1: \documentclass[]{article}
2: \setlength{\textheight}{8.5in}
3: \setlength{\textwidth}{6.5in}
4: \setlength{\oddsidemargin}{-0.0in}
5: \setlength{\topmargin}{-.5in}
6: %\usepackage{latexsym}
7: \usepackage{amsmath}
8: %\usepackage{english}
9: \usepackage{amssymb}
10: \newcommand{\be}{\begin{equation}}
11: \newcommand{\ee}{\end{equation}}
12: \newcommand{\calO}{{\cal O}}
13: \newcommand{\calL}{{\cal L}}
14: \newcommand{\calG}{{\cal G}}
15: \newcommand{\calH}{H }
16: \newcommand{\calM}{{\cal M}}
17: \newcommand{\calS}{{\cal S}}
18: \newcommand{\calD}{{\cal D}}
19: \newtheorem{theorem}{Theorem}[section]
20: \newtheorem{corollaire}[theorem]{Corollaire}
21: \newtheorem{lemma}[theorem]{Lemma}
22: \newtheorem{proposition}[theorem]{Proposition}
23: \newtheorem{question}{Question}
24: \newtheorem{defi}[theorem]{Definition}
25: \newtheorem{definition and theorem}[theorem]{Definition and Theorem}
26: \newtheorem{remark}[theorem]{Remark}
27: \newtheorem{*remark}[theorem]{$^* $Remark}
28: \newtheorem{exercise}[theorem]{Exercise}
29: \newtheorem{*exercise}[theorem]{$^* $Exercise}
30: \newtheorem{**exercise}[theorem]{$^{** } $Exercise}
31: \newtheorem{example}[theorem]{Example}
32: \newtheorem{examples}[theorem]{Examples}
33: \newtheorem{properties}[theorem]{Properties}
34: \newtheorem{corollary}[theorem]{Corollary}
35: \newtheorem{spv}[theorem]{Sneak preview}
36:
37: \begin{document}
38: \title{Value-at-Risk and Expected Shortfall for Quadratic Portfolio of Securities with Mixture of Elliptic Distributed Risk Factors }
39:
40: \author{Jules SADEFO KAMDEM
41: \thanks{ Paper based on the M. Jules SADEFO KAMDEM Universit\'e de Reims Phd
42: Thesis.\quad Author address: BP 1039 Moulin de la Housse 51687
43: Reims Cedex FRANCE. sadefo@univ-reims.fr }
44: \thanks{The author is currently temporary lecturer to the Mathematics Department of the ( Universit\'e d'evry val d'essonne FRANCE).}
45: \\
46: Laboratoire de Math\'ematiques\\
47: CNRS UMR 6056\\
48: Universit\'e De Reims \\
49: }
50:
51: \maketitle
52: \begin{abstract}
53: Generally, in the financial literature, the notion of quadratic VaR is implicitly confused with
54: the Delta-Gamma VaR, because more authors dealt with
55: portfolios that contained derivatives instruments. In this paper,
56: we postpone to estimate both the expected shortfall and
57: Value-at-Risk of a quadratic portfolio of securities (i.e
58: equities) without the Delta and Gamma Greeks, when the
59: joint log-returns changes with multivariate elliptic distribution. To illustrate
60: our method, we give special attention to mixture of normal
61: distributions, and mixture of
62: Student t-distributions.
63: \end{abstract}
64: \noindent {\it Key Words:} Classical analysis, Computational
65: Finance, Elliptic distributions, Risk Management .
66:
67:
68: \section{\bf Introduction}
69: \medskip
70:
71:
72:
73: Value-at-Risk is a market risk management tool that permit to measure the maximum
74: loss of the portfolio with certain confidence probability $1-\alpha$, over a certain time horizon such as one day. Formally ,
75: if the price of portfolio's $ P(t, S(t))$ at time t is a random variable where $S(t)$ represents a vector
76: of risk factors at time t, then VaR be implicitly given by the formula
77: $$ Prob\{ -P(t,S(t))+P(0,S(0)) > VaR_\alpha \}=\alpha . $$
78:
79:
80: Generally, to estimate the $VaR_{\alpha}$ for portfolios depending non-linearly on the return
81: , or portfolios of non-normally distributed assets, one turns to Monte Carlo
82: methods. Monte Carlo methodology has the obvious advantage of being almost universally
83: applicable, but has the disadvantage of being much slower than
84: comparable parametric methods, when the latter are available.
85:
86:
87: In this paper we are concerned with the numerical
88: estimation of the losses that the portfolio of equities faces due
89: to the market, as a function of the future values of S. Following
90: the quadratic Delta-Gamma Portfolio, we introduce the notion of
91: quadratic portfolio of equities due to the analytic approximation
92: of Taylor in $2^{nd}$ order of log-returns for very small
93: variations of time. Quadratic approximations have also been the
94: subject of a number of papers dedicated to numerical computations
95: for VaR ( but these have been done for portfolio that contains
96: derivatives instruments). We refer the reader to Cardenas and
97: al.(1997) \cite{CFKM} for a numerical method to compute quadratic
98: VaR using fast Fourier transform . Note that in \cite{BCQS},
99: Brummelhuis, Cordoba, Quintanilla and Seco have dealt with the
100: similar problem, but their work have been done analytically for
101: Delta-Gamma Portfolio VaR when the joint underlying risk factors
102: follow a normal distribution which is a particular case of elliptic distribution . All our
103: calculus will be done according to the assumption that the joint
104: securities (i.e equities) log-returns follow an elliptic
105: distribution. To illustrate our method, we will take some examples
106: of elliptic distributions as mixture of {\em{ multivariate
107: $t$-student}} and mixture of normal distributions. Note also that,
108: Following RiskMetrics, Sadefo-Kamdem \cite{S2}(2003) have
109: generalized the notion of $\Delta$-normal VaR by introducing the
110: notion of $\Delta$-Elliptic VaR , with special attention to
111: $\Delta$-Student VaR , but this concerned the linear portfolio. In
112: this paper, we will do the same for nonlinear
113: quadratic portfolios without derivatives instruments.\\
114:
115:
116: The rest of the paper is organized as follows: In section
117: 2, we introduced the notion of quadratic portfolio of
118: securities(i.e equities) due to the $2^{nd}$ order Taylor
119: approximation of log returns . Our calculus is made with the more
120: generalized assumptions that the joint underlying log-returns
121: follow an elliptic distribution. That is why in section 3,
122: following \cite{EMS}, we recall the definition of elliptic
123: distribution and we show that under the hypothesis of elliptic
124: distribution the VaR estimation of such portfolios is reduced to a
125: multiple integral equation. Next following the paper of Alan Genz
126: \cite{G} , we recall the notions of symmetric interpolar rules for
127: multiple integrals over hypersphere and we use this method to
128: reduce our problem to one dimensional integral equation. In the
129: same section, we illustrate our method by giving an explicit
130: equation with solution VaR ( Value-at-Risk ) when the joint
131: log-returns follow some particular mixture of elliptic
132: distributions named mixture of multivariate Student
133: $t$-distributions or the mixture of normal distributions, in these
134: cases the VaR estimation is reduced to finding the zero's of a
135: certain specials functions. In section 5 we treats the expected
136: shortfall for general elliptic quadratic portfolios of securities
137: without derivatives instruments and we illustrate with the special
138: case of normal distribution. Finally, in section 6 we give a
139: conclusion.
