FULL TEXT: GRAVELLE LOOKS AT LOWER CAPITAL GAINS TAX, LIBERAL IRA RULES.

Jane G. Gravelle Senior Specialist in Economic Policy

September 26, 1989

SUMMARY

Two tax reduction proposals are currently being considered: a reduction in capital gains tax by indexing of the basis for inflation, and a partial restoration of deductible IRA's. One of the arguments advanced for adopting these tax incentives is to increase the savings rate in the economy. This study assesses what both economic theory and empirical evidence suggest about the effect of these proposals on savings.

Economic theory cannot determine whether a tax incentive will increase or decrease savings, because there are offsetting income effects (individuals whose taxes are reduced will consume part of the reduction) and substitution effects (individuals will prefer to save more because of the higher rate of return). If the tax incentives are financed by increasing the deficit (and thereby decreasing government savings), there is a strong possibility that such tax changes will actually reduce savings in the economy. Moreover, the price effect which does increase savings will not necessarily be present for individuals whose otherwise planned savings exceed the ceilings on individual retirement accounts. If the incentives are financed by an across the board tax increase, an increase in savings becomes more likely because of the reduced consumption by those individuals arising from the general tax increase. In this case, the income effects would suggest that IRA's would increase savings more than a capital gains tax cut, for an incentive with an equal reduction in effective tax rate.

Econometric evidence suggests that the savings response to a tax incentive is relatively small. Some calculations of the effects of the capital gains tax cut in the context of a life cycle model suggest that such a change could actually reduce the long run savings rate, although by only a negligible amount. Even in those cases where the treatment most conducive to a large savings effect is assumed, the effects on the savings rate are quite modest. Similarly modest effects would be expected for Individual Retirement Accounts. These results are not surprising given the small magnitude of estimated savings responses and the small magnitude of the tax changes themselves.

Some studies which have focused on IRA's have suggested that, despite the conventional wisdom as to these small and uncertain savings responses, much of the contribution to IRA's was new savings. A review of that literature suggests that such conclusions are not justified by these studies. In general, the data sets used in these studies cannot provide good evidence about the effects of IRA's.

CONTENTS

BACKGROUND ON THE ISSUES

HOW TAX INCENTIVES CAN AFFECT SAVINGS

ECONOMETRIC EVIDENCE ON SAVINGS BEHAVIOR

CONCLUSION

CAPITAL GAINS TAXES, IRAS, AND SAVINGS.

The Tax Reform Act of 1986 cut tax rates across the board, but these tax rate reductions were particularly deep for higher income individuals. The top rate was reduced by almost one half, from 50 percent to 28 percent. To offset these relatively larger rate reductions, there were a number of provisions which reduced tax benefits available to higher income individuals. Among these were the passive loss restriction and the full taxation of capital gains (formerly capital gains were eligible for a 60 percent exclusion), income sources which were very heavily concentrated at higher incomes. The availability of up front deductions for individual retirement accounts (IRA's) for higher income individuals was also restricted with the deduction phased out between $40,000 and $50,000 of adjusted gross income ($25,000 to $35,000 for single returns).

Interest has recently developed in restoring some of these benefits, and one of the arguments advanced for doing so is to increase savings. The Ways and Means Committee has proposed to exclude 30 percent of capital gains for assets sold by the end of 1991 and to index assets for inflation thereafter. This measure, which has been narrowly approved, is quite controversial. Senator Bentsen, Chairman of the Finance Committee, has proposed as an alternative savings incentive to restore part of the treatment of individual IRA accounts as a substitute for the capital gains change. The Democratic leadership in the House has also proposed an amendment to substitute an IRA deduction for the capital gains provision when the measure is considered in the House, and to finance the IRA treatment by an increase in tax rates for the top income levels. The IRA proposal under discussion would restore part of the deduction which was eliminated in 1986 and allow penalty-free withdrawal from IRA accounts when used for education or first time home purchase. Although there are a number of issues surrounding the desirability of these policies, both proposals have also been argued as ways of increasing savings, and it is this issue which is the focus of this paper.

The first section of the paper explains the background of the two provisions. The next section explains the basic economic theory surrounding the effects of these proposals on savings, and explains why economic theory cannot unambiguously determine the effect of savings incentives on the level of savings. The following section discusses the econometric literature on these effects, including general studies of savings responses and some specific studies of the effect of IRA's. This section suggests that the effects of either of these policies on aggregate savings s likely to be quite small.

BACKGROUND ON THE ISSUES

Historically, capital gains have been eligible for special treatment and prior to the Tax Reform Act, sixty percent of gain could be excluded from income. In addition, gains held until death, which constitute a large fraction of gains, escape taxation entirely because the heirs take the market value at time of death as the basis for determining any future capital gains. Under the Tax Reform Act of 1986, capital gains were taxed in full. Because, however, the tax rates were cut deeply, the capital gains tax rate for the high income individuals who receive most of the gain is not much higher than it has been through most of history. Indeed, it is lower than it has been in many periods. Because gains are not indexed some gain was taxed at effective rates higher than the statutory rate; gains on relatively short lived assets with low real appreciation rates can be taxed quite heavily for that reason. Capital gains on assets held for a relatively long period of time or assets which have high appreciation rates, may on the other hand, be subject to tax rates below the statutory rate, and, of course, gains held until death are not subject to tax.

