CRS STUDY IS CRITICAL OF TREASURY'S REVENUE ESTIMATES FOR A CAPITAL GAINS CUT.

                       CRS Report for Congress

The Administration's tabulation of the empirical literature contains a number of problems in the reporting and calculation of elasticities which have led to an overstatement of these estimated effects. When adjustments are made, the Treasury's elasticity appears to be relatively high compared with the literature. Moreover, there are a number of reasons to believe that the estimates from cross section studies are very unreliable, and for a variety of reasons it also appears that most of the estimates are likely to be biased upward.

The Administration has also argued that the induced growth in the capital stock resulting from the capital gains tax cut would offset the revenue shortfall predicted by the Joint Committee on Taxation. This finding does not hold up under scrutiny. The Administration's cost of capital estimate which drives the effect on capital formation is overstated and inconsistent with the revenue loss and the implicit savings elasticities are far too large. When the dynamics of capital accumulation and debt accumulation are taken into account, the result is a magnification of the revenue loss.

It appears unlikely that the capital gains tax cut could pay for itself. The Joint Committee's estimates appear more reasonable predictions of the likely revenue consequences than those of the Administration and may, themselves, substantially understate the revenue loss associated with cutting capital gains taxation.

                              CONTENTS

INTRODUCTION

The Administration proposes to restore a tax benefit for capital gains by allowing thirty percent of these gains to be excluded from income. ¹ For the budget period FY 1990 to FY 1995, the Joint Committee on Taxation estimates that the proposal will lose $11.4 billion while the Administration's Office of Tax Analysis (OTA) estimates that it will gain $12.5 billion. While there are differences in the projected magnitudes of capital gains realizations under current law, the difference in sign of these estimates derives from differences in the estimated increase in realizations induced by the tax cut.

Both estimates include a substantial behavioral response. The Joint Committee on Taxation (JCT) estimates that the static revenue loss from the proposal (assuming no change in behavior) would be $100.2 billion over the five year period, but taxes on new realizations will be $78.4 billion. Another $10.6 billion in increased taxes will accrue from miscellaneous features of the change, primarily the recapture of depreciation. By 1995, the proposal is estimated to lose $3.1 billion: a loss of $20.9 billion offset by $13.8 billion in taxes due on increased realizations, with another $4 billion arising from miscellaneous features. The total loss in 1995 is $3.1 billion.

The OTA estimates show the static loss at $74.7 billion, offset by $84.6 billion in taxes on induced realizations. There is a further loss of $5 billion due to the switching from ordinary income to capital gains assets. Another $7.3 billion would be gained from miscellaneous features, again primarily recapture of depreciation. For 1995, the proposal is estimated to lose $15.5 billion, offset by $16.3 billion due to induced realizations. There would be a loss of $2 billion due to shifting of ordinary income into capital gains, and a gain of $2.5 billion due to miscellaneous provisions. The total gain is $1.4 billion.

The Administration argues that its higher estimates of tax induced realizations are more consistent with economic studies than the JCT's, but are nevertheless extremely conservative. In a recent development, the Administration has also argued that even if the more conservative JCT estimates are used as a starting point, increased revenues from the economic stimulus will produce offsetting revenues which will cause the proposal to pay for itself over the first five years and more than pay for itself over the first ten years. We review both of these arguments in the following sections.

THE REALIZATIONS RESPONSE

As the figures cited above indicate, both estimates assume a substantial response of taxpayers in the level of capital gains realizations. Although the Administration has assumed a more pronounced response, they have also claimed that their responses are quite conservative. Assistant Treasury Secretary Kenneth W. Gideon states in his testimony of March 6, 1990 before the Senate Finance Committee:

"OTA's revenue estimates were made after a careful review of the major empirical studies by experts in the government and the academic community. Compared to the results in most of the studies, OTA's estimate of induced realizations is conservative. Table 1 provides detail on these studies . . . By any reasonable standard, OTA has endeavored to err on the side of caution when estimating these behavioral results."

A similar view is expressed in a letter dated March 6 written to several Members of Congress Michael Boskin, Chairman of the Council of Economic Advisors, who also refers to these studies:

". . . I strongly believe that the Treasury estimates are superior to the JCT, and that they probably understate the increase in revenue resulting from a cut in the capital gains tax."

Boskin stresses that the Treasury estimates are smaller than the estimates in nine of the existing twelve studies.

Two witnesses at a hearing on February 27, 1990 by the House Budget Committee had a different perspective. Henry Aaron of the Brookings Institution stated:

My own view is that the Administration's proposal should be treated as the upper limit of any possible revenue gain, but that the JCT estimate is nowhere near the bottom end of the range of plausible estimates of the revenue loss."

Similarly, Alan Auerbach of the University of Pennsylvania stated at the same hearing:

". . . I find the alternative revenue estimates produced by the Joint Committee on Taxation to be much more plausible than those of the Administration."

The Administration has appealed to the econometric literature on this issue to support their case that a capital gains tax cut will pay for itself and clearly this is not a unanimous view. The following sections are a review of this issue. We first suggest that the numerical results cited by the Administration do not allow a proper comparison of the OTA elasticity with those estimated in the literature and are not entirely representative of the empirical findings in a number of studies. ² Secondly, our assessment of this literature suggests that we should be quite cautious in relying on these studies. In particular, there are very serious problems with the micro-data studies. Micro-data studies examine the characteristics of different taxpayers rather than the relationship between overall capital gains and taxes over time. In light of this assessment, the JCT estimates seem more appropriate and indeed may very well understate the revenue loss associated with the capital gains tax cut, a view also suggested by Aaron.

