Chapter 4: Multifactor Pricing Models

Contents

1 Review of the CAPM 2

1.1 Sharpe-Linter CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Black CAPM (Zero-Beta CAPM) . . . . . . . . . . . . . . . . . . . 3

1.3 Empirical Implementation of CAPM . . . . . . . . . . . . . . . . . 4

2 Multifactor Pricing Models 5

3 Estimation and Testing 6

3.1 Time-Series Regressions . . . . . . . . . . . . . . . . . . . . . . . . 7

3.1.1 Sharpe-Lintner CAPM . . . . . . . . . . . . . . . . . . . . . 7

3.1.2 Black CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.1.3 Multifactor Pricing Models . . . . . . . . . . . . . . . . . . . 11

3.2 Cross-Sectional Regressions . . . . . . . . . . . . . . . . . . . . . . 17

3.3 Fama-Macbeth Approach . . . . . . . . . . . . . . . . . . . . . . . . 17

3.4 Summary of Fama and MacBeth (1973) . . . . . . . . . . . . . . . . 20

1
1 Review of the CAPM
This chapter draws on chapter 5 and chapter 6 in Campbell et al. (1997).

Markowitz (1959) cast the investor's portfolio selection problem in terms of

expected return and variance of return. He argued that invetors would optimally

hold a mean-variance ecient portfolio, that is, a portfolio with the highest

expected return for a given level of variance. This seminal research laid the

groundwork for the CAPM.

1.1 Sharpe-Linter CAPM

Sharpe (1964) and Lintner (1965) showed that if investors have homogeneous

expectations and optimally hold mean-variance ecient portfolios then, in the

absence of market frictions, the portfolio of all invested wealth, or the market

portfolio, will itself be a mean-variance ecient portfolio. The usual CAPM

equation implies the mean-variance eciency of the market portfolio.

The Sharpe-Lintner CAPM assumes

1. All agents have homogenous expectations.

2. Agents maximize expected return relative to the standard deviation of the

portfolio.

3. Agents can borrow or lend unlimited amounts at the riskfree rate.

4. The market is in equilibrium at all times.

For this version of the CAPM we have for the expected return of any asset i,

E[Ri ] = Rf + βi (E[Rm ] − Rf ) (1.1)

Cov[Ri , Rm ]
βi = ,
Var[Rm ]

2
where Rm is the return on the market portfolio, and Rf is the return on the riskfree

asset. Let Zi represent the return on the ith asset in excess of the riskfree rate,

Zi ≡ Ri − Rf . Then

E[Zi ] = βi E[Zm ] (1.2)

Cov[Zi , Zm ]
βi = ,
Var[Zm ]

where Zm is the excess return on the market portfolio of assets (i.e. the market

risk premium). Equations (1.1) and (1.2) are equivalent, because riskfree rate

is constant. However, in empirical implementations, proxies for the riskfree rate

are stochastic and thus the beta can dier. Most empirical work relating to the

Sharpe-Lintner CAPM employs excess returns and thus uses 1.2.

1.2 Black CAPM (Zero-Beta CAPM)

Black (1972) derived a more general version of the CAPM. This version, known

as the Black version or the zero-beta version, does not assume the existence of

riskfree rate. The expected return of asset i in excess of the zero-beta return is

linearly related to its beta.

E[Ri ] = E[R0 ] + βi (E[Rm ] − E[R0 ]) (1.3)

= E[R0 ](1 − βi ) + βi E[Rm ]

Cov[Ri , Rm ]
βi = ,
Var[Rm ]

where Rm is the return on the market portfolio, and E[R0 ] is the return on the zero-
beta portfolio. This portfolio is dened to be the portfolio that has the minimum

variance of all portfolios uncorrelated with m. Any other uncorrelated portfolio

would have the same expected return, but a higher variance.

