Professional Documents
Culture Documents
Lecture 6
Lecture 6
Index model
Fall, 2023
2. Index model
2.1 The market index model
2.2 Alpha and beta
Part 1. Non-diversifiable and Diversifiable risk
Part 1.1. Variance of a portfolio’s returns
1. Non-diversifiable and Diversifiable risk 2. Index model
N 2 N N 2
X 1 X X 1
σp2 = σi2 + Cov (Ri , Rj )
N N
i=1 i=1 j=1,j̸=i
N N N
!
1 1 X 2 N −1 1 X X
= σi + Cov (Ri , Rj )
N N N N(N − 1)
i=1 i=1 j=1,j̸=i
1 2 N −1
= σ + Cov (Ri , Rj )
N i N
A N grows large,
σp2 −→ Cov (Ri , Rj )
If we further simplify to the case by assuming that all stocks have the
same variance and all the pairwise covariances are equal
▶ σi = σj = σ for all i and j
▶ Cov (Ri , Rj ) = cov for all (i, j)
Cov (Ri ,Rj ) cov
▶ ⇒ pairwise correlation is also the same σi σj = σ2
We can then simplify the previous equation further:
1 2 N −1
σp2 = σ + cov
N N
σp2 1 N − 1 cov
= +
σ2 N N σ2
σp2 cov
As N grows big, σ2
→ σ2
= Corr (Ri , Rj ), for any pair (i, j)
Ways simplify the estimation process (to make fewer estimations) are
desired
Assume:
There exists a single macro factor that:
▶ Summarizes all relevant macro condition
▶ Moves the market as a whole
▶ Is the only source of returns correlation
Imagine, your favorite stock goes up 10%. Why? This realized return has
three components:
1 The fair expected return for that stock, foreseeable level ex-ante
2 News about in the market as a whole, and
3 News affect this stock alone
Ri = E0 [Ri ] + mi + ei
ei is specific to stock i:
▶ Formally, Corr (ei , ej ) = 0 for any stock j other than i
Cov (mi , ei ) = 0
mi is too abstract
Economists want more structure to formulate the index model:
▶ The market index: used to capture the macro condition, or the
common source of risk, MKT
▶ Beta representation: used to express mi in a linear way, βi MKT
mi is too abstract
Economists want more structure to formulate the index model:
▶ The market index: used to capture the macro condition, or the
common source of risk, MKT
▶ Beta representation: used to express mi in a linear way, βi MKT
Formally,
▶ Rie : stock i’s excess return
e
▶ Rm : market index (portfolio)’s excess return
▶ The market index model formula of Rie is:
Rie = αi + βi Rm
e
+ ϵi
Return decomposition:
e
▶ βi R m : market-related return
▶ ϵi : non-market related return
▶ αi : part of expected return
αi = E [Rie ] − βi E [Rm
e
]
Random: Rie , Rm
e , ϵ ; constant: α , β
i i i
By construction,
e
Cov (ϵi , Rm ) = 0 and E [ϵi ] = 0
PN
αp = i=1 ωi αi
PN
βp = i=1 ωi βi
PN
ϵp = i=1 ωi ϵi
The market index has a standard deviations of 22% and the risk-free rate
is 8%. Suppose that we were to construct a portfolio with proportions:
A 30%
B 45%
T-bill 25%
The market index has a standard deviations of 22% and the risk-free rate
is 8%. Suppose that we were to construct a portfolio with proportions:
A 30%
B 45%
T-bill 25%
The market index has a standard deviations of 22% and the risk-free rate
is 8%. Suppose that we were to construct a portfolio with proportions:
A 30%
B 45%
T-bill 25%
e e
Ri,t = αi + βi Rm,t + ϵi,t
e
▶ αi is the regression line’s intercept; part of the expected Ri,t
▶ βi is the regression line’s slope
▶ ϵi,t is the regression residuals
e ,R e )
Cov (Ri,t
e , Re ) = √
Given Corr (Ri,t √ m,t
m,t e )
Var (Ri,t e )
Var (Rm,t
e )
Std.Dev .(Ri,t e e
βi = e ) × Corr (Ri,t , Rm,t )
Std.Dev .(Rm,t
e ,R e )
Cov (Ri,t
e , Re ) = √
Given Corr (Ri,t √ m,t
m,t e )
Var (Ri,t e )
Var (Rm,t
e )
Std.Dev .(Ri,t e e
βi = e ) × Corr (Ri,t , Rm,t )
Std.Dev .(Rm,t
e ] − β E [R e ]
Alpha estimates: αi = E [Ri,t i m,t
R2 e )
βi2 Var (Rm,t
=
1 − R2 Var (ϵi,t )
R 2 = {Corr (Ri,t
e e
, Rm,t )}2
R 2 = {Corr (Ri,t
e e
, Rm,t )}2
Economically:
▶ R 2 describes how much portion of a stock’s variation comes from the
macro factor
▶ βi describes how sensitive of a stock’s return is to the macro factor’s
movement
Now you want to estimate the second size decile’s beta, r-squared,
the correlation with the market
You know:
▶ σ(Rm ) = 0.051
▶ σ(Rdec2 ) = 0.056
▶ Corr (Rm , Rdec2 ) = 0.88
Calculate:
▶ βdec2 =?
Now you want to estimate the second size decile’s beta, r-squared,
the correlation with the market
You know:
▶ σ(Rm ) = 0.051
▶ σ(Rdec2 ) = 0.056
▶ Corr (Rm , Rdec2 ) = 0.88
Calculate:
▶ residual standard deviation: σ(ϵdec2 ) =?
Calculate:
▶ residual standard deviation: σ(ϵdec2 ) =?
σ 2 (Rdec2 )= βdec2
2
σ 2 (Rm ) + σ 2 (ϵdec2 )
⇔ σ 2 (ϵdec2 )= σ 2 (Rdec2 ) − βdec2
2
σ 2 (Rm )
= 0.0562 − 0.96632 × 0.0512 = 0.0007
√
⇔ σ(ϵdec2 )= 0.0007 = 2.66%
Now you want to estimate the second size decile’s beta, r-squared,
the correlation with the market
You know:
▶ σ(Rm ) = 0.051
▶ σ(Rdec2 ) = 0.056
▶ Corr (Rm , Rdec2 ) = 0.88
Calculate:
▶ R-squared of the market index model for decile 2?
Now you want to estimate the second size decile’s beta, r-squared,
the correlation with the market
You know:
▶ σ(Rm ) = 0.051
▶ σ(Rdec2 ) = 0.056
▶ Corr (Rm , Rdec2 ) = 0.88
Calculate:
▶ R-squared of the market index model for decile 2?
R 2 = Corr (Rdec2 , Rm )2
= 0.882
= 0.7744
However, stocks’ true betas are also changing - so we cannot use the
entire history of a stock to estimate OLS beta.
In practice, with monthly data, 5-7 years seems to give good trade-off
between the two
▶ 5-7 years allow greater accuracy arising from more observations
▶ also higher accuracy to capture time-varying true betas
In practice, with monthly data, 5-7 years seems to give good trade-off
between the two
▶ 5-7 years allow greater accuracy arising from more observations
▶ also higher accuracy to capture time-varying true betas
We can also use higher frequency returns to estimate betas, e.g., daily
returns
▶ allows a shorter window in calendar time
▶ It can be problematic for stocks that are not very liquid