The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
.................................................................
Business
Review
Federal Reserve Bank o f Philadelphia
S eptem ber •O cto b er 1991
Q
\
ISSN 0 0 0 7 -7 0 1 1
M
V
Understanding
National and Regional
Housing Trends
&
Leonard Mills
Premium
Puzzle
Andrew B. Abel
Business
Review
The BUSINESS REVIEW is published by the
Department of Research six times a year. It is
edited by Patricia Egner. Artwork is designed
and produced by Dianne Hallowell under the
direction of Ronald B. Williams. The views
expressed here are not necessarily those of this
Reserve Bank or of the Federal Reserve System.
SUBSCRIPTIONS. Single-copy subscriptions
for individuals are available without charge. Insti
tutional subscribers may order up to 5 copies.
BACK ISSUES. Back issues are available free of
charge, but quantities are limited: educators may
order up to 50 copies by submitting requests on
institutional letterhead; other orders are limited to 1
copy per request. Microform copies are available for
purchase from University Microfilms, 300 N. Zeeb
Road, Ann Arbor, MI 48106.
REPROD U CTIO N . Perm ission must be
obtained to reprint portions of articles or whole
articles. Permission to photocopy is unrestricted.
Please send subscription orders, back orders,
changes of address, and requests to reprint to
Publications, Federal Reserve Bank of Philadelphia,
Department of Research, Ten Independence Mall,
Philadelphia, PA 19106-1574, or telephone (215)
574-6428. Please direct editorial communications
to the same address, or telephone (215) 574-3805.
SEPTEMBER/OCTOBER1991
THE EQUITY PREMIUM PUZZLE
Andrew B. Abel
In the nearly 100 years from 1889 to 1978,
the inflation-adjusted return on stocks
averaged nearly 7 percent per year. Mean
while, short-term bonds returned less than
1 percent per year. How can equities have
paid such a premium for so long? Tradi
tionally, economists have looked to a so
phisticated asset-pricing model for the
answer, but that model is no longer be
lieved equal to the task.
UNDERSTANDING
NATIONAL AND REGIONAL
HOUSING TRENDS
Leonard Mills
As the recession unwinds, housing starts
will rebound from their low levels. But
don't look for any boom. The main deter
minant of the housing trend won't be the
economy's ups and downs, but rather
this decade's slow population growth.
Undoubtedly, the effects will differ across
regions. So policymakers, builders, and
others interested in the housing outlook
should keep a watchful eye on this impor
tant demographic change.
The Equity Prem ium Puzzle
T he basic paradigm used by financial econo
JL
mists to explain rates of return on assets was
called into question a few years ago by econo
mists Rajnish Mehra of the University of Cali
fornia at Santa Barbara and Edward Prescott of
the University of Minnesota. In a 1985 article
published in the Journal of Monetary Economics,
*Andrew B. Abel is a Professor of Finance at the Wharton
School, University of Pennsylvania, and a Visiting Scholar
at the Federal Reserve Bank of Philadelphia. He thanks
Stephen Cecchetti, Dean Croushore, Leonard Nakamura,
Jeremy Siegel, Herb Taylor, and Stephen Zeldes for detailed
comments, and Pat Egner for editorial assistance.
Andrew B. Abel*
Mehra and Prescott presented a powerful argu
ment that commonly used economic models
were incapable of accounting for the histori
cally observed rates of return on stocks and
short-term bonds (bills). Specifically, they found
that, in the 90 years from 1889 to 1978, the
average real rate of return on stocks was 6.98
percent per year, while the average real rate of
return on bills was only 0.80 percent per year.
The rate of return on stocks minus the rate of
return on b ills— the so-called equity
premium—averaged an astonishing 6.18 per
cent per year.
Why was the equity premium so large? The
3
BUSINESS REVIEW
obvious answer is "risk." Stocks are much
riskier than bills, and investors would not want
to hold stocks unless they were compensated
for the higher risk by earning a higher average
rate of return. This basic insight—that invest
ments with higher risk should earn higher
average returns—underlies the capital asset
pricing model (CAPM), initially developed in
the 1960s and refined considerably in the last
three decades.
Perhaps the most significant refinement, the
consumption capital asset pricing model
(CCAPM), recognizes that the ultimate reason
for holding wealth is to provide for future
consumption; as a result, the equity premium
should depend on the variability of consump
tion and its relation to stock returns. In light of
the small fluctuations in U.S. real consumption
per capita, however, Mehra and Prescott found
that the CCAPM could account for an equity
premium of only 0.35 percent per year, a tiny
fraction of the historically observed equity pre
mium. To describe this large discrepancy, they
coined the term "equity premium puzzle."
Trying to explain average rates of return
over a historical time period is a much less
formidable task than, say, trying to forecast the
returns on stocks or bills in any particular year.
Indeed, economists readily admit their limited
ability to forecast asset returns. But the
CCAPM's inability to account for average rates
of return on stocks and bills, even after the fact,
is a serious indictment of this model's practical
value.
Moreover, the basic CCAPM is essentially
the same as the model underlying the theory of
long-run economic growth and the new strand
of classical macroeconomics known as realbusiness-cycle theory. If the CCAPM has to be
discarded or even drastically altered, then much
of grow th theory and new classical
macroeconomics may need a major overhaul.
Indeed, the equity premium puzzle could lead
economists to reformulate basic models of
decisionmaking in the presence of risk.
4
SEPTEMBER/OCTOBER1991
THE CONSUMPTION CAPITAL
ASSET PRICING MODEL
The CCAPM is a sophisticated economic
model of the prices and rates of return on
assets. To understand its basic workings, let's
first see how asset prices would be determined
if investors did not care about the riskiness of
their investments.
Risk-Neutral Investors. Confronted with
two assets offering different expected rates of
return, risk-neutral investors would buy the
asset with the higher expected rate of return
and sell the asset with the lower expected rate
of return. These purchases and sales by inves
tors, however, ultimately affect the expected
rates of return. The asset with the higher
expected rate of return would attract buyers,
and its price would be bid upward. Of course,
when the price of the asset increases, its rate of
return falls because investors must pay more to
receive its payoffs. Similarly, the asset with the
lower expected rate of return would fall in price
as investors sold it. The fall in price would
increase the asset's expected rate of return by
allowing investors to acquire ownership and
future payoffs at a lower price.
The adjustment of asset prices and rates of
return would cause the gap between the rates of
return to shrink. When there is no more up
ward or downward pressure on asset prices,
the asset markets are said to be in equilibrium,
and the expected rates of return on both assets
will be the same. Thus, with risk-neutral inves
tors, the basic model of asset pricing predicts
that asset prices will adjust until all assets offer
equal expected rates of return.
