Some Considerations for Empirical Research on Tax-Preferred Savings Accounts.

Some Considerations for Empirical Research on Tax-Preferred Savings Accounts. Kevin Milligan Department of Economics University of British Columbia Prepared for: Frontiers of Public Finance National Tax Association Annual Meetings Orlando FL November 15, 2002 Acknowledgement: This paper is inspired by my dissertation research at the University of Toronto. I thank Dwayne Benjamin, Jack Mintz, and Michael Smart for their advice and guidance as I completed my dissertation. I also thank participants at the National Tax Association meetings in Orlando for helpful comments. 1

1.0 Introduction A vast body of theoretical and empirical research has examined the economics of taxpreferred savings accounts over the last two decades. Accounts such as 401(k) plans, Individual Retirement Accounts and 529 plans for education savings are important components in the financial planning decisions of households. The emergence of this area of research mirrors the growing importance of tax-preferred accounts in the economy as a whole. For example, funds drawn from IRAs and 401(k) plans will provide an increasing share of retirement income in the decades to come. 1 Empirical research seeking to understand the various aspects of behavioral response to these tax policies is important both to further economists understanding of the effects of taxation and to provide a base for pertinent advice to policymakers. Since the pioneering empirical contributions of Hubbard (1984) and Venti and Wise (1986), empirical investigation into the effect of tax-preferred savings accounts on economic behavior was productive, but controversial. The core of the debate concerns the degree to which contributions to tax-preferred savings accounts crowd out savings in other forms; whether the accounts create new savings. Research has studied the saving impact of IRAs, Keogh plans, and 401(k) plans, as well as Canada s Registered Retirement Savings Plans and Registered Home Ownership Savings Plans. Bernheim (2002) offers a complete survey of this literature, concluding that the evidence is not decisive in favor of either side of the debate. Empirical attempts to reconcile the evidence can be found in Poterba, Venti, and Wise (1998a) and Engen and Gale (2000). 2

In any of the approaches discussed in the literature, key assumptions must be made about either the functional form of the structural model estimated, or the behavior of households in response to a natural experiment. These assumptions provide the motivation for the discussion that follows here a re-examination of theory raises questions about some of the key assumptions used in the empirical literature to date. In this paper, I provide some considerations for future empirical research on tax-preferred savings accounts, drawing from the work in my dissertation. 2 After giving a very brief overview of the state of empirical knowledge, I proceed to discuss three issues that can influence the choice of empirical strategy and the interpretation of empirical results. First, I argue that any analysis of the savings response to a reform of tax-preferred savings limits must consider the response within a life-cycle framework. Second, I explain how the presence of multiple tax-preferred savings accounts may confound estimation. Finally, I explore some empirical implications of behavioral economics for estimations of behavioral response to tax-preferred savings accounts. Common to all of the three considerations is the importance of a careful understanding and exploration of the theory that underlies the chosen empirical approach. 2.0 Contribution Room Over the Life-Cycle The life-cycle approach to savings and consumption decisions posits that households accumulate wealth during their working life in order to fund consumption during their 3

retirement years. 3 I adopt the life-cycle approach as an organizing principle for my analysis of tax-preferred savings accounts. Consider a household with two account choices for the location of its savings. One account (the tax-preferred account) offers tax-exempt accrual on capital income while the other account (the taxable account) has its capital income taxed as it accrues. Savings through the tax-preferred account are constrained by a contribution limit, while savings in the taxable account are not. Given this institutional framework, a household will plan a lifetime path for consumption and for saving in the two accounts. Milligan (2003) examines this two-account savings decision. The model shows that households having their desired contributions to the tax-preferred account constrained by the contribution limit in a given period will shift contributions into periods where the constraint is not binding. This generates a use-it-or-lose-it pattern for savings in the tax-preferred account. Contributions in unconstrained periods will be higher in reflection of the binding constraints in other periods, as households use the contribution room while it is available to them. How many households are bound by existing contribution limits? Evidence in Engen, Gale, and Scholz (1994) suggests that 63.3 per cent of IRA contributors were at the family contribution limit over the 1982 to 1986 period relevant for the literature that evaluated the savings impact of IRAs in the 1980s. More recently, Gokhale et al. (2001) estimate that the only households likely to be constrained are high-earnings, late-saving, and late-retiring households. However, even if households are not empirically likely to 4