ADJUSTMENTS IN THE SURVEY OF ELASTICITIES

Both Gideon and Boskin refer to a table surveying the elasticities from the twelve studies included in Gideon's testimony and reproduced below in Table 1. The Administration's tabulation does indeed suggest that the OTA estimate is conservative. In terms of ranking the studies from high (1) to low (12) elasticities, the OTA estimate ranks exactly at number 9 and the Joint Committee's is between 9 and 10.

The Administration's tabulation contains, however, a number of problems in representing or reporting the results of the empirical studies. Moreover, one can assess the merits of these studies more easily by characterizing the research methods used. Table 2 reclassifies the studies by type and presents the corrected numbers, along with adding the 1982 Auten study and the 1986 Congressional Budget Office (CBO) study. An alternative result reported by the CBO in their 1988 study is included because, for reasons suggested below, it may help to illustrate the importance of accounting for changes in accrued gains.

A number of studies estimate the values for an equation which yields a different elasticity result depending on size of the tax rate. Some of the numerical corrections are to seek out the values which are appropriate for assessing the current proposal. The most appropriate value to use for a given proposed change would be a value which is midway between the old and new tax rates, a value equal to 85 percent of the current marginal tax rate of .257, or approximately 22 percent. This correction tended to lower the elasticities because most of them were derived from an earlier table included in Auten, Burman, and Randolph (1989) which valued the elasticities at higher rate than that appropriate to the proposed thirty percent exclusion. When the OTA reported its own elasticity, however, it was valued at a lower rate than appropriate, 20 percent. These differences in choice of tax rate make the OTA elasticity look relatively low. These tax rate corrections were made in all cases where they were possible; no correction was possible for the micro-data studies because of insufficient data or for the Treasury 1985 time series data which was run in a form which makes it difficult to adjust to current circumstances. (In some cases, especially for the Auten and Clotfelter study of a middle income sample, the elasticity will be understated because it was measured at lower tax rates; on the other hand, two of their three basic equations produced long run elasticities which were lower still and not statistically significant).

The second reason for correcting the elasticities is that a number of them are reported as the midpoint of a range reported in the survey table in the Auten, Burman, and Randolph study. Their purpose in specifying this range was to illustrate the variation in point estimates as a background for a discussion of the causes of variation in the econometric studies. Reporting the midpoint of a range may, however, misrepresent the general findings of studies if there is an extreme value. Moreover, authors may report functional forms which they find questionable to illustrate some point they are making, even though they state clearly that they do not prefer a particular equation. The elasticities reported in Table 2 reflect either the averages of those equations which the authors themselves indicate that they find to be reasonable specifications, a stated preferred equation, or the equation chosen for further use in simulation work.

Finally, in one case (Jones) there appeared to be a simple mistake in reporting the elasticity. The specific details of the corrections are presented in Appendix I.

The OTA elasticity was also altered to include the elasticity before portfolio effects. The OTA estimates that realizations will rise for two reasons: an unlocking of gains which will increase revenues and a shift into capital gains yielding assets from other assets which will lose revenues. For example, realizations may go up because individuals sell assets they otherwise would not have sold (the unlocking effect) or they may go up because individuals will shift assets from ordinary income into capital gains (the portfolio effect). Table 1 reports the total realizations effect, but the revenue estimates treat a small part of that effect as leading to a revenue loss.

      TABLE 1: SURVEY TABLE FROM TESTIMONY OF KENNETH W. GIDEON,

ASSESSMENT OF THE STUDIES

As indicated by the classification of the studies in Table 2, most econometric studies fall into two major classes: aggregate time series and micro-data cross section. The time series studies generally show the elasticity to be below one (the estimate necessary, roughly, to yield no revenue loss), and well below one in many cases. The cross section studies vary widely. Moreover, in both cases there are instances where the tax variable was not statistically significant, which means that we cannot reject the hypothesis that the realizations response is zero.

These differences reflect the basic data sources used to generate the estimates. Aggregate time series data sets use observations of total gains, average marginal tax rates, and other aggregate variables over time while micro-data cross section studies use observations of gains, tax rates, and other variables for a sample of individual taxpayers within a single time period. Panel studies are micro-data studies that add some limited time dimension to the cross section study of gains by tracing individuals over a few years. The Gillingham, Greenlees, and Zeishchang study includes micro data for several years but does not trace the individuals themselves, and thus we have classified it with the two other single year cross section studies by Feldstein, et al., and Minarik. The Lindsey study includes many years, but also disaggregates taxpayers by income class; this hybrid treatment makes the estimated elasticities of this study difficult to classify.

If these studies were measuring the same phenomenon, they should yield similar results. The fact that they are so divergent suggests that they are not. The following discussion is designed, therefore, to explain the major differences between the types of studies and their shortcomings. We conclude that it is probably safer to rely on time series estimates, simply because the problems with cross section studies appear to be crippling. Moreover, there are reasons to believe that the time series estimates may actually be overstated, suggesting a relatively low realizations elasticity.

