Sei sulla pagina 1di 40

ChapLer 13 Lmplrlcal 1esLs of

Lqulllbrlum Models
1he ModelsLxAnLe LxpecLaLlons and LxosL 1esLs
Lmplrlcal 1esLs of Lhe CAM
1esLlng Some AlLernaLlve lorms of Lhe CAM Model
1esLlng Lhe osL1ax lorm of Lhe CAM Model
Some 8eservaLlons abouL 1radlLlonal 1esLs of Ceneral
Lqulllbrlum 8elaLlonshlps and Some new 8esearch
Concluslon
Appendlx 8andom Lrrors ln 8eLa and 8las ln Lhe
arameLers of Lhe CAM
The Models Ex-Ante Expectations and Ex-Post
Tests
The Modelsare based on expectations. All variables are expressed in terms
of future values.
However all tests of the CAPM have been performed using ex-post or
observed values for the variables.
The simple form of CAPM model is:
Testing this model ex-post requires three assumptions be satisfied:
1. The market model holds in every period.
2. The CAPM model holds in every period.
3. The beta is stable over time.
) (
Z M i Z i
R R R R .
) (
M i i
R R R R .
it Mt i it
e R R R R )
~
(
~
.
Lmplrlcal 1esLs of Lhe CAM
Some Hypotheses of the CAPM
1. High risk (Beta) should be associated with high
return
2. Linear relationship between beta and return
3. No added return for bearing non-market risk.
A Slmple 1esL of Lhe CAM
Sharpe and Copper (1972) sLudy Lo lnvesLlgaLe
Lhe relaLlonshlp beLween 8eLa and reLurn
Figure 15.1
Estimated Security Market Line.
Sharpe and Copper (1972) SLudy
Test of whether shares with higher betas carry higher returns.
New York Stock Exchange, time period: 1931-1967.
Construct portfolios with different betas
1. First regressing individual asset returns on market returns (based on
five years).
2. Once a year all shares were divided into ten categories ranked by
their beta.
3. A portfolio for each category was formed (equally weighted).
4. nvestment strategy: hold shares in one category only, over the entire
period (1931-1967).
5. Regress portfolio return on its beta: Ri = a
1
+ a
2
b
i
+ q
i
They find a
1
=5.54 and a
2
=12.75.
Sharpe and Copper (1972) SLudy
Conclusions:
Shares with higher betas generate higher returns: a2>0.
Buying shares with higher forecasted betas lead to holding
portfolios with higher betas (i.e. betas are stable over time).
More than 95% in the variation in expected return
explained by differences in beta (looks as if there is a linear
relationship between beta and realized average return).
The intercept greater than riskless rate of return
(5.54>2%).
The slope (a
2
) is smaller than the market premium.
Lintner (reproduced by Douglas (1968)
First regression
R
it
= d
i
+ b
i
R
mt
+ e
it
yearly returns, 301 common stocks, 1954-1963
Second regression R
i
= a
1
+ a
2
b
i
+ a
3
S
2
ei
+ q
i
S
2
ei
is the variance of the fitted residual from the first regression.
f the standard CAPM is true then a
1
should equal R
f
,
a
2
should equal Rm-Rf, and a
3
should equal zero.
f the Zero-Beta CAPM is true then a1 should equal R
z
, a
2
should equal (R
m
-R
z
)
and a
3
should equal zero.
The result was
a1 = 0.108
a2 = 0.063
a3 = 0.237 (both a
2
and a
3
are significantly different from zero at the 0.01 level)
Seems to violate the CAPM!
Empirical Evaluation of CAPM
Long-run average returns are significantly related to beta:
The dots show the actual average risk premiums from portfolios with
different betas.
high beta portfolios generated higher average returns
high beta portfolios fall below SML
low beta portfolios land above SML
a line fitted to the 10 portfolios
would be flatter than SML.
(Source: Fisher Black, "Beta and return.)