3
1.3 Empirical Implementation of CAPM

The CAPM is a single-period model; hence equations (1.2) and (1.3) do not have

a time dimension. For econometric analysis for the model, we need to add an

assumption about the time-series behaviour of returns and estimate the model

over time. We assume that returns are independently and identically distributed

through time and jointly multivariate normal. The assumption applies to excess

returns for the Sharpe-Lintner version and to real returns for the Black versions.

The usual estimator of beta in the Sharpe-Lintner version is the OLS estimator

of the beta in the regression equation

Zit = αi + βi Zmt + it , (1.4)

where i denotes the asset and t denotes the time period. Zit and Zmt are the

realized excess returns in time period t for asset i and the market portfolio m,
respectively.

In the black version, the estimator of beta is the OLS estimator of the beta in

the following regression

Rit = αi + βi Rmt + it (1.5)

Rit and Rmt are the realized returns in time period t for asset i and the market

portfolio m, respectively.

Typically, for the analysis of US stocks, the Standard and Poor's 500 index

proxies for the market portfolio, and the US Treasury bill rate proxies for the risk

free return. The equations are most commonly estimated using 5 years of monthly

data.

The CAPM can be useful for applications requiring a measure of expected stock

returns. Applications include cost of capital estimation, portfolio performance

evaluation, and event-study analysis. For example, given an estimate of the beta,

4
the cost of equity capital is calculated using a historical average for the excess

return on S&P 500 over Treasury bills. Note that this is only justied if the

CAPM provides a good description of the data.

2 Multifactor Pricing Models

The CAPM can be thought of as a one-factor model, i.e. the market portfolio is

a factor which determines the price of securities. However, empirical evidence

indicates that the market beta does not completely explain the cross section

of expected asset returns. Theoretical arguments also suggest that more than

one factor is required, because the CAPM is a single-period model and only

under strong assumptions will it apply period by period. Two main theoretical

approaches, the Arbitrage Pricing Theory (APT) developed by Ross (1976) and

Intertemporal Capital Asset Pricing Model (ICAPM) developed by Merton (1973)

allow for more than one risk factor. Hence, these models are more general than

the CAPM.

The APT assumes that markets are competitive and frictionless. A frictionless

market is one that has no transaction costs (e.g. taxes) and no restrictions on

trade (e.g. short sale constraints). A competitive market is one where any agent

can buy or sell unlimited quantities of an asset without changing the asset's price.

The ATP assumes that that returns are generated according to

R = α + Bf +  (2.1)

E[|f ] =0
0
E[ |f ] = Σ

where R is an (N × 1) vector of asset returns with R = [R1 R2 · · · RN ]0 , α is

an (N × 1) vector of intercepts, B is an N ×K matrix of parameters with B=

5
[b1 b2 · · · bK ]. f is a (K ×1) vector of factor realizations with f = [f1 f2 · · · fK ]0 .
The parameters in matrix B represent the sensitivities of each asset to the factor

realizations (i.e. the beta-values). For example, b1 is vector of sensitivities of

each asset to the rst factor f1 .

Given this structure, Ross (1976) shows that the absence of arbitrage in large

economies implies that

E[R] ≈ ιγ0 + BλK (2.2)

where E[R] is the N × 1 expected return vector; γ0 is the model zero-beta portfolio

expected return and is equal to the riskfree return if such an asset exists, and λK
is a K×1 vector of factor risk premia. Since the APT relation (2.2) is only an

approximation, it is not possible to test the relation explicitly.

The aproximation becomes exact when we impose additional structure. For

example, the additional requirements proposed by Connor (1984) are that the

market portfolio is well-diversied and that the factors are pervasive. Hence, we

can analyze models where we have exact factor pricing, that is,

E[R] = ιγ0 + BλK (2.3)

The factors can be but need not be traded portfolios. Some macroeconomic

variables have been shown to be important pricing factors empirically. Also, most

empirical implementations choose a proxy for the market portfolio as one factor.

3 Estimation and Testing

Given the asset pricing models (1.2), (1.3) and (2.3), we will focus on three

implications of a candidate asset pricing model.

