a higher net of tax return only if the expense differential is under 25 basis points. 2.3 Insurance Features of Variable Annuities Variable annuities offer a range of potential insurance features. In particular, if the variable annuity owner dies before converting to a life annuity, the insurance company typically provides a minimum guar-

Household Ownership of Variable Annuities

175

anteed death beneﬁt. Milevsky and Posner (2001) explain that a typical beneﬁt stipulates that at least the original investment will be returned to the estate or the beneﬁciary of the policy, regardless of the performance of the underlying assets in the account. Thus, a variable annuity buyer has a put option that has a nominal strike price equal to their cumulated nominal contributions. Milevsky and Posner (2001) suggest that the put option is the least valuable option that variable annuities provide. Many providers offer a guaranteed death beneﬁt that set beneﬁts at various high water marks, meaning that they lock in some portion of past investment returns. For example, a maximum anniversary value feature guarantees the maximum value that the investment achieves on a speciﬁed date, usually the contract anniversary date. The insurer guarantees to pay out the higher of the value of (1) the purchase price, (2) the highest value on any anniversary date, or (3) the value of the account at the date of death. Alternatively, insurers may offer minimum growth guarantees for the assets held in the variable annuity by promising a death beneﬁt that is equal to the actual account balance or the value of the premiums compounded at a speciﬁed rate of interest. This particular death beneﬁt is often offered as a rider at additional cost to the annuity buyer. Milevsky and Posner (2001) use option pricing techniques to compute the actuarially fair value of the insurance component of these guarantees. They ﬁnd that ‘‘a simple return of premium death beneﬁt is worth between one to ten basis points, depending on purchase age. In contrast to this number, the insurance industry is charging a median Mortality and Expense Risk charge of 115 basis points, although the numbers do vary widely for different companies and policies.’’ In evaluating this claim, however, one should remember that the ‘‘one to ten basis point’’ valuation is only of the simplest death beneﬁt. In recent years, the array of insurance beneﬁts offered through variable annuity products has become more diverse and complex, with features that are often ﬁrm- or contract-speciﬁc. During 2001 and 2002, after a sharp decline in equity markets, some of these insurance components proved very expensive for insurance companies. Policy provisions that guaranteed variable annuity buyers the value of their portfolio at past policy anniversaries committed insurance ﬁrms to substantial payouts in a declining equity market. For example, Treaster (2003) reports that the Hartford Financial Services Group, which in 2000 paid out only

176

Brown and Poterba

$5.4 million as a result of variable annuity guarantees, faced payouts of $258 million during the bear market of 2002. 3.

Summary Patterns of Variable Annuity Ownership

We explore the cross-sectional patterns of variable annuity ownership using the 1998 and 2001 Surveys of Consumer Finances (SCF). The 1998 SCF is the ﬁrst to distinguish household ownership of variable annuity products from ownership of several other investment products. Beginning in 1998, the SCF asks, ‘‘Do you (or anyone in your family living here) receive income from or have assets in an annuity?’’ Respondents are speciﬁcally told to exclude employment-related pensions as well as any assets that have been recorded earlier in the survey. They are then asked to distinguish between annuities set up to provide only income and those that ‘‘have an equity interest.’’ We identify variable annuity owners as those who report having an equity interest that is invested in ﬁnancial assets, such as stocks, bonds, money market accounts, and real estate. We exclude those who report that their annuity is invested in life insurance/ﬁxed contracts, tangible assets other than real estate, intangible assets, and other assets since these are unlikely to be standard variable annuities. We suspect that our deﬁnition is conservative and that we have excluded some households who hold variable annuities. Using our deﬁnitions of variable annuity ownership, there are 4.8 million variable annuity owners in the 1998 SCF. The total value of the variable annuities reported in the 1998 survey is $255 billion. For the same year, the American Council of Life Insurers (ACLI) (1999) reports that there were 14.6 million variable annuity policies in force, with total asset reserves of $354 billion. National Association for Variable Annuities (NAVA) (2005) reports $343.0 billion of variable annuity assets in non-qualiﬁed accounts for 1999. There is no reason to think that the number of households should match the number of policies because households may have multiple policies even with the same insurer. However, we would ideally want the value of assets in the SCF to match the asset reserves reported by life insurers. The SCF measure is roughly three-quarters of the ACLI or NAVA number. We suspect that this is because some variable annuity owners reported the assets in these accounts elsewhere on the survey, perhaps as other ﬁnancial assets. We are not aware of any evidence suggesting that misreporting

Household Ownership of Variable Annuities

177

rates vary by income, wealth, or age, or in any systematic fashion that might affect our analysis. 3.1 Demographic Patterns of Ownership Table 5.3 presents summary information on the characteristics of households that owned variable annuities in 1998 and 2001. The second and third columns indicate the percentage of households with various characteristics that own variable annuities, while the fourth and ﬁfth columns show the percentage of all variable annuities that are owned by households in each category. Columns two and three show that just under 4 percent of households reported owning a variable annuity in 1998. By 2001, this ﬁgure had increased to 4.65 percent. The small fraction of households owning variable annuities, and the high correlation between variables such as household income and net worth, makes it difﬁcult to obtain robust ﬁndings when we carry out multivariate statistical analysis. Therefore, we restrict our analysis in this paper to univariate analysis, which still provides valuable evidence on the large and growing variable annuity market. Table 5.3 indicates that variable annuity ownership is highly correlated with income and net worth. In the bottom half of the income distribution, for example, just over 2 percent of the population own variable annuities. In the top decile, the ownership rate is over 10 percent in 2001. Ownership is even more highly concentrated by networth deciles, with 16 percent of the top net-worth decile owning variable annuities. From 1998 to 2001, overall growth in ownership rates appears to be largely concentrated at the top of the net-worth distribution, particularly in the top quintile. Older households are more likely to own variable annuities. Less than 2 percent of households under the age of 45 own variable annuities. This rate rises to 5.7 percent in the pre-retirement ages of 45–64, and nearly 9 percent for age 65þ households. Variable annuity ownership is also steeply rising with education level, with 12.45 percent of households with more than a college education reporting ownership of an annuity, compared with less than 3 percent of those with a high school education or less. Variable annuity ownership is highly concentrated among highincome and high-net-worth groups. In 2001, 38 percent of variable annuities were held by households in the top decile of the income distribution, and more than half were held by those in the two top deciles.

178

Brown and Poterba

Table 5.3 Summary Statistics on Variable Annuity Ownership, 1998 and 2001

All households

Percentage of Households Owning Variable Annuities

Percentage of Variable Annuities Held by Households in Each Category

1998

1998

2001

100.0%

100.0%

2001

3.98%

4.65%

<35

1.02

1.74

3.5

4.5

35–44 45–64

2.38 4.50

2.06 5.68

15.2 51.5

6.5 39.3

65þ

8.13

8.88

29.9

49.7

Households grouped by age

Households grouped by education
1.16

3.41

3.9

3.5

High school

3.07

2.13

9.8

14.9

Some college

3.22

4.11

17.2

22.9

College

3.96

5.48

21.8

27.8

12.45

47.4

30.9

>College 12.28 Households grouped by income decile Lowest decile

1.76

0.53

1.3

1.5

Decile 2

1.86

2.22

2.2

3.5

Decile 3

0.97

1.06

1.7

2.5

Decile 4

3.48

3.67

5.8

1.6

Decile 5

3.45

3.28

5.8

3.7

Decile 6

2.82

6.01

4.7

14.4

Decile 7 Decile 8

3.43 7.03

5.77 6.71

3.7 18.0

12.7 8.4

Decile 9

5.61

7.52

8.1

14.9

Highest decile

9.50

10.43

48.7

38.1

Households grouped by wealth decile Lowest decile

0.45

0.00

0.0

0

Decile 2

0.00

0.00

0.0

0

Decile 3

0.43

0.36

0.0

0

Decile 4 Decile 5

0.00 2.23

1.06 1.75

0.0 0.5

0 0.2

Decile 6

2.18

3.11

0.9

1.3

Decile 7

5.92

3.14

4.9

0.7

Decile 8

6.91

9.01

9.6

9.0

Decile 9

7.61

12.23

12.9

16.1

Decile 10

13.81

15.93

71.1

72.7 (continued)

Household Ownership of Variable Annuities

179

Table 5.3 (continued) Percentage of Households Owning Variable Annuities

Percentage of Variable Annuities Held by Households in Each Category

1998

1998

2001

Households grouped by Marginal Federal Income Tax Rate (MTR) MTR < .075 2.82 2.50 10.5 .075 < MTR < .215

2.05

2001 6.3

3.50

8.9

23.7 42.2

.215 < MTR < .299

6.43

6.24

37.9

.299 < MTR < .350

9.33

11.00

14.3

12.7

.350 < MTR

9.71

8.27

28.4

15.1

Households grouped by level of risk tolerance ‘‘Substantial’’ risk

6.40

5.36

4.77

3.42

Above average risk Average risk

5.02 5.76

6.31 6.66

20.46 60.14

29.05 58.31

No risk

1.41

1.89

14.64

9.22

Source: Authors’ tabulations using 1998 and 2001 Survey of Consumer Finances.

Since many retired households may have current income that does not reﬂect their lifetime earnings position, ranking by current income may provide an incomplete indicator of the concentration of variable annuity holdings. The statistics on the net worth of variable annuity holders may be more revealing. More than 70 percent of variable annuities are held by households in the top 10 percent of the wealth distribution, and only 15 percent are held by households who are not in the top ﬁfth of the wealth distribution. 3.2 Marginal Tax Rates and Variable Annuity Ownership Table 5.3 also stratiﬁes households by their marginal federal income tax rate and then tabulates the probability of owning a variable annuity. Our tax rate variable is the marginal income tax rate on ordinary investment income for each household in the 1998 and 2001 SCF. We use an updated version of the algorithm developed by Poterba and Samwick (2003), which estimates a ‘‘ﬁrst dollar’’ marginal tax rate on investment income. The algorithm was developed for use with all of the available Surveys of Consumer Finances, including those from the 1980s. The set of variables that might be used to compute tax rates for SCF households varies over time and has become more elaborate in recent surveys. In particular, beginning in 1995, SCF respondents were

180

Brown and Poterba

asked if they itemized deductions on their income taxes. Recent surveys also include information on adjusted gross income (AGI) that was not recorded in early surveys. We are currently in the process of updating the tax rate algorithm to incorporate this information. The present analysis, however, is the same as that in Poterba and Samwick (2003); it does not utilize the reported information on itemization status or taxable income. The tax rate is computed in two steps. First, we set interest and dividend income to zero and ﬁnd the household’s federal income tax payment. Then we assume that the household receives interest income equal to the maximum of $100, or 5 percent of its total ﬁnancial assets, and we recompute its tax liability. The marginal tax rate on investment income is then deﬁned as the difference in the tax liability divided by the amount of investment income imputed to the household. Our tax rate algorithm uses SCF data to impute as many items on the 1040 tax form as possible. Filing status is determined by the household’s marital status, with all married households assumed to ﬁle a joint return. Personal exemptions are estimated based on marital status and the number of dependents in the household under age 18. The SCF reports information on many of the components of total income. Wages and salaries, tax-exempt interest, alimony received, rents and royalties, business income, and farm income are all deﬁned similarly in the SCF and for tax purposes. Unfortunately, many other income and deduction items, such as IRA distributions and refunds of state and local taxes, are not reported in the SCF. We make several calculations and imputations to estimate adjustments to total income, which in turn affect tax liability. Selfemployment tax applies to all business and farm income. IRA and Keogh contributions can be imputed based on information in the survey, but we set these contributions to zero in computing our marginal tax rates. The SCF also includes data on alimony paid, and this is an adjustment to income. There is no data on other adjustments that are allowed on form 1040, such as moving expenses, so we set these items to zero. Subtracting the total adjustments from total income gives the household’s AGI. We also estimate whether each household will itemize deductions on Schedule A. The SCF reliably reports information on interest payments and charitable contributions. Deductions for local taxes are based on the reported value of real estate and personal property subject to tax. Itemization is determined by comparing the sum of these deductions to the standard deduction appropriate for the house-

