CRS ECONOMIST DISPUTES DRIVE TO LOWER CAPITAL GAINS RATE FOR DEFICIT REDUCTION

CRS REPORT FOR CONGRESS

An argument has been made that a cut in capital gains taxes would raise revenue and that such an option be considered in achieving deficit targets. This study examines the evidence presented to support this proposition. While these issues are highly uncertain, the assessment of the evidence suggests that such a change would be likely to lose rather than raise revenue, particularly in the long run.

                                   by

CONTENTS

THE BASIC THEORY OF CAPITAL GAINS REALIZATIONS

THE ECONOMETRIC STUDIES: A SURVEY

ASSESSMENT OF THE LINDSEY STUDY

The author would like to acknowledge valuable discussions with Gerald Auten, Donald Kiefer, Lawrence Lindsey, Andrew Lyon, Joseph Minarik, Eric Toder, and Dennis Zimmerman.

A PROPOSAL FOR RAISING REVENUE BY REDUCING CAPITAL GAINS TAXES?

Recent interest has developed in an argument that we could actually raise revenue in order to meet the deficit targets by cutting the capital gains tax rate. The argument for this proposal as a revenue raising option is based largely on a simulation study by Lawrence Lindsey. ¹ Lindsey used a number of previous studies to estimate the effects of the increase in capital gains tax rates in the Tax Reform Act of 1986. The simulations indicated that in four out of five cases, the increase will actually lose revenue, even in the long run, while in the remaining case long run revenues were virtually unchanged. Because long run responses are smaller than short run ones, the loss of revenue in the short run would be expected to be even greater.

These results would seem to imply that the weight of evidence suggests that cutting back the capital gains tax would raise revenue. This study reviews the theoretical and empirical evidence on this issue, and the particular simulations done by Lindsey.

The first section of the study discusses the basic theory of capital gains realizations and how they might be affected by taxes. This theory suggests that a mere change in timing of realizations is unlikely to change long run levels of realizations. The major source of increased realizations would arise from selling assets that would otherwise be held until death. Thus, if a revision in capital gains taxes is desired, one which would be virtually certain to raise revenue is taxation of capital gains at death. This section also explains why changes in realizations would be much more pronounced in the short run than the long run.

The second section reviews the econometric evidence and the general controversies which have surrounded this body of research. Many of these equations have been subject to substantial criticism, and the estimated effects have varied widely. The most important insight which arises from a general review of this literature is that the estimates of responses from the cross section studies tend to be larger than the estimates of responses from the time series studies. Theory would predict the opposite. There are some reasons to believe that the cross section estimates may be overstated because one cannot control properly for transitory (temporarily high or low) tax rates. Moreover, the tax rate in cross section studies is not entirely independent of taxpayer behavior, which raises a number of questions about the accuracy of these estimates. These reasons suggest that the cross section estimates that Lindsey relied on are Likely to be overstated.

The third section addresses the Lindsey study directly. The Lindsey study transformed the regression equations into a different functional form, so that strictly speaking, his simulations do not really represent the extrapolations from any of the studies. It is not, clear, however, how his simulations would differ from those done from the original functional form.

Lindsey does not report certain time series studies, which tend to have lower responses than those he simulates; nor does he acknowledge that several regression equations have not shown a statistically significant relationship.

The transformation in the equations, the exclusion of certain regressions, and the questionable reliability of the cross section studies for theoretical reasons suggest that any claim that the body of academic research supports revenue increases from lowering capital gains tax rates is incorrect. Indeed, while this area is one of uncertainty, the assessment in this study suggests that a cut in capital gains taxes would lose revenue in the long run, and possibly in the short run as well.

THE BASIC THEORY OF CAPITAL GAINS REALIZATIONS

Basic economic theory offers some crucial insights into evaluating the econometric evidence, as well as thinking in general about the consequences of capital gains tax changes on revenues. We begin by noting two important issues -- first the source of effects on long run realizations, and, secondly, the relationship between short run and long run effects.