Risk-Averse Investors. Most investors are
anything but risk-neutral, demanding a higher
expected rate of return in order to hold a riskier
asset. But how do we measure the riskiness of
an asset? The CCAPM offers a very precise
answer. Instead of measuring the riskiness of
an asset simply by the variability of its returns,
the CCAPM uses the relationship between the
asset's returns and the value an investor places
FEDERAL RESERVE BANK OF PHILADELPHIA
The Equity Premium Puzzle
on having an additional dollar of funds.1 When
the investor's overall wealth is low, his con
sumption is low and he places a relatively high
value on an additional dollar of funds. And
when the investor's overall wealth is relatively
high, his consumption is relatively high and the
value he places on an additional dollar of funds
is relatively low.
According to the CCAPM, an asset is risky if
its low payoffs occur when consumption is low
(and the value of additional funds is high), and
its high payoffs occur when consumption is
high (and the value of additional funds is low).
On the other hand, an asset would have nega
tive risk if its high returns occur when con
sumption is low and its low returns occur when
consumption is high; in this case, rather than
being risky, the asset would provide insurance
by offering high returns when the investor
values additional funds most highly (when
consumption is low).
The CCAPM predicts that risk-averse inves
tors will choose assets with the highest expected
value of returns weighted by the value placed by
investors on additional funds. As in the case of
risk-neutral investors, prices will adjust until
equilibrium is reached. In equilibrium, the
expected rates of return weighted by the value
of additional funds will be the same for all
assets.2 Nevertheless, assets with relatively
high risk will have higher average returns than
assets with relatively low risk. The higher
average return of a risky asset is offset by the
fact that the high returns occur when additional
funds have low value to investors.
’The value of additional funds is measured by what
economists call "the marginal utility of consumption."
Equilibrium is represented by the following technical
condition: E {(1 + q ) * MU) = E {(1 + r2) * MU), where MU is
the marginal utility of consumption (the value of additional
funds), q and q are the real rates of return on assets 1 and 2,
respectively, and E { ) denotes the expectation of the term
that appears inside the brackets.
Andrew B. Abel
If we apply the CCAPM to stocks and bills,
the average rates of return weighted by inves
tors' value of additional funds should be equal
for stocks and bills. To the extent that stock
returns (which comprise dividends plus capi
tal gains or losses resulting from changes in the
prices of stocks) are riskier than bill returns, the
average rate of return on stocks should be
higher than the average rate of return on bills.
How much higher depends quantitatively on
two factors: (1) the covariances of consumption
growth with stock returns and bill returns,
which measure the sizes of fluctuations in re
turns and how strongly these fluctuations are
related to the fluctuations in consumption
growth;3 and (2) the coefficient of relative risk
aversion, A, which indicates how much the
value of additional funds increases when con
sumption falls.4
*
Mehra and Prescott combined a simple eco
nomic model conventionally used in growth
theory and real-business-cycle theory with the
actual historical variability of U.S. consump
tion to capture the covariances of consumption
with asset returns. The value of A is an impor
tant ingredient in this analysis, and, based on
their reading of theoretical and empirical re
search, Mehra and Prescott argued that con
ventionally accepted values for A lie between 0
and 10. Using a variety of values for A in this
range, they found that, in the framework of the
technically, the covariance of stock returns with con
sumption growth equals the product of the correlation
coefficient between stock returns and consumption growth,
the standard deviation of stock returns, and the standard
deviation of consumption growth.
4If the coefficient of relative risk aversion equals A, then
a 1 percent fall in consumption increases the value an
investor places on an additional dollar of funds by A per
cent. For example, if A = 6, then a 2 percent fall in consump
tion increases the value of an additional dollar of funds by
12 percent.
5
BUSINESS REVIEW
CCAPM, they could not simultaneously ac
count for an average equity premium higher
than 0.35 percent per year and an average
return on bills of less than 4 percent per year.
For the average equity premium to be as large
as the historically observed equity premium,
the value of A would have to be extremely high,
around 30 or 40, which is much higher than the
conventionally accepted values for A.
To see why values of A around 30 or 40 are
conventionally viewed as implausibly high,
suppose that you face a risky situation that will
either raise your total wealth by 50 percent or
lower it by 50 percent, and each of these out
comes has a 50-50 chance of occurring.5 How
much would you be willing to pay for insurance
to avoid this risky situation? If you were riskneutral, so that A = 0, you would not care about
risk and would pay zero for such insurance.
However, if risk-averse, you would be willing
to pay something for this insurance, and the
amount would depend on the strength of your
risk aversion measured by A.6 With A = 2, you
would be willing to pay 25 percent of your
wealth; with A = 10, you would be willing to
pay 46 percent; and with A = 30, you would be
willing to pay 49 percent. Because it seems
implausible that you would pay 49 percent of
your wealth to avoid an even chance of losing
50 percent of your wealth or gaining 50 percent
of your wealth, many economists reject as im
plausible values of A as high as 30.
5In addition to financial assets, total wealth includes all
other tangible assets, such as real estate and consumer
durables, and also human capital, which is the present
value of a person's current and future labor income.
SEPTEMBER/OCTOBER1991
REEXAMINATION OF THE DATA
One approach to reconciling the gap be
tween the CCAPM and the average actual eq
uity premium reported by Mehra and Prescott
is to reexamine the historical data. The average
rates of return on bills (0.80 percent per year)
and stocks (6.98 percent per year) reported by
Mehra and Prescott are based on 90 years of
U.S. data. However, recent research by Jeremy
Siegel (1991) indicates that the rates of return in
the years between 1889 and 1978 may not have
been truly representative of the underlying
rates of return over a longer span of time. Siegel
compiled annual rates of return on stocks and
bills for the period from 1802 to 1990, starting 87
years before and ending 12 years after the
period examined by Mehra and Prescott.7 The
variability of stock returns is much greater than
the variability of bill returns, which is consis
tent with the notion that stocks are much riskier
than bills.
Although the greater variability of stock
returns is clear from Figure 1, the difference in
the average rates of return on stocks and bills is
not. To get a clearer view of the average rates of
return, we can calculate the 30-year moving
average rate of return, which, for any given
year, is the average of the rates of return over
the previous 30 years. In Figure 2, the differ
ence between the 30-year moving averages of
returns for stocks and for bills is the average of
the equity premium over the previous 30 years.
The 30-year moving average equity premium
increased substantially during the 1940s and
1950s and remained high during the 1960s and
1970s.
The average rates of return calculated by
Siegel for the period examined by Mehra and
Prescott (1889-1978) differ somewhat from the
values reported by Mehra and Prescott (see
^ h e general formula is
y = 1 - [(V 2) (1 - x)1-A + (V 2) (1 + x)1A]1/(1A,; where x is the
fraction of your wealth that you could gain or lose with a 5050 chance, A is the coefficient of relative risk aversion, and
y is the fraction of your wealth that you would pay to avoid
this risk.
6
7A s in Mehra and Prescott, the average rates of return are
arithmetic averages (rather than geometric averages) of
annual rates of return.
