Lecture Note 3

(BKM Ch. 9)
Investments - Insper
Ruy M. Ribeiro
Chapter 9: Capital Asset Pricing
Model
 We now know that investors will hold efficient
portfolios. What does this mean?

 If investors hold efficient portfolios, what sort of

relation between expected return and risk can we
expect to observe?
Capital Asset Pricing Model
 We now know that investors will hold efficient
portfolios. What does this mean?

 Investors will buy just two benchmark investments:

 The risky portfolio T
 The risk-free security (either borrowing or lending)
Capital Asset Pricing Model
 Two-fund separation:

~
E ( Rp )

T

RF

p
What is portfolio T?
 In a competitive (efficient) market, you're unlikely
to have monopolistic access to information. It is
“reasonable” to assume that all investors have
homogeneous (identical) beliefs about expected
returns for all securities.
 But this implies that everyone will hold the same
portfolio of risky assets (T) in some combination
with Rf.
What is portfolio T?
 In equilibrium supply = demand, all securities issued
and outstanding must be held.
 The previous picture implies that all investors hold
the same portfolio of risky assets.
 But this portfolio must contain all securities if the
market is in equilibrium.
 In equilibrium, tangency portfolio T is the market.
What is the market?
Simple Example
 If GE is 1% of T (xGE=0.01), then GE will
represent 1% of the risky portion of each
individual’s portfolio.
 In the aggregate portfolio, GE will represent 1% of
investors’ wealth. This is because everybody holds
GE in the same proportion.
 We have put everyone’s portfolios of invested
wealth together, this big portfolio is the market. The
portfolio T will have the same weights in each
security as the aggregate market portfolio.
Is this reasonable?
 Proliferation of index funds.
 Vanguard, ETF’s, SPDR’s

 Even small investors hold well-diversified portfolios

(of multiple mutual funds) that are highly correlated
with the market.
The Risk of a Security in a Portfolio
 Investors are concerned primarily with the risk and
return of their portfolios.
 Marginal impact of an asset on the risk and return of
the portfolio.
 Investors are not interested in the variance of the
returns for the security.
Contribution of a Security
 Contribution of security i to expected returns to
E ( R p ) is xi E ( Ri )
 To variance of the portfolio is:
N
xi  x j ij
 But j 1

 x j ij   ip
j 1
Contribution of Security to Portfolio
 Contribution of security i to variance of the portfolio:

 But xi ip

 ip
  ip
 Then, p 2

xi ip p
Contribution of Security in Equilibrium
 Contribution of security i to variance of the
portfolio:
xi im
 But
 im
  im
m 2

 Then,

xi im m
Relation between Risk and Return
 All investors hold the same portfolio of risky assets
-- the market.
 The relevant measure of security risk is the market
beta.
 Thus, investors require compensation for bearing
market beta risk.
 It is, therefore, reasonable to consider the relation
between market beta risk and expected return.
Assumptions I
 Security returns are normally distributed.
 Other assumption can replace this one, for example
when investors only worry about the mean and
variance of their end-of period wealth.
Assumptions II (secondary)
 Perfect Capital Markets; there are no frictions in the
market:
 no transactions costs;
 no personal income taxes;
 no restrictions on short sales;
 all assets are marketable;
 securities are infinitely divisible.
 Investors are price takers.
 Investors are in complete agreement regarding the
distribution of stock returns
Is Complete Agreement
Reasonable?
Proof of CAPM: Optional!

~
E ( Rp ) Capital
Market
~
E ( Rm ) M Line

rf

~ ~
 ( Rm )  ( Rp )
Proof of CAPM: Optional!
 Capital Market Line is:

 E ( RT )  R f 
E ( R p )  R f    P
 T 

 In equilibrium, slope of the efficient set = slope of

CML.
 Equating these slopes yields the CAPM
Proof of CAPM: Optional!
 What is the slope of the efficient set at M? i.e., what
is
E ( R p )
?
 p

E ( R p )  xi E ( Ri )  (1  xi ) E ( R m )
1
 p   xi2 i2  (1  xi ) 2  m2  2 xi (1  xi ) im  2
Proof of CAPM: Optional!
 What is the slope of the efficient set at M? i.e., what
is
E ( R p )
?
 p

E ( R p )
E ( R p ) xi

 p  p
xi
Proof of CAPM: Optional!
 What is the slope of the efficient set at M?

