What explains early withdrawals from retirement accounts? Evidence from a panel of taxpayers.

INTRODUCTION

Over the past 20 years, tax-deferred accounts (TDAs) such as Individual Retirement Accounts (IRAs) and employer-sponsored plans (ESPs) have become a key component of individual retirement planning, (1) Although such accounts can take on many guises, they generally share two features--tax-exempt account earnings and withdrawal restrictions before retirement. (2) Early withdrawals are generally subject to a 10 percent penalty, and, in the case of 401(k) plans, are prohibited outright except in cases of economic hardship or separation from the employer. A recent Administration proposal for establishing a new type of TDA with no early withdrawal penalties has brought into sharp relief the role of such penalties in shaping household behavior. Would households with free access to their TDA assets continually run them down, or would the increased liquidity encourage new savings by households with precautionary motives? One way to shed some light on this question is to examine what leads to pre-retirement withdrawals under the current system.

A number of recent studies have found that when ESP assets become available upon leaving a job, many individuals cash out the accounts and spend the assets, rather than rolling them over to another TDA. (3) This result is particularly prevalent among households with low levels of ESP assets. Engelhardt (2002) and Poterba, Venti, and Wise (1995, 1999) find that the aggregate size of these cash-outs are small i.e., a majority of assets in dollar terms are preserved in TDA form. Nevertheless, there is a certain sense of alarm that precisely those households that are worst prepared for retirement end up raiding their pension savings when given a chance.

We offer a somewhat different perspective on the failure of households to preserve TDA assets by investigating the economic circumstances under which such pre-retirement withdrawals are made. Due to a number of data limitations, previous empirical studies considered the TDA rollover decision in isolation, without taking into account the existence and severity of shocks to income or consumption needs. For example, many ESP distributions are associated with job separations, not all of which are voluntary. Thus, much of the previous work has focused on individuals who were disproportionately likely to suffer from adverse income shocks, and thus more likely to need the assets to smooth current consumption.

By contrast, our unique panel data set of 1987-1996 individual tax returns allows us to control for a variety of changes in household finances, including income shocks, job loss, divorces, changes in the number of earners or dependents, home purchases, and high medical expenses. Our goal is to gauge the importance of such shocks in determining early withdrawals from TDAs. In addition to ESP withdrawals, we also study early withdrawals from IRAs, which feature fewer access restrictions than employer-sponsored accounts.

Our findings indicate that penalized withdrawals are significantly more likely among households with low levels of non-retirement financial wealth, job loss, income shocks, divorce, and home purchases increase the likelihood of early withdrawals by 3 to 10 percentage points each. In addition, households with low non-retirement financial wealth are significantly more likely than other households to access their retirement accounts in response to such shocks. We conclude that a significant portion of early withdrawals from retirement accounts reflects consumption-smoothing behavior by liquidity-constrained households who experience financial shocks, rather than squandering of pension assets simply because they have become available.

The rest of the paper is organized as follows. The second section discusses the tension between current consumption and retirement saving in the context of lifetime utility maximization. The third section summarizes the restrictions on IRA and ESP withdrawals, while the fourth section describes the data set. The fifth section contains the discussion of empirical results, and the sixth section summarizes our findings and offers directions for future research.

CONSUMPTION SMOOTHING AND RETIREMENT SAVINGS

The central feature of a standard multi-period model of consumption under uncertainty is that households seek to smooth marginal utility of consumption over time and over different states of the world. As a result, households forgo current consumption in order to accumulate retirement savings. During their working years, households face uncertain labor earnings that can dramatically affect the amount of current income available to finance consumption. Thus, in order to smooth consumption across different states of the world, households build up stocks of precautionary assets. If borrowing is constrained or excessively costly, negative income or demographic shocks may compel households to dissave from accumulated precautionary or retirement assets. A similar result may occur if a household faces an increase in current consumption needs, such as an additional dependent.