140:
141:
142:
143:
144:
145: \section{Quadratic Portfolio of Securities(i.e Equities)}
146:
147:
148: A portfolio of n securities is a vector $\theta \in
149: \mathbb{R}^{n}$; the component $\theta_{i}$ represents the number
150: of holdings of the $i^{th}$ instruments, which in practise does
151: not need to be an integer.
152: So at time t the price of the portfolio of n securities is given by:
153: \be P(t)=\sum_{i=1}^{n} \theta_{i}S_{i}(t) \label{e1} \ee where
154: $ S(t)=( S_{1}(t) ,\ldots, S_{n}(t)) $ such that
155: $$
156: P(t) - P(0) =\sum_{i=1}^{n} \theta_i ( S_{i}(t) -S_{i}(0) ) =
157: \sum_{i=1}^{n}S_{i}(0) . \theta_{i} . (\frac{S_{i}(t)}{S_{i}(0)}
158: -1 ) \label{e3} $$ For small fluctuations of time and market, we
159: assume that log-return is given by :
160:
161: \be log(S_{i}(t)/ S_{i}(0)) = \eta_{i}(t) \label{eq2} \ee
162: therefore
163: $$S_{i}(t) - S_{i}(0) = S_{i}(0) ( \frac{S_{i}(t)}{S_{i}(0)} - 1)
164: = S_{i}(0) (exp(\eta_{i}(t)) - 1) $$
165: Then we have that
166: $$S(t)=(S_{1}(0) exp(\eta_{1}), \ldots,S_{n}(0) exp(\eta_{n})). $$
167: By using Taylor's expansion of the exponential's due to small
168: fluctuations of returns with a time, we have that :
169: \be
170: exp(\eta_{i}(t)) - 1 \approx \eta_{i}(t) +
171: \frac{{\eta_{i}(t)}^{2}}{2}. \label{eq3}
172: \ee
173: If we assume that $\eta =(\eta_{1},\ldots,\eta_{n})$ is an elliptic distribution note by $ N_{n}(\mu
174: ,\Sigma ,\phi )$ then
175:
176: \be P(t) -P(0) =\sum_{i=1}^{n}S_{i}(0)\cdot \theta_{i}\cdot
177: (exp(\eta_{i}(t)) - 1 ) \approx \sum_{i=1}^{n} S_{i}(0)
178: .\theta_{i} ( \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} ).
179: \label{e6} \ee
180:
181: By following the usual convention of recording portfolio losses
182: by negative numbers, but stating the Value-at-Risk as a positive
183: quantity of money, The $VaR_{\alpha}$ at confidence level of
184: $1-\alpha$ is given by solution of the following equation:
185: $$Prob\{ |P(t) - P(0) | \geq VaR_{\alpha} \} =\alpha $$
186: In the probability space of losses $|P(t) - P(0) | =-P(t) + P(0)
187: $, therefore recalling (\ref{e6}), we have
188:
189: \be
190: Prob\{ \sum_{i=1}^{n}S_{i}(0) . \theta_{i} ( \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} ) \leq - VaR_{\alpha } \} = \alpha
191: \label{e7}
192: ,\ee
193: then an elementary mathematical tool give that :
194: \be
195: \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} =\frac{1}{2}
196: {(\eta_{i}(t) +1)}^{2} - 1 \label{eq7} \ee
197: therefore
198: \be
199: Prob\{ \sum_{i=1}^{n} S_{i}(0) . \frac{\theta_{i}}{2}
200: {(\eta_{i}(t) +1)}^{2} ) \leq - VaR_{\alpha } +
201: \sum_{i=1}^{n} \frac{\theta_{i}}{2}\cdot S_{i}(0) \} = \alpha .
202: \label{e8}
203: \ee
204: By posing $X =(\eta_{1}+1 , ... , \eta_{n}+1 ) $, it is straightforward that , X
205: is an elliptic distribution due to the fact that it is a linear
206: combination of elliptic distribution $\eta $. We note $X\sim N(\mu
207: +1,\Sigma ,\phi^{'})$ with a continuous density function
208: $h_{1}(x)$. Remark that $$\sum_{i=1}^{n} \alpha_{i} {(\eta_{i}(t)
209: +1)}^{2} =(x,\Lambda.x)$$ with $\Lambda={( \alpha
210: _{ii})}_{i=1..n} $ is a diagonal matrix with diagonal values $
211: \alpha_{ii}=\frac{ S_{i}(0) . \theta_{i}}{2} \geq 0$ and $
212: \mu^{'} =(\mu_{1}+1,\ldots,\mu_{n} +1) =\mu + \rm 1
213: \hspace{-0.25em} I$ is the mean vector of X , therefore (\ref{e8})
214: becomes
215: \be Prob\{ (X,\Lambda.X) \leq k
216: \} =\alpha \label{eq8} \ee
217:
218: with $ k= -VaR_{\alpha } + \sum_{i=1}^{n} \alpha_{ii}
219: =\frac{P(0)}{2}- VaR_{\alpha } $.
