RANK REVERSALS AND THE TAX ELASTICITY OF CAPITAL GAINS REALIZATIONS**

Transcription

1 RANK REVERSALS AND THE TAX ELASTICITY OF CAPITAL GAINS REALIZATIONS** JOELSLEMROD* Introduction proportionate increase in capital gains realizations and calculating the change in OW responsive capital gains realiza- total capital gains by income group. Even H tions are to tax rates is a subject of an equi-proportionate increase in realizasome controversy. One piece of evidence tions for all taxpayers can generate data that has been cited recently in support of which apparently suggest a greater rea negative tax effect on realizations (the sponse of higher-income taxpayers. "lock-in" effect) is the differential re- This phenomenon is explained in Secsponse of taxpayers to the capital gains tion I of the paper. Section 2 describes a tax cuts that were instituted in Ta- simple empirical exercise that illustrates ble 1, reproduced from a study conducted how much of the apparent differential reby the Congressional Budget OTice (CBO, sponse in capital gain realizations be- 1986), summarizes thir, evidence. Be- tween 1980 and 1983 could be explained tween 1980 and 1983, the marginal tax by this phenomenorl Section 3 offers some rate on capital gains fell for nearly all in- concluding comnients and suggestions for come groups, but fell among the top 1 per- further research. cent of income earners by more than twice as much as any other group. In the same period, all groups' capital gains realiza- 1. Rank Reversals tions increased, but rose for the top 1 percent by a rate nearly twice as high as for The problem of rank reversals arises any other group. That capital gains re. because, comparing 1980 to 1983, the taxalizations increased most for the group Payerr, represented in any percentile group whose tax rate declined most has been are not identical Moreover, what percencited by CBO (1986, p. 56) and Lindsey tile income group a taxpayer belongs to (1988) as possible evidence for a lock-in depends on how much capital gains were effect. Lindsey states "if the change in the realized. Thus if a taxpayer realizes more distribution of capital gains resembles the capital gains, he or she is more likely to distributional change predicted by behav- appear in a higher income group than in ioral simulations of the response of cap- previous years. This can make it appear ital gains realizations to tax changes, then that the rate of increase of capital gains the tax changes are the likely cause of the is higher for liigher-income taxpayers than rise in capital gains" (p. 150). it actually is. This paper argues that a large fraction A simple example should make the point of the apparent differential response of clear. In the four-person economy illustaxpayers may be illusory. The pattern of trated in Table 2, all individuals doubled the data apparent in Table I can be gen- their capital gains between years 1 and 2. erated almost entirely by the phenome- However, because in the process individnon of "rank reversal," which is unre- ual 3 moved from the bottom half of the lated to any lock-in effect of taxation on income distribution to the top half,2 the realizations, and related to the fact that Percentage increase in capital gains of the taxpayers are assigned to income groups top 50 percent of the income distribution based on income measures that include is apparently 180 percent, while the persome or all of realized capital gains.' This centage increase of the lower 50 percent is demonstrated by simulating an equi- appears to be exactly zero! This apparent differential response obtains even tliough *The University of Michigan and National Bureau all individuals in fact had exactly the of Economic Reseamh. percentage change in capital gains. 503

2 504 NATIONAL TAX JOURNAL [Vol. XLII Table I Percentage Change in Capital Gains and Marginal Tax Rate, , By Expanded Adjusted Gross Income Group 1% 2-5% 6-25% 26-50% 51-95% TOTAL % Growth in Capital Gains Per Return % Change in Marginal Tax Rate Note: 1. Reproduced from Congressional Budget Office (1986), Table rv.3, p Expanded adjusted gross income is defmd in footnote The labelling of percentile groups is CBO'S. More precisely, the 2-5% percentile group includes taxpayers in the 1.0 to 5.0 percentiles, the 6-25% group refers to the 5.0 to 25.0 percentiles, etc. 2. Rank Reversals and Capital Gains realizations between 1980 and The data base is the 1980 Treasury Tax Model, This section presents a simple exercise which contains tax return information on to illustrate how important rank re- a stratified random sample of 171,508 tax versals can be in explaining the apparent retums for Accompanying each redifferential responsiveness of capital gains tum is a sample weight equal to the in- Table 2 ExuMle of Rank Reversal Y-Mi xw_z Non Capital Capital Total capital Total Percentage Change Individual Gains Income Gains Income Gains Income in Capital Gains 1 35,000 20,000 55,000 40,000 75, ,000 5,000 45,000 10,000 50,000 1()0 3 25,000 15,000 40,000 30,000 55, ,000 5,000 25,000 10,000 30, Capital Gains of top 1/2 of, [-AMIrs 25,000 (Individuals 1&2) 70,000 (Individuals 1&3) 180 Capital Gains of bottom 1/2 20,000 20,000 0 of income (Individuals 3&4) (Individuals 2&4) eaniers I I I