be bound by lifetime contribution limits, a richer model incorporating uncertainty over future incomes could lead to households engaging in use-it-or-lose-it contribution behavior. In Canada, Milligan (2003) uses the expansion of contribution limits for Registered Retirement Savings Plans in the early 1990s to examine the impact of changing contribution limits on contribution behavior using a panel of taxpayer data. Consistent with the use-it-or-lose-it model, the paper presents evidence that households responded to an increase in their future contribution room with a decrease in their current contributions. The effect was strongest among households observed to be constrained in future years. In the presence of the use-it-or-lose-it motivation for contributing to tax-preferred savings accounts, the structural assumptions underlying much of the existing empirical methodology is invalidated. For example, Gale and Scholz (1994) compare the savings of IRA limit contributors (those who are contributing the maximum amount to their IRAs) to interior contributors (those contributing less than their maximum). By their assumption, a change in the annual contribution limit affects only the limit contributors and not the interior contributors. However, if the interior contributors are making contributions out of a use-it-or-lose-it motivation, then a change in their lifetime contribution limit will cause the contribution behavior of the interior contributors to change. Importantly, this does not help to distinguish the two sides of the 1990s debate about IRAs, as the Venti and Wise (1990) approach also makes a similar assumption. 5

Use-it-or-lose-it contribution behavior also carries implications for empirical assessments of savings behavior that attempt to exploit natural experiments in contribution limits. Two implementations of the natural experiment approach are found in Engelhardt (1996) who studies the cancellation of the Registered Home Ownership Savings Plan in Canada, and Attanasio and DeLeire (2002) who study the effects of the Tax Reform Act of 1986 on IRA contributions. With the recent expansions of tax-preferred contribution room in the Economic Growth and Tax Relief Reconciliation Act of 2001, the opportunity to exploit natural experiments will arise again soon as data become available. The implication of use-it-or-lose-it behavior for this type of empirical approach is that the short-run responses of households to changes in contribution limits will be complicated by their adjustment to their new endowment of lifetime contribution room. As found by Milligan (2003), an increase in the lifetime endowment of contribution room will decrease the current contributions of any contributor who may be constrained at some point in the future, not just those who are currently constrained by the contribution limit. Any empirical strategy should therefore consider the implications of use-it-or-lose-it behavior for the interpretation of empirical estimates. 3.0 Multiple Savings Accounts A second complication for the analysis of the savings impact of tax-preferred savings accounts arises in the presence of multiple, discretionary tax-preferred accounts. The most obvious current example is the recent expansion of federal tax benefits for 529 plans 6

for savings for advanced education. In Canada, the same issue arises for Registered Retirement Savings Plans and Registered Education Savings Plans. 4 In the presence of multiple accounts, the expansion of tax-preferred contribution room has different implications for household savings than when only one account exists. In order to analyze the savings impact of an additional tax-preferred savings opportunity on household savings behavior, a researcher must first address the effect of the new opportunity on the taxation of the household s marginal savings opportunity. In other words, the researcher must ask if the new savings account affects the marginal dollar of household savings. If the household is currently saving well in excess of the combined sum of its tax-preferred savings limits, then the additional contribution room will have only an infra-marginal wealth effect on the household rather than a marginal affect on its savings opportunities. In contrast, if additional tax-preferred savings opportunities are extended to households that are currently beneath their existing contribution limits, then the new tax-preferred account will only affect savings through a use-it-or-lose it affect as discussed above. The existence of multiple accounts has clear implications for empirical researchers. A researcher must consider the possibility that funds contributed to the new tax-preferred account would not have been contributed to another tax-preferred account in the absence of the new opportunity. Households may simply take a desired amount of savings and go shopping for the most attractive tax-preferred opportunity available to it. Care must be 7

taken that measured savings effects represent true gains to household saving rather than re-allocations across tax-preferred accounts. 4.0 Psychology and Savings The third and final challenge confronting empirical researchers of tax-preferred savings accounts arises from research in behavioral economics. Motivated by empirical and experimental research in psychology, economists have developed new tools for the analysis of savings. 5 Time-inconsistent preferences, the framing and salience of choices, and other alternative models of savings can both change the interpretation of existing evidence and generate new and interesting testable predictions to take to the data. Most existing research on tax-preferred savings accounts is based on the assumption that the economic incentives provided by the tax preferences underlie any observed change in economic behavior. Recently, it has been argued that the operative channel through which tax-preferred accounts affect behavior is through psychology. 6 Choi et al. (2001, 2002) provide evidence that the framing of choices about participation in a 401(k) plan can have a strong influence on the decisions that are made investors appear to follow the path of least resistance in making their choices. For example, participation in a firm s 401(k) plan increases dramatically when employees are automatically enrolled in the plan and must actively choose not to participate. If these behavioral channels represent the operative mechanism through which taxpreferred savings accounts affect behavior, then empirical methodologies must adjust. 8

