CAPM Cheatsheet

Charlie Wang
April 29, 2009

1 CAPM Background
Builds on mean-variance model of Markowitz (1959)

– Mean-variance utility Investor chooses portfolio at t 1 with stochastic return at t.

– Rational investors choose mean-variance e¢ cient portfolios (p), satisfying the following minimum variance
condition:

E (Ri ) = E (RZP ) + iP [E (RP ) E (RZP )] 8i = 1; :::; N (1)

where E (Ri ) = expected return of any asset in the investment opportunity set, E (RP ) = expected return of any min-
imum variance frontier portfolio, E (RZP ) = expected return on an asset with 0 correlation to the portfolio, and RiP
Cov (Ri ; RP ) =V ar (RP )
Sharpe (1964) and Lintner (1965) add 2 assumptions

– Complete Agreement: Given market clearing asset prices at t 1, investors agree on the joint distribution of asset
returns from t 1 to t, and this is the true distribution from which returns are drawn.
– Borrowing and lending at risk-free rate: All investors can borrow/lend unlimited at the risk-free.

Punchline:

– Complete agreement ) investors see the same opportunity set, and they combine the same risky tangency
portfolio with risk-free lending or borrowing.
– By market clearing of asset market, since everyone holds the same risky asset ) the tangency portfolio must be
the value-weight market portfolio of risky assets (i.e. the "Market" portfolio)!
– Unlimited borrowing and lending at risk-free ) that expected return on assets uncorrelated with market return
E (RZM ) = Rf .
– Thus we obtain the famous equation:

1
E (Ri ) = Rf + iM [E (RM ) Rf ] 8i = 1; :::; N (2)

1 There are 2 standard interpretations of CAP M

1. = sensitivity of asset’s return to variation in market return
2. It can be shown that for a portfolio p s.t. Rp = ! T R;
@V ar (Rp )
= Cov (Ri ; Rp )
@! i
So we can think of CAPM in the following way:
E (Rm ) Rf
E (Ri ) Rf = Cov (Ri ; Rm )
| {z } V ar Rf
c o ntrib u tio n o f i to p o rtfo lio risk | {z }
risk -re tu rn tra d e o ¤ in m k t

1
Black (1972) develops a version of the CAPM without risk-free borrowing or lending, but rather obtains the same result
(market portfolio is MV e¢ cient) by allowing for unrestricted short sales of risky assets.

– Black di¤ers from Sharpe-Lintner in E (RZM )

Black: E (RM ) E (RZM ) > 0
Sharpe-Lintner: E (RZM ) = Rf .
– In either case, the Sharpe-Lintner-Black CAPM ) positive premium for .

Sharpe-Lintner and Black versions of CAPM share the following predictions

– Market portfolio is mean-variance e¢ cient

– has positive premium

CAPM and SDF

– In general, beta representation with f as factors (see below) ,a¢ ne SDF: m = a+bT f: (See Cochrane p.106~108)
Beta representation takes the following form:
K
X
ei =
E R + ik k 8i (3)
|{z} |{z}
i=1
factor sens. factor risk pr

K
X
ei = ai +
R ik fek + e
"i 8i (4)
|{z} |{z}
i=1
factor sens. risk factor

"i ) = 0 and E feke

with E (e "i Cov fek ; e
"i = 0 (5)

– Form of : From the above formulation It can be shown that (See Cochrane p.108)...
if the factor fk is a gross return on a portfolio (e.g. CAPM), and there exists risk-free asset ) factor risk
premium always takes the form of a risk premium on that portfolio, i.e. k = E R ek ef
R
if the factor fk is an excess return on a portfolio ) factor risk premium always takes the form of the expected
excess return, i.e. k = E fek
– Examples:
CAPM: k=1, fe = R
em , em
=E R Rf ; = Rf ; i
ei ; R
= Cov R em =V ar R
em

FF3factors: fe1 = R
em Rf ; fe2 = R
eS eB ; fe3 = R
R eH eL ,
R k = E fek 8k
– Note on Factor Mimicking Portfolios (See Cochrane p.109)
When factors are not already returns or excess returns, it is often convenient to express a pricing model in
terms of its factor-mimicking portfolios rather than the factors themselves (FF3Factors).
That is, rather than the true factors f , use one of the following...

factor-mimicking payo¤s: f = proj (f jX) (6)

proj (f jX)
factor-mimicking gross returns: f = (7)
p (proj (f jX))
factor-mimicking excess returns: f = proj (f jRe ) (8)
e
where X is the space of payo¤s in the economy and R the space of excess returns.
m = bT f is a valid SDF, since

p = E (mx) = E (proj (f jX) x) = E (m x) (9)

2
2 Empirical Tests of CAPM
2.1 Methodology
2 Approaches for testing beta models (e.g. CAPM)

1. Time-Series Approach (FF ’93)

When factor is a excess return, then factor risk premium is = E fe so that representation in (3) reduces to

ei
E R Rf = iE fe (10)

where i de…ned through he time-series regression

Ri;t Rf;t = i + i ft + "i;t (11)

The model has one and only one implication for the data:

i = 0 8i (12)

To Test: Run time-series regressions (11) for each test asset/portfolio, then test whether all pricing errors ( i ) are
0
Problems with methodology: Requires the factor to be a return.

2. Cross-Sectional Approach (FF ’92)

Generally, if risk factors are not returns, our representation model is written as:

ei
E R Rf = i + i (13)

We can …t this model as follows:

– 1. Identify iM from time series regression for each security

regress Rit Rf t = i + iM (RM;t Rf t ) + "it (14)

– 2. Run the following cross sectional regression of (time-series) average asset returns on estimates of asset betas
^ for each stock i:
iM

regress Ri Rf = + ^ iM + ui (15)

CAPM Predictions in this model are that: = 0 and > 0 (Black) or eM

=E R Rf (Sharpe-Lintner)

Problems with above methodology

– 1. estimates for individual assets are imprecise )measurement error.

(Solution: Estimate in portfolios improve precision. Blume (1970), Friend and Blume (1970), Black, Jensen,
and Scholes (1972). )
– 2. Cross sectional regression (15) has correlated errors (e.g. common industry e¤ects in average returns) )
standard errors too small!
(Solution: Fama-MacBeth (1973) - estimate month-by-month cross-section regressions of monthly returns on s,
use the time-series means and standard errors of the means for inference. Assuming no time-series correlation, the
standard errors of coe¢ cients capture the e¤ects of residual correlations via repeated sampling of the regression
coe¢ cients. In other words, we are bootstrapping from the sampling distribution of ’s, where each bootstrap
draw comes from a monthly cross-section.)
– 3. Second stage standard errors does not account for …rst stage estimation ) standard errors too small!
(Solution: Explicitly correct for the …rst stage estimation in standard error - Shanken (1992))

3
2.2 Findings in Tests of CAPM
3 Testable Predictions of CAPM to test: 1) expected returns on all sets are linearly related to their , and no other variables
have explanatory power, 2) premium is positive, and 3) asset uncorrelated with market is Rf and premium is E (RM ) Rf
or +.

Early tests of the CAPM (Fama MacBeth (1973), Gibbons (1982), and Stambaugh (1982)) …nd...

– appears to su¢ ce in explaining expected returns and premium on is +

– too large and too small (see (15))

Later tests of the CAPM (Fama French (1992)) reject earlier …ndings, using cross-sectional Fama-MacBeth approach

– does not explain cross section of average returns. + premium does not exist in the post 1963 period.
– Size and B/M explain cross section of average returns