There are, to be sure, shortcomings in both types of studies. Therefore, to some extent the public policy question is how to use highly imperfect information to decide how to accommodate a tax provision in the budgetary process. If we use elasticities that are too high we will have increases in deficits that reduce national savings and retard economic growth. If we use elasticities that are too low, we achieve somewhat more deficit reduction than otherwise planned. If the latter error is considered less damaging than the former, then we will wish to be extremely conservative in choosing our elasticities. Indeed, the Administration's stress on using conservative estimates suggests that is the view they hold. By that standard, even the JCT elasticity would probably be seen as too high.

There are a host of both econometric and theoretical problems associated with all of these studies, many of which are detailed in the studies themselves. Many of these problems are common to both types of studies. For example, none of the studies really captures well the basic theory of realizations behavior, in part because that theory itself is not really developed. Individuals may realize gains for consumption purposes which would require an extremely complex over-lapping generations life cycle model. They may wish simply to switch assets either because they have changed expectations or because they wish to re-balance their portfolios. These theories do, however, tend to suggest that the major source -- in some models the only source -- of permanent changes in realizations is the selling of assets otherwise held until death. ³ If individuals are not very willing to sell assets they otherwise intend to hold until death, then a cut in the capital gains tax might yield a temporary response, but not a permanent one. This occurs primarily because the initial increased realizations reduce the basis for future realizations and thus offset the increases arising from higher turnover rates. That is, the same income cannot be taxed twice. Yet, none of the studies really capture these dynamic elements, and with one exception they do not include changes in accrued unrealized gains as an explanatory variable.

Moreover, the studies try to capture the effect on realizations through a single variable, the capital gains tax rate, when the effect depends in part on the array of possibilities for investment and timing of gains. For example, to the extent that realizations reflect a portfolio response as well as an unlocking effect, changes in the capital gains tax which derive from changes in ordinary income have uncertain effects on realizations arising from portfolio shifts. In general, taxpayers are more likely to move into tax favored investments when ordinary tax rates are high. But the effect of changing the ordinary tax rate (as opposed to changing the preference for capital gains income) on realizations through this portfolio effect depends on where capital gains yielding assets rank among tax favored assets (i.e. they are favored relative to fully taxable assets, but disfavored relative to tax exempt bonds and perhaps tax shelters). Moreover, other features of the tax law (such as depreciation) and influences such as inflation which can alter relative degrees of tax preference also influence this portfolio effect, as well as the lock in effect itself. This simplified representation of a more complex model, therefore, causes problems for all of the empirical studies.

There are, however, some key differences in the types of studies. These differences are so important that they should affect how much we rely on different estimates and which category of estimates we choose as the most valid. The following discussion is a comparison of the problems with micro-data studies and time series studies. The reader is referred to the studies themselves and to some of the critiques cited below for more detail.

THE MICRO-DATA STUDIES

In many ways, cross section data seem initially more attractive than time series. The sample size is much larger, leaving more scope for the study of many explanatory variables. The initial studies used this data source. There are two important basic criticisms of these micro-data studies: failure to fully account for transitory effects, and failure to account for individual specific effects. ⁴

The transitory effects issue is relatively straightforward. An individual's tax rate may vary from one year to the next because of fluctuations in earnings or in measured taxable income. When an individual is in a temporarily low tax rate compared to his permanent rate he has a strong incentive to realize gains at that time, including speeding up realizations he would otherwise have deferred. Indeed, if we needed any proof that individuals respond to temporary tax differences, we need only examine 1986 when realizations rose dramatically because of the expectation that tax rates were going up. Similarly, when tax rates are above average he has a strong incentive to delay realization until his tax rate falls. Thus, even if we observe a strong inverse relationship between realizations and tax rates, this will not tell us what happens when tax rates are permanently lowered. Cross section studies cannot deal with this phenomenon at all, and thus must be considered quite suspect (and likely to overstate the elasticity, other things being equal).

Panel studies can try to address this problem, albeit imperfectly. This point was one of the important contributions of Auten and Clotfelter's study. They separated taxes and income into permanent and transitory elements, by treating the permanent values as the averages over the previous three year period and including a transitory element as the difference between the current value and the average value. They found the transitory tax rate to have a much larger and statistically significant effect on gains, while the permanent rate resulted in a much lower, and frequently statistically insignificant effect, suggesting that this issue of transitory tax effects is extremely important. The Treasury (1985) panel study used the same approach, while the Treasury (1989) study by Auten, Burman and Randolph effectively did so by including lagged tax rates. Because they included only one lag, their correction involves an average over two years rather than three.

Even so, the panel data corrections are not entirely satisfactory because two or three years are not enough to establish a permanent rate. Suppose for example, two individuals have a permanent tax rate of .25 percent but one individual's tax rate drops temporarily to .20 percent. The true deviation of his current tax rate (the transitory component) is .05 but the deviation will be measured as .033 and his permanent tax rate will be measured as .233 (the average of .25, .25 and .20) when a three year averaging approach is used. Thus, part of the transitory effect will still be ascribed to a permanent effect. This situation will be worse in the case of only a two year average, where the transitory component will be measured as .025 and the permanent tax rate as .225. Thus, the panel studies done to date must be viewed as only partially dealing with this transitory tax effect. In both of these cases, we would still expect that the realizations response would be overstated. This problem of transitory tax rates is further complicated by the fact that realizations may be affected by expectations of future changes.