FEDERAL RESERVE BANK OF PHILADELPHIA
Andrew B. Abel
The Equity Premium Puzzle
FIGURE 1
Real Returns on Stocks and Bills
Annual returns 1802-1990
Percent per year
FIGURE 2
Real Returns on Stocks and Bills
30-year moving average, 1831-1990
Percent per year
7
BUSINESS REVIEW
SEPTEMBER/OCTOBER1991
table). The differences arise because
Siegel used a different stock price
Rates of Return and
index, a different measure of infla
the Equity Premium
tion, and, for part of the period, a
(Percent per year)
different short-term interest rate.8
Despite these differences, the basic
Real Return
Real Return
Equity
result is the same: the average equity
Premium
Period
on Bills
on Stocks
premium from 1889 to 1978 was very
large— well over 6 percent per year.
7.52
1802-1888
5.62
1.90
But including the additional 99
0.91
7.87
6.96
1889-1978
years of data in Siegel's study re
duces the average equity premium
1979-1990
2.73
9.44
6.71
from 6.96 percent per year to 4.62
percent per year. The reason for this
3.19
4.62
1802-1990
7.81
drop is that the average real rate of
return on bills rises to 3.19 percent
per year when we include data over the entire are to the underlying rates of return investors
1802-1990 period; the average real rate of re expect when making their portfolio decisions.
turn on stocks is virtually the same over that Applying statistical techniques to data from
period as over the period studied by Mehra and 1892 to 1988, Stephen Cecchetti, Pok-sang Lam,
Prescott. However, even this lower value of the and Nelson Mark (1991) found that the average
equity premium is much higher than that pre equity premium was 6.03 percent, but that the
dicted by the CCAPM examined by Mehra and equity premium expected by investors could
Prescott.9
have been anywhere from 2.35 percent to 9.71
0
Another way to examine the reliability of the percent.1 *Even the low value of 2.35 percent for
historical average rates of return is to estimate the equity premium is higher than the CCAPM
how close the historical average rates of return studied by Mehra and Prescott can explain.
Because the equity premium still appears
large after reexamining the historical data on
returns, the next step is to reexamine the basic
^ h e short-term real interest rate is intended to measure
CCAPM.
the short-term riskless rate of return, which is the real rate of
return that can be earned on a short-term asset that has no
risk of default or price variation. Siegel, as well as Mehra
and Prescott, used the interest rate on short-term Treasury
bills to measure the short-term riskless rate from 1920
onward. To measure the riskless interest rate before 1920,
Mehra and Prescott used the short-term commercial paper
rate, but Siegel adjusted the commercial paper rate to adjust
for the risk of default by issuing companies.
9The predictions from the CCAPM studied by Mehra
and Prescott are based on the variability of consumption
growth during the period 1889-1978. Strictly speaking, we
should use the variability of consumption growth during
the period 1802-1990 to compare the predicted equity pre
mium with the actual average equity premium reported by
Siegel. However, there are no reliable annual data on
consumption prior to 1889.
8
EXTENSIONS OF THE BASIC CCAPM
The other approach to explaining the equity
premium puzzle is to see if the basic CCAPM
can be modified to produce a realistic value of
10More precisely, their statistical analysis indicates that
if the expected equity premium was constant, then we can
be 95 percent confident that it was in the range of 2.35
percent to 9.71 percent. As for the riskless rate, its average
value was 1.15 percent, and we can be 95 percent confident
that the expected value of the riskless rate was between
-0.47 percent and 2.77 percent.
FEDERAL RESERVE BANK OF PHILADELPHIA
Andrew B. Abel
The Equity Premium Puzzle
the average equity premium using a value of
the coefficient of relative risk aversion, A, in the
conventionally accepted range of 0 to 10. Sev
eral potential modifications are discussed be
low.
Richer Models of Underlying Risk. In their
version of the CCAPM, Mehra and Prescott
assumed that consumption fluctuations be
haved according to a simple model that does
not allow for the possibility of a large, sudden
drop in consumption as might occur during a
sharp depression. In addition, Mehra and
Prescott assumed that fluctuations in stock
dividends were matched exactly by fluctua
tions in consumption, and they used historical
data on consumption to measure the variability
of dividends.1 Subsequent research, discussed
1
below, has studied the importance of these
assumptions by allowing for large, sudden
drops in consumption and by allowing fluctua
tions in dividends to differ from fluctuations in
consumption.
In a recent study, Thomas Reitz (1988) ar
gued that if there is some possibility of a large,
sudden drop in consumption accompanied by
a large, sudden drop in dividends, then inves
tors would be willing to hold stocks only if
compensated by a high average equity pre
mium. He found that extending the CCAPM to
include the possibility of depressions with large,
sudden drops in consumption could account
for the historically observed equity premium.
However, Mehra and Prescott (1988) point out
that the potential depressions analyzed by Reitz
11Dividends differ from stock returns because of changes
in the price of stocks. The return on a stock equals the
dividend plus the increase in the price of the stock (capital
gain) or minus the decrease in the price of the stock (capital
loss). In the CCAPM, the price of a stock is related to the
current and future dividends weighted by the current and
future marginal utilities of consumption. Given the behav
ior of consumption and dividends, we can compute the
price of stock, and the rate of return on stock, using the
CCAPM.
involved declines in consumption of 25 percent
or more during a single year. While it is true
that consumption during the Great Depression
fell 22 percent between 1929 and 1933,1 Mehra
2
and Prescott point out that in no single year did
consumption fall as much as 9 percent.1 Thus,
3
they conclude that the drops in consumption in
Reitz's study are too large to provide a realistic
solution to the equity premium puzzle.
An alternative approach to modeling the
riskiness of stocks is to incorporate in the model
spans of good years (high consumption growth)
and spans of bad years (low consumption
growth), with unpredictable switches between
the two. Shmuel Kandel and Robert Stambaugh
(1990) and Cecchetti, Lam, and Mark (1991)
used this approach, but concluded that a high
value of A was still needed to explain the
historically observed equity premium. Al
though this richer process of underlying risk
did not help explain the average rates of return
on stocks and bills, Kandel and Stambaugh
point out that it helps explain other statistical
features of returns, such as their predictability.
Another way to enrich the model of risk is to
relax the assumption that fluctuations in divi
dends are matched exactly by fluctuations in
consumption. One approach, followed by
Cecchetti, Lam, and Mark (1991) and Kandel
and Stambaugh (1990 and 1991), is to account
for the fact that stocks are leveraged claims on
firms. Firms generally raise capital by issuing
both stocks and bonds. Because firms must pay
their obligations to bondholders before they
can pay dividends to stockholders, leverage
tends to increase the riskiness of a stock and
would increase the equity premium in the
CCAPM. However, even taking account of
historically observed degrees of leverage, a
high value of A is still needed to account for the
12Reitz (1988), footnote 9, p. 125.
13Mehra and Prescott (1988), p. 134.
9
BUSINESS REVIEW
historically observed value of the equity pre
mium.
A more empirical approach to relaxing the
assumption that fluctuations in dividends are
matched exactly by fluctuations in consump
tion is simply to use historical data on divi
dends to measure dividend variability, and
historical data on consumption to measure
consumption variability. As pointed out by
Cecchetti, Lam, and Mark (1991), dividends are
much more variable than consumption.1 Us
4
ing the actual variability of dividends in the
CCAPM raises the equity premium predicted
by the CCAPM by about 50 percent for any
given value of A.1
5
The general conclusion is that richer models
of underlying risk can raise the value of the
equity premium predicted by the CCAPM.
