statistique, théorie et gestion de portefeuille - Docs at ISFA
statistique, théorie et gestion de portefeuille - Docs at ISFA
statistique, théorie et gestion de portefeuille - Docs at ISFA
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
386 12. Gestion <strong>de</strong>s risques grands <strong>et</strong> extrêmes<br />
r<strong>et</strong>urns on the in<strong>de</strong>x r<strong>et</strong>urns, which shows th<strong>at</strong> there is significantly less<br />
linear correl<strong>at</strong>ion b<strong>et</strong>ween P 1 and the in<strong>de</strong>x (correl<strong>at</strong>ion coefficient of 0.52<br />
for both the equally weighted and the minimum variance P 1 ) compared<br />
with P 2 and the in<strong>de</strong>x (correl<strong>at</strong>ion coefficient of 0.73 for the equally weighted<br />
P 2 and of 0.70 for the minimum variance P 2 ). Theor<strong>et</strong>ically, it is possible<br />
to construct two random variables with small correl<strong>at</strong>ion coefficient<br />
and large λ and vice versa. Recall th<strong>at</strong> the correl<strong>at</strong>ion coefficient and the<br />
tail-<strong>de</strong>pen<strong>de</strong>nce coefficient are two opposite end-members of <strong>de</strong>pen<strong>de</strong>nce<br />
measures. The correl<strong>at</strong>ion coefficient quantifies the <strong>de</strong>pen<strong>de</strong>nce b<strong>et</strong>ween<br />
rel<strong>at</strong>ively small moves while the tail-<strong>de</strong>pen<strong>de</strong>nce coefficient measures the<br />
<strong>de</strong>pen<strong>de</strong>nce during extreme events. The finding th<strong>at</strong> P 1 comes with both<br />
the smallest correl<strong>at</strong>ion and the smallest tail-<strong>de</strong>pen<strong>de</strong>nce coefficients suggests<br />
th<strong>at</strong> they are not in<strong>de</strong>pen<strong>de</strong>nt properties of ass<strong>et</strong>s. This intuition is<br />
in fact explained and encompassed by the factor mo<strong>de</strong>l since the larger<br />
β is, the larger the correl<strong>at</strong>ion coefficient and the larger the tail <strong>de</strong>pen<strong>de</strong>nce.<br />
Diversifying away extreme shocks may provi<strong>de</strong> a useful diversific<strong>at</strong>ion<br />
tool for less extreme <strong>de</strong>pen<strong>de</strong>nces, thus improving the potential<br />
usefulness of a str<strong>at</strong>egy of portfolio management based on tail <strong>de</strong>pen<strong>de</strong>nce<br />
proposed here.<br />
As a final remark, the almost i<strong>de</strong>ntical values of the coefficients of tail<br />
<strong>de</strong>pen<strong>de</strong>nce for neg<strong>at</strong>ive and positive tails show th<strong>at</strong> ass<strong>et</strong>s th<strong>at</strong> are the<br />
most likely to suffer from the large losses of the mark<strong>et</strong> factor are also<br />
those th<strong>at</strong> are the most likely to take advantage of its large gains. This<br />
has the following consequence: minimising the large concomitant losses<br />
b<strong>et</strong>ween the stocks and the mark<strong>et</strong> means renouncing the potential<br />
concomitant large gains. This point is well exemplified by our two portfolios<br />
(see figure 3): P 2 obviously un<strong>de</strong>rwent severe neg<strong>at</strong>ive co-movements<br />
but it also enjoyed large gains with the large positive movements<br />
of the in<strong>de</strong>x. In contrast, P 1 is almost compl<strong>et</strong>ely <strong>de</strong>coupled from the<br />
large neg<strong>at</strong>ive movements of the mark<strong>et</strong> but is also insensitive to the large<br />
positive movements of the in<strong>de</strong>x. Thus, a good dynamic str<strong>at</strong>egy seems<br />
to be: invest in P 1 during bearish or trend-less mark<strong>et</strong> phases and prefer<br />
P 2 in a bullish mark<strong>et</strong>. ■<br />
Yannick Malevergne is a PhD stu<strong>de</strong>nt <strong>at</strong> the University of Nice-Sophia<br />
Antipolis and <strong>at</strong> the <strong>ISFA</strong> Actuarial School – University of Lyon. Didier<br />
Bingham N, C Goldie and J Teugel, 1987<br />
Regular vari<strong>at</strong>ion<br />
Cambridge University Press, Cambridge<br />
Boyer B, M Gibson and M Laur<strong>et</strong>an, 1997<br />
Pitfalls in tests for changes in correl<strong>at</strong>ions<br />
Intern<strong>at</strong>ional Finance Discussion Paper 597, Board of the Governors of the Fe<strong>de</strong>ral<br />
Reserve System<br />
Coles S, J Heffernan and J Tawn, 1999<br />
Depen<strong>de</strong>nce measures for extreme value analysis<br />
Extremes 2, pages 339–365<br />
Embrechts P, A McNeil and D Straumann, 1999<br />
Correl<strong>at</strong>ion: pitfalls and altern<strong>at</strong>ives<br />
Risk May, pages 69–71<br />
