Factor Model

Finance: Lecture 5 - Factor models Chapters 6,7,13 of DD, Chapters 8-9 of Cochrane (2001)
Prof. Alex Stomper

MIT Sloan, IHS & VGSF
March 2010
Alex Stomper (MIT, IHS & VGSF)
Finance
March 2010
1 / 22
Generic factor models
This class
Generic factor models Example: the CAPM Conditional vs. unconditional models The Arbitrage Pricing Theory
Finance
March 2010
2 / 22
A generic factor model

We pose that expected returns are determined by: E[Rj ] = + j where and the vector are/contain coecients and is determined by the regression: Rj,t = aj + j Ft + j, t where F is a vector of random risk factors. Why is there substance to this model?
Finance
March 2010
3 / 22
A generic factor model

We pose that expected returns are determined by: E[Rj ] = + j where and the vector are/contain coecients and is determined by the regression: Rj,t = aj + j Ft + j, t where F is a vector of random risk factors. Why is there substance to this model? The regression model allows for cross-sectional variation in the intercept. The coecient will be the same for all assets. If there is a risk-free asset, the return of this asset will have zero exposure to any risk-factors and = rf .
Alex Stomper (MIT, IHS & VGSF) Finance March 2010 3 / 22
Factor representation of pricing kernels

We have already seen a factor representation of the CCAPM: ERj = + j,M M How can we, more generally, relate pricing kernels and factor models? Proposition: Given the model M = a + b(F E[F ]) we can nd and such that E[Rj ] = + j where j are coecients of a regression of Rj on a constant and the factors F . Conversely, given and , we can nd a and b such that M = a + b(F E[F ]).
Proof
Since E[Rj ] =
Cov[M, Rj ] 1 , E[M] E[M]
M = a + b(F E[F ]) implies that

E[Rj ] = = = 1 E[Rj F ]b a a 1 E[Rj F ](E[FF ])1 E[FF ]b a a 1 j E[FF ]b = + j a a
(because j = (E[FF ])1 )E[FRj ]).
Finance
March 2010
5 / 22
Example: the CAPM
This class
Finance
March 2010
6 / 22
Example: the CAPM
The CAPM
A capital asset pricing model (CAPM) is a model in which: Mt+1 = at + bt RW ,t+1 Such a model can be derived in a number of models. We will cover: A two period quadratic utility model Rubinsteins (1976) log utility model Mertons (1973) intertemporal capital asset pricing model ICAPM
Finance
March 2010
7 / 22
Example: the CAPM
A two period quadratic utility model

Investors maximize: 1 1 (ct c) E[(Ct+1 c)2 ] 2 2 s.t. Ct+1 = Wt+1 , Wt+1 = RW ,t+1 (et ct ), and RW ,t+1 = wi Ri,t+1 .
Mt+1 = = Ct+1 c RW ,t+1 (et ct ) c = ct c ct c c (et ct ) + RW ,t+1 = at + bt RW ,t+1 ct c ct c
Now, use the last result: E[Ri,t+1 ] = t + i,t t where i,t = (Et [RW ,t+1 RW ,t+1 )1 Et [RW ,t+1 Ri,t+1 ]. How can we nd ?
Example: the CAPM
The market risk premium
Et [RW ,t+1 ] is determined by: E[RW ,t+1 ] = t + 1t Thus: E[Ri,t+1 ] = t + i,t (E[RW ,t+1 ] t ) where t is the expected return of a zero-beta asset. If a risk-free asset exists, t = rf ,t .
Finance
March 2010
9 / 22
Example: the CAPM
Rubinsteins (1976) log utility model

Dene the wealth portfolio W as a claim to all future consumption. What is the price of this portfolio if the RAs utility from consumption c is ln[c]? u [Ct+ ] Mt+ = u [ct ]

pW ,t = Et
=0
M Ct+ = Et
=0
ct Ct+ = ct Ct+ 1
What is the return on the wealth portfolio? pW ,t+1 + Ct+1 = = pW ,t