220: We will suppose that $k >0$, this means that the Value-at-Risk
221: of our portfolio's is not greater than P(0)/2 .
222:
223: \begin{remark}
224: We remark that to estimate the Value-at-Risk of a
225: portfolio of securities (i.e equities), the computation of our model need as inputs the quantity $\theta_{i}$ and the initial security price
226: $S_{i}(0)$ for each $i=1..n$ as given in (\ref{e1}). Recall
227: that in literature, the computation of Quadratic
228: Delta-Gamma ($\Delta$-$\Gamma$) VaR need as inputs the sensitivity vector
229: $\Delta$ and the sensitivity matrix $\Gamma$, because of
230: the presence of derivatives products in the portfolio.
231: \end{remark}
232:
233:
234:
235:
236: \section{Reduction to an Integral Equation}
237:
238: In this section, we will reduce the problem of computation of the Value at Risk for quadratic portfolio
239: of equities to the study of the asymptotic behavior of the density function distribution over the hyper-sphere.
240:
241:
242: \subsection{ Notions of Elliptic Distributions}
243:
244:
245: The following definitions will be given as in \cite{EMS}(2002) .
246: \subsubsection{Spherical Distribution}
247:
248: \begin{defi}
249:
250: A random vector $ X=(X_1 , X_2 ,..., X_n )^{t} $ has a spherical distribution if for every orthogonal map
251: $U\in R^{n \times n}$ (i.e. maps satisfying $ U U^t = U^t U = I_{n \times n} ) $
252: $$UX =_d X .\footnote{ $=_d $ denote equality in
253: distribution}$$
254: we note that: $ X\sim S_{n}(\phi ) $.If X has a density f(x) then this is equivalent to $ f(x)=g(x^t x ) = g( \| x\| ^2 ) $ for some function
255: $g: R_{+} \longrightarrow R_{+}$, so that the spherical distributions are best interpreted as those distributions
256: whose density is constant on spheres.
257: \end{defi}
258:
259:
260: Elliptical distributions extend the multivariate normal
261: $N_{n}(\mu ,\Sigma )$, for which $\mu$ is mean and $\Sigma $ is
262: the covariance matrix. Mathematically, they are the affine maps
263: of spherical distributions in $\mathbb{R}^n$.
264:
265: \subsubsection{Elliptic Distribution}
266: \begin{defi}
267: Let $ T: R^n \longrightarrow R^n $ ,
268: $ y \longmapsto Ay+\mu $ , $A \in R^{n\times n} $ , $ \mu \in R^{n} $ .
269: X has an elliptical distribution if X = T ( Y ) and $Y\sim S_{n}
270: (\phi ) $. If Y has a density $f(y) = g(y^t y) $ and if A is
271: regular (det(A) $\not = 0 $ so that $\Sigma=A^t A$ is strictly
272: positive),then $X= AY + \mu $ has a density
273: \be
274: h(x) = g((x-\mu )^t {\Sigma}^{-1} (x-\mu )) /\sqrt{det(\Sigma)} \label{pdf1}
275: \ee
276: and the contours of equal density are now ellipsoids. An
277: elliptical distribution is fully described by its mean, its
278: covariance matrix and its characteristic generator.
279: \end{defi}
280:
281:
282: \begin{itemize}
283: \item Any linear combination of an elliptically distributed
284: random vector is also elliptical with the same characteristic
285: generator $\phi$ . If $Y\sim N_{n}(\mu ,\Sigma ,\phi ) $ , $
286: b\in \mathbf{R}^m $ and $B \in \mathbf{R}^{m \times n} $ then
287: $ B.Y + b \sim N_{m}(B\mu + b ,B\Sigma B^{t} ,\phi ) $ .
288: \end{itemize}
289:
290: \subsection{ \bf{ Integral Equation with solution VaR} }
291:
292: Since X is an elliptic distribution, its density take the following form:
293:
294: $$ h_{1}(x) = h( (x-\rm 1 \hspace{-0.25em} I)) $$
295: $\rm 1 \hspace{-0.25em} I $ is the vector of unities and h is the density
296: function of $\eta $ which take the form :
297: $$h(x) = g((x-\mu ) {\Sigma}^{-1} (x-\mu )^{t})/\sqrt{det( \Sigma )} , $$
298: therefore we have the following equation
299: $$Prob\{ (X,\Lambda X ) \geq - VaR_{\alpha } +\sum_{1}^n \alpha_{ii} \} = 1-\alpha= I(k) $$
300: with $VaR_{\alpha} $, as solution such that $ k=-VaR_{\alpha }
301: +\sum_{1}^n \alpha_{ii} $.