Empirical Evaluation of CAPM
(Source: Fisher Black, "Beta and return.)
Empirical Evaluation of CAPM
The dots show the actual average risk premiums from portfolios with
Different betas over different periods. The relation between beta and
actual average return has been much weaker since the mid-1960s.
CAPM does not seem to work well over the last 30 years:
(Source: Fisher Black, "Beta and return.)
Fama and French and The Three Factor Model
Fama and French examined 9,500 stocks between 1963
and 1990, concluding that a stock's risk, measured by beta,
was not a reliable predictor of performance.
They found factors describing "value (HML) and "size
(SML) to be the most significant factors, outside of market
risk, for explaining the realized returns of publicly traded
stock.
.
S
measures the level of exposure to size risk and .
H
measures the level of exposure to value risk.
ML SMB R R R R
S M i i
. . . ) (
Fama and French SMB and HML Factors
%he SMB Factor: Accounting for the Size
Premium
SMB, which stands for Small Minus Big, is
designed to measure the additional return
investors have historically received by
investing in stocks of companies with
relatively small market capitalization.
Fama and French SMB and HML Factors
HML, which is short for High Minus Low, has been
constructed to measure the "value premium provided to
investors for investing in companies with high book-to-
market values
FAMA and FRENCH AND THE THREE FACTOR
MODEL
Fama and French-Factors other than beta seem important in pricing assets:
Source: G. Fama and K. French, "The Cross-Section of Expected Stock Returns.
FAMA and FRENCH AND THE THREE FACTOR
MODEL
FAMA and FRENCH AND THE THREE FACTOR
MODEL
nterpretations of the Factors
n reality, the SMB and HML factors first drew attention
and continue to be the most commonly used simply
because they workthey have shown a greater
predictive power of any two additional factors that
researchers have tested often yielding an R
2
value of
approximately 0.95. %hat
Being said, causal explanations for SMB are appealing
from a theoretical perspective, but for HML, the labeling
of it as a "risk factor has spurred much discussion.
Problem with Fama and French Three Factor Model
Robert D. Coleman- Asset Pricing Simultaneity, Three-
Factor Model and Cost Analysis, ndian Journal of
Economics and Business, 2005
He argues that size, value and other pricing entailing
factors are not scientifically valid in an asset pricing
model of return estimated and rested by statistical
methods.
n his 2006 paper, he shows that asset pricing models of
return with risk factors that entail either shares or
dividends are logically circular simultaneities and thus
are fallacious, meaningless, non-interpretable,
indeterminate and not valid when tested and estimated
by scientific statistical methods
roblems wlLh MeLhodology
AnoLher paper (Paugen and Pelns 8lsk and
Lhe 8aLe of 8eLurn on llnanclal AsseLs Some
Cld Wlne ln new 8oLLles concluded 1he
resulLs of our emplrlcal efforL do noL supporL
Lhe convenLlonal hypoLhesls LhaL rlsk
sysLemaLlc or oLherwlse generaLes a speclal
reward
The Black-Jensen-Scholes (1972) Test
A. The errors-in-variables problem
1. f 5-10 years of data is used to estimate .
i
, the estimate is very noisy.
2. Using longer estimation period won't work because .
i
's change over
time.
3. Solution is to form portfolios, whose .
i
's can be estimated much
more precisely.
4. Problem is how to choose the composition of the portfolios:
f portfolios are formed randomly, portfolio .
i
's will all be close to 1, and
test will lack power.
Need a method to get sufficient variation in the portfolio .
i
's.
The Black-Jensen-Scholes Test
B. BJS Portfolio selection technique (slightly modified):
1. Each year, calculate .
i
's using a regression of each securities return on the
market for the past five years.
2. Form 10 portfolios based on estimated .
i
's
Smallest portfolio has the 10% of the stocks with the lowest estimated .
i
's, etc.
3. Re-estimate the portfolio .
i
's after the formation period.