However, the CCAPM still predicts a value for
the equity premium that is much lower than the
actual historical average value, if we continue
to use a coefficient of relative risk aversion less
than or equal to 10.
Differences Among Investors. The research
discussed so far has assumed that investors are
identical in all respects. Like other assump
tions used in economic models, this one was
made for the sake of simplicity. The question is
14In addition, the unpredictable components of divi
dend growth and consumption growth have a correlation
coefficient of 0.443, which is lower than the value of 1.0 that
is assumed in the Mehra-Prescott model.
15If the growth rates of consumption and dividends are
jointly identically and independently distributed, the eq
uity premium is approximately proportional to A times
Covfconsumption growth, dividend growth). Using con
sumption growth to measure dividend growth in the
CCAPM, the equity premium is approximately propor
tional to A times Var(consumption growth). Using data
from Cecchetti, Lam, and Mark, Cov(consumption growth,
dividend growth)=0.002053 and Var(consumption growth)
= 0.001398. Therefore, using actual dividend growth in
creases the equity premium by about 47 percent because
0.002053 is about 1.47 times as large as 0.001398.
10
SEPTEMBER/OCTOBER1991
whether this assumption is responsible for the
small predicted value of the equity premium in
most applications of the CCAPM.
To get an idea of the differences among
investors and their portfolios, N. Gregory
Mankiw and Stephen Zeldes (1991) studied the
asset holdings of 2998 families. They found a
striking degree of variation in the portfolios
held. In particular, 72.4 percent of the families
in the survey held no stocks at all.1 Even among
6
families that held more than $100,000 in other
liquid assets, only 48 percent held stock. This
finding is important because, to determine the
prices of assets, the CCAPM typically uses the
covariance of stock returns and aggregate con
sumption per capita.
But with almost three-fourths of the families
holding no stock at all, the covariance should be
calculated using the consumption not of all the
families but only of those that hold stocks.
Having made this change, Mankiw and Zeldes
find that the covariance of stock returns and
consumption per family triples, reflecting the
facts that, compared to nonstockholders, stock
holders have more volatile consumption and
their consumption is more closely related to
stock returns. This tripling of the covariance of
stock returns and consumption reduces by
about two-thirds the value of A needed to
account for the equity premium. This finding
is appealing, but leaves us asking why so many
consumers—especially wealthy consumers
16For the purposes of this study, a family that held stocks
in a pension fund but did not directly own stocks was
considered a nonstockholding family. Mankiw and Zeldes
argue that this treatment is appropriate because only 49
percent of the labor force had a pension fund, and only 31
percent of these people had defined-contribution (rather
than defined-benefit) plans. Thus, only 16 percent of the
labor force had defined-contribution plans. In definedbenefit plans, the stocks held by the pension fund are more
appropriately regarded as being owned by the employing
firms rather than the worker because the firm bears the risk
of changes in the value of stocks.
FEDERAL RESERVE BANK OF PHILADELPHIA
The Equity Premium Puzzle
Andrew B. Abel
with large amounts of liquid wealth—hold no pend on how much wealth the investor has.
Utility functions with these features are conve
stock.
A more fundamental question is why con nient, but have an important limitation: they do
sumption behaves so differently for different not distinguish an investor's aversion to risk
groups of consumers. The CCAPM is based on from his aversion to switching some consump
the assumption that even though individuals tion from one year to another year.
Kandel and Stambaugh (1991), Narayana
face idiosyncratic risks that do not hit everyone
in the economy, they can protect their con Kocherlakota (1990), and Philippe Weil (1989)
sumption from such risks by various sorts of have investigated rates of return in the CCAPM
risk-sharing and insurance arrangements. For using a more flexible utility function that distin
example, life insurance, disability insurance, guishes aversion to risk from aversion to sub
fire insurance, and so on protect an individual's stituting consumption between different years.
consumption against various idiosyncratic risks. However, they all conclude that, even with this
But problems such as the costs and difficulties more flexible structure, a very high value of the
of writing and enforcing various contracts pre coefficient of relative risk aversion is needed to
vent complete sharing of idiosyncratic risks. account for the historical value of the equity
Theoretical studies have examined the impact premium.
Moreover, Kandel and Stambaugh (1991)
of idiosyncratic risks on the equity premium,1
7
but these studies do not provide empirical have suggested that the search for a version of
evidence of the importance of these factors in the CCAPM that can explain a large equity
accounting for the equity premium puzzle. premium with a value for A of less than 10 is
perhaps misdirected. They argue that the con
Further research in this area is needed.
Attitudes Toward Risk. Investors' attitudes ventional view that A is small (less than 10) is
toward risk are represented in economic mod based on an unconvincing body of evidence.
els by utility functions that specify how much Furthermore, they point out that for risks that
utility, or satisfaction, an investor gets for each represent a relatively small portion of total
possible level of consumption.1 *The most com wealth, high values of A may be plausible. For
8
monly used version of the CCAPM is based on example, to avoid a risky situation that in
a particular utility function with two important volves either a 1 percent gain or 1 percent loss
features: (1) consumption in any year affects of wealth with equal probabilities, a person
utility in that year only; and (2) the utility with A = 30 would be willing to pay an insur
function has a constant coefficient of relative ance premium of 0.15 percent of his wealth (15
risk aversion, which implies that the share of percent of the amount at risk), which is not
the portfolio held in risky assets does not de implausible. Because high values of A (around
30) may be plausible for small risks, the impor
tant issue for asset pricing considerations is the
degree of risk aversion appropriate for the
17See Mankiw (1986), Weil (1990), and Kahn (1988). A
magnitude of the risks investors bear in their
different aspect of differences among investors— different
portfolios. The value of A is extremely impor
beliefs about future payoffs to risky assets— is examined in
tant for the equity premium puzzle because the
Abel (1989). That theoretical study shows that such differ
ences tend to increase the equity premium predicted by the
CCAPM will produce a high value of the equity
CCAPM.
premium if A is large.
Until this issue is resolved, Kandel and
18The marginal utility of consumption, discussed ear
Stambaugh urge us not to rule out high values
lier, is the derivative of the utility function with respect to
consumption.
of A, if we continue to use utility functions that
11
BUSINESS REVIEW
have a constant coefficient of relative risk aver
sion. In light of the difference in plausible
values of A for small and large risks, it may be
appropriate to use more general utility func
tions for which the coefficient of relative risk
aversion is not constant. Future research may
pursue this suggestion.
Another modification of the attitude toward
risk is to assume that an investor cares about his
level of consumption relative to a benchmark or
accustomed level of consumption attained in
the recent past. So far, studies have taken two
approaches to modeling an accustomed level of
consumption. In one approach, dubbed "Catch
ing up with the Joneses," an investor cares
about the level of his consumption relative to
the accustomed national average level of con
sumption (modeled as the level of national
consumption per capita in the previous year).