302:
303:
304: In terms of our elliptic distribution parameters we have to solve
305: the following equation:
306:
307: \be I(k)= \int_{\{ (x,\Lambda .x ) \geq k \} }
308: h_{1}(x) dx =1-\alpha \label{eq1} \ee
309:
310: with $ X \sim E_{n}( \mu+1 , \Sigma , \phi ) $, $ AA^t = \Sigma $
311: and $I(k)$ given as follow:
312: $$I(k)= \int_{\{ (x,\Lambda .x ) \geq k \} } g((y-\mu -\rm 1 \hspace{-0.25em} I)^{t} {\Sigma}^{-1} (y-\mu-\rm 1 \hspace{-0.25em} I )) \frac{dy}{\sqrt{det( \Sigma )}} $$
313: since $\Lambda $ is a diagonal matrix with all positive diagonal
314: values, we decompose $\Lambda = \Lambda^{1/2} \cdot \Lambda^{1/2}$
315: therefore the equation (\ref{eq1}) becomes
316: $$I(k)= \int_{\{ <\Lambda^{1/2} (Az+\mu + \rm 1 \hspace{-0.25em} I),\Lambda^{1/2} (Az + \mu + \rm 1 \hspace{-0.25em} I)> \geq k \} } g( \| z\|^{2} ) dz = \int_{\ {\| \Lambda^{1/2} (Az+\mu + \rm 1 \hspace{-0.25em} I)\|}_{2}^2 \leq k \} } g( \| z\|^{2} ) dz . $$
317: Cholesky decomposition states that $ \Sigma = AA^{t} $ when $\Sigma$ is
318: suppose to be positive, therefore if we changing the variable
319: $z=A^{-1}(y-\mu -\rm 1 \hspace{-0.25em} I)$, the precedent
320: integral becomes :
321: $$I(k)= \int_{\ (Az+\mu + \rm 1 \hspace{-0.25em} I)^{t} \Lambda (Az + \mu + \rm 1 \hspace{-0.25em} I) \geq k \} } g( \| z\|^{2} ) dz . $$
322: If we do the following decomposition $(Az+\mu + \rm 1
323: \hspace{-0.25em} I)^{t} \Lambda (Az + \mu + \rm 1 \hspace{-0.25em}
324: I) =(z+v)^{t}D(z+v) + \delta $, with $D=A^{t}\cdot \Lambda
325: \cdot A$ , $ v=A^{-1}(\mu +\rm 1 \hspace{-0.25em} I ) $ and
326: $\delta =0$, after some elementary calculus
327: $$I(k)= \int_{\{ (z+v)^{t}D(z+v) \geq k - \delta \} } g( \| z\|^{2} ) dz .$$
328: If we suggesting $k_{1}= k - \delta=R^2 $, $ z+v=u $, $ dz = du$,
329: we find that
330: $$ I(k) = \int_{\{ u^{t}. D.u \geq R^2 \} } g( \|u-v\|^{2} ) dz = 1-\alpha . $$
331: By introducing the variable $z=D^{1/2}u/R$, we have that:
332:
333: $$ I(R) = R^n \int_{\{ \| z\| \geq 1 \} } g(\| R{D^\frac{-1}{2}}z-v\|^2 ) \frac{dz}{\sqrt{det(D)}}$$
334:
335: next, by using spherical variables $z=r.\xi $ with $\xi \in
336: S_{n-1}$ and $ dz = r^{n-1} dr d\sigma(\xi )$, where $d\sigma
337: (z) $ is a elementary surface of z on $ S_{n-1} = \{ \xi | \xi
338: \in {\mathbb{R}}^{n} , \xi _{1}^{2} + \xi_{2}^{2} + ... + \xi
339: _{n}^{2} = 1 \} $ and by introducing the function $J(r,R)$ such
340: that
341: \be
342: J(r, R) = \int_{S_{n-1}} g(\| rR{D^\frac{-1}{2}}\xi -v\|^2 )
343: d\sigma (\xi ) \label{eq10}
344: \ee
345: we obtain
346: \be
347: R^{-n}\cdot I(R) = \int_{1}^{\infty } r^{n-1} \Big{[} \int_{S_{n-1}} g(\| rR{D^\frac{-1}{2}}\xi -v\|^2 ) d\sigma (\xi ) \Big{]} \frac{dr}{\sqrt{det(D)}} = \int_{1}^{\infty } r^{n-1} J(r,R) \frac{dr}{\sqrt{det(D)}} . \label{eq0}
348: \ee
349:
350: Next By introducing the function
351: \be H(s)= s^{n} \int_{1}^{\infty
352: } r^{n-1} J(r,s) dr ,\label{funct} \ee
353:
354: our goal will be to solve the
355: following equation \be H(s)=(1-\alpha )\sqrt{det(D)}
356: .\label{equat}\ee
357:
358: In the following section, we propose to approximate J(r,R) by
359: applied the numerical methods giving in the paper of Alan Genz
360: (2003), ( see \cite{G} for more details).
361:
362:
363:
364:
365: \subsection{\bf{ Numerical approximation of J(r,R)} }
366:
367:
368: In this section, we estimate the integral J(r,R) by a numerical
369: methods given by Alan Genz in \cite{G} .
370:
371: \subsection{\bf{Some interpolation rules } on $S_{n-1}$}
372:
373: The paper \cite{G} of Alan Genz, give the following method.
374: Suppose that we need to estimate the following integral
375: $$J(f)=\int_{S_{n-1}} f(z) d\sigma (z) $$
376: where $d\sigma (z) $ is an element of surface on $S_{n-1} = \{ z |
377: z \in {\mathbb{R}}^{n} , z_{1}^{2} + z_{2}^{2} + ... + z_{n}^{2}
378: = 1 \} $.
379:
380: In effect, let be the n-1 simplex by $T_{n-1}= \{ x |
381: x \in {\mathbb{R}}^{n-1} , 0 \leq x_{1} + x_{2} + ... + x_{n-1}
382: \leq 1 \} $ and for any $ x\in T_{n-1}$, define $x_{n} = 1-
383: \sum_{i=1}^{n-1} x_{i} $ . Also $t_{p}=(
384: t_{p_{1}},...,t_{p_{n-1}})$ if points $t_{0}$, $t_{1}$,\ldots ,
385: $t_{m}$ are given, satisfying the condition :
386: $|t_{p}|=\sum_{i=1}^n t_{p_{i}} = 1 $ whenever $\sum_{i=1}^n
387: p_{i} = m $, for non-negative integers $p_{1}$,\ldots, $p_{n}$ ,
388: then the Lagrange interpolation formula (sylvester \cite{SY} for a
389: function g(x) on $T_{n-1}$ is given by
390:
391: $$ L^{(m,n-1)} (g,x) = \sum_{|p|=m} \prod_{i=1}^{n} \prod_{j=0}^{p_{i}-1} \frac{ x_{i}^{2} - t_{j}^{2}}{ t_{p_{i}}^{2} - t_{j}^{2}} g(t_{p}) $$
392: $L^{(m,n-1)} (g,x) $ is the unique polynomial of degree m which
393: interpolates g(x) at all of the $C_{ m+n-1}^m $ points in the
394: set $\{x | x=(t_{p_{1}},...,t_{p_{n-1}}) , |p|=m \}$. Silvester
395: provided families of points, satisfying the condition $|t_{p}|=1
396: $ when $|p|=m$, in the form $ t_{i}=\frac{i+\mu}{m + \theta n}$
397: for i=0,1,\ldots,m, and $\mu$ real. If $ 0\leq \theta \leq 1$ ,
398: all interpolation points for $L^{(m,n-1)} (g,x) $ are in
399: $T_{n-1}$. Sylvester derived families of interpolators rules for
400: integration over $T_{n-1}$ by integrating $L^{(m,n-1)} (g,x) $ .