Individual Retirement Accounts (IRA's) were first introduced into the tax law in 1974 when individuals not covered by an employer's pension plan were allowed to deduct contributions to an individual account. In 1981, this provision was made universally available, allowing individuals to deduct contributions of up to $2,000. Earnings on the account would not be subject to tax; rather taxes would be deferred until contributions were withdrawn for use in retirement. A penalty of 10 percent applied if contributions were withdrawn prior to age 59 and 1/2. If the individual had the same tax rate both at the time of contribution and the time of withdrawal, the effect is a zero tax rate.

The Tax Reform Act of 1986 restricted the full benefits of the IRA's for individuals not covered by private pension plans by requiring eligibility to phase out as adjusted gross income exceeded $40,000 for a married couple ($25,000 for a single individual). The phase out was completed at $50,000 for a married couple ($35,000 for an individual) so that married taxpayers with adjusted gross incomes of more than $50,000 would not be able to deduct contributions to IRA's. These individuals could, however, still open IRA accounts and defer taxes on the earnings in these accounts until funds are withdrawn. These IRA's still receive substantial benefits because of tax deferral; tax rates are quite low if assets are held over a long period of time. Essentially, the deferral still provides significant tax benefits, depending on the holding period and the rate of return. In the case of a 28 percent tax rate, a fully taxable account is subject to the full nominal rate; a deductible IRA is subject to a zero tax rate. A tax deferred IRA with contributions not initially deductible, would fall somewhere in between. For example, with a ten percent interest rate, the effective tax rate on an IRA is 19 percent if held for ten years and 14 percent if held for twenty years.

Although the IRA proposal has not been formally reported on, suggestions have been made to restore one half of the up front deductibility and to allow penalty free withdrawals for education and first time home purchases.

HOW THE INCENTIVES CAN AFFECT SAVINGS

It is well known in economic theory that a reduction in taxes on income from capital can either decrease or increase savings. Economists commonly refer to two routes through which these policies can influence savings as income effects (which normally reduce savings) and substitution effects (which increase savings). An income effect simply means that when individuals receive income because taxes are lower they will spend a portion of that income -- consumption will rise and savings will fall. The substitution effect means that because the return net of taxes rises, individuals will want to save more. It is the offsetting of these effects which leads to ambiguity in the effect of changes in the rate of return on savings. If, however, the government maintains a fixed budget balance AND does not redistribute income, there is no income effect and the substitution effect (which increases savings) will be the only effect. The question then becomes how large this substitution effect is.

We use some simple examples below to explain these effects. Several points are made in this section. First, the income effect alone, assuming deficit finance for the revenue loss, would unambiguously reduce savings. The income effect of a capital gains tax cut would unambiguously reduce private savings, while the income effect of an IRA would unambiguously reduce total savings (public and private), but not necessarily private savings. For an incentive of equal magnitude, the income effect on total savings is identical for identical individuals. Thirdly, in both cases substitution effects will increase savings, but there may be cases under an IRA where this effect does not operate because of the dollar ceilings. Fourthly, if the tax incentive were financed by an across the board tax increase, income effects would likely increase savings for the IRA approach but not for the capital gains cut. The basic conclusion of this study is that it is very difficult to determine which provision might have the larger effect on savings, but the effects would be quite small in either case.

INCOME EFFECTS WITH DEFICIT FINANCE

HOW INCOME EFFECTS REDUCE SAVINGS: CAPITAL GAINS

We use some simple illustrations to demonstrate this point, and initially consider a change financed by an increase in the deficit. Consider savings of $100 held in an account for ten years at which time it is consumed, a fifty percent tax rate, and a ten percent interest rate. $100 thus invested will grow after tax (that is, at an after tax rate of return of five percent) to $162.89. Consider first eliminating the tax entirely. While this effect is not exactly like a capital gains exclusion, it is analogous in that the tax benefit is to be provided in the future, not in the present. Suppose that the individual is a target saver; that is he simply wants to plan to have the amount of $162.89 in ten years. With a higher interest rate he needs to save only $62.80, since that amount compounded at ten percent over ten years will grow to $162.89. Thus, his consumption will go up -- and savings fall -- by $37.20.

