Stock Prices and Fundamentals: January 2000

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/5033933
Stock Prices and Fundamentals
Article · January 2000

DOI: 10.1162/08893360052471455 · Source: RePEc
CITATIONS READS
180 8,647
2 authors, including:
Deborah J. Lucas
Massachusetts Institute of Technology
76 PUBLICATIONS 6,490 CITATIONS
SEE PROFILE
All content following this page was uploaded by Deborah J. Lucas on 21 July 2014.
The user has requested enhancement of the downloaded file.

This PDF is a selection from an out-of-print volume from the National
Bureau of Economic Research
Volume Title: NBER Macroeconomics Annual 1999, Volume 14
Volume Author/Editor: Ben S. Bernanke and Julio J. Rotemberg, editors
Volume Publisher: MIT
Volume ISBN: 0-262-52271-3
Volume URL: http://www.nber.org/books/bern00-1
Conference Date: March 26-27, 1999
Publication Date: January 2000
Chapter Title: Stock Prices and Fundamentals
Chapter Author: John Heaton, Deborah Lucas
Chapter URL: http://www.nber.org/chapters/c11048
Chapter pages in book: (p. 213 - 264)

JohnHeatonandDeborahLucas
NORTHWESTERN
UNIVERSITY
KELLOGGGRADUATESCHOOLOF
MANAGEMENTAND NBER
Stock Prices and Fundamentals
1. Introduction
While stock returns in the United States this past century have exceeded
Treasury returns by an average of about 6% annually, in the last few years
they have done so by more than 12% annually. Commentators have sug-
gested a variety of explanations for the dramatic stock-market run-up that
accompanied these high returns. The baby boom is entering peak savings
years, productivity has escalated worldwide due to technological im-
provements and political change, and stock-market participation rates are
on the rise. The growth of mutual funds has lowered transaction costs and
made diversification feasible. Public awareness of the benefits of stock-
market investing is high. On the other hand, irrational exuberance could
be fueling the price rise, with inexperienced investors expecting double-
digit returns to continue indefinitely or at least long enough to reap a
substantial gain.
Whether the price rise is due primarily to fundamentals or whether it
is the result of a bubble is important to policymakers concerned with
avoiding the real disruption a sharp stock-market decline could precipi-
tate. It is also important to the academic debate over the determinants of
stock valuations. Because this paper is about the relations between stock
prices and fundamentals, we emphasize three broad categories of expla-
nations for the recent price rise: changes in corporate earnings growth,
changes in consumer preferences, and changes in stock-market participa-
tion patterns. The goal in qualifying the importance of fundamental
effects is to better understand whether a combination of fundamentals
and statistical fluctuation can plausibly explain the observed magni-
tudes, or whether a bubble is the likely cause of the price rise.
The paper has benefited from the comments of John Campbell, Annette Vissing-
J0rgensen, and participants of the 1999 NBER Macroeconomics Annual Conference. We
thank the National Science Foundation for financial support.
214 *HEATON& LUCAS
Although the paper touches on a variety of issues, its main contribution

is to look more closely at how participation patterns have changed, and at
how they are expected to affect required returns in a stochastic equilib-
rium model. We interpret participation broadly to include both the frac-
tion of the population that holds any stocks, and the degree of diversifica-
tion of a typical stockholder. To review the evidence, we use data from the
Survey of Consumer Finances (SCF) to document changes in stock-
holding patterns and reported attitudes toward risk from 1989 to 1995.
Consistent with previous studies (e.g., Poterba, 1993; Vissing-Jorgensen,
1997), we see an increasing rate of stock-market participation over time.
Participation rates among the wealthy, who own the majority of stock,
however, have increased only slightly. Foreign participation changes may
also influence required returns. Using data from the U.S. Treasury, we
find that net purchases of stocks by foreigners have been relatively high in
recent years, but small in comparison with total trading volume. Finally,
flow-of-funds data show that diversification has increased markedly, with
large outflows of individual stocks from household portfolios moving into
mutual funds and other institutional accounts.
To quantify the potential impact of these changes, we calibrate an
overlapping-generations model that allows for considerable heterogene-
ity in the cross section of nonmarketable income risk, preferences, diversi-
fication, and participation. This extends the analyses of Basak and Cuoco
(1998), Saito (1995), and Vissing-Jorgensen (1997), all of whom consider
the effect of participation when traded securities span income realiza-
tions. We use this framework to experiment with changes in stock-market
participation rates, changes in background risk, changes in preferences,
and changes in the expected dividend process reflecting changes in diver-
sification. We find that for realistic changes in raw participation rates,
expected stock returns change very little. Within the range of risk-
aversion parameters normally considered, preference changes also have
little effect on expected return differentials. Changing the rate of time
preference has a significant effect on the level of all returns, but not on
the differential between stock and bond returns. One factor that appears
to have a significant effect on required returns is the degree of assumed
diversification. This suggests that one fundamental reason for the stock
price run-up may be the rapid growth of mutual funds and the accompany-
ing large increase in diversification.
The remainder of the paper is organized as follows. In Section 2 we
review the statistical evidence on whether the current stock price level
is anomalous. In Section 3, we discuss some possible explanations for
the stock price increase in the context of a simple discounted-cash-flow
model, and present some evidence from the SCF and other sources on
*215
StockPricesandFundamentals
changes in stock-market participation patterns. The influence of partici-

pation rates, extent of diversification, background income risk, and
preferences on stock prices is examined in Section 4 in an overlapping-
generations model. By considering a variety of scenarios reflecting si-
multaneous changes in several of these factors, we show that changes
in fundamentals can account for perhaps half of the observed increase
in price-dividend ratios in the model. Section 5 concludes.
Facts
2. Empirical
Historically stocks have returned a substantial premium over bonds. Over
the period 1871 to 1998, the average annual (log) real return on a broad-
based index of U.S. stocks was 7.3%, compared to an average (log) real
return on bonds of about 3%.1The return on stocks over the last few years
has exceeded this historical average. For example, since 1991, the average
real return on stocks was 17% per year. This has led many observers to
question whether expected returns looking forward are lower than they
have been in the past.
A related issue is the composition of recent returns, which have been
mostly the result of capital gains rather than increased dividend pay-
ments. To illustrate this, Figure 1 plots the ratios of prices to dividends
and prices to earnings for aggregate U.S. stocks. (For the years since
1926 this is based on the S&P 500 index.) Notice that the price-dividend
ratio for this index has increased to an unprecedented level since about
1995. The increase in this ratio is significant because in a discounted-
cash-flow model of stock valuation, it indicates a reduction in the ex-
pected rate of return or an increase in the dividend growth rate (see
Section 3). Because dividends are discretionary and only one of the ways
in which corporations distribute cash to shareholders, it may be more
informative to look at price-earnings ratios. Figure 1 also shows the ratio
of prices to earnings. This ratio is also at a relatively high level, but the
change has not been as dramatic as for dividends.
A notable aspect of the rise in the price-dividend ratio is that there is
substantial evidence that a large value of the price-dividend ratio pre-
dicts lower stock returns in the future. For example, Table 1 reports the
results of regressing annual (log) stock returns on a constant and the log
of the price-dividend ratio lagged one year for the period 1887 to 1998.
Notice that the coefficient on the dividend-price ratio is negative. This is
consistent with a large body of evidence (e.g., Campbell and Shiller,
1988; Hodrick, 1992; Lamont, 1998). At the current high level of the
1. Source: Robert Shiller's data, available at http://www.econ.yale.edu/shiller/chapt26.html.
216 *HEATON& LUCAS
Figure 1 PRICE-DIVIDENDRATIOAND PRICE-EARNINGSRATIO,

1871-1998
Price-DividendRatio, _ .. Price-EarningsRatio
2000
price-dividend ratio, this regression predicts a substantial decline in the

stock market over the next year. In fact, since 1995 this regression has
consistently predicted a decline in the stock market.
On the other hand, due to the substantial variability in stock returns, it
is possible that the recent returns are within the bounds of normal statis-
Table 1 REGRESSION
OF ONE-
YEARSTOCKRETURNS
ON LAGGEDP/D OVER
THEPERIOD1871TO 1998
Coefficient Estimate StandardErrora
a 0.28 0.02
/3 -0.07 0.05
logRS+1 = a + ,3log(Pt/D,) + Et.
aCorrectedfor conditional heteroskedasticity
and autocorrelation using the procedure of
Newey and West (1987)and two years of lags.
*217
tical fluctuations, without any change in the underlying driving pro-

cesses. For example, the standard deviation of the annual premium of
stock returns over bond returns over the period 1871 to 1998 was 18%.
Therefore, it is not improbable that one would observe several years of
premiums in excess of 20% per year, even with no change in the underly-
ing statistical process. Since there is not a statistically definitive answer
to the question of whether returns have been abnormally high, we focus
below on whether recent changes in various aspects of the economy are
large enough to suggest a fundamental change in expected returns.
3. PossibleExplanations
In this section, we discuss some of the potential explanations that have
been offered for the stock price run-up, and begin to evaluate their likely
quantitative importance in the context of a simple discounted-cash-flow
model. We also present some evidence on changes in market participa-
tion patterns that may be influencing required returns.
3.1. GORDONGROWTHMODEL
The Gordon growth model is perhaps the simplest fundamentals-based
approach to predicting stock prices.2 In this model, stock prices are
based on the discounted present value of future expected dividend pay-
ments. It is assumed that dividends grow, on average, at a constant rate,
g, and investors discount dividends at a constant rate, r. Dividends,
earnings, and growth are connected by two equations: DIV = (1 - p)E
and g = p(ROE), where DIV is dividends, E is earnings, p is the propor-
tion of earnings reinvested, and ROE is the marginal physical product of
capital. If the marginal physical product of capital is constant, and if the
fraction of reinvested earnings is constant, then, all else equal, dividend
growth is constant. Then the price-dividend ratio equals 1/(r - g).
The model highlights two of the fundamental reasons that the price-
dividend ratio can change. The first is due to changes in dividend
growth, reflected in the choice of g. The second is due to changes in
preferences that affect the subjective rate of time preference or the pre-
mium demanded for risk, reflected in the choice of r.
Expectations of g may be higher than in the past for several reasons. A
major determinant of dividend growth is the availability of profitable
investment projects. The potential for sustained economic growth in
excess of historical precedent has been attributed to the opening and
2. This valuation model, a staple of marketanalysts, is described, for instance, in Brealey

and Myers (1996).
218 *HEATON& LUCAS
integration of world markets, continuing technological advances, and an

increasingly educated labor force. In fact, U.S. per capita GDP growth
has been slightly higher than average in recent years, averaging 2.3%
from 1995 to 1998, compared with 2.0% from 1947 to 1998.
Other considerations suggest that r may be lower than in the past.
One possibility is that aggregate preferences have changed. Either a
decrease in risk aversion or an increase in patience could contribute to
the run-up in stock prices. Risk aversion could vary across generations
due to their varying experiences and circumstances. For example, baby
boomers do not share their parents' first-hand experience with the Great
Depression. Some have argued that the economy is more stable, reduc-
ing the exposure to background risk, and possibly reducing the risk to
dividends. Davis and Willen (1998) show, for example, that the income
risk for households with various educational attainments has changed
over time. Reduced transaction costs in financial markets make diversifi-
cation easier, which, as discussed below, can reduce effective aversion to
the risk of holding stocks as people hold more diversified portfolios.
It should be noted that these types of changes affect the risk-free rate
as well as the expected return on stocks. Since the risk-free rate has been
relatively stable over the period of the recent stock price run-up, in much
of what follows we focus on factors that affect the equity premium,
rather than the absolute level of rates.3
The Survey of Consumer Finances (SCF) is one of the few data sources
that provides some direct survey evidence on peoples' attitude towards
financial risk and how it has changed over time. Respondents to the SCF
answer detailed questions, both quantitative and qualitative, about their
financial situation. The survey is conducted by the Federal Reserve
Board every three years, with different households in each survey year.
Here we focus on the question:
Whichof the statementson this page comesclosest to the amount offinancial risk
that you (and your husband/wife) are willing to take when you save or make
investments? If more than one box checked,code smallest category #.
1. takesubstantialfinancial risks expectedto earn substantial returns

2. takeaboveaveragefinancial risks expecting to earn aboveaveragereturns
3. takeaveragefinancial risks expecting to earn averagereturns
4. not willing to takeany financial risks
3. See Blanchard (1993) for an analysis of historical trends in the equity premium and risk-
free rate.
*219
Table2 AVERAGERESPONSEBYAGE AND SURVEYYEARTO

