Thank you for inviting me to testify today on the issue of expanding Individual Retirement Accounts. My testimony consists of two parts: a summary of the main findings and a more detailed discussion of the basis for the conclusions drawn.
The low level of private and national saving is one of the most important economic problems facing our country today and in the future. American saving rates have been very low in recent years, compared to other countries and by historical standards. On a national level, more saving could finance increased investment. This in turn can make workers more productive, and raise their wages and standards of living. At the household level, increased saving helps people prepare for retirement, provides a cushion for financial downturns, and assists in meeting other financial goals.
Many potential factors have been offered to explain the saving decline. These include: increased intergenerational transfers to the elderly; expansions of government programs that reduce the need to save (including Social Security, Medicare, Medicaid, unemployment insurance, workers' compensation, housing guarantees, and student loans); liberalization of debt markets; demographic changes; and the slowdown in income growth since the mid-1970s. Tax considerations are notably absent from this list; indeed, the general tax and inflation environment facing savers may be at least as favorable today as it has been in the past. The highest marginal tax rates are relatively low by historical standards and inflation, which raises the effective tax rate on financial assets, is quite low. Despite these considerations, tax policy is sometimes claimed to be an effective way to raise the saving rate substantially.
Tax policy toward saving is inconsistent. Some assets are taxed at high effective rates, while a large number are taxed at rates that are very low and can even be negative. There is no shortage of tax-preferred methods of saving. Current options include IRAs, defined benefit pensions, defined contribution pensions, 401(k) plans, Keoghs, 403(b) plans, 457 plans, federal government thrift saving plans, SIMPLE plans, SEP plans, fixed and variable annuities, and life insurance saving. Moreover, housing and municipal bonds are also tax-favored, as are the capital gains that accrue to unincorporated businesses. Over the last several decades, as the personal saving rate has fallen, tax-favored saving (via pensions, 401(k)s, IRAs, Keoghs, and life insurance) has become an ever more important component of total personal saving. Between 1986 and 1993, saving in tax-preferred accounts constituted about 100 percent of net personal saving (Table 1). This does not mean there was no other saving activity, it just means that any gross saving in other accounts was fully offset by withdrawals from those accounts or by increases in borrowing.
Wide variations in effective tax rates on saving creates opportunities for investors to shift funds into the most tax-preferred accounts. The variation in rates, coupled with the tax-deductibility of interest payments, creates opportunities to game the system further by borrowing, deducting the interest payments, and investing in a tax-preferred asset.
IRAs are just one more patch in the crazy quilt of saving policy. Contributions of up to $2,000 per year are tax-deductible for households with income up to prescribed limits. Deductibility is then phased out as income rises further. Balances accrue tax-free. Ordinary income taxes are due on any withdrawals, and a 10 percent penalty is also assessed on withdrawals that are not related to death or disability, but occur before the account holder is 59.5 years old.
Allowing penalty-free (and income-tax-free) withdrawals for specified purposes such as education, medical expenses, first-time home purchases, long-term unemployment, or business start-up expenses.
These proposals involve issues of tax policy, budget policy, retirement income security, and saving policy.
Tax Policy Considerations
Expanding IRAs would be counterproductive tax policy. The IRA proposals would make the tax system more complex and intrusive. Serious consideration of how the IRS would verify that a particular withdrawal was made for a particular purpose suggests compliance and enforcement difficulties. Enforcing the combined limits on IRAs and elective deferral plans would cause further compliance headaches. Tax debates in 1996 correctly emphasized the importance of broadening the base, removing loopholes, and reducing rates in a revenue-neutral manner. As we move into 1997, proposals that expand IRAs move in exactly the opposite direction.