Basically, there are two different types of selling activities which should be distinguished. Individuals may sell assets in order to consume the proceeds. Individuals may also sell assets in order to reinvest in other assets, either capital gains or other types of assets. In selling for reinvestment, individuals may either be selling those assets earlier than they would otherwise have or selling assets they would have otherwise held until death, when these assets are passed on to heirs. It is the selling motivated by reinvestment that is associated the "lock-in effect." The basic notion here is that there are other investments which earn a higher rate of return or whose characteristics are more attractive to the taxpayer, but the attractiveness must be great enough to offset the payment of the tax. If the tax rate is lowered, some investments which were not sufficiently desirable will now become desirable. For some investments, the tax acts as a barrier currently, but not necessarily indefinitely as new investment prospects appear over time. For others, the tax acts as a permanent barrier and individuals will simply hold on to these assets. A number of these theoretical issues are discussed in a study by Cook and O'Hare. ²

In the case of sales to consume, a mathematical model developed by Stiglitz indicates that if a tax causes individuals in the economy to delay selling without reinvestment, realizations are actually likely to RISE in the long run. ³ That result occurs because there is simply a timing shift. As long as the individual expects to recognize the gain at sore point, what is lost in realizations today is recovered, with additional appreciation, in the future. It seems, however, relatively unlikely that sales for the purpose of consumption are very responsive to tax changes in any case, since research on consumption functions suggests that they are relatively insensitive to the rate of return.

In the case of gains with reinvestment, models developed by Bailey and Stiglitz suggest that mere changing of the timing is unlikely to produce any major effect on long run realizations. ⁴ The issue becomes a bit more murky when gains are realized to reinvest in other assets that don't produce capital gains, since it depends on how those other assets are taxed.

In the case of gains which would otherwise be held until death, there would be both a short term and a long term realization. These gains would otherwise go entirely untaxed, since under current law the basis for figuring capital gains by heirs is the market value at the time of death.

This theoretical discussion indicates that the major source of long term increases in realizations is from the unlocking of capital gains held until death. Thus, in order for a tax reduction to actually increase long run revenues, the increase in realizations of gains which would otherwise be held until death multiplied by the tax applied to those gains must more than offset the reduction in the tax rate multiplied by preexisting realizations. Such a result is theoretically possible, since large amounts of assets are passed on at death; but its magnitude is a matter for empirical investigation. ⁵ At the same time, this theoretical observation suggests that an obvious revenue raising device is to impose an income tax on capital gains at death. Such a tax would effectively decrease the tax price of selling, causing realizations to increase permanently, and would also collect a tax on unrealized capital gains as well. A more modest approach would be to keep the original basis when the asset is passed on so that heirs would pay the capital gains tax when and if they sell it, again converting a tax forgiveness into a tax deferral.

The second major theoretical point is that short run increases in realizations should be considerably larger than long run gains. This effect occurs for the reasons mentioned above -- that is, a change in the timing of realizations will only temporarily increase these realizations. This effect also occurs because individuals in the face of a tax reduction have a large stock of assets where the tax rate at the higher level was a barrier to selling, but the lower tax rate removes this barrier. Once this portfolio has been adjusted, the future annual amount of realizations which are induced by the rate reduction will be much smaller. Thus, even for unlocking of capital gains held until death, the short run effects would be larger than the long run effects. Because of these theoretical considerations, Lindsey, in fact, argues that the revenue losses he estimates for the capital gains tax increases in the Tax Reform Act of 1986 are understatements of the short run revenue effects.

THE ECONOMETRIC STUDIES: A SURVEY

The econometric studies must be evaluated in the light of the general theory presented above. Initially, it is important to distinguish between two types of econometric studies -- cross section studies and time series studies. A cross section study observes taxpayers at the same time period who face different tax rates, to find a statistical relationship between tax rate and capital gains realizations, while controlling for other variables such as income. If the tax rates facing these individuals are their permanent tax rates, any such statistical relationships would indicate long run effects of realizations since these individuals are in a steady state. The time series studies look at how realizations change as tax rates change over time, again controlling for other variables such as income. Lindsey's study reports five estimates, four of which are basically cross section studies, along with his own which is both time series and cross section (including 18 years and five different income classes).