401: Fully symmetric interpolar integration rules can be obtained by substitute $ x_{i}=z_{i}^2 $ ,
402: and $t_{i}=u_{i}^2 $ in $L^{(m,n-1)} (g,x) $, and define
403: $$M^{(m,n)} (f,z) = \sum_{|p|=m} \prod_{i=1}^{n} \prod_{j=0}^{p_{i}-1} \frac{ z_{i}^{2} - u_{j}^{2}}{ u_{p_{i}}^{2} - u_{j}^{2}} f\{u_{p}\} $$
404: where $f\{u\}$ is a symmetric sum defined by
405: $$f\{u\}= 2^{-c(u)} \sum_{s} f(s_{1}u_{1},s_{2}u_{2},\ldots,s_{n}u_{n})$$
406: with c(u) the number of nonzero entries in $(u_{1},\ldots,u_{n})$,
407: and the $\sum_{s} $ taken over all of the signs combinations that
408: occur when $s_{i}=\pm 1$ for those i with $u_{i}$ different to
409: zero.
410:
411: \begin{lemma}
412: If
413: $$ w_{p} =J( \prod_{i=1}^{n} \prod_{j=0}^{p_{i}-1} \frac{ z_{i}^{2} - u_{j}^{2}}{ u_{p_{i}}^{2} - u_{j}^{2}} ) $$
414: then
415: $$
416: J(f) = R^{(m,n)}(f) = \sum_{|p|=m} w_{p} f\{u_{p}\}
417: $$
418:
419: $$f\{u\}= 2^{-c(u)} \sum_{s} f(s_{1}u_{1},s_{2}u_{2},...,s_{n}u_{n})$$,
420: with c(u) the number of nonzero entries in $(u_{1},...,u_{n})$,
421: and the $\sum_{s} $ taken over all of the signs combinations that
422: occur when $s_{i}=\pm 1$ for those i with $u_{i}$ different to
423: zero.
424:
425: \end{lemma}
426:
427:
428: The proof is given in \cite{G} by (Alan Genz (2003)) as follow:\\
429: Let $z^{k}=z_{1}^{k_{1}}z_{2}^{k_{2}}\cdot z_{n}^{k_{n}}.$ J and R are both linear functionals, so it is sufficient to show that
430: $R^{(m,n)}(z^{k})= J(z^{k})$ whenever $|k|\leq 2m+1$. If k has
431: any component $k_{i}$ that is odd, then $J(z^{k})=0$, and
432: $R^{(m,n)}(z^{k})= 0$ because ever term $u_{q}^{k}$ in each of the
433: symmetry sums $f\{u_{p}\} $ has a cancelling term - $u_{q}^{k}$ .
434: Therefore, the only monomials that need to be considered are of
435: the form $z^{2k}$, with $|k|\leq m$. The uniqueness of
436: $L^{(m,n-1)} (g,x) $ implies $L^{(m,n-1)} (x^{k},x)=x^{k} $
437: whenever $|k|\leq m$, so $M^{(m,n)}(z^{2k},z)=z^{2k}$,
438: whenever $|k|\leq m$.
439: Combining these results:\\
440: $$
441: J(f)=M^{(m,n)}(f,z)) =\sum_{|p|=m} w_{p} f\{u_{p}\}
442: =R^{(m,n)}(z^{k})
443: $$
444: whenever $f(z)=z^{k}$, with $|k|\leq 2m+1$, so $R^{(m,n)}(f)$ has
445: polynomial degree 2m+1. For more details ( cf. Genz \cite{G}).
446:
447: \subsection{Application to Numerical approximation of $J(r,R)$}
448:
449:
450: Since our goal is to estimate the integral (\ref{eq1}), it is straightforward that the theorem (4.2.1)
451: is applicable to the function f such that
452:
453: $$f(z) = g(\| rR{D^\frac{-1}{2}}z -v\|^2 ).$$
454:
455: Next, since we have that
456: $$f\{u_{p}\} = g(\| rR{D^\frac{-1}{2}}(s.u_{p})^{t} -v\|^2 ),$$
457: by introducing the approximate function $J_{u_{p}}$ that depend to
458: the choice of $ u_{p}$, (\ref{eq1}) becomes
459:
460: \be J(r,R) \approx \sum_{|p|=m} \sum_{s} w_{p} \quad g(\|
461: rR{D^\frac{-1}{2}}(s.u_{p})^t -v\|^2 ) =J_{u_{p}}(r,R)
462: \label{approx}\ee
463:
464: we note $s.u_{p}={(s_{1}u_{1},...,s_{n}u_{n})}^t$.
465: \begin{remark}
466: $ J_{u_{p}}(r,R)$ is the numerical approximation of $J(r,R)$ as
467: given in (\ref{approx}), is depend to the choice of interpolation
468: points $u_{p}$ on hypersphere. Recall that $J(r,R)$ was a fixed
469: function that depend to R and the density function of our
470: elliptic distribution .