3 No. 41 CAPITALGAINS REALIZATIONS 505 verse of its sampling probability, i.e. how of taxpayers than before. Row 2 of Table many similar returns from the universe 3 shows, for each percentile group, the of taxpayers the sample return repre- percentage increase in capital gains per sents. Using this information the sample return that results from this simulation. data can provide reliable estimates of the CBO's results concerning the actual total taxpayer population. changes in capital gains for each income The first step of the exercise is toorder group are repeated in Row 1 for the sake taxpayers by their expanded adjusted gross of comparison with the results from the 3 income (EAGI), and calculate the amount simulations. of capital gain realized by income group. Next, for each taxpayer separately, the Even though by construction each taxpayer experienced exactly the same per realized capital gain is multiplied by centage increase in capital gains realiza , the factor of increase of aggregate tions, Row 2 of Table 3 makes it appear capital gains per return between 1980 and that there was a greater responsiveness 1983." Finally, the taxpayers are re-or- of higher-income taxpayers. The appardered according to their new expanded ent percentage increase in capital gains adjusted gross income, which reflects the for the top 1 percent is 80.1, significantly changed capital gain amounts. The per- higher than the overall average of 61.2 centile groups now include different sets percent. The percentage increase for the Table 3 Percentage Change in Capital Gains By Income Group, , Using Actual Data and Data from Simulation of Equi-Proportionate Increases in Capital Gains Percentile Group 1% 2-5% 6-25% 26-50% 51-95% TX)TAL % Actual Growth in Capital Gains Per Return, Ranked by EAGI* Results from Simulated EQui:pIQRgnionate Increase of 61.2% in CQiial Gains 2. Ranked by EAGI, no increase in other income 3. Ranked by AGI, no increase in other income 4. Ranked by EAGI, % proportional increase in other incorm 5. Ranked by AGI, 19.5% proportional increase in other income *Source: CBO (1986), Table rv.3.