It is important to realize that the statistical studies of the change in realizations as a result of tax changes are all very recent studies, and there is still a considerable debate about the validity of these studies. Lindsey's paper presents these estimates without critique; yet it is crucial to evaluate these estimates in a critical fashion. Moreover, Lindsey's study relies primarily on cross section studies; the time series estimates should also be considered as part of the body of research.

In interpreting the results of these studies, an elasticity with an absolute value less than one with respect to the tax rate (all elasticities are negative) would indicate that small increases in tax rates would result in revenue gains, while an elasticity of greater than one would indicate that small increases in tax rates would lead to revenue losses. The elasticity is the ratio of the percentage change in realizations to the percentage change in the tax rate. For large discrete changes, values less than one can result in revenue gains with tax increases, depending on the functional form of the equations.

In addition, econometric studies are also evaluated by tests of statistical significance. These tests give one an idea of whether the results obtained are likely to be representative of true relationships or whether they are just random accidents. Any regression equation will force the data to fit to a line or curve; but observations may fit very closely on that curve or may be widely scattered. If they are widely scattered, the results are not statistically significant and suggest that the relationship measured is not a true relationship, so that we cannot say that we have a relationship. (In technical terms, we cannot reject the hypothesis that there is no relationship, in this case, between capital gains realizations and tax rates).

The first study Lindsey reports was a pioneering study done by Feldstein, Slemrod, and Yitzhaki (hereafter referred to as FSY) which examines gains on the sale of corporate stock. ⁶ This study reports a very large elasticity of -3.75. FSY pointed out, however, that their study did not deal with the problem of transitory tax rates, discussed below. Because their data set covered a single year, such an adjustment was not possible.

As one of the first studies done using cross section data from tax returns, the FSY results were subsequently subject to a number of criticisms. Later studies attempted to correct for these criticisms in various ways. The following general points should be made about this study. First, when a general sample was taken from tax returns, no statistically significant relationship appeared. It was only when the sample was restricted to individuals with very high incomes and assets that a strong relationship emerged. Secondly, the sample was a stratified sample, where more observations were taken from the higher income taxpayers in the sample. Thus, the sample was not a general sample for both of these reasons.

The Minarik study, which Lindsey also reports, was actually an attempt to correct the FSY study, including dealing with the stratification issue by weighting the regression, and his estimate produced a markedly lower coefficient of -0.44. ⁷ (Minarik's sample was also somewhat different from the FSY sample.) In their reply, FSY acknowledged that their results might be overstated for a general across the board tax rate change such as occurred in the Tax Reform Act of 1986. ⁸ They disagreed, however, with Minarik's weighting technique.

Even before the Minarik study was published, Auten and Clotfelter published a study which tried to deal explicitly with the transitory tax issue by using panel data which traced taxpayers over several years. ⁹ The problem is that individuals who face a temporary dip in their tax rate would find a period of low tax rates an opportune time to realize capital gains when the tax rate would be lower. Thus, much of the relationship between low tax rates and capital gains realizations, for individuals in the same income class, would reflect a temporary timing effect rather than a permanent relationship. Using the estimates to measure response to a permanent tax change would, therefore, overstate the response. Because their data covered taxpayers over more than one year, they constructed a permanent and a transitory tax rate, with the permanent rate reflecting an average over three years. In addition, the Auten/Clotfelter study considers all capital gains, not just those on corporate stock.

Auten and Clotfelter report a number of different specifications; Lindsey reports one which is statistically significant and yields an elasticity of -0.55. Auten and Clotfelter report another regression which is statistically significant and has a much higher elasticity, although the dependent variable in that case includes losses which should not be subject to these lock-in effects. Two other specifications, however, produce relationships which are not statistically significant and which would suggest that we cannot reject the hypothesis of a zero elasticity. ¹⁰

The Auten/Clotfelter results do point strongly to the notion that one needs to separate the transitory and permanent tax rate issues, because the effects of transitory tax rates were much larger than those of permanent tax rates. In their equation which was statistically significant, producing a -0.55 elasticity for the permanent tax rate, the result was a -1.24 elasticity for the temporary tax rate. All of their results stress this relationship, which was not taken into account in the FSY study, or the Minarik study (and could not be because of the data limitations).