The final basic problem with cross section studies is the presence of unmeasured individual specific effects. Cross section studies work well if the individuals being compared only vary due to measurable factors such as marital status and age which are included as explanatory variables. Otherwise it is not possible to tell how a given individual is affected when his tax rate changes. If individuals differ in their preferences, then it may be invalid to use findings from cross section to make inferences about the response to an exogenous change. Indeed, since all individuals face the same tax law, the fact that individuals with similar economic incomes have different tax rates strongly implies that they are different, particularly in their tastes in investment which tend to drive tax differentials for the high income individuals who realize most gains. These different tastes may simultaneously affect both their tax rate and their level of realizations. This is another way of saying that the independent variable, the tax rate, is not truly independent, but rather itself determined by behavior (endogenous).

Some aspects of this problem might be dealt with. For example, there is a technique which can be used in panels to deal with constant effects (such as a taste for risk), called fixed effects. One can study the deviation of current gains from the average gains over the period of the panel as a function of the deviation of other variables from their averages, so that the fixed effects drop out. Unfortunately, as Slemrod and Shobe ⁵ have pointed out, permanent tax rates are indistinguishable from other fixed effects because they are constant over time. Thus, only the transitory effect, which is meaningless for estimating the effect of a permanent tax change, may be consistently estimated if fixed effects are present.

It is largely for these reasons that considerable doubt as to the validity of cross section studies, even those with panels, has developed. Auerbach, for example, concludes that these studies simply cannot deal with the transitory and individual specific effects. ⁶ Moreover, those panel studies which span periods of exogenous tax change such as the Auten, Burman, and Randolph study are also affected by some of the issues of dynamic adjustment which complicate time series analysis. That is, individuals may be adjusting to a permanent as well as a temporary change in their tax rates, and it not possible to separate these components. Thus, while recognizing that this study was an extraordinarily sophisticated one in dealing with a number of complex econometric and theoretical problems, the results cannot be viewed as very reliable for estimating the response to a permanent tax change for all of the reasons cited above as the authors themselves suggest in their paper. ⁷

DIFFICULTIES WITH TIME SERIES

If micro-data studies seem inherently unreliable, can we then turn to time series studies? These studies all suggest that the elasticity is below one (the value roughly needed for the capital gains tax cut to pay for itself). Time series studies largely avoid some of the major problems of cross section since most of the variation in tax rates across time is clearly exogenous -- a result of tax law changes rather than taxpayer behavior. ⁸ Moreover, like micro-data studies the capital gains tax treatment is represented by a single variable, In aggregate studies, this tax variable is further collapsed into a single economy wide tax rate, which may result in aggregation bias. Time series also suffer from the inherent problems of small sample size. This small sample size limits the number of variables which can be included and causes an incomplete representation of the dynamics of adjustment. Some of the remaining problems, such as problems with time trends, have been addressed in some of the studies, such as Auerbach's and Jones's, by using differenced data.

Briefly, aggregation bias occurs because a lot of individual tax rates are averaged into one rate. This can create a problem if the responses of individuals in different income brackets are different and if the tax law changes did not proportionally alter the tax rates across income classes. And by averaging out variations, it may cause the estimated elasticity to be too low. Much has been made of aggregation bias by some critics of time series studies. The 1988 CBO study explored this issue by separately estimating the response for the top one percent and bottom 99 percent, finding that this effect was apparently unimportant. This result may not be surprising given the heavy concentratiori of capital gains among high income taxpayers. Thus while aggregation bias may be a problem it does not appear to be a significant one.

Finally, it is very difficult to capture many of the variables which influence realizations. This shortcoming, however, appears on the whole more likely to result in the time series estimates being overstated. There are three elements of this argument.

If asset prices are not changing erratically then realizations would tend to be a constant fraction of assets absent changes in other variables, and all values would grow at the steady state rate. For this reason, one of the variables which is usually introduced into a time series equation is asset values, either of corporate stock or some measure of tradeable wealth. However, when price changes of assets deviate from the rate of change in the economy this relationship would not be stable. To consider a simple example, suppose an asset would normally sell for $100 and the basis is $50. But suppose the price of stock now doubles. Asset value will rise to $200 but basis will not change. While the asset price has doubled, realized gains have almost tripled. This point is extremely important because much of the reason for finding a statistically significant relationship between realizations and tax rates is that realizations rose after 1978 in excess of what could be accounted for by other variables such as wealth and income and this rise was accompanied by a reduction in marginal tax rates. This was also a period, however, when the stock market was rising rapidly, and it might be that the rapid rise in realizations was reflecting not a response to tax cuts but rather the natural response to a surge in asset values.

Thus, time series estimates (and cross section, for that matter) should include a variable to reflect the effect of changes in accrued gains rather than changes in assets. But such a variable is difficult to construct because, although we know what changes in asset values are, we cannot observe changes in aggregate basis. Nevertheless, CBO did try to construct an accruals variable in one of their alternatives, which is included in Table 2. The elasticity dropped considerably and the coefficient on the tax term was not statistically significant. This finding is consistent with the discussion above but must be considered with caution because the stock of accruals is measured with error.

Another reason for overstatement of the responsiveness of realizations to tax rates is that there have been some institutional changes not captured in the regressions: the reduction in brokerage fees, the growing popularity of mutual stock funds which tend to have higher turnover rates (owing perhaps to both professional management and lower brokerage fees due to buying and selling in blocks), the growth in leveraged buyouts which may induce some realizations that might not otherwise have occurred, and introduction of reporting requirements for capital gains which may have increased compliance. The last issue was explored by the CBO (1988) study by using a dummy variable for the years of the reporting requirements. The dummy variable was not statistically significant, but the tax rate coefficient did fall, implying an elasticity of .58. While such an equation is not conclusive, it is suggestive.