While IRAs are often described as tax-deferred saving, the effective tax rate on IRAs is typically zero or negative. The effective rate is zero if the tax rate that applies to the deductible contribution is equal to the rate that applies to the withdrawal. However, since marginal tax rates have fallen since 1986, and since people typically face lower marginal tax rates in retirement than during working years, the effective tax rate for many IRA holders is likely to be negative. For example, a household that deducts a $2,000 IRA contribution at a 28 percent tax rate, holds the asset for 20 years at a 10 percent annual return, and withdraws the funds at a 15 percent tax rate pays an effective tax rate of negative 9 percent on the IRA. Punching a hole in the tax code to generate more assets with negative effective tax rates is inefficient and inequitable. Good tax policy would even out the taxation of all forms of saving, and possibly reduce the overall level of taxation on saving.
Budget Policy Considerations
Expanding IRAs would also be counterproductive budget policy. First, it would create a new entitlement for anyone with enough funds to place money in a designated account. The fact that IRAs are tax rules rather than spending programs should not blind us to the essential equivalence of an entitlement set in the tax code and one set on the spending side. Tax entitlements are just as costly (and often more difficult to discern) than spending entitlements. The IRA entitlement would accrue largely to households in the top part of the income distribution, and would provide larger entitlement payments (i.e., tax cuts) to wealthier households who contributed more or faced higher tax rates. The key to long run budget control is to eliminate or reduce entitlement obligations rather than increase them.
Second, current budget procedures understate the cost of back-loaded IRAs. The requirement of a 5-year holding period before penalty-free withdrawals are allowed effectively places most of the costs beyond the five-year budget window. Budget policy should move toward more complete accounting of the costs of government programs.
Third, for any given amount of contributions, allowing both traditional front-loaded IRAs and back-loaded IRAs will prove more expensive in revenue terms than having either one. Other things equal, people who believe their tax rate will be lower when they withdraw the funds than it is now will tend to choose front-loaded IRAs, so they can take the deduction at the relatively higher current tax rate. Likewise, people who believe that their tax rate upon withdrawal will be lower than their current rate will tend to choose back-loaded IRAs to obtain the biggest tax cut.
Retirement Income Considerations
Expanding the conditions for penalty-free IRA withdrawals would undermine the retirement income goals of IRAs, and could reduce both saving and tax revenue. One can imagine the list of favored uses of IRA funds expanding indefinitely. One can also imagine the list of favored accounts expanding as well: if IRA funds can be tapped, why not Keoghs, SIMPLE plans, SEPs, 401(k)s, pensions, or fixed and variable annuities? Moreover, there would be difficult administrative problems associated with minimizing abuse of these provisions. These problems will make the tax code more complex, and will require the IRS to gather more information, which could be quite intrusive, or risk not enforcing the provisions.
If withdrawals are allowed for new, favored uses of funds, two considerations are paramount. First, the withdrawals should be allowed only for funds contributed after legislation is enacted. As of the end of 1995, IRA and Keogh balances totalled $1.2 trillion. These funds were placed in the accounts with the understanding that they were to be held until retirement or would face a penalty. If these funds become eligible for penalty-free withdrawal, the saving rate could actually drop. For example, suppose that in one year, 5 percent of these funds were removed for other purposes. That would represent about a withdrawals of about $60 billion, or about 20 percent of personal saving. Second, funds withdrawn from deductible IRAs should face income taxes, even if the penalty is waived. Otherwise, the entire withdrawal will never have been taxed, which would create obvious inequities and inefficiencies.
Saving Policy Considerations
All of these problems in tax policy, budget policy, and retirement income policy might be worth the cost if IRA expansions were certain to raise private and national saving substantially. The effect of IRAs on saving is the subject of considerable controversy, however, so it is useful to start with some basics.
The single most important factor is that IRAs do not provide incentives to save. Instead, IRAs provide incentives to place funds in a designated account. The distinction is crucial.
There are many ways to finance IRA contributions. One way, of course, is to raise saving. This involves consuming less, or to put it bluntly, reducing one's current standard of living. This is the "painful" way of taking advantage of the tax breaks afforded by IRAs. There are, however, relatively painless ways to capture the tax break as well. For example, the contribution may be financed by transferring existing taxable assets into IRAs, by reallocating into an IRA current or future saving that would have been done outside the IRA, or by increasing household debt. These painless methods of contributing to an IRA do not raise overall private saving. Thus, IRAs and other so-called "saving incentives" do not require that contributors save, or save more than they would have otherwise.