The individual might prefer, however, to consume more in the future. He might wish to increase both his present consumption and his future consumption, and as one can see the income effect allows him to do both. Indeed, we would normally expect that an increase in income would be shared by both present and future consumption. For example, if the individual normally saved ten percent of his income he might wish to save ten percent of the $37.50 and enjoy a proportional increase in both present and future consumption. This is the outcome which would occur if there were no substitutability between present and future consumption and individuals prefer to consume the same proportions; savings would still fall in this example because there is only an income effect and income effects alone will always lead to a reduction in savings. Because future consumption, however, has become cheaper -- forgoing a dollar of consumption today allows $2.59 in future consumption (at a ten percent rate) rather than $1.63 (at a five percent rate) -- he might wish to consume relatively more in the future. For example, suppose he wishes to keep current consumption fixed (a targeted consumption in the present). In this case savings will not change at all. This outcome implies a relatively pronounced tendency to respond to price effects. If the substitution effect is even more powerful, savings will increase.

HOW INCOME EFFECTS REDUCE SAVINGS: THE IRA CASE

Consider now the alternative of the IRA approach in which savings are deducted today and withdrawals taxed in the future. Again, if the individual is a target saver and wishes to consume the same $162.89 in the future, he will now save $125.60 -- at a fifty percent tax rate, exactly twice his savings under the exclusion example above. His accumulated savings plus interest in the future are twice as much as in the exclusion example above, but he must pay half the proceeds in taxes. Note in this case that personal savings has increased -- by $25.60. If, however, we take the government's behavior into account, we find that total savings has fallen. The individual will receive a deduction for this contribution with tax savings equal to $62.80. His current consumption is no different from the previous case. The government will have a revenue loss of $62.80 and the net effect on savings -- the $62.80 reduction in the government's case and the $25.60 increase by the individual -- will involve a net reduction in overall savings of $37.50.

As in the case of the tax cut in the future, the individual might save more because he desires to increase consumption in the future -- but each additional dollar of saving will provide him with an increased current tax saving -- offset by a government loss -- and future tax liability. For example, if the individual wants to keep present consumption fixed, his savings will double to $200, but the additional $100 of savings will be financed not by consumption but by the tax savings -- government revenue loss will be $100. Private savings will double but total savings -- public and private -- will be fixed. If he is quite willing to substitute present for future consumption the IRA could increase not only private saving but total saving. These two alternatives are only different in the source of the changed savings -- public versus private -- and not in their overall effect on savings.

In both of these cases, of course, the government has lost revenue in the future which also must be financed, presumably through borrowing. Thus in these simple examples, it is very easy to imagine that a savings incentive might depress the overall savings level in the economy indefinitely. Hence, the ambiguity regarding the effect of tax incentives on savings in the economy.

Note that the new IRA treatment in addition to imposing a partial tax on the return to IRA's also shifts the timing of the tax benefits from the present to the future. Thus, new IRA's would increase individual savings through both a price and an income effect, but reduce government savings. If the revenues are made up by increased taxes, however, different results could occur.

PRICE EFFECTS

As noted above, it is the price effects that would make it possible to for a tax incentive to increase savings; thus, in the deficit finance case only the operation of a price effect would increase the savings rate. Questions can arise, however, as to whether the price effect always operates in the case of a tax benefit with a dollar limit, as in the case of an IRA. It is these points which have given rise to some criticism of IRA's, particularly for higher income individuals and that are addressed in some of the IRA studies. For example if individuals can borrow or shift assets to finance IRA's, or use already planned savings to contribute to IRA's, they may easily contribute up to the limit without changing their overall level of savings. In that case, there is no price effect operating. Once the limit is reached, income from an additional dollar of savings is fully taxed and the price of consumption does not change. In that case, one would expect IRA's financed by government borrowing to unambiguously reduce the economy's savings rate.

In practice, however, there are many reasons to believe that IRAs can still have price effects even for higher income individuals. First, for most individuals the rate paid on borrowed funds is likely to be above the rate on savings, suggesting that the opportunity to borrow would not eliminate price effects, particularly since IRA accounts cannot be used as security. In addition, there are now restrictions on the deductibility of interest, although the effectiveness of those restrictions is subject to some question. Secondly, over a period of time individuals may essentially run out of existing assets which can be shifted into IRA's. Moreover, for many individuals the IRA levels may exceed normally planned savings. Finally, IRA's may not be perfect substitutes for other forms of savings since they are subject to a penalty for withdrawal before retirement. This penalty makes IRA's unattractive for savings which assets which are directed towards shorter lived assets. In that case, for the IRA to have a marginal effect the dollar limit must exceed savings which are otherwise sufficiently long term.

INCOME EFFECTS WITH TAX FINANCE

In addition to the complex issues surrounding the effects of IRA limits, the examples discussed above are in fact dramatic oversimplifications of how one might set up such a model of savings behavior. One such possibility is that the government might choose to finance the revenue losses with other taxes. If all individuals were identical, then such a financing mechanism would simply eliminate all of the income effects, assuming that individuals could correctly anticipate the government's future tax collections. In that case, savings would unambiguously rise, with the amount of that increase depending on the substitution effect.