Household Ownership of Variable Annuities

181

hold’s age and ﬁling status. The lack of reported information on other possible deductions, such as medical expenses, state and local income taxes, casualty losses, and job expenses, is the biggest handicap in using this algorithm to calculate marginal tax rates in the SCF. The household’s exemptions and deductions are then subjected to the applicable income-based limits, and they are subtracted from AGI to compute taxable income. Applying the appropriate tax rate schedule to taxable income gives the household’s tax liability. Total taxes equal this liability measure, plus self-employment taxes and alternative minimum taxes. We did not compute tax credits since the SCF does not contain the information needed to evaluate most of them. The rows in the ﬁfth panel of table 5.3 present the results, with households stratiﬁed by marginal tax rates. In both 1998 and 2001, the probability of owning a variable annuity is higher for households in high marginal tax brackets than for those in low tax brackets. In both years, the lowest probabilities of ownership are for those with tax rates of less than 21.5 percent. These households do not have ownership rates exceeding 3.5 percent. Most of these households would either be in the 15 percent income tax bracket or would be in a zero tax bracket group. In 2001, households facing the highest income tax rates, those above 35 percent, have a lower probability of owning variable annuities (8.3 percent) than households with tax rates just below the top range (11 percent). The pattern is different in 1998, when the highest probability of owning a variable annuity is observed among households with the highest marginal tax rates. The last two columns and last four rows of table 5.3 show the fraction of variable annuities held by households in different tax rate categories. The 1998 data show that more than one-quarter of all variable annuities are held by households with marginal tax rates of 35 percent or greater. Only 10 percent of these assets are held by households who are assigned very low marginal tax rates by our algorithm. The results for 2001 suggest a shift in the concentration of variable annuity ownership toward lower-income tax brackets. Only 15 percent of variable annuities are held by those with marginal tax rates of 35 percent or above, while there is an increase in the share of variable annuities that are reported by households with tax rates between 7.5 and 21.5 percent. The share of variable annuities held by those with very low marginal tax rates was lower in 2001 than in 1998. One difﬁculty in evaluating the results on marginal tax rates and variable annuity ownership patterns is that retired households may

182

Brown and Poterba

have low marginal tax rates, even if their lifetime income placed them in higher marginal tax brackets when they purchased their variable annuity contract. To explore this issue, we stratiﬁed households by age of the household head and then repeated our analysis of the ownership probabilities by marginal tax rates. For the 45–64 age group in 2001, the age group for which variable annuity ownership becomes substantial, there is a monotonic relationship between marginal tax rate and variable annuity ownership probability. For households with a marginal tax rate of 35 percent or greater, the ownership probability is 11.8 percent. For those in the 30–35 percent tax rate category, this probability is 10.8 percent, while for those between the 21.5 and 30 percent marginal tax rates, it is 7.6 percent. For older households, those headed by someone aged 65 or older, the variable annuity ownership probability peaks in the 30–35 percent marginal tax rate category, where the ownership rate is 19.3 percent. For the elderly households in the highest marginal tax rate category, the ownership probability is 5.2 percent. Similar declines are seen in the ownership probability at the highest marginal tax rate category for those in the 35–44 and the under 35 age groups. 3.3 Risk Aversion and Variable Annuity Ownership A household’s risk aversion may affect its demand for the insurance component of variable annuity products. We test for a positive association between self-reported risk aversion and variable annuity holdings using the responses to the following question in the SCF: Which of the following statements on this page comes closest to the amount of ﬁnancial risk that you and your (spouse/partner) are willing to take when you save or make investments? 1. 2. 3. 4.

Take substantial ﬁnancial risks expecting to earn substantial returns. Take above-average ﬁnancial risks expecting to earn above-average returns. Take average ﬁnancial risks expecting to earn average returns. Not willing to take any ﬁnancial risk.

These measures have been used in Weisbenner’s (2002) study of stock ownership. We deﬁne four indicator variables corresponding to each of the four responses above. The results suggest that the probability of owning a variable annuity is much lower for households that are not willing to take any ﬁnancial risk than for households that are willing to take average, above-average, or substantial ﬁnancial risk. The ownership probability is roughly four percentage points higher for those in these three cate-

Household Ownership of Variable Annuities

183

gories than for those in the not willing to take risk category. Most households are in the average or the above-average risk tolerance categories. Those in the average risk group own 60 percent of variable annuities. Those in the above-average risk category own 21 percent of variable annuities in 1998, and 29 percent in 2001. 4. Ownership Patterns for Variable Annuities Compared with Other Assets While variable annuity ownership is a strongly increasing function of income, net worth, age, and education, this is true for most ﬁnancial instruments. Table 5.4 presents information on the probabilities of holding several ﬁnancial asset classes other than variable annuities. These asset classes include taxable bonds, corporate stock, mutual funds, and tax-free assets such as tax-exempt bonds and mutual funds. The sharply rising probability of asset ownership by income and networth categories is evident for taxable bonds and tax-exempt bonds as well as for variable annuities. For example, in 2001, the top income decile’s ownership of variable annuities was approximately twenty times that of the bottom decile, while the analogous statistic was twenty-six for bonds, twenty-ﬁve for stocks, eighteen for mutual funds, and twenty-two for tax-exempt assets. The wealth-ownership proﬁle is also steeply rising for variable annuities and other asset classes. The age-ownership proﬁle for variable annuities is much steeper than that for stocks or mutual funds. There is a four-fold increase in the ownership probability for variable annuities between ages 35–44 and age 65þ, compared with an increase of less than 20 percent for mutual funds. Table 5.4 shows clearly that the variable annuity ownership probability is substantially lower than the analogous probabilities for stocks or mutual funds but that it is comparable to the ownership probability for both taxable and tax-exempt bonds. Table 5.5 presents information from the 2001 SCF on the percentage of various asset classes that are held by households at different points in the age, wealth, income, education, and marginal tax rate distribution. The table shows that variable annuity ownership is less concentrated than the ownership of the other asset types. For example, 73 percent of variable annuities are held by households in the top decile of the wealth distribution, compared with more than 96 percent of taxable bonds, 90 percent of corporate stock, and 90 percent of tax-exempt

184

Brown and Poterba

Table 5.4 Probability of Owning Variable Annuities and Other Assets, 2001 Survey of Consumer Finances Variable Annuities Total

Taxable Bonds

Corporate Stock

Mutual Funds

TaxExempt Bonds

4.65%

3.00%

21.75%

17.90%

6.75%

<35 35–44

1.74 2.06

0.40 2.04

17.57 21.76

11.50 17.64

2.66 5.17

45–64

5.68

4.12

24.50

20.90

8.96

65þ

8.88

5.00

21.84

20.26

9.30

Households grouped by age

Households grouped by education
3.41

0.28

6.37

3.41

1.40

High school

2.13

0.89

13.15

12.44

3.72

Some college

4.11

3.06

21.28

14.97

6.34

5.48 12.45

5.45 8.44

38.54 41.84

29.43 41.31

10.31 17.61

College >College

Households grouped by income deciles 1

0.53

0.49

2.50

2.78

1.08

2

2.22

0.22

6.01

4.90

2.28

3

1.06

1.44

9.20

6.87

2.50

4

3.67

1.45

16.20

13.81

5.40

5

3.28

1.35

14.76

12.92

4.28

6 7

6.01 5.77

0.81 2.86

18.28 23.01

17.98 17.44

3.98 4.63

8

6.71

5.73

30.32

25.10

10.11

9

7.52

3.27

38.65

28.88

10.59

10

10.43

12.62

61.26

50.25

23.61

Households grouped by wealth deciles 1

0.0

0.0

6.88

2.08

1.74

2

0.0

0.0

0.48

0.0

0.0

3 4

0.36 1.06

0.0 0.24

4.00 4.86

0.34 4.48

0.20 2.48

5

1.75

0.76

11.86

11.42

2.23

6

3.11

0.74

21.89

12.91

3.71

7

3.14

1.54

21.68

17.73

5.81

8

9.01

2.22

31.47

28.74

6.11

9

12.23

5.83

47.87

44.24

15.82

10

15.93

18.72

66.79

57.27

29.53 (continued)

Household Ownership of Variable Annuities

185

Table 5.4 (continued) Variable Annuities

Taxable Bonds

Corporate Stock

Mutual Funds

TaxExempt Bonds

Households grouped by federal marginal income tax rate MTR < .075 .075 < MTR < .215

2.50 3.50

1.38 1.82

7.20 15.20

6.44 12.83

2.61 4.36

.215 < MTR < .299

6.24

3.60

31.51

24.41

9.26

.299 < MTR < .350

11.00

6.76

46.73

47.24

14.38

8.27

13.14

60.59

45.14

23.48

.350 < MTR

Households grouped by self-reported risk tolerance Substantial risk

5.36

2.89

32.77

25.50

5.08

Above average risk

6.31

4.22

39.53

32.14

9.79

Average risk No risk

6.66 1.89

4.28 1.22

28.00 6.27

23.21 5.35

9.22 3.19

Source: Authors’ tabulations using 2001 Survey of Consumer Finances.

bonds. For mutual funds, the most broadly held of the ﬁnancial asset categories, the top tenth of the wealth distribution holds 79 percent. Table 5.5 shows that older households hold a higher fraction of variable annuities than of other ﬁnancial assets. Households headed by someone over the age of 65 hold nearly half of all variable annuities, compared with roughly one-third of the other ﬁnancial asset categories that we consider. The concentration of stock, bond, and tax-exempt bond ownership among the highest marginal income tax rate households is also greater than the analogous concentration for variable annuities. Tables 5.6 and 5.7 explore the cross-ownership patterns between variable annuities, other tax-deferred savings vehicles, stocks, mutual funds, and tax-free assets. If variable annuities are viewed primarily as vehicles for tax-favored asset accumulation, but with higher expense ratios than the investment vehicles that can be held in IRAs, Keogh plans, and other tax-deferred accounts, then a high fraction of variable annuity owners should also hold these accounts. The results in table 5.6, which reports the probability that investors who hold one asset also hold another, offer only limited support for this prediction. Sixty-three percent of the households with variable annuities also hold IRAs or Keoghs. While this is consistent with substantial use of tax-deferred saving vehicles by variable annuity investors, the share of variable annuity holders with IRAs or Keoghs is no

186

Brown and Poterba

Table 5.5 Share of Various Assets Held by Different Population Subgroups, 2001 Survey of Consumer Finances Variable Annuities

Taxable Bonds

Corporate Stock

Mutual Funds

TaxExempt Bonds

3.6%

4.3%

Households grouped by age <35

4.5%

3.6%

6.3%

35–44 45–64

6.5 39.3

4.8 55.5

9.3 47.7

13.0 49.8

5.7 55.1

65þ

49.7

36.2

36.7

33.6

34.9

Households grouped by education
3.5

0.2

1.2

1.3

1.8

High school

14.9

3.2

5.4

12.9

6.7

Some college

22.9

10.9

13.4

12.3

11.7

College

27.8

35.6

37.7

34.2

32.5

50.1

42.3

39.3

47.4

>College 30.9 Households grouped by income deciles 1

0.1

0

1.3

0.2

0.4

2

3.5

0.3

0.1

1.2

0.3

3

2.5

0.7

1.1

1.4

0.3

4

1.6

0.3

1.7

2.8

1.4

5

3.7

0.2

1.4

3.0

2.6

6

14.4

2.0

3.2

6.1

1.6

7 8

12.7 8.4

3.1 5.4

5.4 5.1

5.0 7.7

4.8 6.1

9

14.9

3.6

8.4

12.5

6.3

10

38.1

84.4

73.5

58.9

76.2

Households grouped by wealth deciles 1

0

0

0

0

0

2

0

0

0

0

0

3

0

0

0

0

0

4 5

0 0.2

0 0.1

0 0.1

0 0.3

0.1 0.1

6

1.3

0.1

0.4

0.5

0.3

7

0.7

0.3

0.7

2.0

1.3

8

9.0

0.6

2.1

4.1

1.4

9

16.1

2.2

6.4

13.7

6.9

10

72.7

96.7

90.2

79.4

90.0 (continued)

Household Ownership of Variable Annuities

187

Table 5.5 (continued) Variable Annuities

Taxable Bonds

Corporate Stock

Mutual Funds

TaxExempt Bonds

4.9 16.7

2.3 17.2

Households grouped by Federal Marginal Income Tax Rates (MTR) MTR < .075 .075 < MTR < .215

6.3 23.7

1.0 12.4

1.6 11.2

.215 < MTR < .299

42.2

16.4

23.8

29.7

18.8

.299 < MTR < .350

12.7

15.0

16.2

16.6

13.5

.350 < MTR

15.1

55.3

47.3

32.0

48.1

Households grouped by self-reported risk tolerance Substantial risk

3.42

4.26

8.90

4.73

3.60

Above average risk

29.05

34.96

37.88

36.71

28.32

Average risk No risk

58.31 9.22

55.99 4.79

45.33 7.89

50.33 8.23

60.04 8.04

Source: Authors’ tabulations using the 2001 Survey of Consumer Finances.