In this case, what an investor needs to guard
against is not a decline in his own consumption
per se, but a decline in consumption relative to
the national level of consumption per capita
attained in the previous year. With the level of
consumption per capita generally growing over
time, stocks that have a risk of occasional nega
tive rates of return appear very risky; investors
would be willing to hold stocks only if they
offer a large expected equity premium. Using
this modification of the utility function in simu
lations of the CCAPM can produce average
rates of return of 6.70 percent per year on stocks
and 2.07 percent per year on bills, with a value
for A equal to only 6.1
9
In the other approach to modeling an accus
tomed level of consumption—known as "habit
formation"—an individual investor's utility in
any year depends on his level of consumption
in that year compared to the level of his own
consumption in the recent past.2 Like the
0
"Catching up with the Joneses" model, habit
19These calculations are reported in Abel (1990).
12
SEPTEMBER/OCTOBER1991
formation makes investors more loath to hold
risky assets that could earn negative net rates of
return. Thus, stocks will have to offer a sizable
equity premium for investors to be willing to
hold them in their portfolios. Abel (1990) and
George Constantinides (1990) have used habit
formation in the CCAPM with low values of A
to generate fairly realistic values for the equity
premium.
CONCLUSION
Rather than discouraging use of the CCAPM,
the equity premium puzzle has provided the
impetus for new lines of research aimed at
making the statistical predictions of the CCAPM
conform more closely to the statistical behavior
of actual rates of return. One line of research
has focused on producing additional data on
asset returns and characterizing the statistical
behavior of the actual rates of return on stocks
and bills. This line of research has produced
useful new information about the statistical
properties of asset returns over an extended
period of time.
Another line of research has focused on
modifications of the basic CCAPM. Some of the
modifications, such as taking account of differ
ences among investors and incorporating more
general attitudes toward risk, seem to help
account for part of the large historically ob
served value of the average equity premium.
But accounting for the equity premium is only
a first step in accounting for the statistical
behavior of asset returns. A good model of
asset returns should also account for other
statistical properties, such as the variability or
20Another modification of the attitude toward risk is
studied by Nason (1988), who introduces a time-varying
lower bound on consumption in the utility function. This
formulation has some analytic similarities to "Catching up
with the Joneses" and habit formation, though it differs
from these formulations.
FEDERAL RESERVE BANK OF PHILADELPHIA
Andrew B. Abel
The Equity Premium Puzzle
predictability of returns.2 In addition, a model
1
that relates asset returns to consumption should
be tested to see whether it is consistent with
21Some of the research discussed in this article, notably
Cecchetti, Lam, and Mark (1990), Kandel and Stambaugh
(1990,1991), and Constantinides (1990), has already begun
to examine other statistical properties of returns, but more
remains to be studied.
data on consumption by individuals and by the
economy as a whole.
If incorporating differences among inves
tors or more general attitudes toward risk can
explain the various statistical properties of as
set returns—and if the results are consistent
with data on consumption—then the theories
of both long-run economic growth and real
business cycles will need to take account of
these modifications.
REFERENCES
Abel, Andrew B. “Asset Prices Under Heterogeneous Beliefs: Implications for the Equity Premium,"
Working Paper 9-89, Rodney L. White Center for Financial Research, University of Pennsylvania
(February 1989).
Abel, Andrew B. "Asset Prices Under Habit Formation and Catching Up With the Joneses," American
Economic Review 80 (May 1990), pp. 38-42.
Cecchetti, Stephen, Pok-sang Lam, and Nelson Mark. "The Equity Premium and the Risk-Free Rate:
Matching the Moments," mimeo, Department of Economics, Ohio State University (June 1991).
Constantinides, George M. "Habit Formation: A Resolution of the Equity Premium Puzzle," Journal
of Political Economy 98 (June 1990), pp. 519-43.
Kahn, James A. "Moral Hazard, Imperfect Risk-Sharing, and the Behavior of Asset Returns," Journal
of Monetary Economics 26 (August 1988), pp. 27-44.
Kandel, Shmuel, and Robert F. Stambaugh. "Expectations and Volatility of Consumption and Asset
Returns," Review of Financial Studies 3 (1990), pp. 207-32.
Kandel, Shmuel, and Robert F. Stambaugh. "Asset Returns and Intertemporal Preferences," Journal
of Monetary Economics T7 (February 1991), pp. 39-71.
Kocherlakota, Narayana R. "Disentangling the Coefficient of Relative Risk Aversion From the
Elasticity of Intertemporal Substitution: An Irrelevance Result," Journal of Finance 45 (March 1990),
pp. 175-90.
Mankiw, N. Gregory. "The Equity Premium and the Concentration of Aggregate Shocks," Journal of
Financial Economics 17 (1986), pp. 211-19.
Mankiw, N. Gregory, and Stephen P. Zeldes. "The Consumption of Stockholders and Non
stockholders," Journal of Financial Economics (forthcoming 1991).
13
SEPTEMBER/OCTOBER1991
BUSINESS REVIEW
REFERENCES (continued)
Mehra, Rajnish, and Edward C. Prescott. "The Equity Premium: A Puzzle," Journal of Monetary
Economics 15 (March 1985), pp. 145-61.
Mehra, Rajnish, and Edward C. Prescott. "The Equity Risk Premium: A Solution?" Journal of Monetary
Economics 22 (July 1988), pp. 133-36.
Nason, James N. "The Equity Premium and Time-Varying Risk Behavior," Finance and Economics
Discussion Paper 11, Board of Governors of the Federal Reserve System (February 1988).
Reitz, Thomas A. "The Equity Risk Premium: A Solution," Journal of Monetary Economics 22 (July
1988), pp. 117-31.
Siegel, Jeremy J. "The Real Rate of Interest From 1800-1990: A Study of the U.S. and the U.K.,"
Working Paper 9-91, Rodney L. White Center for Financial Research, University of Pennsylvania
(March 1991).
Weil, Philippe. "The Equity Premium Puzzle and the Riskfree Rate Puzzle," Journal of Monetary
Economics 24 (1989), pp. 401-21.
Weil, Philippe. "Equilibrium Asset Prices With Undiversifiable Labor Income Risk," mimeo,
Harvard University (July 1990).
14
FEDERAL RESERVE BANK OF PHILADELPHIA
U nderstanding N ational and
R egional H ousing Trends
A
L s the recession unwinds, housing starts
will rebound from their current low levels. But
over time, cyclical influences on housing starts
will be overshadowed by the demographic
factors that largely determine the trend in hous
ing starts. The slowdown in adult population
growth in the 1990s is a key factor in forecasts
of a lower future level of housing starts.
This slowdown in population growth, howj l j
*Leonard Mills is Manager of Financial Economics at the
Federal National Mortgage Association and a former Senior
Economist at the Federal Reserve Bank of Philadelphia.
Leonard Mills *
ever, won't be uniform across all regions; in
deed, in some areas, it is expected to be quite
pronounced. And so, given the strong link
between population and housing, it is reason
able to expect the decline in the number of
housing starts to affect some regions more than
others.