QUESTIONSABOUTRISKAVERSIONFROMTHESCF.
Responsea
Year Age < 35 35-65 >65
1989 3.14 (0.88) 3.32 (0.77) 3.63 (0.61)
1992 3.19 (0.84) 3.26 (0.81) 3.64 (0.60)
1995 3.07 (0.87) 3.18 (0.82) 3.58 (0.68)
aImpliedpopulationstandarddeviationsin parentheses.
Table 2 reports the average response by age and survey year. The implied
population standard deviation across responses is reported in parenthe-
ses. Since the population represented by the survey totals approximately
90 million households, the standard errors of the estimates of the means
are quite small. Consistent with the idea that risk tolerance has in-
creased, the average reported aversion to risk has decreased slightly for
each age category over time. Older households own significantly more
stock than younger households, and reported risk aversion increases
with age in each survey year. When a similar tabulation (not reported
here) is done conditional on households that own at least $500 in stocks,
the same patterns emerge with respect to age and time. The average
reported level of risk tolerance, however, is higher when we condition on
stockholders. For instance, in 1995 the average risk attitude for stockhold-
ers over age 65 was 3.17, as compared to 3.58 for all households over 65.
This suggests that those who already own stocks are more risk-tolerant
as a group than nonparticipants. Hence, the entry of new stockholders
may slightly decrease the average level of risk tolerance. One would
expect this to mitigate the effect of wider participation in reducing the
equity premium.
There are objective reasons why the underlying subjective rate of time
preference also may be changing. Increases in life expectancy beyond
retirement would likely increase the incentive to save and thereby re-
duce required returns. Mortality, for example, has declined at an average
annual rate of 3.3% over the period 1900 to 1988 (Social Security Adminis-
tration). Past improvements in health and life expectancy might under-
state expected improvements in these factors that are premised on con-
tinued medical progress.4 As with the other explanations considered for
the stock price run-up, however, it is hard to point to events that would
4. In fact, there is a lively debate in the demographic literature on these questions, with
some authors claiming that a life expectancy at birth of 100 years will be realized early in
the next century.
220 *HEATON& LUCAS
trigger a large change in aggregate preferences over the course of only a

few years.
Calibrating the Gordon growth model gives a rough sense of how far
earnings growth rates or stock returns would have to deviate from their
historical averages to justify current price levels. This approach has the
advantage that it allows one to avoid taking a definitive stand on the
magnitude of technology or preference parameter changes. In the tabula-
tions presented here, we focus on earnings-adjusted price-dividend ra-
tios rather than actual price-dividend ratios because earnings are likely to
be a more stable proxy for long-run payments to shareholders. Consistent
with the average ratio of dividends to earnings over the period 1947 to
1997, we assume an average reinvestment rate of 50%. Hence, the ad-
justed price-dividend ratio is defined as twice the price-earnings ratio.5
Over the past century, real earnings growth has averaged about 1.4%
annually, with a standard deviation of about 25%. Table 3 shows the
required growth rate in the future to match current and historical ad-
justed price-dividend ratios, for various levels of required returns. For r
ranging from 5% to 15%, column 2 reports the growth rate g that is
consistent with the adjusted price-dividend ratio of 28 for the period
1872 to 1998. Column 3 reports the growth rate necessary to match the
adjusted price-dividend ratio of 48 in January 1998 (the ratio in January
1999 is even higher at 58). For instance, to realize a real stock return of
7% (consistent with a 6% equity premium and a 1% real risk-free rate)
and to match the average historical adjusted price-dividend ratio of 28
requires growth of 3.4%. To match the 1998 adjusted price-dividend
ratio of 48, assuming a real risk-free rate of 3% and an equity premium of
6%, requires perpetual growth of 6.9%. This is a large number by histori-
cal standards, suggesting that, at least in this simple model, a plausible
increase in the expected long-run growth rate is unlikely to be the sole
explanation for the increase in stock prices.
The growth rate and required return enter symmetrically in these
calculations. Therefore, another interpretation of the results in Table 3 is
that if growth rates are expected to be similar to historical averages, the
expected real return on the stock market is now less than 5%. Again
assuming a risk-free return of 3%, this implies an equity premium below
2%. This large a change in expected returns also seems unlikely to have
taken place over the period of only a few years.
One shortcoming of this model is the restriction that the expected
5. While stock prices depend on the long-run behavior of dividends, properly measured,
in the short run dividends can vary due to temporary changes in payout policy (for
instance, in response to changes in the tax law). Therefore, it is common to focus on the
price-earnings ratio, adjusted for reinvestment rates, to approximate long-run price-
dividend ratios.
*221
Table 3 GROWTHRATESIMPLIEDBYTHEGORDONGROWTHMODEL
Ten-Yearg to
Long-Rung to Long-Rung to Match1998P/E
rs MatchHistoricalP/E Match1998P/E with2%Tail
0.05 0.014 0.029 0.064
0.07 0.034 0.049 0.134
0.09 0.054 0.069 0.189
0.11 0.074 0.089 0.236
0.13 0.094 0.109 0.280
0.15 0.114 0.129 0.320
growth rate is constant. This assumption, however, can be relaxed quite

easily. A minor variation on the model is to assume a higher growth rate
for some number of years, followed by a return to a lower long-run
growth rate. Column 4 of Table 3 reports, for each value of r in column 1,
the growth rate over 10 years necessary to explain the adjusted price-
dividend ratio in January 1998. The calculation assumes that the growth
rate returns to the long-run average of 2% from year 10 onward. In this
case achieving a 9% average rate of return requires a growth rate of
18.9% for ten years!
Although these calculations are admittedly primitive, more detailed
analyses along similar lines produce qualitatively similar conclusions.
For instance, Lee and Swaminathan (1999) estimate the value of individ-
ual stocks in the Dow Jones Industrial Average, projecting cash flows
using accounting data and analysts' forecasts, and discounting using the
CAPM. They conclude that the index is about 1.6 times the fundamental
value predicted by their analysis.
Despite the apparently large changes in parameters necessary to ex-
plain current price levels, these results do not preclude a fundamentals-
based explanation. It is possible that there have been a simultaneous
increase in expected growth rates and a reduction in required returns.
For instance, if the long-run growth rate is realistically expected to be
about 2.4% and if expected returns fall to about 6.6%, current prices are
in line with fundamentals. Our focus in the rest of the paper is on
whether such a change in expected returns can be attributed to measur-
able changes in the economy, in the context of an equilibrium model.
One factor of particular interest is the change in stock-market participa-
tion patterns, which is the topic of the next subsection.
3.2 STOCK-MARKET
PARTICIPATION
PATTERNS
It is well documented that a large fraction of the U.S. population holds
little or no stocks (Bertaut and Haliassos, 1995; Blume and Zeldes, 1993)
222 *HEATON& LUCAS
and that participation varies systematically with factors such as wealth

and age (Gentry and Hubbard, 1998; King and Leape, 1987). As noted
in several recent studies (e.g., Basak and Cuoco, 1998; Constantinides,
Donaldson, and Mehra, 1998; Saito, 1995; Polkovnichencko, 1998; Vis-
sing-Jorgensen, 1997), an increase in the stock-market participation rate
has, in theory, the potential to decrease the required risk premium on
stocks because it spreads market risk over a broader population.6 Not
only has the number of participants been rising, but the nature of
participation has changed. A typical stockholder today has a more di-
versified portfolio than in the past, presumably due to the lower cost of
diversification. Thus, the effective risk of the typical portfolio may have
declined. In this subsection, we review some of the evidence on these
changes.
The best source of data on market participation rates in the United
States is perhaps the SCF, which reports detailed information about
household wealth composition every three years. Using these data,
Poterba (1998) reports that in 1995 there were approximately 69.3 million
shareholders in the United States, compared to 61.4 million in 1992 and
52.3 million in 1989. There is also evidence that people are entering the
market at a younger age. Poterba and Samwick (1997) show that baby
boomers are participating more heavily in the market than previous
generations at a similar age. Baby boomers are entering peak savings
years and directing some of their savings into stocks. More generally, the
aging of the population should result in a greater demand for stocks,
since older people hold proportionally more of their wealth in the mar-
ket than do younger people (see, e.g., Heaton and Lucas, 1998). Finally,
foreign participation in the U.S. markets has increased, further spread-
ing the risk across a broader population.
Market participants are also holding more diversified portfolios, which
reduces their exposure to risk from their stockholdings. This is potentially
important, since holders of diversified portfolios may demand a lower
average return. Historical evidence on this phenomenon of improving
diversification is summarized in Allen and Gale (1994). Friend and Blume
(1975) found that a large proportion of investors had only one or two
stocks in their portfolios and very few had more than ten. At the time, this
lack of diversification in individual stockholdings could not be justified by
the claim that these investors achieved diversification through unre-
ported mutual-fund holdings. In fact, King and Leape (1984) found that
6. Bakshi and Chen (1994) note that to the extent that demographic changes have an effect
on the demand for stocks, they will have a predictable effect on asset prices. Bodie,
Merton, and Samuelson (1992) provide a theoretical justification for the demand for
stocks to vary with age.
*223
only 1% of investors' wealth was in mutual funds at around that time. In

contrast, Poterba (1998) reports a sharp increase in the proportion of stock
held in mutual funds over time and a reduction in directly held stocks. For
instance, while the total number of individuals holding stock increased
from 61.4 million to 69.3 million from 1992 to 1995, the number of individu-
als holding stock directly fell from 29.2 million to 27.4 million over the
same period. In the calibrations presented below, we will look at whether
this diversification effect is significant by comparing stock prices with
different underlying assumptions about dividend volatility, where high
assumed dividend volatility proxies for less diversification.
Although these statistics point to an increase in participation and a
reduction in risk exposure, it is questionable how significant these ef-
fects are quantitatively. The change from 52 million to 69 million partici-
pants is a 33% increase, but when the numbers are wealth-weighted, the
increase is much smaller. Now as in the past, the vast majority of stocks
are held by wealthy individuals. For instance, Poterba (1998) finds that in
the 1995 SCF, 82% of stock was held by households with a stock portfolio
exceeding $100,000, and 54% of stock was held by households with
annual income over $100,000. This suggests that stockholdings remain
extremely concentrated. Figures 2 and 3 present a more complete picture
of how the distribution of stockholdings vs. wealth and income has
changed over the period 1989 to 1995 (see also Table 4). Using data from
the SCF, we plot the share of stocks held against the share of income or
wealth. Stockholding looks more democratic when measured relative to
income than relative to wealth, since as noted in Vissing-J0rgensen
(1997), lower-labor-income households own a larger share of the market
than in the past. When the metric is wealth, however, there has been
very little change-holdings were and are extremely concentrated.
Table4 PROPORTIONOF POPULATIONTHATHOLDSSTOCKBY

WEALTHCOHORT
Percentile Range Proportion Range Proportion
Range(%) ($) (%) ($) (%)
<25 <801 2.3 <1101 4.7
26-50 801-40,051 13.0 1102-40,500 17.8
51-75 40,052-121,500 21.6 40,501-126,251 28.7
76-90 121,501-279,001 36.7 126,252-309,501 47.8
91-95 279,001-456,000 55.4 309,502-574,000 62.8
96-99 456,001-1,767,730 65.8 574,001-1,814,330 78.3
>99 >1,767,730 84.3 >1,814,331 82.0
Figure 2 PERCENTAGEOF STOCK HELD BY INCOME PERCENTILE
0: 1995,X: 1989
.g
5)
0
U-
40 50 60 100
Income Percentile
Figure 3 PERCENTAGEOF STOCK HELD BY WEALTH PERCENTILE

0: 1995, X: 1989
I 1111
vv
I
I 1 i i 1 I i I 'Jf
90
80
70
e)
0
0
60
c
50
"o
0
40
LL
30
I
20
10
J KS)'
~ ' ]I'-,I I
0 10 20 30 40 50 60 70 80 90 100
Wealth Percentile
*225
Table5 NET PURCHASESOF STOCKSBYINDIVIDUALS

Purchases(billionsof dollars)
Year Net ThroughMutualFunds OutsideMutualFunds
1995 -116.1 91.3 -207.4
1996 -101.9 218.4 -320.3
1997 -179.0 190.2 -369.2
Ideally, one would like to measure the net investment in the stock
market in recent years on behalf of households. If net inflows were large,
one could perhaps conclude that the demand for stocks had significantly
increased. The fact that aggregate savings rates are low is indirect evi-
dence that these net inflows cannot be large. Still, there could be substitu-
tion out of money and bonds into stocks, increasing the net flow into
stocks. According to the flow-of-funds accounts, U.S. Treasury securities
are the only category of fixed-income investment that had a large net
outflow from the household sector in recent years. Calculating flows into
stocks directly is tricky because there have been large changes in the
institutional structure of the investment industry. Table 5, using data
from the Investment Company Institute, shows net purchases of stocks,
purchases made through mutual funds, and purchases made outside
mutual funds, by households, from 1995 to 1997. While purchases made
through mutual funds increased significantly over the period, net pur-
chases of equities by households were actually negative in each year.
This is because households were net sellers of equities to institutions.
Changes in foreign participation in the U.S. market may also affect
expected returns. Assuming that foreign participants are similar to U.S.
stockholders in their attitude towards risk and their ex ante risk expo-
sure, an increase in foreign participation should lower expected returns
by increasing opportunities for diversification. Net foreign purchases of
stock have spiked sharply in recent years (see Figure 4), and these in-
flows, over the period January 1988 to February 1999, have a correlation
of 0.13 with monthly returns on the S&P 500. The average monthly net
inflow between January 1996 and February 1999 is $3.8 billion, compared
to only $349 million from the period January 1988 to December 1995.
Although the inflows have increased significantly, they still represent a
small fraction of total market transactions, which totaled approximately
$479 billion per month in 1997 on the New York Stock Exchange alone.7
7. Data on foreign purchasesand sales of U.S. stocks are from the U.S. Departmentof the
Treasury'stable "TICCapital Movements, U.S. Transactionswith Foreignersin Long
TermSecurities."S&P 500 monthly returns data are from RobertShiller.Totaltrading
volume is from the NYSE 1997 Fact Book.
226 - HEATON& LUCAS
Figure4 FOREIGNNET INVESTMENTAND RETURNS

Foreign Net Stock Investment
- - S&P 500 Return
tc
a:
a
c
CO
a)
Date
4. An Overlapping-Generations
Model
In this section, we ask whether changing stock-market participation pat-
terns and increased diversification can have a quantitatively important
effect on stock prices in an equilibrium model. We calibrate an overlap-
ping-generations (OLG) model in which agents face both aggregate and
idiosyncratic income risk, and a variable subset of agents has limited
access to financial markets.
The effects of limited participation in financial markets has been con-
sidered by a number of authors, including Basak and Cuoco (1998),
Saito (1995), and Vissing-Jorgensen (1997). In these papers, aggregate
consumption is completely traded in financial markets in the form of
dividends. Only a limited number of agents can trade claims on this
dividend flow directly. The other agents participate in financial markets
only by trading claims to risk-free bonds. The result is incomplete
sharing of aggregate risk, with stockholders often taking leveraged posi-
*227
tions to accommodate the demand for bonds by nonparticipants. Be-

cause of this, a larger risk premium is necessary to induce those in the
stock market to hold all of the aggregate risk. It is difficult to justify the
magnitude of the observed equity premium in these models, however,
unless one assumes very high risk aversion or very low participation
rates.
One way to increase the effects of limited participation is to include
other sources of uninsurable income risk. For instance, income from
wages and/or privately held businesses constitutes the majority of in-
come for most households (Heaton and Lucas, 1998). These income
flows are difficult to contract upon, and a large component of this
income risk is specific to each individual of household. We refer to the
sum of labor income and privately held business income as non-
marketed income. Potential differences in the properties of this income
for participants versus nonparticipants are likely to influence the ef-
fects of limited participation on asset returns. This is consistent with
the empirical observation that the consumption of stockholders is
more volatile than that of nonstockholders (Mankiw and Zeldes, 1991).
Polkovnichenko (1998) demonstrates how differential income risk and
risk aversion can affect asset prices in a model with infinite-lived
heterogeneous agents. He shows that a small fixed transaction cost
that endogenously limits stock-market participation can interact with
idiosyncratic risk to result in a bigger equity premium than in a repre-
sentative agent model, although matching the observed premium is
still elusive.
The model presented here allows us to examine the effect of participa-
tion and diversification while considering a greater degree of cross-
sectional heterogeneity than in the previous literature, due to the simpli-
fying assumption of two-period lives. Unlike the papers discussed
above, which focus on whether limited participation can explain the
historical equity premium, we focus on the question of to what extent
observed changes in participation rates can explain the recent run-up in
stock prices. We also emphasize the effect of changes in the degree of
portfolio diversification.
4.1 STRUCTURE
OF THEMODEL
At each time period, t, a generation of J "young" agents are born and live
for two periods.8 Let C(j,t) be the consumption of agent j when young,
8. Storesletten, Telmer, and Yaron (1998) consider an OLG model in which the agents face
nontradable idiosyncratic risk and live for a large number of periods. We limit ourselves
to a smaller number of periods to make numerical solution of the model easier.
228 *HEATON& LUCAS
and C?(j,t + 1) be the consumption of agent j when old (j = 1,2, . . ., ).