The fourth and final cross section study reported by Lindsey was done by the Treasury Department, also using panel data (with a different panel). ¹¹ This study reported elasticities ranging from -2.2 to -1.2, all above 1 in absolute value (implying that a small increase in tax rate would cause revenues to fall). Lindsey uses the one they chose for their own simulation, at a value of -1.29. That is also the only equation for which tests demonstrating statistical significance are reported. A weighting technique from the stratified sample is used, but it is different from the Minarik studies, and is claimed to be more appropriate.

The Treasury study also provided estimates of the elasticities for different types of capital gains. Corporate stock had the highest elasticity -- -2.07. Real estate had a lower elasticity of -0.71 and other assets an elasticity of -0.43. If these estimates are correct, then the FSY and Minarik studies would have elasticities which are overstated, since both these studies are confined to corporate stock.

These studies represent all of the true cross section studies, and perhaps all that can be said of them is that the estimates vary widely depending on specification, including some cases where the results are not statistically significant. The body of this research might, in sum, be said not to yield very clear results. If we exclude the FSY study, where elasticities are probably overstated due to the restrictive sample, the transitory tax rate influence, and the restriction to corporate stock, the range of results is from zero (the null hypothesis when the results are not statistically significant) to slightly over -2.

Lindsey's study is a pooled cross section, time series study, where taxpayers are aggregated into broad income classes. ¹² Lindsey actually reports a number of regressions in his study. The one he uses here has an elasticity of -0.8. This estimate included two tax rate variables -- the current tax rate and the change from the previous year. This specification was an attempt to separate the tax rates into a permanent and transitory element. Lindsey's study is quite different from the other cross section studies because of the aggregation of taxpayers. It would seem likely, however, that most of variation in tax rates is across income classes rather than across time. In any case, it is very difficult to interpret this type of regression.

Two time series studies have also been done. The Treasury time series study, reported in the same volume as the cross section study mentioned above, results in an elasticity of -0.77. The Congressional Budget Office reports a time series estimate with a considerably lower elasticity of -0.25. The results of the Congressional Budget Office equation are not statistically significant, although some preliminary results of a new equation, with a slightly higher elasticity, are significant. ¹³ The Treasury time series regression differs in the number of years of coverage and in some other aspects from the Congressional Budget Office regression.

There is one aspect of this research which is quite striking. The cross section studies have tended to produce estimates that are higher than the time series estimates. Yet, the theory presented earlier suggests that the initial higher realizations will be much reduced as time passes. Thus, the relationship between these two sets of estimates is the reverse of what theory suggests. ¹⁴ Indeed, if the time series elasticities are in the range of -0.25 to -0.77 (the Treasury and the Congressional Budge Office studies), then we would expect the cross section estimates to be quite low, implying that a gain in revenue with a tax rate reduction is highly unlikely. ¹⁵

A possible explanation of this discrepancy has to do with certain basic shortcomings in the cross section data which are taken from tax files. The first is whether the issue of transitory versus permanent tax rate has been adequately dealt with, or can be adequately dealt with. A taxpayer would be expected to realize unusually large gains during a year when his tax rate is particularly low. The technique used in the two studies which dealt with the transitory tax issue took the permanent tax rate as the average of the current year and the preceding two years; the transitory tax variable was the difference between the observed tax rate and this estimated permanent tax rate. If, however, the taxpayer is in a temporarily low or high tax rate for one year the permanent tax rate will reflect one third of this difference. Moreover, the individual may be in a temporarily low tax rate for more than one year. Thus, there is a very real possibility that some of the relationship observed is a response to a temporarily low tax rate rather than a permanent tax rate.