Table 5.6 Cross-Asset Ownership Patterns, 2001

Population Variable annuity

Variable Annuity

IRA/ Keogh

Corporate Stock

Mutual Fund

Tax Free Assets

0.05 1.0

0.32 0.63

0.22 0.42

0.18 0.50

0.07 0.22

IRA/Keogh

0.09

1.0

0.42

0.37

0.13

Corporate stock

0.09

0.62

1.0

0.40

0.18

Mutual fund

0.13

0.65

0.48

1.0

0.25

Tax free assets

0.15

0.63

0.57

0.66

1.0

Notes: Tax free assets include tax-exempt bond and money market funds as well as municipal bond funds. All entries are based on population-weighted tabulations from the 2001 Survey of Consumer Finances.

higher than the share of mutual fund or corporate stock investors. Moreover, only 42 percent of variable annuity investors hold corporate stock, and 22 percent hold tax-free bonds, even though these are the other tax-favored asset classes that one would expect to ﬁnd in the portfolios of investors who are trying to maximize tax-free asset accumulation. Table 5.7 presents information similar to that in table 5.6, but instead of reporting the probability that investors in a given asset class hold another asset, it reports the fraction of assets in a given asset class that

188

Brown and Poterba

Table 5.7 Asset-Weighted Cross-Asset Ownership Patterns, 2001 Variable Annuity

IRA/ Keogh

Corporate Stock

Mutual Fund

Tax Free Assets 0.26

Variable annuity

1.0

0.73

0.57

0.55

IRA/Keogh

0.15

1.0

0.61

0.41

0.25

Corporate stock

0.12

0.76

1.0

0.54

0.42

Mutual fund

0.18

0.77

0.67

1.0

0.51

Tax free assets

0.10

0.84

0.74

0.63

1.0

Notes: Each entry shows the fraction of the asset indicated in the row that is held by households that also hold the asset in the column. Tax free assets include tax-exempt bond and money market funds as well as municipal bond funds. All entries are based on population-weighted tabulations from the 2001 Survey of Consumer Finances.

are held by investors who also hold another asset. Thus, in the ﬁrst row, 73 percent of all variable annuities are held by investors who also hold assets in an IRA or a Keogh. Fifty-seven percent of variable annuity assets are held by households with some holding of corporate stock. The results are broadly similar to those in table 5.6 in that they do not suggest large differences in the share of variable annuity investors and investors in other asset categories who hold assets in tax-deferred accounts. The critical question that such cross-asset ownership probabilities raise is whether households turn to variable annuities after they have exhausted other opportunities for tax-deferred saving. The SCF does not provide data on contributions to tax-deferred saving accounts, so we cannot identify households who are constrained by the contribution limits for these accounts. Without more detailed information on the other options available to each household, it is difﬁcult to test for a hierarchy of investment choices. 5.

Conclusions and Future Directions

This paper documents the rapid growth during the 1990s of the market for variable annuities. The gross volume of annuity sales rose faster than the net volume of sales because many variable annuity contracts were terminated and the assets were transferred to new annuities. We identify two factors, the opportunity for tax deferral and the insurance features of variable annuities, that might contribute to the growth of variable annuities. We then evaluate the importance of these factors using data from the ﬁrst two waves of the Survey of Consumer Fi-

Household Ownership of Variable Annuities

189

nances, which included questions that identify variable annuity holders in the United States. We ﬁnd that variable annuity ownership is strongly increasing with income, wealth, age, and education. We ﬁnd, however, that ownership of variable annuities across the education, income, and net worth distribution is less concentrated than ownership of most other ﬁnancial assets. Compared to other ﬁnancial assets, variable annuities are more heavily concentrated at older ages. With regard to marginal tax rates, we ﬁnd that higher marginal tax rates are associated with a greater probability of variable annuity ownership at low and moderate tax rates but that this monotone progression breaks down for the taxpayers in the highest tax brackets. Our proxy variables for risk aversion suggest that households that report a low tolerance for risk are much less likely to hold variable annuities than are households with greater risk tolerance. This may reﬂect greater willingness of such households to invest in the assets that are held in variable annuities, rather than demand for the insurance component of variable annuities. The SCF and other household surveys do not collect detailed information on the payout phase of variable annuities, in particular the use of lump-sum payouts and annuitized streams. We hope that this data deﬁciency will be addressed in future surveys, and that we can then determine how many households are choosing lump-sum payouts rather than various annuity options. Our analysis offers a useful starting point for the analysis of variable annuity demand, but many issues are left unresolved. For example, as our numerical examples illustrate, the 2003 federal income tax reform substantially reduced the incentive for households to invest in variable annuities relative to alternative investment products. A careful examination of how this tax change affects incentives and behavior may shed additional light on this issue. A second issue involves the separate analysis of the group and the individual markets for variable annuities. Group variable annuities are often purchased as part of employer-based retirement planning, and consequently they may not reﬂect the particular portfolio demands of the taxable households that own them. Individual purchases of variable annuities, however, are more likely to be driven by the speciﬁc after-tax portfolio needs of the purchaser. One useful enhancement of the household-level data on annuity ownership would be distinguishing group and individual annuity purchases. Another issue that requires future analysis is the

190

Brown and Poterba

computation of effective load factors on variable annuities and the comparison between these loads and those on other insurance products. For example, Mitchell et al. (1999) ﬁnd loads on ﬁxed life annuity products of around 15 percent, while Brown and Finkelstein (2004) ﬁnd loads on long-term care insurance as high as 50 percent for men. Computing the loads on variable annuities is more difﬁcult than computing the loads on some other insurance products because the return to a variable annuity investor depends on the investor’s behavior. If the variable annuity is held for many years, and if the payouts are withdrawn as annuity payments, the net after-tax and after-expense return may be substantially greater than if the variable annuity is transferred to another insurance carrier after just a few years. Notes We thank Hui Shan and Amir Suﬁ for outstanding research assistance; Andrew Samwick for assistance with marginal tax rate calculations in the Survey of Consumer Finances; Cynthia Saccocia of Cerulli Associates for helpful discussions and data; Peter Davis, Moshe Milevsky, Peter Merrill, Austin Nichols, Kent Smetters, and participants at the 2004 AEA meeting and the NBER Tax Policy and the Economy meeting for helpful comments; and the Social Security Administration and the Boston College Center for Retirement Research for research funding.

References American Council of Life Insurers (1999). Life Insurers Fact Book. Washington, D.C.: ACLI. Bodie, Zvi, and James Pesando (1983). ‘‘Retirement Annuity Design in an Inﬂationary Climate,’’ in Zvi Bodie and John Shoven (eds.), Financial Aspects of the U.S. Pension System. Chicago, Ill.: University of Chicago Press. Brown, Jeffrey R., and Amy Finkelstein (2004). ‘‘Supply or Demand? Why Is the Market for Long Term Care Insurance So Small?’’ NBER working paper no. 10782. Brown, Jeffrey R., Olivia S. Mitchell, and James M. Poterba (2001). ‘‘The Role of Real Annuities and Indexed Bonds in an Individual Accounts Retirement Program,’’ in John Y. Campbell and Martin Feldstein (eds.), Risk Aspects of Investment-Based Social Security Reform. Chicago, Ill.: University of Chicago Press, 321–360. Brown, Jeffrey R., and Mark J. Warshawsky (2004). ‘‘Longevity Insured Retirement Distributions from Pension Plans: Market and Regulatory Issues,’’ in W. Gale, J. Shoven and M. Warshawsky (eds.), Private Pensions and Public Policies. Washington, D.C.: Brookings Institution, 332–369. Cerulli Associates, Inc. (2001). The State of The Annuity Industry. Boston, Mass.: Cerulli Associates. Gentry, William M., and Joseph Milano (1998). ‘‘Taxes and the Increased Investment in Annuities,’’ NBER working paper no. 6525.

Household Ownership of Variable Annuities

191

Milevsky, Moshe A., and K. Panyagometh (2001). ‘‘Variable Annuities versus Mutual Funds: A Monte-Carlo Analysis of the Options,’’ Financial Services Review, 10:145–161. Milevsky, Moshe A., and Steven E. Posner (2001). ‘‘The Titanic Option: Valuation of the Guaranteed Minimum Death Beneﬁt in Variable Annuities and Mutual Funds,’’ The Journal of Risk and Insurance, 68:93–128. Mitchell, Olivia S., James M. Poterba, Mark Warshawsky, and Jeffrey R. Brown (1999). ‘‘New Evidence on the Money’s Worth of Individual Annuities,’’ American Economic Review, 89(December):1299–1318. National Association for Variable Annuities (2005). 2005 Annuity Fact Book. Reston, Va.: NAVA. Poterba, James M., and Andrew A. Samwick (2003). ‘‘Taxation and Household Portfolio Composition: U.S. Evidence from the 1980s and 1990s,’’ Journal of Public Economics, 87:5– 39. PricewaterhouseCoopers (2000). Variable Annuities and Mutual Fund Investments for Retirement Planning: A Statistical Comparison. Washington, D.C.: PWC National Economic Consulting. Reichenstein, William (2000). ‘‘An Analysis of Non-Qualiﬁed Tax-Deferred Annuities,’’ Journal of Investing, Summer:1–12. Toolson, R. B. (1991). ‘‘Tax Advantaged Investing: Comparing Variable Annuities and Mutual Funds,’’ Journal of Accountancy, May:71–77. Treaster, Joseph B. (2003). ‘‘Mutual Fund Reports: Annuity Death Beneﬁt Starts Paying Its Way,’’ New York Times, April 6, Section 3, 16. Weisbenner, Scott (2002). ‘‘Do Pension Plans with Participant Investment Choice Teach Households to Hold More Equity?’’ Journal of Pension Economics and Finance, 1:223–248.

6 Fiscal and Generational Imbalances: An Update Jagadeesh Gokhale, CATO Institute Kent Smetters, Wharton School

Executive Summary This paper provides an update of the U.S. ﬁscal and generational imbalances that we originally calculated in Gokhale and Smetters (2003) and presents the calculations in several alternative ways. We ﬁnd that a lot has changed in just a few years. In particular, the nation’s ﬁscal imbalance has grown from around $44 trillion as of ﬁscal year-end 2002 to about $63 trillion, mostly due to the recent adoption of the prescription drug bill (Medicare, Part D). The imbalance also grows by more than $1.5 trillion (in inﬂation adjusted terms) each year that action is not taken to reduce it. This imbalance now equals about 8 percent of all future gross domestic product (GDP) and it could, in theory, be eliminated by more than doubling the employer-employee payroll tax from 15.3 percent of wages to over 32 percent immediately and forever—assuming, quite critically, no reduction in labor supply or national saving and capital formation. Massive cuts in government spending would also be required to achieve ﬁscal balance: the total federal ﬁscal imbalance now equals 77.8 percent of non–Social Security and non-Medicare outlays. 1.