The utility specification for agent j distinguishes between risk aversion
and the elasticity of intertemporal substitution. We use the parametric
form proposed by Epstein and Zin (1989) and Weil (1990):
U(j,t) logC(j,t) + 1 logE[CO(j,t + 1)1- i(t)],

| (1)
1-aj
where 3j> 0 and aj > 0. Here ;(t) is the information available at time t and
is assumed to be common across agents. As discussed by Epstein and Zin
(1989), the parameter aj is the coefficient of relative risk aversion. The
elasticity of intertemporal substitution equals one. In the experiments
considered below, changes in participation affect the equilibrium volatil-
ity of individual consumption in the second period. In general, this affects
both the level of interest rates and the equity premium. By distinguishing
between risk aversion and the elasticity of intertemporal substitution, the
effect of the income process on the equity premium is to some extent
separated from its effect on the risk-free rate.
Each agent j(j = 1,2, . . ., J) is endowed with random nonmarketed
income Y(j,t) at time t and random nonmarketed income Y?(j,t + 1)
when old at time t + 1. The details of the individual income processes
are described below. The agents trade in financial markets in an attempt
to smooth consumption over time. There are two securities that can be
traded: a stock and a risk-free bond. At time t the stock represents a
claim to future dividends {D(t + r): r = 1,2, . . . }. The total supply of
stock is normalized to one. The bond is assumed to be in zero net supply.
Each agent is exposed to nonmarketed income risk that has both an
aggregate component and an idiosyncratic component. Aggregate non-
marketed income at time t is denoted by Ya(t),where
J J
ya(t) = Y(j,t) + E Y(j,t). (2)
j=l j=1
In equilibrium, this endowment plus dividends equals aggregate con-

sumption at time t. The properties of individual nonmarketed income
Y(j,t) and Y?(j,t) will be potentially important in assessing the effects of
changing participation an'd background income risk on equilibrium
returns.
At time t each young agent maximizes utility (1) subject to a constraint
?229
that depends on the agent's access to financial markets. Let PS(t)be the
price of the stock at time t, and Pb(t) be the price of a bond that pays one
unit of consumption at time t + 1 for sure. If agent j has access to both
financial markets, then the agent's flow wealth constraints are
C(j,t) = Y(j,t) - S(j,t)P(t) - B(j,t)Pb(t), (3a)

C?(j,t + 1) = Y?(j,t + 1) + S(j,t)[Ps(t + 1) + D(t + 1)] + B(j,t), (3b)
where S(j,t) gives the stockholdings of agent j, and B(j,t) gives the
bondholdings of agent j. A subset of the agents is assumed to have only
limited access to financial markets and can trade only in bonds. In this
case the constraints (3) are replaced by
C(j,t) = Y(j,t) - B(j,t)Pb(t), (4a)

C?(j,t + 1) = Y?(j,t + 1) + B(j,t). (4b)
An equilibrium is given by processes for stock and bond prices {Ps(t): t =

0,1, . . } and {Pb(t): t = 0,1, . . . } such that
S*(j,t) = 1, (5a)
j=l
B*(j,t) = 0, (5b)
j=l
where {S*(j,t),B*(j,t)} maximizes (1) subject to (3) if the agent can trade in
both markets, or subject to (4) if the agent can trade only in the bond
market.
We assume that nonmarketed income {Ya(t): t = 0,1, . . . } and divi-
dend income grow over time in such a way that the growth rate of
aggregate income is a stationary process. Consistent with this, we as-
sume that at time t we have Y(j,t) = y(j,t)Ya(t) and D(t) = d(t)Ya(t),where
y(j,t) denotes the share of individual j's income in aggregate income,
and d(t) the dividend relative to aggregate nonmarketed income. Simi-
larly we assume that Y?(j,t) = y?(j,t)Ya(t).This implies that one can look
for an equilibrium in which the stock price also scales with aggregate
income, so that PS(t) = ps(t)Ya(t).Finally, we assume that the face value of
a bond purchased at time t is given by Ya(t), so that B(j,t) = b(j,t)Y'(t),
where b(j,t) gives the quantity of these "rescaled" bonds purchased by
agent j.
230 *HEATON& LUCAS
4.2 CALIBRATION
In this subsection, we calibrate the model in order to revisit quantita-
tively some of the questions discussed in Section 3. How much do as-
sumed changes in participation rates affect the predicted equity pre-
mium and expected returns (and hence prices)? How does the degree of
portfolio diversification affect required returns? Can small changes in
preference parameters, reflecting changes in patience or risk aversion,
result in large changes in required returns? How important is hetero-
geneity in income risk? To answer these questions, the model is solved
numerically using standard techniques. Although we assume consider-
able heterogeneity in the cross section, the fact that agents only live for
two periods makes the problem numerically tractable.9
We begin by describing the parameterization of the income processes
and preferences. Parameters are chosen to reflect limited stock-market
participation, and to try to match gross features of the data with respect
to stock returns, the risk-free rate, and the driving processes for non-
marketed income and dividends.
As in most exercises of this type, the equity-premium puzzle remains
a serious problem. For income and dividend processes and participation
rates based on historical data, the model predicts an unrealistically small
equity premium. We have increased the assumed volatility of aggregate
income to increase the predicted premium, but want to emphasize that
this may not be a neutral adjustment with respect to the other quantities
of interest.10
4.2.1 Incomeand Preferences Let y(t) = log[Ya(t)/Y(t - 1)] be the growth

rate of aggregate nonmarketed income at time t. Then the aggregate state
of the economy is given by z(t) = [y(t) d(t)]', which is assumed to be
generated by a Markov chain. To calibrate a process for z(t) we assume that
a period corresponds to 25 years. The first period roughly corresponds to
the working years between age 40 and retirement, and the second period
is the time in retirement. Over the period 1889 to 1985, the average annual
(log) growth rate in real aggregate consumption was 1.7% with a standard
deviation of 3.5%. So that the model will produce a nonnegligible equity
premium, we assume that the standard deviation of the aggregate growth
rate in the model is 1.5 times the historical standard deviation of aggregate
9. The Matlab code is available upon request.
10. Recently Campbell and Cochrane (1998) suggested that time-varying habit provides a
higher estimate of the equity premium in a model based on aggregate consumption.
However, Cochrane (1997) claims that this preference specification cannot account for
the recent run-up in stock prices.
*231
consumption. For the same reason, we assume that annual income

growth is independently and identically distributed over time, although
in fact it is slightly negatively autocorrelated. This implies a 25-year aver-
age (log) growth rate of 42.5% with a standard deviation of 17.5%. This
distribution is discretized by assuming that y takes on the values 0.16 and
0.69 with equal conditional probability.
The capital share in total income averages approximately 30%. Consis-
tent with the aggregate statistics reported in Heaton and Lucas (1998),
we assume that only half of this capital income is actually tradable. The
nontradable portion, generated by private business holdings, is ac-
counted for in nonmarketed income. Since dividends in the model are
scaled relative to nonmarketed income, this means that we require d(t) to
average 18%. In most of the calculations d(t) is fixed at 18%. In other
experiments described below, we assume a more volatile dividend pro-
cess to proxy for a lack of diversification.1
The relative nonmarketed income of young agent j and of old agent k
at time t are given by
y(j,t) = E (j,t)[1 - 1(t)], (6)

y?(k,t) = E?(k,t)n(t), (7)
where
J J
E(j,t) = 1 and f(k,t) =1. (8)
j=1 k=l
Under this normalization, r(t) gives the share of old individuals' non-
marketed income in total nonmarketed income. The analysis is sensitive
to this parameter because the amount of nonmarketed income influ-
ences agents' attitude towards the risk of investment income. For the
basic analyis we assume that r1= 0.2 for all t, reflecting the observation
that noninvestment wealth is relatively small for retirees. In the sensitiv-
ity analysis, this parameter is varied to a maximum of 0.3.
The process for E(j,t) and E?(k,t)captures idiosyncratic income risk
across agents. We know from earlier work (e.g., Constantinides and
Duffie, 1996) that asset returns are potentially sensitive to the persis-
tence of idiosyncratic income shocks and to the correlation and condi-
tional covariance of idiosyncratic and aggregate shocks. We assume a
process for individual income risk based in part on the estimations re-
11. In constructing the total dividend series, we always normalize the level of dividends so
that they average 15% of GDP.
232 - HEATON& LUCAS
ported in Deaton (1992) and adjusted for the assumed 25-year period
length. Deaton reports a standard deviation of shocks for an MA(1)
specification of individual income growth of 15%, and an MA coefficient
of -0.4. Based on this, the idiosyncratic income shocks for both the
young and the old are assumed to have a standard deviation of 45% over
each 25-year period. The shock when young is assumed to be completely
persistent, so that
E?(j,t+ 1) = e(j,t)co(j,t + 1), (9)
where o(j,t + 1) is the further 45%-standard-deviation shock to relative

nonmarketed income that agent j faces when old. In experiments below,
we also consider the situation in which the idiosyncratic shocks of a
subset of the population are correlated with dividends. This captures the
possibility that certain classes of agents, such as business owners or
executives who own large shares of stock in their own corporation, face
risks that are more correlated with the market than a typical individual.
Because preferences are homothetic, when agents are assumed to be
homogeneous only the o-shock affects prices and portfolio choice. When
the wealth and income of participants and nonparticipants differ, how-
ever, the income distribution of the young can affect predicted returns.
For most of the analysis, preferences are parametrized with 3j= 0.9525
and ao = 5 for all j. These parameters are also varied in the sensitivity
analysis.
4.2.2 VaryingParticipationRates Table 6 shows what happens when the

assumed participation rate in the stock market is varied between 30%
and 100% of the population, assuming the preference specification and
processes for individual and aggregate income described above. As one
would expect, increased participation lowers the equity premium. No-
tice, however, that the effect is small in the region of participation rates
that correspond to the data. For example, when participation increases
from 50% to 80% of the population, the equity premium and the absolute
level of equity returns are reduced by less than a tenth of a percent.
Changing participation also has a small effect on the level of the risk-free
rate, with an increase in participation increasing the average rate of
return. This can be attributed to a precautionary effect that decreases
when risk is spread more evenly over the population. Although small,
this effect is in keeping with the observation that the risk-free rate has
risen in recent years.
Consistent with the literature on the equity-premium puzzle, aggre-
?233
Table6 AVERAGEANNUAL RETURNSAS A FUNCTION

OF PARTICIPATION
Returns(%)
Returns(%)
Percentage
of
Stockholders E(rb) E(r5) E(rs- rb)
100 4.42 5.47 1.05
90 4.40 5.48 1.08
80 4.38 5.49 1.11
70 4.37 5.50 1.13
60 4.35 5.51 1.16
50 4.33 5.52 1.19
40 4.32 5.53 1.21
30 4.32 5.55 1.23
gate income and dividend risk alone are not sufficient to generate a
sizable equity premium. This is true even under the assumption of ex-
tremely limited participation, inflated aggregate risk, and nonmarketed
income risk. Still, the premium predicted here is higher than in Mehra
and Prescott (1985) by about 1%. Experiments not reported here indicate
that this difference is due primarily to the assumption that aggregate risk
is higher than that observed in the data, rather than to limited participa-
tion or exposure to idiosyncratic income risk.
In the experiments that follow, we examine other stochastic steady
states based on different degrees of diversification, risk aversion, etc.
Although looking across steady states does not allow one to watch re-
turns gradually changing over time as parameters gradually change, it
does provide an upper bound on the size of these effects. Thus, one can
give a temporal interpretation to some of the experiments. For instance,
we will compare the stylized historical past, with low diversification and
low participation rates, to the stylized present, with greater diversifica-
tion, more complete participation, and greater patience.
4.2.3 Increasing Diversification As a proxy for the increased diversifica-

tion of a typical market participant over time, we vary the assumed
volatility of the dividend process. It is an empirical fact that the variabil-
ity of returns falls dramatically as diversification increases. Based on
CRSP monthly data from 1962 to 1997, Table 7 shows the effect of diversi-
fication on a typical portfolio's annual standard deviation. In monthly
data, we find an average individual stock standard deviation of 16% and
an average pairwise covariance of 0.01. The portfolio standard devia-
234 *HEATON& LUCAS
Table7 THEEFFECTOF DIVERSIFICATION