Undoubtedly, there will be cyclical swings
in housing in the years ahead, and these swings
will affect regions differently. But policymakers,
builders, and others concerned about the hous
ing outlook should keep an eye on the slower
population growth and its effect in lowering
the number of housing starts.
15
SEPTEMBER/OCTOBER1991
BUSINESS REVIEW
LINKING POPULATION AND HOUSING
Housing researchers usually analyze the link
between the adult population and the number
of housing units in terms of two ratios: 1) the
vacancy rate, the number of vacant houses di
vided by the total number of houses; and 2) the
headship rate, the number of households di
vided by the adult population. These two
components are affected by different economic
and sociological factors.
Vacancy rates are partly affected by busi
ness-cycle conditions. For example, when in
come growth slows during a recession, people
are less willing or able to afford the higher
mortgage payments associated with new
homes. To the extent that the recession is
unanticipated by builders, the inventory of
new homes—which are vacant homes—rises.
The same is true for apartments; more of them
become vacant during recessions.
The vacancy rate is also subject to long-term
structural changes, such as laws that affect the
cost of carrying a vacant housing unit. For
example, any tax law change that accelerates
depreciation deductions will lower the after
tax cost of carrying a vacant unit; stretching the
deductions out over more years will raise the
cost.1 Any shifts in the vacancy rate due to tax
law changes—or other structural changes—gen
erally last longer than those due to the business
cycle because tax laws change relatively infre
quently.2
The other factor affecting the link between
population and housing is the headship rate.
The headship rate measures adults' tendencies
to form households. For example, two adults
could choose to live together to form one house
hold and reside in the same housing unit. For
these two adults, the headship rate would be
0.5. Alternatively, they could choose to live in
two separate housing units. In this case, two
households would be formed and the headship
rate for the two adults would be 1.
Like vacancy rates, headship rates are sub
ject to both business-cycle changes and long
term changes. When incomes are low during a
recession, adults are more likely to join to
gether to form a single household because they
may be unable to afford living alone.3 Thus, the
headship rate has a tendency to fall during
recessions. Longer-term changes in the
headship rate include such factors as a fall in
marriage rates or an increase in divorce rates,
both of which decrease the tendency for adults
to get together and form households.
The link between population and housing
can be summarized by combining these two
components. Specifically, the number of hous
ing units per adult (HPA) can be computed as
follows:
Headship
Rate
(1 - Vacancy Rate)
’In 1981, depreciation deductions were accelerated, but
five years later the depreciable life of residential real estate
was lengthened. For a discussion of these tax law changes,
see Stephen A. Meyer, "Tax Cuts: Reality or Illusion?" this
Business Review (July/August 1983), and Theodore Crone,
"Housing Costs After Tax Reform," this Business Review
(March/April 1987).
2For a discussion of some other factors affecting the
natural vacancy rate in rental housing units, see Stuart A.
Gabriel and Frank E. Nothaft, "Rental Housing Markets and
the Natural Vacancy Rate," Journal of the American Real Estate
and Urban Economics Association 16 (1988), pp. 419-29.
16
Housing Units
Per Adult (HPA)
For example, the average U.S. vacancy rate for
housing units since 1973 has been 3.2 percent,
3For a discussion of the effect of income and other factors
on household formation decisions, see Lawrence B. Smith et
al., "The Demand for Housing, Household Headship Rates,
and Household Formation: An International Analysis,"
Urban Studies 21 (1984), pp. 407-14. Also see Patric H.
Hendershott and Mark Smith, "Household Formations," in
The Level and Composition of Household Saving, Patric H.
Hendershott, ed. (1985).
FEDERAL RESERVE BANK OF PHILADELPHIA
Understanding National and Regional Housing Trends
Leonard Mills
and the headship rate—for the population 21
and over—has averaged 53.8 percent.4 (See
Components of Housing Unit Per Adult.) These
two components lead to the calculation of 0.556
housing units per adult [0.538 f (1-.032)]. In
other words, there have been about 1.8 (1 f
0.556) adults living in each housing unit.
Given that the number of housing units per
adult has averaged 0.556 and that the 21-and-
4In this article, total housing units are defined as the sum
of owner-occupied units, renter-occupied units, units for
sale, and units for rent. Thus, units rented or sold but not
occupied, units held for occasional use, and other vacancies
are excluded from this analysis.
over population has increased by 39 million
since 1973, the trend increase in the housing
stock has been 22 million units (0.556 x 39
million). Meanwhile, the actual increase in the
housing stock was 23 million units, a level close
to the trend.
In summary, then, HPA is simply the link
that allows one to translate population growth
into housing growth.
The Link Between Population and Hous
ing Has Been Stable. The arithmetic shown
above is simple enough, but there is a compli
cation: the link between population and hous
ing may not be stable. The HPA will change
whenever one of its two components changes.
Components of Housing Unit Per Adult
For Total Housing Stock
Nation
Northeast
Midwest
South
West
.032
.538
.556
.025
.521
.534
.030
.541
.558
.038
.539
.559
.033
.554
.573
.562 (87)
.550 (80)
.542 (85)
.525 (73)
.570 (81)
.548 (73)
.576 (87)
.542 (83)
.589 (78)
.560 (83)
Averages (1973-87)
Vacancy Rate
Headship Rate
Housing Units Per Adult
Ranges for Housing
Units Per Adult
High (year)
Low (year)
For Owner-Occupied Housing Stock
Nation
Northeast
Midwest
South
West
Averages (1973-87)
Vacancy Rate
Headship Rate
Homeowner Rate
Housing Units Per Adult
.015
.602
.646
.395
.580
.603
.354
.014
.609
.689
.426
.017
.604
.669
.411
.016
.619
.600
.377
.405 (81)
.388 (87)
.366 (81)
.339 (73)
.444 (81)
.408 (87)
.421 (79)
.402 (83)
.393 (76)
.357 (87)
.011
Ranges for Housing
Units Per Adult
High (year)
Low (year)
17
BUSINESS REVIEW
SEPTEMBER/OCTOBER1991
An increase in the vacancy rate means that from its long-term average. For example, be
there are relatively more houses with no one cause the headship rate tends to fall during a
living in them, which raises the number of recession, HPA tends to fall below its historical
houses per adult. HPA will also vary with the average and consequently the housing stock
headship rate. A higher headship rate is asso falls below its trend line.5 In other words, any
ciated with a higher number of housing units deviations in the actual housing stock from its
per adult, and a lower headship rate is associ population-driven trend line appear as tempo
*
ated with a lower number of units per adult. rary cyclical deviations.6
Because both the vacancy and headship rates
depend on a variety of economic and sociologi
cal factors that can change over time, the HPA
^ h e tendency for the residential vacancy rate to rise
can also change over time. Accordingly, a
during a recession will raise the HPA, which offsets the
constant HPA, such as the historical average effect of the declining headship rate during the recession.
used above, might not provide an accurate However, even though vacancy rates are subject to more
assessment of population-related trends in cyclical variability than headship rates, the effect of the
vacancy rate variability on HPA is generally smaller be
housing.
cause the vacancy rate is so low.