Introduction

The oldest baby boomers will attain Social Security’s early retirement age of 62 in 2008, and will become eligible for Medicare beneﬁts by 2011. As this generation enters retirement, the population share of retirees will climb rapidly, increasing from about 20 percent today to 37 percent by 2035. Projected longevity improvements mean that the retiree population share will continue to increase gradually during the

194

Gokhale and Smetters

remainder of this century. This ongoing and irreversible process of population aging in the United States will exert tremendous pressure on government budgets in terms of both their size and composition. Combined with the politically inﬂexible eligibility and beneﬁt rules of entitlement programs, population aging will induce a shift in federal budget priorities from discretionary spending such as defense, infrastructure, education, and research and development to mandatory outlays such as Social Security, Medicare, and Medicaid. If the increase in these mandatory outlays cannot be controlled, maintaining growth in discretionary outlays to keep pace with overall economic growth will require higher taxes. An important additional factor that is likely to cause the share of government in the economy to grow is the rapid projected increase in health-care costs—which have historically grown at a much faster pace than general price inﬂation. We have argued elsewhere that ﬁscal policymaking would become easier if the impending change in federal budget priorities were preceded by an adjustment in our ﬁscal vocabulary—that is, by adopting new measures to gauge the federal government’s ﬁscal health (Gokhale and Smetters 2003).1 Traditional measures—such as annual deﬁcits and debt held by the public projected for a limited number of future years—are not adequate for providing lawmakers with the information necessary for enacting new policies in the presence of the age wave. The backward-looking nature of these measures makes it difﬁcult to gauge whether the future ﬁscal commitments created by laws that Congress enacts are affordable or not. These measures also bias Congress’s decisions, inducing rejection of reforms that could improve the nation’s long-term ﬁscal outlook by undertaking a shortterm sacriﬁce. The two measures that we have proposed in the past are called the ﬁscal imbalance (FI) and the generational imbalance (GI) (Gokhale and Smetters 2003). The most important differences between traditional ﬁscal measures and our proposed measures are that the latter are forward-looking and apply a time discount to future dollar ﬂows. The FI measure equals the current level of debt held by the public (representing past overspending) plus the present discounted value of future federal non-interest expenditures less the present discounted value of future federal receipts.2,3 In other words, FI shows the extent to which current U.S. federal ﬁscal policy is not sustainable. FI equals zero for a sustainable (or balanced) policy—wherein outstanding debt held by the public plus future spending commitments are balanced

Fiscal and Generational Imbalances

195

with future receipts in present value. While FI encompasses all federal programs, it can also be calculated separately for speciﬁc federal programs, including Social Security. The FI measure includes all future federal ﬁnancial shortfalls without a time limit. Of course, it can also be calculated under a ﬁnite time horizon. But truncating the calculation in this way could seriously misstate the size of the total FI because it would ignore the present value of shortfalls accruing outside the particular choice of budget window. Under current U.S. ﬁscal policy, our estimates suggest that even if the federal budget window were extended from the normal ﬁve-year or ten-year window to seventy-ﬁve years (the standard projection window used by the Social Security and Medicare trustees), the projected shortfall would miss over half of the true present value imbalance. Restricting attention to such truncated calculations of ﬁscal shortfalls could signiﬁcantly bias policymaking toward obtaining short-term beneﬁts at the expense of policies with short-term costs but larger long-term gains. This short-term policy bias would make current generations better off at the expense of future ones. Even the FI measure, however, does not fully reﬂect this policy bias. For example, a strict pay-as-you-go ﬁnanced retirement beneﬁt has no effect on either traditional budget measures or on FI since the costs of such a program are, by construction, ﬁnanced out of contemporaneous receipts. Still, such a program would transfer resources toward older people who would receive a beneﬁt without having paid much in taxes when working. Such a program would reduce national savings and increase interest rates, as was pointed out in a seminal work by Feldstein (1974). Under a dynamically efﬁcient economy (one in which the steady-state interest rate exceeds the growth rate), this transfer to older generations is ﬁnanced by younger and future generations, who pay more taxes under this program relative to their beneﬁts in present value. To capture the intergenerational redistributive effects of such payas-you-go policies, we also proposed a second, complementary measure—the generational imbalance (GI). This measure calculates the contribution of past and current generations to FI, that is, the amount of overspending by past and current generations under current law. In other words, whereas FI shows the extent to which current ﬁscal policy is not sustainable, GI measures the amount by which beneﬁts to past and current generations (including prospective beneﬁts of current generations) exceed their tax payments (including prospective tax

196

Gokhale and Smetters

payments by current generations) in present value. The GI measure is also useful in estimating the amount by which such obligations induce a reduction in national saving and capital formation. The GI measure is calculated under projections of taxes and beneﬁts and assuming continuation of current policies throughout the lifetimes of current generations. Therefore, GI can be interpreted as the amount of implicit debt under current ﬁscal policy that past and current generations are passing to future generations, who must ﬁnance it through tax payments in excess of their beneﬁts in present value. The amount of implicit debt can be changed, however, by changing current ﬁscal policy. Most policy changes will affect both GI and FI. As noted earlier, however, a strictly pay-as-you-go-ﬁnanced program—wherein higher beneﬁts to older generations are fully ﬁnanced out of higher taxes levied on working generations—would, by construction, have no impact on FI. But such a program would cause a potentially large increase in GI. Thus, while GI provides important information on the effect of ﬁscal policy on national savings, it also provides a complementary measure of policy sustainability. For instance, one could conceive of policies that are sustainable in a traditional static-scoring sense (i.e., for which FI ¼ 0) but involve a very high implicit debt, as reﬂected in a high value of GI, which would produce unrealistically large tax hikes or beneﬁts cuts. Unfortunately, the GI measure can be cleanly estimated only for certain federal programs whose beneﬁts and taxes can be easily distributed across the recipient population. For such programs, the GI measure indicates the contribution of past and current generations to the program’s total FI. This paper reports updated calculations of the inﬁnite-horizon FI and GI measures. Our calculations are based on long-term federal spending and revenue projections made for the Budget of the United States Government for ﬁscal year 2005, the latest long-term budget projections available to us. We report the calculations—particularly Medicare’s estimates—in several alternative ways, and we report the sensitivity of our results to different economic assumptions. We also report limitedhorizon FI measures over budget windows of ﬁve, ten, twenty-ﬁve, ﬁfty, and seventy-ﬁve years, and we show how those calculations would potentially severely bias ﬁscal policy decision-making. Since the publication of our book (Gokhale and Smetters 2003), the nation’s ﬁscal position has dramatically worsened, even relative to the

Fiscal and Generational Imbalances

197

alarmingly large estimates that we presented two years ago. In particular, the FI has increased from around $44.2 trillion (expressed in constant 2002 dollars) to about $63.3 trillion (expressed in constant 2004 dollars). Restating the 2002 estimate of FI in 2004 dollars makes it equal to $45.9 trillion. About $3.4 trillion of the difference between this and FI as of ﬁscal year-end 2004 ($63.3 trillion) arises from the accrual of interest over two years (calculated in inﬂation adjusted terms). The enactment in 2003 of the prescription drug beneﬁt (Medicare, Part D) adds $24.2 trillion to FI as of ﬁscal year-end 2004 (including one year’s interest cost since enactment). However, the Ofﬁce of Management and Budget’s (OMB’s) more favorable long-term productivity outlook reduced FI on the rest-of-federal-government account by $6.2 trillion, arising mainly from higher non-payroll-tax revenues. The remaining difference is explained by changes in revenue and outlay projections for Social Security and Medicare—especially reductions in Medicare Parts A and B outlays resulting from the introduction of Medicare Part D. The GI measure indicates massive overspending by past and current generations in just the Social Security and Medicare programs— to the tune of $33.6 trillion. Again, this is under the assumption that general-revenue transfers are appropriated by the federal government for Medicare Parts B and D. Alternatively, if general-revenue transfers were viewed as dedicated to the Medicare program, the total GI value for Social Security and Medicare would equal $26.1 trillion. Achieving ﬁscal balance would require either massive tax increases (e.g., more than doubling the employer-employee combined payroll tax immediately and forever) or massive cuts in government outlays, for example, a 77.8 percent immediate and permanent reduction in all non–Social Security and non-Medicare outlays. Such a sharp increase in taxes would likely send the U.S. economy into a tailspin and, therefore, pass along to future generations an economy that is in worse shape than the economy that baby boomers inherited from their parents. A sharp decrease in Social Security, health care, and other beneﬁts, however, could entail signiﬁcant hardship for retirees unless beneﬁts could be reduced in a sufﬁciently progressive manner. The FI and GI measures have now also been published by Social Security’s and Medicare’s trustees in their annual reports, starting with Social Security in 2003 and then both programs in 2004 and 2005. These presentations have been endorsed by the 2003 Social Security advisory board’s technical panel on assumptions and methods (Social

198

Gokhale and Smetters

Security Advisory Board 2003), which is composed of several prominent economists and actuaries outside the Social Security Administration. The calculations reported herein differ from the trustees’ estimates because our calculations are based on long-term budget projections made under the administration’s economic assumptions, whereas the trustees use their own set of assumptions, including a smaller interest rate and a smaller rate of productivity growth. As a result, the imbalances that we report for the Social Security and Medicare programs are actually somewhat smaller than what they ﬁnd. In addition to the Social Security trustees, the Federal Accounting Standards Advisory Board (FASAB) is actively considering ways to broaden the deﬁnition of liabilities associated with social insurance programs for purposes of ﬁnancial reporting by the federal government. Doing so would be consistent with representing more fully the future implications of current laws—such as those of entitlement programs—that prescribe criteria for beneﬁt eligibility and beneﬁt amounts payable to those eligible until such time that Congress acts to change those laws. Finally, the current administration appears to have endorsed, in principle, the formal reporting of future federal obligations and anchoring the legislative budget process on such measures. The ﬁscal year 2006 Budget of the United States Government calls for the following reforms: First, the Administration proposes a point of order against legislation which worsens the long-term unfunded obligation of major entitlements. The speciﬁc programs covered would be those programs with long term actuarial projections, including Social Security, Medicare, Federal civilian and military retirement, veterans disability compensation, and Supplemental Security Income. Additional programs would be added once it becomes feasible to make longterm actuarial estimates for those programs. Second, the Administration proposes new reporting requirements to highlight legislative actions worsening unfunded obligations. These requirements would require the Administration, as part of the President’s Budget, to report on any enacted legislation in the past year that worsens the unfunded obligations of the speciﬁed programs.4

2.

Estimates of U.S. Federal Fiscal Imbalances

This section presents calculations of the U.S. federal government’s ﬁscal imbalances, using the Ofﬁce of Management and Budget’s longrange projections (made through the year 2080) as a starting point,

Fiscal and Generational Imbalances

199

that are consistent with the federal budget for ﬁscal year 2005. Our long-range assumptions underlying our projections include an annual labor productivity growth rate (change in hourly labor compensation) of 1.8 percent per year and a consumer price inﬂation of 2.5 percent per year. Present values are calculated using a discount rate of 3.65 percent per year—consistent with the rates on outstanding thirty-year Treasury securities.5 Table 6.1 presents FI estimates for the entire federal government as well as separately for Social Security, Medicare, and rest-of-federalgovernment account. The federal government’s total ﬁscal imbalance amounts to more than $63 trillion in 2004. Social Security contributes ‘‘only’’ $8 trillion to total federal FI. Total federal FI equals 8.2 percent of the present value of future GDP. Some analysts prefer this measure of the total imbalance because it compares FI to the economy’s resource base. However, because only about half of GDP is subject to taxation, the imbalance-to-GDP ratio measure severely understates the difﬁculty in ﬁnancing such a large ﬁscal imbalance. Indeed, as shown in table 6.1, FI equals to 18.0 percent of all future uncapped payrolls—the present value of Medicare’s tax base, which, unlike Social Security, does not impose a taxable wage ceiling. In other words, even with a zero labor supply elasticity—a heroic assumption that almost all economists would dispute at existing tax rates—balancing the federal government’s intertemporal budget constraint would require more than doubling the employer-employee combined payroll tax of 15.3 percent to more that 33.3 percent permanently and forever. Note, however, that the vast majority of the current 15.3 percent tax rate is levied only on earnings below the wage ceiling. In other words, both a large tax rate increase and a base broadening would be required to achieve ﬁscal balance under this hypothetical policy scenario. More realistically, of course, labor supply would sharply fall in response to such a tax increase (Feldstein 1996, Prescott 2004). We conjecture that federal tax increases alone could never be successful in reducing the federal FI to zero. This view is only strengthened when we consider that many states are facing budget crises of their own due to rising Medicaid costs—ﬁscal imbalances that the calculations reported here ignore.6,7 The extent to which federal taxes can be increased are therefore further limited by the need to increase revenues from the same tax base for state balancing budgets. That suggests

772,260

4.8

Memo: Present Value of GDP (billions of constant 2004 dollars)

2.3

Rest of federal government

790,733

4.9

2.4

8.0 0.7

7.9

1.1

8.3

38,808

18,768

0.7

1.0

Medicare

Panel D: general revenue transfers are dedicated to Medicare Parts B and D

Rest of federal government

Panel C: general revenue transfers are annually appropriated for Medicare Parts B and D Medicare

Social Security

Total ﬁscal imbalance—U.S. federal government

8.2

37,282

Rest of federal government

As a Percentage of the Present Value of GDP

17,997

Medicare

Panel B: general revenue transfers are dedicated to Medicare Parts B and D

63,381 5,805

60,886 5,608

8,352

65,928

Rest of federal government

8,006

63,284

2005

Medicare

Social Security Panel A: general revenue transfers are annually appropriated for Medicare Parts B and D

Total ﬁscal imbalance—U.S. federal government

Present Values in Billions of Constant 2004 Dollars

2004

Fiscal Years

Table 6.1 U.S. Federal Fiscal Imbalance and Its Components Under Alternative Assumptions

812,819

5.0

2.4

0.7

8.1

1.1

8.4

40,369

19,554

5,951

65,875

8,709

68,633

2006

834,656

5.0

2.4

0.7

8.2

1.1

8.5

41,917

20,333

6,071

68,321

9,067

71,317

2007

855,240

5.1

2.5

0.7

8.3

1.1

8.6

43,445

21,101

6,171

70,717

9,422

73,968

2008

874,525

5.1

2.5

0.7

8.4

1.1

8.8

44,988

21,876

6,258

73,122

9,784

76,648

2009

893,283

5.2

2.5

0.7

8.5

1.1

8.9

46,583

22,676

6,339

75,599

10,158

79,417

2010

200 Gokhale and Smetters

Source: Authors’ calculations.