E ( R p )
 E ( Ri )  E ( Rm )
xi
 p 1 2
1

   

 p 2 2 xi i2  2(1  xi ) m2  2 im  4 xi im
xi 2
xi ( i2   m2  2 im )   im   m2

p
Proof of CAPM: Optional!
 What is the slope of the efficient set at M?

E ( R p )
E ( R p ) xi  E ( Ri )  E ( R m )   p
 
 p  p xi ( i2   m2  2 im )   im   m2
xi
Proof of CAPM: Optional!
 What is the slope of the efficient set at M?

E ( R p )  E ( Ri )  E ( Rm )   m

 p  im   m2
xi  0
Proof of CAPM: Optional!
 As the two slopes have to be equal. The slope of
efficient set at m, and right-hand side is slope of
capital market line must be equal at the point of
tangency

 E ( Ri )  E ( R m )   m  E ( R m )  rf 

 im   m 2
m
Proof of CAPM: Optional!
 E ( Ri )  E ( R m )   m2  ( im   m2 )  E ( R m )  rf 

 im
 E ( Ri )  E ( Rm )   ( 2  1)  E ( R m )  rf 
 
 m

 E ( Ri )  E ( R m )   ( im  1)  E ( R m )  rf 

 E ( Ri )  rf    E ( R m )  rf   im
 This is not a regression line!
Another (but simpler) Proof
 Two assets with the same beta must have the same
expected return. cov( R1 , Rm )
 GE’s beta is 1 

var( Rm )
Apple’s beta is cov( R 2 , R m )
 2 
var( R m )
 Suppose GE and Apple have the same beta, or
equivalently their covariances with the market are the
same.
 Let’s prove that GE and Apple must have the same
expected returns.
Another (but simpler) Proof
 Take the market portfolio M. Recall that:
N
E ( R m )   xi E ( Ri )
i 1

N
var( R m )   xi cov( Ri , R m )
i 1
Another (but simpler) Proof
 If GE’s expected return were strictly greater that
Apple’s, then instead of investing in M, consider
increasing slightly the share invested in GE (from
x1 to x1 + d) at the expense of Apple (whose share
declines from x2 to x2 - d). The new portfolio M’
has a rate of return:
N
R m '  ( x1  d ) R1  ( x2  d ) R 2   xi Ri
i 3

 R m  d ( R1  R 2 )
 Another (but simpler) Proof
 Then,

E ( Rm ' )  E ( R m )  d ( E ( R1 )  E ( R2 ))

var( R m ' )  var( Rm )  d 2 var( R1  R 2 )  2d .cov( R1  R 2 , R m )

cov( R1  R 2 , Rm )  cov( R1 , Rm )  cov( R2 , R m )  0

var( Rm ' )  var( Rm )

Comments
 Note that the CAPM is a condition of equilibrium. If
we have equilibrium, then the CAPM holds.
 The CAPM holds for all individual securities or
portfolios of securities that are contained in the market
portfolio.
Comments

~
E ( Ri ) Security
Market
Line

~
E ( Rm )
RF

0 1.0 
Comments
 Is it possible to have a negative beta? If so, what is
the risk premium in this case?
 Important: SML and CML are different relations.
 Recall that CML graphs the expected return of
optimal portfolios invested in the market portfolio
(since this is the tangent portfolio) and the T-bill,
against their standard deviations.
 The SML graphs the expected return of assets
against their beta.
Comments
 The total risk of an asset is always measured by its
standard deviation (or equivalently variance).
 However, when the asset is part of the market
portfolio, the only part of the asset’s total risk that is
being taken into account for computing its CAPM
equilibrium risk premium is its covariance with the
market.
Comments
 Indeed, in equilibrium, everybody invests in the
market portfolio (and not in assets taken
individually), therefore assets should be priced
according to their beta and not their variance.
 For example, consider an asset that has an enormous
standard deviation but zero beta.
 If all you held in your portfolio were this asset, then
you would face a very large amount of risk.
Comments
 But if you follow the portfolio theory and invest in the
market portfolio (and the T-Bill) then this asset
contributes nothing to your portfolio’s risk (because
its beta is zero).
 Therefore (in equilibrium) you will not be
compensated for holding this asset, and it will not
earn anything beyond the risk-free rate.
Alpha
Extensions of the CAPM
 Zero-Beta Model
 Helps to explain positive alphas on low beta stocks and
negative alphas on high beta stocks
 Consideration of labor income and non-traded assets
No Riskless Borrowing: Zero-Beta
Model
Extensions of the CAPM
 Merton’s Multiperiod Model and hedge portfolios
 Incorporation of the effects of changes in the real
rate of interest and inflation