In fact, the HPA has varied only slightly
since 1973, ranging from a high of 0.562 in 1980
6For statistical evidence on the use of the historical
to a low of 0.550 in 1987. Thus, the high and the average HPA in projecting housing trends, see Theodore
low for HPA are only about 1 percent from the Crone and Leonard Mills, "Forecasting Trends in the Hous
historical average of 0.556. More important, ing Stock Using Age-Specific Demographic Projections,"
Journal of Housing Research 2 (1991), pp. 1-20. This study
there has been no noticeable trend in the HPA found that, over the 1965-89 period, the HPA-based trend
since 1973.
was more precise for owner-occupied units than for total
Since the link between housing and popula units.
tion seem s fairly
stable, the trend in
FIGURE 1
housing can be reli
U.S. Housing Stock
ably computed by
multiplying the av
(1973-2000)
erage HPA by the 21 Millions o f Units
and-over p op u la
tion. This popula
tion-driven trend in
housing is in te r
preted as the num
ber of houses re
quired by the size of
the adult population
(Figure 1). Of course,
the actual housing
stock has differed
from this long-term
trend. This differ
ence occurs when
ever HPA d iffers
18
FEDERAL RESERVE BANK OF PHILADELPHIA
Understanding National and Regional Housing Trends
PROJECTING THE FUTURE
HOUSING STOCK
By making some assumptions about the fu
ture size of the adult population and HP A, we
can project the future housing stock. Since the
bulk of the people who will make up the adult
population have already been bom, adult popu
lation projections for the next 10 to 20 years are
considered very reliable. The Census Bureau
regularly projects the future adult population
by making assumptions about other determi
nants of the adult population, such as immigra
tion and deaths.7
These other determinants are relatively easy
to project over a span as short as a decade. For
HP A, a useful baseline is to assume that the link
between population and housing will remain
as it has been. On the basis of historical evi-
7Both the regional and national population projections
used in this study are provided by the Census Bureau. The
national projections are consistent with the "middle" (inter
mediate) series of the Census Bureau.
FIGURE 2
Housing Starts
Annual Averages
Leonard Mills
dence, using the average HPA has provided a
reliable housing trend. For the 1990s, then, the
trend growth in the housing stock is projected
to slow because the adult population is pro
jected to grow more slowly. The lower trend
growth tells us that fewer houses will need to be
added to the housing stock.
Less Need for Additional Housing Stock
Means Lower Housing Starts. Changes in the
housing stock occur when new housing units
are built or old units are removed. Net addi
tions to the housing stock are defined as the
number of new housing starts minus the num
ber of removals. In other words, in any given
year, the number of housing starts must equal
net additions to the stock plus removals.8
*
For the 1990s, the net additions required to
accommodate the number of new households
should average about 900,000 units per year.
Adding the historical average of 300,000 re
movals per year results in an annual average of
about 1.2 million housing starts in the 1990s.
This is substantially below the annual level of
housing starts in the 1970s and 1980s (Figure 2).
Trend, as opposed
to actual, housing
starts are expected
to fall from about
1.4 million in 1991
Thousands o f Units
2000 -.
* Actual
** Trend
8H ousing starts are
measured directly, but re
movals must be mea
sured indirectly by sub
tracting the number of
housing starts in a given
year from the actual
change in the housing
stock. Since 1973, remov
als have averaged about
300,000 units per year.
Removals were higher in
the 1960s because of the
large number of urban
renewal programs.
19
SEPTEMBER/OCTOBER1991
BUSINESS REVIEW
to about 1.1 million by the mid-decade.9
Single-Family Starts Are an Important Com
ponent. About two-thirds of the nation's house
holds own their homes. Moreover, the owneroccupied segment of the housing stock is gen
erally regarded as more stable because it seems
less susceptible to waves of speculative invest
ment. In addition, tax laws for owner-occupied
housing have changed less frequently than those
for rental housing. For these and other reasons,
the owner-occupied segment of the housing
market is often analyzed separately.
Because there is less turnover in the owneroccupied housing market, the vacancy rate is
naturally lower than that for renter-occupied
units. The owner-occupied vacancy rate has
averaged 1.5 percent since 1973. Moreover, a
very large proportion of home owners are in the
25-and-over age category, so that is the most
relevant population segment for calculating
the headship rate for owner-occupied units.
Since 1973, the headship rate for the population
25 and over has averaged 60.2 percent.
In addition to headship and vacancy rates,
the home-ownership rate is a necessary factor
in analyzing the owner-occupied segment. The
home-ownership rate is the number of home
owning households divided by the total num
ber of households. The historical average for
the home-ownership rate has been 64.6 percent.
Multiplying the headship rate by the homeownership rate, and dividing by 1 minus the
vacancy rate, determines the number of owneroccupied houses per adult. Since 1973, the
average has been 0.395, varying little.
A projection for single-family housing starts
can be obtained by following the procedure
outlined above for total housing starts.1 First,
0
9With the recession, actual housing starts through the
first half of 1991 averaged only 956,000 at an annual rate.
That the trend level of starts in 1991 is so much higher is one
reason why many forecasters expect a rebound in housing
activity once the recession ends.
20
multiply the change in the projected 25-andover population by the average number of
owner-occupied houses per adult, 0.395. Then
add the historical average number of removals
for the owner-occupied housing units (about
200, 000).
On the basis of this calculation, single-family
starts are also projected to decline in the 1990s.
The trend in single-family starts falls to an
annual average of 875,200 in the 1990s, com
pared to 1 million in the 1980s. The trend level
of single-family starts falls from about 1 million
in 1991 to slightly more than 740,000 by the end
of the decade.
The population cohort born during the "birth
dearth" period between 1972 and 1978 will, of
course, enter the 21-and-over population be
fore it enters the 25-and-over population. Con
sequently, the drop in the trend for single
family starts occurs later in the 1990s than the
drop in the trend for total starts. In fact, the
decline in the population-driven trend for single
family starts is not expected to be noticeable
until 1996, while the decline in total starts will
begin in 1993.1
1
Other Housing Market Aspects Are More
Difficult to Predict. Even though the number
of housing starts is expected to fall in the 1990s,
that does not imply that the residential con
struction sector of the U.S. economy will de
cline. Although new housing measured in
physical units (housing starts) will decline, each
housing unit may be larger or more elaborate.
And larger, more elaborate housing units are
generally more expensive to build. Thus, new
10Single-family starts and additions to owner-occupied
units are closely related because most occupied single
family units are owner occupied (85 percent in 1985) and
few single-family units are built as rentals.
11For a broader discussion of the economic impact of the
birth-dearth generation, see Theodore Crone, "The Aging of
America: Impacts on the Marketplace and Workplace," this
Business Review (May/June 1990).