1. We will analyze if a candidate asset pricing model is sucient for explaining

the expected asset returns by testing the signicance of the model's pricing

6
 error, which is the deviation from the asset return from the expected return

implied by the model.

2. We will examine if the assets' loadings on the factors can explain the

variations in expected returns across assets.

3. Finally we will test if the factor risk premia are signicantly dierent from

zero. If a factor is not a hedge against a background risk source then we

expect that its risk premium should be positive.

The rst test is usually based on a time series regression analysis (TSR), i.e.

a regression on (joint) time-series data for some rms/portfolios. For the second

test we often use the cross-sectional regressions approach (CSR), i.e. a regression

on data for dierent rms, which can be repeated for dierent time periods. The

third test, depending on how we model the factor risk premium, can be analyzed

by both the CSR and the TSR approaches. Sections 3.1 and 3.2 will present,

respectively, the TSR approach and the CSR approach.

3.1 Time-Series Regressions

This section presents the time-series regression approach of estimating and testing

asset pricing models. We will rely on the maximum likelihood (ML) approach for

a system of time-series regressions for N assets ( or portfolios).

3.1.1 Sharpe-Lintner CAPM

We assume that investors can borrow and lend at a riskfree rate of return. Dene

Zt as an N ×1 vector of excess returns for N assets (or portfolios). For these N

assets, the excess returns can be described using the following unrestricted excess

7
return market model:

Zt = α + βZmt + t (3.1)

E[t ] =0
0
E[t t ] =Σ

E[Zmt ] = µm

E[(Zmt − µm )2 ] = σm
2

Cov[Zmt , t ] =0

Zmt is the time period t market portfolio excess return. β is the N ×1 vector of

betas, and α and t are N ×1 vectors of asset return intercepts and disturbances,

respectively.

The Sharpe-Lintner CAPM implies that all elements of the vector α are zero.

This implication follows from comparing the unconditional expectation of equation

(1.2) to equation (3.1). Non-zero α indicates deviation from the CAPM, therefore,
it is named pricing error. If all elements of α are zero then m is the tangency

portfolio.

To test whether all elements of α are zero, we start by using the ML approach

to develop estimators of the unrestricted model (3.1). Given the assumption that

excess returns are IID through time and are jointly normal, the log-likelihood

function of (3.1) is

NT T
ln L(α, β, , Σ) = − ln(2π) − ln |Σ| (3.2)
2 2
T
1X
− (Zt − α − βZmt )0 Σ−1 (Zt − α − βZmt )
2 t=1
T
NT T 1 X 0 −1
=− ln(2π) − ln |Σ| −  Σ t
2 2 2 t=1 t

The maximum likelihood estimators are the values of the parameters which

8
maximize ln L:

α̂ = µ̂ − β̂ µ̂m (3.3)
PT
(Zt − µ̂)(Zmt − µ̂m )
µ̂ = t=1 PT (3.4)
2
t=1 (Zmt − µ̂m )
T
1X
Σ̂ = (Zt − α̂ − β̂Zmt )(Zt − α̂ − β̂Zmt )0 (3.5)
T t=1
T
1X 0
= ˆt ˆt
T t=1

where

T T
1X 1X
µ̂ = Zt and µ̂m = Zmt .
T t=1 T t=1

It should be noted that ordinary least squares regression asset by asset lead to the
same estimators for α, µ and Σ.
When testing the Sharpe-Lintner CAPM, we can use the unconstrained

estimators above to form a nite-sample F-test statistic of the null hypothesis

H0 : α=0

against the alternative

H1 : α 6= 0.

The F-test statistic is

−1
µ̂2m

T −N −1
J= 1+ 2 α̂0 Σ̂−1 α̂ ∼ FN,T −N −1 . (3.6)
N σ̂m
2 1
PT 2
where σ̂m = T t=1 (Zmt − µ̂m ) . This test is proposed by Gibbons et al. (1989)

and is named the GRS test.

