NBER WORKING PAPER SERIES

PROGRESSIVITY OF CAPITAL
GAINS TAXATION WITH
OPTIMAL PORTFOLIO SELECTION

Michael Haliassos

Andrew B. Lyon

Working Paper No. 4253

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
January 1993

We wish to thank Carol Bertaut for data from the Survey of Consumer
Finances and helpful discussions, and Bill Gale, Jim Poterba, Joel Slemrod,
and conference participants for helpful suggestions. We owe special thanks to
John O'Hare of the Joint Committee on Taxation for providing us with
tabulated data from the Sales of Capital Assets studies. This paper is part of
NBER's reseaith program in Public Economics. Any opinions expressed are
those of the authors and not those of the National Bureau of Economic Research.
 NBER Working Paper #4253
January 1993

PROGRESSIVITY OF CAPITAL
GAINS TAXATION WITH
OPTIMAL PORTFOLIO SELECTION

ABSTRACT

We provide new data on capital gains realizations using a five-year

stratified panel of taxpayers covering 1985-1989. We find, as earlier studies
have, that capital gains realizations are very concentrated among the highest
income groups. We use these data and data from the Federal Reserve Board
Survey of Consumer Finances to draw inferences from a simulation model of
the effects on progressivity and efficiency of alternative tax treatment of capital

gains.
Tax payments alone are not an accurate indication of the burden of a
tax. Taxes generally create costs beyond the dollar value collected by causing
persons to change their behavior to avoid the tax. Risk is also affected by the
tax system. Beneficial risk-sharing characteristics of the tax system are
frequently overlooked when examining the treatment of capital gains, We find
that reforms comprising reductions in the capital gains tax rate offset by
increases in the tax rate on other investment income are efficiency reducing.
Surprisingly, we find that for taxpayers for whom loss limits are not binding
a switch to accrual taxation is also efficiency reducing. For those taxpayers for
whom loss limits are potentially binding, we find that large efficiency gains can
be achieved by increasing the amount of capital losses that may be deducted
against ordinary income. These results are partly attributable to changes in
risk-sharing encompassed in these reforms.
Michael Haliassos Andrew B. Lyon
Department of Economics Department of Economics
University of Maryland University of Maryland
College Park, MD 20742 College Park, MD 20742
and NBER
 1. Introduction

The debate over capital gains taxes has served as a focal point for
the larger debate over the desired progressivity of the tax system. Some
argue that a reduction in the capital gains tax rate predominantly benefits
upper-income taxpayers. Others argue that taxpayers in all income groups
receive capital gains and that once-in-a-lifetime gains artificially distort the
concentration of gains by income class. In addition to the concern over
progressivity, there are debates on the efficiency consequences of capital
gains taxes: the degree to which they hinder or promote saving and
investment and their effects on portfolio behavior.
In this paper we present new data from the Internal Revenue
Service on capital gains realizations between 1985 and 1989 and data from
the Federal Reserve Board Survey of Consumer Finances on stockholding.
These data show a high concentration of stockholding and capital gains
realizations in the highest income groups.
Next, we draw upon these data to examine distributional effects
and portfolio responses to changes in the tax treatment of capital gains.
A simple analysis of tax burden focussing only on tax payments is not an
accurate indication of the burden of a tax. Taxes generally create costs
beyond the dollar value collected by causing persons to change their
behavior to avoid the tax. Economists have long noted the "lock-in' effect
as one of these additional costs. Taxes also reduce the variance in after-
tax returns. For risk-averse investors this reduction in variance is a
benefit.
We examine capital gains taxes using a model based solely on
utility-maximizing behavior. Econometric estimates of realization behavior
in response to changes in tax rates are frequently criticized for relying on
artificial or ad hoc specifications of the realization response. Kiefer
(1990) presents a simulation model which demonstrates how the realization
response can be modeled endogenously. This model, however, is
atemporal: risk-neutral agents maximize the expected value of terminal
wealth. Intertemporal models of risk-averse agents are modeled by
Slemrod (1983), Cook and O'Hare (1992), and Galper, Lucke, and Toder
(1988). This latter model is modified to estimate changes in portfolios and
trading behavior by Hendershott, Toder, and Won (1991, 1992). This
model has a considerable amount of disaggregation, including 147 separate
households, five financial assets, owner-occupied housing, and consumer
durables. A corporate sector is modeled and its financial behavior with
respect to debt-equity ratios and dividends versus retained earnings is
endogenous. A cost of working with a model of this size is that portfolio
selection and capital gains realizations decisions have to be simplified.1
We have opted instead to use a smaller, stylized model, but allow
asset demands and realizations to be derived as solutions to the
households' maximization of intertemporal utility of consumption. For
this purpose, we incorporate alternative systems of capital gains taxation
into a consumption-based model of portfolio selection. We consider a
three-period model with two financial assets: a risk-free bond and risky
stock. Our model is in the spirit of Auerbach's (1992) analysis based on
a similar model. We differ from his analysis by explicitly allowing for a
dividend component to stock returns, which analyses of historical data

1
For example, exponential utility functions are used, because they allow
derivation of explicit solutions. This utility function, however, displays
constant absolute risk aversion (i.e., no change in the amount of risky asset
holdings for different wealth levels). In addition, changes in taxation induce
realizations in an ad hoc way, and its dependence on other factors such as
incomes and risk aversion is ignored.

2
have found to be sizable. We also differ by providing a capital gains asset
which is risky in both periods that it may be held, rather than only in the
second. This seemingly slight difference is important, because the
possibility of a capital loss affects initial asset holdings and realizations.
A shortcoming of our model is that the constant relative risk aversion
parameter, though popular for analysis of portfolio behavior, does not
distinguish between risk aversion and elasticity of intertemporal
substitution.2 We hope to incorporate other forms of the utility function
where this feature is not present in future research.
We find that reforms comprising reductions in the capital gains tax
rate offset by increases in the tax rate on other investment income are
efficiency reducing. Surprisingly, we find that for taxpayers for whom
loss limits are not binding a move to accrual taxation is also efficiency
reducing. These results are partly attributable to a reduction in risk-
sharing encompassed in these reforms. We find that for those taxpayers
for whom loss limits are potentially binding, large efficiency gains can be
achieved by increasing the amount of capital losses that may be deducted
against ordinary income.
In the next section we present data on capital gains realizations
from tax return data. In section 3 we outline our model. Section 4
presents calculations of portfolio holdings and realizations under the
benchmark tax system. In section 5, we examine the effects of alternative

2 The same parameter in this utility function affects both the willingness to
substitute consumption across states of the world (within a single time period)
and substitute consumption over time periods. Hall (1987) has noted the
difficulty in separating these parameters in common utility functions. Epstein
and Zin (1991) investigate a class of utility functions for which the effects of
these parameters on behavior are better separated.

3
capital gains tax cuts. Section 6 examines accrual taxation. The final
section presents conclusions.

2. Distribution of Capital Gains Income

This section presents data on the distribution of capital gains
income using the Sales of Capital Assets panel data set for 1985-1989.
Presently, the panel contains tax information on 11,452 taxpayers for each
of the years between 1985 and 1989 (57,260 tax returns).3 These data
are stratified to oversample high-income taxpayers based on their income
in 1985. Taxpayers in the highest income decile represent approximately
70 percent of the sample.
Two recent studies have examined capital gains realizations using
a panel of tax return information in the public domain. Feenberg and
Summers (1990) use a nonstratified panel of returns covering 1979 to
1984. Slemrod (1992) uses a more recent version of the nonstratified
panel which includes 1985 and 1986.
These studies find that capital gains realizations in any given year
are highly concentrated among taxpayers in the highest income groups.
Gains tend to be much more heavily concentrated among these high-
income taxpayers than other forms of income.
These two studies also examine one standard criticism of studies
based on cross-sectional analyses of capital gains: if capital gains are
realized only infrequently, a taxpayer of moderate income who has a large

The data set is not available in a public-use format. Data presented in

this section are based on computations by John O'Hare and Cathy Koch of the
Joint Committee on Taxation. The authors greatly appreciate the cooperation
of these individuals and the Joint Committee on Taxation in performing these
calculations.

