The Capital Asset Pricing Model and the

CAPM Literature
Marshall E. Blume
Howard Butcher 11/ Professor of Finance
The Wharton School, University of Pennsylvania

A notable accomplishment of the capital asset pricing model is that it turned the focus
of investing from single assets to whole portfolios. Although its assumptions are
unrealistic and many empirical anomalies have come to light, theoretical work continues
to generalize the CAPM to relax some assumptions, to add price-explanatory factors, and
to broaden the measurement of risk.

The controversy over the CAPM has many ingredients. Although dissecting a controversy and examining its parts always involves some degree of simplification, this presentation separates the controversy
into three ingredients: the asset-allocation model, expected return relationships, and volatility models.
First, I will define the CAPM and describe some of the
empirical contradictions to the efficient market hypothesis, and then I will comment on where beta
stands in this controversy.

Origins of the CAPM

The literature on asset pricing is substantial. Jeremy
Siegel and I recently reviewed this body of literature
beginning with Graham and Dodd's (1934) Security
Analysis. Implicit in Graham and Dodd's original
theory was the idea that a stock has an intrinsic value.
If an investor purchased an asset or stock at a price
below its intrinsic value, the asset over time would
move up to its intrinsic value without risk. Graham
and Dodd recognized that people hold different expectations of the future, but they had little to say about
diversification. The basic idea was that, if every stock
bought was below its intrinsic value, the overall portfolio would be a good one and would make money as
the values of the component stocks rose to their intrinsic values.
The legal profession translated this intellectual
idea into the Prudent Man rule for investing personal
trusts. According to this rule, a trust manager must
Note: This presentation is based on Blume and Siegel (1992).

invest in each asset on its own merit. If each asset is

safe, then the total portfolio will be safe. For example, futures cannot be used under the Prudent Man
rule because they are inherently risky-even though
investment managers now know that when futures
are combined with other assets, they can reduce portfolio risk.
Markowitz (1959) then developed mathematics
for the efficient set. He postulated that any portfolio
could be evaluated in terms of two parameters: the
expected return on the portfolio and the volatility (or
uncertainty) of getting that expected return; volatility
was measured by the variance, or standard deviation,
of portfolio return. He developed a parametric quadratic programming algorithm. At each level of expected return, this algorithm minimized the variance;
at each level of risk, it maximized the expected return.
This concept of looking at an entire portfolio changed
the way investors think about investing.
Markowitz focused on the portfolio as a whole,
not explicitly on the individual assets in the portfolio,
which was clearly at odds with the Prudent Man rule
for personal trusts. In fact, until the Employee Retirement Income Security Act passed in the mid-1970s,
investing in derivatives to reduce the risk of a portfolio was, for the most part, legally imprudent.
The optimality conditions for the efficient set depend on certain variables. The expected risk premium on an individual asset is defined as the difference between the expected return of the asset as perceived by the investor and the risk-free rate of return,
which is typically the T-bill rate (a taxable investor
would use after-tax returns). As a condition for opti5