An alternative test method is the likelihood ratio (LR) test. To implement this

test, we need to estimate the parameters under the restricted model (the model

9
without intercept α), which results in a new log-likelihood value ln L∗ . The LR

test statistic is then given by

LR = −2(ln L∗ − ln L) (3.7)

The test statistics is distributed chi-square with degrees of freedom equal to the

number of restrictions. The number restrictions placed by the Sharpe-Lintner

CAPM is N, as α is restricted to be zero for N assets.

3.1.2 Black CAPM

In the absence of a riskfree asset we consider the Black version of the CAPM

in (1.3). The expected return on the zero-beta portfolio E[R0 ] is treated as an

unobservable and hence becomes an unknown model parameter. Dening the

zero-beta portfolio expected return as γ0 , the Black version is

E[Rt ] = ιγ0 + β (E[Rmt − γ0 ]) (3.8)

= (ι − β)γ0 + β E[Rmt ]

where Rt denotes the vector of real returns for N assets (or portfolios), Rmt is the

real return on the market portfolio, ι is an N ×1 vector of ones.

The Black version of CAPM can be investigated in a similar manner as the

Sharpe-Lintner version. The restricted market model implied by (3.8) is

Rt = (ι − β)γ0 + βRmt + t (3.9)

And the unrestricted market model is

Rt = α + βRmt + t (3.10)

Comparing the unconditional expectation of (3.10) with that of (3.9), we get

the null hypothesis to be tested

H0 : α = (ι − β)γ0

10
and the alternative

H1 : α 6= (ι − β)γ0 .

This restricted market model and the hypothesis are more complicated than those

of the Sharpe-Lintner version because the parameters β and γ0 enter in a nonlinear

fashion. Because of the nonlinear relation, OLS regression asset by asset is not

feasible. We estimate and test the Black CAPM using the ML approach. A

likelihood ratio can be constructed in the same manner as the test constructed

from the Sharpe-Lintner version.

LR = −2(ln L∗ − ln L) (3.11)

Notice that the degree of freedom of the null distribution is N − 1. Relative to the

Sharpe-Lintner version, the black version loses one degree of freedom because the

zero-beta expected return is a free parameter.

3.1.3 Multifactor Pricing Models

In this section we consider the estimation and testing of dierent forms of the

multifactor pricing models. The assumption for the econometric analysis of the

model is analogous to that made of the CAPM. We assume that the returns

conditional on the factor realizations are IID through time and jointly multivariate

normal. Also, it is assumed that the number of factors, K, is given, and that the

identication of these factors is known.

There are dierent cases of interest depending on the existence of a riskfree

asset and whether the factors are portfolios are not. We discuss two cases:

1. Portfolios as factors with the existence of a riskfree asset

2. Factors are not portfolios of traded assets

11
 Given the assumptions on the returns, we can use ML to estimate the following

unrestricted K -factor model in order to test the exact multifactor pricing (2.3 )

Rt = α + Bft + t (3.12)

E[t ] =0
0
E[t t ] =Σ
0
Cov[ft , t ] = 0.

Rt is an N ×1 vector of time t asset returns (or excess returns, depending on

the case), ft is a K×1 vector of time t factor values, and B is the N ×K

matrix of factor sensitivities. As is done for the CAPM, estimating (3.12) with

maximum likelihood produces the ML estimators α and B and Σ, and yields

the unconstrained log-likelihood value. The exact asset pricing relation can then

be tested using a likelihood ratio test, after having estimated the corresponding

restricted model.

Case 1: Portfolios as Factors with a Riskfree Asset

In this case factors are traded portfolios and there exists a risk free rate. The

unrestricted model (3.12) will be expressed in excess returns:

Zt = α + BZKt + t . (3.13)

Zt is an N ×1 vector of excess returns for N assets (or portfolios). ZKt is the

K ×1 vector of factor portfolio excess returns. Exact factor pricing then implies

H0 : α = 0.