Individuals, however, are not identical. The examples presented above consider two individuals who are earning income in the present and saving for the future. There are also individuals who are now consuming the proceeds of their savings -- individuals who would receive a tax reduction under a capital gains tax cut but not under an IRA. That is, the economy is composed of many over-lapping generations of individuals in different stages of their life cycle. If individuals are not identical, savings might rise or fall depending on whose taxes are increased (and what other effects those alternative taxes might have). ¹ For example, if taxes are lowered on individuals with a high propensity to save and raised on those with a low propensity to save, aggregate savings would rise. In a stylized life cycle model of the economy, savings would tend to depend on age, so that savings could be increased by taxing the old and subsidizing the young. Since the tax savings of IRA's are more directed toward the younger savers and tax savings from capital gains are more directed toward the older holders of capital, such life cycle models would tend to find IRA's to have a larger effect on saving than capital gains tax incentives of similar magnitude. In such a life cycle model, an IRA provision would increase savings, even in the absence of a price effect, while a capital gains tax cut could easily reduce savings. ²

ECONOMETRIC EVIDENCE ON SAVINGS BEHAVIOR

SAVINGS BEHAVIOR IN GENERAL

Economists normally use statistical analysis to try to estimate behavioral relationships. There are two kinds of data sets used: time series and cross section. In time series studies one would look at how the dependent variable -- in this case savings or the savings rate -- changed over time as other relevant variables such as income, price (in this case the net rate of return), and wealth changed. The life cycle model would also suggest that demographic variables be included in such studies. Most studies of savings behavior have relied on time series. It is perhaps obvious from the discussion of the theory above that such estimation can be quite complex. Moreover, time series studies have a limited number of observations which makes statistical inference more difficult and constrains the number of explanatory variables. Some studies attempt to estimate the interest elasticity of savings (the percentage change in savings divided by the percentage change in net return) without considering the complex structure of life cycle models. Others have attempted to use more complex models to directly estimate the inter-temporal substitution elasticity (the percentage change in the ratio of present to future consumption divided by the percentage change in the ratio of prices of present and future consumption). This latter measure isolates the substitution effect from the income effect.

While a full review of the literature on the savings response is beyond the scope of this paper, most studies have found changes in the interest rate to have little effect on the savings rate and have estimated the inter-temporal substitution elasticity to be quite small, well below one. ³ There is also the long standing observation that U.S. savings rates have remained relatively constant over time through most of history, despite significant changes in interest rates and other variables.

The other type of data set used is cross section, where one examines individuals with different income and other characteristics. If one assumes these individuals are otherwise identical in their preferences, then the effects of changing these prices and incomes could be studied by examining the behavior of individuals with different incomes and prices. These data sets seem very attractive because they have many observations. Nevertheless, they have a fundamental flaw in their use for the purpose of explaining price effects. In the absence of taxes, the prices facing all individuals are identical so that there would be no way to estimate a price elasticity. Because of progressive income taxes, however, individuals do face different prices because they face different marginal tax rates. These marginal tax rates are, however, highly correlated with income and it is difficult, therefore, to obtain reliable estimates of both price and income effects. ⁴ In particular, it would be possible that there is no true price effect, but to have the tax price variable statistically significant if the true functional relationship between the dependent variable and income (e.g. linear, quadratic,) is not used. In addition, to the extent that tax rates vary for individuals with the same income levels, the implicit assumption that individuals have the same preferences is violated. To use a simple illustration, if two individuals had identical incomes and one had a lower marginal tax rate because of extraordinary medical costs, the individual with the lower marginal tax rate might be drawing down savings because of cash flow needs. Hence, lower marginal tax rates might be associated with reduced IRA contributions (or with increased capital gains realizations, for that matter) because of an unrelated variable. There are a host of other examples of this problem.

ESTIMATES FROM A LIFE CYCLE MODEL

Although these observations suggest that it is very difficult to estimate the appropriate savings elasticity, it may nevertheless be useful to consider what type of effect on savings a life cycle model might produce given a reasonable measure of the substitution elasticity. Auerbach and Kotlikoff, after surveying the literature, conclude that an elasticity of .25 best represents the state of knowledge about this relationship.

Gravelle reports estimates of the effects of a full wage or income substitution using several different versions of a life cycle model. ⁵ One model is quite similar in structure to that developed by Summers /6/, in that it assumes that labor supply is fixed and that the government will finance its revenue loss by an increase in other taxes. Two versions of the model are considered. In one version, the uncompensated version, there a considerable effect from redistribution across the generations. This version allows the income effects to operate. The second version, the compensated version, eliminates these redistributional effects by assuming that tax increases exactly offset tax savings for each individual. She estimates that a complete substitution of wage taxation for the income tax without redistribution would REDUCE the capital stock by 5 percent in the long run, while complete substitution of a consumption tax would increase the capital stock by 30.7 percent. The capital stock falls in the wage tax substitution because income is taken from the young who are savers and given to the old who are no longer saving. In this case the income effect more than offsets the substitution effect. In the case of the consumption tax, the redistribution occurs the other way and reinforces the substitution effect. The capital gains tax cut would be modeled as a wage change, while the IRA benefit would be modeled as a consumption tax. In the compensated version of the model where these income effects are eliminated, the effect on capital is the same -- an increase of 12.4 percent. ⁷ This increase in the capital stock arises from the operation of the substitution effect alone.