4
one-time realization of capital gains will be classified as "rich' in a study
focussing on only a single year's income and realizations. With a single
year's data one could address this problem only by classifying taxpayers
by an income concept that does not include capital gains. This approach,
however, would understate the concentration of gains among taxpayers
who have large recurring realizations.
The availability of panel data allow these studies to follow a
taxpayer for a number of years to compare cumulative realizations to the
taxpayer's average income over this extended period. Slemrod (1992),
using the seven-year panel ending in 1985, finds that a one-year study
does overstate the concentration of capital gains income accruing to high-
income taxpayers. However, even using cumulative capital gains over the
seven-year period for each taxpayer and classifying each taxpayer by his
average income over this seven-year period, Slemrod finds capital gains
still are heavily concentrated among the highest income groups. Slemrod
finds that the highest income percentile received 52 percent of all capital
gains using the single-year approach, while using income measured over
the seven-year period they accounted for 44 percent of all capital gains
earned. In contrast, the highest income percentile accounts for 8 percent
of income from all sources, regardless of whether measured using the
single-year or seven-year measure.
The findings that we report in this section are based on the most
recent data available. Our ability to use a stratified random sample of
returns reduces sample variance since previous studies have shown capital

Slemrod omits 1986 from the study in case of anomalous results due to
the large quantity of capital gains realized in advance of the new provisions
contained in the 1986 Tax Reform Act.

5
gains to be concentrated among high-income taxpayers. Our analysis
generally confirms earlier findings of Feenberg and Summers (1990) and
Slemrod (1992). In fact, we find even a slightly greater concentration of
gains among the highest income decile than in the earlier studies. This
may be due to our use of a stratified sample of returns or reflect a change
in concentration over time.
We first examine the extent to which gains realizations tend to be
a recurring event for taxpayers during the 1985-1989 period. Sales of
stock, other securities, and partnerships comprise about one-half of capital
gains. Except for the market crash of October 1987, this period was
generally one of rising stock prices. By the close of 1989, the Standard
& Poor's 500 composite index had increased by more than 100 percent
relative to 1984 and by over 200 percent since June 1982.
Table 1 presents data on the frequency with which net gains were
realized by taxpayers between 1985 and 1989. Capital gains for this
purpose include net capital gains reported on Schedule D, the excluded
portion of long-term gains (for 1985 and 1986), and other distributions
listed directly on Form 1040.6 Over three-fourths of all taxpayers had no
realized net gains over this period. Of those who reported a gain in at

An important legislative event in this period was the enactment in

September 1986 of the Tax Reform Act of 1986. This Act eliminated the 60
percent long-term capital gains exclusion effective January 1, 1987, and may
have given some taxpayers an incentive to accelerate the realization of gains in
1986. Because realizations in 1986 may be unusual, all findings reported in
the text were also examined for the four-year period excluding 1986. The
exclusion of this year does not affect the annual rate of realizations across
taxpayers or alter the general distribution of gains.
6
Taxpayers who receive capital gain distributions from mutual funds and
who engage in no other transactions may report these gains directly on Form
1040.

6
least one year, nearly 60 percent also reported a net gain in at least one
other year. Thus, to some extent gains realizations appear to be a
recurring event. Only 2.5 percent of taxpayers reported positive net gains
in each of the five years, yet these taxpayers accounted for 42 percent of
all net gains. For these taxpayers, capital gains realizations are not only
frequent, but constitute a quantitatively important recurring source of
income.
Table 2 examines the distribution of capital gains income across
income groups. Capital gains are defined here to include both net gains
and net losses . Income is defined to equal AGI plus (i) excluded gains
and dividends (for 1985 and 1986) and (ii) statutory adjustments.8 The
panel data are used to examine the discrepancy between annual measures
of the distribution of capital gains income and those based on longer
periods of observation. In column 2 the data in the panel are treated as if
they represented five different single-year observations. In column 3, the
data are properly treated as a panel and the five-year distribution of capital
gains income is examined relative to five-year income of the taxpayer.
Columns 4 and 5 of Table 2 show the distribution of the income measure
across income percentiles using both the single-year average and the five-
year measure, respectively.

'
Reported net capital losses in any year are limited to $3,000. Excess
losses may be carried forward and reported in subsequent years by the
taxpayer. Certain losses on the stock of qualifying small businesses which
would otherwise be classified as capital losses are allowed as ordinary losses
up to $100,000. These losses are reported as other income and not on
Schedule D.
8
These adjustments consist mainly of IRA and Keogh contributions and
alimony paid. For 1985 and 1986, they additionally include moving expenses,
employee business expenses, and the two-earner deduction.

7
 Examining the highest income percentile, one can observe how the
single-year measure of the distribution of capital gains income overstates
the concentration of gains in this group. Column 2 shows that on average,
a single-year study over this period would have estimated that the highest
income percentile received 66 percent of all capital gains reported in a
given year. When the same taxpayer is followed for each of the five
years, column 3 shows that taxpayers in the highest five-year income
percentile received 58 percent of all gains reported over the five-year
period. Column 5 shows that this highest income percentile received 13
percent of income from all sources over the five-year period.9
Although we would prefer to use a longer time period to
approximate permanent income, the high concentration of gains in this
income percentile and the other groups comprising the top decile
(accounting for 82.5 percent of all gains over the five-year period) leads
one to conclude as Slemrod (1992) does: capital gains income appears to
be "largely a phenomenon of the upper-income classes."10
Table 3 presents an alternative examination of the distribution of
capital gains income across income groups. In this table, income is

Excluding 1986 data from the panel, the highest income percentile
accounts for 55.5 percent of all gains.
10 One notable exception to the general tendency for gains to be
concentrated among the higher income groups is the concentration of gains in
the lowest decile (less than $8,000 of income in 1989). This group of
taxpayers receives 3.7 percent of five-year capital gains, while accounting for
less than 1 percent of total income. Below, we note the concentration of gains
in this income group is substantially eliminated when individuals with other tax
losses are removed from this category. Even without this adjustment most
taxpayers in this decile are unaffected by the tax treatment of capital gains since
they either have negative income or the standard deduction and personal
exemption eliminate all of their tax liability.

8
defined to exclude all capital gains or losses. Thus, the realization of a
large capital gain will not cause a taxpayer to be classified in a higher
income group. While this understates the true income of taxpayers with
recurring gains, it may be a closer measure of the permanent income of
taxpayers with infrequent gains than the measure used in Table 2. Because
we exclude capital gains from this income measure, we also increase
income by the amount of losses claimed by the taxpayer on business or
partnership activities. Losses claimed from these activities may be
associated with gains received in other years. For example, an active
participant in real estate may claim operating losses on some properties
while simultaneously receiving capital gains on other properties (or on the
same property in a future year). Eliminating these losses gives a better
understanding of gains realizations among taxpayers with low permanent
income rather than among those taxpayers with large business losses.
Using this measure of income, column 3 of Table 3 shows that the
highest income percentile receives 49 percent of five-year capital gains and
the highest decile cumulatively receives 75.6 percent of gains, down from
82.5 percent in Table 2.11 Income groups between the 60th and 80th
percentiles show a modestly higher amount of capital gains. Removal of
losses from the definition of income substantially reduces the amount of
gains reported among those in the lowest income decile.
Finally, in Table 4 we present data on the realization of capital
gains by age group. The Internal Revenue Service matched taxpayer
social security numbers for returns included in this panel with social

Excluding 1986 data from the panel, the highest income percentile
continues to receive 49 percent of all gains, and the highest decile receives
72.4 percent of gains.