- - - - - - - - - -

mality, the expected risk premium on each asset must

be proportional to the expected risk premium on the
efficient portfolio; the constant of proportionality is
the beta coefficient. (Markowitz assumed no transaction costs and no impediments to short sales.) The
beta coefficient, then, is defined relative to an efficient
portfolio, which can differ from one investor to another, and different efficient portfolios will imply
different beta coefficients. The beta coefficient for a
particular security measures the contribution of that
security to the overall risk of the efficient portfolio. If
an asset's beta is greater than 1 and more of that asset
is added to the portfolio, the volatility of the portfolio
will initially increase. If beta is less than 1 and the
proportion of that asset increases, volatility is initially
reduced.
With the Markowitz model, each asset will have a
different beta depending on the efficient portfolio implied by an investor's judgment about future expected
returns and risks of individual assets. Mathematically,
the result looks like a CAPM, but it is not one; it is an
optimality condition for an efficient portfolio.
People with different views will have different
optimality conditions. The model is consistent, for
example, with investors having different tax rates.
Suppose the short-term interest rate is 6 percent. A
tax-free investor would plug 6 percent into the algorithm; an investor who pays taxes at 50 percent would
plug 3 percent into the algorithm. Thus, efficient sets
will differ from one individual to another. Individuals
can even have different views of before-tax expected
returns and risk characteristics of individual assets.
The most important assumption in Markowitz's
efficient-set algorithm is that a portfolio can be evaluated by two parameters: the expected return of the
portfolio and the standard deviation of the portfolio.
The algorithm has no assumption that the market got
the prices right, so security analysts have a role: If one
analyst has better judgment than others, that analyst
will do better in the long run. Markowitz said nothing about market efficiency.
Markowitz's theory is not a model of equilibrium
in the capital markets. It is a way of combining judgments to pick a good portfolio. Different investors
could have different portfolios, and a portfolio that is
efficient to one investor may be inefficient to another.
The expected risk premium of every asset in a portfolio will be proportional to the expected risk premium
of the portfolio as judged by the investor in that
portfolio. One investor may construct a portfolio and
say the expected risk premium is 6 percent; another
may look at the same portfolio and say it is only 5
percent.
Sharpe (1964) described the nature of portfolio
selection under the assumption that all investors at6

..

_--

tribute the same expected returns and risk characteristics to all assets (see also Mossin 1966 and Lintner
1965). Sharpe demonstrated that, under this condition of homogeneous expectations, the best portfolio
of risky assets to hold is a totally diversified one in
which each security is held in proportion to its value
in the market. In an efficient market, the risk premium on each stock is proportional to the risk premium on the entire stock market (the expected return
on the market minus the risk-free rate), where the
constant of proportionality (called the beta coefficient of the stock) is related to the covariance of the
individual stock and market returns. Sharpe developed these ideas into a formal model to explain the
differences in risk premiums of individual stocks,
which came to be known as the capital asset pricing
model.
According to the CAPM, every investor holds the
market portfolio of risky assets in conjunction with a
long or short position in the safe asset. If investors
want more risk than the market portfolio, they short
(borrow) at the risk-free rate; if they want less risk,
they put some of their money into the market portfolio and put the rest in very safe assets. Therefore, the
market portfolio of risky assets is efficient and
Markowitz's optimality conditions apply. The expected risk premium on every asset is proportional to
the expected risk premium on the market, and its
constant proportionality is the beta. With the CAPM,
beta coefficients are unique because the efficient risky
portfolio for each investor is the same portfolio,
namely, the market portfolio, with respect to which
betas are measured.
The CAPM is based on several simplifying assumptions. Some of the critical assumptions include
the homogeneous beliefs of investors, no taxes, complete liquidity, and no transaction costs. The model
assumes that people look at a portfolio in terms of
only two parameters-the expected return and the
variance of the return. Taxes are not included because
taxes would cause expected returns to differ. All
assets must be completely liquid so as to be readily
marketable.
Given these assumptions, investors will hold the
same market portfolio and the CAPM will follow
Markowitz's mathematics. These assumptions do
not describe the real world, but that does not mean
that the economic model is wrong. The question is
whether the model captures enough reality to be
useful.
_

Empirical Contradictions to the CAPM

The popularity of beta as a means of measuring the
risk of a portfolio spread rapidly through the invest-

ment community, but researchers soon began to find

some troubling discrepancies between the predictions of an efficient market and the empirical data.
The early tests of the CAPM found that the relationship between expected returns and betas measured
relative to a market portfolio was flatter than expected. Low-beta assets had higher expected returns
than predicted, and high-beta assets had lower returns. All of these relationships were extremely
weak and varied from period to period. Sometimes
for periods of 10 or 20 years, the relationship would
be wrong; then, something would happen and it
would be right for a while. This phenomenon was
not overly disturbing; perhaps the beta was slightly
mismeasured. Basically, the relationship between
expected returns and betas was positive.
As Roll (1977) strongly maintained, any empirical test of the CAPM would encounter theoretical
difficulties. In theory, the market portfolio in the
CAPM consists of all risky assets, including real estate, bonds, and other nonequities. In practice, the
empirical tests of the CAPM measured the return on
the market portfolio of only a subset of risky assetsoften proxying this market portfolio by a stock index
such as the S&P 500. Roll showed that the use of an
incomplete measure of the market portfolio resulted
in mismeasuring the beta coefficients of individual
assets, and he then argued that this mismeasurement
made tests of the CAPM meaningless.
Despite these problems, by the end of the 1970s,
the CAPM was the most sophisticated pricing mechanism available to portfolio managers. It was widely
used in investment analysis.