Rest of federal government Memo: Present value of uncapped payrolls (billions of constant 2004 dollars)

Medicare

Panel F: general revenue transfers are dedicated to Medicare Parts B and D 10.6 352,529

5.1 10.8 360,875

5.2

17.6 1.6

17.3 1.6

2.3

18.3

Rest of federal government

2.3

18.0

2005

Medicare

Social Security Panel E: general revenue transfers are annually appropriated for Medicare Parts B and D

Total ﬁscal imbalance—U.S. federal government

As a Percentage of the Present Value of (Uncapped) Payrolls

2004

Fiscal Years

10.9 370,810

5.3

1.6

17.8

2.3

18.5

2006

11.0 380,586

5.3

1.6

18.0

2.4

18.7

2007

11.1 389,750

5.4

1.6

18.1

2.4

19.0

2008

11.3 398,302

5.5

1.6

18.4

2.5

19.2

2009

11.5 406,604

5.6

1.6

18.6

2.5

19.5

2010

Fiscal and Generational Imbalances 201

202

Gokhale and Smetters

federal spending reductions will have to play an important role in resolving the federal government’s ﬁscal imbalance. 2.1 Alternative Presentations of Medicare’s Portion of FI In our book, Gokhale and Smetters (2003), we presented Medicare’s portion of the total FI by ignoring the general revenue transfers received by Medicare Part B (Supplementary Medical Insurance). About 75 percent of Medicare Part B’s outlays are ﬁnanced out of general revenues. Moreover, Medicare Part D (prescription drug coverage), which was enacted after the publication of our book, is entirely ﬁnanced out of general revenue transfers. Some commentaries correctly disputed our representation of the entire burden of Medicare’s general revenue ﬁnancing as Medicare’s ﬁscal imbalance. That’s because Medicare Parts B and D are not intended to be fully ﬁnanced from dedicated federal receipts; to ignore general revenue contributions is to essentially ignore this aspect of current law and therefore to disregard the explicit intent of the U.S. Congress to partly ﬁnance Medicare out of general revenues. Auerbach, Gale, and Orszag (2004), for example, consider several alternative methods of presenting Medicare’s shortfalls. We still believe that the best way of presenting Medicare’s shortfalls is to offset outlays by only its dedicated payroll taxes. The reason for this—based on budget accounting principles and not political or economic ones—is that the reported contribution of any program to the federal government’s overall FI should reﬂect the budgetary savings (reduction in FI) generated by eliminating that program. Of course, we are not advocating Medicare’s elimination. Rather, we favor accounting for Medicare’s contribution to the FI by measuring the total amount of burdens generated by that program. Otherwise, the purpose of the calculations (measuring budgetary costs arising from operating federal programs) would become unclear. Nonetheless, for the sake of completeness and to acknowledge that Congress intended Medicare to be partly ﬁnanced out of general revenues, we present Medicare’s contribution to total federal FI under two alternative views. First—and the approach we prefer—general revenue transfers are ignored by assuming that these transfers are annually appropriated for Medicare Parts B and D. Medicare’s FI in 2004 is about $61 trillion under this perspective. Under the second view, we include general revenue transfers by assuming that they are dedicated to these two subprograms, in which case Medicare’s FI is substantially lower—$18 trillion.

Fiscal and Generational Imbalances

203

Regardless of one’s view of how Medicare’s ﬁnances should be represented, however, total federal FI remains unchanged at $63 trillion, as shown in table 6.1. When general revenue transfers are included as part of Medicare’s ﬁnances, the contribution to FI by the ‘‘rest of federal government’’ simply increases by about $43 trillion, the difference between the two alternative measures of Medicare’s FI.8 2.2 A Growing Fiscal Problem Table 6.1 also shows that the ﬁscal imbalance is growing by about $2 trillion each year, or by about 20 percent of this year’s GDP. Like a corpus of debt, an outstanding total federal FI accrues interest over time.9 Under current estimates, its value will grow from $63.3 trillion at yearend 2004 to $79.4 trillion by year-end 2010 if policies and projections remain unchanged in the interim. However, a seemingly more optimistic view, also shown in table 6.1, indicates that the imbalance grows from 8.2 percent of all future GDP in 2004 to 8.3 percent of all future GDP in 2005. Relative to all future uncapped payroll, FI grows from 18.0 percent in 2004 to 18.3 percent in 2005. The advantage of dividing FI by the present value of all future GDP or uncapped payroll is that this measure accounts for the fact that not only does FI grow over time but GDP and uncapped payrolls grow as well. Indeed, if the economy’s capital stock were exactly at or above the golden rule level—implying that the economy’s interest rate is less than or equal to the economy’s growth rate—the ratio of FI relative to all future GDP (or uncapped payrolls) would not grow over time (see the discussion in the appendix). In that case, of course, federal deﬁcits would not matter either—in fact, reducing national saving would be Pareto improving. The U.S. economy would be in Paul Samuelson’s (1958) hypothetical world in which Ponzi games are feasible in the long run. Empirical studies, however, have rejected the hypothesis that the U.S. economy is dynamically inefﬁcient (e.g., Abel et al. 1989). The time-path of the ratio of FI to GDP or payrolls shown in table 6.1 indicates the trade-off available to policymakers between adopting smaller policy changes (tax increases or beneﬁt reductions) effective immediately, or larger ones that would become effective after some years have passed. Nonetheless, exactly how to report FI’s growth over time—whether as a dollar ﬁgure or relative to the present value of GDP or uncapped payroll—has generated a heated debate. For example, Dr. Paul Krugman, a well-known economist and columnist at the New York Times, has repeatedly criticized President Bush for claiming that Social

204

Gokhale and Smetters

Security’s contribution to FI worsens by about $600 billion each year, as now estimated by the Social Security trustees.10 Dr. Krugman’s argument apparently rests on the fact that the growth of Social Security’s FI relative to the present value of future GDP or payroll does not appear as alarming.11 However, Dr. Krugman can only reject the $600 billion ﬁgure if he also rejects the budget accounting system currently being used by the federal government in favor of the FI and GI metrics. But elsewhere, Krugman has referred to Social Security’s FI estimate, which is now being produced by the Social Security trustees, as a scare tactic.12 His positions, therefore, seem inconsistent to us. Indeed, the president’s claim that Social Security’s problem worsens by $600 billion each year is consistent with the standard deﬁcit language that indicates the dollar amount that the national debt grows each year. The president’s message simply emphasizes the need to look ahead rather than restrict attention to conventional cash-ﬂow deﬁcit as a guidepost for ﬁscal policymakers. 3.

Social Security’s Fiscal and Generational Imbalances

Table 6.2 shows a decomposition of Social Security’s FI into two components—GI, which shows the contribution to FI on account of past and current generations (those age 15 and older during the ﬁscal year being considered), and FI minus GI, which shows the contribution to FI that future generations are scheduled to make under current policies. The ﬁrst row of table 6.2 repeats Social Security’s FI shown in table 6.1 (in constant 2004 dollars) for the sake of comparison. The second row of table 6.2 shows the generational imbalance on account of Social Security. As it turns out, Social Security’s GI is larger in present value than its FI, indicating that more than 100 percent of this program’s FI is accounted for by the excess of beneﬁts over payroll taxes in present value scheduled to be awarded to past and living generations. The third row of table 6.2 shows that Social Security’s GI can be decomposed into two parts: the ﬁrst part is the present value of prospective excess beneﬁts over payroll taxes that those age 15 and older will receive. As of ﬁscal year-end 2004, this part equals $11.2 trillion. The second part is the accumulated (present) value of excess beneﬁts paid in the past compared to payroll taxes received by the Social Security system. It includes the present value of excess beneﬁts over payroll

0.44

Future generationsb

0.42

0.53

3.29

2.35 2.77

0.19

0.24

1.50

1.26

1.07

1,547

12,205 1,949

10,256

8,709

2006

0.41

0.56

3.34

2.38 2.79

0.18

0.25

1.53

1.27

1.09

1,543

12,729 2,120

10,609

9,067

2007

0.39

0.59

3.40

2.42 2.81

0.18

0.27

1.55

1.28

1.10

1,535

13,255 2,297

10,958

9,422

2008

Source: Authors’ calculations. a Those born 15 years ago and earlier. In the year 2004, for example, this category includes people born before 1990. b Those born 14 years ago and later. In the year 2004, for example, this category includes people born during 1990 and later.

0.43

3.24 0.50

3.17 0.46

Future net beneﬁts of living generationsa

2.31 2.74

Trust fund

2.27 2.71

Total ﬁscal imbalance in Social Security Past and living generations (GI)

0.20

0.20

Future generationsb

As a Percentage of the Present Value of (Uncapped) Payrolls

1.48 0.23

1.45

Future net beneﬁts of living generationsa

1.25

1.06

0.21

1.24

1,547

11,686 1,787

9,899

8,352

2005

Trust fund

1.04

Past and living generations (GI)

1,543

11,182 1,634

9,549

8,006

2004

Fiscal Years

Total ﬁscal imbalance in Social Security

As a Percentage of the Present Value of GDP

Future generationsb

Future net beneﬁts of living generationsa Trust fund

Past and living generations (GI)

Total ﬁscal imbalance in Social Security

Present Values in Billions of Constant 2004 Dollars

Table 6.2 Social Security’s Fiscal and Generational Imbalances

0.38

0.62

3.46

2.46 2.84

0.17

0.28

1.58

1.29

1.12

1,527

13,787 2,476

11,310

9,784

2009

0.37

0.65

3.53

2.50 2.87

0.17

0.30

1.61

1.31

1.14

1,518

14,338 2,662

11,676

10,158

2010

Fiscal and Generational Imbalances 205

206

Gokhale and Smetters

taxes paid to those alive today (age 15 and older) and those no longer alive since the system’s inception in 1935. As of 2004, this value is negative $1.6 trillion, indicating past accruals of payroll tax surpluses in the Social Security Trust Fund. Adding these two parts yields the ﬁscal year-end 2004 value of GI—$9.5 trillion. Because Social Security’s FI is smaller than its GI, the difference, FI minus GI, is negative. Thus, under current policies, future generations (those age 14 and younger and those that will be born in the future) will pay more in the present value of payroll taxes compared to the present value of their Social Security beneﬁts. The present value of future generations’ excess payroll tax payments equals $1.5 trillion. Despite this overpayment, they will be asked to pay even more (or receive even less)—about $8 trillion more—in order to produce a sustainable system unless Social Security is reformed soon. 4.