These estimates are for the long run effect on savings. In life cycle models, the savings rate will rise much more quickly in the short run than in the long run.

The tax changes under consideration, however, are much more limited than a full tax substitution. Consider first the indexation of capital gains. Capital gains taxes are only one component of taxation in the economy -- taxes on interest and dividends, the net corporate tax, and taxes on unincorporated business profits are also part of the capital income tax structure. The current overall aggregate effective tax rate on capital income in the economy is estimated at 30.85 percent. ⁸ The indexation of capital gains on corporate stock is estimated to reduce that tax rate to 29.16 percent for a reduction of 1.69 percentage points. Since corporate stock accounts for about a third of gains, the reduction would decrease the overall tax rate by three times that amount or 5.07 percent. To replace the revenue, however, assuming a uniform income tax increase, would also require some increase in capital income taxes as well as labor income taxes. Assuming that capital income is responsible for a quarter of income, we take the reduction to be 75 percent of that amount or approximately 3.80 percentage points. This reduction reduces the overall effective tax rate on capital income by 12.3 percent. (A similar calculation for a permanent thirty percent exclusion would yield a smaller reduction -- only 55 percent as large as indexation).

Since a capital gains tax cut is similar to a wage substitution -- it exempts current income -- we could simply apply the estimates to the 5 percent reduction in the full tax substitution and obtain an estimate of a reduction in the capital stock and the savings rate of .6 percent -- a negligible and negative change. Since the capital gains tax benefit is limited to newly purchased stock, we might also wish to consider the compensated effect. (This approach also avoids the issue of revenue feedback). In this case the capital stock would increase by 1.5 percent. Since the savings rate is proportional to the capital stock in the steady state, the estimated effect on the savings rate, using the rather generous fully compensated estimates, is an increase of 1.5 percent as well. If we take the overall savings rate (net savings out of net national product) as typically around 8 percent, the long run effect of this change would be to increase the savings rate from 8 percent to 8.12 percent. This must be considered a quite modest change in the savings rate. Nor would increasing the elasticity of substitution change this conclusion very much: a unitary inter-temporal substitution elasticity which would be considered quite generous by any review of the literature would increase the rate to around 8.24 percent. Moreover, as noted above this is an upper limit to the estimate; redistributional effects, depending on how the revenues are financed, could easily cause this effect on the savings rate to be negative.

To reemphasize that the price elasticity is this model is reasonably conducive to large savings effects, it may be useful to note that a direct application of one the largest savings elasticity, the .4 estimated by Michael Boskin, ⁹ would produce a "back of the envelope" calculation in the same range. The capital gains benefit, assuming some general source of financing and holding the interest rate constant would result in an increase in net rate of return of 5.49 percent (the tax savings divided by the after tax return, or .038/(1-.3085)). Multiplying by the .4 elasticity would produce a 2.2 percent increase in the savings rate, which, assuming an initial savings rate of 8 percent, would increase the savings rate to 8.18 percent.

It is much more difficult to estimate the effects of IRA changes because IRA's are relatively new and the share of assets held in them in the long run steady state is difficult to determine, although the effect on the savings rate should be unambiguously positive. Moreover, both price and income effects are difficult to assess because of dollar limits on these accounts and because deferral is already available on IRA accounts. There are limits on the amount of assets which can be held in IRA's (obviously owner occupied housing and the equity portion of non-corporate business cannot be). Because of the consumption tax nature of the change, the uncompensated effect would be larger than for a capital gains tax benefit of equal magnitude (that is, an equal proportional reduction in effective tax rate); the difficulty is in determining the magnitude. Some rough calculations based on the 1986 share of IRA's in personal savings suggest that the magnitude of reduction in tax rate is smaller than that for capital gains, however.

These estimates are from a life cycle model which is highly simplified. These estimates are those for the long run steady states; initial changes in savings tend to be larger, and then fall as the expanding capital stock drives down the interest rate. Thus, such estimates should be seen not as predicting the long run effects of these policies but rather suggesting in what general ball park they might reasonably be expected to be. What is striking about the above estimates is not so much how they might compare, but that they are quite modest. This conclusion should, however, not be surprising since they involve rather small revenue effects compared to GNP although they may involve large tax savings for the individuals affected. If the goal of government policy is to increase savings, this goal could probably be accomplished with more certainty, more impact, and less redistribution across the income classes with an increase in government savings -- i.e., lowering the deficit. Any forced savings, however, redistributes from the current to future generations.

THE IRA STUDIES

While the preceding section suggests that standard analysis of savings behavior suggests quite modest effects on savings rates from these proposals there have been a number of studies specifically directed at IRA's. Some of these studies have claimed to show or have been characterized as claiming to show that IRA's were extremely successful in increasing savings. There have been a number of these studies which use a cross section rather than a time series approach. We review the two most commonly discussed cross section studies; other cross section studies face the same difficulties. ¹⁰ We also consider a time series study which has also had some attention.