A final reason that the time series estimates may be overstated involves dynamics. To the extent that increased realizations responses are decreasing holding periods of assets that would have been sold anyway rather than resulting in the realization of gains on assets held until death, the short run response will be considerably larger than the long run response. Moreover, if the response arises from portfolio rebalancing, the sales of the existing stock of assets in the first year or so will be much larger than the increase in turnover rates in the long run, leading to an initial surge of realizations which will then decline. Indeed both the Treasury and the Joint Committee on Taxation incorporate such differentials between short run and long run elasticities. Time series estimates may be primarily capturing a shorter term response which will be larger than the permanent response. ⁹ This issue of dynamics is complicated, however, by asymmetries in the response to tax reductions versus tax increases. Because only a small fraction of assets are sold each year, the short term response to a tax decrease might be more pronounced than the short term response to a tax increase, the latter being limited for any taxpayer to the annual level of gains already realized.

CONCLUDING NOTES ON THE REALIZATIONS RESPONSE

Although the critique in this section suggests that all of the estimates are unreliable, the flaws of cross section, including panel studies, appear to be more disabling than those in the time series. Certainly, all studies are not of equal validity and one's estimate is best guided not by an average but by an understanding of the weaknesses of different study techniques. Moreover, the omitted variables in time series do appear by and large to suggest that the time series estimates, themselves lower than the cross section estimates, may nevertheless be overstated. On the whole, therefore, the body of evidence does not suggest that elasticities are high enough for a capital gains tax cut to pay for itself through a realizations response; indeed, it suggests that the JCT estimates may well understate the revenue loss.

FEEDBACK FROM ECONOMIC GROWTH

This growth feedback issue raised in Assistant Secretary Gideon's testimony was also raised in another letter to several Members of Congress, also dated March 6, jointly signed by Michael Boskin, Chairman of the Council of Economic Advisors and Robert R. Glauber, Undersecretary for Finance of the Treasury Department. These statements claimed that the revenue loss from the capital gains tax cut would be made up by increased output arising from the stimulus to capital from the capital gains tax cut. This letter states:

Even using the extremely pessimistic JCT estimates and making lower bound assumptions, the capital gains tax cut would increase revenue over the next five years, once economic growth is considered.

The letter states that a conservative estimate is that the President's proposal would lower the cost of capital for businesses by 3.6 percent. There is apparently some recognition that the capital stock cannot adjust instantaneously. In any case, the conclusion is that over the next five years, the lower cost of capital will lead to an increase in GNP of $61 billion and raise revenue of roughly $12 billion, approximately offsetting the net revenue loss estimated by the JCT. In the next ten years, the lower cost of capital is said to increase GNP by $274 billion, with a revenue increase of $55 billion, which would swamp any loss.

This claim is considerably overstated for several reasons. First, a 3.6 percent reduction in the cost of capital is a major overstatement. Appendix II shows that this estimate is inconsistent with the magnitude of the revenue that would be lost in the absence of a realizations response. For corporations, we estimate that the percentage reduction in the required pre-tax return, holding net after tax return to equity constant, is only .9 percent. For aggregate capital, including owner occupied housing, it is about .8 percent. The primary reason for the overstatement in the estimate appears to be a failure to account for the large fraction of capital gains which are never realized due to step up in basis at death. Obviously, overstating the effect on the cost of capital by a magnitude of four would make it much more likely that the revenue gain from growth would appear to offset the cost of the tax cut, as a simple matter of arithmetic. Or, put another way, if the cost of capital measure used in the Council's analysis were reduced to .9 or .8 percent, the feedback would be less than $3 billion even without any of the other corrections discussed subsequently.

Moreover, if one is calculating the investment effects of a change in taxes holding net return constant, the change should be measured relative to the user cost of capital, the sum of the pre-tax return and depreciation rate, since this price drives investment. For the corporate sector, the depreciation rate is approximately seven percent, or almost half the total user cost. Thus, our calculations would suggest that the percentage change in the user cost of capital would be .5 percent. For capital as a whole, with a lower depreciation rate of .03 (owing to more structures and less equipment as compared to the corporate sector), the percentage change would be about .55 percent.

Furthermore, the growth estimate appears to assume an extremely elastic supply of capital. Growth is produced by an expansion of the capital stock. But a given percentage cut in the cost of capital, holding the net rate of return constant will not produce as large an effect on the capital stock if individuals require an increase in rate of return to supply more capital. Even the highest estimates of the supply elasticity of savings are no more than .4 to .6, and some evidence suggests that an increase in the rate of return will cause the savings to fall rather than rise, because the higher return to savings will allow individuals to consume more both today and in the future, even if their consumption in the future rises proportionally more (i.e. the income effects dominate substitution effects). For example, setting the elasticity to .6 would cut the projected increase in the capital stock by half even if one could attain the steady state instantaneously and not be concerned with the financing of deficits.

If we really want to answer the question as to whether a capital gains tax cut can pay for itself through a combination of realization responses and growth feed back effects, it is necessary to take into account the fact that the capital stock grows very slowly even if the savings rate increases. For example, net savings is typically only about two to three percent of the capital stock. Thus a 3 percent increase in investment in the first year would only increase the capital stock by less than one tenth of a percent. In addition, some of the savings during the adjustment period must be used to finance the deficit, and as the deficit grows, the interest on the debt and the need for borrowing grows. Normally, private savings would not be expected to rise enough to both finance the deficit and increase the private capital stock, since some large fraction of the tax savings is expected to be consumed. Therefore, if one traces the effect of a deficit financed savings incentive, even with a generous assumption of savings elasticity, the deficit simply grows without limit, and the capital stock declines continuously. The question then is whether this analysis can be altered by simultaneously including a realizations response and the feedback effects.