The focus of most existing research assumes that the structure of taxation drives behavior rather than psychological aspects of savings. However, there are also potential pitfalls for testing behavioral hypotheses. For there to be a true savings effect and not simply a savings displacement effect, it must be shown that the funds placed in the 401(k) through a behavioral mechanism displaced consumption rather than savings in another form. For example, if employees save more through the 401(k) because of automatic enrollment, a researcher must be careful to ensure that these households do not simply reduce their IRA or 529 plan contributions or the rate they pay back their household debts. The framing and salience effects may simply distract the saver s attention from one form of saving to another. Another frequently discussed behavioral approach to saving involves mental accounts and target savers. Thaler (1990) describes the idea that households may save funds specifically with a target in mind rather than as a unified portfolio of wealth. The marginal propensity to consume out of accounts established for different purposes may differ, meaning that a windfall gain in one account may not affect consumption of goods for which that account is not intended. Some empirical evidence of this type of targeting behavior comes from a study of New York taxi drivers by Camerer et al. (1997). The researchers find that taxi drivers respond to higher hourly wages by working fewer hours, suggesting that the taxi drivers may have an income target in mind and quit after reaching it. 9

The presence of target savers has large implications for research on tax-preferred savings accounts. If households behave as target savers, then any increase in the return to their savings will necessarily decrease their level of savings to reach any given target, fewer dollars of savings are needed at a high rate of return than when the rate of return is lower. For example, in Canada the government pays a matching grant of 20 percent on the first $2,000 of annual contributions to a Registered Education Savings Plan. If households act as target savers, then the effect of this policy would be to decrease household savings, as the matching grant simply displaces other savings as the family attempts to reach its fixed target level of savings. In summary, the implications of recent developments in behavioral economics for research on tax-preferred savings accounts are potentially very important. A more thorough understanding of the active mechanisms that drive savings behavior would be very helpful for policy makers. For example, if behavioral rather than tax mechanisms explain the pattern of contributions more successfully, then policies that serve to activate behavioral channels should be favored. 5.0 Conclusion Given the large share of household wealth held in tax-preferred accounts, a better understanding of the contribution and overall savings behavior of households is warranted. This paper has outlined three challenges for researchers interested in continuing this important area of inquiry. I argue that life-cycle contribution limits, multiple tax-preferred accounts, and developments in behavioral economics should be taken into account as researchers approach empirical work into the future. More and 10

better empirical work will provide economists with a better understanding of savings behavior and policy makers with an improved environment for the design and structure of tax-preferred savings accounts. 11

References Angeletos, George-Marios, David Laibson, Andrea Repetto, Jeremy Tobacman and Stephen Weinberg (2001) The Hyberbolic Consumption Model: Calibration, Simulation, and Empirical Evaluation. Journal of Economic Perspectives 15(3), pp. 47-68. Attanasio, Orazio P. and Thomas DeLeire (2002), The Effect of Individual Retirement Accounts on Household Consumption and National Saving, The Economic Journal, Vol. 112 (July), pp. 504-538. Bernheim, B. Douglas (2002), Taxation and Saving, in Alan J. Auerbach and Martin Feldstein (eds.) Handbook Of Public Economics Vol. 3. Amersterdam: Elsevier Science. Bernheim, B. Douglas, Daniel M. Garrett and Dean M. Maki (2001) Education and Saving: The Long-Term Effects of High School Financial Curriculum Mandates. Journal of Public Economics 80(3), pp. 435-465. Browning, Martin, and Thomas F. Crossley (2001), The Life-Cycle Model of Consumption and Saving. Journal of Economic Perspectives 15(3), pp. 3-22. Camerer, Colin, Linda Babcock, George Loewenstein, and Richard Thaler (1997), Labor Supply of New York City Cabdrivers: One Day at a Time, Quarterly Journal of Economics, Vol. 112, No. 2, pp. 407-441. Choi, J. James, David Laibson, Brigitte C. Madrian, and Andrew Metrick (2001), For Better or For Worse: Default Effects and 401(k) Savings Behavior, NBER Working Paper No. 8651. Choi, J. James, David Laibson, Brigitte C. Madrian, and Andrew Metrick (2002), Defined Contribution Pensions: Plan Rules, Participant Decisions, and the Path of Least Resistance, Tax Policy and the Economy, Vol. 16, pp. 67-114. Engelhardt, Gary V. (1996), Tax Subsidies and Household Saving: Evidence from Canada, Quarterly Journal of Economics, Vol. 111, pp. 1237-1268. Engen, Eric M. and William G. Gale (2000), The Effects of 401(k) plans on Household Wealth: Differences Across Earnings Groups, NBER Working Paper No. 8032. Engen, Eric M., William G. Gale, and John Karl Scholz (1994), Do Savings Incentives Work? Brookings Papers on Economic Activity, pp. 85-180. Gale, William G. and John Karl Scholz (1994), IRAs and Household Saving, American Economic Review, Vol. 84, pp. 1233-1260 12