FEDERAL RESERVE BANK OF PHILADELPHIA
Understanding National and Regional Housing Trends
housing measured in dollar units (residential
investment) may not decline as much as hous
ing starts. So residential construction workers
and suppliers of residential building materials
may not experience as large a drop in their
business activity.
In addition, the trend in the size of the adult
population tells us nothing about the long-term
price of each housing unit.1 Of course, a
2
temporary overabundance of housing units
relative to the population-driven trend, either
regionally or nationally, will depress housing
prices temporarily. Similarly, if the housing
stock falls below the population-driven trend,
there will be a tendency for house prices to be
bid up. But these effects are only temporary
because eventually the housing stock does re
turn to its long-term trend. The long-term price
of a housing unit appears to be determined not
so much by changes in housing demand due to
population trends, but by such factors as the
cost of land, materials, and labor.1
3
HOUSING PROJECTIONS DIFFER
BY REGION
The average HPA varies across different
regions of the country. For the total housing
12This is the consensus among housing analysts. How
ever, for an argument that population trends have a large
impact on housing prices, see N. Gregory Mankiw and
David N. Weil, "The Baby Boom, the Baby Bust, and the
Housing Market," Regional Science and Urban Economics 19
(1989), pp. 235-58. Arguments against any impact of popu
lation trends on housing prices are more prevalent. For
example, see Denise DiPasquale and William Wheaton,
"Housing Market Dynamics and the Future of Housing
Prices," Joint Center for Housing Studies of Harvard Uni
versity, Working Paper W90-3; A. Steven Holland, "The
Baby Boom and the Housing Market: Another Look at the
Evidence," Regional Science and Urban Economics (forthcom
ing); and "Will Home Prices Collapse?" various special
reports compiled by the National Association of Home
Builders (1990).
13See James R. Follian, "The Price Elasticity of the LongRun Supply of New Housing," Land Economics (1979), pp.
190-99.
Leonard Mills
stock, the Northeast has the lowest average
HPA and the West has the highest. An impli
cation of this regional difference in HPA is that
even if the Northeast and West experience
identical increases in their adult populations,
more houses would have to be built in the West.
The regional differences in owner-occupied
houses per adult are larger than those for the
total housing stock because home-ownership
rates also differ across regions. The Northeast
also has the lowest average number of owneroccupied houses per adult. But the Midwest
has the highest average number of owneroccupied houses per adult, primarily because it
has the highest home-ownership rate.
Even if we consider only the adult segment
of the population, regional populations are
more difficult to project than the national popu
lation because we must also make assumptions
about internal migration within the country.
Using one set of Census Bureau assumptions
regarding internal migration and assuming the
number of houses per adult in each region
equals its historical average, we can use re
gional population projections to predict the
trend in housing starts by region.1 Under this
4
set of assumptions, the Midwest will experi
ence the largest percentage decline in total
housing starts, followed by the Northeast. This
regional pattern holds for single-family starts
as well, with the drop occurring later in all
regions. (See Projected Trends in Housing Starts,
p. 22.)
14The Census Bureau's "Series A" for regional popula
tion projections is used in the housing projections. For a
description of the assumptions underlying these projec
tions, and alternative projections, see Signe I. Wetrogan,
"Projections of Population of States by Age, Sex, and Race,
1989 to 2010," Current Population Reports: Population Esti
mates and Projections, Series P-25, No. 1053 (1990). Of course,
regional housing projections will differ under different
migration assumptions.
21
SEPTEMBER/OCTOBER1991
BUSINESS REVIEW
Projected Trends in Housing Starts
(Thousands)
Total Housing Starts
Nation
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
1415
1438
1259
1138
1072
1107
1062
1126
1131
1186
Northeast
158
164
127
105
95
104
99
111
115
126
167
168
113
78
61
70
58
73
75
88
South
West
621
642
594
554
528
537
519
543
545
565
Midwest
470
465
425
401
387
395
385
398
395
407
West
Single-Family Housing Starts
Nation
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
Northeast
Midwest
South
1000
950
914
928
942
968
838
753
716
743
123
115
108
107
109
112
87
72
67
72
145
126
111
111
112
118
80
57
46
54
449
433
423
432
441
457
419
389
373
382
283
276
272
277
279
280
252
236
229
234
Note: The sum of the regions may not add up to the nation because of rounding.
COULD THESE HOUSING
PROJECTIONS BE WRONG?
By simply analyzing trends in the size of the
adult population, we can construct a histori
cally reliable baseline for the number of houses
to be built in the 1990s. Inevitably, the projec
tions will differ from the number of houses
actually built, for a variety of reasons. Housing
starts have a large cyclical component, which
leads to deviations from trend. For example,
Digitized for22
FRASER
housing starts so far in 1991 have been substan
tially below their trend level of 1.4 million units,
partly because of the recession. And adjust
ments to these cyclical deviations may take
some time to correct themselves. But over time,
history leads us to believe that housing will
return to its population-driven trend.
However, it is possible that the populationdriven trend itself may change. First, national
population projections must make some as
FEDERAL RESERVE BANK OF PHILADELPHIA
Understanding National and Regional Housing Trends
sumptions about immigration and other fac
tors affecting the adult population. And re
gional projections must make additional as
sumptions about internal migration within the
country. The assumptions underlying the popu
lation projections may turn out to be inaccu
rate, which in turn would make the housing
projections inaccurate.
Second, long-term changes in headship, va
cancy, and home-ownership rates may also
occur, and such changes would alter the link
between population and housing. For example,
some analysts think the headship rate may rise
slightly in the 1990s as the baby-boom genera
tion ages further.1 They argue that, histori
5
cally, headship rates have continued to rise as
people grow older, even though the increase
after age 25 is very gradual. Since a higher
headship rate would raise the HPA, the trend in
housing starts may be higher in the 1990s. This
argument assumes that the baby-boom genera
tion will have the same tendencies to form
15See, for example, Gretchen A. Armijo et al., "Demo
graphic and Economic Trends," Journal of Housing Research
1 (1990), pp. 21-42.
Leonard Mills
households as their parents did. However, an
alternative assumption is that each generation,
including the baby-boomers, behaves uniquely
and therefore long-term changes in headship
rates are difficult to predict.
In any case, prospective changes in the va
cancy, headship, and home-ownership rates
will have very little impact on the trend in
housing starts compared to the effect of the
slower population growth. For example, even
if headship rates were to rise by 10 percent—a
figure far greater than any analyst expects—the
annual average for total housing starts would
rise only by 90,000, to about 1.3 million, still far
below the level of starts in the 1970s and 1980s.1
6
Thus, housing analysts, though they may dis
agree on the precise numbers, are unanimous in
their belief that housing starts will decline sub
stantially in the 1990s. Consequently, builders
will build fewer houses in this decade, and
policymakers and others concerned with the
housing outlook should not be surprised by the
lower number of starts.
16This calculation continues to assume that the vacancy
rate and level of removals remain unchanged.
23
FEDERAL
RESERVE BANKOF
PHILADELPHIA
BUSINESS REVIEW Ten Independence Mall, Philadelphia, PA 19106-1574
Address Correction Requested
@FedFRASER