9
security records to determine the age of the taxpayer as of January 1 in the
year covered by the return.12 The data show that most capital gains are
received by taxpayers who are under age 60, but that taxpayers over age
60 account for a sizable proportion of gains. Taxpayers age 60 and older
in 1985 accounted for 16.7 percent of all tax returns (column 2), but 26.7
percent of all taxpayers claiming gains or losses (column 3) in the five-
year period. These older age groups also accounted for 39.4 percent of
all gains claimed over the five-year period (column 5). Despite the
heavier concentration of gains among older taxpayers, less than half of the
taxpayers in these older age groups realized capital gains in at least one of
the five years examined.
Taken together, Tables 2 and 3 both show that while all income
groups receive some capital gains, the largest part of gains is received by
those in the highest income decile of the population. We also find that
while it is not true that most capital gains are received by the elderly, or
that most elderly receive capital gains, some elderly account for a sizable
portion of gains relative to their numbers in the taxpaying population.
Although this section has focussed primarily on the distribution of capital
gains across income groups, the finding that only one-fourth of all
taxpayers received any capital gains suggests that examination of the
heterogeneity of taxpayers within income groups would be an interesting
area for future research.

3. The Model
The results examined in the previous section indicate that high

12 Public-use data sets provide only data reported on the return and
therefore do not include this variable.

10
income taxpayers realize most capital gains. As a result, most capital
gains taxes are paid by high income taxpayers. The distributional effects
of capital gains taxes are incompletely described, however, by focussing
only on tax payments. In examining the burden of a tax, one would like
to examine both the burden borne directly through the payment of a tax
and the burden borne indirectly by altering one's behavior to avoid the tax.
These indirect costs are frequently ignored in distributional analyses, but
they are potentially significant in magnitude. As an extreme example,
consider the effects of two alternative taxes on the purchase of a good: a
small tax and a tax so large that none of the good is purchased. The large
tax in this case collects no revenue. Of course, no one would claim that
the large tax is less burdensome than the small tax. A distributional
comparison of tax payments under the two alternative taxes would be
uninformative. Because the indirect costs of a tax system are as
fundamental to any system of taxation as the direct tax payments, we
believe that distributional analyses should incorporate these costs.
By modeling the decision to save and allocate savings among
various assets we can begin to measure these indirect costs of taxation.
In this section we describe our model used to examine both the direct and
indirect costs of capital gains taxation. We compute the total burden of
the capital gains tax by finding the lump sum tax which would give an
individual the same level of utility as the existing capital gains tax. If the
capital gains tax creates indirect net costs to the individual, the
hypothetical lump sum tax will collect more revenues than the existing tax.
In this case, tax payments understate the burden of the capital gains tax.
If the capital gains tax creates indirect net benefits, for example, by
reducing the riskiness of an asset, then the lump sum tax will collect

11
smaller revenues than the existing capital gains tax.
An advantage of modeling capital gains behavior in a utility-
maximizing framework is that we can also examine the distributional
effects of alternative tax schemes, after taxpayers have modified their
behavior in response to the new system of taxation. This allows us to
examine, for example, the claim that a lower rate of tax can result in more
tax revenues and make individuals better off. Such a claim is dependent
on the alternative scheme imposing smaller indirect costs than the current
tax system.
Our model economy consists of four types of agents: two groups
of expected-utility maximizers, the firm that employs them and issues
stocks, and the government which issues flat money and riskless bonds.
Households live for three periods (each period representing twenty years),
and they have a choice between four different assets: currency,
government bonds, private loans, and stocks. Currency bears zero
nominal interest, it is dominated in return by other riskless assets, and it
is therefore optimally held in an amount just sufficient to cover purchases
of the consumption good. Since we do not intend to study the monetary
transmission mechanism, money is assumed to be neutral in our "cash-in-
advancet' economy. Its only function is to define the nominal price level.
We consider a simple version, where the inflation rate is taken to be
known with certainty and to occur at an annualized rate of 3 percent.
Government bonds and private loans are perfect substitutes in
agents' portfolios. Perfect substitutability implies a common riskless
interest rate. We thus abstract from liquidity constraints in the form of a
wedge between the borrowing rates available to the government and to
households. This simplifies the portfolio selection problem by allowing

12
us to consider one "riskiess asset and a real riskiess rate set at a
historically based level. We assume an annual real riskiess rate of 3.13%,
equal to Siegel's (1992) estimate of the average real rate on bonds in the
u_S. over the period 1800-1990. Our simplifying assumption that changes
in tax rates do not affect the pre-tax real rate is tantamount to considering
a small open economy with a real rate tied to the world interest rate
through capital mobility. In view of the absence of inflation uncertainty,
constancy of the real rate on bonds does not require the assumption that
debt is indexed.
The historical average of annual pre-tax real rates of return on
stocks over the period 1800-1990 was calculated by Siegel as 7.77% per
annum. Our procedure for calculating appropriate 20-year stock returns
for use in our numerical solutions is the following. First, we compute
equiprobable "high" and "low' annual stock returns, which match both
the average annual return computed by Siegel and its standard deviation
(18.36). Since each realization can occur with probability 0.5, we
compute the expected value and standard deviation of the return over a
twenty-year period from this binomial process. Finally, we choose a
"high" and a "low" 20-year realization which match these two moments.
The 20-year cumulative return is thus set to roughly 7.39 in the good state
and -0.46 in the bad state.
The return on stocks consists of capital gains and dividends. We
model each of these components separately, given their potentially
different tax rates and the ability of the taxpayer to defer capital gains
taxes but not dividend taxes. Schwert (1990) has found that dividend
yields (i.e., the ratios of end-of-period dividend to beginning-of-period
stock price) represent about half the total average stock return in his 1802-

13
1987 sample. Their standard deviation is about 1/20 as large as that of
capital gains returns, and there is no evidence of a secular increase in
dividend yields since 1800. We have chosen dividend yields for the
"high" and "low" states, such that their expected value is half the total
expected pretax return on equity. Although it would be natural to assume
zero variance of dividend yields, this causes a technical problem: in a two-
state world, the total return in the low state is so low relative to the
average dividend yield, that it would require implausibly large capital
losses, implying negative stock prices in the low second-period state. To
avoid this, we allowed for essentially minimum variability in dividend
yields consistent with positive stock price in the low state. The resulting
coefficient of variation over each 20-year period was .84.
Households choose portfolios of these assets to maximize expected
lifetime utility of consumption, assumed to be time separable. Utility in
each period exhibits constant relative risk aversion (CRRA), with degree
of relative risk aversion denoted by A.13 Although the direction of
behavioral responses to a tax change turns out to be independent of the
risk aversion parameter, its magnitude frequently does depend on A. The
range of A values considered by Mehra and Prescott (1985) as plausible
for representative-agent models was between 2 and 10, and this is the
range we consider. In general, how "plausible" a particular risk parameter
is depends on the size (coefficient of variation) of the gamble considered
[Kandel and Stambaugh (1991)]. Evidence from the Survey of Consumer
Finances suggests strongly that the proportion of highly risk-averse

13 Utility in a given state of nature may be written as:

A), A> 0, where c represents consumption in that state. Utility in each period
is time separable. We set the discount factor to .818 for each 20-year period.

14
individuals is much larger among low-income households than among
high-income groups, based on their responses to direct questions about
attitudes towards risk.14 Column 2 of Table 5 shows that the percentage
of households who responded that they are not willing to undertake any
financial risk is virtually a monotonically declining function of income.
It ranges from 67% for very low incomes to 6% of households in the 99th
percentile. Although we present a number of model predictions for the
entire range of degrees of risk aversion, we will often contrast results for
high-income households at low A's to those for low-income households
with A's at the high end of the spectrum.
After-tax labor income is taken to be exogenous in the model. In
our simulations, high-income households are assumed to earn (net of tax)
twice as much as low-income households in each period of their life. An
element of life-cycle behavior is built into the model by assuming that
after-tax labor income in the third period drops to one-fourth of what it
was in the first two periods. This generates an incentive to save for the
final period of life. In the third (and final) period of life, all assets are

14 Indeed, Shaw (1989) has suggested that high risk aversion discourages
individuals from investing in risky firm-specific human capital and reaching
high income levels. It is interesting to note that limited stockholding among
low-income households could also be observed if low- and high-income agents
had the same degree of risk aversion, but those with low incomes also faced
more income uncertainty. Kimball (1989) has derived conditions under which
increased background (income) risk reduces exposure to a risky asset in an
atemporal model, even though it is uncorrelated with stock returns. He termed
this property "standard risk aversion". This is a possibility we cannot explore
in the present paper, since we do not incorporate nondiversifiable individual
risk.