How are people likely to react to IRAs? Common sense suggests that people will try to capture the tax breaks in the least painful way possible. A reasonable conjecture is that one reason IRAs are so popular with taxpayers is precisely because taxpayers do not need to reduce their standard of living (raise their saving) to claim the tax break.
Research findings back up this claim at the most general level. Economists Joel Slemrod of the University of Michigan, and Alan Auerbach of the University of California, surveying a broad range of studies of the effects of the tax reform act of 1986, have concluded that similar phenomena arise in a host of tax-related activities. They find that decisions concerning the timing of economic transactions are the most clearly responsive to tax considerations. The next tier of responses involves financial and accounting choices, such as allocating a given amount of saving to tax-preferred saving versus other saving. The least responsive category of behavior applies to agents' real decisions, such as changes in the level of saving. This hierarchy of responses, applied to IRAs, suggests that most IRA contributions are not new saving.
(A) What proportion of IRA contributions is new saving?
In recent years, a number of studies have examined the effects of IRAs on saving and reached a variety of conclusions.
The crucial issue in this literature is determining what households who had IRAs would have saved in the absence of these incentives.
Several factors, however, make this a difficult problem and one subject to a series of biases that overstate the impact of IRAs on saving. Analyses that ignore these issues overstate the impact of IRAs on saving. No study that corrects for these biases finds that IRAs raise saving. Rather, Engen, Gale and Scholz (1996a, b) show that accounting for these factors largely or completely eliminates the estimated positive impact of IRAs on saving found in some studies.
First, saving behavior varies significantly across households. Households that hold IRAs have systematically stronger tastes for saving than other households. Thus, a simple comparison of the saving behavior of households with and without IRAs will be biased in favor of "showing" that IRAs raise saving. To oversimplify somewhat, suppose there exist two groups: "large" savers and "small" savers. We would expect to see that IRA holders (where large savers are overrepresented) would save more than non-IRA holders (where small savers were overrepresented). But this would provide no information about the effects of IRAs per se, unless there is a way to control for the observable and unobservable differences between large and small savers.
Even researchers that claim that IRAs raise saving recognize that the heterogeneity of saving behavior is a crucial factor in this literature. What is often overlooked, however, is that the implication of heterogeneity is that findings such as "households with IRAs saved more than households without IRAs," do not imply anything about whether IRA contributions represent new saving, since those households would have been expected to save more to begin with.
Due to heterogeneity in saving, studies that compare IRA contributors with noncontributors tend to "find" that IRAs raise saving (Hubbard 1984, Feenberg and Skinner 1989, Venti and Wise, 1987, 1988, 1990, 1991). However, statistical tests reject the validity of such comparisons (Gale and Scholz 1994.) In contrast, studies that compare one group of contributors to another tend to find much smaller or negligible effects of IRAs, or expansions of IRAs, on saving (Gale and Scholz 1994, Attanasio and De Liere 1994, Joines and Manegold 1995). By comparing two groups of contributors, these studies more effectively isolate groups with similar propensities to save and hence provide a more valid comparison.
A second problem is that saving and wealth are net concepts and are broad concepts. If a household borrows $1000 and puts the money in a saving incentive account, net private saving is zero. The data indicate that households with saving incentives have taken on more debt than other households. Hence, studies should focus on how saving incentives affect wealth (assets minus debt), not just assets. Because financial assets are small relative to total assets, studies that focus only on the effects of saving incentives on financial assets may have particularly limited significance.