471: \end{remark}
472:
473:
474: By introducing $H_{u_{p}}$, the approximate
475: function of $H$ as define in (\ref{funct}) that depend of $J_{u_{p}}$, such that
476:
477: \be H_{u_{p}}(s)=s^{n} \int_{1}^{\infty
478: } r^{n-1} J_{u_{p}}(r,s) dr \approx H(s) .\label{funct1} \ee
479:
480: By replace $H(s)$ in (\ref{equat}) by $H_{u_{p}}(s)$ we then
481: prove the following result:
482: \begin{theorem}\label{Quad-VaR-elliptic}
483: If we have a quadratic portfolio of securities (i.e equities)
484: such that the Profit \& Loss function over the time
485: window of interest is, to good approximation, given by $\Delta \Pi
486: \approx \sum_{i=1}^{n} S_{i}(0)
487: \cdot\theta_{i} ( \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} ) $,
488: with portfolio weights $\theta_{i}$. Suppose moreover that the
489: joint log-returns is a random vector $(\eta_{1},\ldots,\eta_{n})$
490: that follows a continuous elliptic distribution, with probability
491: density as in $(\ref{pdf1})$,where $\mu$ is the vector mean and
492: $\Sigma$ is the variance-covariance matrix, and where we suppose
493: that $g(s^2 )$ is integrable over $\mathbb{R}$, continuous and
494: nowhere 0. Then the approximate portfolio's quadratic elliptic
495: $VaR_{\alpha,u_{p}}^g$ at confidence ($1-\alpha$) is
496: given by
497: \be
498: VaR_{\alpha,u_{p}}^g= \frac{P(0)}{2} - R_{g,u_{p}}^2
499: \label{theo01}
500: \ee
501: where $R_{g,u_{p}}$ is the unique solution of the equation
502: \be H_{u_{p}}(s) =(1-\alpha)\cdot \sqrt{det(D)}=
503: \frac{(1-\alpha)}{2^{n/2}}\sqrt{det(\Sigma)\prod_{i=1}^n
504: {\theta_{i}\cdot S_{i}}(0)}.\label{theo1} \ee
505:
506: In this case, we assume that our losses will not be greater than
507: half-price of the portfolio at time 0.
508: \end{theorem}
509:
510: \begin{remark}
511: The precedent theorem give to us an approximate Quadratic Portfolio Value-at-Risk ($VaR_{\alpha,u_{p}}^g$) that depend to our choice of interpolation
512: points on hypersphere ,$\alpha$ and the function g. Therefore it is
513: clear that the best choice of interpolation point will depend
514: to the g function in (\ref{pdf1}).
515: \end{remark}
516:
517:
518:
519:
520: With some simple calculus we have
521: the following remark
522:
523: \begin{remark}\label{rem}
524:
525: \be J_{u_{p}}(r,R)= \sum_{|p|=m} w_{p} \sum_{s}
526: g\Big{(}a(s,u_{p},R)\cdot r^{2} -2\cdot b(s,u_{p},R,D,v)\cdot r +
527: c(v)\Big{)}
528: \label{notJ} \ee
529: for which
530: $a(s,u_{p},R,D)=\|R{D^\frac{-1}{2}}(s.u_{p})\|^{2}$,
531: $b(s,u_{p},R,D,v)=R<{D^\frac{-1}{2}}(s.u_{p}),v>$,$ c=\|v\|^2$.
532: Sometimes, for more simplification we will note a,b,c.
533: \end{remark}
534:
535:
536:
537: Since inequality of Schwartz give that $b^2-ac <0$ , we use the
538: change of variable by posing $b1 =\frac{b^2 -ac}{a}<0$,
539: $u=r-\frac{b}{a}$, by using using the binom of Newton, and by introducing the function $G^{j,g}$ for $j=0,..,n-1$, such that we have the following
540: remark
541: \begin{remark}\label{rem1}
542:
543: \be
544: J(r,R)=\sum_{|p|=m}
545: w_{p} \sum_{s} \sum_{j=0}^{n-1} \binom{n-1}{j} { (b/a)}^{n-1-j}
546: G_{u_{p},s}^{j,g} (R) \label{cour1}
547: \ee
548: with
549: \be
550: G_{u_{p},s}^{j,g} (R)=\int_{1-\frac{b}{a}}^{\infty } z^j \cdot g(a
551: z^2 -b_{1} ) dz \label{note3}\ee
552: for which a,b and c are defined in (\ref{rem}).
553: \end{remark}
554:
555: By replace $d= b/a=\frac{<{D^\frac{-1}{2}}(s.u_{p}),v>}{R\|{D^\frac{-1}{2}}(s.u_{p})\|^{2}}$ by its value in
556: (\ref{rem1}),we obtain
557: \be
558: G_{u_{p},s}^{j,g} (R)= \int_{1-\frac{<{D^\frac{-1}{2}}(s.u_{p}),v>}{R\|{D^\frac{-1}{2}}(s.u_{p})\|^{2}}}^{\infty }
559: z^j \cdot g \Big{(}R^2 \cdot\|{D^\frac{-1}{2}}(s.u_{p})\|^{2} z^2
560: -
561: \frac{<{D^\frac{-1}{2}}(s.u_{p}),v>^2}{\|{D^\frac{-1}{2}}(s.u_{p})\|^{2}}
562: +\|v\|^2 \Big{)} dz \label{not2} \ee
563:
564: we then have the following theorem
565: \begin{theorem}
566: If we have a quadratic portfolio of securities (i.e equities) for which the joint
567: securities log-returns changes with continuous elliptic
568: distribution with pdf distribution as in $(\ref{pdf1})$, then the approximate portfolio's quadratic elliptic
569: $VaR_{\alpha,u_{p}}^g$ at confidence ($1-\alpha$) is
570: given by
571: \be
572: VaR_{\alpha,u_{p}}^g= \frac{P(0)}{2} - R_{g,u_{p}}^2
573: \label{theo01}
574: \ee
575: where $R_{g,u_{p}}$ is the unique solution of the equation
576: \be
577: \sum_{|p|=m} w_{p} \sum_{s} \sum_{j=0}^{n-1}
578: \binom{n-1}{j} R^{j+1} \cdot
579: {(\frac{<{D^\frac{-1}{2}}(s.u_{p}),v>}{\|{D^\frac{-1}{2}}(s.u_{p})\|^{2}})}^{n-1-j}
580: \cdot G_{u_{p},s}^{j,g} (R) =
581: \frac{(1-\alpha)}{2^{n/2}}\sqrt{det(\Sigma)\prod_{i=1}^n
582: {\theta_{i}\cdot S_{i}}(0)} \label{cor1} .\ee
583: In this case, we assume that our losses will not be greater than
584: half-price of the portfolio at time 0.