 Consumption-based CAPM
 Rubinstein, Lucas, and Breeden
 Investors allocate wealth between consumption
today and investment for the future
Liquidity and the CAPM
 Liquidity: The ease and speed with which an asset can be
sold at fair market value
 Illiquidity Premium: Discount from fair market value the
seller must accept to obtain a quick sale.
 Measured partly by bid-asked spread
 As trading costs are higher, the illiquidity discount will be
greater.
The Relationship Between
Illiquidity and Average Returns
Liquidity Risk
 In a financial crisis, liquidity can unexpectedly dry up.
 When liquidity in one stock decreases, it tends to
decrease in other stocks at the same time.
 Investors demand compensation for liquidity risk
 Liquidity betas
Liquidity Risk
 In a financial crisis, liquidity can unexpectedly dry up.
 When liquidity in one stock decreases, it tends to
decrease in other stocks at the same time.
 Investors demand compensation for liquidity risk
 Liquidity betas
Roll’s Criticism
 The only testable hypothesis is whether the market
portfolio is mean-variance efficient.
 Sample betas conform to the SML relationship because all
samples contain an infinite number of ex post mean-
variance efficient portfolios.
 CAPM is not testable unless we know the exact
composition of the true market portfolio and use it in the
tests.
 Benchmark error due to proxy for M
Measurement Error in Beta
 Problem: If beta is measured with error, then the
slope coefficient of the regression equation will
be biased downward and the intercept biased
upward.
 Solution: Construct P with large dispersion of
beta. Then, by ranking them, they yield insightful
tests of the SML
 Fama and MacBeth
CAPM and World
 Academic world
 Cannot observe all tradable assets
 Impossible to pin down market portfolio
 Attempts to validate using regression analysis
 Investment Industry
 Relies on the single-index CAPM model
 Most investors don’t beat the index portfolio
How can we use the CAPM
 The CAPM gives us economic structure that
explains differences in expected returns across
securities.
 Use the CAPM to estimate expected returns
 But CAPM requires estimates of two unknown
quantities (market risk premium and beta) and could
be incorrect.
 Why not estimate expected returns with historical mean
returns?
Estimating Expected Returns
Why bother with CAPM?

 Based on the past 60 months of data for a common

stock, you estimate the following
Mean Std. Deviation
Stock Return 1.7% 12.0%
S&P-Rf 0.6% 5.0%
 You also estimate the stock’s beta running a time
series regression.

Rit  0.80  1.20 Rmt   it

(1.70) (2.70)
Estimating Expected Returns
 The current T-Bill return is 0.7% monthly.
 Using the historical average return, you would
estimate the expected return as 1.7%.
 Using CAPM your estimate is:

E ( Rit )  0.70  1.20(0.6)  1.32%

Which Estimate is Better?
 One response might hinge on the relative precision
of the estimates.
 compare standard errors of the two estimates.
 For historical mean:
s2 122
std .error    1.55%
T 60
 For CAPM-based estimate:

std .error  std .deviation( ˆ .rm )

Which Estimate is Better?
 Then,

Var ( ˆi .rm )  Var ( ˆi ).Var ( rm )  ˆi 2Var ( rm )  rm 2Var ( ˆi )

 52
  52

Var (i .rm )  (.27)    (1.2)    (0.6) .27  0.6566
ˆ 2 2 2 2

 60   60 
 So that std. dev. is 0.81%
 Standard error of CAPM-based estimate of the stock's expected
return is about half of standard error of the historical average return
because the greater precision in the estimate of the market risk
premium more than offsets the additional estimation error in beta.
How well does the CAPM describe
average returns?