ON
PORTFOLIOVOLATILITY
No.of
No. of StandardDeviation(%)
Stocks Monthly Annual
1 16.0 55.4
2 11.7 40.4
3 9.8 33.9
4 8.7 30.2
5 8.0 27.7
10 6.3 21.9
20 5.3 18.3
100 4.3 14.9
500 4.1 14.1
tions reported in the table assume equal value weights on each stock.
Monthly returns are annualized under the assumption that they are
independent. These calculations show that holding a one-stock portfolio
results in an annual standard deviation of 55%, while increasing hold-
ings to five stocks decreases the standard deviation to 28%, and holding
500 stocks brings it down to 14%.
The above statistics on portfolio returns do not translate directly into
parameter values, since the inputs into the model are income and divi-
dend processes, whereas returns are endogenous. One assumption
about dividends that produces returns consistent with those observed in
CRSP data is that d(t) is variable over time, taking on the values 0.11 and
0.25 with equal probability. This level of variation essentially brackets the
variation in dividends' share in total income, based on the S&P 500
dividend flow and U.S. gross domestic product since 1947. We call this
the case of high dividend volatility. It implies variation that is approxi-
mately consistent with a three-stock portfolio under the parametriza-
tions we focus on.
Second, we consider a situation referred to as correlatedhigh dividend
volatility. Here the aggregate dividend is assumed to be correlated with
nonmarketed income, taking on the value 0.11 in the low-nonmarketed-
income state and 0.25 in the high-nonmarketed-income state. These first
two cases bracket two views of the relation between dividend growth
and income growth. The first is that there is very little correlation be-
tween income growth and dividend growth on an annual basis. The
second is that over longet time periods, such as the 25-year periods
?235
considered here, there is a positive correlation between dividends and

income.12
Finally, we represent the increased volatility in a poorly diversified
portfolio by assuming a skewed distribution of dividends. The dividend
share, 8, is fixed at 0.1865 for 95% of the time, but falls to 0.06 for 5% of the
time, independent of the aggregate state. This skeweddividendcase repre-
sents bankruptcy of a poorly diversified portfolio. It is further assumed
that zero is an absorbing state for the value of a bankrupt portfolio after
this small dividend is paid. To maintain stationarity, bankrupt shares are
replaced by new shares in the new generation. These new shares are held
in the portfolios of the young, but cannot be sold until the following
period. The reason to consider a more skewed distribution of payoffs is
twofold. First, although catastrophic outcomes are rare for the U.S. stock
market as a whole, individual firms fail quite frequently. Secondly, the
properties of the utility function suggest that skewed outcomes will have
a much different effect on asset prices than a symmetric distribution with
the same variance. In fact, the implied volatility of returns in this case is
set to be similar to that in the case of high dividend volatility.
Table 8 is similar to Table 6, but reports results under the assumptions
of high dividend volatility, correlated high dividend volatility, and
skewed dividends. Panel A reproduces the predicted returns under the
base-case set of assumptions for participation rates of 50% and 100%.
Relative to panel A, assuming high dividend volatility (panel B) has the
effect of decreasing the risk-free rate by 0.61% and 0.82% for participa-
tion levels of 100% and 50% respectively. It increases the equity premium
by 0.71% and 0.97%, respectively, for the same participation rates. These
results are consistent with the view that increased diversification has
significantly reduced the required equity premium, although for these
parameters it suggests only a slight decrease in the level of the required
return on equities. For the case of correlated high dividend volatility
(panel C), the effect on the equity premium of an increase in dividend
volatility is even larger.
Notice that for high dividend volatility an increase in participation re-
sults in a larger decline in the equity premium than for low dividend
volatility. This occurs in part because with high dividend volatility case
there is more risk to be shared, and hence a greater benefit from spreading
12. As one would expect, predicted returns are sensitive to the assumed degree of correla-
tion between dividends and nonmarketed income. It is not obvious, however, whether
the dividends from a poorly diversified portfolio are likely to be more or less highly
correlated with nonmarketed income than for a well-diversified portfolio. If, for in-
stance, households own stock primarily in the companies for which they work (a
common phenomenon), the correlation may be relatively high.
236 *HEATON& LUCAS
AND
Table8 AVERAGERETURNSAS A FUNCTIONOF PARTICIPATION
THEDIVIDENDPROCESS
Returns
Returns(%)
of
Percentage
Stockholders E(rb) E(rs) E(rs - rb)
A. LowDividendVolatility
100 4.42 5.47 1.05
50 4.33 5.52 1.19
B. HighDividendVolatility
100 3.81 5.57 1.76
50 3.51 5.67 2.16
HighDividendVolatility
C. Correlated
100 3.38 5.87 2.49
50 3.36 5.95 2.59
D. SkewedDividends
100 1.54 6.17 4.63
50 0.47 6.93 6.46
this risk to new participants with no initial exposure to market risk. For
correlated high dividend volatility, however, changes in participation
have a smaller effect on the equity premium than for uncorrelated high
dividend volatility. This is because the new entrants are less willing to
bear stock-market risk when it is correlated with their nonmarketed risk.
Finally, the much more dramatic results in the skewed-dividend case
are shown in panel D. The small risk of a catastrophic outcome reduces
the risk-free rate to 0.47% and increases the equity premium to 6.46%
with 50% participation. With 100% participation, the risk-free rate equals
1.54% and the equity premium is 4.63%. This assumption therefore
allows one to match the historical equity premium. It also suggests that
in this region of parameter space the premium is more sensitive to
changes in participation rates. This points to changes in diversification
as a potentially large factor in explaining changes in expected returns.
4.2.4 Preference-Parameter Changes The potential effects of changing risk

attitudes are explored by changing the coefficient of relative risk aver-
sion, aj. Recall that in all the results reported above aj is set to 5. If aj is
increased to 10, with all else as in the high-dividend-volatility case and at
a 50% participation rate, the equity premium rises by only 0.16%. The
risk-free rate falls by 0.3%. It is clear that over the range of risk-aversion
coefficients usually considered, a change in risk aversion does not ac-
count for large changes in stock prices in this model.
*237
As discussed in Section 3, increases in life expectancy may affect the

subjective rate of discount. Varying f3j is a proxy for these changes.
Unlike in an infinite-horizon model, where /3 generally does not have a
first-order effect on the equity premium, varying P3here influences the
equity premium as well as the general level of returns. The reason is that
when /3j increases, the value of future dividends and nonmarketable
income increases relative to the value of first-period income. This
changes the share of capital in wealth and increases the importance of
second-period income risk. For the parameters we consider, this results
in a lower equity premium in levels, but a higher premium relative to the
risk-free rate. For instance, increasing / from 0.9525to 0.9625, with 50%
participation, high dividend volatility, a equal to 5, and i] equal to 0.2,
moves the equity premium from 2.16% to 2.02%, and the risk-free rate
from 3.51% to 3.05%. We interpret the increase in the relative premium
as a response to the increased exposure to market risk. The reduction in
the absolute premium reflects the increased precautionary demand for
savings with the increase in risk, which lowers all required rates of
return.
Varying 71,the share of nonmarketed income accruing to the elderly,
similarly affects risk and hence returns. For instance, increasing 77from
0.2 to 0.3, with i3jequal to 0.9525and all else as in the case above, moves
the equity premium from 2.16% to 2.49% and the risk-free rate from
3.51% to 4.14%.
4.2.5 Heterogeneityin IdiosyncraticIncomeShocks An interesting question

is whether background income risk (i.e., nonmarketable risk) is different
for stockholders and nonstockholders, and whether this difference inter-
acts with the effect of participation changes on asset returns. In Heaton
and Lucas (1998) we present evidence that many large stockholders de-
pend more heavily on income from privately held businesses than on
labor income. Business income is more volatile and more highly corre-
lated with stock returns than is labor income. Hence, the equity pre-
mium is likely to fall more sharply if new entrants who are otherwise
similar to stockholders depend predominantly on labor income. As dis-
cussed in Section 3.2, in recent years there has been an increase in
participation by middle-income households, which are likely to contain
wage earners. We investigate the potential quantitative effect of this
change by assuming a different idiosyncratic income process for a subset
of the stockholders and for nonstockholders.13
13. Ideally we would make participation endogenous and hence a function of the assumed
income process, as in Polkovnichenko (1998). This tends reduce the risk-sharing capac-
ity of new entrants, since the most risk-tolerant agents already hold stocks when entry
238 *HEATON& LUCAS
Table9 AVERAGERETURNSAS A FUNCTIONOF PARTICIPATION

AND
THEDIVIDENDPROCESS(HETEROGENEOUS INCOMERISK)
Returns
Returns(%)
Percentage
of
Stockholders E(rb) E(rs) E(r - rb)
A. HighDividendVolatility
100 3.74 5.49 1.75
50 3.42 5.58 2.16
B. Correlated
HighDividendVolatility
100 3.32 5.79 2.47
50 3.28 5.85 2.57
C. SkewedDividends
100 1.18 6.12 4.94
50 0.17 6.82 6.65
To implement this, we assume that a fixed number of participants

have nonmarketed income that is correlated with the dividend flow from
stocks. New entrants to the stock market and nonparticipants have a less
correlated income process. More precisely, we assume that 25% of the
population receive idiosyncratic income when old, with a standard devia-
tion of 67.5% and a correlation with dividends' share in aggregate in-
come of 0.2. this group is always assumed to hold stocks.14 The rest of
the population receives idiosyncratic shocks that have a standard devia-
tion of 45% as before, and a correlation with dividends of -0.1. This
negative correlation is necessary to produce an average correlation that
is consistent with data. In annual data, one does in fact see a slight
negative correlation between labor income and stock returns.
Table 9 reports results under these assumptions for high dividend
volatility (panel A), correlated high dividend volatility (panel B), and
skewed dividends (panel C). In each case, the experiment is to move
from a situation in which 50% of the stockholding population is exposed
to high background risk to one in which 25% has this exposure. Changes
in participation now have only slightly more effect than in Table 8 for
high dividend volatility. The effect of a change in participation is slightly
smaller for skewed dividends. The effect of an increase in participation
on the premium relative to the risk-free rate is higher in each case,
is endogenous. For simplicity, and to put an upper bound on this effect, we assume
that participation is completely exogenous.
14. These parameter assumptions are consistent with the estimates reported by Heaton
and Lucas (1998).
*239
however, because of the greater volatility of the nonmarketed income of

stockholders. As in Table 8, the greatest effect of participation is in the
case of skewed dividends, where the equity premium falls by 1.71%
when participation increases from 50% to 100%.
4.2.6 Simultaneous Changes As discussed in the introduction, each of

the factors that we have looked at individually has been suggested as a
fundamental reason for the stock price run-up. We have seen that none
of these factors alone is sufficient to produce a large change in required
equity returns, and hence the large run-up in stock prices. Here we
examine the best case for the model, simultaneously changing a number
of parameters. The stylized historical past is characterized by a 1Bof
0.9525,dividends as described in the skewed-dividend case, and a partici-
pation rate of 50%. Income processes are heterogeneous as described in
the previous subsection, so that 50% of stockholders have highly volatile
income that is correlated with dividends. The risk aversion a is fixed at
5, and r is fixed at 0.2. The stylized present is described by a 3 of 0.9625,
reflecting an upward revision of expected life expectancy, low-volatility
dividends as in Table 6 reflecting a considerable increase in diversifica-
tion, and a participation rate of 80%. All else is as in the past. This results
in a risk-free rate that rises from 0.17% to 3.73%, and an expected return
on stocks that decreases from 6.82% to 4.84%. The equity premium is
substantially reduced, from 6.65% to 1.11%. We conclude, then, that
assuming reasonable changes in a number of variables simultaneously
can account for changes in expected returns in keeping with what ap-
pears to be the case in the U.S. economy.
5. Conclusions
In this paper, we have looked at a number of potential fundamentals-
based explanations for the recent stock price run-up. In particular, we
focused on whether changes in market participation patterns or changes
in portfolio diversification are likely to account for a substantial fraction
of the rise in stock prices. We conclude that the changes in participation
that have occurred over this decade are unlikely to be a major part of the
explanation. This conclusion is based both on the data, which suggest
only small changes in participation for wealthy households, and the
model, which implies that participation changes have to be quite ex-
treme to substantially affect expected returns. Increased portfolio diversi-
fication, however, is likely to have had a larger effect. There is empirical
evidence that households have significantly diversified their portfolios,
selling individual stocks and buying mutual funds. An important differ-
240 * HEATON & LUCAS
ence between poorly diversified portfolios and a market index is the

