A Layman's Approach to Black-Litterman Model

Eola Investments, LLC

info@eolainvestments.com
First draft in March 2008
Revised in October 2009
All rights reserved

Quantitative Discovery
Page 1 of 7 Eola Investments LLC All Rights Reserved
EXECUTIVE SUMMARY
The expected returns in the Black-Litterman Model can be regarded as the weighted average of
the market view and the investor view. The weightings are the inverses of the covariance matrices,
which models the certainty of the views. Furthermore, the assumption of the statistical distributions of
the asset returns is not needed.

Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 2 of 7
THE MODEL
There are n basis assets in the market, each with expected return r
i
, i = 1, 2, , n. We write all
the expected returns in a vector
r=

r
1
r
2

r
n

[1]
The expectation of the returns needs to be elaborated. The market has expectations of the returns of the
basis assets and their covariance. The expectations are implied by the equilibrium weight of each basis
asset. The set of market expected returns and the market expected covariance among returns are
denoted by H and 2 , respectively
H=r
M
=

r
1
M
r
2
M

r
n
M

[2]
2=
|
var(r
1
M
, r
1
M
) cov (r
1
M
, r
2
M
) cov(r
1
M
, r
n
M
)
cov( r
2
M
, r
1
M
) var ( r
2
M
, r
2
M
) cov(r
2
M
, r
n
M
)

cov( r
n
M
, r
1
M
) cov (r
n
M
, r
2
M
) var (r
n
M
, r
n
M
)

[3]
In addition to the expectation of the general market, an investor could have own expectation of the
returns of the basis assets and their correlations. The market is a big voting machine after all. An
investor's expectation of asset returns and their correlations can be written as
r
I
=

r
1
I
r
2
I

r
n
I

[4]
Eola Investments LLC All Rights Reserved
Quantitative Discovery
Page 3 of 7
2
I
=
|
var( r
1
I
, r
1
I
) cov(r
1
I
, r
2
I
) cov(r
1
I
, r
n
I
)
cov (r
2
I
, r
1
I
) var (r
2
I
, r
2
I
) cov(r
2
I
, r
n
I
)

cov (r
n
I
, r
1
I
) cov(r
n
I
, r
2
I
) var (r
n
I
, r
n
I
)

[5]
The investor could form a new expectation of the returns of basis assets by taking a weighted
average of own view and the view of the market. A reasonable choice of the weights should favor the
view that has more certainty than the other. We know that the covariance matrix is related to the
uncertainty, considering the larger the variance the more uncertain the expected value is. Then the
inverse of the covariance must positively related to the certainty. Therefore, a rational weighted average
of the views could be
r
*
=(2
1
+(2
I
)
1
)
1
( 2
1
H+(2
I
)
1
r
I
)
[6]
More generally, one could choose a weighting more favorable to one view than the other. This can be
done by introducing a scaling factor t
r
*
=((t2)
1
+(2
I
)
1
)
1
((t2)
1
H+(2
I
)
1
r
I
)
[7]
Although an investor has own views, the views may not necessarily be the explicit expectation
of returns of basis asset r
I
. The investor's implicit view about returns of basis assets can be hidden in
views on m different portfolios with a weight matrix P. The investor also has uncertainty, D ,
assigned to each of the view of portfolio returns. Mathematically,
q=P r
I
[8]
q=

q
1
q
2

q
m

,
P=
|
w
11
w
12
w
1n
w
21
w
22
w
2n

w
m1
w
m2
w
mn

, D=
|
c
( q1)
2

c
( q2)
2


c
( qm)
2
m x m
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 4 of 7
Case 1: m < n.
In plain language it means the number of portfolios an investor has views on is less than the
number of individual asset. Therefore there is no way one can infer from the investor's views on
portfolio returns to the returns of basis assets. That is, one cannot solve a unique set of
r
I
from
Equation [8], simply because the number of unknowns is greater than the number of equations.
However, one could always use the market expectations to make up the missing views so that m = n
and equation [8] is solvable, as long as the investor's views are self consistent. This leads to Case 2.
Case 2: m=n
As long as the investor's views of the portfolio returns are self consistent and the number of
equations is equal to the number of unknowns, one can infer the expectation of the investor on the
returns of the basis assets from the views of portfolio returns, i.e., Equation [8] is solvable and there
exists the inverse of P matrix,
r
I
=P
1
q [9]
Now we need to know the covariance matrix of
r
I
, i.e, 2
I
, in order to form a weighted average
with the market view. We know q has a covariance matrix D and P
1
is a constant matrix, then
by the liner relationship of Equation [9] between
r
I
and q , we have
2
I
=P
1
D( P
1
)
T
[10]
Inserting Equation [10] into Equation [7], we have
r
*
=((t 2)
1
+(2
I
)
1
)
1
((t2)
1
H+(2
I
)
1
r
I
)
r
*
=((t 2)
1
+( P
1
D( P
1
)
T
)
1
)
1
((t 2)
1
H+( P
1
D( P
1
)
T
)
1
r
I
)
r
*
=((t 2)
1
+P
T
D
1
P)
1
((t2)
1
H+P
T
D
1
Pr
I
)
[11]
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 5 of 7
Inserting Equation [8] into Equation [11] we obtain the Black-Litterman result
r
*
=((t2)
1
+P
T
D
1
P)
1
((t2)
1
H+P
T
D
1
q)
[12]
DISCUSSIONS
From the above derivation we can see the thought process presented here requires an investor
have implied views on each basis asset. By so doing one of the advantages the Black-Litterman model
offers is lost. It has been shown that a complete view of the returns of the basis asset is not needed.
Therefore, the derivation here is only a special case of the Black-Litterman model. Nevertheless, it
provides insights to help understand the essence of the Black-Litterman model.
Wait! Is it truly a special case of the Black-Litterman model? In the above derivation we have
not assume any distribution of r
I
and q ! While other approaches such as Bayesian approach and
Sampling Theory approach universally assume the asset returns follow normal distributions. Does that
forming a complete views on the returns of basis assets free us from the form of the statistical
distributions that asset returns follow?
Continuing the thought process presented here, we can have more results.
Case III. m > n
In this case, an investor have more views than the number of basis assets. Then the implicit
views of the returns of the basis assets must subject to some sort of random error. One can readily infer
a set of returns of basis assets by minimizing the random error using the Least Square method. In this
case, the Equation [9] is changed to
r
I
=( P
T
P)
1
P
T
q [13]
Let
X =( P
T
P)
1
P
T
[14]
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 6 of 7
The covariance of r
I
in this case is
2
I
=X DX
T
[15]
And the final result is
r
*
=((t 2)
1
+(2
I
)
1
)
1
((t2)
1
H+(2
I
)
1
r
I
)
r
*
=((t 2)
1
+( X DX
T
)
1
)
1
((t2)
1
H+( X DX
T
)
1
r
I
)
r
*
=((t 2)
1
+( X DX
T
)
1
)
1
((t2)
1
H+( X DX
T
)
1
X q)
[16]
Given the results of the derivation, now the fun part of estimating the value of each parameter
begins. Unfortunately, the Black-Litterman model is a conjoint hypothesis in that the model and the
view of the investor cannot be separately tested. One really needs to either believe that the Black-
Litterman model is a superior model, or find a better alternative oneself.
POSTSCRIPT
The Black-Litterman model assumes normal distributions of returns. Therefore in principle it
should be better to use log excess returns than simple returns in practice.
Quantitative Discovery
Eola Investments LLC All Rights Reserved Page 7 of 7