Imposing this restriction and estimating the model using ML produces the

constrained log-likelihood value, ln L∗ . The hypothesis can then be tested with

the likelihood ratio statistic with N degree of freedom.

12
 Given the exact factor pricing µ = ιγ0 + BλK , we also need to estimate the

factor risk premia λK in order to estimate the expected return on a given asset.

In the case where the factors are excess returns on portfolios of traded assets,

the factor risk premia is simply given by the sample averages of the factor excess

returns:

T
1X
λ̂K = ZKt . (3.14)
T t=1

An estimator of the variance-covariance matrix of λ̂K is

T T T
1 1 X 1X 1X
V ar(λ̂K ) = Ω̂K = 2
d (ZKt − ZKt )(ZKt − ZKt )0 . (3.15)
T T t=1 T t=1 T t=1

Note that Ω̂K is the sample covariance matrix of the factors ZKt .
If the factors have been specied on the basis of theoretical arguments, we

are interested in testing whether the individual factors are priced in the market

and thus have non-zero risk premia. This can be done by testing whether λ̂k is

signicantly dierent from zero using the conventional statistic

λ̂k
φk = 1
∼ N (0, 1), (3.16)
Vd
ar[λ̂k ] 2

where Vd
ar(λ̂k ) is the element (k, k) in the matrix Vd
ar[λ̂K ].
A joint test of whether the factors are jointly priced is given by

1 0 d
φ= λ̂ V ar[λ̂K ]−1 λ̂K (3.17)
K K

Important empirical example for this case are the Fama-French three factor

model, the Carhart four factor model, and the Fama-French ve model. According

to studies such as Banz (1981),Basu (1983), Bhandari (1988), and Rosenberg

et al. (1985), rm characteristics like size (the market value of a rm's equity,

ME), leverage, earnings/price (E/P), and book-to-market equity (the ratio of the

13
book value of a rm's common stock, BE, to its market value, ME) are proxies

for nancial distress and consequently measure the rms risk. The observed

higher returns for highly leveraged, small, or high book to market rms should be

considered as compensation for additional sources of risk that are not included in

CAPM. Fama and French (1993) used rm characteristics to form factor portfolios

and proposed a three-factor model

Zt = α + βm Zmt + βSM B SM Bt + βHM L HM Lt + t (3.18)

Fama and French (1993) formed 6 value-weight portfolios on size and book-to-

market Factor SMB and HML are then constructed using the 6 portfolios. SMB

Figure 1: Six Portfolios Formed on Size and Book-to-Market (from Kenneth R.

French - Data Library)

(Small Minus Big) is the average return on the three small portfolios minus the

average return on the three big portfolios. HML (High Minus Low) is the average

return on the two value (high BE/ME) portfolios minus the average return on the

two growth (low BE/ME) portfolios. The size factor, SMB, accounts for the size

premium, which means that small stocks tend to outperform large stocks. HML

accounts for the value premium, the spread in returns between value and growth

14
stocks. Stocks with high book-to-market ratios, known as value stocks, tend to

outperform stocks with lower book-to-market values, known as growth stocks.

Fama and French (1993) and Fama and French (1996) nd that the model

captures much of the variation in average return for portfolios formed on size,

book-to-market equity and other price ratios that cause problems for the CAPM.

Fama and French (1998) show that an international version of the model performs

better than an international CAPM in describing average returns on portfolios in

13 major markets.

Further, studies such as Jegadeesh and Titman (1993) nd that strategies which

buy stocks that have performed well in the past and sell stocks that have performed

poorly in the past generate signicant positive return. From this viewpoint,

Carhart (1997) extends the Fama and French model by adding a fourth factor,

namely the momentum factor (MOM) which is equal to the dierence between

returns on one-year winners (i.e. stocks with the highest returns in the previous

12 months) and returns on one-year losers (i.e. stocks with the lowest returns in

the previous 12 months).

