Econ 424/CFRM 462
Single Index Model
Eric Zivot
August 19, 2014
Sharpe’s Single Index Model
= + +
= 1 ; = 1
where
are constant over time
= return on diversified market index portfolio
= random error term unrelated to
Assumptions
• cov( ) = 0 for all
• cov( ) = 0 for all 6= and
2 )
• ∼ iid (0
2 )
• ∼ iid (
Interpretation of
= + +
cov( )
= = 2
var()
captures the contribution of asset to the volatility of the market index
(recall risk budgeting calculations).
Derivation:
cov( ) = cov( + + )
= cov( ) + cov( )
= var() since cov( ) = 0
cov( )
⇒ =
var()
Interpretation of :
= − −
• Return on market index, captures common “market-wide” news.
• measures sensitivity to “market-wide” news
• Random error term captures “firm specific” news unrelated to market-
wide news.
• Returns are correlated only through their exposures to common “market-
wide” news captured by
Remark:
• The CER model is a special case of Single Index (SI) Model where = 0
for all = 1
= +
In this case, = [] =
• In the CER model there is only one source of news
• In the Single Index model there are two sources of news: market news and
asset specific news
Single Index Model with Matrix Algebra
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1 1 1
⎜ .. ⎟ ⎜ ⎟ ⎜
.. ⎠ + ⎝ .. ⎟ ⎜ . ⎟
⎝ ⎠=⎝ ⎠+⎝ . ⎠
or
R = + +
×1 ×1 ×1 1×1 ×1
where
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1 1 1
R = ⎜
⎝
.. ⎠ = ⎝ .. ⎟
⎟ ⎜ ⎜
⎠ =⎝
.. ⎠ = ⎝ .. ⎟
⎟
⎜
⎠
Statistical Properties of the SI Model (Unconditional)
= + +
• = [] = +
• 2 = var() = 2
2 + 2
2
• = cov( ) =
• ∼ ( 2) = ( + 2
2 + 2 )
Derivations:
var() = var( + + )
= 2var() + var() + 2cov( )
= 2var() + var() (assume cov( ) = 0)
= 2
2 + 2
where
2
2 = variance due to market news
2 = variance due to non-market news
Next
= cov( )
= cov( + + + + )
= cov( ) + cov( )
+ cov( ) + cov( )
= cov( )
2
=
Implications:
• = 0 if = 0 or = 0 (asset i or asset j do not respond to market
news)
• 0 if 0 or 0 (asset i and j respond to market news
in the same direction)
• 0 if 0 and 0 or if 0 and 0 (asset i and j
respond to market news in opposite direction)
Statistical Properties of the SI Model (Conditional on )
= + +
Given that we observe, =
• [| = ] = +
2
• var(| = ) =
• cov( | = ) = 0
2 )
• | = ∼ ( +
Decomposition of Total Variance
= + +
2 = var() = 2
2 + 2
total variance = market variance + non-market variance
Divide both sides by 2
2
22
1= 2 + 2
= 2 + 1 − 2
where
2 22
= 2 = proportion of market variance
1 − 2 = proportion of non-market variance
Sharpe’s Rule of Thumb: A typical stock has 2 = 30%; i.e., proportion of
market variance in typical stock is 30% of total variance.
Return Covariance Matrix
3 asset example
= + + = 1 2 3
2 = var() = 2
2 + 2
2
= cov( ) =
Covariance matrix
⎛ ⎞
2
1 12 13
⎜ ⎟
Σ = ⎝ 12 22 23 ⎠
13 23 32
⎛ ⎞
2 2 2
+ 1 2
12 2
13
⎜ 1 2 2 2 2 2 ⎟
⎜
= ⎝ 12 2 + 2 23 ⎟ ⎠
2
13 2
23 2 2
3 + 32
⎛ ⎞ ⎛ 2 ⎞
2
1 12 13 1 0 0
2 ⎜ ⎟ ⎜ 2 ⎟
= ⎝ 12 22 23 ⎠ + ⎜ ⎝ 0 2 0 ⎟
⎠
13 23 32 0 0 32
Simplification using matrix algebra
R = + +
3×1 3×1 3×1 1×1 3×1
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
1 1 1
R = ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝ 2 ⎠ = ⎝ 2 ⎠ = ⎝ 2 ⎠
3 3 3
Then
Σ = (R) = var() 0 + ()
2 ·
= 0+ D
(3×1)(1×3) (3×3)
where
2 · 0 = covariance due to market
2 2 2 ) = asset specific variances
D = diag(1 2 3
SI Model and Portfolios
2 asset example
1 = 1 + 1 + 1
2 = 2 + 2 + 2
1 = share invested in asset 1
2 = share invested in asset 2
1 + 2 = 1
Portfolio return
= 11 + 22
= 1(1 + 1 + 1)
+ 2(2 + 2 + 2)
= (11 + 22) + (11 + 22)
+ (11 + 22)
= + +
where
= 11 + 22
= 11 + 22
= 11 + 22
SI Model with Large Portfolios
= 1 assets (e.g. = 500)
1
= = equal investment shares
= + +
Portfolio return
X
=
=1
X
= ( + + )
=1 ⎛ ⎞
X
X
X
= + ⎝ ⎠ +
=1 =1 =1
⎛ ⎞
1 X X X
1 1
= + ⎝ ⎠ +
=1 =1 =1
= ̄ + ̄ + ̄
where
1 X
̄ =
=1
1 X
̄ =
=1
1 X
̄ =
=1
Result: For large
1 X
̄ = ≈ [] = 0
=1
2 )
because ∼ iid (0
Implications
In a large well diversified portfolio, the following results hold:
• ≈ ̄ + ̄ : all non-market risk is diversified away
• var() = ̄ 2var() : Magnitude of portfolio variance is propor-
tional to market variance. Magnitude of portfolio variance is determined
by portfolio beta ̄
• 2 ≈ 1 : Approximately 100% of portfolio variance is due to market
variance