Secondly, there is a general problem with the fact that the tax rate is not really independent of taxpayer behavior. A regression equation includes a dependent variable -- in this case capital gains -- and a number of independent variables such as income, price, and so forth. The price in this case is the tax rate. Because capital gains realizations are associated with income, these studies control for income. The effect of tax rate variables, therefore, is derived from observations of taxpayers with the same income but different tax rates. At high income levels, most of the variation in tax rates for individuals in the same income class are due to differences in investment behavior since factors such as marital status, family size, and so forth become relatively unimportant. But the individual controls his investment decisions and therefore the tax rate is determined by him. As a result we expect that individuals with different tax rates in the same income brackets have different basic tastes in investment. It is easy, therefore, to speculate that a relationship between capital gains and low tax rates may occur because individuals who choose more active, risky portfolios might simultaneously be actively trading in stocks and, say, involved in tax shelter investments as well. Thus, low tax rates could be associated with capital gains realizations because of a taste in portfolio, rather than from a true lock-in effect.

The influence of this effect depends on how income is specified and whether these effects mismeasure income or tax rate. When adjusted gross income is the income measure, portfolio choices which produce deductions which lower adjusted gross income will distort the measure of economic income rather than tax rate, while portfolio choices which result in deductions from adjusted gross income will produce a lower tax rate while measuring income correctly. The more serious problem with this portfolio effect would occur with a broader measure of income such as that used in the Treasury study. Indeed, when one excludes the FSY regression, the Treasury study produced the highest elasticities.

Indeed, different paths of causality could occur with many other decisions. Extraordinary medical expenses could simultaneously create a need for cash and the realization of an unusually large gains from disposing of assets, while also lowering tax rates.

These illustrations suggest that the cross section studies may not be measuring the realization of capital gains due to unlocking but may be picking up other relationships between low tax rates and realizations. If this analysis is correct then the cross section studies which Lindsey reported in his study are too high and may suggest revenue gains from cutting capital gains taxes which would not materialize. Both the observed relationships between the time series and cross section estimates, as well as the potential problems of transitory tax rates and endogeneity of the independent variable in cross section studies lends some support to this notion. Thus, these studies may be regarded as suggestive, but not conclusive. ¹⁶

A final issue raises questions not so much about the studies themselves, but about their applicability to the current situation. There is substantial underreporting of capital gains realizations. Individuals with high tax rates may have more of an incentive to underreport gains than individuals with low tax rates. ¹⁷ Such tax effects would, nevertheless, have to be taken into account in assessing the effects on revenues. Recent changes in the tax law have, however, required reporting of capital gains transactions to the Internal Revenue Service, which should increase taxpayer compliance and also reduce any variability in compliance due to changes in tax rates. To the extent that this compliance issue is important, estimates derived from historical data would overstate the effect on reported realizations under current regulations.

ASSESSMENT OF LINDSEY'S STUDY

With this background in mind, we can now look specifically at the Lindsey study. Lindsey's estimates are based on the four cross section studies discussed above, plus his own regression. He then performed a micro simulation of the results, by estimating for each taxpayer the change in realizations and the resulting tax collection. The simulation was done for changes in the Tax Reform Act of 1986. Of the five studies, four produced revenue losses from the rate increase and one produced a slight revenue gain (the Minarik equation).

Before assessing his study in the context of the issues discussed above, however, one important aspect of Lindsey's analysis should be explained. Each regression was originally estimated in a particular functional form. The FSY and Minarik regressions were linear (that is, the relationship between gains and tax rates is assumed to fall on a straight line). The other equations were curvi- linear. In performing the simulations, Lindsey took the average taxpayer, and transformed the relationship so that while the average taxpayer was at the same point on the line or curve, the other taxpayers were at different point (sic). That is, he transformed the original regression function into a different functional form. Thus, his simulations do not reflect, strictly speaking, a true simulation of the original regression equations.

One reason Lindsey gave for making this transformation was to convert the FSY equation, where the high elasticities would yield extreme results for simulations of such a larger percentage change in tax, into a more reasonable result. But this transformation was also applied to the other equations which would yield more reasonable results. In addition, some type of transformation would be necessary because not enough information is reported in the some of the studies to perform a micro-simulation. Because we do not have the capability to perform the micro simulations, it is not possible to determine what the effects of this transformation was in general.

More importantly, however, one cannot simply draw conclusions from averaging together the results of these simulations. The FSY equation has been subject to considerable critique and the authors have themselves acknowledged that their estimate is likely to be overstated. ¹⁸ Moreover, there are several instances of regressions which did not produce a statistically significant response, and this information is just as relevant (i.e., pointing towards no response) as those equations which did produce statistically significant results.