Embrechts P, A McNeil and D Straumann, 2002<br />
Correl<strong>at</strong>ion and <strong>de</strong>pen<strong>de</strong>nce in risk management: properties and pitfalls<br />
In Risk Management: Value <strong>at</strong> Risk and Beyond, edited by M Dempster, pages<br />
176–223, Cambridge University Press, Cambridge<br />
Hull J and A White, 1987<br />
The option pricing on ass<strong>et</strong>s with stochastic vol<strong>at</strong>ilities<br />
Journal of Finance 42, pages 281–300<br />
Hult H and F Lindskog, 2002<br />
Multivari<strong>at</strong>e extremes, aggreg<strong>at</strong>ion and <strong>de</strong>pen<strong>de</strong>nce in elliptical distributions<br />
Forthcoming in Advances in Applied Probability 34(3)<br />
Ledford A and J Tawn, 1998<br />
Concomitant tail behaviour for extremes<br />
Advances in Applied Probability 30, pages 197–215<br />
REFERENCES<br />
Appendix: empirical estim<strong>at</strong>ion of the<br />
coefficient of tail <strong>de</strong>pen<strong>de</strong>nce<br />
We show how to estim<strong>at</strong>e the coefficient of tail <strong>de</strong>pen<strong>de</strong>nce<br />
b<strong>et</strong>ween an ass<strong>et</strong> X and the mark<strong>et</strong> factor Y rel<strong>at</strong>ed by the rel<strong>at</strong>ion<br />
(6) where ε is an idiosyncr<strong>at</strong>ic noise uncorrel<strong>at</strong>ed with X.<br />
Given a sample of N realis<strong>at</strong>ions {X 1 , X 2 , ... , X N } and<br />
{Y 1 , Y 2 , ... , Y N } of X and Y, we first estim<strong>at</strong>e the coefficient β using<br />
the ordinary least square estim<strong>at</strong>or. L<strong>et</strong> β ^ <strong>de</strong>note its estim<strong>at</strong>e. Then,<br />
using Hill’s estim<strong>at</strong>or, we obtain the tail in<strong>de</strong>x ν^ of the factor Y:<br />
⎡ k 1<br />
ˆ = ⎢ ∑ logY −logY<br />
⎣⎢<br />
k j=<br />
1<br />
νk j, N k, N<br />
where Y 1, N ≥ Y 2, N ≥ ... ≥ Y N, N are the or<strong>de</strong>r st<strong>at</strong>istics of the N realis<strong>at</strong>ions<br />
of Y. The constant l is non-param<strong>et</strong>rically estim<strong>at</strong>ed with<br />
the formula:<br />
FXu X<br />
l = lim ≃<br />
u→1<br />
−1<br />
F u Y<br />
for k = o(N), which means th<strong>at</strong> k must remain very small with<br />
respect to N but large enough to ensure an accur<strong>at</strong>e d<strong>et</strong>ermin<strong>at</strong>ion<br />
of l. Figure 2 presents l ^ as a function of k/N.<br />
Finally, using equ<strong>at</strong>ion (7), the estim<strong>at</strong>ed λ ^ is:<br />
ˆ<br />
max , ˆ<br />
+ 1<br />
λ =<br />
νˆ<br />
{ 1<br />
βˆ<br />
} l<br />
Sorn<strong>et</strong>te is a CNRS research director <strong>at</strong> the University of Nice-Sophia<br />
Antipolis and professor of geophysics <strong>at</strong> the University of California <strong>at</strong><br />
Los Angeles. They acknowledge helpful discussions with Jean-Paul<br />
Laurent. This work was partially supported by the James S McDonnell<br />
Found<strong>at</strong>ion twenty-first century scientist award/studying complex<br />
system. e-mail: Yannick.Malevergne@unice.fr and sorn<strong>et</strong>te@unice.fr<br />
Malevergne Y and D Sorn<strong>et</strong>te, 2002a<br />
Investig<strong>at</strong>ing extreme <strong>de</strong>pen<strong>de</strong>nces: concepts and tools<br />
Working paper, available <strong>at</strong> http://papers.ssrn.com/sol3/papers.cfm?abstract_id =<br />
303465<br />
Malevergne Y and D Sorn<strong>et</strong>te, 2002b<br />
Tail <strong>de</strong>pen<strong>de</strong>nce of factor mo<strong>de</strong>ls<br />
Working paper, available <strong>at</strong> http://papers.ssrn.com/sol3/papers.cfm?abstract_id =<br />
301266<br />
Man<strong>de</strong>lbrot B, 1997<br />
Fractals and scaling in finance: discontinuity, concentr<strong>at</strong>ion<br />
Springer-Verlag, New York<br />
Nelsen R, 1998<br />
An introduction to copulas<br />
Lectures Notes in St<strong>at</strong>istics 139, Springer Verlag, New York<br />
Ross S, 1976<br />
The arbitrage theory of capital ass<strong>et</strong> pricing<br />
Journal of Economic Theory 17, pages 254–286<br />
Rubinstein M, 1973<br />
The fundamental theorem of param<strong>et</strong>er-preference security valu<strong>at</strong>ion<br />
Journal of Financial and Quantit<strong>at</strong>ive Analysis 8, pages 61–69<br />
Sharpe W, 1964<br />
Capital ass<strong>et</strong> pricing: a theory of mark<strong>et</strong> equilibrium un<strong>de</strong>r conditions of risk<br />
Journal of Finance 19, pages 425–442<br />
Taylor S, 1994<br />
Mo<strong>de</strong>ling stochastic vol<strong>at</strong>ility<br />
M<strong>at</strong>hem<strong>at</strong>ical Finance 4, pages 183–204<br />
−1<br />
Y<br />
( )<br />
( )<br />
kN ,<br />
kN ,<br />
⎤<br />
⎥<br />
⎦⎥<br />
WWW.RISK.NET ● NOVEMBER 2002 RISK 133