Medicare’s Fiscal and Generational Imbalances

In the following discussion we will adopt the convention of representing Medicare’s imbalance under our preferred perspective, which does not assume that general revenue transfers represent a free revenue source to Medicare. Table 6.3 shows FI and GI values for Medicare and its component programs [hospital insurance (Part A), supplementary medical insurance (or Part B), and the prescription drug beneﬁt (Part D)]. Medicare’s overall imbalance equals $60.9 trillion under current policies. Similar to the procedure used by the Center for Medicare and Medicaid Services in making long-range health care projections, the Ofﬁce of Management and Budget’s long-range budget projections assume that future federal health care outlays per capita will grow about 1 percent faster than GDP per capita through the next 75 years. Thereafter, this growth rate wedge is tapered down to equal GDP growth per capita.13 Table 6.4 shows, however, that total (economywide) medical spending per capita has increased by 1.6 percent per year since 1980 and federal health-care outlay growth has averaged 1.8 percent (calculated exponentially) during the same period. This is much faster than assumed in ofﬁcial long-range federal budget projections used to calculate the FI and GI values of table 6.3. That makes the FI and GI estimates reported here considerably more conservative compared to those that would be obtained under a health care growth assumption closer to its historical average.

Fiscal and Generational Imbalances

207

We estimate Medicare’s overall FI to be $60.9 trillion as of ﬁscal yearend 2004. That equals 7.9 percent of GDP—almost equaling the entire federal FI of 8.2 percent. Medicare’s FI equals 17.3 percent of the present value of uncapped payrolls. Despite the very conservative assumption about health care outlay growth, that’s more than seven times larger than Social Security’s FI. Waiting until 2010 to change policies on Medicare’s revenues or beneﬁts would increase the program’s FI to $75.6 trillion—increasing it as a share of GDP to 8.5 percent. Viewed alternatively, the additional resources required through policy changes in 2010 would be equivalent to imposing a tax of 18.6 percent of uncapped payrolls instead of 17.3 percent were the policy change undertaken immediately. Table 6.3 also decomposes Medicare’s FI into those computed on account of its sub-programs. Medicare Part A’s and Part B’s FIs are almost identical—between $18.0 and $19.0 trillion each as of ﬁscal yearend 2004. The Medicare prescription drug program’s FI is larger by 25 percent—valued at $24.0 trillion. A noteworthy difference between Social Security and Medicare is that GI constitutes a much smaller share in FI for Medicare than for Social Security.14 Recall that more than 100 percent of Social Security’s FI is accounted for by generous beneﬁts awarded to past and scheduled for current generations compared to their payroll taxes, whereas future generations are projected to pay more in Social Security payroll taxes than they will receive in beneﬁts. In contrast, Medicare’s GI contributes only two-ﬁfths of its total FI of $60.9 trillion and, under current Medicare tax and beneﬁt rules, future generations are projected to receive $36.8 trillion more in future health care beneﬁts than they will pay in present value of taxes. This result arises because much of Medicare’s large FI is caused by rapid growth in future health care costs and outlays. Indeed, the conservative assumptions used in making future health care outlay projections suggest that these estimates may signiﬁcantly understate Medicare’s FI and signiﬁcantly overstate the percentage contribution of Medicare’s GI to its FI. 5. Comparison with Estimates by the Social Security and Medicare Trustees Table 6.5 compares this paper’s FI and GI estimates with those of Social Security and Medicare’s trustees that are published in their 2005 annual reports. The Social Security program’s FI is estimated at $11.1

295

11,163

11,507

28

21

Trust fund

Future generationsb

7,815

7,787

19,295

10,998

7,467

7,447

18,610

10,629

Future net beneﬁts of living generationsa

Past and living generations

Fiscal imbalance

Medicare Part B

Future generationsb

8,136 268

7,722 261

18,866 7,869

Trust fund

18,090 7,462

37,951

282 36,791

25,726

25,431

63,381

2005

24,376

24,094

60,886

2004

Fiscal Years

Future net beneﬁts of living generationsa

Fiscal imbalance Past and living generations

Medicare Part A

Future generationsb

Trust fund

Future net beneﬁts of living generationsa

Past and living generations

Total ﬁscal imbalance in Medicare (Parts A, B, and D)

Present Values in Billions of Constant 2004 Dollars

Table 6.3 Medicare’s Fiscal and Generational Imbalances

11,847

41

8,177

8,136

19,983

11,365

279

8,572

19,658 8,292

39,097

320

27,098

26,778

65,875

2006

12,171

43

8,537

8,494

20,665

11,719

292

9,014

20,441 8,722

40,190

335

28,466

28,131

68,321

2007

12,479

45

8,894

8,850

21,328

12,058

306

9,461

21,213 9,155

41,232

350

29,835

29,485

70,717

2008

12,783

46

9,255

9,209

21,992

12,393

319

9,918

21,992 9,599

42,261

366

31,227

30,862

73,122

2009

13,094

49

9,629

9,580

22,674

12,735

333

10,394

22,797 10,062

43,310

381

32,670

32,289

75,599

2010

208 Gokhale and Smetters

Trust fund Future generationsb

Future net beneﬁts of living generationsa

Past and living generations

Fiscal imbalance

Medicare Part A

Future generationsb

Trust fund

Future net beneﬁts of living generationsa

Past and living generations

Percentage of the Present Value of GDP Total ﬁscal imbalance in Medicare (Parts A, B, and D)

Future generationsb

Trust fund

Future net beneﬁts of living generationsa

Past and living generations

Fiscal imbalance

Medicare Part D

1.03 0.03 1.39

1.00

1.00

2.39

4.80

0.03 1.38

0.97

2.34

4.76

3.25 0.04

3.16

3.22

8.02

15,446

0

9,775

9,775

25,220

2005

0.04

3.12

7.88

15,000

0

9,186

9,186

24,186

2004

Fiscal Years

0.03 1.40

1.05

1.02

2.42

4.81

0.04

3.33

3.29

8.10

15,885

0

10,349

10,349

26,234

2006

0.03 1.40

1.08

1.04

2.45

4.82

0.04

3.41

3.37

8.19

16,301

0

10,915

10,915

27,216

2007

0.04 1.41

1.11

1.07

2.48

4.82

0.04

3.49

3.45

8.27

16,695

0

11,480

11,480

28,176

2008

0.04 1.42

1.13

1.10

2.51

4.83

0.04

3.57

3.53

8.36

17,084

0

12,054

12,054

29,138

2009

(continued)

0.04 1.43

1.16

1.13

2.55

4.85

0.04

3.66

3.61

8.46

17,480

0

12,647

12,647

30,128

2010

Fiscal and Generational Imbalances 209

Future generationsb

Trust fund

Past and living generations Future net beneﬁts of living generationsa

Total ﬁscal imbalance in Medicare (Parts A, B, and D)

Percentage of the Present Value of Uncapped Payrolls

10.52

0.08

0.08 10.44

7.05 7.13

17.56

1.95

0.00

1.24

1.24

3.19

0.00 1.46

0.99

0.98

2.44

2005

6.83 6.91

17.27

1.94

0.00

Future generationsb

1.19

Trust fund

1.19

3.13

0.00 1.45

0.97

0.96

2.41

2004

Fiscal Years

Future net beneﬁts of living generationsa

Past and living generations

Fiscal imbalance

Medicare Part D

Trust fund Future generationsb

Future net beneﬁts of living generationsa

Past and living generations

Fiscal imbalance

Medicare Part B

Table 6.3 (continued)

10.54

0.09

7.22 7.31

17.77

1.95

0.00

1.27

1.27

3.23

10.56

0.09

7.39 7.48

17.95

1.95

0.00

1.31

1.31

3.26

1.02 0.01 1.46

1.01

1.02

2.48

2007

0.01 1.46

1.00

2.46

2006

10.58

0.09

7.57 7.65

18.14

1.95

0.00

1.34

1.34

3.29

0.01 1.46

1.04

1.03

2.49

2008

10.61

0.09

7.75 7.84

18.36

1.95

0.00

1.38

1.38

3.33

0.01 1.46

1.06

1.05

2.51

2009

10.65

0.09

7.94 8.03

18.59

1.96

0.00

1.42

1.42

3.37

0.01 1.47

1.08

1.07

2.54

2010

210 Gokhale and Smetters

0.00 4.25

2.61

2.61

6.86

3.17

0.00 4.28

2.71

2.71

6.99

3.19

2.17 0.01

2.12

2.16

5.35

3.05

0.01

2.11

5.28

3.01

2.25 0.07

2.19

2.18

5.23

0.07

2.12

5.13

2005

0.00 4.28

2.79

2.79

7.07

3.19

0.01

2.21

2.19

5.39

3.07

0.08

2.31

2.24

5.30

2006

0.00 4.28

2.87

2.87

7.15

3.20

0.01

2.24

2.23

5.43

3.08

0.08

2.37

2.29

5.37

2007

0.00 4.28

2.95

2.95

7.23

3.20

0.01

2.28

2.27

5.47

3.09

0.08

2.43

2.35

5.44

2008

Source: Authors’ calculations. a Those born 15 years ago and earlier. In the year 2004, for example, this category includes people born before 1990. b Those born 14 years ago and later. In the year 2004, for example, this category includes people born during 1990 and later.

Trust fund Future generationsb

Future net beneﬁts of living generationsa

Past and living generations

Fiscal imbalance

Medicare Part D

Future generationsb

Trust fund

Future net beneﬁts of living generationsa

Past and living generations

Medicare Part B Fiscal imbalance

Future generationsb

Trust fund

Future net beneﬁts of living generationsa

Past and living generations

Fiscal imbalance

Medicare Part A

2004

Fiscal Years

0.00 4.29

3.03

3.03

7.32

3.21

0.01

2.32

2.31

5.52

3.11

0.08

2.49

2.41

5.52

2009

0.00 4.30

3.11

3.11

7.41

3.22

0.01

2.37

2.36

5.58

3.13

0.08

2.56

2.47

5.61

2010

Fiscal and Generational Imbalances 211

212

Gokhale and Smetters

Table 6.4 Growth in Health Care Expenditures per Capita, 1980 and 2003 Exponential Growth Rate (percent)

National health expenditures

1980

2003

Nominal

Real

$1,067

$5,670

3.2

1.6

Private

612

3,084

3.1

1.5

Public

455

2,586

3.3

1.8

Federal

310

1,829

3.4

1.8

State and local

146

757

3.2

1.6

52

605

4.7

3.2

37,176 188.9

2.1

Prescription drugs Memo items: GDP per capita Consumer Price Index

12,130 82.4

0.6 1.6

Source: Authors’ calculations based on data from the Centers for Medicare and Medicaid Services (see http://www.cms.hhs.gov/statistics/nhe/historical/t1.asp). Figures for the Consumer Price Index (CPI-U, current series) are taken from the Bureau of Labor Statistics’ web site.

Table 6.5 Comparison with Ofﬁcial Estimates for Social Security and Medicare (Present Values in Trillions of Constant 2004 Dollars) Ours Social Security FI

Social Security and Medicare Trustees

8.0

11.1

9.5

12.0

FI

18.1

24.1

GI

7.5

9.4

18.6

25.8

7.4

9.7

FI

24.2

18.2

GI

9.2

6.7

GI Medicare Part A

Medicare Part B FI GI Medicare Part D

Fiscal and Generational Imbalances

213

trillion by the trustees, whereas it is just $9.5 trillion under the economic assumptions of the Ofﬁce of Management and Budget. When we lower the discount rate from 3.65 percent to 3.30 percent, Social Security’s FI increases to $11.5 trillion—higher than the trustees’ estimate. That is, adopting the trustees’ discount rate assumption would result in an even higher estimate of that program’s unfunded obligation. Why the difference? The answer is OMB’s higher productivity growth rate assumption—1.8 percent per year compared to 1.6 percent. Faster economic growth results in higher future tax revenues but also larger beneﬁt obligations because of Social Security beneﬁts. As it turns out, faster economic growth increases rather than reduces Social Security’s unfunded obligations. In the words of the program’s trustees: While faster real wage growth . . . results in increased tax revenue somewhat before it increases beneﬁt levels, the cumulative additional growth in wage levels eventually results in greater dollar increases in the relatively large projected cost of the OASDI [Old Age, Survivors, and Disability Insurance] program than in the smaller projected tax revenues. Thus, eventually, faster real wage growth, alone, results in an increase in the unfunded obligation of the program.15

The Medicare trustees’ estimate of the inﬁnite horizon unfunded obligations for Medicare Part A (hospital insurance) equals $24.1 trillion, much higher than our estimate of $18.1 trillion. However, the proportion of FI contributed by GI is about 40 percent under both sets of estimates. For Medicare Part B, the trustees report an unfunded obligation of zero. That’s because their reporting convention counts general revenue transfers to Medicare Part B as dedicated rather than appropriated for the program. Using our preferred approach of viewing general revenue transfers as appropriated, Medicare Part B’s unfunded obligation equals $25.8 trillion. Again, GI contributes just under 40 percent of Medicare Part B’s unfunded obligations under both sets of estimates. Estimates of Medicare Part D (prescription drug coverage), shown in table 6.5, differ considerably. The trustees’ estimate is smaller at $18.2 trillion, whereas ours is $24.2 trillion. Our estimate is based on OMB’s projections of growth in prescription drug outlays. Reportedly, these projections are based on higher growth rates through 2040, as seen in ﬁgure 6.1. Again, however, the ratios of GI to FI are quite similar under both sets of estimates.