The most important point, however, is that there are reasons to believe that the cross section estimates are, by their basic nature, likely to be overstated, as discussed above. The cross sections estimates tend to be higher than the time series estimates, whereas theory would predict the opposite to occur. Since some of the time series equations would predict a revenue loss, this observation suggests that cutting the capital gains tax rate seems more likely to lose rather than raise revenue in the long run.

² Cook, Eric W. and John F. O'Hare. Issues Relating to Taxation of Capital Gains. Forthcoming, National Tax Journal.

³ Stiglitz, Joseph E. Some Aspects of the Taxation of Capital Gains. National Bureau of Economic Research. Working Paper No. 1094. March 1983.

⁴ Bailey, Martin J. Capital Gains and Income Taxation. In Harberger, Arnold C., and Martin J. Bailey, eds. The Taxation of Income from Capital. Washington, The Brookings Institution, 1969. p. 11-49.

⁵ Data supplied by Lawrence Lindsey to the author suggest that during the period 1977-1984 only 16 percent of changes in asset value were realized as gains. Of course, some of these assets were in owner occupied housing which would not typically be recognized in any case; even excluding real estate, however, gains recognized are only 30 percent of changes in asset value. Thus, there is clearly a large continual pool of unrealized gains in the economy. This observation only suggests that significant increases are possible, not that they are plausible. At any one time there would be only a portion of these gains where the individual would be better off selling even in the absence of taxes, because of transactions costs or because no better opportunity presents itself. Cook and O'Hare, in Issues Relating to Taxation of Capital Gains, present some regression equations which suggest the effect of taxes on lock-in of assets held at death is small; this research has not yet been subject to critical review, however.

⁶ Feldstein, M., J. Slemrod, and S. Yitzhaki. The Effects of Taxation on the Selling of Corporate Stock and the Realization of Capital Gains. Quarterly Journal of Economics, June 1980. p. 777-91.

⁷ Minarik, J. 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.

⁸ Feldstein, M., J. Slemrod, and S. Yitzhaki. Reply. Quarterly Journal of Economics, February 1984. p. 111-120.

⁹ Auten, G., and C. Clotfelter. Permanent vs. Transitory Effects and the Realization of Capital Gains. Quarterly Journal of Economics, November 1982. p. 613-632.

¹⁰ The authors also report a variety of other specifications which show varying elasticities and varying degrees of significance.

¹¹ U.S. Department of Treasury. Office of Tax Analysis. Report to the Congress on the Capital Gains Tax Reductions of 1978. September 1985.

¹² Lindsey, Lawrence B. Capital Gains: Rates, Realizations, and Revenues. National Bureau of Economic Research, Inc. Working Paper 1893. April 1986.

¹³ The regression includes a current and lagged tax rate; the current tax rate coefficient is significant at the 90 percent but not the 95 percent confidence level. U.S. Congressional Budget Office. Effects of the 1981 Tax Act on the Distribution of Income and Taxes Paid. Staff Working Paper. August 1986.

¹⁴ One possibility is that there is an adjustment lag; this does not appear likely, however, as both the Treasury and Congressional Budget Office used both current and lagged tax rates and find the effect of the current tax rate much more pronounced. These results strongly suggest that taxpayers can react very quickly to tax changes.

¹⁵ The elasticities reported here for the Treasury and Congressional Budget office studies are the sum of the coefficients for the current and lagged tax rates. The coefficients for the current tax rates are higher, -1.5 and -0.6 respectively.

¹⁶ Of course, time series estimates are not without problems as well. It is impossible to capture all of the events over time which might affect realizations. Moreover, because of the limited number of observations, it is not possible to specify a theoretically appropriate lag structure.

¹⁷ This issue was studied by Poterba, James M., Tax Evasion and Capital Gains Taxation. National Bureau of Economic Research, Working Paper 2119, January 1987. Poterba argued that a significant component of the relationship between capital gains and tax rates reflected differential tax compliance.

¹⁸ Feldstein, et al., Reply.