Anomaly Literature
A large body of evidence supported the early
version of the efficient market hypothesis, but a
growing number of studies raised questions about its
validity. Many of these studies identified so-called
anomalies, inconsistencies, in the efficient market
hypothesis. These anomalies are hard to rationalize
with any reasonable theory of equilibrium in the
context of the CAPM, although people have tried.
Vallie Lille tests. The first anomaly was reported by Black in an article in the Fillallcial Allalysts
JOllnlal in 1973. He found some predictability in the
recommendations of the Vallie Lille rankings; that is,
the stocks ranked number one tend to do better than
stocks ranked number five. That finding did not
disturb too many people, because many of the companies ranked number one in Vallie Lille were small.
Investors could not put a lot of money into them.
Some, but not many, people at the margin could
make some money.
Closed-end illvestment companies. Another

anomaly is the discount on closed-end investment

companies. Some people have tried to rationalize
this discount as the value of the management fee
option the managers have. These discounts are
large, however, and it is hard to believe managers
could waste that much money.
Size and calendar anomalies. Several researchers have found a difference between the returns on
large- and small-capitalization stocks. Liquidity effects might account for some difference among
stocks, but the problem is that, in some tests, largecapitalization stocks did better than small-capitalization stocks in some periods. Then, researchers found
that most of that size effect occurred in the first three
days of January.
Another calendar-related anomaly is the dividend-yield effect. This effect refers to the V-shaped
relationship between expected returns and dividend
yields. Zero-dividend stocks have very high expected returns; then the return drops, and it goes up
again for high-dividend-yield stocks. Virtually all of
this pattern is attributable to returns in January.
These anomalies are at odds with the CAPM
predictions. In recent years, however, some of them
may have diminished or disappeared. Whether this
change is the result of investors acting on the results
of these studies or is just a statistical happenstance
has yet to be determined.

Volatility Tests
A further setback to the efficient market hypothesis consists of the growing body of research on the
volatility of financial markets. Although markets
may often seem to the casual observer to be extremely volatile, proponents of the efficient market
hypothesis claim that rapid price movements are just
a consequence of new information being rapidly incorporated into the valuation of securities.
The value of any asset is the present value of the
future cash flows that it spins off. Shiller (1989),
however, found statistical evidence that financial
markets, and particularly the stock market, are too
volatile to be explained by the behavior of these cash
flows. This finding has spawned considerable debate, and the outcome is not yet settled.

Returns Predictability
An issue related to the excess volatility of the
market was evidence that the returns to stock prices
display "mean reversion"-that is, periods of high
returns followed by periods of low returns and vice
versa. If stock prices follow a random walk, stock
returns would not tend to revert to some statistical
mean. Several studies have demonstrated a tendency, however, for stock returns to revert to some
7

average value over long periods. Early evidence of

mean reversion led to tests of the predictability of
future returns based on past returns.
Recent studies have found that stock returns are
far more predictable over long horizons than originally thought. Supplementing the evidence of positive short-run correlations of returns, this research
showed that long periods of high returns are followed by periods of low returns and vice versa. This
negative long-run correlation applies to individual
stocks as well as the market. Not only do past returns
influence future returns, but recent research has confirmed that both the price-to-earnings ratio and the
dividend yield have strong explanatory power in
long-run statistical tests.
These findings focused the debate on whether
return predictability represents movements of the
market away from fundamental intrinsic value or
whether it is related to long-term changes in expected
returns, perhaps associated with changes in real economic conditions. The fact that these long-term market movements appear correlated across assets and
economies does little to resolve the debate. Clearly,
more sophisticated models of asset pricing than the
CAPM are needed to identify the determinants of the
expected returns on financial assets.