214

Gokhale and Smetters

Figure 6.1 Medicare Part D’s Projected Outlays as a Percentage of Projected GDP

Except for the magnitude of Medicare Part D outlay projections, the comparison of the two sets of estimates suggests broad agreement regarding the future projections for Social Security and Medicare and their allocation across past and current versus future generations. This is not, of course, surprising because OMB usually receives projections for both programs’ revenues and outlays from their respective administrative agencies based on OMB’s economic assumptions. For Medicare Part D expenditures projected in the ﬁscal year 2005 budget, OMB staff assumed higher outlay growth through the year 2040 (see ﬁgure 6.1).16 These growth rates appear to be consistent with historical growth in economy-wide prescription drug expenditures (see table 6.4). 6.

Estimates Under Alternative Budget Windows

Table 6.6 shows FI for selected budget windows. The last column of table 6.6 repeats the inﬁnite horizon FI measure. It is clear from the numbers that calculating FI over short budget windows signiﬁcantly understates the ﬁnancial shortfall that the federal government faces. For example, the regularly reported budget window for the OMB is ﬁve years into the future. Over this period, the sum of Social Security’s, Medicare’s, and the rest of the federal government’s ﬁscal imbalances amounts to $4.5 trillion. Over the Congressional Budget Ofﬁce’s (CBO’s) regular budget-reporting horizon of ten years, the total

Fiscal and Generational Imbalances

215

Table 6.6 Fiscal Imbalances for Selected Budget Windows as of Fiscal Year-End 2004 (Billions of Dollars) 5 Years Total federal government

10 Years

25 Years

50 Years

75 Years

Inﬁnite Horizon

4,593

4,125

5,185

13,568

23,580

2,051

2,430

2,136

45

1,742

8,006

405

1,178

4,978

15,010

25,282

60,886

Rest of federal government

6,239

5,377

2,343

1,397

3,444

5,608

Rest of federal government—outlays

6,034

11,244

24,416

41,048

52,900

81,323

Social Security Medicare

Rest of federal government—revenues

63,284

6,131

12,203

28,408

48,781

62,680

93,266

Federal liabilities to Social Security

1,915

1,915

1,915

1,915

1,915

1,915

Debt held by the public

4,421

4,421

4,421

4,421

4,421

4,421

Source: Authors’ calculations.

federal imbalance equals $4.1 trillion. Longer horizon ﬁscal imbalances are larger. For example, over the next ﬁfty years, total federal FI equals $13.5 trillion. The short-term estimates of FI are much smaller because they ignore ﬁnancial shortfalls accruing after the budget window’s terminal year. Even the seventy-ﬁve-year FI estimate for the entire federal government equals only about one-third of the FI calculated in perpetuity. 7. FI and GI Estimates Under Alternative Productivity Growth Assumptions Table 6.7 shows estimates of FI for ﬁscal year 2004 under alternative assumptions of productivity growth and discount rates. Variation around the labor-productivity growth assumption equals G50 basis points. Thus, the ‘‘high’’ and ‘‘low’’ productivity growth estimates correspond to labor productivity growth rates of 2.3 percent and 1.3 percent per year, respectively. Variation around the discount rate assumption equals G25 basis points per year. Thus, estimates under high and low discount rates reﬂect discounting at 3.9 and 3.4 percent per year, respectively. Finally, variation around the health care growth wedge assumption equals G50 basis points: estimates under the high and low assumptions reﬂect health care growth rate wedges of 1.5 and 0.5 percentage points, respectively.

Source: Authors’ calculations.

Present value of GDP Present value of payrolls

772,260 348,416

4,421

Debt held by the public

Memo:

1,915

665,833 305,153

4,421

1,915

10,713

11,943

Liability to Social Security and Medicare

PV of excess of outlays over receipts

48,054 4,378

60,886 5,608

49,356 5,679

Rest of federal government

63,284 8,006

918,267 416,991

4,421

1,915

13,400

7,064

79,794

84,236 11,506

999,295 451,181

4,421

1,915

22,722

16,391

89,489

85,795 12,697

626,711 288,189

4,421

1,915

6,497

156

41,951

47,278 5,482

Low

High

High

Low

Productivity Growth

Discount Rate

Medicare

Total ﬁscal imbalance—U.S. federal government Social Security

Baseline Assumptions

Table 6.7 Sensitivity of Fiscal Imbalances To Economic Assumptions (Present Values in Billions of Constant 2004 Dollars)

772,260 352,529

4,421

1,915

9,061

2,726

70,539

75,819 8,006

High

772,260 352,529

4,421

1,915

14,449

8,113

52,530

52,423 8,006

Low

Excess Health Care Outlay Growth per Capita

216 Gokhale and Smetters

Fiscal and Generational Imbalances

217

The FI for ﬁscal year 2004 is quite sensitive to variations in the discount rate. It is estimated to be $84 trillion under the low discount rate (3.4 percent) and $49 trillion under the high one (3.9 percent). The wide variation in FI estimates for small changes in the discount rate is only to be expected because a large share of total federal FI accrues after several decades have passed. Normally, such wide variations in FI arising from small changes in the discount rate are taken as an indication that the FI measure is not reliable or useful. However, as we have argued in Gokhale and Smetters (2003), wide variations in FI triggered by discount rate changes conﬁrm the need to adopt longer-term calculations because they indicate that a large fraction of the imbalance accrues after several decades have passed—a component that would be ignored under truncated horizons. For example, table 6.6 shows that about two-thirds of the total federal FI would arise under current policies after another seventyﬁve years have passed, and it is well known that a given change in the interest rate imposes a larger discount effect on fund ﬂows that occur further out into the future. Table 6.7 shows that FI equals $85.7 trillion under the high productivity growth rate assumption (2.3 percent). Social Security’s ﬁscal imbalance increases from $8.0 trillion under baseline assumptions to $12.7 trillion when high productivity growth assumption is introduced. A considerable increase in FI also emerges in Medicare under the high productivity growth assumption. Note that increasing productivity growth also leads to higher growth in federal health care outlays because those outlays are assumed to grow 1 percentage point faster than growth in output per worker. The opposite result obtains when productivity growth is lowered to 1.3 percent per year. In that case, Medicare’s FI is estimated to be $41.9 trillion. Increasing or reducing the health-care growth wedge also considerably affects the total federal FI. Increasing the wedge by 50 basis points (from 1 percentage point to 1.5 percentage points) increases total federal FI by more than $12 trillion, to $75.8 trillion, and reducing the wedge by 50 basis points (to 0.5 percentage point) reduces federal FI to $52.4 trillion. These wide variations in dollar estimates of FI may make this measure appear to be unsuitable as a guide for policymakers. However, a more stable measure of the size of the federal government’s ﬁnancial shortfall under current policies may be to view it as a ratio to GDP or

218

Gokhale and Smetters

Table 6.8 Sensitivity of Total Federal Fiscal Imbalance as a Percentage of the Present Values of Total Payrolls and GDP GDP

Total Payrolls

Baseline

High

Low

Baseline

High

Low

Discount rate

8.2

7.4

9.2

18.0

16.2

20.2

Productivity growth per capita

8.2

8.6

7.5

18.0

19.0

16.4

Health care outlay growth per capita

8.2

9.8

6.8

18.0

21.5

14.9

Source: Authors’ calculations.

a tax base. When expressed relative to the present value of taxes, this ratio shows the size of the tax increase that would be needed to create a sustainable ﬁscal federal policy. Table 6.8 shows federal FI in perpetuity as a ratio, alternatively, to the present value of GDP and the present value of total payrolls. These ratios exhibit less volatility than dollar estimates because the denominator (the present value of GDP or payrolls) changes in the same direction as does FI in response to changes in each of the three assumptions. For example, although FI under high productivity growth ($85.8 trillion) is roughly double its size under the low productivity assumption ($47.3 trillion), the difference in its ratio to GDP is much less divergent—8.6 under the high-productivity assumption and 7.5 under the low-productivity assumption. Table 6.8 shows that, as a ratio of the present value of payrolls, FI ranges between 16.2 and 19.0 percent in response to the changes in productivity and discount rate assumptions considered here. In sum, while FI expressed in dollars is sensitive to the choice of interest rate and productivity, the size of the policy change itself that is necessary to eliminate the imbalance is fairly stable. The variation in this ratio, however, is much larger under alternative assumptions on the size of the health-care growth rate wedge. Were health care outlays to grow 50 basis points faster immediately and permanently—something that the historical evidence on health care growth suggests is quite feasible (see table 6.4)—resolving the federal ﬁscal imbalance would require appropriating 21.5 percent of all future payroll. In contrast, were it possible to reduce the growth of healthcare costs by 50 basis points, 14.9 percent of future payroll would still be needed to create a sustainable ﬁscal system.

Fiscal and Generational Imbalances

219

There is an interesting difference between the estimates reported in table 6.8 and those reported in Gokhale and Smetters (2003). In our 2003 book, FI as a share of the present value of payrolls was smaller compared to the baseline when productivity growth was assumed to be faster and, symmetrically, was larger when productivity growth was assumed to be slower. However, the estimates reported in Gokhale and Smetters (2003) were made prior to the enactment of sizable additional beneﬁts made available through the Medicare prescription drug law (Medicare Part D). The estimates reported here, however, include the effects of that law. With the addition of the drug beneﬁts, the higher health-care growth rate accompanying the assumption of higher overall productivity growth results in a larger increase in projected beneﬁt outlays compared to the increase in total projected payrolls. The reason is that the prescription drug beneﬁt will begin paying beneﬁts more quickly than the rate at which the payroll tax base grows. 8.

Conclusion

This paper updates calculations of U.S. federal ﬁscal and generational imbalances. The result published in Gokhale and Smetters (2003) of a $44 trillion total federal ﬁscal imbalance as of ﬁscal year-end 2002 is now revised to $63 trillion. A small part of the increase arises from the accrual of interest on the existing ﬁscal imbalance. A large part of the increase comes from the enactment of signiﬁcant additional Medicare beneﬁts through the new prescription drug beneﬁt. That law alone accounts for an increase in FI by $24 trillion. The nation faces an extremely difﬁcult challenge in implementing ﬁscal adjustments to reduce the ﬁscal imbalance built into today’s ﬁscal policies. Given the large magnitude of the overall ﬁscal imbalance, its resolution from higher taxes alone is likely to trigger negative economic effects and does not appear to be feasible. Hence, a sizable part of the adjustment will be required through cuts in discretionary federal outlays and reductions in future entitlement obligations. 9.

Appendix

This Appendix shows how the ratio of Fiscal Imbalance relative to GDP as well as the ratio of Generational Imbalance relative to GDP change over time when there are no changes in ﬁscal policy and projections.