Since the expansion of IRAs in the early 1980s, financial markets, pensions, and Social Security have undergone major changes. Pension coverage (other than 401(k)s) fell over the 1980s, and social security wealth was reduced in the 1983 reforms. Both of these factors would have caused people to have accumulated more assets in the late 1980s or early 1990s than in the early 1980s. Moreover, the reduction in inflation and tax rates that occurred over the 1980s made financial assets relatively more attractive than tangible assets (such as housing). This led to strong increases in the stock market and to shifts of wealth from nonfinancial to financial forms. For all of these reasons, it is important to study the impact of IRAs on broad wealth measures and to control for other events that occurred during the 1980s.
Studies that examine only financial assets often "find" a large impact of IRAs on saving (Venti and Wise 1992, 1996). But extensions of those studies indicate that the effects disappear when the analysis examines the impact on broader measures of wealth that include debt or nonfinancial assets and include the impact of events that occurred during the 1980s (Engen, Gale and Scholz 1996a, b).
Third, IRA balances represent pre-tax balances; one cannot consume the entire amount because taxes and perhaps penalties are due upon withdrawal. In contrast, contributions to other accounts are generally not deductible and one may generally consume the entire balance in a taxable account. Therefore, a given balance in a saving incentive account represents less saving (defined either as reduced previous consumption or increased future consumption) than an equivalent amount in a conventional account.
Analyses that correct for these biases indicate that little if any of the overall contributions to IRAs have raised private or national saving. This conclusion arises consistently from evidence and estimates from a wide range of methodologies, including time-series data, cross-sections, panel data, cohort analysis, simulation models, and analysis of evidence from Canada (Engen, Gale, and Scholz 1996a, b).
(B) Who Contributed to IRAs and Why it Matters
Supporting evidence for this view comes from data on who contributed to IRAs. Table 2 shows that households with IRAs in 1986 were very different from households that do not have IRAs. In particular, compared to households without IRAs, the typical IRA holder had seven times the non-IRA financial assets, four times the overall net worth, and eight times the saving. Although some of these differences are due to observable characteristics, there is widespread agreement that households with IRAs tend to have stronger unobservable tastes for saving than do observationally equivalent households without IRAs.
Two types of households will be most able and hence most likely to make painless contributions, that is, contributions that do not raise private saving. The first is households that have large amount of other assets. These households have more existing assets to shift, typically have more current saving to shift, and have less of a need to maintain all of their assets as precautions against emergencies. The second is older households, who are less likely to face a binding early withdrawal penalty. In the extreme, people older than 59.5 years face no early withdrawal penalties. For each group, IRAs are good substitutes for the saving those households would do anyway, so the IRA contribution will be unlikely to represent new saving.
Data from the 1980s show that households with non-IRA financial assets over $20,000 in 1986 (about $28,600 in 1996 dollars) or who were 59 or older made more than two-thirds of all IRA contributions in the 1983-6 period.
Households who had non-IRA financial assets in excess of $40,000 (about $57,200 in 1996 dollars) or where the head was 59 or older made half of all IRA contributions during this period. Thus, while some people have argued that many of the accounts were held by middle class households, the data show that most contributions were made by households that would consider IRAs and other saving good substitutes. This suggests that the overall effects of IRAs on saving were likely to have been small at best.
In contrast, contributions will represent a net addition to saving only when they are financed by reductions in consumption, which will occur only when IRAs and other saving are poor substitutes for one another. This is more likely to occur for households that have lower asset holdings, and are younger. Thus, if IRAs are to be expanded, the expansion should be targeted to lower-income groups. Higher income groups will typically have higher assets and will find it easier to substitute other assets into IRAs.
(C) Aggregate Effects of Expanded IRAs on Saving
How much would expanding IRAs raise national and private saving? One can get some perspective on this issue by noting that net national saving has fallen from 8 percent of net national product in the 1950s, 1960s, and 1970s, to 4.1 percent in the 1990s. Personal saving has fallen from 7 percent of personal disposable income between 1950 and 1980, to under 5 percent in the 1990s.