THE VENTI AND WISE STUDY

Perhaps the most widely cited studies in support of IRA contributions being net savings are a series of studies done by Venti and Wise. All of the studies involve the same theoretical framework, and the discussion here is concentrated on the most recent one, "Have IRAs Increased U.S. Savings? Evidence from Consumer Expenditure Surveys". ¹¹ The conclusion of this study is straightforward: an increase in the IRA limit would come about one third from tax savings and two thirds from consumption, with virtually no reduction in other savings, and the effect of an increased limit would be quite pronounced. Their results, therefore, suggest that IRA's are very effective in expanding savings and that the contributions to IRA's in excess of the revenue loss are largely net additions to savings.

Given the complex theoretical problems described above and the magnification of these problems by the particular aspects of IRAs -- contribution limits and penalties -- it seems difficult to see how such a study could be done using cross sectional data which essentially does nothing more than report patterns of IRA and non-IRA savings for a group of individuals. Essentially, however, Venti and Wise's conclusions are largely predetermined by the functional relationship between consumption and savings that they begin with. Their results depend very much on the notion that IRAs, presumably because of the penalty for early withdrawal, are imperfect substitutes for types of saving. While it is this preference function which allows them to actually use their data set, it is a highly questionable one.

In order to make their problem tractable, they first abstract from the complications of the life cycle model or even the simplified present and future consumption model discussed above by writing a specific functional form of the relationship between consumption and savings (technically a Cobb-Douglas relationship.) If there were only one type of saving and IRA's were perfectly substitutable this model would predict that consumption would not change at all, that savings would increase by the amount of the government's revenue loss, and that if the government only allowed the tax benefit in the future that no change in savings would occur. In terms of the examples used earlier to depict income and substitution effects, this functional form would predict that savings would move from $100 to $200. IRA's would be funded partially from existing savings and partially from government revenue losses (one half from each, if the tax rate is fifty percent). Since the model does not deal at all with the tax consequences or any occurrences in the future, there is no implicit price elasticity; if the tax benefits were fully compensated the implication is of a zero price elasticity and if they are not the implication is probably of a quite substantial price elasticity. Rather, their model is simply one with a zero savings elasticity without specifying whether this result is the result of offsetting income and substitution effects or simply a zero substitution effect with a fully income compensated change.

How, then, do they move from a zero savings elasticity model to one which predicts that increasing the limit on IRA's will increase savings -- that is one which predicts a very powerful net savings effect? This result arises from the way they enter the IRA into the expenditure function. Essentially, they treat IRAs as commodities which are new goods whose substitutability with the other commodities is uncertain. This new good is, however, constrained by the government to be zero in certain years and to have dollar limit in others. They now have a preference function not for two items -- consumption and savings -- but for three, consumption, ordinary savings and savings through IRA's.

If the new good is a perfect substitute for other savings, the original conditions of the model will hold and all increases in IRA savings come from either the government revenue loss or non-IRA savings. But, they use the data to estimate a substitution elasticity between IRA and non-IRA savings and then use that data along with estimation of the parameters of the preference function to produce their estimates. This approach raises several questions. First, in order to calculate the effects of changes in the IRA limits they have to determine what the parameters of the preference function are in the case where there are no IRA limits. They do this by assuming a normal distribution of desires for IRA's, and observing the distribution of IRA contributions. This allows them to predict IRA contributions if there were no limits, a necessary step to setting the parameters of their preference function.

Secondly, they use observations of savings outside of IRA's and IRA contributions to estimate the substitution elasticity between the two. If IRA's and non IRA savings were perfect substitutes then individuals should first save through the tax favored IRA's; only if their saving exceeds the limit would they should save through other means. The observation, then, that individuals frequently save outside of IRA's without reaching the IRA limit would, in the context of the model, be evidence that IRA's and non-IRA savings are not very close substitutes. Indeed, their estimates indicate a relatively low substitution elasticity and suggest that IRA saving is as close a substitute to consumption as it is to non-IRA saving.

It is this finding that then sets the stage for the large savings effects. First individuals are constrained in purchasing this desired commodity through dollar limits and if the dollar limits are increased the desired amount of IRAs will increase substantially because they are constrained to be below their desired level. For individuals at the limit, they will increase dollar for dollar for a small change. Moreover, they will increase substantially for a large change as well because the estimates of unconstrained desired IRAs are very large. Indeed, they estimate that IRA's if allowed without limit would expand to 24 percent of a marginal disposable income! (Actual IRA contributions in 1983 were only 1.3 percent of disposable income.) Needless to say, such an estimate seems obviously implausible, in light of history where savings rates have been remarkably stable and always under ten percent of income. But these increases will now come at the expense of both consumption and other savings for which they are equally good substitutes -- and they will be in proportion to the shares spent on those goods. Since the share of income spent on other savings is generally quite small, most of the increase in IRA contributions will come out of consumption.