likelihood of catastrophic outcomes. When this difference is reflected in
model parameters, the expected equity premium falls by more than 4%.
More generally, we can construct scenarios that are loosely consistent
with the data in which the required return on stocks falls by 2%. As shown
in Section 3.1 using a calibrated Gordon growth model, this amount of
change in expected returns goes at least halfway towards justifying the
current high level of the price-dividend ratio in the U.S. market. We
interpret this as quite a positive result, especially because it is difficult to
produce much variation in the predicted equity premium in this class of
models. The model also predicts an increase in the real risk-free rate,
which also appears to be consistent with the data.
These results depend in an important way on changes in diversifica-
tion and, to a lesser extent, on income heterogeneity. There is evidence
that entrepreneurs and managers tend to be large stockholders who bear
a sizable amount of undiversifiable risk in the form of their own busi-
nesses. Still, we do not have a complete picture of the income and
wealth characteristics of large stockholders, and we are uncertain about
the extent of their diversification. We also do not have a satisfactory
understanding of how older stockholders, who own a substantial frac-
tion of the market, view the risk of stock ownership. Looking even more
closely at the characteristics of large stockholders remains a useful direc-
tion for future research.
REFERENCES
Allen, E, and D. Gale. (1994).Limitedmarketparticipationand volatility of asset
prices. AmericanEconomicReview84:933-955.
Bakshi, G., and Z. Chen. (1994). Baby boom, population aging, and capital
markets.Journalof Business67:165-202.
Basak, S., and D. Cuoco. (1998). An equilibrium model with restricted stock
marketparticipation.Reviewof FinancialStudies11:309-341.
Bertaut,C., and M. Haliassos. (1995).Why do so few hold stocks? TheEconomic
Journal, 105:1110-1129.
Blanchard, 0. (1993). Movements in the equity premium. Brookings Papers on
Activity2:75-138.
Economic
Blume, M., and S. Zeldes (1993). The structure of stockownership in the U.S.
Wharton.Manuscript.
Bodie, Z., R. Merton, and W. Samuelson (1992). Labor supply flexibility and
portfolio choice in a life-cycle model. Cambridge, MA: National Bureau of
Economic Research. NBER Working Paper 6.
Brealey, R., and S. Myers (1996). Principles of CorporateFinance, 5th ed. New York:
McGraw-Hill.
Campbell, J., and J. Cochrane (1998). By force of habit: A consumption-based
explanation of aggregate stock market behavior. Journalof PoliticalEconomy
107:205-251.
Stock Prices and Fundamentals*241
Cochrane, J. (1997). Where is the market going? Uncertain facts and novel theo-
ries. FederalReserveBankof ChicagoEconomicPerspectivesXXI(6):3-37.
Campbell, J., and R. Shiller. (1988). Stock prices, earnings, and expected divi-
dends. Journalof Finance 43:661-676.
Constantinides, G., J. Donaldson, and R. Mehra. (1998). Junior can't borrow: A
new perspective on the equity premium puzzle. University of Chicago. Manu-
script.
, and D. Duffie. (1996). Asset pricing with heterogeneous consumers.
Journalof Political Economy104(2):219-240.
Davis, S., and P. Willen. (1998). Using financial assets to hedge labor income
risks: Estimating the benefits. Princeton University. Manuscript.
Deaton, A. (1992). Savings and liquidity constraints. Econometrica59:1221-1248.
Epstein, L., and S. Zin (1989). Substitution, risk aversion and the temporal
behavior of consumption and asset returns: A theoretical framework. Econo-
metrica57:937-969.
Friend, I., and M. Blume. (1975). The demand for risky assets. AmericanEconomic
Review 65:900-922.
Gentry, W. M., and R. G. Hubbard. (1998). Why do the wealthy save so much?
Columbia University. Manuscript.
Heaton, J., and D. Lucas (1998). Asset pricing and portfolio choice: The impor-
tance of entrepreneurial risk. Northwestern University. Manuscript.
Hodrick, R. (1992). Dividend yields and expected stock returns: Alternative pro-
cedures for inference and measurement. Review of Financial Studies 5:357-386.
King, M., and J. Leape (1984). Asset accumulation, information, and the life
cycle. Cambridge, MA: National Bureau of Economic Research. NBER Work-
ing Paper 2392.
Lamont, 0. (1998). Earnings and expected returns. Journal of Finance 53:1563-
1587.
Lee, Charles M. C., and B. Swaminathan. (1999). Valuing the Dow: A bottom-up
approach. Cornell University. Manuscript.
Mankiw, N. G., and S. Zeldes (1991). The consumption of stockholders and non-
stockholders. Journalof Financial Economics29:97-112.
Mehra, R., and E. Prescott (1985). The equity premium: A puzzle. Journal of
Monetary Economics15:145-161.
Newey, W, and K. West (1987). A simple positive semi-definite heteroskedasti-
city and autocorrelation consistent covariance matrix estimator. Econometrica
55:703-708.
Polkovnichenko, V. (1988). Heterogeneity and proprietary income risk: Implica-
tions for stock market participation and asset prices. Kellogg Graduate School
of Management. Manuscript.
Poterba, J. (1993). Who owns corporate stock?: A report to the New York Stock
Exchange. MIT. Manuscript.
. (1998). Population age structure and asset returns: An empirical investi-
gation. Cambridge, MA: National Bureau of Economic Research. NBER Work-
ing Paper 6774.
, and A. Samwick. (1997). Household portfolio allocation over the life
cycle. Cambridge, MA: National Bureau of Economic Research. NBER Work-
ing Paper 6185.
Saito, M. (1995). Limited market participation and asset pricing. Department of
Economics, University of British Columbia. Manuscript.
242 *J0RGENSEN
Storesletten,K., C. Telmer,and A. Yaron.(1998).Asset pricingwith idiosyncratic

risk and overlappinggenerations. Carnegie Mellon University.Manuscript.
Vissing-J0rgensen,A. (1997).Limitedstock marketparticipation.Departmentof
Economics,MIT.Manuscript.
Weil, P. (1990). Nonexpected utility in macroeconomics.QuarterlyJournalof Eco-
nomics 105:29-42.
Comment
ANNETTE VISSING-J0RGENSEN
Universityof Chicago
1. Introduction
During the period 1995-1998 the U.S. stock market experienced four
consecutive years with real stock returns above 20%. Suppose as a rough
approximation that annual real log gross stock returns are normally dis-
tributed and independent over time. With a mean and variance of this
distribution equal to the historical values for the period 1871-1994, the
probability of observing four years of above 20% returns is 0.4%.1 The
high returns have come primarily from capital gains driving price-
dividend and price-earnings ratios to historical highs (the latest num-
bers from August 9 for the S&P 500 are P/D = 78.5 and P/E = 31.9). Thus,
even taking into account statistical fluctuation, it is becoming increas-
ingly unlikely that nothing has changed. The only period since 1871 with
as impressive returns was 1924-1928, with five years of above 20% real
stock returns. Over the three-year period following that event, real stock
returns averaged -15.4% annually.
In the present paper Heaton and Lucas ask whether the recent stock-
market boom can be explained by changes in economic fundamentals.
Three candidates are considered: changes in corporate earnings growth,
changes in consumer preferences, and changes in stock-market participa-
tion patterns. Participation is defined broadly as concerning both the
level of stock-market participation and the amount of diversification
among participants. Poor diversification is found to have large effects on
equilibrium returns in an overlapping generations exchange economy.
The main conclusion of the paper is that increased diversification by
itself can explain at least half of the increase in the adjusted P/D ratio.
This is an interesting finding, not only for interpreting the recent past
but also seen in the context of the literature on the equity premium
1. Using the data from Robert Shiller's home page, the 1871-1994 mean and standard
deviation of log(l + rt?ck'real)
are 0.067 and 0.17.
Comment 243
puzzle. Increased participation is found to have only small effects. For

readers who attended the presentation of the paper at the NBER
Macroeconomics Annual conference, I should mention that the part of
the paper which concerns diversification is new and thus was not dis-
cussed at that time.
My discussion focuses first in Section 2 on whether the increase in
diversification is sufficiently recent and sufficiently large for this to be
considered the main reason for the recent stock-market boom. In Section
3 I turn to the overlapping-generations model to address whether the
theoretical results regarding large effects of diversification and smaller
effects of participation are likely to be robust. Section 4 comments on the
authors' calibration of the Gordon growth model and contains current
and historical data for analyst earnings forecasts to determine if high
earnings growth expectations rather than lower required stock returns
could be driving the stock-market boom. Section 5 concludes.
2. Is theIncreasein Diversification
Largeand
RecentEnough?
In the overlapping-generations model calibrated by Heaton and Lucas a
shift from a three-stock portfolio to full diversification generates a decline
in the mean real stock return of 1.41 percentage points for participation
fixed at 50% (compare cases D and A in Table 8). This is in fact more than
needed to explain current valuation ratios according to my calculations
in Section 4 below. Should we conclude from this that increased diversifi-
cation is the main reason behind the stock-market boom? From an empiri-
cal perspective it would need to be established that diversification has in
fact increased from something close to the level of a three-stock portfolio
to close to full diversification and that the timing of the increase coincides
to a reasonable extent with the stock-market boom. The evidence pre-
sented below shows that the trend in diversification started long before
the recent stock-market boom. Thus if diversification is as important as
suggested, valuation ratios should have reached historical highs long
before the 1990s. P/D and P/E ratios have trended upward since the early
1980s, but much of this was a return to normal levels from very low values
in the beginning of the 1980s.2
2. In Vissing-J0rgensen (1998) I documented the upward trend in stock-market participa-

tion from around 6% of households in the beginning of the 1950s to around 41% in 1995.
It is too early to say whether the trend in participation has strengthened significantly
since then. It will be interesting to see the latest numbers when the 1998 Survey of
Consumer Finances becomes available. However, the increase in participation since 1995
would have to be dramatic, since the effect of increased participation is likely to be
244 *J0RGENSEN
Figure 1 STOCKOWNERSHIPSHARES,1952-1999, FLOWOF

FUNDS ACCOUNTS
1.00 -
0.90
0.80 -
Househods and nonpaf
0.70 organatns
0.60 Bankpersonatmusts
C and estates
a) R_, ofthevworld
0.50 -
Insu[ance ompanies
Q.
040 itepensionfuds,
-Prtiv defied benefit
Private pensionfunds,
0.30 d-e ned contruton
eand bcatgovemrnnet
0.20 - emr_ ,byee rret ent ds
0.10 - Muttjalfunds
0.00 - - r . .
1950 1955 1960- 1965 1970 1975 1980 1985 1990 1995 2000
Year
The values shown are for the end of the first quarterof the year, except for the last data point for the
split of privatepension plans between defined contributionand definedbenefits,which is forthe end of
1998.The dataarenot seasonallyadjusted.Mutualfunds include closed-endfunds. The category"bank
personaltrusts and estates"was added in 1969.Beforethis it was lumped togetherwith directowner-
ship by households and nonprofitorganizations.Four small categoriessumming to less than 1.5%of
the totalfor all yearsare left out of the graphfor simplicity,but areincludedin the numbersgiven in the
text. These are state and local governments, commercialbanking, savings institutions, and security
brokersand dealers.The measureof equityin the Flow of FundsAccountsis the totalU.S. stock-market
capitalizationincludingclosely held companies.
Figure 1 updates Table 5 of Poterba and Samwick (1995), which shows

the proportions of the stock market owned through various channels.
The source of the data is the Flow of Funds Accounts. Consistent with
Heaton and Lucas's Table 4, the share of stocks held through mutual
funds has increased over the period since 1995. However, direct stock
ownership has declined steadily throughout the period. The correspond-
ing increases are mainly in the shares for pensions and for mutual funds.
Stockholding by private and governmental pension plans increased
from a negligible share in the beginning of the 1950s to a maximum of
27.0% in 1986:1. It has been fairly stable since then. The upward trend in
stockholding through mutual funds started around 1982 after a slight
decrease in the 1970s. The increase was 8.9 percentage points from
1982:1 to 1995:1 and 4.9 percentage points from 1995:1 to 1999:1. The
mutual fund share for 1999:1 was 16.5%.
nonlinear, with a bigger effect of a given increase in participation at initially low partici-
pation levels. It seems more plausible that changes in participation have contributed to a
gradual trend in returns than that they are responsible for the recent boom.
Comment 245
Up to the beginning of the 1980s the growth in stockholding through

pension plans represented purchases by defined-benefit pension plans.
It is likely that stock purchases by governmental defined-benefit plans
represent a significant increase in risk sharing. The bearers of the invest-
ment risk in this case are the taxpayers. Thus stockholding by these
pension plans spreads risk over a broad group of households and thus
increases risk sharing.
Since defined-benefit plans are managed by investment professionals,
one would expect them to be well diversified. It is however not clear that
increased stock ownership by private defined-benefit plans has in-
creased the diversification level of the typical stockholder to a significant
extent. Consider a world with workers and capitalists where stocks are
in unit net supply, and bonds are in zero net supply. Initially workers
save for retirement out of wages. Workers do not like risk and save in the
form of bonds issued by capitalists (directly or indirectly through com-
pany debt). Capitalists bear all output risk. A defined-benefit plan is
then introduced. Workers must accept a reduction in wages in exchange
for the pension benefits. The shareholders of each company take this
part of wages and invest it in other companies. They pay workers a
riskless stream when retired. Thus capitalists still end up bearing all the
risk. Workers get riskless retirement benefits in either case. In the situa-
tion with a pension plan each shareholder is more diversified: He still
only owns a small number of stocks directly, but now there is cross
ownership of stocks by companies via the pension plans. Buying one
share of a given company now gives you the right to a payment stream
representing partly this company's earnings and partly other companies'
earnings. However, with cross holding at most equal to the share of
private benefit plans in total stock-market capitalization, this effect is
small.
Since the introduction of individual retirement accounts (IRAs) in 1981
and 401(k) plans in 1978, there has been a shift from defined-benefit to
defined-contribution plans. In these, individual beneficiaries fully or par-
tially decide how to invest their assets. The increased share for 401(k)s
and similar plans represents a significant increase in diversification,
since these plans typically offer choices of stock portfolios rather than
allowing employees to pick individual stocks.3
Overall, while the effects are hard to quantify, increased stock owner-
ship by pension plans most likely contributed to increased diversifica-
tion and risk sharing long before the recent stock-market boom.
The share of stocks held directly remains large. Thus it is important to
3. Since 1993, 401(k) providers have been required by law to include a broad range of
equity funds in the investment choices. Large holdings of own-company stock remain
an issue for diversification of 401(k) stockholdings.
246 - J0RGENSEN
consider whether diversification has increased significantly for stock-

holders who hold all or most of their stocks directly. Table 1 gives various
measures of diversification based on data from the Survey of Consumer
Finances (SCF) for 1983, 1989, and 1995. The numbers for 1971 are from
Blume and Friend (1978, Chart 2-5) and are based on a sample of 17,056
federal income tax forms. There is a clear trend towards increased diversi-
fication of directly held stocks. The share of directly held equity which is
held by households with less than 10 stocks has decreased from 56.5% in
1971 to 37.9% in 1995. For 1989 and 1995 the SCF contains information
about how much equity is held in indirect form (mutual funds, pension
plans, trusts, and managed investment accounts). If we assume that all
such stockholdings are well diversified and that households with 20 or
more directly held stocks are well diversified, then 73.8% of household-
owned equity is owned by well-diversified investors, up from 60.8% in
1989.
Counting the number of directly held stocks overstates the level of
diversification if portfolios are unbalanced. Blume, Crockett, and Friend
(1974) found this to be important. Even for high-income households
with on average 18.7 different stocks, the level of diversification only
corresponded to an equal-weighted portfolio of about two stocks. The
SCF contains information about holdings of stock in the company where
household members work or have worked (I refer to these as own-
company stock). The bottom part of Table 1 shows that own-company
stockholding is likely to be a main cause of poor diversification. House-
holds with positive holdings of own-company stock owned 40.2% of
directly held equity in 1995. Of their direct stockholdings the mean
percentage held in own-company stock was 47.8%. Even if one allows
for indirect stockholding of these households, they still on average hold
30.8% of their equity portfolio in their own company [this number does
not include own-company stockholdings via 401(k) or similar plans].
This suggests that a substantial share of the stock market remains owned
by poorly diversified households. It furthermore emphasizes that under-
standing why so many households hold substantial amounts of wealth
in own-company stock is crucial for understanding the effects of poor
diversification. Are the results driven by rich households holding large
shares of companies they founded? Is delaying payment of capital gains
taxes a key reason they do not sell part of these stocks and invest in a
more diversified portfolio? Do rich households choose to hold large
shares of few companies to have influence on company decisions? Or
are people simply overly optimistic about their own company, in which
case poor diversification would not warrant higher returns? A better
understanding of these issues is crucial for determining the general equi-
librium effects on asset returns caused by poor diversification.
Comment 247
Table 1 TRENDS IN DIVERSIFICATIONOF DIRECTLYHELD EQUITYa
Number
of stocks 1971 1983 1989 1995
1. Percentage of equity held directly - 59.5 41.4