4 506 NATIONAL TAX JOURNAL [Vol. XLH top 2 percent to 5 percent of income earn- capital gains would appear in the data as ers is also higher than the overall aver- an even greater differential response than age. The percentage increase of the low- the actual data show. This suggests that est three income groups appears to be the actual behavioral response may have much lower than the overall average, and been greater in the lower income classes as low as 5.4 percent for the per- than in the higher classes. cent group. When the income concept used to classify taxpayers is acbusted gross income 3. Conclusion and Directions for rather U= expanded adjusted gross in- ' Future Research come, the results are similar, as shown in Ms exercise establishes that the mes- Row 3 of Table 3. The most noticeable dif- sage conveyed by comparing the behavior ference between Rows 2 and 3 is that the of percentile groups across years may give apparent increase in realizations of the a very misleading picture of the actual lowest three income groups is much higher, differential behavioral response across so that the discrepancy between the high groups. It does not establish that no difand low -groups is not as large. This is be- ferential realization response occurred cause only 40 percent of long-term gains between 1980 and 1983; rather it atis included in adjusted gross income, while taches a strong caveat to using this sort 100 percent of these gains is included in of evidence to establish that fact. expanded adjusted income, so that the ex- Much more light could be shed on this tent of rank reversals will be less when issue by examining panel data about taxadjusted gross income is the classifier. payersrealization behavior over the The results reported so far may be biased 3 period. This would allow us to compare because they allow for no increase in the the same taxpayers' behavior over time components of income other than capital as the marginal tax rate changed, and gains, when in fact these components did avoid having to compare the behavior of increase between 1980 and Of different sets of taxpayers in different course, if all components of income in- years.5 Unfortunately, the one publicly creased at the same rate for everyone, available panel data set that spans this there would be no rank reversals at all. period-the Ernst & Young/University of Rows 4 and 5 present the results of re- Michigan panel-contains an insufficient peating the earlier exercises when all other number of upper-income taxpayers on components of income are increased by which to base reliable conclusions about 19.5 percent, the increase in the non-cap- the relative importance of rank reversals ital gains component of aggregate ad- and behavioral response for the taxpayer justed gross income between 1980 and population as a whole. Micro-econometric As expected, the rank reversal ef- analysis of this panel (Slemrod and Shobe, fect is dampened somewhat. However, the 1989) and a stratified panel data set used apparent differential responsiveness of internally by the U.S. Treasury Departhigh and low-income UxTayers is still very ment (Auten, Burman, and Randolph, evident. 1989) have generally concluded that there The pattern of apparent differential re- is a large rhort-run responsiveness of sponse presented in Rows 2-5 of Table 3 capital gains realizations to their taxais clearly not identical to what the data tion, although the long-rim responsivereveal, as shown in Row 1. In the simu- ness is more controversial. lation exercises, the percentage increase in the 2-5 percent group (and for the 6-25 percent group, when AGI is the clas- FOOTNOTES sifier) is much higher than the data show, and the simulated increase in the lowest **I am grateful to Paul McCarthy for able reseanh two groups is generally less than what the assistanceand to the National Bureau of Economic data show. What is striking is that the Researchfor fmancial support. 'The problem of rank reversal was noted by CBO simulated equi-proportionate change in (1986,p. 6) but not investigated further. CBO (1988,

5 No. 41 CAPITAL GAINS REALIZATIONS 507 pp ) attempted to correct its time-series exu- REFERENCES mates of the tax responsiveness of realizations by adjusted gross income claw for the bias due to rank reversals, and concluded that the bias was mnall. Auten, Gerald E., Leonard E. Burman and William 'Me example assumes that capital gains are fully C. Randolph, "Estimation and Interpretation of included in the income measure used to order tax- Capital Gains Realization Behavior- Evidence from parers' income. Panel Data." Department of the Treasury, Office of Expanded adjusted gross income is equal to ad- Tax Analysis Working Paper No. 67, May justed gross income plus the excluded portion of div- Congressional Budget Office, "Effects of the 1981 Tax idends, the excluded portion of long-term capital gains Act on the Distribution of Income and Taxes Paid," in excess of short-term capital losses, and deductions Staff Working Paper, Washington, D.C., August for contributions to IRA's and self-employment tirement plans. Congressional Budget OTice, How Capit4al Gains T= 4 The measure of capital gains used here, and in the Rates Affect Revenues: The Historical Evidence, CBO report, is net long-term gains in excess of short- Washington, D.C., March term losses plus short-term gains, for those returns Lindsey, Lawrence B., "Did ERTA Raise the Share of with net gains, plus capital gains distributions. Re- Taxes Paid by Upper Income Taxpayers? Will TRA86 turns that showed net losses were assigned a capital be a Repeat," in L. Summers, ed., T= Poli4ey and gains value of zero. the Economy, VoL 2, Cambridge, MA: The Mrr Press 'Such an analysis would still have to face some dif- and National Bureau of Economic Research, ficult issues related to how best to measure the in- Slemrod, Joel and William Shobe, "The Tax Elasticcome of taxpayers, such as to what extent to include ity of Capital Gains Realizations: Evidence from in income the capital gains realizations and whether Panel Data," The University of Michigan, Februto average several yean' annual income. ary 1989.

6 I j ri I-, I ' I, :,,:,