The fundamental flaw in their reasoning, aside from the extremely truncated representation of a more complex life cycle model, is assuming that IRAs are a new commodity. They, of course, are not -- saving for retirement or for any other purpose has always been available and IRA's simply change the price. Because of the penalty for withdrawal, IRA's are more costly for short term savings; they are less costly for long term savings. Thus, the proper way to write the constraint at least in a world of certainty (in highly simplified form) would be to denote short term and long term savings as the arguments and treat IRA's as perfect substitutes for long term savings.

To clarify this point, suppose consumers eat a mix of apples, oranges and bananas and the government introduces a limited number of lower priced bananas. Their model lumps oranges and regularly priced bananas together as an identical commodity and treats lower priced bananas as an entirely new good. They aggregate not by the nature of the commodity but by the price, which is an incorrect method of modeling preference functions. If they aggregated in the normal fashion, the remaining structure of the model would then predict that a small change in the IRA limit would have no effect on savings aside from the additional tax savings; it would replace non-IRA long term savings, at least for those cases where long term savings is greater than the IRA limit. The data series would be largely irrelevant to this conclusion which would be determined a priori by the basic functional relationship.

This observation does not mean that their model, properly specified, would demonstrate that IRA's do not increase savings, because that result would then derive from the structure of their model, which does not take into account the theoretical relationships -- income and substitution effects. It simply suggests that the application of the data to their model cannot really answer the question.

There are some other criticisms of a technical nature which might be made of their results, ¹² but this basic one obviates the rest. Essentially, the Venti and Wise results are predetermined by how they set up the theory in the first place -- by treating IRA's as a new commodity rather than a change in the price of an existing commodity.

THE FEENBERG AND SKINNER STUDY

Although the Feenberg and Skinner study, "Sources of IRA Saving" has been interpreted as presenting a strong case for IRA's expanding savings, the authors' actual claims are much more modest. The authors use a panel data set over five years taken from the IRS- Michigan sample, which allows the tracing of the behavior of families. They run a regression to estimate IRA contributions as a function of income, wealth, tax rates, and other variables. Many of the findings they discuss are actually, however, from simply examining some of the data.

They first tackle the issue, discussed above, as to whether IRAs are funded by switching existing assets or by borrowing. Their data suggest that at least in the aggregate, IRA's are not financed by either borrowing or switching assets. They actually find that interest and dividend income expands more rapidly for those who invest in IRAs, outside of taxpayers with wealth in excess in $50,000 where essentially little difference is observed. If IRA had been replacing existing assets, they suggest, then the growth in dividend and interest income should have been slower for those with IRAs. Similar results are found when examining the changes in the level of debt.

Although these results suggest that in the aggregate individuals did not finance IRAs by reducing assets or by borrowing, such results do not prove that additions to IRAs were net savings or that some of this asset shifting and borrowing did not occur. Recall that IRAs can be financed from four sources (outside of the tax savings itself): shifting of existing assets into IRAs, borrowing, diverting new savings into IRAs, or reducing consumption. If on average individuals who are investing in IRAs would have been increasing their savings by greater than the average savings rate in the economy, one could easily observe increases in IRAs without those increases necessarily reducing other savings. Thus, their observations do not necessarily demonstrate that IRA's increase the overall savings rate, a point that they acknowledge.

Feenberg and Skinner also make some points about another issue which complicates the assessment of IRAs -- the notion that IRAs have increased savings because of factors not normally considered in setting up economic models. For example, they suggest that the gratification of an immediate up front deduction might play an important role in encouraging individuals to save, and that advertising which is highly related to the availability of the deduction also encourages individuals to save. They cite a couple of observations to support this contention. One is that many married working couples who are eligible for $2250 (non-working spouse) or $4000 limits (working couples) contribute exactly the $2000 limit, which suggests that they may not be aware of the larger limit. Secondly, they note a positive correlation between owing taxes to the IRS and amounts contributed to IRAs. They suggest that there may be a psychological effect operating.

Although these are interesting observations, there are alternative explanations. Married couples may nevertheless do some of their financial planning independently. Under-withholding is correlated with a number of factors which might also be correlated with larger IRA contributions -- higher incomes, married working couples, self employed individuals, and a larger fraction of capital income. Although they have a control for this last effect, it is a relatively crude one. Moreover, even if individuals open IRA's because of non-economic factors, they may still finance them from other savings rather than from consumption.

A somewhat more persuasive piece of evidence on this advertising effect was not available to Feenberg and Skinner, and that is that IRA contributions by eligible individuals also fell from 1986 to 1987. This effect could have been because marginal tax rates fell at least for some individuals; it also could, however, have reflected a perception that IRA's were no longer available at all because of reduced advertising by financial institutions. But again there is no way of determining whether these reductions were substitutes for consumption or for non-IRA savings.