Households that dissave in order to finance current consumption would be expected to dissave out of non-TDA assets before TDA assets. TDA assets are more costly because they grow at a pre-tax rate of return, while non-TDA assets grow at an after-tax rate. In addition, since many TDA contributions are made with pre-tax dollars, income taxes are due upon withdrawal of the assets. Finally, early access to TDA assets generally comes with an additional 10 percent penalty. Thus, pre-retirement TDA withdrawals can result in significant taxes and penalties, as well as the loss of future tax benefits on account earnings. Nonetheless, if few other assets are available for consumption smoothing, households may well use costly retirement savings for this purpose. While such choices may be unfortunate from the standpoint of retirement security, they would not necessarily be inconsistent with utility maximization over the lifetime.

In this context, the decision to cash out ESP distributions (or to withdraw IRA assets) becomes a function of the strength of a household's consumption-smoothing motives, its income, and the size and composition of its savings. This modeling framework links the likelihood of TDA dissaving to the occurrence of adverse income shocks and the level of non-retirement assets. Income and demographic shocks trigger the need for consumption smoothing, while the level of non-TDA assets determines the household's ability to avoid drawing on its retirement savings.

Previous studies of TDA cash-outs have been based (usually implicitly) on comparisons between current and expected future tax rates. (4) If the expected marginal tax rate at retirement exceeds the sum of current marginal tax rate and the 10 percent penalty, then it is optimal not to roll over the lump-sum distribution. This decision framework makes rollovers much more attractive to households that are at or near the top of the schedule of statutory marginal tax rates, because such households have little reason to believe that their future tax burdens will exceed their present marginal tax rates by more than the penalty wedge. Empirically, households with high MTRs are also likely to have high TDA balances. Thus, this decision framework fits well the robust empirical finding that the likelihood of rollover is increasing in the size of the TDA.

However, the tax-rate comparison model ignores other factors that affect the decision to cash out a TDA. In an intertemporal consumption model, the decision to withdraw TDA assets before retirement would be positively related to the strength of consumption-smoothing motives on part of households, which (for a given level of risk aversion) depends on the severity of current income or demographic shock. By extension, the withdrawal likelihood is a negative function of household current and expected income and liquid financial wealth. This set of testable restrictions guides our empirical analysis.

RESTRICTIONS ON ACCESS TO TDA ASSETS

All TDAs currently feature some restrictions on access to the funds before retirement age. (5) IRAs permit withdrawals at any age, but taxpayers must pay a 10 percent penalty tax on withdrawals taken before age 59 1/2. The penalty does not apply in certain cases, including death or disability, or withdrawals made in substantially equal periodic payments over the life of the participant or beneficiary. ESPs (including 401(k) plans) are also subject to the 10 percent penalty, but offer a longer list of exceptions, including (in addition to those listed above) withdrawals after reaching age 55 if the participant has separated from the employer, distributions to a spouse under a divorce-court order, deductible medical expenses, and certain ESOP distributions. (6)

Special restrictions apply to 401(k) plans, which prohibit pre-retirement access outright, except in the cases of death or disability, separation from the employer, termination of the plan, or employee hardship. Hardship withdrawals are limited to amounts necessary to satisfy an immediate and heavy financial need, including medical expenses, home purchase, post-secondary tuition, and expenditures required to prevent eviction or foreclosure. Hardship withdrawals are taxable and subject to the 10 percent penalty (unless an exception applies), and also trigger additional "implicit penalties": the suspension of new contributions (by either the employee or the employer) for at least 12 months, and a reduced contribution limit in the following year. (7)

Many ESPs (including 401(k) plans) offer loans against the account balance. Participants do not need to prove hardship to take a loan. A loan is generally not treated as a distribution (and thus not subject to taxes and penalties) as long as it is required to be repaid within five years (or is used for a home purchase) and is less than the smaller of $50,000 or 50 percent of the account balance. Loans are not permitted from IRAs.

DATA

Clearly, not all of the nuances available in the withdrawal rules described above are observable in the tax data. We use a large panel of tax returns covering the years 1987 to 1996. (8) The panel is based on a …