Zt = α + βm Zmt + βSM B SM Bt + βHM L HM Lt + βM OM M OMt + t (3.19)

In recent years, Novy-Marx (2013) and Aharoni et al. (2013), among others,

document evidence for the variation in stock returns in protability and invest-

ment. Such variation is unexplained by the Fama-French three factor model. This

leads Fama and French (2015) to develop a new model that adds protability

and investment factors to the market,size, and BE/ME factors of the three-factor

model.

The data on portfolios formed on size, BE/ME, momentum, protability and

investment and the corresponding factors can be found for the US and several

international countries and regions on Kenneth R. French - Data Library.

15
 Case 2: Factors are not traded portfolios

One class of non-portfolio factors commonly used in asset pricing is macroeconomic

variables. Variables are selected because it is believed that they have signicant

impact on the returns on risky assets. Such variables include measures of business

conditions, bond yields and measures of ination. The values of these variables are

taken to be the factor realizations Ft in the unrestricted model. The return vector

Rt is the return on the assets, and the matrix B now contains the sensitivities of

the assets to the macroeconomic factors.

The unrestricted model is

Rt = α + BFKt + t . (3.20)

In the restricted model we specify a parameter vector, γ1 , as the riskfree part of

factor, which by denition is the expected value of the factors minus the factor

risk premia:

γ1 = E[FK ] − λK . (3.21)

This gives the following restricted model,

Rt = ιγ0 + B(FKt − γ1 ) + t (3.22)

= (ιγ0 − Bγ1 ) + BFKt + t

where γ0 is the zero-B expected return. The null hypothesis associated with this

restricted model is

H0 : α = ιγ0 − Bγ1 (3.23)

which can be tested using a LR test. Under the null, the distribution has N −K −1
degrees of freedom.

16
 The factor risk premia for the case that the factors are not asset portfolios is

given by

T
1X
λ̂k = FKt − γ̂1 . (3.24)
T 1

The estimated variance-covariance matrix of the risk premia for this case is

1
Vd
ar(λ̂K ) = Ω̂K + Vd
ar[γ̂1 ] (3.25)
T

where Ω̂K is the sample estimate of the factor covariance matrix and ar[γˆ0 1 ]
Vd is

the covariance matrix of the γ̂1 vector and comes from the ML estimation of the

model.

A famous exploration of macroeconomic variables as factors is done by Chen

et al. (1986), in which

Rt = α + bM P M Pt + bDEI DEIt + bU I U It + bU P R U P Rt + bU T S U T St + t .
(3.26)

MP stands for the monthly industrial production growth, DEI is the change in

expected ination, UI is the unexpected ination, UP R is 'Baa and under' bond

portfolio return less the return on a portfolio of long-term government bonds, which

measures degree of risk aversion in the market, and UT S is the spread between

long-term government bonds and short-term T-bill rate.

3.2 Cross-Sectional Regressions

3.3 Fama-Macbeth Approach

So far, we focused on the implications of factor pricing models for α. An alternative

approach is to verify if risk sensitivities β (in CAPM) and B (in multifactor

models) can explain the cross-sectional variation in stock returns.

17
 Fama and MacBeth (1973) rst developed the cross-sectional regression

approach. The basic idea is, for each cross section, to project the returns on
the betas and then aggregate the estimates in the time dimension. Assuming that

the β is known for all the assets, the cross-sectional regression for CAPM for N
assets at time t is

Rt = γ0t ι + γ1t βm + ηt (3.27)

ι is a N ×1 vector of ones, and βm is the N ×1 vector of CAPM betas. In the

Sharpe-Lintner version, Rt stands for the N ×1 vector of excess returns for time

period t. In the Black version, Rt is the real returns. Note that γ1 here stands

for factor risk premia.

The Fama-MacBeth approach is implemented as a two-pass cross-sectional

regression.