E ( Ri )  RF   E ( R m )  RF  im (Sharpe-Lintner)

E ( Ri )  E ( R Z )   E ( R m )  E ( R Z )  im (Black)

 If the model is correct and security markets are

efficient, then returns will on average conform to
such a linear relation.
 How well do these models describe the behavior of
capital markets?
How well does the CAPM describe…
 Testable Implication: Is the market mean-variance
efficient?

(1) Is the risk-return relation linear?

(2) Does the risk-return relation have positive slope?
(3) Does the return on a zero-beta portfolio equal RF?
How well does the CAPM describe…
 Alternative Hypotheses:
 Does β’s play a role? (Is this relation curvilinear?)
~
E (R )

 Are investors compensated for bearing nonmarket risk? i.e.,

does unique or firm-specific risk affect rewards?
Testing CAPM - Traditional
Approach
 For all the securities in the market, estimate a first
pass regression to estimate the beta of security i, bi
(using, say, 60 months of data)

Rit  ai  bi Rmt   it for t=1,2,…,T

 If the market is MV-Efficient, then the expected
returns on the N securities will be linearly related to
the bi’s:
for n=1,2,…,N
E ( Ri )   0   1bi
Testing CAPM - Traditional
Approach
 If Sharpe-Lintner CAPM is correct,
 0  RF
 1  E ( Rm )  RF
 To test these implications, regress sample security (or
portfolio) returns on estimated betas in the following
second-pass regression

Ri   0   1bi  ui
Testing CAPM - Traditional
Approach
 This regression is estimated in each of many time
periods, and average values of the ’s. Tests are then
conducted to assess whether the average 0 is 0 or RF
and the average 1 >0.
Testing CAPM - Traditional Approach
 Issues to consider in implementing this two-pass
approach:
 CAPM is ex ante model; how valid are tests with ex post
data?
 Measurement error in individual security betas.
 The use of portfolios reduces the estimation error of the
betas in the first-pass regressions.
 But forming portfolios can conceal violations of the model
that might be detectable using individual stocks
Testing CAPM - Traditional Approach
 Early evidence was interpreted as supportive of the
CAPM,
 But the intercept was too large
 And the slope too flat
 Market beta is marginally important in explaining
cross-sectional differences in average returns.
Testing CAPM - Traditional Approach
 For example, from Fama and MacBeth (1973):
1 2
1935-1968 0.0061 (3.24) 0.0085 (2.57)
1935-1945 0.0039 (0.86) 0.0163 (1.92)
1946-1955 0.0087 (3.71) 0.0027 (0.70)
1956-1968 0.0060 (2.45) 0.0062 (1.73)

 Note: (t-stats)
Testing CAPM - Traditional Approach
Testing CAPM - Traditional Approach
 Expected rates of return are linear and increase with
beta, the measure of systematic risk.
 Expected rates of return are not affected by
nonsystematic risk.
Testing CAPM
 Testing for Alternative Hypotheses
N
Ri   0   1bi    j cij ui
j 2

 where cij’s can be any characteristic (size, earnings-

price ratio, price-book ratio, dividend yield, etc.).
 If the CAPM is correct, then the time-series average
of j =0 for j>1.
Testing CAPM
 Recent Cross-Sectional Tests of CAPM alternatives
find:
N
Ri   0   1bi    j cij ui
j 2

 Significant explanatory power for variables not

predicted by the model (i.e., the time-series average of
j ≠0 for j>1).

 Insignificant role for beta

Testing CAPM - Critiques
 Subset of all securities contained in the market
 CAPM must always hold with respect to ex post
data if the proxy chosen for the market is ex post
efficient.
 The only real test of the CAPM requires
measurement of the true market portfolio, which is
impossible.
 Stambaugh (1982) adds corporate and governmental
bonds real estate consumer durables. Conclusions
aren't materially affected.