One way to gauge the effect of all tax policy on saving is to consider the effects of replacing the income tax with a consumption tax. Estimates by Engen and Gale (1996) suggest that a cold-turkey switch to a pure consumption tax--with no personal exemptions or transition relief--would raise the saving rate by about 1.5 percentage points in the short run and by about 0.5 percentage points in the long run. Output per capita would rise by about 1.5 percentage points over the first 10 years. These effects are positive, but are modest compared to the decline in saving noted above.
The results also provide a useful perspective on what targeted tax policy changes can achieve. If a complete overhaul of the income tax system raises the saving rate by at most 1.5 percentage points, only a much smaller impact can be expected of policies that tinker around the edges of the system.
The aggregate impact of expanding IRAs would be tiny. From 1982 to 1986, IRA contributions constituted about 1 percent of GDP. Since then, however, tax rates have fallen and other saving incentives have proliferated. Moreover, expansion would only affect a small portion of the population. If contributions rose by 0.5 percentage points of GDP and--splitting the difference among the studies--about half of those contributions were new saving, private saving would rise by 0.25 percentage points. But, assuming an effective federal and state tax rate of about 25 percent, government saving would fall by about one fourth of the contributions, so the net increase in national saving would be about 0.12 percentage points over the next few years.
Note that this estimate does not include the impact of allowing penalty-free (and income-tax-free) withdrawals for specified purposes. If these withdrawals are allowed from pre-existing balances, or if the withdrawals are made free of income tax, the impact on private and national saving of expanding IRAs could well be negative.
(D) Short-run versus Long-run Effects of IRAs on Saving
Some commentators (including Engen and Gale 1993) have made the point that the short-term effects of IRAs are likely to be less favorable than the long-term effects. The idea is that when IRAs are introduced, people will shift funds from taxable sources into IRAs so the contributions at first will not be new saving. After awhile, the people who contribute to IRAs may run out of funds to shift so that IRA contributions may eventually become new saving. For example, in a simulation model in Engen, Gale, and Scholz (1994), IRAs reduce short-term saving, but raise the long-term saving rate by 0.2-0.3 percentage points.
The crucial issue then becomes "how long does it take until the saving rate rises?" In Engen, Gale, and Scholz (1994) it takes 49 years for the wealth to income ratio to exceed its original (pre-IRA value). Some IRA proponents have reasoned that since the typical household has very little in pre-existing financial assets, the transition period will be very short: a year or less.
The logic of a short transition period is misleading for two reasons. The first is simply that the typical household in 1986 did not have an IRA, so the typical household is irrelevant to the debate about how long the transition will last. The relevant households are those that contributed to IRAs and in particular those that continued to contribute to IRAs: Did these households have many pre-existing assets that they could shift into IRAs? The answer here is a resounding "yes." Table 2 shows that pre-existing asset balances are high among household with IRAs. The typical IRA household in 1986 had over $20,000 in non-IRA financial assets. Among households that contributed to the limit for three years in a row, typical financial asset balances were $40,000. It is clear that for these households, IRAs could be financed from pre-existing asset balances for several years without raising saving.
The second problem with the proponents' logic is even more important: it ignores IRA contributions that are financed by current or future saving that would have been done even in the absence of IRAs. These contributions do not represent new saving. The table shows that typical IRA households and 3-year limit contributors have extremely high levels of other saving relative to their IRA contributions and so could easily finance contributions out of saving that would have been done anyway. The median 3-year saving level for 3-year limit contributors in the SCF was $60,000. Surely, it would not be difficult for many of them simply to shift $12,000 of that into an IRA. The median 3-year saving level for the typical IRA contributor was $23,000. This is certainly large enough to fund all or most of a typical three years worth of contributions. These figures suggest that among households that did contribute to IRAs, there was a large on-going source of funds from which IRA contributions could be financed without raising saving. There is every reason to think the transition period could take a very long time.
A second reason IRAs may raise long-term saving is that workers who leave jobs often roll their pension balances over into an IRA. Thus, the IRA provides a convenient way to keep the money "tied up" rather than encouraging people to spend the funds prematurely. Over long periods of time, the cumulative effect of having fewer people cash out their pension could raise the saving rate. Two caveats, however, should be noted. First, any such effect does not seem to have occurred yet. Second, this factor is already fully operable under the existing IRA system. No expansion of IRAs is needed.