To explore this issue, we calculated an illustrative adjustment path, using the myopic adjustment model in Appendix II, under the assumption that the realizations response offsets all of the revenue loss in the first year, eighty percent in the second year, and seventy percent thereafter, and with one percent initial change in the pre-tax rate of return. ¹⁰ Normalizing our results to reflect a $12 billion loss after realizations response but before growth effects, we obtained the following results. Even for an extremely high savings elasticity of .6, these effects would increase the $12 billion cost over the first five years to $12.06 billion. In this case, the growth effects on the deficit are negligible. In the first ten years, however, the loss would rise from $30.23 billion with no feedback effects to $33.79 billion. These deficits would rise continuously. The capital stock would expand by a negligible amount for a brief period (to a maximum of 3/100 of one percent), and eventually fall as the deficit crowds out private investment despite the increase in savings. For a zero savings elasticity, the $12 billion cost would rise to $12.58 billion in the first five years; in the first ten years the cost would rise from $30.23 billion to $36.49 billion. Crowding out of private capital begins almost immediately with the capital stock 4/100 of one percent lower after five years and 12/100 of a percent lower after ten years.

While the assumption of a revenue feedback moderates the results of the normal crowding of private investment by the deficit, it does not reverse them. Thus, if a capital gains tax cut adds to the deficit, it will likely reduce growth and productivity in the economy.

CONCLUSION

The basic question in this study is whether the capital gains tax is likely to pay for itself through realizations and growth. The analysis in this study suggests that such an outcome is highly unlikely. The realizations response does not appear adequate to make up for the static revenue loss. Indeed, it seems quite likely that even the revenue loss predicted by the JCT may be understated. Moreover, even with the losses predicted by the JCT, borrowing to finance these losses will be larger than any induced savings, causing the capital stock to contract rather than expand. Thus, taking into account feedback effects in the economy will merely increase the projected negative effect on the deficit.

APPENDIX I: CORRECTIONS TO ESTIMATES

The Council of Economic Advisors, in constructing the table, relied in many cases on choosing the midpoint of a range of estimates listed in the Auten, Burman, and Randolph (1989) study. Choosing a midpoint can easily misrepresent the findings if there is one extreme value; a more appropriate procedure would have been to average the results where several elasticities are presented. Moreover, in some cases, the range is not correctly reported, or the authors were simply experimenting with different forms and clearly state preferences that do NOT include some of the experiments that were performed. In the case of Jones, there appears to be a simple mistake in the reported number. The 1986 CBO regression was apparently overlooked.

Since the elasticity can vary with the tax rate for most functional forms, the elasticities are all corrected where possible to reflect the midpoint of the range between the old and new tax rate; the Auten, Burman, and Randolph study derived elasticities at a somewhat higher rate. When corrections are made, they are adjusted to a tax rate of 85 percent of the Treasury's reported marginal tax rate of 25.7 percent (slightly under 22 percent). Specifics of the corrections follow. In some cases no corrections were made. For a number of the studies there is no way to determine an appropriate elasticity because of lack of information. These include the Treasury (1985) time series studies, and all of the micro-data studies; the only modifications for these studies are those where the reported elasticity in the table did not appear to reflect appropriately the elasticity in the study.

(1) Auerbach's number is taken from his most recent regression, reported in his 1989 study, which he states to be his preferred estimate.

(2) Darby, Gillingham and Greenlees' number is taken from the average of their three modified regressions excluding the regression which uses a quadratic form of the tax variable and which is not statistically significant; including it would lower the estimates. These three regressions include a semi-log form, a log, log form with the price variable being the after tax share, and a log, log form with tax rate variable. The latter yielded the highest elasticity of .67. The original Auten, Burman, and Randolph table includes a higher number from the inclusion of a large elasticity from a modification of the Treasury (1985) regression which the authors clearly find unsatisfactory because the elasticity is sensitive to the level of realizations. Such a regression cannot be easily interpreted for the current period. Note however that since their regression was run using the high income marginal tax rate, the elasticities for the semi-log form and the log form with the price variable being the after tax share, their elasticities should probably be increased to account for using higher overall marginal tax rates.

(3) CBO's 1988 number is the average of their four regressions evaluated at the 22 percent tax rate.

(4) CBO's 1986 number is a single regression evaluated at the 22 percent tax rate.

(5) Jones (1989) ran numerous regressions in a specification search. His preferred estimate, as indicated in the text, is .89, consistent with the number reported in the 1989 summary provided by Treasury in releasing the three 1989 studies. His regression was run in log, log form so that the elasticity is constant at every tax rate.

(6) CBO's alternative regression is evaluated at the 22 percent tax rate; it was not, however, statistically significant.

(7) Auten and Clotfelter reported four basic regressions with elasticities of .36, .37, .55, and 1.45. The last regression included losses which the authors did not appear to believe to be reasonable. The first two regressions were not statistically significant. We report the .55 elasticity. One could make a case that this elasticity over-represents the response they found, since lack of statistical significance technically means that you cannot reject the hypothesis of no response. On the other hand the average marginal tax rate in the sample was quite low, suggesting that the elasticity from the equation which is significant is understated.