1
+ 1 Ct+1
1 ct
RW ,t+1
1 Ct+1 1 = ct Mt+1
Finance
March 2010
10 / 22
Example: the CAPM
Discussion
So, Mt+1 = 1/RW ,t+1 . Why do we get this result? Because the eects of news of higher future consumption Ct+ on the value of the wealth portfolio are oset by changes in the discount factor Mt+ . How about a factor representation? The discrete time log utility model can only represented as Mt+1 = at + bt RW ,t+1 by means of a Taylor approximation. An exact linearization is often possible in continuous time since diusion processes are locally normally distributed - we will see this when we talk about the ICAPM. Or, we assume normally distributed returns and factors...
Finance
March 2010
11 / 22
Example: the CAPM
Steins lemma
Steins lemma: If F and R are bivariate normal, is a dierentiable function and E[| [F ]|] < , then Cov[[F ], R] = E[ [F ]]Cov[F , R]
Finance
March 2010
12 / 22
Example: the CAPM
Factor models for normally distributed returns

Suppose that Mt+1 = [Ft+1 ]. Then, for Xt+1 = pt Rt+1 :
pt = Et [Mt+1 Xt+1 ] = Et [[Ft+1 ]Xt+1 ] = Et [[Ft+1 ]]Et [Xt+1 ] + Covt [[Ft+1 ]Xt+1 ] = Et [[Ft+1 ]]Et [Xt+1 ] + E[ [Ft+1 ]]Cov[Ft+1 , Xt+1 ] = = 1 = Et [(Et [[Ft+1 ]] + E[ [Ft+1 ]](Ft+1 , Et [Ft+1 ])Xt+1 ] Et [(Et [[Ft+1 ]] E[ [Ft+1 ]]Et [Ft+1 ] + E[ [Ft+1 ]]Ft+1 )Xt+1 ] Xt+1 Et (Et [[Ft+1 ]] E[ [Ft+1 ]]Et [Ft+1 ] + E[ [Ft+1 ]]Ft+1 ) pt
i.e.
Mt+1 = = Et [[Ft+1 ]] E[ [Ft+1 ]]Et [Ft+1 ] + E[ [Ft+1 ]]Ft+1 at + bt Ft+1
for any dierentiable function .

Example: the CAPM
Mertons (1973) Intertemporal CAPM

Investors maximize: V [et , t ] = max u[ct ] + Et [V [Wt+1 , t ]]
ct ,wt
where Wt+1 = RW ,t+1 (et ct ), RW ,t+1 = wt Rt+1 and t are state variables that may describe relative price changes, changes in the investment opportunity set, etc. We will consider a continuous time model. Recall: Et Assume that t = exp[t]VW [Wt , t ]. dt Wt VWW [Wt , t ] dW VW [Wt , t ] = dt + + dt t VW [Wt , t ] W VW [Wt , t ]
dpj,t dj,t dt dpj,t + dt = rf ,t dt Et pj,t pj,t t pj,t
Example: the CAPM
A factor model
Let: RRt = Then,

Et dpj,t dj,t + dt pj,t pj,t = = Et [Rj,t+1 ]
Wt VWW [Wt , t ] . VW [Wt , t ]
dt dpj,t t pj,t VW [Wt , t ] dpt dWt dpj,t Et dt rf ,t dt + RRt Et Wt pj,t VW [Wt , t ] pt rf ,t + RRt Covt [RW ,t+1 , Rj,t+1 ] + t Covt [t+1 , Rj,t+1 ] rf ,t dt Et
Finance
March 2010
15 / 22
Conditioning
This class
Finance
March 2010
16 / 22
Conditioning
Conditional and unconditional models