2. Percentage of equity held in mu- m- - 9.3 20.2
tual funds
3. Percentage of directly held equity 3 18.3 18.0 15.9 15.2
owned by households
with less than this number of 5 - 26.1 24.4 22.8
stocks
10 56.5 48.9 44.3 37.9
15 - 60.9 55.3 51.7
20 74.9 71.9 65.4 64.9
4. Percentage of equity held by 3 - - 91.4 94.1
households with half or
more of their equity holdings in 5 - - 86.0 91.3
indirect form or at
least this number of stocks 10 74.0 85.0
15 66.8 79.6
20 60.8 73.8
5. Own-company stock as percent- 23.4 17.1 19.2
age of directly held equity
6. Own-company stock as percent- - - 10.2 7.9
age of directly and
indirectly held equity
Households with positive holdings of
own-company stock:
7. Percentage of directly held equity - 36.0 31.0 40.2
owned by these households
8. Percentage of directly and indi- ~- - 23.1 25.7
rectly held equity owned
by these households
9. Mean percentage of own- 65.0 55.3 47.8
company stock in direct equity
portfolio for these households
10. Mean percentage of own- - - 44.1 30.8
company stock in direct and
indirect equity portfolio for these
households
Note: For the numbers based on the Survey of Consumer Finances (SCF) observations are weighted
using SCF weights. For 1989 and 1995 the numbers shown are averages of the numbers obtained for
each of the five SCF imputations. For 1983 the edited and imputed SCF data file is used. "Indirect
stockholding" refers to stockholding in mutual funds (half of holdings in combined mutual funds are
assumed to be equity), in IRAs, in thrift-type plans as defined in the SCF net-worth program, and in
trusts, annuities, and managed investment accounts. In line 9 values are weighted by size of direct
stockholdings; in line 10 values are weighted by size of direct and indirect stockholdings.
"Taxreturn sample 1971, Survey of Consumer Finances 1983, 1989, 1995.
248 JJ0RGENSEN
In sum, the empirical evidence raises two concerns for the theory
that the recent stock-market boom is due to increased diversification:
firstly, that the trend in diversification started much earlier than 1995;
secondly, that although diversification has improved, the share of eq-
uity owned through mutual funds is still only 16.5% and a substantial
share of the stock market remains owned by poorly diversified house-
holds. It would be interesting to see how large effects on returns
Heaton and Lucas's model generate for changes in diversification more
in line with this.
3. Robustness ResultsfromtheOLGModel
of Theoretical
I was surprised by the way the authors calibrate the poor-diversification
cases (cases B-D). They first use firm-level return data to determine the
effects of holding a larger number of stocks on the standard deviation of
a portfolio (Table 7). The amount of idiosyncratic dividend risk (and for
case D bankruptcy risk) is then chosen such that the model generates a
stock return volatility equal to that observed for a typical three-stock
portfolio. For this amount of volatility the expected stock return and the
equity premium are much higher than in the full-diversification case
(case A). But how do we know whether this is a reasonable amount of
idiosyncratic dividend and bankruptcy risk? This could be checked
against the firm-level data. In fact, a more standard approach would be
to first use the firm-level data to determine the dividend risk and the
bankruptcy probability for a typical firm, then assume that a portfolio of
three such stocks was bought, and determine if the model generates a
stock return (a return to such a portfolio) which is much higher than the
one for full diversification.
It should also be emphasized that the assumption of risk aversion
equal to 5 may be crucial for the large effects of diversification on the
mean stock return. If risk aversion were set to 1, there would most likely
be little or no effect on the mean stock return of either diversification or
participation. To see this, consider the following special case of the
model for which a simple closed-form solution for the stock price is
available. Suppose there is no idiosyncratic labor income risk and no
labor income when old. With risk aversion set to one (and thus equal to
the assumed elasticity of intertemporal substitution), Epstein-Zin prefer-
ences specialize to CRRA preferences. Then each young agent consumes
the constant fraction a = P3/(1+ ,3) of wages, independent of asset re-
turns. Let A denote the proportion who are stockholders, wt their portfo-
lio share for stocks at time t, and P,t the stock price at t. The equilibrium
conditions for the stock and the bond market at t are then as follows.
Comment 249
Stocks: Awt(l - a)(l - dt)Yt= Pst, (1)

Bonds: A(1 - co)(1 - a)(1 - dt)Yt + (1 - A)(1 - a)(l - dt)Yt =0. (2)
Equation (2) implies Awt= 1. Inserting this in (1) gives Pst = (1 - a)(l -
dt)Yt.Thus in this special case the stock price is unaffected by the level of
stock-market participation. Furthermore, the stock price is affected by
diversification only because this is modeled by a stochastic aggregate
dividend share (causing the wage share to be stochastic) rather than
using several different stocks. In other words, for this special case in-
creased participation [diversification] affects the equity premium only
[mainly] via the bond rate, with no [little] effect on stock price and the
stock return. Given this, I would expect much smaller effects of diversifi-
cation on the mean stock return if risk aversion were set equal to 2 or 3
rather than 5 (in the end the right number may turn out to be 5 or higher,
but given that we do not have precise knowledge about this parameter,
sensitivity analysis is relevant).
The underlying reason that the stock price is unaffected by participa-
tion or diversification in the log utility case is that the model is an
exchange economy. With log utility the propensity to save is the same for
all households. The bond market therefore requires that stockholders in
equilibrium be willing to lend to the nonstockholders as much as stock-
holders wish to save. This implies Awt = 1 and thus, along with a con-
stant, leaves the stock price to be determined by the wage income of the
young. In an exchange economy wages are exogenous to both participa-
tion and diversification. This suggests that an alternative way of generat-
ing a higher stock price upon entry or diversification, even in the log
utility case, is to change the model to one in which the resources of the
young can be affected by increased participation or diversification. In
Vissing-Jorgensen (1998) I analyze an OLG model with production to
study the general equilibrium effects of limited participation. In that
model (for the log utility case) the riskless rate is unaffected by participa-
tion, and the full effect on the equity premium is due to a lower mean
stock return. Wages, the capital stock, and the stock price are higher for
higher levels of participation. I recalibrated the model to have similar
amounts of output risk to Heaton and Lucas's low-dividend-volatility
case. Increasing participation from 10% to 60% then decreases the mean
stock return by around 0.5 percentage point.4
4. The model is fairly standard. The results are not sensitive to whether the production
function is assumed to exhibit constant or decreasing returns to scale. One assumption
which is central for the results is that in each period, factor input levels and wages are
set before the realization of uncertainty. Factors are paid after output is realized. Thus
the labor share of output is countercyclical, since workers do not take any of the output
risk. Countercyclical labor shares are well documented in the business-cycle literature.
250 *J0RGENSEN
4. TheGordonGrowthModel
Although I agree that a decrease in the required return is needed to
explain recent valuation ratios, the authors' calculation based on the
Gordon growth model to some extent overstates the necessary change.
The required stock return in the formula P/E = l/(r - g) is net of transac-
tion costs, and these have declined significantly. While it is hard to evalu-
ate costs of direct investment, Bogle (1991) finds that equity mutual funds
underperformed the S&P 500 by an average of 2.1 percentage points over
the period 1969-1989. Rea and Reid (1998) find that the sales-weighted
average of total shareholder costs for equity mutual funds has decreased
from 2.25% in 1980 to 1.49% in 1997. Indeed, declining transaction costs
both for direct investment and for investment via mutual funds are likely
to have been a key factor behind the increases in diversification and
participation (the issue of lower transaction costs does not arise in the
overlapping-generations model, since diversification or participation is
changed exogenously). Assuming a 0.75-percentage-point decline in
transaction costs, the change in r - g needed to imply a movement in the
adjusted P/D ratio from 28 historically to 48 at the end of the authors'
sample is not 2 - 1 = 0.015 but 0.0075, or 0.75 percentage points. It is
worth pointing out in this context that without transaction costs it is very
difficult to reconcile the Gordon growth model with the historical-mean-
adjusted P/D ratio. With a historical value of g around 2% the model
implies a historical required real stock return of 2 + 0.02 = 0.056. The
actual real stock return was much higher at 8.5% (arithmetic average) for
1871-1994, and 9.1% if we include the recent period up to 1998.
The authors' calibration of the Gordon growth model furthermore
assumes that the riskless rate has increased by 2 percentage points.
Therefore an increase in g of 3.5 percentage points (or a g of 13.4% for 10
years and 2% thereafter) is required to explain an adjusted P/D of 48 with
a constant equity premium. The most relevant interest rate in this con-
text is the real interest rate on long-term bonds. These rates are currently
high (around 4% for long-term inflation-indexed U.S. Treasury bonds),
but it may be premature to conclude that they are as much as 2 percent-
age points higher than their historical mean. Blanchard (1993) and the
discussion of it by Siegel show that fluctuations in long-term real bond
rates have historically been quite dramatic.
As for the dividend growth g, it has in fact been higher than its histori-
cal average lately. The geometric and arithmetic averages were 1.5% and
2.4%, respectively, for 1871-1994, but 3.9% and 4.0%, respectively, for
1995-1998.5 An alternative to considering the recent past is to look at
5. The Gordon growth model assumes that g is nonstochastic and thus does not recom-
mend whether to use geometric or arithmetic means. The numbers for real earnings
Comment*251
EARNINGSGROWTHFORECASTS,
Figure2 I/B/E/STWO-YEAR-AHEAD
1982-1999
40.00
P/E
30.00
20.00
Anayst forecast of
10.00
^~~--^
/y/~
^\ ~I^~ J
/~
e/
\ \\ real earnings groth,
e/ p ye
000
Actual real earnings grrth,

-10.00 - / peroetperyear
-20.00
-30.00 1 . .
1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000
Year
forecasts from market participants. It is well known that analyst earnings

forecasts tend to be upward biased. Therefore it is useful to consider
earnings forecasts for which historical data are available and focus on
whether forecasts are higher than usual. Figure 2 shows I/B/E/S forecasts
for two-year-ahead S&P 500 earnings growth for 1982-1999 as well as the
subsequent realization and the P/E ratio at the time of the forecast.
Forecasts for long-run growth were only available for a smaller number
of analysts. The forecasts shown are top-downforecasts. This means that
the analysts were asked for a single forecast for the index rather than
forecasts for each of the companies which make up the index. The latest
bottom-upforecasts for S&P 500 earnings growth are much higher than
the top-down forecasts, but I do not have a time series to determine if
they are higher than their previous values. The average number of ana-
lysts reporting per year is 10. The minimum number is 4. The forecasts
plotted are the means across analysts. For each year the values are for
the first month after the previous annual earnings realization is known,
usually March. Thus the forecast value for 1999 shows the expected
percentage increase in year 2000 earnings over 1998 earnings, right after
1998 earnings became known. The 1999 forecasts are from March. The
analysts provide nominal earnings forecasts but no inflation forecasts.
To convert the forecasts to real terms, I used the annual inflation rate
over the previous five-year period.
growth are as follows. Geometric means: 1.9% for 1871-1994, 4.0% for 1995-1998.
Arithmetic means: 5.0% for 1871-1994, 4.1% for 1995-1998.
252 *J0RGENSEN
Several points are worth noticing. First, analyst earnings growth fore-
casts are quite good. The R2 from a regression of the realized earnings
growth rates on the forecasts is 0.59 for the real-earnings growth rates
and 0.52 for the nominal-earnings growth rates. Second, until 1995 the
correlation between the P/E ratio and the two-year-ahead real-earnings
growth forecast is surprisingly high, 0.87. But third, this correlation
breaks down after 1995. Earnings growth forecasts have stayed essen-
tially constant while the P/E ratio has increased sharply. Thus, if the
expectations of the analysts asked by I/B/E/Sare representative of current
market expectations, it looks like the stock market boom since 1995 ei-
ther is driven by a sudden decrease in required returns or is a bubble.
In sum, the required change in r - g to explain an increase in the
adjusted P/Dfrom 28 historically to 48 recently is around 0.75 percentage
points. Dividend growth and (geometric) earnings growth has been
higher since 1995 and thus might warrant an increase in the expected
dividend growth rate. However, at least according to one source, market
participants have not increased their dividend growth expectations. If,
therefore, a change in the required real stock return is left as the sole
factor explaining the increase in valuation ratios, the necessary change is
around 0.75 percentage points. For given long-term real bond rates the
necessary change in the equity premium is of the same size. If we believe
long-term real bond rates will be higher in the future, the required de-
crease in the equity premium is correspondingly larger.6
5. Conclusion
Heaton and Lucas address an important but difficult question: What
caused the recent stock-market boom? They focus on changes in stock-
market participation and diversification. Having worked on limited
stock-market participation, I found the analysis of the related issue of
diversification very interesting. The references given by the authors and
the numbers in Table 1 above indicate that poor diversification is in fact a
pervasive phenomenon which should be considered seriously in general
equilibrium asset pricing models. Understanding why many households
concentrate large amounts of wealth in own-company stock seems cru-
cial in this respect.
More work is needed to determine exactly how large the effects on
6. The latest P/D ratio of 78.5 is higher than the value 48 used by Heaton and Lucas,
suggesting that larger changes in r -g are needed. This depends on how the P/D ratio is
adjusted. Campbell and Shiller (1998) refer to studies suggesting adjustments to D/P of
80 basis points for 1996 and 1997. This would make the latest adjusted P/Dequal to 48.2,
close to the value 48 used in Heaton and Lucas's calibration.
Comment 253
equilibrium returns of poor diversification are. In the calibration of the