15
liquidated and consumed.15
The model incorporates taxation at two rates, tjb and t8, for each
group of households i = 1,2. Rate tib is applied to interest earnings on
loans to the government and to other households, as well as to dividends
received on stocks.16 The tax rate on capital gains is t8, and it is applied
at realization. The U.S. tax code imposes a limit of $3,000 on the net
capital losses which can be declared for tax purposes. We incorporate a
loss limit equal to 5% of the after-tax labor income received by high-
income households.17 This constraint turns out to be binding only for
high-income households with degrees of risk aversion of four or less. The
Sales of Capital Assets panel data indicate that just 4.5 percent of
taxpayers with gains or losses had $3,000 or more in net capital losses in
1985 and 1986. This number increased to 12.9% in 1987, 11.9% in 1988,
and 14.9% in 1989.18 Of course, what matters from a utility-maximizing
perspective is not whether the limits bind for all states of nature, but
whether they are potentially binding constraints for at least a single state

We assume all asset sales occur while the individual is alive. Thus, we
do not consider bequest motives and the possible advantages to deferring
realizations until death to allow heirs a step-up of basis.

16 We do not consider a differential tax rate between long-term and short-

term gains. Constantinides (1984) suggests trading strategies involving
realization of gains immediately upon their eligibility for long-term treatment.
Lyon (1990) fmds that this strategy is successful only for high variance stocks
and offers only slight benefits over a deferral strategy.

17 Since each period in the model represents 20 years, this percentage was
chosen to approximate the effects of the $3,000 limit applied on an annual
basis.

18 Poterba (1987) finds that in 1982 11.2% of taxpayers with gains or

losses had more than $3,000 in capital losses.

16
of nature.
In our simulations of the benchmark case, tax rates for high-
income households are equal to 30% on both capital gains and interest
income, while those for low-income households are set at 15%. We
consider several policy experiments involving reductions in t1 (and
endogenously determined increases in tlb), which preserve the expected
utility of the corresponding household group at the benchmark level. We
also consider a switch to accrual taxation. Each of these policy changes
is assumed to occur in the first period before any saving and portfolio
decisions have been made. The excess burden in the benchmark case is
compared to that under the different policy experiments. All computations
take into account optimal portfolio responses to the tax rates facing the
household.
The policy experiments examined either preserve the expected
utility of each class of agents or the expected present value of tax
collections on the basis of information available at the beginning of the
first period. In effect, we are considering cases where decreases in
investor risk resulting from higher tax rates are absorbed by the
government. The government's ability to reduce risk has received
19
considerable attention in other research dealing with risk sharing.
A frequently adopted alternative assumption is to focus on policies
which preserve the amount of real tax collections in each future time
period and state of the world.20 The motivation for considering such

19 Examples in the context of income taxation include Eaton and Rosen

(1980), Gordon and Varian (1988), and Varian (1980).

20 For example, this assumption is used in Slemrod (1983) and Galper,

Lucke, and Toder (1988).

17
policies is that they leave the government budget constraint unaffected,
though this presumption is not strictly accurate to the extent that policy
changes influence interest rates on government debt.21 We intend to
investigate the consequences of this and of a range of alternative
assumptions about government finance in future research. What is
particularly interesting for assessing effects of capital gains taxation on risk
taking is that such policies are revenue neutral but not portfolio neutral.
Specifically, the alternative assumption used by others supposes
that the government has access to lump sum taxes or transfers, which are
imposed in an amount necessary to preserve tax revenues after returns on
risky assets are realized in the future. As a result, households are faced
both with risky asset returns and with risky lump sum taxes and transfers.
Because lump sum taxes and transfers are by definition independent of the
amount of risky investment undertaken by the household, the government
imposes risk even on those who would not have chosen to bear any asset
return risk. Tax rate changes result in reallocation of private sector risk
between these two sources. The rational response to this risk reallocation
is to alter portfolios in order to hedge against unwanted risk.22
Conditions under which our conclusions would be modified by alternative
modeling assumptions remain to be examined. The present paper shows

21 The assumption ensures that the primary budget deficit in each period
and state of the world is the same across policy regimes. However, a change
in interest rates resulting from the tax policy experiment still affects the total
budget deficit and the time path of government monetary and nonmonetary
debt.
22 This may not be possible for households subject to binding short sales
constraints. For example, a household which does not desire any stockholding
risk would find it optimal to hedge by adopting a short position in stocks, since
lump sum taxes are expected to be high when stock returns are low.

18
that such modifications would not arise from the cut in capital gains tax
rates per Se, but from the assumption that the government maintains a
given level of tax revenue risk by redistributing risk among households.
We make no attempt to predict the short term effects of changes
in tax regime. Unanticipated changes in tax rates during an investor's
lifetime have contemporaneous lump-sum tax or subsidy effects which
depend on the accumulated capital gains or losses up to that point. It is
then optimal for investors to respond to the new tax regime by rebalancing
their portfolios in directions suggested by our analysis. The magnitude of
these portfolio changes is sensitive to the amount of accrued gains at the
time of the legislative change. In this paper, we focus on the permanent
effects of such tax changes by comparing regimes in which there are no
unanticipated changes in tax rates throughout the investment horizon of
each household.

4. The Benchmark Tax System

Under the benchmark system, capital gains are taxed at the same
rate as dividend and interest income, but at realization. High-income
households face a tax rate of 30%, which is twice that of their low-income
counterparts. Since stocks offer higher expected returns than riskless
assets and -- starting from a portfolio with no stocks -- have zero
covariance with consumption, it is always optimal for expected utility
maximizers to hold some stocks.23 The extent of stockholding, however,

23
Despite their idiosyncratic risk, stocks have zero covariance with a
nonstockholder's consumption, which is the relevant risk characteristic for
portfolio behavior. This attitude of expected utility maximizers towards
divisible risky projects was first noted by Arrow (1970), termed by Segal and
Spivak (1990) "second order risk aversion", and explored in the context of
stockholding behavior by Haliassos and Bertaut (1992) and Bertaut (1992).

19
does depend both on incomes and on risk aversion.
In their first period of life, both high- and low-income households
hold stocks and bonds in relative amounts that depend crucially on risk
aversion. At low levels of the risk aversion parameter A, it is optimal to
engage in arbitrage by borrowing at the low (riskless) rate to invest in
stocks. At slightly higher risk aversion levels (around 3 in our
simulations), it is optimal for both types of assets to be held in positive
amounts, with optimal stockholding still dominating holdings of riskless
assets. At risk aversion levels around 5, the covariance of stock returns
with consumption becomes an important consideration, the equity premium
notwithstanding; investors characterized by these or higher levels of risk
aversion want to hold more riskless assets in their portfolios than stocks.
The levels of risk aversion separating each of these three regimes
are slightly lower for the low-income group. This is a consequence of the
asymptotic behavior of marginal utility as consumption approaches zero.
Households with CRRA are disproportionately interested in marginal
utility at low states, and this concern limits their exposure to stockholding
risk. As shown in columns 5 and 6 of Table 5, data from the SCF show
that households in the 95th percentile of income or above hold on average
more stocks in their portfolios than riskiess assets; the opposite is true for
all other income groups examined. Thus, low risk aversion among high-
income households is not only corroborated by their responses to
attitudinal questions, but also by their portfolio behavior.
In the second period of life, portfolio behavior is state dependent.
In our simulations, the "good" state involves capital gains for stockholders,
while the "bad" state involves capital losses. In the bad state, it is optimal
for both groups to realize their losses (up to the allowable limit) in order