Recall: pj,t = Et [Mt+1 Xj,t+1 ] We can condition down: E[[pj,t ] = E[Mt+1 Xj,t+1 ] since your forecast of your forecast at t is your current forecast. Does the same work for factor models? Suppose 1 = Et [(at + bt Ft+1 )Rj,t+1 ]
1 = Et [at ]Et [Rj,t+1 ]+Et [bt ]Et [Ft+1 Rj,t+1 ]+Covt [at , Rj,t+1 ]+Covt [bt , Ft+1 Rj, so, we need that Covt [at , Rj,t+1 ] = Covt [bt , Ft+1 Rj,t+1 ] = 0 for 1 = Et [(Et [at ] + Et [bt ]Ft+1 )Rj,t+1 ]
Conditioning
Implications
A model that implies a conditional linear factor model with respect to a particular information set that cannot be observed. As a consequence, such models cannot be tested. Examples: the two period quadratic utility model, models based on normally distributed returns (Steins lemma),... Exception: models like the log-utility model (M = 1/RW ) can be tested using GMM. A partial solution: we could try to model the variation in at and bt in Mt+1 = at + bt Ft+1 : at = at and bt = bt Mt+1 = at + bt Ft+1 is a model with factors t , Ft+1 , t Ft+1 . The variables zt are referred to as instruments.
The APT
This class
Finance
March 2010
19 / 22
The APT
A statistical model
Suppose that asset returns can be described by the following factor decomposition: Rj = aj + j F + j Rj = E[Rj ] + j (F E[F ]) + j where E[
j k]
= 0. (This restriction is the real content of the model.)
Finance
March 2010
20 / 22
The APT
No arbitrage
Lets assume that we can set up well-diversied portfolios s.t. RP = wR E[wR] + P F where w is the vector of portfolio weights and R is a vector of asset returns. No arbitrage requires that:
1 0 0 E[M](E[RP ] rf ) E[RP ] rf E[RP ] rf w E[R] rf E[Rj ] rf
= = = = = = = =
E[MRP ] E[M(RP rf )] E[M(E[RP ] + P F rf )] P (E[MF ]) E[MF ] P E[M] P w j

Finance March 2010 21 / 22
The APT
Factor risk premia
Lets construct a portfolio Qi with a unit beta vector Qi (i.e. a vector the ith component of which equals one while all other components are zero). Line 6 of our last derivation implies: E[RQi ] rf = i Thus, E[Rj ] rf =
i
j,i (E[RQi ] rf )
One open question remains: what are the factors?
Finance
March 2010
22 / 22

Factor Model

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Factor Model

Hochgeladen von

Copyright:

Verfügbare Formate

Finance: Lecture 5 - Factor models Chapters 6,7,13 of DD, Chapters 8-9 of Cochrane (2001)

Prof. Alex Stomper

Alex Stomper (MIT, IHS & VGSF)

Generic factor models

Alex Stomper (MIT, IHS & VGSF)

Generic factor models

A generic factor model

Alex Stomper (MIT, IHS & VGSF)

Generic factor models

A generic factor model

Generic factor models

Factor representation of pricing kernels

Generic factor models

Cov[M, Rj ] 1 , E[M] E[M]

M = a + b(F E[F ]) implies that

(because j = (E[FF ])1 )E[FRj ]).

Alex Stomper (MIT, IHS & VGSF)

Example: the CAPM

Alex Stomper (MIT, IHS & VGSF)

Example: the CAPM

Alex Stomper (MIT, IHS & VGSF)

Example: the CAPM

A two period quadratic utility model

Example: the CAPM

The market risk premium

Alex Stomper (MIT, IHS & VGSF)

Example: the CAPM

Rubinsteins (1976) log utility model

What is the return on the wealth portfolio? pW ,t+1 + Ct+1 = = pW ,t

Alex Stomper (MIT, IHS & VGSF)

Example: the CAPM

Alex Stomper (MIT, IHS & VGSF)

Example: the CAPM

Alex Stomper (MIT, IHS & VGSF)

Example: the CAPM

Factor models for normally distributed returns

for any dierentiable function .

Example: the CAPM

Mertons (1973) Intertemporal CAPM

dpj,t dj,t dt dpj,t + dt = rf ,t dt Et pj,t pj,t t pj,t

Example: the CAPM

Let: RRt = Then,

Wt VWW [Wt , t ] . VW [Wt , t ]

Alex Stomper (MIT, IHS & VGSF)

Alex Stomper (MIT, IHS & VGSF)

Conditional and unconditional models

Alex Stomper (MIT, IHS & VGSF)

= 0. (This restriction is the real content of the model.)

Alex Stomper (MIT, IHS & VGSF)

E[MRP ] E[M(RP rf )] E[M(E[RP ] + P F rf )] P (E[MF ]) E[MF ] P E[M] P w j

Factor risk premia

One open question remains: what are the factors?

Alex Stomper (MIT, IHS & VGSF)

Das könnte Ihnen auch gefallen