Introduction

Black-Litterman
Meuccis Model
Mutilvariate Skew-t
Conclusion

Models for Quantitative Portfolio Management

Part 2 : Black Litterman and Meuccis Model

Pierre GARREAU

Florida State University

September 17th 2009

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman
Asset Allocation
Meuccis Model
Markowitz model
Mutilvariate Skew-t
Conclusion

Credit Portfolio

Investment Universe : Corporate Bonds, CDS, Equities and Options.

Structural Models for CDS depend on Qualitative Data : Balance
Sheet.
Managers decisions based on Morning signals, News and Intraday
Info.

Need to Specify views in a Markowitz type Framework.

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman
Asset Allocation
Meuccis Model
Markowitz model
Mutilvariate Skew-t
Conclusion

Framework

Choice of a fully invested portfolio h = [h1 , h2 , . . . , hN ]T in a

universe of N securities.
Location and Dispersion measure are Mean f and Variance .
The investors are Risk Averse with utility function
U(h) = hT f hT h.

h = argmaxhH U(h)
where X
hT f
H= hi = 1, h > c, r 2 < q
B

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman
Asset Allocation
Meuccis Model
Markowitz model
Mutilvariate Skew-t
Conclusion

Figure: Efficient frontier for a universe of 5 Stocks : Exxon, City, Boeing, Ford,
SnP - November 2008
P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman
Asset Allocation
Meuccis Model
Markowitz model
Mutilvariate Skew-t
Conclusion

Figure: Efficient frontier for a universe of 5 Stocks : Exxon, City, Boeing, Ford,
SnP - November 2008 - Shortsell 20 % - Perturbation
P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman
Asset Allocation
Meuccis Model
Markowitz model
Mutilvariate Skew-t
Conclusion

Limitations
The risk premium for positive deviations is the same as negative
deviations : Approximation of Investors satifaction
Optimizer overweigths high expected returns and negatively
correlated securities.
Optimizer subject to estimation errors. Relevance of the risk
aversion parameter ?
Covariances estimation.

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

The Market The market is represented as a random variable X.

fX N (, )

The market distribution is affected by estimation risk.

The Managers selection The manager asses the outcome of the market V
and expresses a conditional distribution V|x or V|g (x) . This allows to get

fV|g (x) (v |x)fX (x)

fX |v (x | v ) = R (1)
fV|g (x) (v |x) fX (x) dx

The views are specified through a Pick Matrix

g (x) = Px

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

Distribution of Views Evaluation of the managers track record

V|Px N (Px, )

where
1
= ( 1)PPT
c

The Managers opinion Asked target views on his pick : v

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

The Black-Litterman Model (1990)

The Black-Litterman distribution is normal

X |v N (BL , BL )

where the expected value and covariance matrix are expressed as

BL = + P(PPT + )1 (v P) (2)
BL = PT (PPT + )1 P (3)

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

Portfolio and Confidence

Full confidence in the managers view : c = 1

BL = + P(PPT )1 (v P)
0
BL = PT (PPT )1 P

Conditional Model PX = v
Null confidence in the managers view : c = 0

BL =

BL =

Market Model : Markowitz Portfolio

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

Investment Universe is composed of 5 Stocks : Aegon, Air Liquide,

Allianz, Alstom, Arcelor Mittal.
Covariance Matrix estimated with the method of moments.

25, 57% 1, 94% 5, 68% 2, 16% 4, 33%
1, 94% 3, 59% 0, 28% 0, 46% 0, 50%

= 5, 68% 0, 28% 8, 42% 1, 53% 1, 75%
2, 16% 0, 46% 1, 53% 13, 37% 5, 32%
4, 33% 0, 50% 1, 75% 5, 32% 20, 15%

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

Returns are given by the characteristic portfolio of weights

hw = w2 1 w

9, 63% 7, 31%
1, 39% 20, 97%

= 7, 83% w = 34, 96%

7, 15% 10, 33%
13, 81% 26, 43%

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

The targets of the views are different from the impled returns and chosen
to show extreme cases.

3, 00%
8, 50%

v = 3, 50%

11, 00%
12, 50%

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

Markowitzs Portolio

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

No confidence in the manager

(a) Black-Litterman c = 0 (b) Black-Litterman c = 1

Equilibrium portfolio.
Pocket for Air Liquide is increased since 8.5% return and volatility
only 3.59% with low or negative correlation with other stocks.

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
The Set Up
Black-Litterman
A Posteriori Distribution
Meuccis Model
Discussion
Mutilvariate Skew-t
Discussion
Conclusion

Limitations

Non-Normal Markets and the problem of Asset Correlation.

Express Views on more than just returns.
Stress Testing.
Correlation of Views ?