(E) Did Advertising Make IRA Contributions New Saving?
Some commentators have asserted that the heavy advertising of IRAs means that IRA contributions were new saving. However, while it seems likely that IRAs were advertised heavily by the financial industry in the 1982-6 period, that fact provides no information as to whether the source of IRA contributions was new saving (reduction in living standards) or shifted assets, redirected saving, or increases in debt. There is certainly no evidence to support the notion that advertising for IRAs affected the level of saving.
Looking at the ads themselves, however, suggests that advertising may actually encourage asset shifting, rather than new saving. Some ads explicitly advocated financing IRAs with debt as an "easy" way to obtain the tax break (see Feenberg and Skinner 1989). Aaron and Galper (1984, p. 5) report the following ad from the New York Times in 1984:
Were you to shift $2,000 from your right pants pocket into your left pants pocket, you wouldn't make a nickel on the transaction. However, if those different "pockets" were accounts at The Bowery, you'd profit by hundreds of dollars ....Setting up an Individual Retirement Account is a means of giving money to yourself. The magic of an IRA is that your contributions are tax-deductible."
For obvious reasons, advertising seems more likely to emphasize the possibility of painless contributions, which don't raise saving, rather than painful contributions that do raise saving.
A second perspective on advertising is provided by the recent avalanche of ads for mutual funds and the accompanying massive inflows into those funds. Figure 1 shows that as mutual funds have increased dramatically in recent years, personal saving has not. Figure 2 shows that the increase in mutual fund saving has been matched by a decline in individual holdings of equities and bonds. That is, to a large extent households appear to have shifted their assets from one form to another. This is in no way a criticism of the mutual fund industry, which is supplying a product that the public demands. The point is just that the presence of massive advertising does not imply that the subsequent contributions are new saving.
A similarly unproven assertion is that IRAs created a "culture of saving," or would have if they had not been curtailed in 1986. To some extent, this notion is based on evidence about the persistence of IRA contributions over time. Households that contributed in one year had a very high probability of contributing in the next year as well. This led to speculation that IRAs helped people create good saving habits over time (Skinner 1992, Thaler 1994). The problem with this conclusion is that the data on persistence are perfectly consistent with standard models (Engen and Gale 1993). There is nothing surprising about the persistence of contributions over time. A purely rational model with no "habit formation" generates the same persistence as the data.
Moreover, other evidence makes it hard to believe that IRAs created a culture of saving. The early 1980s featured lower inflation, lower tax rates, high real interest rates, cuts in social security as well as expanded IRAs, yet the saving rate fell rather than rose during the "golden years" of IRAs.
Expanding targeted tax-based saving incentives is unlikely to raise the saving rate by very much if at all, but could have real costs in terms of tax, budget and retirement income policy. Excessive focus on tinkering with tax-based saving incentives obscures other possibilities for raising private and national saving. The surest way to raise national saving is to reduce the budget deficit in ways that do not reduce private saving.
Raising private saving may prove more difficult, but several options are worth exploring. The most obvious candidate is improved financial education of workers. There is serious concern that a substantial fraction of the population will not be adequately prepared for retirement. At the same time, however, a large proportion of households do not use the saving incentives that are already available to them. Everyone, for example, can contribute to an IRA or a fixed or variable annuity if they so choose and receive a tax-preference relative to other saving. Only about two-thirds of workers eligible for 401(k) plans actually participate. Improved education would also be worthwhile to provide needed assistance to American households as the pension system moves away from defined benefit plans and toward defined contribution plans, which place more responsibility on workers, and as social security reform is considered.
Another fruitful area of reform in my view is pension legislation. An improved pension system would feature enhanced pension coverage, simplified nondiscrimination rules with a higher minimum contribution, higher maximum contribution limits, and removal of taxes on excess payouts and excess accumulations.