Further Developments in Asset Pricing

In an attempt to explain the empirical anomalies

found in security prices, theoretical work has continued to generalize the CAPM. Extensions include
restrictions on short selling, an economy that does
not possess a risk-free asset, the effects of taxes, and
the effects of transaction costs. Although these refinements have led to somewhat different allocations
among assets, they have not changed the substantive
results of the CAPM.
The empirical inadequacies of the traditional
CAPM and, later, the difficulty of identifying the
market portfolio led Ross (1976) to offer an alternative theory of asset pricing. Ross proposed that a
limited number of unspecified economic and financial factors, which could include production, interest
rates, and inflation, drive asset returns. He showed
that, unless the expected returns of individual assets
bore a specific relationship to these factors, arbitrage

possibilities would exist. This approach is called

arbitrage pricing theory (APT).
APT generates a variance model focused on volatility-how to measure volatility in a portfolio and
how to price volatility factors. It can be shown to be
equivalent to the CAPM under certain assumptions.
The underlying theory is that the factors against
which investors wish to hedge are far too complicated to be summarized by some ill-defined market
portfolio. Although empirical tests of APT multifactor models have been promising, the theory does not
rest on as strong a theoretical foundation as the
CAPM.

Where Beta Stands

If an investor is concerned about the variance of
portfolio returns, the betas of individual assets provide useful information, because betas have a role in
measuring a portfolio's volatility. Betas measure the
contribution of an asset to the overall volatility.
Using betas to measure volatility does not imply or
reIy on the validity of the CAPM, so betas can be used
to measure volatility without having a valid CAPM.
In fact, if expected returns and betas are not related
and if a portfolio is to be evaluated solely in terms of
expected returns and variances, the investor would
invest in low-beta stocks and have much less volatility and the same expected return as with a portfolio
with a lot of volatility.
Explaining expected returns requires variables
in addition to beta. Beta is a measure of volatility, but
other factors relate to pricing an asset: tax status,
liquidity, and the ability to hedge against changes in
the consumer price index. If, for example, unanticipated inflation occurs, what will happen to the
price of a long-term bond? One would like an asset
to go up in value in a time of unanticipated inflation.
The behavior of the assets under inflation will thus
be a factor influencing expected returns on assets.

Conclusion
The controversy over the CAPM has many ingredients; some may be palatable, and some not. The
CAPM is like a menu: You do not have to like everything in order to have a good meal.

Question and Answer Session

Marshall E. Blume
Question: Why do you use the
term "expected return" when referring to historical realized returns?
Blume: I am not doing that. If
I want to run Markowitz's efficient set or a tactical asset-allocation model, I want to put in my
best judgment about the future
values of the parameters. One
way of getting those future values is by looking at historical
data; ultimately, however, what
values go into the model is a judgment call, not a simple extrapolation of historical numbers. In
fact, 20 or 30 years of data provide little insight into expected returns of a volatile asset. I would
much rather use a dividend discount model to assess an expected return on equities. With Tbills, at least, I can get an expected return. These models
must use subjective judgments
tempered by the historical past.
Question: To what extent does
increased use of derivatives in equity and fixed-income portfolios
impair the usefulness of the
CAPM for portfolio construction
and return analysis?
Blume: The inclusion of derivatives does not impair the usefulness of these models. In fact, it
probably increases their usefulness, because one of the implicit
assumptions in CAPM-based
models is zero transaction costs,
and the derivatives reduce the
cost of constructing portfolios.
For example, if I wanted a portfolio to have half the volatility of
the S&P 500, I could meet this
goal traditionally by placing 50
percent of my money in the S&P
and 50 percent in safe, risk-free,

assets. Alternatively, if all my

funds were in the S&P 500, I
could sell a sufficient number of
futures contracts on the S&P 500
to reduce the volatility. Mathematically, that move is the same
as investing in the risk-free asset.
If I have a manager who is good
at the S&P risk and adds 1 percent of alpha, then I could swap
out of the S&P Index into the riskfree asset and pick up the entire
alpha. Using the other technique,
I get only half the alpha. Derivatives dramatically expand the horizon.