The second regression:
gave
a
1
= 0.00359
a
2
= 0.01080
it t Mt i it
e R R a R R ) (
1
.
it i it
e a a R R .
2 1
Figure 15.2
Lxcess 8eLurn versus 8eLa
BJS Ex Post Security Market Line (SML)
ResuIts of the BJS Study
The results of the study appeared to be consistent with
the zero beta version of the Capital Asset Pricing Model
(CAPM):
f the standard CAPM was true they should have
found d
i
= 0 for all deciles (portfolios).
However, they found di negative (positive) for large
(small) betas. %his is consistent with the Zero-Beta
CAPM.
Also note that Black, Jensen, and Scholes, estimated
the ex post Security Market Lines in various sub-
periods. n general, the results were similar.
1esLs of lama and Mac8eLh
Sample: All stocks on the NYSE (1926 - 1968)
Market ndex: Equally weighted portfolio of all stocks on the NYSE.
OutIine of the Study
1. Estimate a beta for each stock using monthly returns during the period,
1926 - 1929, (i.e., 48 months).
2. Rank order all of the stock betas, and form 20
portfolios. The top 5% with the highest betas are in
portfolio #1, . . . etc. . . the bottom 5% with the
smallest betas are in portfolio #20.
3. Estimate the beta of each of the portfolios by regressing portfolio monthly
returns against the market index during the period, 1930 - 1934, (i.e., 60
months).
4. For each of the months during the period, 1935- 1938 estimate the ex
post SML by regressing portfolio returns against portfolio betas.
Note: 48 SMLs will be estimated (one for each month).
Summary of Steps 1 Through 4
1926 - 1929
Used to estimate
stock betas and
form 20 portfolios
1930 - 1934
Used to estimate
beta of each of
the 20 portfolios
1935 - 1938
Used to estimate
48 SMLs (One for
each month)

5. For each of the months during the period, 1935 - 1938, estimate two
additional equations:
stocks the oI variance residual Average
m
) (c o
RV : where
return portIolio aIIects stocks oI
variance residual her test whet to c RV a a a a r
ty nonlineari Ior test to c a a a r
m
1 j
j
2
p
p p 3
2
p 2 p 1 0 p
p
2
p 2 p 1 0 p

The FM results appeared to be consistent with the zero beta


version of the CAPM:
a
0
was significantly greater than the mean of the
risk-free interest rate.
a
1
was significantly different from zero. The slope
of the SML was positive.
a
2
and a
3
were not significantly different from zero
(i.e., there was no evidence of nonlinearity or of
residual variance affecting returns).
Difference between BJS and FM Studies
n BJS, betas and average returns were computed
in the same periods.
n FM, betas in one period were used to predict
returns in a later period.
The FM Equation
0
0.1
0.2
0.3
0.4
0 0.2 0.4 0.6
r
j,t
= A
j
+ .
j
r
M,t
+1
j,t
r
j,t
r
M,t
.
j
Roll's Critique of Tests of the CAPM
Tests like those of BJS and FM are
tautological (not necessary). Results like
those reported could be obtained irrespective
of how securities were priced relative to risk.
(e.g., pulling numbers out of a hat would
produce the same results). Therefore, the
CAPM was not really tested. Furthermore,
since we cannot operationally define the
market portfolio, the CAPM can never be
tested.
t is true that if the market portfolio is efficient, the
relationship between expected return and beta will be
perfectly linear and positively sloped:
0
0.25
0 0.48
0
0.25
0 0.5 1 1.5
E(r)
E(r
M
)
E(r
z
)
E(r
M
)
E(r
z
)
E(r)
9(r)
.
SML
CML
M
C
B
A
C
M
B
A
However, given a linear relationship between portfolio
return and beta, it does not necessarily follow that the
market portfolio is efficient. Therefore, we have not tested
the CAPM.