1. In the rst pass, CAPM beta is estimated asset-by-asset using OLS. We

denote this estimator as β̂m

2. In the second pass, using β̂m and the N ×1 vector of asset (excess) returns

for each time period t, regression (3.27) is estimated by OLS for each period

t, t = 1, . . . , T , This results in T estimates of γ0t and γ1t .

Dene γ0 = E[γ0t ] and γ1 = E[γ1t ]. The implication of CAPM is γ1 > 0

(positive market risk premium in a market of risk-averse investors). In the Sharpe-

Lintner version, an extra implication is γ0 = 0 (zero intercept). Given time series

of γ0t and γ1t , we can test these implications using the usual t-test. The t-statistics

is dened as

γ̂j
t(γ̂j ) = (3.28)
σ̂γj

18
where

T
1X
γ̂j = γ̂jt (3.29)
T t=1

v
u T
u 1 X
σ̂γj =t (γ̂jt − γ̂j )2 (3.30)
T (T − 1) t=1

The distribution of t(γ̂j ) is Student t with T − 1 degrees of freedom and

asymptotically is standard normal.

The Fama-MacBeth approach can easily be modied to accommodate addi-

tional risk measures beyond CAPM. By the cross-sectional approach we are able

to test a multifactor pricing model which contains more than one risk measure and

test if the factor loading matrix B completely describe the cross-sectional variation
expected stock returns.

In the rst pass, we estimate factor sensitivities asset-by-asset using OLS. These

estimators represent a measure of the factor loading matrix B which we denote B̂ .

In the second pass, regressing asset returns on B̂ period-by-period, we estimate

factor risk premia period-by-period. The second-pass regression for the multifactor

pricing model is

Rt = γ0t ι + BγKt + ηt (3.31)

If riskfree asset exists, Rt is the N ×1 assets' excess returns and the exact

pricing relation implies that γ0t = 0. In the absence of riskfree asset, Rt is the

N ×1 real returns and the exact pricing relation implies that γ0t is the zero-beta

portfolio return. The output of the regression is a time series of estimated risk

premia γ̂Kt , t = 1, . . . , T , and γ̂0t , t = 1, . . . , T .

The Fama-MacBeth methodology, while useful, does have problems. The

regressions are conducted using betas estimated from the data, which introduces an

19
error-in-variables complication. One approach, adopted by Fama and MacBeth, is

to minimize the errors-in-variables problem by grouping the stocks into portfolios

and increasing the precision of the beta estimates. The portfolio approach,

however, also raises problem. When securities are combined into portfolios, some

of the information in the data about the relationship between risk and expected

return is lost. Portfolio formation may disguise the empirical signicance of

some relevant characteristics. Another approach is to adjust standard errors to

correct for the biases resulting from errors-in-variables. Shanken (1992) suggests

multiplying σ̂γj in (3.28) by an adjustment factor (1+(µ̂m −γ̂0 )/σˆm

2 ). However, this

approach does not eliminate the possibility that variables other than sensitivity to

market portfolio might enter spuriously in (3.31) as a result of the unobservability

of the true betas.

3.4 Summary of Fama and MacBeth (1973)

Data: stocks listed at NYSE between 1926 and 1968.

Rmt = return on the NYSE Index.

• Use the rst four years (1926-1929) of monthly return data to estimate the

market model

Rit = αit + βi Rmt + t (3.32)

and get β̂i for each stock i.

• Rank securities on the basis of β̂i and form them into 20 portfolios.

• Use the following ve years (1930-1934) of data to recompute β̂i in 1935.

These β̂i are averaged across stocks within portfolios to obtain 20 portfolio

β̂pt . t stands for months in the following year 1935. These β̂pt are used

20
 for estimating the following cross-sectional risk-return relationships in each

month in 1935.

Rpt = γ0 ι + γ1 βpt + γ2 (βpt )2 + ηt γ2 : (non-linearity?) (3.33)

Rpt = γ0 ι + γ1 βpt + γ2 (βpt )2 + γ3 IVpt + ηt (idiosyncratic risk? ) (3.34)

where IVpt is the variance of least squares residuals of eq(3.32) averaged over

stocks within portfolio p.