Gokhale, Jagadeesh, Laurence J. Kotlikoff, and M. Warshawsky (2001), Life-cycle Saving, Limits on Contributions to DC Pension Plans, and Lifetime Tax Benefits, NBER Working Paper 8170. Laibson, David (1998), Comment, in David A. Wise (ed.) Frontiers in the Economics of Aging. Chicago, IL: The University of Chicago Press. Hubbard, R. Glenn (1984), Do IRAs and Keoghs Increase Saving? National Tax Journal, Vol. 37, No. 1, pp. 43-54. Milligan, Kevin (2002), Tax Preference for Education Savings: Are RESPs Effective? Commentary, C.D. Howe Institute, Toronto. Milligan, Kevin (2003), How do Contribution Limits Affect Contributions to Tax-Preferred Savings Accounts? Journal of Public Economics, Vol. 87, No.2, pp. 253-281. Mullainathan, Sendhil and Richard H. Thaler (2000) Behavioral Economics. Working Paper 7948, National Bureau of Economic Research. Poterba, James M., Steven F. Venti, and David A. Wise (1998a), Personal Retirement Savings Programs and Asset Accumulation: Reconciling the Evidence, in David A. Wise (ed.) Frontiers in the Economics of Aging. Chicago, IL: The University of Chicago Press. Poterba, James M., Steven F. Venti, and David A. Wise (1998b), Implications of Rising Personal Retirement Saving, in David A. Wise (ed.) Frontiers in the Economics of Aging. Chicago, IL: The University of Chicago Press. Thaler, Richard H. (1990), Saving, Fungibility, and Mental Accounts. Journal of Economic Perspectives 4(1), pp. 193-205. Venti Steven F., and David A. Wise (1986), Tax-Deferred Accounts, Constrained Choice, and Estimation of Individual Saving, Review of Economic Studies, Vol. 53, pp. 579-601. Venti, Steven F. and David A. Wise (1990), Have IRAs Increased U.S. Saving?: Evidence from Consumer Expenditure Surveys, Quarterly Journal of Economics, Vol. 105, pp. 661-698. 13

Endnotes: 1 See Poterba, Venti, and Wise (1998b) for simulations on the impact of tax-preferred accounts on the retirement wealth of future cohorts. 2 I refer in this paper to tax-preferred savings accounts. While other authors prefer different terminology, I favor tax-preferred for two reasons. First, it is a more general term that can encompass more traditional tax-deferred accounts such as IRAs as well as tax-prepaid accounts such as the Roth IRA. Second, the organizing framework for taxation in both the United States and Canada is annual income. Departures from this base, of which there are many, can therefore reasonably be considered preferences with respect to the notional base. If the organizing framework were changed to have either consumption or lifetime income as the base for taxation, then exemptions of capital income would no longer be preferences. 3 Browning and Crossley (2001) explore the predictions, strengths, and limitations of the life-cycle approach to consumption and savings. 4 See Milligan (2002) for more detail and analysis of Registered Education Savings Plans. 5 See the short review article by Mullainathan and Thaler (2000) or a description of the theory and implications of hypberbolic discounting in Angeletos et al (2001). Bernheim (2002) also discusses the implication of behavioral theories for savings incentives. 6 Laibson (1998) describes a behavioral economics perspective on savings incentives. 14