20
to receive an immediate reduction in tax liability provided by the capital
loss, and then to purchase anew all the stocks they want to hold to the
third period.24 In the good state, unconstrained expected utility
maximizers find it optimal never to realize capital gains but to engage in
short sales of stocks instead, in order to reduce their net stock holdings
and smooth their consumption over time. Such a transaction results in the
same net stock position as an outright sale, but defers tax liability. We
limit the extent by which taxpayers may defer taxes through short sales by
imposing quantity constraints on such sales. For the results presented in
this paper, short sales are ruled out completely. As a result, households
are encouraged to realize some capital gains in the second period of their
life even in a "good" state. We regard this restriction as bringing the
model closer to actual trading behavior.
The proportion of capital gains realized in the second period of
life is fairly similar across income groups for any given risk aversion
parameter, but is heavily dependent on that parameter. Households whose
A is at the low end of the spectrum realize slightly more than 65% of their
gains (Figure 1). The proportion increases at a decreasing rate, reaching
85% for A= 10. This, in conjunction with the SCF self-reported data on
risk aversion, implies that the proportion of capital gains deferred: on
average by high-income households should substantially exceed the

24 Technically, such a wash-sale" is disallowed under the U.S. tax law.

If more than thirty days elapse between buy and sell dates, the tax loss is
permitted. Because trading periods in our model are to represent periods of 20
years, we do not believe that inclusion of this constraint would alter our
results. Stiglitz (1983) discusses strategies of avoiding this and various other
legal restrictions pertaining to capital gains taxes.

21
corresponding average proportion for low-income households.25 In our
solutions, the proportion of deferrals by high income households with
A= 3 is about 50 percent larger than for low-income households with
A= 10. Combined with the much larger stockholding by high-income, less
risk-averse households, this implies a very substantial difference in the size
of unrealized capital gains across the two income groups.
In computing the burden of the benchmark system, we compare
it to one of lump sum taxation of (first-period) income. The equivalent
variation is the lump sum tax on first-period income which results in the
same expected utility as the benchmark system of proportional taxation.
Alternatively, it is the amount that households faced initially with no taxes
would be willing to pay the government to prevent it from imposing
proportional taxation (and the associated loss limits). This measure may
be the best yardstick to use in comparing the burden of alternative tax
instruments. Naturally, this measure is positive, since taxation reduces the
households' "feasible set" of consumption opportunities. Excess burden
is a comparison of the equivalent variation to the present value of tax
revenues that are collected from the proportional tax system.26 An
interesting finding in the context of the expected utility model is that tax
revenues far outweigh the amount households would be willing to pay to

25 1989 Survey of Consumer Finances includes questions allowing one

to calculate directly unrealized gains by household and to test this implication
of the model. Unfortunately, a properly weighted data set is not yet publicly
available.
26 The present value of government tax revenues is computed at the rate
relevant for government borrowing, namely the riskless rate on government
bonds. In the absence of borrowing constraints for individuals, the excess
burden is independent of the period in which the equivalent variation is
computed.

22
avoid proportional taxation, so that excess burden is negative. In other
words, actual capital gains, dividend, and interest tax payments
collectively overstate the burden of these taxes relative to lump sum
taxation. Further, excess burden is more negative for those with low risk
aversion, even when expressed as a fraction of tax revenues from that
group.
The negative excess burden of the proportional tax system arises
primarily from two related factors: (i) the desirable effects capital gains
taxation has on stockholding risk (even in the presence of loss limits) and
(ii) the portfolio response of households to the introduction of proportional
taxation. While it is true that capital gains taxation lowers the expected
return on stocks, it also reduces its variance, by taxing realizations in the
good state and subsidizing losses in the bad one.27 As explained below,
we assume in this model that the government can absorb the risk
associated with a risky flow of tax revenues. The finding that excess
burden is most negative for households with a low risk aversion parameter
and high incomes arises primarily from the fact that both high incomes and
low risk aversion are associated with more substantial stockholding.
In our model, we find capital gains taxation generally encourages
stockholding relative to lump sum taxes which provide the same expected
utility.28 Households make full use of tax loss opportunities in the bad

27 This utility increasing feature of proportional taxation was first noted by

Domar and Musgrave (1944). Atkinson and Stiglitz (1980) provide a more
recent exposition in an atemporal model.

28 The one exception is for high-income taxpayers with A= 2 (for whom

the loss limit binds). Replacement of proportional taxation with lump sum
taxation increases stockholding by 2.5% in the first period. Excess burden,
however, is still negative for this group.

23
state by holding more stocks to begin with and selling as many as the loss
limit allows them to in order to reap the immediate reduction in tax
liability. They then repurchase a (larger than otherwise) amount of stocks
to hold to the third period. Recalling that households' concern with bad
states is disproportionate, one can understand why the amount they should
be willing to pay to eliminate (or reduce) the distorting proportional
taxation is less than the tax revenues collected from them.
The increased stockholding associated with proportional taxes
yields substantial tax revenues to the government. Interestingly, the
increase in taxes collected under proportional taxation does not arise
simply from higher tax collections in the last period of life: even the
amount of capital gains realizations in the second period is higher because
of the higher overall stockholding. Movement to the lump sum tax system
would cause tax liabilities to fall by 50% for high-income taxpayers (A =3)
and by 23% for low-income taxpayers (A= 3). These changes in tax
liability are noted in the first row of Table 6. Of course, the true burden
of the proportional tax system is identical to that of the lump sum tax
system since equal utilities are attained. Again, this illustrates the
shortcoming of focussing only on tax payments to assess the distributional
effects of taxation. Because of the risk-sharing element of the tax system,
a simple analysis of the burden of capital gains taxes based solely on the
amount of taxes collected particularly overstates the utility costof taxation
for higher income taxpayers.
The importance of risk-sharing considerations can also be
demonstrated by comparing our benchmark system to one without loss
limits. High-income taxpayers subject to a 30% tax rate on capital gains
and a loss limit can achieve the same utility with a higher capital gains tax

24
rate if the loss limit is removed. At A= 3, high-income taxpayers would
be indifferent between the benchmark system with a loss limit or a
counterfactual situation in which the loss limit was eliminated and the
capital gains tax rate was increased to 40%. As shown in the second row
of Table 6, tax revenues from these taxpayers would increase by over 25%
at this higher rate. This indicates that there is a range of feasible tax rates
for which both utility and revenue can be increased by eliminating loss
limits (or increasing the amount of losses that may be deducted) •29 In
a distributional analysis, of course, these higher tax payments by high-
income investors would not be an indication of an increased tax burden.
In the next section we examine a policy experiment involving a
reduction in capital gains tax rates and in section 6 we consider a switch
to accrual taxation.

5. Effects of a Reduction in Capital Gains Tax Rates

In this section, we examine the effects of a simultaneous reduction
in capital gains tax rates and an increase in tax rates on dividend and
interest income which preserves the expected utility each income group has
under the benchmark system. These simulations also allow us to draw
implications for the utility effects of alternative revenue-neutral tax
schemes. We assume these tax changes take place at the beginning of the
first period, before any portfolio decisions have been made. For the main
body of our analysis, we consider a 20% reduction in capital gains tax

29 In a model with multiple capital gains assets (especially negatively

correlated ones) this result might not hold. If two negatively correlated assets
existed, an investor could always sell the asset that declined in value and defer
gain on the asset that appreciated. The portfolio could be partially rebalanced
by increasing investment in the asset that was sold rather than by selling shares
of the appreciated asset.