• Update β̂i for months in year 1936 using the 1931-1935 data. Calculate

portfolio β̂pt . Estimate the cross-sectional risk-return relationships in each

month in 1936.

• The month-by-month estimation of the cross-sectional relationships through

1935 to 1968 results in a time series of γ̂0t , γ̂1t , γ̂2t , γ̂3t .

• Use the time series of γ̂t to test the following implications of CAPM

1. E[γ0t ] = Rf t in Sharpe-Lintner CAPM.

2. E[γ1t ] > 0 (positive expected return-risk tradeo ).

3. E[γ2t ] = 0 (linearity).

4. E[γ3t ] = 0 (no systematic eect of non-β risk).

References
Aharoni, G., Grundy, B., and Zeng, Q. (2013). Stock returns and the Miller

Modigliani valuation formula: Revisiting the Fama French analysis. Journal of

Financial Economics, 110(2):347357.

Banz, R. W. (1981). The relationship between return and market value of common

stocks. Journal of nancial economics, 9(1):318.

21
Basu, S. (1983). The relationship between earnings' yield, market value and return

for nyse common stocks: Further evidence. Journal of nancial economics,

12(1):129156.

Bhandari, L. C. (1988). Debt/equity ratio and expected common stock returns:

Empirical evidence. The Journal of Finance, 43(2):507528.

Black, F. (1972). Capital market equilibrium with restricted borrowing. The

Journal of Business, 45(3):444455.

Campbell, J., Lo, A., and MacKinlay, A. C. (1997). C. mackinlay, 1997, the

econometrics of nancial markets.

Chen, N.-F., Roll, R., and Ross, S. A. (1986). Economic forces and the stock

market. Journal of business, pages 383403.

Connor, G. (1984). A unied beta pricing theory. Journal of Economic Theory,

34(1):1331.

Fama, E. F. and French, K. R. (1993). Common risk factors in the returns on

stocks and bonds. Journal of nancial economics, 33(1):356.

Fama, E. F. and French, K. R. (1996). Multifactor explanations of asset pricing

anomalies. The journal of nance, 51(1):5584.

Fama, E. F. and French, K. R. (1998). Value versus growth: The international

evidence. The Journal of Finance, 53(6):19751999.

Fama, E. F. and French, K. R. (2015). A ve-factor asset pricing model. Journal

of Financial Economics, 116(1):112.

Fama, E. F. and MacBeth, J. D. (1973). Risk, return, and equilibrium: Empirical

tests. The journal of political economy, pages 607636.

22
Gibbons, M. R., Ross, S. A., and Shanken, J. (1989). A test of the eciency

of a given portfolio. Econometrica: Journal of the Econometric Society, pages

11211152.

Jegadeesh, N. and Titman, S. (1993). Returns to buying winners and selling losers:

Implications for stock market eciency. The Journal of nance, 48(1):6591.

Lintner, J. (1965). The valuation of risk assets and the selection of risky

investments in stock portfolios and capital budgets. The review of economics

and statistics, pages 1337.

Markowitz, H. (1959). Portfolio Selection: Ecent Diversication of Investments.

J. Wiley.

Merton, R. C. (1973). An intertemporal capital asset pricing model. Econometrica:

Journal of the Econometric Society, pages 867887.

Novy-Marx, R. (2013). The other side of value: The gross protability premium.

Journal of Financial Economics, 108(1):128.

Rosenberg, B., Reid, K., and Lanstein, R. (1985). Persuasive evidence of market

ineciency. The Journal of Portfolio Management, 11(3):916.

Ross, S. A. (1976). The arbitrage theory of capital asset pricing. Journal of

economic theory, 13(3):341360.

Shanken, J. (1992). On the estimation of beta-pricing models. Review of Financial

studies, 5(1):133.

Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under

conditions of risk. The journal of nance, 19(3):425442.

23