(8) Treasury's (1985) panel study included five regressions. We report the one chosen for simulation analysis and the only one for which information on statistical significance was reported. The study suggests a variety of problems with the other regressions.

(9) Lindsey's estimate was corrected for the 22 percent tax rate.

(10) The elasticity actually used by the Treasury was changed to reflect response before portfolio effect, a value of .9 rather than .8. It is increased to .98 to reflect the higher marginal tax rate.

(10) The elasticity for the Joint Committee on Taxation was supplied to the author.

The following are references to the studies:

Alan J. Auerbach. Capital Gains Taxation and Tax Reform. National Tax Journal, September 1989, pp. 391-401.

Auten, Gerald E. Capital Gains Taxes and Realizations: Can a Tax Cut Pay for Itself? Policy Studies Journal, Autumn 1980, pp. 53-60.

Auten, Gerald E. , Leonard E. Burman, and William C. Randolph, "Estimation and Interpretation of Capital Gains Realization Behavior: Evidence from Panel Data." National Tax Journal, September 1989, pp. 353-374. (This study was also released by the U.S. Department of Treasury. OTA paper 67, May 1989).

Auten, Gerald E., and Charles Clotfelter. Permanent vs. Transitory Effects and the Realization of Capital Gains. Quarterly Journal of Economics, November 1982, pp. 613-632.

Congressional Budget Office. Effects of the 1981 Act on the Distribution of Income and Taxes Paid, Staff Working Paper, August 1986.

Congressional Budget Office. How Capital Gains Tax Rates Affect Revenues: The Historical Evidence. March 1988.

Feldstein, Martin, Joel Slemrod and Shlomo Yitzhaki. The Effects of Taxation on the Selling of Corporate Stock and the Realization of Capital Gains. Quarterly Journal of Economics, June 1980, pp. 777-91.

Gillingham, Robert, John S. Greenlees, and Kimberly D. Zieschang, "New Estimates of Capital Gains Realization Behavior: Evidence from Pooled Cross-Section Data." U.S. Department of Treasury, OTA Paper 66, May 1989.

Jones, Jonathan D. An Analysis of Aggregate Time Series Capital Gains Equations, U.S. Department of Treasury, Office of Tax Analysis Paper 65, May 1989.

Lindsey, Larry. Capital Gains: Rates, Realizations, and Revenues. National Bureau of Economic Research Working Paper No. 1893, April, 1986.

Minarik, Joseph. The Effects of Taxation on the Selling of Corporate Stock and the Realization of Capital Gains: Comment. Quarterly Journal of Economics, February, 1984. p. 93-110.

U.S. Department of Treasury, "The Direct Revenue Effects of Capital Gains Taxation: A Reconsideration of the Time Series Evidence." Prepared by Michael Darby, Robert Gillingham, and John S. Greenlees. Treasury Bulletin, June 1988.

U.S. Department of Treasury. Office of Tax Analysis. Report to the Congress on the Capital Gains Tax Reductions of 1978. September 1985.

APPENDIX II: MODELING THE MACROECONOMIC EFFECTS

A. Initial Cost of Capital Estimates

This appendix provides details on the measurement of the net cost of capital effects of the capital gains tax cut, why they differ from those of the Council of Economic Advisors, and presents a simple model to trace the dynamics of the economy wide response. The latter assumes a closed economy; with an open economy, the effects on the net cost of capital would be smaller for the corporate sector and larger for the noncorporate sector.

The cost of capital, R, is a weighted average of debt and equity costs of the following form:

(1) R = ((f(i*(1-u)-p) + (1-f)E)/(1-u)

where:

f = share of debt finance

i = nominal interest rate

u = corporate tax rate

E = real return to all stockholders before personal tax

p = inflation rate

We now define E* as the after tax rate of return to stockholders:

(2) E* = E(1 - st -(1 - s)gtx)-pgtx

where t is the average marginal tax rate on stockholders, s is the share of real income received as dividends, g is the modifier to account for deferral and exclusion of capital gains at death, and x is the fraction of capital gains included in income.

We measure the effect of the capital gains tax cut on the cost of capital by changing x from 1 to .7, and holding E* constant. The following values were used in the calculation. The debt share, f, was set at .4 based on flow of funds data. The nominal interest rate was set at .08; the inflation rate at .04, the corporate tax rate at .34, the initial real return to equity at .0883, the value of s at .67 and the value of t at .18, the marginal tax rate on capital gains reported by the Treasury after accounting for the holding of approximately thirty percent of equity by tax exempt shareholders (based on flow of funds data). The value of g is set at .21. Two thirds of gains are assumed to never be taxed because they are realized at death, based on a rough midpoint between the 76 percent share reported by Gravelle and Lindsey and the 50 percent reported by Auerbach. ¹¹ The effective rate was further reduced by taking into account the deferral advantage, assuming corporate stocks are held for seven years on average. This formula does not account for the fact that the exclusion rate is lower for assets held less than three years. The result for this formula was a .9 percent reduction in the cost of capital, equivalent to 8 basis points.

This number was cross checked against a direct cash flow estimate, by assuming that 35 percent of capital gains are on corporate stocks and dividing that share of the gain by the sum of corporate interest payments and corporate profits. That calculation yielded a very similar estimate of 1 percent.

To extrapolate the gain to the entire capital stock, the entire revenue effect before any realizations response was divided by capital income in the economy, assuming it is a quarter of net national product. This same share was used in the dynamic calculation model below. The result was a percentage change of .8 percent.

