You are on page 1of 9

Fama Macbeth

Anish S. Menon

March 16, 2015

Anish S. Menon

Fama Macbeth

March 16, 2015

1/9

Fama and Macbeth (1973) developed a method by which to test how risk
factors describe portfolio or asset returns. The main goal of the
Fama-Macbeth two step regression is to find the premium from exposure
to these factors.

Anish S. Menon

Fama Macbeth

March 16, 2015

2/9

In the first step, each portfolios return is regressed against one or


more factor time series to determine how exposed it is to each one of
the factor exposures.
In the second step, the cross section of portfolio returns is regressed
against the factor exposures, at each time step, to give a series of risk
premia coefficients for each factor.
The insight of Fama-Macbeth is to then average these coefficients,
once for each factor, to give the premium expected for a unit
exposure to each risk factor over time.

Anish S. Menon

Fama Macbeth

March 16, 2015

3/9

In equation form, for n portfolio or asset returns and m factors, in the first
step, the factor exposure s are obtained by calculating n regressions, each
one on m factors (each equation in the following represents a regression).
R1,t = 1 + 1,F1 F1,t + 1,F2 F2,t + + 1,Fm Fm,t + 1,t
R2,t = 2 + 2,F1 F1,t + 2,F2 F2,t + + 2,Fm Fm,t + 2,t
..
.
Rn,t = n + n,F1 F1,t + n,F2 F2,t + + n,Fm Fm,t + n,t

(1)

where Ri,t is the return of the portfolio or asset i (n total) at time t, Fj,t
is the factor j (m total) at time t, i,Fm are the factor exposures, or
loadings, that describe how returns are exposed to the factors, and t goes
from time t through T . Notice that each regression uses the same factors
F , because the purpose is to determine the exposure of each portfolios
return to a given set of factors.
Anish S. Menon

Fama Macbeth

March 16, 2015

4/9

The second step is to compute T cross-sectional regressions of the returns


calculated from the first step.
on the m estimates of the s (call them )
Notice that each regression uses the same s from the first step, because
now the goal is the exposure of the n returns to the m factor loadings over
time (e.g., does a larger factor exposure mean a higher return?).

Anish S. Menon

Fama Macbeth

March 16, 2015

5/9

The second stage regression is as follows:


Ri,1 = 1,0 + 1,1 i,F1 + 1,2 i,F2 + + 1,m i,Fm + i,1
Ri,2 = 2,0 + 2,1 i,F1 + 2,2 i,F2 + + 2,m i,Fm + i,2
..
.

Ri,T = T ,0 + n,1 i,F + n,2 i,F + + n,m i,F + i,T


1

(2)

where the returns R are the same as those used in the first stage
regression, are regression coefficients that are used to calculate the risk
premium for each factor, and in each regression i goes, from 1 through n.

Anish S. Menon

Fama Macbeth

March 16, 2015

6/9

In the end there are m + 1 series (including the constant in the


second step), for every factor, each of length T . If the  are assumed
to be independently and identically distributed (i.i.d.), calculate the
risk premium m for factor Fm by averaging the mth over T , and
also get standard deviations and t statistics.

T
1 X
t
=
T

i =

Anish S. Menon

1
T

(3)

t=1
T
X

i,t

(4)

t=1

Fama Macbeth

March 16, 2015

7/9

The standard errors are calculated as follows


T
1
1 X
(
) = var (
) = 2
(t )2
T
T
2

(5)

t=1

T
1 X
1
cov (
) = cov (t ) = 2
(
i,t i )(
j,t j )
T
T

(6)

t=1

This methodology ensures that the standard errors are corrected for
cross-sectional correlation. However it does not correct for time series
autocorrelation since it assumes that yearly estimates of the coefficients
are independent over time. A correction as proposed by Peterson(2009) is
to follow a double clustering methodology where there is clustering by
both firm and time.

Anish S. Menon

Fama Macbeth

March 16, 2015

8/9

The t statistics for the mth risk premium:


t statistic =

m / T

(7)

This follows a student t distribution with T 1 degrees of freedom.


The Fama Macbeth regression is a rolling regression.

Anish S. Menon

Fama Macbeth

March 16, 2015

9/9