25
rates (reducing the rate to 24% for high-income households and to 12%
for low-income households). The increase in tib required to maintain the
expected utility of income group i differs across income groups and it
depends on risk aversion. We carry out sensitivity analysis by considering
the entire range of risk aversion parameters from A=2 to A= 10.
For the high-income group, the utility preserving tibranges from
31.6% for low risk aversion (A=2) to 30.2% for highly risk-averse
households (A = 10), given a 20% cut in capital gains tax rates. This is to
be compared to a rate tib of 30% assumed under the benchmark system
without differential taxation. For low-income households, the utility-
preserving t2b is slightly above 15% -- rising to just 15.5% for households
with A=2. For both groups, the utility-preserving tib is inversely related
to risk aversion, primarily because holdings of the riskless asset are
smaller (often negative) among the less risk-averse.
We find that this differential taxation entails an increase in excess
burden relative to the benchmark, for both income levels, for the entire
range of risk aversion parameters, and for varying sizes of capital gains
tax cuts. In Figure 2, we express this increase in excess burden as a
proportion of the present value of tax revenues collected from each group
in the new equilibrium. In the region where loss limits are not binding
(A>4), this measure is uniformly higher for high-income households than
for low-income households; and it is inversely related to the degree of risk
aversion for any given income level. For the cases we focus on, namely
high-income households with A =3 and low-income households with
A= 10, the changes in excess burden relative to tax revenues are 6.0% and
3.1 % respectively. Tax revenues fall for these groups by 5.6% and 3.0%
as shown in the third row of Table 6.

26
 Because the cut in capital gains tax rates induces some undesirable
effects on the risk properties of capital gains assets, the tax rate on
alternative sources of income, tib, cannot increase substantially if it is to
preserve the same expected utility of each group. Although holdings of
riskless assets by the least risk-averse are also increased, the increase in
tib is insufficient to make up for the loss in capital gains tax revenues.
To illustrate the importance of portfolio adjustments for our
findings, consider how they affect total capital gains realizations. Total
realizations depend both on the level of stockholding and the proportion
of gains realized. We examine the effects of the capital gains rate
reduction on both components. For taxpayers for whom loss limits are not
initially binding, a cut in capital gains taxes induces increased bondholding
and discourages stockholding. In the region where the loss limit is
binding, only for the least risk-averse (A 2 and A =3) does stockholding
increase. Even here, the increase in stockholding is less than 0.5%.
As expected, at the lower capital gains tax rates the proportion
of capital gains realized by each group in the good' state of the second
period of life increases. However, this increase is relatively small. In our
benchmark simulation, low-income households (with A= 10) realize 85.3%
of their capital gains in the second period. If capital gains tax rates are
reduced by 20% of their benchmark level, this proportion increases, as
expected, but by less than 0.5%. The proportion realized by high-income
households is only slightly more sensitive. But even for these high-income
taxpayers (with A =3), 75.7% of capital gains are realized under the
benchmark system in the second period and this amount increases by only
2.7% when capital gains taxes are reduced.
As a result, for taxpayers for whom the loss limits are not

27
binding, total realizations decline with the cut in capital gains taxes. For
those taxpayers for whom the loss limits are binding, total realizations
increase, but by a far smaller percentage than the reduction in the tax rate.
In either case, tax revenues fall substantially compared to the benchmark
system.
In view of the evidence of an inverse relationship between incomes
and risk aversion, our findings suggest that the increase in excess burden
resulting from a utility-preserving shift to differential taxation will be
larger for high-income households than for those with low incomes.
Portfolio size turns out to matter significantly for the change in tax
revenues and hence for excess burden. The implication is that if the
government wants to cut capital gains taxes without affecting welfare, it
will suffer a large reduction in tax revenues. If, on the other hand, it
decides to raise tib sufficiently to preserve tax revenues, expected utility
drops and both income groups are worse off than under the current
system.
While the findings we discuss primarily relate to permanent effects
of tax changes, our model is consistent with a short-run increase in capital
gains realizations following a capital gains tax cut. To see this, consider
households who have accumulated capital gains and are now at the
beginning of the second period of life. Their optimal portfolios at the
current capital gains tax rate would dictate realizing a fraction of capital
gains in the second period. If it is announced that the applicable tax rate
has been reduced, this is a welcome bonus to these households. On the
one hand, they experience a lump sum tax bonus on accumulated gains
which they were already planning to realize. On the other, the now lower
tax rate induces a portfolio response which makes them better off from

28
that point on. Because this portfolio response generally implies lower
stockholding than at the higher tax rate, the tax cut will actually generate
larger capital gains realizations in the second period than what households
had planned for at the beginning of their lifetime.
This should reconcile our findings with the observation that
households with accumulated capital gains tend to favor a reduction in
gains tax rates, and with the common perception that such a reduction
should encourage realizations in the short run. What appears questionable
is the notion that a pennanent move to lower capital gains tax rates will
sufficiently promote stockholding and realizations to provide larger tax
revenues over the longer run.

6. Accrual Taxation of Capital Gains

Using our model, we calculate the change in excess burden that
results if capital gains were taxed on an accrual basis rather than at
realization. We assume that the accrual tax system has no limit on the
amount of losses that may be claimed against ordinary income.30 As in
our previous experiments, the policy change holds utility constant. The
resulting accrual tax rate thus differs conceptually from the traditional
measure which holds the present value of tax payments constant assuming
unchanged realizations and portfolios. For taxpayers for whom loss limits
are not binding, we find that a move to accrual taxation increases excess
burden. In contrast, if loss limits are binding constraints, we find that a
move to accrual taxation together with the elimination of loss limits can
reduce excess burden. This latter reform is dominated, however, by the

30If an accrual tax system were enacted, retention of loss limits would
appear to be less justified since a taxpayer could no longer claim losses on
some assets while deferring gains on other assets.

29
retention of taxation at realization but the elimination of loss limitations
considered in section 4.
It may at first appear surprising that we find cases where the
movement from realization taxation to accrual taxation generates excess
burdens. Public finance economists have long emphasized the excess
burden created from the "lock-in" effect on capital gains. As our results
presented earlier show, the lock-in effect exists in our model, too. It must
be remembered, however, that the lock-in only occurs when stocks have
appreciated in value. When stocks decline in value, shareholders can
claim these losses immediately in the absence of loss limitations. In our
model, these taxpayers then repurchase stock after claiming their losses.
For taxpayers not constrained by the loss limits, the accrual tax
rate required to make them as well off as under the benchmark system
must be lower than the rate under realization taxation. Allowing for
behavioral response to the accrual tax system, taxpayers become better off
from the elimination of the lock-in effect, but they become worse off by
the increase in the variance of returns caused by the lower tax rate. In
theory, either effect can dominate. In our simulations, we find that the
accrual tax rate necessary to hold utility constant must be sufficiently low
that tax revenues fall relative to the benchmark case.31 For taxpayers
for whom the loss limits are binding, the switch to accrual taxation
together with the removal of loss limits can generate greater risk-sharing
in the bad state than under the benchmark case. These taxpayers benefit

31 Our results differ from those of Auerbach (1992) who finds the move
to accrual taxation generates efficiency gains. Our results differ, in part,
because stock retums in the first period of Auerbach's model are certain.
Therefore, the reduction in the accrual tax rate has no effect on variances in
retums earned in this period.