585: \end{theorem}
586:
587: \begin{remark}
588: We have reduced our problem to one dimensional integral equation. Therefore, to get an explicit equation to
589: solve, we need to estimate $G_{u_{p},s}^{j,g} (R) $ that depend to
590: R with parameters $g$,$u_{p}$,v and D.
591: \end{remark}
592:
593: Therefore, in the case of normal distribution or $t$-distribution,
594: it will suffices to replace g in the expression of (\ref{note3}),
595: an to estimate the one dimensional integral (\ref{note3}).
596:
597:
598:
599:
600: \subsubsection{\bf{ The case of normal distribution}}
601:
602:
603:
604: In the case of normal distribution, the pdf is given by:
605: \be f(x)=\frac{1}{\sqrt{(2\pi)^n |\Sigma|}}exp(-\frac{1}{2}
606: (x-\mu)\Sigma^{-1}(x-\mu)^t )\label{pdfnorm}\ee
607: and specific is given as follow
608: $$ g(x)=(2\pi)^{-\frac{n}{2}} e^{-\frac{x}{2}}=C(n).e^{-\frac{x}{2}}$$
609: therefore it suffices to replace g in ($\ref{note3}$) then
610:
611: \be
612: G_{u_{p},s}^j (R)=(2\pi)^{-\frac{n}{2}} e^{\frac{b_{1}}{2} }
613: \int_{1-\frac{b}{a}}^{\infty } u^j e^{-\frac{a u^2}{2}} du
614: .\label{eq12} \ee
615:
616:
617: If $1-\frac{b}{a} >0 $ ( it is the case when R is
618: sufficiently big such that $|v| < R
619: \|{D^\frac{-1}{2}}(s.u_{p})\|$ ).
620:
621: \be
622: \frac{G_{u_{p},s}^j (R)}{\exp({-\frac{\|v\|^2}{2}})(2\pi)^{-\frac{n}{2}}}=
623: \exp{(\frac{<{D^\frac{-1}{2}}(s.u_{p}),v>^2}{R\|{D^\frac{-1}{2}}(s^{t}u_{p})\|^{2}})}
624: {(2/a)}^{\frac{1+j}{2}} \Gamma(\frac{j+1}{2},\frac{(R\|{D^\frac{-1}{2}}(s^{t}u_{p})\|)^2}{2}{(1-\frac{<{D^\frac{-1}{2}}(s.u_{p}),v>}{R\|{D^\frac{-1}{2}}(s^{t}u_{p})\|^{2}})}^2
625: ) \label{equa12} \ee
626:
627: therefore, since $ a =(R\|{D^\frac{-1}{2}}(s.u_{p})\|)^2$ we have
628: the following theorem
629:
630:
631:
632: \begin{theorem}
633: If we have a portfolio of securities (i.e equities), such that the
634: Profit \& Loss function over the time window of interest is, to
635: good approximation, given by $\Delta \Pi
636: \approx \sum_{i=1}^{n} S_{i}(0)
637: \cdot\theta_{i} ( \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} ) $,
638: with portfolio weights $\theta_{i}$. Suppose moreover that the
639: joint log-returns is a random vector $(\eta_{1},\ldots,\eta_{n})$
640: that follows a continuous multivariate normal distribution with
641: density function in (\ref{pdfnorm}), vector mean $\mu$ , and
642: variance-covariance matrix $\Sigma $, the Quadratic Value-at-Risk
643: $(VaR_{\alpha,u_{p}}$ at confidence $1 - \alpha $ is given by the
644: following formula
645: $$R_{u_{p},\alpha}^2 = -VaR_{\alpha,u_{p}}+\frac{P(0)}{2}$$
646: for which $R_{u_{p},\alpha}$ is the unique solution of the
647: following transcendental equation.
648:
649: \be
650: 2(1-\alpha)\frac{\sqrt{det(D)}}{(2\pi)^{\frac{n}{2}}}=\sum_{|p|=m}
651: w_{p} \sum_{s}\frac{(<{D^\frac{-1}{2}}(s.u_{p}),v>)^{(n-j-1)}
652: }{\|{D^\frac{-1}{2}}(s.u_{p})\|^{(2n-1-j)}}
653: e^{\frac{b_{1}}{2}}\sum_{j=0}^{n-1} \binom{n-1}{j}
654: \Gamma(\frac{j+1}{2},\frac{a}{2}{(1-\frac{b}{a})}^2 )
655: \label{equa12} \ee
656: for which
657: $b_{1}=\frac{(<{D^\frac{-1}{2}}(s.u_{p}),v>)^2 }
658: {\|{D^\frac{-1}{2}}(s.u_{p})\|^{2}}-\|v\|^2 $,
659: $\frac{b}{a}=\frac{<{D^\frac{-1}{2}}(s.u_{p}),v> }{R\|
660: D^\frac{-1}{2} (s.u_{p})\|^2 } $, $a=R^2
661: \|{D^\frac{-1}{2}}(s.u_{p})\|^{2}$. In this case, we implicitly
662: assume $VaR_{\alpha;u_{p}} \leq P(0)/2$. $\Gamma$ is the
663: incomplete gamma function.