OLG model in the present paper, a concern is whether the amount of
idiosyncratic dividend and bankruptcy risk is consistent with the data.
More analysis regarding sensitivity of the results to changes in risk aver-
sion and the extent of poor diversification would also be useful. From an
empirical perspective any explanation of the current boom which relies
on changes in either participation or diversification will have difficulties
with timing. The upward trend in both participation and diversification
started long before the current boom, suggesting that valuation ratios
should have reached historical highs much earlier. However, valuation
ratios have historically fluctuated substantially, making it difficult to dis-
cern gradual trends. Aside from patiently awaiting more data for the
United States, it would be interesting to consider evidence from other
countries.
REFERENCES
Blanchard,O. J. (1993). Movements in the equity premium. BrookingsPaperson
Economic Activity1993(2):75-118,137-138.
Blume, M. E., and I. Friend. (1978). TheChangingRoleof theIndividualInvestor:A
TwentiethCenturyFundReport.New York:Wiley.
Blume, M. E., J. Crockett,and I. Friend. (1974). Stock ownership in the United
States:Characteristicsand trends. Surveyof CurrentBusiness54(11):16-40.
Bogle, J. C. (1991).Investing in the 1990s.JournalofPortfolioManagement,17(3):5-
14.
Campbell,J. Y, and R. J. Shiller. (1998).Valuationrates and the long-run stock
marketoutlook. Journalof PortfolioManagement, 24(2):11-26.
Poterba, J. M., and A. A. Samwick. (1995). Stock ownership patterns, stock
market fluctuations, and consumption. BrookingsPaperson EconomicActivity
1995(2):295-357,368-372.
Rea, J. D., and B. K. Reid. (1998).Trendsin the ownership cost of equity mutual
funds. InvestmentCompanyInstitutePerspective 4(3).
Vissing-J0rgensen,A. (1998).Limitedstock marketparticipation.Departmentof
Economics,MIT.Ph.D. Thesis.
Comment
JOHN Y. CAMPBELL
Harvard University
1. Introduction
The dramatic bull market of the late 1990s has challenged economists to
explain why stock prices are so high relative to historical valuation lev-
els. John Heaton and Deborah Lucas begin their interesting paper by
254 *CAMPBELL
reviewing some facts and then using theory to interpret them; I shall
organize my discussion in a similar fashion.
2. HowHighIs theStockMarket?
Popular commentary often uses the Dow Jones Industrial Average index
(around 11,000 as I write) or the Standard and Poor 500 index (around
1,300) to track the level of stock prices. Of course, index levels can
increase because of general price inflation, or growth in the real econ-
omy, or changes in the size of the publicly traded corporate sector rela-
tive to the economy, or changes in the size of index-included firms rela-
tive to other publicly traded firms. Intelligent analysis of stock index
levels must begin by scaling them in some way.
Recognizing this point, Heaton and Lucas discuss price-dividend and
price-earnings ratios for the S&P 500 index. Both ratios are high relative
to historic norms, but the price-dividend ratio is far more extreme; it is
almost two-thirds higher than its previous peak in the early 1970s,
whereas the price-earnings ratio is close to levels reached earlier in this
decade and in several previous decades.
Heaton and Lucas focus on the price-earnings ratio (scaled by the
historical average payout ratio of dividends to earnings) rather than
the price-dividend ratio. They claim that "earnings are likely to be a
more stable proxy for long-run payments to shareholders" (Section 3.1)
and that "in the short run dividends can vary due to temporary
changes in payout policy (for instance, in response to changes in the
tax law). Therefore, it is common to focus on the price-earnings ratio,
adjusted for reinvestment rates, to approximate long-run price-
dividend ratios" (footnote 5).
It is certainly true that changes in corporate financial policy can affect
the price-dividend ratio. Most notably, a shift from paying dividends to
repurchasing shares can permanently increase the price-dividend ratio.
The Gordon growth model, discussed in the paper, says that the price-
dividend ratio is the reciprocal of the difference between the discount
rate and the growth rate of dividends per share. A share repurchase
program causes the number of outstanding shares to shrink over time;
this increases the growth rate of dividends per share and increases the
price-dividend ratio. Share repurchases account for some of the increase
in the price-dividend ratio over the last decade, although direct esti-
mates of the effect are fairly modest. Cole, Helwege, and Laster (1996),
for example, suggest that net repurchases have increased the growth
rate of dividends per share by about 0.8%. Their calculation assumes
that shares are issued and repurchased at the market price; to the extent
Comment 255
that shares are issued at below-market prices as part of executive com-

pensation, then the true repurchase effect is smaller.
Despite these difficulties with the price-dividend ratios, I do not agree
that the price-earnings ratio is a superior measure of stock-market valua-
tion. The problem is that earnings are subject to short-term noise arising
from the business cycle. One can see the importance of this by inspect-
ing Figure 1 in the paper. Previous peaks of the price-earnings ratio,
close to levels today, were reached in the early 1990s, the mid-1930s, the
early 1920s, and the 1890s. None of these were peaks in stock prices;
instead, they were recession years when corporate earnings temporarily
declined.
The issue of noise in current earnings has been recognized at least
since the work of Graham and Dodd (1934), who in their famous text-
book SecurityAnalysis recommended that analysts should use an average
of earnings over "not less than five years, preferably seven or ten years"
(p. 452). Campbell and Shiller (1998) follow Graham and Dodd's advice
and smooth earnings over ten years. They find that the ratio of price to
smoothed earnings behaves more like the price-dividend ratio than like
the conventional price-earnings ratio. It is currently far above its previ-
ous peak reached in 1929.
Heaton and Lucas use the Gordon growth model, adjusting the cur-
rent price-earnings ratio for the long-run average payout ratio of divi-
dends to earnings, to characterize combinations of earnings growth rates
and discount rates that could rationalize the current level of stock prices.
They conclude that real earnings growth of 2.4% (1% above the historical
average) and a real discount rate of 6.6% (4.1% below the historical
average) could do the job. In the rest of the paper, they use alternative
theoretical models to try to hit this target.
The problem with this analysis is that the cyclical noise in earnings
should lead earnings growth forecasts to be adjusted downwards at
cyclical peaks when earnings are temporarily high, and upwards at cycli-
cal troughs when earnings are temporarily low. Rapid earnings growth
from a starting point in 1999, after many years of robust economic
growth, is less likely than Heaton and Lucas admit.1 Heaton and Lucas
could correct for this problem by using the price-smoothed-earnings
ratio instead of the conventional price-earnings ratio.
1. One factor that can produce higher long-run earnings growth is a reduction in the
payout ratio. As Heaton and Lucas point out, the earnings growth rate should be the
fraction of earnings that is retained (one minus the payout ratio) times the return on
equity. If the payout ratio falls, earnings growth should be expected to increase. Unfortu-
nately this effect also increases Heaton and Lucas's adjusted price-earnings ratio, so it
does not make it easier to account for the level of stock prices.
256 *CAMPBELL
Even though rapid earnings growth following a period of strong eco-

nomic performance would be historically unusual, some commentators
do appear to believe that it will occur. Interesting evidence on this point
is provided by Steven Sharpe (1999). Sharpe studies the consensus fore-
casts of stock analysts, and finds that since 1994 forecasts of two-year
nominal earnings growth have been high and stable (between 10% and
15%), even though realized two-year earnings growth has been declin-
ing. He also finds that forecasts of long-term (five-year) nominal earn-
ings growth have increased from 10.5% in 1989 to over 13% in 1998.
Over the same period forecasts of long-term (ten-year) inflation have
decreased from 4.5% to 2.5%, implying a remarkable increase of 4.5
percentage points in the expected long-run growth rate of real earnings.
Of course, analysts' earnings forecasts are hard to interpret. It may be
that they reflect a rational assessment of the prospects for a "new era" of
corporate profitability in the twenty-first century. It may be, as Sharpe
suggests, that analysts have failed to adjust their nominal earnings fore-
casts for the effects of declining inflation and thus are subject to a form of
money illusion first proposed by Modigliani and Cohn (1979). Finally, a
cynic might say that Wall Street analysts do not have incentives to pro-
duce the most accurate earnings forecasts, but rather to produce fore-
casts that justify the current level of stock prices.
3. ModelingDecliningDiscountRates
While reasonable people can disagree about the prospects for future
earnings growth, it is almost impossible to rationalize the current level of
stock prices without some decline in the discount rate applied to inves-
tors to future earnings. Heaton and Lucas devote most of their paper to
an exploration of alternative mechanisms that could produce such a
decline. They rightly concentrate on effects that could reduce the equity
premium (the expected excess return on equities over short-term debt),
since real interest rates have not historically moved closely with the
stock market.
Heaton and Lucas first consider an increase in the stock-market partici-
pation rate. Intuitively, if aggregate equity risk is now shared more
broadly, then the amount of risk borne by any single investor has de-
clined, justifying a decline in the equity premium. In thinking about this
effect, it is important to keep in mind that investors should be weighted
by their wealth. The right measure of the participation rate is not the
fraction of individuals who invest in stocks, but the fraction of wealth
controlled by individuals who invest in stocks. As Heaton and Lucas
admit, wealthy individuals have always tended to participate in the
Comment*257
stock market, so there is little evidence for a dramatic increase in the

wealth-weighted participation rate.
Heaton and Lucas take the participation rate as exogenous, determined
by unmodeled forces such as transaction and information-processing
costs. They build a fairly realistic, but correspondingly complicated,
model to explore the effects of the participation rate on the equity pre-
mium. Unfortunately they find it very hard to generate a large equity
premium when the participation rate is above 30% or so. The reason for
this is hard to see in their model, but Gollier (1999) suggests a simpler
framework that can be used to gain insight.
Gollier assumes a static atemporal market in which a claim to random
output y is traded for a riskless claim. The price of the output claim is P
Agents have utility u over final wealth and choose the portfolio share of
the output claim, a, to maximize
V(a) = E[u(P + a(y -P))]. (1)
The first-order condition is
E[(y - P)u'(P + a(y - P))] = 0. (2)
Equilibrium requires that the total supply of the output claim (normal-
ized to one) be held. When all agents participate in the financial market,
this requires a = 1, or
E[(y -P)u'(9)] = 0. (3)
If only a fraction k of wealth is controlled by agents who can hold equity,

however, then for these agents equilibrium requires a = 1/k, so we get
0- (4)
E[- P)' (k ( k) )]
Gollier calibrates these equations to data on real per capita output in

the United States over the period 1963-1992. Consistent with the results
of Heaton and Lucas, he finds little effect of the participation rate k on
the expected return of the output claim for RRA = 2 and k > 0.3.
To understand the source of this result, I now take a second-order
Taylor approximation of marginal utility around the mean of output, g:
u'(9) = u'(y) + u "(y)( - y) + 'u"'()(Y - )2. (5)

258 *CAMPBELL
Substituting into (3), assuming that y has a symmetric distribution, and

assuming constant relative risk aversion y, I find that with full equity
participation (k = 1), the expected return on equity is
1
- 1= - 1 y, (6)
P 1 - y~[1 + y(y + 1)2/2]
where o2 Var(Q)2 is the proportional volatility of output, and the

second approximation is accurate for small o2. This can be understood by
recalling the well-known rule of thumb that the optimal portfolio share
in a risky asset is the expected excess risky return, divided by relative
risk aversion times the variance of the excess risky return.2 To achieve an
optimal portfolio share of one, the expected excess risky return must
equal relative risk aversion times the variance.
Similar analysis of the case with limited participation (k < 1) shows
that in general,
9 - yo2
1- . (7)
P k
Limited participation by investors who control a fraction k of wealth is

equivalent to scaling up the variance of dividends by a factor 1/k. Once
again, this can be understood by using the rule of thumb for optimal
risky investment. To achieve an optimal risky portfolio share of 1/k, the
expected excess risky return must be 1/k times larger than it would be if
the optimal risky portfolio share were only one.
Equation (7) has two important implications. First, a change in equity
participation has a larger effect on the equity premium if the participa-
tion rate is initially low than if it is already high. A doubling of participa-
tion from 5% to 10% cuts the equity premium in half in just the same
way as a doubling from 50% to 100%; and the absolute change in the
equity premium is much larger in the former case. This explains why
both Gollier, and Heaton and Lucas in their more elaborate model, find
little participation effect for k larger than about 1. Second, limited equity
participation has a larger effect on the equity premium if relative risk
aversion and dividend volatility are high than if they are low. Limited
participation can amplify a high equity premium caused by high divi-
dend volatility or high risk aversion, but an unrealistically small k is
2. This rule of thumb is exact in a continuous-time model in which the risky asset's price
follows a geometric Brownian motion (Merton, 1969). Friend and Blume (1975) used this
approach to estimate risk aversion.
Comment*259
required to produce a high equity premium in the absence of these