30
both from greater risk-sharing advantage and from the elimination of the
lock-in effect.
The loss limitation in our model is binding only for high-income
taxpayers with risk aversion coefficients of four and less. For taxpayers
with the lowest risk aversion coefficients, desired stockholding is large and
the inability to deduct all losses in the event the bad state occurs results in
a large loss in utility. These taxpayers are willing to accept a very high
accrual tax rate in exchange for removal of the loss limitations. For the
least risk-averse (A =2), an accrual tax rate of 50 percent leaves them the
same utility as in the benchmark case. At higher rates of risk aversion,
stockholding is smaller so that the inability to deduct losses is less costly.
At A=4, the accrual tax rate falls to 27.4 percent. In each case here, the
accrual tax rate results in higher tax revenues than in the benchmark so
that excess burden is reduced. As shown in the fourth row of Table 6, tax
payments for high-income taxpayers (with A =3) are 16% greater than in
the benchmark.
Figure 3 shows the change in excess burden that results from the
shift to accrual taxation. The change in excess burden is expressed as a
fraction of tax payments made by each group in the new equilibrium. At
low levels of risk aversion and high income the welfare gain from
switching to accrual taxation appears quite large, approaching 50 percent
of tax revenues for the high income group at A =2. These gains appear
to be entirely from the elimination of the loss limit. As noted above, the
switch to accrual taxation is dominated by the elimination of loss limits
and the retention of taxation at realization.
For taxpayers for whom the loss limit is not binding, the increase
in excess burden is declining with risk aversion. The change in excess

31
burden ranges from 10 percent of revenues for low-income taxpayers at
A = 2 to less than 2 percent of tax revenues for low income taxpayers at
A=10.

7. Conclusions
In this paper we have examined new data on capital gains
realizations by taxpayers in the 1985-1989 period. These data show a high
concentration of capital gains among taxpayers in the highest income
groups. More than half of all capital gains are realized by the richest one
percent of the population. As a result, reductions in capital gains tax rates
would reduce tax payments the most for these taxpayers.
We present a stylized model of savings and portfolio behavior.
We caution the reader that this model makes a number of simplifying
assumptions, but these simplifications allow us to explore in full detail the
changes in behavior that occur in response to permanent changes in the
taxation of capital gains. Our model suggests that reductions in capital
gains tax rates will lead in many cases to a reduction in stockholding.
This occurs partly because of the increase in risk which results as capital
gains tax rates are lowered. For these taxpayers, even though the lock-in
effect is reduced slightly, the increase in the proportion of gains realized
is not sufficient to offset the decline in stockholding so that total
realizations decline. Only for taxpayers currently constrained by loss
limits do we find an increase in stockholding when capital gains tax rates
are reduced. The resulting increase in realizations, however, is not
sufficient to make up for the much larger reduction in tax rates, so that
total revenues decline. We find that a revenue-neutral reduction in capital
gains tax rates offset by increases in taxes on risk-free investment income
results in a decline in taxpayer utility. In future research we seek to

32
examine the sensitivity of our results to alternative assumptions about the
extent to which the government absorbs fluctuations in tax revenues.
We find that the elimination of loss limits offers a potential
welfare gain. The removal of loss limits increases the risk-sharing
characteristics of the tax system. With greater risk-sharing, taxpayers in
our model respond by holding greater amounts of stock. The removal of
loss limits causes the government to collect less tax revenues (in
expectation) on stocks that would have been held anyway, but additional
tax revenues are collected by the increase in stockholding. Taxpayers are
made better off by the elimination of loss limits so that they would be
willing to accept slightly higher tax rates on capital gains in return. As a
result, net tax revenues can increase and still leave these taxpayers better
off.
This example of an efficiency enhancing reform and our other
examples of efficiency decreasing tax changes illustrate the trouble with
conducting traditional distributional analyses based on the tax payments
collected from a household. Table 6, which summarizes the effects of the
various policy changes, indicates substantial variation in tax liabilities for
each income group. Yet in each experiment, utility of each group is held
unchanged. Traditional distributional analysis, which focusses only on tax
revenues, would give quite different impressions of the distributional
effects of these reforms. As models of taxpayer behavior improve, better
distributional analyses of taxes should include measures of excess burden
(whether positive or negative) when evaluating the true costs of a tax.
For the case of capital gains taxation considered in this paper, we
find that dollar amounts of capital gains taxes paid overstate the burden of
these taxes relative to lump sum taxation. Our results suggest that

33
distributional analyses of the overall progressivity of the tax system based
on tax payments probably overstate progressivity, since capital gains tax
payments are concentrated in the highest income brackets. Our results
further suggest that reductions in the tax liability of capital gains taxes are
likely to benefit taxpayers by less than equal size reductions in other taxes.

34
 References

Arrow, Kenneth. Essays in the Theory of Risk Bearing. Chicago:

Markham (1970).

Atkinson, Anthony B. and Joseph E. Stiglitz. Lectures on Public

Economics. New York: McGraw-Hill (1980).

Auerbach, Alan J. "On the Design and Reform of Capital Gains

Taxation," American Economic Review, vol. 82 (1992): 263-267.

Bertaut, Carol. "Who Holds Stocks in the U.S.? An Empirical

Investigation," mimeo (1992).

Cook, Eric W. and John F. O'Hare. "Capital Gains Redux," National Tax
Journal, vol. 45 (1992):53-76.

Constantinides, George M. "Optimal Stock Trading with Personal Taxes:

Implications for Prices and the Abnormal January Returns,"
Journal of Financial Economics, vol. 13 (1984):65-89.

Domar, Evsey D. and Richard A. Musgrave. "Proportional Income

Taxation and Risk Sharing," Quarterly Journal of Economics,
vol. 58 (1944):388-422.

Eaton, Jonathan and Harvey S. Rosen. "Labor Supply, Uncertainty, and

Efficient Taxation," Journal of Public Economics, vol. 14
(1980):365-374.

Epstein, Larry G. and Stanley E. Zin. "Substitution, Risk Aversion, and

the Temporal Behavior of Consumption and Asset Returns: An
Empirical Analysis," Journal of Political Economy, vol. 99
(1991) :263-286.

Feenberg, Daniel and Lawrence Summers. "Who Benefits from Capital

Gains Tax Reductions?" in Lawrence H. Summers, ed., Tax
Policy and The Economy, vol. 4 (1990): 1-24, Cambridge: MIT
Press.

Galper, Harvey, Robert Lucke, and Eric Toder. "A General Equilibrium
Analysis of Tax Reform" in Henry Aaron, Harvey Galper, Joseph

35
 Pechman, eds., Uneaxy Compromise. Washington, D.C.:
Brookings Institution (1988): 59-108.

Gordon, Roger H. and Hal R. Varian. "Intergenerational Risk Sharing,"

Journal of Public Economics, vol. 37 (1988):185-202.

Haliassos, Michael and Carol C. Bertaut. "Why Do So Few Hold Stocks?

The View from Asset Pricing Theory," miineo (1992).

Hall, Robert E. "Consumption." National Bureau of Economic Research,

working paper no. 2265 (1987).

Hendershott, Patric, Eric Toder, and Yunhi Won. "Effects of Capital

Gains Taxes on Revenue and Economic Efficiency," National Tax
Journal, vol. 44 (1991):21-40.

Hendershott, Patric, Eric Toder, and Yunhi Won. "A Capital Gains
Exclusion and Economic Efficiency," Tax Notes, February 17,
1992, 881-885.

Kandel, Shmuel and Robert F. Stambaugh. "Asset Returns and

Intertemporal Preferences," Journal of Monetary
Economics, vol. 27 (1991):39-71.

Kiefer, Don. "Lock-in Effect with a Simple Model of Corporate Stock

Trading," National Tax Journal, vol. 43 (1990):75-94.

Kimball, Miles. "Standard Risk Aversion" mimeo, 1989.

Lyon, Andrew B. "Capital Gains Tax Rate Differentials and Tax Trading
Strategies," mimeo (1990).

Mehra, Rajnish and Edward C. Prescott. "The Equity Premium: A

Puzzle," Journal of Monetary Economics, vol. 15 (1985): 145-161.

Poterba, James M. "How Burdensome Are Capital Gains Taxes?"

Journal of Public Economics, vol. 33 (1987):157-172.

Schwert, G. William. "Indexes of U.S. Stock Prices from 1802-1987,"

Journal of Business, vol. 63 (1990):399-426.

36
Segal, Uzi and Avia Spivak. 'First Order versus Second Order
Risk Aversion," Journal of Economic Theory, vol. 51
(1990) : 111-125.