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Modelisation and Views
Meuccis Model Copula Opinion Pooling
Mutilvariate Skew-t Joint Posterior Distribution
Conclusion

The Market The market is represented as an d-dimensional random

variable X and will be specified thanks to its cdf

X FX
The apriori Market M is represented by J simulations of X : J d
Matrix

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Modelisation and Views
Meuccis Model Copula Opinion Pooling
Mutilvariate Skew-t Joint Posterior Distribution
Conclusion

The Views
The manager expresses an opinion on K d securities so that

V = PX
is the completed and invertible matrix.
where P
Each view is specified individually and reprensented by its cdf

ck (v ) = P(Vk v )
F k = 1, . . . , K

Managers view is different from the Market Implied Distribution : COP

approach

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Modelisation and Views
Meuccis Model Copula Opinion Pooling
Mutilvariate Skew-t Joint Posterior Distribution
Conclusion

Posterior Marginal CDF

The posterior marginal CDF of each security is given by a linear
combination :
fk (v ) = ck F
F ck (v ) + (1 ck )Fk (v ) k = 1, . . . , K

where ck represents the confidence level in view k.

In order to get the joint distribution, we need to estimate the market
implied copula.

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Modelisation and Views
Meuccis Model Copula Opinion Pooling
Mutilvariate Skew-t Joint Posterior Distribution
Conclusion

Copula A d-dimensional Copula C is a d-dimensional distribution

function on [0, 1]d with standard uniform marginal distributions.
Sklars Theorem Every cumulative distribution function F with margins
F1 , . . . , Fd can be written as

F (v1 , . . . , vd ) = C (F1 (v1 ), . . . , Fd (vd ))

with C unique if the margins are continuous.

Kendalls Tau For bivariate random vectors (X , Y ) Kendalls Tau is
expressed as
(X , Y ) = E sign(X Xe )(Y Y e)

e, Y
where (X e ) is a second independent pair with the same distribution.

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Modelisation and Views
Meuccis Model Copula Opinion Pooling
Mutilvariate Skew-t Joint Posterior Distribution
Conclusion

Joint Posterior Distribution The market implied copula is then

d
C = (F11 (V1 ), . . . , Fd1 (Vd ))0

The joint posterior distribution of the views is given by :

1
V1 Fe1 (F1 (V1 ))
.. d ..
. = .


1
Vd fd
F (Fd (Vd ))

Posterior Distribution of the Market We need to apply backward the

definition of the views to get X e
fM :
1 V
X P

where the remaining d K entries of V remain unchanged.

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Representation
Meuccis Model Calibration
Mutilvariate Skew-t Skewed-t Copula C
Conclusion

Market Representation

The market is represented as an d-dimensional random variable

X Td (, , , )

X + W + W Z
with
Z (0, ) Multivariate Gaussian random variable with 0 mean and
covariance matrix .
W Inv-Gamma( 2 , 2 ) Inverse Gamma random variable.
Z W.

Closed formula for f,,, and C,,,

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Representation
Meuccis Model Calibration
Mutilvariate Skew-t Skewed-t Copula C
Conclusion

Views Representation

The manager has a view on each of the stocks, and represents it

with a target window so that

0 v ak
ck (v ) =
F v ak
v [ak , bk ]
bk ak
1 v bk

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Representation
Meuccis Model Calibration
Mutilvariate Skew-t Skewed-t Copula C
Conclusion

Maximum Likelihood
Taking a set of data that is independent and identically distributed, the
log-likelihood function L of the vector parameter is defined as
J
X
L() = log f (xj )
j=1

Estimation of the location and skewness parameters for the marginals

(i , i ) = argmax, LXi (, ; ri1 , . . . , riJ ) i = 1, . . . , d

Estimation of the degree of freedom for the Copula

= argmax LC (, , ; ui1 , . . . , uiJ )

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Representation
Meuccis Model Calibration
Mutilvariate Skew-t Skewed-t Copula C
Conclusion

From the Market distribution XJd , estimate each empirical cdf by

J
1 X
Fbi (x) = 1{Xj,i x} i = 1, . . . , d
J +1
j=1

Form the pseudo-sample from the copula

uj,1 , . . . , uj,d ) = (Fb1 (Xj,1 ), . . . , F

( cd (Xj,d ))

Can estimate the dispersion matrice thanks to Kendalls tau

j,i = sin (Xj , Xi )
2
1 X
2
(Xj , Xi ) = sign(Xl,j Xk,j )(Xl,i Xk,i )
J
1klJ

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman Market Representation
Meuccis Model Calibration
Mutilvariate Skew-t Skewed-t Copula C
Conclusion

Generate a Multivariate Skewed-t distribution Td (, 0, P, ) from the

Normal Mixture.
Return
U = (T,1 (X1 ), . . . , T,d (Xd ))0

(c) (0.4,-0.4) (d) (0,0) (e) (-0.4,0.4)

Figure: J = 10, 000 = 0.8 = 5.5

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman
Maximum Likelihood
Meuccis Model
Extension
Mutilvariate Skew-t
Conclusion

Calibration of AEGONs daily returns

(a) Likelihood function, (b) 1 = 1, 1 = 1.72, 1 = 0.64, = 6

on x-axis for several
value of 1

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M
 Introduction
Black-Litterman
Maximum Likelihood
Meuccis Model
Extension
Mutilvariate Skew-t
Conclusion

Extensions

Need a relevant Risk Measure to perform an optimization on the a

posteriori Market : E[X |X < q].
More robust Algorithm for Calibration : EM Algorithm.
Possible Extension to Levy Copulas.

P. Garreau Models for Quantitative Portfolio Management Part 2 : Black Litterman and M