How did the Council obtain such a large number for the net cost of capital effect? The measure used by the Council was interpolated from an existing study of a different capital gains effect by Yolanda Henderson, Capital Gains Taxation and the Cost of Capital for Mature and Emerging Firms, presented at a Conference of the American Council for Capital Formation, October 13, 1989. Her calculations apparently did not account for the reduction in the capital gains effective tax rate due to the step up in basis when assets are held until death. This error was magnified because she used the new view of dividends which treats dividend taxes as irrelevant, rather than the traditional view employed in equation (2). (If the new view were used in equation (2) the cost of capital effect would be slightly larger at 1.4 percent; we do not employ the new view because of its counter- factual implications). These overstatements were then again magnified by extrapolating the calculation to the entire capital stock (where the new view of dividends would play no role).

B. The Model of Capital Accumulation

This section presents the simple model of capital accumulation which is used to trace the path of the capital stock, revenues and other variables when the capital gains tax change is introduced assuming a Cobb Douglas production function, a fixed labor supply, and myopic expectations. The basic equations are:

(3) Q(t) = AKtaL(1-a)

(4) K(t+1) - Kt = (st(Qt-dKt)(1-t) - nKt - Dt)/(1+n)

(5) St = b(Rt(1-u))E

(6) Rt = aQt/Kt - d

Equation (3) is the production function relating gross output (Q) at time t to the capital stock (K) as time t. Equation (4) is the equation of change in the capital stock. This change is equal to net savings less normal investment to keep the capital fixed less government borrowing, all divided by (1+n), where n is the growth rate in the economy. Net savings is s(Qt - dKt)(1 - t), where s is the savings rate out of income after depreciation and taxes, d is the depreciation rate, and t is the aggregate tax rate in the economy. The deficit, Dt is the actual revenue loss plus interest on accumulated debt. Equation (5) is a formula for the savings rate as a function of the after tax rate of return, where R is the marginal product of capital and u is the capital income tax rate. E is the elasticity of savings with respect to the net interest rate. The final equation, (6) is the first order condition from the production function in (3). To calibrate the model, we set a at .32 (equivalent to capital income 25 percent of net national product), n at .025, d at .03, the initial ratio of the capital stock to output at 3.5, the initial capital income tax at .3 and the wage income tax at .2. The reduction in capital income tax rate is that sufficient to cause the net cost of capital to fall by one percent, holding after tax return fixed. The infiation rate for measuring nominal levels of change is .04.

² The elasticity is the percentage change in realizations divided by the percentage change in tax rates.

³ See for example, Bailey, Martin J. Capital Gains and Income Taxation. In Harberger, Arnold C., and Martin J. Dailey, eds. The Taxation of Income from Capital. Washington, The Brookings Institution, 1969, p. 11-49, and Stiglitz, Joseph E. Some Aspects of the Taxation of Capital Gains. National Bureau of Economic Research. Working Paper No. 1094, March 1983.

⁴ One perhaps less important problem which exists for any cross section study is [trying] separating out the effects of income and price on behavior. The only source of differential price is the tax rate and in a progressive income tax system the tax rate is determined by the income level. If this relationship were linear we would not be able to separate these different effects, but because there are many income levels associated with a different tax level, it is possible to estimate such a relationship. However, we are still not be able to obtain correct estimates unless we know the functional form of the relationships to begin with, as pointed out by Feenberg, who examines this problem in the context of studies on charitable contributions. Feenberg's findings suggest that this bias is probably not very important. See Daniel Feenberg. "Are Tax Price Models Really Identified? The Case of Charitable Giving." National Tax Journal, Vol. 60, December 1987, pp. 629-634.

⁵ Slemrod, Joel, and William Shobe, "The Tax Elasticity of Capital Gains Realizations; Evidence From a Panel of Taxpayers." February, 1989.

⁶ Auerbach, Alan. "Capital Gains Taxation and Tax Reform." National Tax Journal, Vol. 62, September, 1989, pp. 391-401.

⁷ This point is made by one of the authors of the study. See Leonard Burman, "Why Capital Gains Tax Cuts (Probably) Don't Pay for Themselves," forthcoming in Tax Notes, April 2, 1990.

⁸ Endogeneity can be a problem with time series as well as cross section since an average marginal tax rate must be constructed, and that marginal tax rate can be influenced by gains. CBO used some techniques to account for this effect; in any case, this problem appears to be of minor concern. Another problem is called sample selection bias, which occurs because only those actually realizing gains are reflected in the data, but this problem also appears to be a minor one.

⁹ While a completely satisfactory model of capital gains realizations has not been developed, Kiefer has explored the time path of adjustment when trading occurs because of continually changing expectations about rates of return. These time paths can be quite complex with realizations rising, then falling, and then rising again. See Donald W. Kiefer, Lock-In Effect Within a Simple Model of Corporate Stock Trading.

¹⁰ In this calculation, the effect of depreciation is negligible because the expansion of capital is constrained by savings supply. Incorporating depreciation becomes much more important if savings were very elastic because the expansion of capital is then largely governed by the expansion in the investment demand.

¹¹ See Gravelle, Jane G. and Lawrence B. Lindsey. Capital Gains. Tax Notes, January 25, 1988, and Alan Auerbach, Capital Gains and Tax Reform, National Tax Journal, September 1989, pp. 391-401.