THE CARROLL AND SUMMERS STUDY

Another study which has received some attention is the study by Chris Carroll and Lawrence H. Summers comparing savings rates in Canada and in the United States. ¹³ This is a time series study which suggests that the divergence in savings rates may be partially explained by the generous tax sheltered savings plan in Canada. The conclusions of this study seem ambiguous: although the authors suggest that the availability of more generous Canadian sheltered savings plans is statistically significant in explaining the divergence of savings between the U.S. and Canada, tax sheltered plans were not statistically significant in a regression for the U.S. alone. In any case, the degree of reliance one places on these results depends on how valid one feels it is to treat Canada and the United States as similar enough to use a cross country comparison. In this case, the normal difficulties of time series studies are compounded by the difficulty of a cross country comparison.

CONCLUSION

A review of the theory and empirical evidence surrounding the effects on savings of capital gains tax reductions and IRA's suggests that while it is difficult to determine which proposal would have a larger effect on savings, there is reason to believe that the effects on savings would be quite modest. If increasing savings is the major policy object, a reduction in the deficit would appear to be a more certain and more powerful way of achieving such an end.

² IRA's and capital gains cuts differ somewhat in their distribution by income class. While both proposals are directed towards higher income individuals, the capital gains cut is particularly concentrated in the very highest income classes. There is no evidence, however, that the marginal propensity to save differs among the income levels after correcting for life cycle and transitory effects. (See Barry Bosworth, Tax Incentives and Economic Growth, Washington, DC: Brookings Institution, 1984, pp. 85-90.)

³ See A. Lans Bovenberg, "Tax Policy and National Saving in the United States: A Survey," National Tax Journal Vol. 62, June, 1989, pp. 123-138 for a recent survey of the literature. A brief survey of studies of the estimates of the inter-temporal substitution elasticity may be found in Alan Auerbach and Laurence Kotlikoff, Dynamic Fiscal Policy, New York: Cambridge University Press, 1987, who choose a value of .25 for their model based on this survey.

⁴ This point is made in Dan Feenberg, "Are Tax Price Models Really Identified: The Case of Charitible Giving." National Tax Journal 60 (December 1987): 629-634.

⁵ Jane G. Gravelle, "Income, Substitution, and Wage Taxation in a Life Cycle Model: Separating Efficiency from Redistribution." Manuscript. June, 1989.

⁶ Summers, Lawrence. "Taxation and Capital Accumulation in a Life Cycle Growth Model," American Economic Review 71 (September 1981): 533-554. His model is modified to include bequests (which have a very small effect on the results) and to conform the tax rate to Gravelle's estimates.

⁷ A second version of the model is one in which labor is endogenous and is basically the same in structure as the Auerbach- Kotlikoff-Skinner model (full model details are presented in Auerbach and Kotlikoff, Dynamic Fiscal Policy, New York, Cambridge University Press, 1987) again modified for bequests and consistent with her tax rates. In this model the uncompensated substitution of wage for income taxation results in an 11.4 percent increase in the capital stock while the consumption tax substitution results in a 34.2 percent increase. The compensated substitution results in a 15.4 percent increase for the wage tax and a 17.4 percent increase for the consumption tax. While the effects on the capital stock are somewhat larger in this model, and savings go up, the labor supply falls -- by about 2 percent for the consumption tax and about 5 percent for the wage tax. Because of this, except for the uncompensated consumption tax, welfare in the long run steady state actually falls.

⁸ The method of calculation is reported in Jane G. Gravelle, "Non-Neutral Taxation and the Efficiency Gains of the 1986 Tax Reform Act," National Bureau of Economic Research Working Paper No. 2964, May 1989.

⁹ Michael Boskin, "Taxation, Saving, and the Rate of Interest," Journal of Political Economy 86 (1978): S3-S27.

¹⁰ Aside from the studies discussed, see T. Glenn Hubbard, "Do IRAs and Keoghs Increase Savings?" National Tax Journal 38 (March 1984), pp. 43-54; Cherie J. O'Neil and G. Rodney Thompson, "Participation in Individual Retirement Accounts: An Empirical Investigation," National Tax Journal 60 (December 1987), pp. 617-624; Julie H. Collins and James H. Wykoff, "Estimates of Tax Deferred Retirement Savings Behavior," National Tax Journal 61 (December 1988), pp. 562-572; James E. Long, "Taxation and IRA Participations: Re-examination and Confirmation," National Tax Journal 61 (December 1988), pp. 585-590, Cherie J. O'Neil and G. Rodney Thompson, "Taxation and IRA Participation: a Response to Long," National Tax Journal 61 (December 1988), pp. 592-594.

¹¹ Steven F. Venti and David A. Wise, National Bureau of Economic Research Working Paper 2217, April 1987. Venti and Wise have published several other studies of a similar nature which are referenced in that paper.

¹² Comments on an earlier Venti and Wise study by Angus Deaton are included in The Effects of Taxation on Capital Accumulation, ed. Martin Feldstein, Chicago: University of Chicago Press, 1997. Deaton comments that "the results are extremely implausible".

¹³ "Why Have Private Savings Rates in the United States and Canada Diverged?", Journal of Monetary Economics 20 (1987), p. 249- 279.