For example, if betas are computed with
reference to an index portfolio inside the
minimum variance set, the relationship
between security betas and expected returns
will not be linear. Portfolios, however, will plot
closer to the SML because the security
residuals will be averaged in the portfolios.
Therefore, using an inefficient market portfolio
(M'):
(See the graphs that follow)
For ndividual Securities
0
0.25
0 0.48
0
0.25
0 0.5 1 1.5
E(r)
E(r
M
)
E(r
z
)
M
9(r)
E(r
M
)
E(r
z
)
.
E(r)
M'
For Portfolios
0
0.25
0 0.48
0
0.25
0 0.5 1 1.5
E(r)
E(r
M
)
E(r
z
)
M
9(r)
E(r
M
)
E(r
z
)
.
E(r)
M'
Empirical Evaluation of CAPM
WhaL can conclude Slnce mld1960s
- 5moll stocks bove ootpetfotmeJ lotqe stocks
- 5tocks wltb low totlos of motkettobook voloe
bove ootpetfotmeJ sLocks wlLh hlgh raLlos
1he relaLlonshlp beLween reLurn and beLa was
essenLlally flaL Lo negaLlve
1he |mpact of surv|va| b|as does not fu||y exp|a|n the
re|at|onsh|p found between performance and the
pr|ce to book va|ue rat|o
On the Other Hand . . Some Argued . . .
Fama and French study suffers from "survival
bias.
Many stocks with low ratios of price to
book value are in financial stress and wind
up failing. These stocks were excluded
from the Fama and French data. This
produced upward bias in the performance
of low price to book value stocks as a
class.
StiII Others Argued . . .
1he |mpact of surv|va| b|as does not fu||y exp|a|n the
re|at|onsh|p found between performance and the pr|ce
to book va|ue rat|o
Irrespect|ve of the pr|ce to book va|ue rat|o surv|va| b|as
cannot be used to exp|a|n the f|at to negat|ve
re|at|onsh|p between return and beta
Comment on Testing the CAPM
1he ev|dence d|scussed above does not prove
that the CAM |s |nva||d s|nce on|y stocks were
|nc|uded |n the ana|yses 1he "Market ortfo||o"
conta|ns a|| of the cap|ta| assets |n the un|verse
We w||| never be ab|e to observe the returns on
the "true" Market ortfo||o 1herefore the
CAM |s s|mp|y not a testab|e theory
Some Additional Evidence
Estimated betas are very sensitive to the
market index being used.
n risk-return space, indices can be close to
each other, and close to the efficient set, and
still produce different relationships (positive
and negative) between return and beta.
One study suggested using a dual beta
approach to adjust for risk differences in bull
and bear markets.
Some Conflicting Findings in Foreign Stock Markets
A slze effecL and a !anuary effecL exlsLed on Lhe
1okyo SLock Lxchange
lound a slgnlflcanL slze effecL on Lhe Mexlcan
SLock Lxchange
no slgnlflcanL slze effecL was found on Lhe
1oronLo SLock Lxchange
very llLLle slze effecL was found uslng AusLrallan
daLa
Empirical Evaluation of CAPM
Summary of CAPM
CAPM is attractive:
1. t is simple and sensible:
is built on modern portfolio theory
distinguishes systematic risk and non-systematic risk
provides a simple pricing model.
2. t is relatively easy to implement.
CAPM is controversial:
1. t is difficult to test:
difficult to identify the market portfolio
difficult to estimate returns and betas.
2. Empirical evidence is mixed.
3. Alternative pricing models might do better.
SUMMARY OF THE Empirical Study
The empirical studies generally agree upon the following results:
a) The intercept, d
1
, is significantly different from R
f
.
b) There are studies showing that when including other variables in
the regression, their coefficients are significantly different from zero.
c) f including a squared beta in the regression, its coefficient is not
significantly different from zero.
d) The coefficient of beta a
2
is less than Rm - Rf.
e) When the equation is estimated over long periods of time a
2
> 0.