conditions.3
Given their finding that increases in participation from medium to
high levels have little effect on the equity premium, Heaton and Lucas
emphasize an alternative story. They argue that the typical investor used
to hold a poorly diversified portfolio containing only a few stocks. With
the growth of mutual funds and especially index funds over the last few
decades, however, the typical investor is now better diversified. Diversi-
fication makes equities a more appealing investment by reducing the risk
associated with any given average return. Heaton and Lucas show that a
simultaneous increase in participation and reduction in equity risk can
account for a large decrease in the equity premium. In terms of equation
(7), Heaton and Lucas simultaneously reduce o2 (by a factor of 4) and
increase k (by a factor of 2) to get a much more powerful effect on the
equity premium than can be achieved by a change in k alone.
Heaton and Lucas also argue that an undiversified portfolio is likely to
have a negatively skewed return because any single firm can go bank-
rupt. They find that negative skewness further increases the equity pre-
mium. To understand this effect within the simple framework presented
above, one can drop the assumption that 9 has a symmetric distribution.
This adds a term -y(y + 1)SK/2k2, where SK is the proportional
skewness of y, to the equity premium in equation (7). Negative skewness
increases the equity premium, and this effect is more powerful when
stock-market participation is limited.
Although many investors are undoubtedly better diversified today, I
doubt that this is the cause of a major decline in the equilibrium equity
premium. The problem is that diversification, like equity participation,
should be measured on a wealth-weighted basis. Most stocks have al-
ways been held by wealthy investors who are more likely to diversify
their holdings. Even if the typical portfolio has been undiversified, the
typical share of stock is likely to have been held in a diversified portfolio.
Increased diversification by small investors need not have a large effect
on equilibrium asset prices.
Furthermore, diversification can only have had a large impact on the
equity premium if the gains from increased diversification were histori-
cally large, certainly much larger than the direct costs of increasing the
number of stocks held in a typical portfolio. Thus the diversification
story creates a new puzzle-why were investors historically reluctant to
hold diversified portfolios?-and this seems little easier to resolve than
3. In a similar spirit, Campbell (1999) uses the results of Constantinides and Duffie (1996)
to argue that heterogeneous risk in labor income cannot have a large effect on the equity
premium unless risk aversion is high.
260 *CAMPBELL
the original equity-premium puzzle-why were investors historically

reluctant to hold equities?
Both the effects that Heaton and Lucas emphasize-increased partici-
pation and diversification-are long-run trends that may help to explain
why valuation ratios are higher now than they were in the early postwar
period, but do not specifically explain the runup in prices during the late
1990s. An important clue, ignored by Heaton and Lucas, is the fact that
this runup has occurred during a period of robust economic growth.
This is also characteristic of bull markets in previous decades such as the
1920s and the 1960s.
Campbell and Cochrane (1999) present a model of stock-market behav-
ior in which valuation ratios are driven entirely by cyclical variation in
consumption. Increases in consumption drive up risky-asset prices rela-
tive to dividends, not by increasing expected future dividend growth
(which is constant by assumption), nor by decreasing real interest rates
(which are also constant in the model), but by increasing the risk tolerance
of investors. Investors' preferences are assumed to display habitformation:
they have power utility whose argument is not the absolute level of con-
sumption, but the level of consumption relative to a subsistence level,
which is a nonlinear moving average of current and past consumption.
When consumption is close to the subsistence level, only a small fraction
of consumption is available as a surplus to generate utility, and even small
shocks to consumption can have a large effect on this surplus. In such
circumstances, investors become extremely risk-averse. As consumption
increases relative to the subsistence level, however, their risk aversion
declines and the equity premium is driven down.
The use of habit formation to generate time variation in the equity
premium is appealing because there are other reasons to think that
people judge their well-being by relative rather than absolute consump-
tion. For example, it is common to compare a recession period unfavor-
ably with a much earlier period of strong growth, even if the absolute
level of consumption is higher in the recession than in the earlier boom.
Habit formation explains this by the fact that surplus consumption,
which generates utility, may be lower in the recession.
One objection to Campbell and Cochrane's model is that it requires
high risk aversion to explain the historical average value of the equity
premium. Barberis, Huang, and Santos (1999) have recently proposed a
variant of the model in which investors are "loss-averse" (Kahneman and
Tversky, 1979). Investors derive utility from the level of wealth relative to
a reference point, which adjusts only gradually in response to changes in
wealth. Furthermore, there is a kink in preferences at the reference point:
the absolute value of marginal utility is higher for losses than for gains. In
Comment 261
periods of weak economic growth, investors' wealth is close to the refer-

ence point and the kink in the preferences at that point makes them
extremely risk averse. In cyclical expansions, however, wealth increases
far above the reference point and risk aversion declines. The Barberis-
Huang-Santos model uses loss aversion to generate high aversion to
wealth risk even with moderate aversion to consumption risk.
Both these models have an additional advantage relative to the frame-
work used by Heaton and Lucas. Because risk aversion varies in these
models, risky asset prices more relative to dividends and so the volatility
of stock returns can be much higher than the volatility of dividend
growth. This is a feature of the data that is not easily matched by models
with constant risk aversion. Heaton and Lucas do not report the volatil-
ity of stock returns in their constant-risk-aversion model, but it is proba-
bly close to the underlying volatility of dividend growth. Even though
Heaton and Lucas calibrate their model with greater dividend volatility
than has historically been observed, the model probably understates the
volatility of stock returns, and this makes it harder to generate a large
equity premium.
In my view cyclical factors of the sort emphasized by Campbell and
Cochrane and by Barberis, Huang, and Santos are at least as important for
stock prices as the secular changes in participation and diversification
emphasized by Heaton and Lucas. But in the end, it is important to
recognize that the recent runup in stock prices is so extreme relative to
fundamental determinants such as corporate earnings, stock-market par-
ticipation, and macroeconomic performance that it will be very hard to
explain using a model fit to earlier historical data. The relation between
stock prices and fundamentals appears to have changed, and it may be a
long time before a definitive interpretation of this change is possible. In
the meantime investors should keep in mind that a return to historical
valuation ratios would imply extremely large negative returns, while a
continuation of current ratios would imply mediocre returns unless there
is a historically unprecedented acceleration of corporate earnings growth.
REFERENCES
Barberis,N., M. Huang, and T. Santos. (1999).Prospect theory and asset prices.
Cambridge,MA: National Bureauof EconomicResearch. NBERWorkingPa-
per 7220.
Campbell,J. Y. (1999).Asset prices, consumption, and the business cycle. Forth-
coming in Handbookof Macroeconomics, J. B. Taylorand M. Woodford (eds.).
Amsterdam:North-Holland.
, and J. H. Cochrane. (1999). By force of habit: A consumption-based
explanation of aggregate stock market behavior. Journalof PoliticalEconomy
107:205-251.
262 *DISCUSSION
,and R. J. Shiller. (1998).Valuationratios and the long-run stock market

outlook. Journalof PortfolioManagement, Winter,pp. 11-26.
Cole, K., J. Helwege, and D. Laster.(1996).Stock marketvaluation indicators:Is
this time different?FinancialAnalystsJournal,May/June,pp. 56-64.
Constantinides, G., and D. Duffie. (1996). Asset pricing with heterogeneous
consumers. Journalof PoliticalEconomy104:219-240.
Friend, I., and M. E. Blume (1975). The demand for risky assets, AmericanEco-
nomicReview,65:900-922.
Gollier,C. (1999). TheEconomics of RiskandTime.Forthcoming.Cambridge,MA:
MITPress.
Graham, B., and D. L. Dodd. (1934). SecurityAnalysis, 1st ed. New York:
McGraw-Hill.
Kahneman,D., and A. Tversky (1979).Prospecttheory:An analysis of decisions
under risk, Econometrica, 47:313-327.
Merton, R. (1969). Lifetime portfolio selection under uncertainty:The contin-
uous-time case. Reviewof Economics andStatistics51:247-257.
Modigliani, E, and R. A. Cohn. (1979). Inflation, rational valuation, and the
market.FinancialAnalystsJournal,March-April,pp. 24-44.
Sharpe, S. (1999).Stock prices, expected returns, and inflation. Boardof Gover-
nors of the FederalReserve System. FEDSPaper99-02.
Discussion
In replying to the discussants, both authors agreed that increasing par-

ticipation rates alone could not explain the current level of stock prices.
John Heaton emphasized that much of the effect of increasing participa-
tion occurs as the economy moves from low to moderate participation,
but we are now moving from moderate to high participation. Deborah
Lucas expressed skepticism that adding capital accumulation to the
model would change this result. As an alternative explanation, which is
more fully developed in the published version of the paper, she noted
that stock investors are now holding more diversified portfolios, which
increases their aggregate risk-bearing capacity.
Mark Gertler asked whether there might be some benefit to studying
stock prices at a more disaggregated level. For example, there is a great
deal of variation among stocks in price-earnings ratios, with internet
stocks like Amazon.com at the upper extreme. John Campbell remarked
that explaining the pricing of "growth" stocks raises interesting issues:
One hypothesis is that such stocks are priced high because optimistic
investors with upward-biased earnings expectations are more likely to
hold them. Another story is that, as the discount rate falls for whatever
reason, growth stocks experience large effects because of the long aver-
age duration of their expected earnings streams.
Discussion 263
Martin Feldstein argued that it is important to incorporate tax consider-

ations. Because of their tax treatment, share buybacks are a much more
efficient way to pay out to individual shareholders. To the extent that
buybacks are becoming a more important share of payouts, net-of-tax
returns have increased.
Martin Eichenbaum asked whether participation is defined to include
holding stocks in a retirement account. Heaton responded that the contri-
bution data reflect holdings of defined-contribution retirement accounts,
and that the increase in such accounts may explain part of the measured
rise in participation. Jonathan Parker noted that some people who are
technically participants in the market at earlier times held only one or
two stocks, whereas today they might hold one or two well-diversified
mutual funds. This would support the idea that average diversification
rather than participation per se is what is important. As another example
of how financial innovation can reduce required yields, Michael Mussa
mentioned the "liquification" of the below-investment-grade bond mar-
ket by Michael Milken.
The identity of the "the marginal stockholder" was the subject of some
discussion. Julio Rotemberg suggested that the marginal stockholder
might be very rich and not very risk-averse. Heaton remarked that the
characteristics of stockholders' noninvestment income are important,
particularly the correlation of this other income with the market. For
example, if the marginal stockholder is a wage earner rather than the
owner of a business, the reduction in the required risk premium may be
greater, all else equal.
View publication stats

Stock Prices and Fundamentals: January 2000

Uploaded by

Copyright:

Available Formats

Stock Prices and Fundamentals: January 2000

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stock Prices and Fundamentals: January 2000

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Stock Prices and Fundamentals

Article · January 2000

The user has requested enhancement of the downloaded file.

Volume Title: NBER Macroeconomics Annual 1999, Volume 14

Volume Author/Editor: Ben S. Bernanke and Julio J. Rotemberg, editors

Volume Publisher: MIT

Volume ISBN: 0-262-52271-3

Volume URL: http://www.nber.org/books/bern00-1

Conference Date: March 26-27, 1999

Publication Date: January 2000

Chapter Title: Stock Prices and Fundamentals

Chapter Author: John Heaton, Deborah Lucas

Chapter URL: http://www.nber.org/chapters/c11048

Chapter pages in book: (p. 213 - 264)

Stock Prices and Fundamentals

Although the paper touches on a variety of issues, its main contribution

changes in stock-market participation patterns. The influence of partici-

Figure 1 PRICE-DIVIDENDRATIOAND PRICE-EARNINGSRATIO,

price-dividend ratio, this regression predicts a substantial decline in the

Coefficient Estimate StandardErrora

tical fluctuations, without any change in the underlying driving pro-

2. This valuation model, a staple of marketanalysts, is described, for instance, in Brealey

integration of world markets, continuing technological advances, and an

1. takesubstantialfinancial risks expectedto earn substantial returns

Table2 AVERAGERESPONSEBYAGE AND SURVEYYEARTO

trigger a large change in aggregate preferences over the course of only a

growth rate is constant. This assumption, however, can be relaxed quite

and that participation varies systematically with factors such as wealth

only 1% of investors' wealth was in mutual funds at around that time. In

Table4 PROPORTIONOF POPULATIONTHATHOLDSSTOCKBY

Figure 3 PERCENTAGEOF STOCK HELD BY WEALTH PERCENTILE

Table5 NET PURCHASESOF STOCKSBYINDIVIDUALS

Figure4 FOREIGNNET INVESTMENTAND RETURNS

tions to accommodate the demand for bonds by nonparticipants. Be-

and C?(j,t + 1) be the consumption of agent j when old (j = 1,2, . . ., ).

U(j,t) logC(j,t) + 1 logE[CO(j,t + 1)1- i(t)],

In equilibrium, this endowment plus dividends equals aggregate con-

C(j,t) = Y(j,t) - S(j,t)P(t) - B(j,t)Pb(t), (3a)

C(j,t) = Y(j,t) - B(j,t)Pb(t), (4a)

An equilibrium is given by processes for stock and bond prices {Ps(t): t =

4.2.1 Incomeand Preferences Let y(t) = log[Ya(t)/Y(t - 1)] be the growth

consumption. For the same reason, we assume that annual income

y(j,t) = E (j,t)[1 - 1(t)], (6)

E?(j,t+ 1) = e(j,t)co(j,t + 1), (9)

where o(j,t + 1) is the further 45%-standard-deviation shock to relative

4.2.2 VaryingParticipationRates Table 6 shows what happens when the

Table6 AVERAGEANNUAL RETURNSAS A FUNCTION

4.2.3 Increasing Diversification As a proxy for the increased diversifica-

Table7 THEEFFECTOF DIVERSIFICATION

considered here, there is a positive correlation between dividends and

4.2.4 Preference-Parameter Changes The potential effects of changing risk

As discussed in Section 3, increases in life expectancy may affect the

4.2.5 Heterogeneityin IdiosyncraticIncomeShocks An interesting question

Table9 AVERAGERETURNSAS A FUNCTIONOF PARTICIPATION

To implement this, we assume that a fixed number of participants

however, because of the greater volatility of the nonmarketed income of

4.2.6 Simultaneous Changes As discussed in the introduction, each of

ence between poorly diversified portfolios and a market index is the