25.06.2013 Views

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

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

400 13. Gestion <strong>de</strong> <strong>portefeuille</strong> sous contraintes <strong>de</strong> capital économique<br />

3 Value-<strong>at</strong>-Risk<br />

3.1 Calcul<strong>at</strong>ion of the VaR<br />

We consi<strong>de</strong>r a portfolio ma<strong>de</strong> of N ass<strong>et</strong>s with all the same exponent c and scale param<strong>et</strong>ers χi, i ∈<br />

{1, 2, · · · , N}. The weight of the i th ass<strong>et</strong> in the portfolio is <strong>de</strong>noted by wi. By <strong>de</strong>finition, the Value<strong>at</strong>-Risk<br />

<strong>at</strong> the loss probability α, <strong>de</strong>noted by VaRα, is given , for a continuous distribution of profit and loss,<br />

by<br />

Pr{W (τ) − W (0) < −VaRα} = α, (47)<br />

which can be rewritten as<br />

<br />

Pr S < − VaRα<br />

<br />

= α. (48)<br />

W (0)<br />

In this expression, we have assumed th<strong>at</strong> all the wealth is invested in risky ass<strong>et</strong>s and th<strong>at</strong> the risk-free<br />

interest r<strong>at</strong>e equals zero, but it is easy to reintroduce it, if necessary. It just leads to discount VaRα by the<br />

discount factor 1/(1 + µ0), where µ0 <strong>de</strong>notes the risk-free interest r<strong>at</strong>e.<br />

Now, using the fact th<strong>at</strong> FS(x) ∼ λ− FZ(x), when x → −∞, and where Z ∼ W(c, ˆχ), we have<br />

<br />

1<br />

Pr S < − VaRα<br />

<br />

√2<br />

c/2<br />

VaRα<br />

1 − Φ<br />

, (49)<br />

W (0)<br />

W (0) ˆχ<br />

λ−<br />

as VaRα goes to infinity, which allows us to obtain a closed expression for the asymptotic Value-<strong>at</strong>-Risk<br />

with a loss probability α:<br />

VaRα W (0) ˆχ<br />

21/c <br />

Φ −1<br />

<br />

1 − α<br />

2/c , (50)<br />

λ−<br />

ξ(α) 2/c W (0) · ˆχ, (51)<br />

where the function Φ(·) <strong>de</strong>notes the cumul<strong>at</strong>ive Normal distribution function and<br />

ξ(α) ≡ 1<br />

2 Φ−1<br />

<br />

1 − α<br />

<br />

. (52)<br />

λ−<br />

In the case where a fraction w0 of the total wealth is invested in the risk-free ass<strong>et</strong> with interest r<strong>at</strong>e µ0, the<br />

previous equ<strong>at</strong>ion simply becomes<br />

VaRα ξ(α) 2/c (1 − w0) · W (0) · ˆχ − w0W (0)µ0. (53)<br />

Due to the convexity of the scale param<strong>et</strong>er ˆχ, the VaR is itself convex and therefore sub-additive. Thus, for<br />

this s<strong>et</strong> of distributions, the VaR becomes coherent when the consi<strong>de</strong>red quantiles are sufficiently small.<br />

The Expected-Shortfall ESα, which gives the average loss beyond the VaR <strong>at</strong> probability level α, is also<br />

very easily computable:<br />

α<br />

ESα = 1<br />

VaRu du (54)<br />

α 0<br />

= ζ(α)(1 − w0) · W (0) · ˆχ − w0W (0)µ0, (55)<br />

where ζ(α) = 1<br />

α<br />

α 0 ξ(u)2/c du . Thus, the Value-<strong>at</strong>-Risk, the Expected-Shortfall and in fact any downsi<strong>de</strong><br />

risk measure involving only the far tail of the distribution of r<strong>et</strong>urns are entirely controlled by the scale<br />

param<strong>et</strong>er ˆχ. We see th<strong>at</strong> our s<strong>et</strong> of multivari<strong>at</strong>e modified Weibull distributions enjoy, in the tail, exactly the<br />

same properties as the Gaussian distributions, for which, all the risk measures are controlled by the standard<br />

<strong>de</strong>vi<strong>at</strong>ion.<br />

12

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!