Shaw, Kathryn L. "An Empirical Analysis of Risk Aversion and Income

Growth," mimeo (1989).

Siegel, Jeremy J. "The Real Rate of Interest from 1800-1990: A Study of

the U.S. and U.K.," Journal of Monetary Economics, vol. 29
(1992) :227-252.

Slemrod, Joel. "A General Equilibrium Model of Taxation with

Endogenous Financial Behavior" in Martin Feldstein, ed.,
Behavioral Sinudation Methods in Tax Policy Analysis. Chicago:
University of Chicago (1983), 427-454.

Slemrod, Joel. "Taxation and Inequality: A Time Exposure Perspective"

in James M. Poterba, ed., Tax Policy and The Economy, vol. 6
(1992): 105-127, Cambridge: MIT Press.

Stiglitz, Joseph E. "Some Aspects of the Taxation of Capital Gains,"

Journal of Public Economics, vol. 21 (1983):257-294.

Varian, Hal R. "Redistributive Taxation as Social Insurance," Journal of

Public Economics, vol. 14 (1980):49-68.

37
 Table 1

FREQUENCY OF NET CAPITAL GAINS REALIZATIONS, 1985-1989

(1) (2) (3) (4)

Number of Years in Percent of All Percent of Percent of
which Taxpayer Taxpayers by Taxpayers Reported Net
Reported Net Gains Frequency of Reporting Gains Gains
Net Gains by Frequency of
Net Gains

No years 77.59 - -

One year 9.11 40.65 7.78

Two years 4.84 21.60 14.60
Three years 3.10 13.83 11.15

Four years 2.89 12.90 24.41

Five years 2.48 11.07 42.06
All taxpayers 100.00 100.00 100.00

Note: Frequency tabulated for taxpayers reporting only net capital gains (sum
of net gains reported on Schedule D and other distributions reported on Form
1040).

Source: Sales of Capital Assets Panel, 1985-1989.

 Table 2

DISTRIBUTION OF CAPITAL GAINS INCOME AND TOTAL INCOME:

SNAPSHOT VS. TIME EXPOSURE, 1985-1989

(1) (2) (3) (4) (5)

Income Single-Year Five-Year Single-Year Five-Year
Percentile Average Capital Average Income
Capital Gains Income
Gains

0- 10 3.25 3.70 0.17 0.78

10 - 20 0.10 0.34 2.66 2.86
20 - 30 0.35 0.98 3.78 4.01
30 - 40 0.62 0.52 4.97 5.12
40 - 50 0.76 1.07 6.28 6.43
50-60 1.08 2.17 7.99 8.04
60 - 70 0.81 1.34 9.94 9.93
70 - 80 2.27 1.83 12.28 12.30
80 - 90 3.21 5.59 15.69 15.43

90-95 5.49 6.48 10.14 10.04

95 - 99 15.76 17.87 12.72 12.45
99 - 100 66.30 58.12 13.38 12.61

Note: Income consists of AGI plus excluded gains and excluded dividends
(1985-1986), and statutory adjustments. Capital gains are those reported on
Schedule D and other distributions reported on Form 1040.

Source: Sales of Capital Assets Panel, 1985-1989.

 Table 3

DISTRIBUTION OF CAPITAL GAINS INCOME AND TOTAL

POSITIVE INCOME: SNAPSHOT VS. TIME EXPOSURE, 1985-1989

(1) (2) (3) (4) (5)

Positive Single-Year Five-Year Single-Year Five-Year
Income Capital Capital Average Positive
Percentile Gains Gains Positive Income
Income

0- 10 0.84 1.05 1.23 1.65

10 - 20 0.97 0.65 2.79 3.00
20-30 2.26 0.31 3.86 4.11
30 - 40 1.23 2.20 5.11 5.27
40-50 2.96 2.26 6.48 6.61
50-60 2.16 1.76 8.24 8.20
60 - 70 4.00 5.09 10.22 10.23
70 - 80 6.52 5.49 12.58 12.53
80 - 90 5.86 5.57 15.94 15.58
90 - 95 6.62 6.32 10.29 10.05
95 - 99 20.37 20.16 12.43 12.31
99 - 100 46.21 49.15 10.81 10.46

Note: Positive Income consists of AGI plus excluded dividends (1985-1986)

and statutory adjustments less (i) net capital gain or loss, (ii) Schedule C
losses, (iii) Schedule E losses, (iv) Schedule F losses, (v) Form 4797 business
property losses, and (vi) negative other income. Capital gains are those
reported on Schedule D and other distributions reported on Form 1040.

Source: Sales of Capital Assets Panel, 1985-1989.

 S

Table 4

AGE DISTRIBUTIONOF CAPITALGAINS, 1985-1989

(1) (2) (3) (4) (5) (6)

Age in 1985 Percent of Percent of All Percent of Percent of 5- Percent of
All Returns Taxpayers Taxpayers Year Net Gain Age Group
(with or with Gain or with Gam or or Loss with Gain
without Loss Loss in 4 or or Loss in
gains) More Years at Least
One Year

less than 30 31.83 15.99 10.95 4.01 12.82

30 - 49 38.55 40.56 31.61 29.35 26.84

50 - 54 6.60 7.78 7.96 14.45 30.10
55 - 59 6.36 8.93 14.74 12.81 35.86
60 - 64 5.14 8.91 8.52 12.87 44.23
65 - 69 3.92 7.14 11.68 10.62 46.45
70 and older 7.60 10.68 14.53 15.91 35.84
All taxpayers 100.00 100.00 100.00 100.00 25.52

Source: Sales of Capital Assets, 1985-1989.

 Table 5

RISK PREFERENCES AND PORTFOUO HOLDINGS

(I) (2) (3) (4) (5) (6)

Average
Income Percent Percent Percent Riskiess Average Stock

Percentile Desiring Holding Holding Financial Holdings

No Financial Riskiess Stock Assets (1992 5)
Risk Assets (1992$)

0-10 67 37 6 1,284 664

10-20 70 48 3 2,607 247

20-30 59 52 7 4,959 760

30-40 54 68 14 10,068 1,587

40-SO 44 72 13 8,031 2,127

50-60 47 77 15 12,218 2,895

60-70 42 84 19 12,041 1,094

70-80 34 84 26 14,335 6,346

80-90 34 92 30 19,493 6,504

90-95 21 94 44 25,612 20,675

95-99 17 93 55 48,770 68,169

99-100 6 96 78 147,467 801,276

Note: Riskless financial assets are defined to include savings accounts (other than checking), money market
funds, CDs, and bonds. Stock ownership includes mutual funds. Holdings in pension funds and [RAs are not
included. Income is the sum of all income reported by the respondent.

Source: 1983 Survey of Consumer Finances.

 Table 6

SUMMARY OF POUCY EXPERIMENTS EFFECTS ON TAX LiABILiTIES

PercentageChange in Actual Tax Liability

High-Income Group Low-IncomeGroup

Policy Change (A=3) (A10)

(I) Replace benchmark with lump -49.7% -22.9%

sum tax

(2) Remove loss limits, and +28.9% 0%

increasecapital gains tax rate

(3) 20% reduction in capitalgains -5.6% -3.0%

tax rate, increase Ifl

(4) Accrual taxation, and remove + 15.7% -1.7%

loss limits

Note: Each policy change holds the utility of each group constant.
 Proportion of Capital Gains Realized
Benchmark

t00a
0
0

Degree of Risk Aversion

 -o

C
ci)
C)

C
0
U)

!
ho- U)

9-
0
'I)
I-
0)
.— 0
,1 C3 0
%J.1 i_
COLL
(:cti
cDv'
0
C

N.

uopjne SSGOX U! OSPOJOUI

 ange iii Ez Bui Aut Tathx
as Fracflon ol
Income
Change inTaxExcess Burden
Payments by Group

C
0)
-
(I)
(I)
w
C)
w
C
0)
0)
C

5 6 7 10
Risk Aversion Coefficient