220

Gokhale and Smetters

9.1 Ratio of Fiscal Imbalance to the Present Value of GDP In Appendix A of Gokhale and Smetters (2003) we show [in equation (6.1)] that absent changes in ﬁscal policies and budget projections, the ﬁscal imbalance measure grows at a rate equal to the rate of interest. That is: FItþ1 ¼ FIt R1

ð6:1Þ

Here, FIt stands for the ﬁscal imbalance calculated as of time period t, and R ¼ 1=ð1 þ rÞ stands for the discount factor, with r as the annual interest rate on long-term government debt. Let Yt stand for the discounted present value of GDP as of period t. If annual GDP in year t, yt , grows at rate g per year, and G represents the growth factor 1=ð1 þ gÞ, we can write: Yt ¼

y X

yt

s¼t

st R G

ð6:2Þ

Therefore: Ytþ1

sðtþ1Þ R ¼ ytþ1 G s¼tþ1 y X

¼

y X

yt G1

s¼tþ1

1

¼G

y X s¼t

¼ G1 Yt

sðtþ1Þ R G

st R yt G ð6:3Þ

Divide both sides of equation (6.1) by Yt and manipulate the expression by using equation (6.3) to get: FItþ1 FIt R 1 ¼ ð6:4Þ Ytþ1 Yt G Under normal conditions the economy is dynamically efﬁcient—that is, g < r, implying that G > R. Hence, we can specify in general that: FItþ1 FIt > Ytþ1 Yt

ð6:5Þ

Fiscal and Generational Imbalances

221

That is, absent changes in policy and projections, the ratio of the ﬁscal imbalance to GDP grows larger over time. Thus, the share of GDP that must be devoted to resolving the ﬁscal imbalance increases if corrective policy changes are postponed. 9.2 Ratio of Generational Imbalance to the Present Value of GDP In equation (A9) of our book (Gokhale and Smetters 2003), we show that: R GItþ1 GIt ¼ R NTtþ1

ð6:6Þ

Here, GIt stands for the generational imbalance in period t, and NTt represents the present value lifetime net transfers to those born in period t as scheduled under current ﬁscal policies. Written alternatively: GItþ1 ¼ GIt R1 þ NTtþ1

ð6:7Þ

Equation (6.7) says that next period’s GI equals this period’s GI accumulated at the rate of interest plus the present value of the lifetime net transfer scheduled to be awarded to next period’s newborn cohort under current ﬁscal policies. Dividing both sides of equation (6.7) by Yt and using equation (6.3) to manipulate the expression, we get: GItþ1 GIt R 1 NTtþ1 ¼ þ ð6:8Þ Ytþ1 Yt G Ytþ1 That is, whether the ratio of GI to GDP grows faster, just as fast, or slower than the ratio of FI to GDP depends on whether NTtþ1 v 0. Notes The authors thank James Poterba for very helpful comments. The opinions and conclusions expressed are solely those of the authors and do not necessarily represent the opinions of the Cato Institute or The Wharton School. The authors thank the Ofﬁce of Management and Budget for providing long-term budget projections and Felicitie Bell of the Social Security Administration for providing demographic projections and related underlying assumptions. 1. Others, for example, Auerbach et al. (2004), have also called for adopting ﬁscal measures based on a forward-looking accounting for the federal budget. 2. This measure has also been used by Auerbach, Gale and, Orszag (2004) and has been advocated by Alan Auerbach for over a decade now (e.g., Auerbach 1994). A key difference with our FI measure is that we focus on the implications of current law using a micro-based estimation model. In contrast, these authors alter future policy in directions they regard as realistic by extending aggregate Congressional Budget ofﬁce projections.

222

Gokhale and Smetters

3. The FI measure is also different from the generational balance measure ﬁrst developed by Auerbach, Gokhale, and Kotlikoff (1991). The generational balance concept involves a hypothetical policy whereby future generations are arbitrarily assigned equal additional ﬁscal burdens except for an adjustment for economic growth. That hypothetical policy balances the government’s intertemporal budget but, unlike the FI measure, is not consistent with a budget concept—that is, it does not reﬂect the implications of continuing current ﬁscal policies. 4. See the Analytical Perspectives, Budget of the United States Government, ﬁscal year 2006, Chapter 15. 5. For technical details of our micro-data-based projections and other details, refer to Gokhale and Smetters (2003). Although OMB’s projected long-term interest rates in the ﬁscal year 2005 budget are slightly higher, we use a 3.65 percent annual rate to make present value estimates comparable with those published in Gokhale and Smetters (2003). 6. See, for example, Baker, Besendorfer, and Kotlikoff (2002). The ﬁscal problems that they measured, however, have likely worsened quite dramatically since their study because economic growth forecasts have been reduced and Medicaid cost forecasts have risen. 7. Our calculations only include the federal share of Medicaid costs (under ‘‘rest of federal government’’ in the calculations reported later). 8. We are currently developing the methodology for decomposing the ‘‘rest of federal government’’ account into GI and FI minus GI components, including defense, transportation, Medicaid, etc. We intend to present those results in a new paper. 9. The effective interest accrual on the total federal FI is a combination of the interest accruing on outstanding government debt, Social Security and Medicare trust funds, and the interest rates assumed to prevail during future years. As mentioned earlier, we use in this calculation the Ofﬁce of Management and Budget’s ﬁscal year 2004 assumption of a 3.65 percent interest rate on the longest-maturity Treasury securities outstanding. 10. For the most recent instance of this criticism, see Paul Krugman ‘‘Social Security Lessons,’’ New York Times, August 15, 2005, page 17. 11. Kamin and Kogan (2005) offer a more thoughtful critique, which likely inﬂuenced Krugman’s thinking. 12. See Paul Krugman, ‘‘The $600 Billion Man,’’ New York Times, March 15, 2005, page 25. 13. This does not necessarily imply that aggregate Medicare outlays will grow no faster than GDP since one of the factors driving Medicare and GDP is demographic change. To the extent that the Medicare beneﬁciary population continues to grow faster than the productive population, aggregate Medicare spending would continue to increase at a faster pace relative to GDP. Our calculations indicate that the difference in the growth rates of the two aggregates is extremely small after the ﬁrst seventy-ﬁve years and makes little difference to FI and GI estimates. 14. This assertion is based on the assumption that general revenue transfers are appropriated by Congress for Medicare each year and these funds are not dedicated to the program. 15. See the Social Security Trustees’ (2005) Annual Report, Chapter IV.B.5, paragraph a.

Fiscal and Generational Imbalances

223

16. Based on a phone conversation with a staff member of the Ofﬁce of Management and Budget.

References Abel, Andrew B., N. Gregory Mankiw, Lawrence H. Summers, and Richard J. Zeckhauser (1989). ‘‘Assessing Dynamic Efﬁciency: Theory and Evidence,’’ Review of Economic Studies, 56( January):1–20. Auerbach, Alan J. (1994). ‘‘The U.S. Fiscal Problem: Where We Are, How We Got Here, and Where We’re Going,’’ in Stanley Fischer and Julio Rotemberg (eds.), NBER Macroeconomics Annual, National Bureau of Economic Research. Cambridge, Mass.: MIT Press. Auerbach, Alan J., Jagadeesh Gokhale, and Laurence J. Kotlikoff (1991). ‘‘Generational Accounting: A Meaningful Alternative to Deﬁcit Accounting,’’ Tax Policy and the Economy, 5:55–110. Auerbach, Alan J., William G. Gale, and Peter R. Orszag (2004). ‘‘Sources of the LongTerm Gap,’’ Tax Notes, May 24:1049–1059. Baker Bruce, Daniel Besendorfer, and Laurence J. Kotlikoff (2002). ‘‘Intertemporal State Budgeting,’’ NBER working paper no. 9067. Centers for Medicare and Medicaid Services (2005). National Health Expenditures tables available at http://www.cms.hhs.gov/statistics/nhe/historical/default.asp. Feldstein, Martin (1974). ‘‘Social Security, Induced Retirement and Aggregate Capital Accumulation,’’ Journal of Political Economy, 82:905–926. Feldstein, Martin (1996). ‘‘The Missing Piece in Policy Analysis: Social Security Reform,’’ The Richard T. Ely Lecture, American Economic Review, 86(2):1–14. Gokhale, Jagadeesh, and Kent Smetters (2003). ‘‘Fiscal and Generational Imbalances: New Budget Measures for New Budget Priorities,’’ pamphlet, Washington, D.C.: American Enterprise Institute. Kamin, David, and Richard Kogan (2005). ‘‘The Administration’s Misleading $600 Billion Estimate of the Cost of Waiting to Act on Social Security,’’ Washington, D.C.: Center on Budget and Policy Priorities. Prescott, Edward (2004). ‘‘Why Do Americans Work More Than Europeans?’’ Federal Reserve Bank of Minneapolis Quarterly Review, 28(1):2–13. Samuelson, Paul (1958). ‘‘An Exact Consumption-Loan Model of Interest with or Without the Social Contrivance of Money,’’ Journal of Political Economy, 66(5):467–482. Social Security Advisory Board (2003). ‘‘The 2003 Technical Panel on Assumptions and Methods Report,’’ available at http://www.ssab.gov/NEW/documents/2003Technical PanelRept.pdf. Social Security Trustees (2005). Annual Report, available at http://www.ssa.gov/OACT/ TR/TR05/index.html.

Tax Policy and the Economy, Volume 20 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 9 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 16 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 18 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 12 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Vol. 21 (NBER Tax Policy and the Economy)

Innovation Policy and the Economy (NBER Innovation Policy and the Economy, Volume 6)

Innovation Policy and the Economy, Volume 3 (NBER Innovation Policy and the Economy)

Tax Policy Reform and Economic Growth, OECD Tax Policy Studies

Tax Policy Reform and Economic Growth (OECD Tax Policy Studies)

Tax and the Economy: A Comparative Assessment of Oecd Countries (Oecd Tax Policy Series, 6)

United States Taxes and Tax Policy

Innovation Policy and the Economy, Vol. 2

Industrial Policy and the International Economy

Tax Policy in the Real World

Sell Globally Tax Locally : Sales Tax Reform for the New Economy (AEI Studies on Tax Reform)

Financing government in the transition--Bulgaria: the political economy of tax policies, tax bases, and tax evasion

OECD Tax Policy Studies No. 02: Tax Burdens: Alternative Measures

Tax Ratios: A Critical Survey (Oecd Tax Policy Studies, 5)

Tax Systems and Tax Reforms in Europe (Routledge Studies in the Modern World Economy, 42)

Tax Systems and Tax Reforms in New EU Members (Routledge Studies in the Modern World Economy)

Using Micro-Data to Assess Average Tax Rates: Oecd Tax Policy Studies No. 8 (Oecd Tax Policy Studies)

Tax policy in sub-Saharan Africa, Volume 19

The X-Tax in the World Economy: Going Global with a Simple, Progressive Tax (AEI Studies on Tax Reform)

Entrepreneurship and the Creative Economy: Process, Practice and Policy

The Political Economy of American Monetary Policy

UK Tax Policy and Applied General Equilibrium Analysis

Tax and Corporate Governance

Aid and the Political Economy of Policy Change

Heaven's Door: Immigration Policy and the American Economy

Migration and the Welfare State: Political-Economy Policy Formation

Tax Policy and the Economy, Volume 20 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 9 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 16 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 18 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Volume 12 (NBER Tax Policy and the Economy)

Tax Policy and the Economy, Vol. 21 (NBER Tax Policy and the Economy)

Innovation Policy and the Economy (NBER Innovation Policy and the Economy, Volume 6)

Innovation Policy and the Economy, Volume 3 (NBER Innovation Policy and the Economy)

Tax Policy Reform and Economic Growth, OECD Tax Policy Studies

Tax Policy Reform and Economic Growth (OECD Tax Policy Studies)

Tax and the Economy: A Comparative Assessment of Oecd Countries (Oecd Tax Policy Series, 6)

United States Taxes and Tax Policy

Innovation Policy and the Economy, Vol. 2

Industrial Policy and the International Economy

Tax Policy in the Real World

Sell Globally Tax Locally : Sales Tax Reform for the New Economy (AEI Studies on Tax Reform)

Financing government in the transition--Bulgaria: the political economy of tax policies, tax bases, and tax evasion

OECD Tax Policy Studies No. 02: Tax Burdens: Alternative Measures

Tax Ratios: A Critical Survey (Oecd Tax Policy Studies, 5)

Tax Systems and Tax Reforms in Europe (Routledge Studies in the Modern World Economy, 42)

Tax Systems and Tax Reforms in New EU Members (Routledge Studies in the Modern World Economy)

Using Micro-Data to Assess Average Tax Rates: Oecd Tax Policy Studies No. 8 (Oecd Tax Policy Studies)

Tax policy in sub-Saharan Africa, Volume 19

The X-Tax in the World Economy: Going Global with a Simple, Progressive Tax (AEI Studies on Tax Reform)

Entrepreneurship and the Creative Economy: Process, Practice and Policy

The Political Economy of American Monetary Policy

UK Tax Policy and Applied General Equilibrium Analysis

Tax and Corporate Governance

Aid and the Political Economy of Policy Change

Heaven's Door: Immigration Policy and the American Economy

Migration and the Welfare State: Political-Economy Policy Formation

Recommend Documents