Question: Do you believe in

time diversification? How does it
enter into the risk-measurement
and beta controversies?
Blume: In time diversification,
the geometric or annual compound rates of return will converge over time into a fixed number. One must blow that up to
see what actually happens to
wealth. Even though the compound annual rates of return contain small errors, when these errors are compounded over long
periods, they result in great differences in final wealth. The more
subtle case for time diversification is that the returns have negative autocorrelation: If you lose
today, you can make it up tomorrow.
Question: You commented on
the empirical finding that the
CAPM line is too flat. Could the
reason be that we are assuming
too Iowa rate on the risk-free
asset?
Blume: That possibility is certainly one rationalization people
have used. In the CAPM, the risk-

free rate is clearly related to the

revision horizon of the portfolio
manager. If a portfolio manager
revises a portfolio every quarter,
then a quarterly risk-free rate
must be used. Even using certificates of deposit instead of T-bills
does not produce enough of a
bang to explain the flatness.

Question: Is semivariance a
more useful risk measure than a
two-sided statistic?
Blume: If a portfolio does
much better than expected, calling that excess return "risk" does
not seem normal, but that information should not be excluded
from a risk analysis. Investors
can gain information about the
downside potential of an investment from the upside, even
though the upside is pleasant.
Whenever a manager has a realized return much better than normal, the chances are good that the
same money manager in another
market could have a return much
less than expected. Because semivariance throws out some very
valuable data, I almost always
prefer to use the variance rather
than a semivariance estimate.
Question: Options and futures
may reduce transaction costs, but
they also create opportunities for
abnormal distribution rather than
a straight mean-variance distribution. Please comment.
Blume: This issue goes to the
heart of the use of Markowitz's efficient-set calculations.
Markowitz assumed that the only
two parameters of a distribution
that matter are the expected return and the variance of that return. To obtain that result theo9

retically, returns must be normally distributed. If they are, the

mean and variance are all that
count. In real investments, other
things may count. For example,
the Pennsylvania State Lottery is
a hard investment to justify on
the basis of a mean-variance
model. The expected return is
about -100 percent. It could be
justified in a mean-variance skewness model, however, because a
small probability of a very high
payoff exists, in which case the
mean-variance viewpoint is inadequate. In many situations, a simple mean-variance model is inadequate and other parameters of
the return distribution must be reviewed. Derivatives are a cheap
way to obtain different return distributions, which might be related to parameters other than the
S&P Index and so forth.

Question: Did you say APT has

no economic underpinning?
Blume: APT is primarily a theory of volatility; the CAPM is
based on a utility framework.
APT assumes that returns are distributed by a model with a limited number of common factors.
It is silent as to which factors, but
given those factors, the model

10

then shows that the elimination

of arbitrage between different
portfolios of assets requires certain relationships between the expected returns on the assets and
the response coefficients to the
common factors of the individual
securities. That is what I consider
APT.
The interesting question
about APT and the CAPM is this:
What are the measures of volatility? The description of these common factors is not really part of
APT. How returns are generated
and what factors are important in
explaining rates and return requires a separate theory. With
those factors, APT places strong
restrictions on how expected returns are related to the response
coefficients.

Question: Taxes on investment

returns are a function of cash
flow, holding period, and asset
class. Taxes are applied differently to the price and income return components of total return.
Why apply tax rates to total returns prior to optimization rather
than after? Should it be done in a
multiperiod planning framework?
Blume: The original Markowitz model exists in a one-pe-

riod world, so a blended tax rate

could be applied to any particular
asset. If more of the income came
from capital gains, assumed to be
realized at the end of the period,
then the model would use a different tax rate from that applied
to an ordinary bond, on which almost all of the return would be ordinary income.
A current major development
on Wall Street with personal investing is to determine the optimal times to realize capital gains
and losses. To do so requires a
dynamic programming approach.
Some of the large New York
firms are already developing
such products for their private clients. Particularly if our tax rates
increase, the difference between
capital gains and ordinary income could make these dynamic
models much more important
than they now are.
The CAPM is a rarefied
model that assumes certain behavior. Using a utility-based
framework to maximize expected
utility while minimizing the impact of taxes is nevertheless a
valid approach, even if the
CAPM does not work. Beta
would still matter in that framework, because the investor is worried about volatility.