664: \end{theorem}
665:
666: \subsubsection{\bf{Case of $t$-student distribution}}
667:
668:
669: If our elliptic distribution is in particular chosen as the
670: multivariate t-student distribution, we will have density
671: function given by
672: \be g(x)=\frac{\Gamma (\frac{\nu +
673: n}{2})}{\Gamma(\nu/2). \pi^{n/2} } {\Big{(}1+\frac{x }{\nu}
674: \Big{)}}^{(\frac{-\nu-n}{2})}=C(n,\nu){\Big{(}1+\frac{x }{\nu}
675: \Big{)}}^{(\frac{-\nu-n}{2})}\label{pdf3}
676: \ee
677: therefore by replacing g in ($\ref{cor1}$), we obtain the equation
678: \be
679: \sum_{|p|=m} w_{p} \sum_{s} \sum_{j=0}^{n-1}
680: \binom{n-1}{j} {(b/a)}^{n-1-j}
681: \int_{1-\frac{b}{a}}^{\infty } u^j {\Big{(}1+\frac{ a u^2 -
682: b_{1}^2 }{\nu} \Big{)}}^{(\frac{-\nu-n}{2})} du = \frac{(1-\alpha)
683: \sqrt{det(D)}}{ C(n,\nu)R^{n}}\label{equstud0} .\ee
684:
685: suggesting $ c_{1}=\nu - b_{1}^2 $ , our equation is reduce to
686: \be
687: \sum_{|p|=m} w_{p} \sum_{s} \sum_{j=0}^{n-1}
688: \binom{n-1}{j} {(b/a)}^{n-1-j}
689: \int_{1-\frac{b}{a}}^{\infty } u^j {\Big{(}a u^2 +c_{1}
690: \Big{)}}^{(\frac{-\nu-n}{2})} du = \frac{(1-\alpha)
691: \sqrt{det(D)}}{\nu^{\frac{\nu+n}{2} }
692: C(n,\nu)R^{n}}\label{equstud1} .\ee
693:
694: changing variable in this integral according to $v=u^2 $ and
695: $\beta=\frac{a}{c_{1}}$, we find that
696: \be
697: R^{n} \sum_{|p|=m} w_{p} \sum_{s} \sum_{j=0}^{n-1}
698: \binom{n-1}{j} {(b/a)}^{n-1-j} c_{1}^{\frac{-n-\nu}{2}}
699: \int_{{(1-\frac{b}{a})^2}}^{\infty } v^{\frac{j+1}{2}-1}
700: {\Big{(}\beta v + 1 \Big{)}}^{(\frac{-\nu-n}{2})} du =
701: \frac{(1-\alpha) \pi^{n/2}\Gamma(\nu/2)
702: \sqrt{det(D)}}{\nu^{\frac{\nu+n}{2} } \Gamma (\frac{\nu + n}{2})
703: }\label{equstud1} .\ee
704:
705: For the latter integral equation, we will
706: use the following formula from \cite{GR}:
707:
708: \begin{lemma} (cf. \cite{GR}, formula 3.194(2)).
709: If $|arg (\frac{u}{\beta})|<\pi $, and $Re(\nu_{1})>Re(\mu)>0$ ,
710: then \be \int_{u}^{+\infty}
711: x^{\mu-1} (1 + \beta x)^{-\nu_{1}}
712: dx= \frac{u^{\mu-\nu_{1}}
713: \beta^{-\nu_{1}}}{\nu_{1} -\mu }
714: {_2F}_{1}(\nu_{1},\nu_{1}-\mu;\nu_{1}-\mu+1;-\frac{1}{\beta \cdot
715: u}).
716: \label{rem13} \ee
717: Here $_2 F _1 (\alpha ; \beta , \gamma ; w ) $ is the
718: hypergeometric function.
719: \end{lemma}
720: In our case,$\nu_{1}=\frac{\nu +n}{2}$, $u=(1-b/a)^2$, $\nu_{1}-\mu=\frac{n +\nu
721: -j-1}{2}$, $\nu_{1}-\mu+1=\frac{n +\nu -j+1}{2}$ therefore
722: If we replace in $(\ref{equstud1})$, we will
723: obtain the following result.
724:
725:
726:
727:
728:
729:
730: \begin{theorem}
731: If we have a portfolio of securities (i.e equities), such that the
732: Profit \& Loss function over the time window of interest is, to
733: good approximation, given by $\Delta \Pi
734: \approx \sum_{i=1}^{n} S_{i}(0)
735: \cdot\theta_{i} ( \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} ) $,
736: with portfolio weights $\theta_{i}$. Suppose moreover that the
737: joint log-returns is a random vector $(\eta_{1},\ldots,\eta_{n})$
738: that follows a continuous multivariate $t$-distribution with
739: density function given by (\ref{pdf3}), vector mean $\mu$ , and
740: variance-covariance matrix $\Sigma $, the Quadratic Value-at-Risk
741: $(VaR_{\alpha,u_{p}}$ at confidence $1 - \alpha $ is given by the
742: following formula
743: $$R_{u_{p},\alpha}^2 = -VaR_{\alpha,u_{p}}+\frac{P(0)}{2}$$
744: for which $R_{u_{p},\alpha}$ is the unique solution of the
745: following transcendental equation.
746: \be \frac{R^{n}}{(1-\alpha)} \sum_{|p|=m} w_{p}
747: \sum_{s}\sum_{j=0}^{n-1}\binom{n}{j}\frac{(b/a)^{n-1-j}}{ {(\nu
748: -b_{1}^2 )}^\frac{-n-\nu}{2}} \frac{ {_2 F}_1 \left[ \frac{n+\nu
749: }{2 } , \frac{n +\nu-j-1 }{2 } ; \frac{n+\nu-j+1 }{2 } ; \frac{
750: b_{1}^2
751: -\nu}{a(1-\frac{b}{a})^2}\right]}{(n+\nu-j-1)\sqrt{det(\Sigma
752: )\prod_{i=1}^n {\theta_{i}\cdot S_{i}}(0)}} =\frac{
753: (\pi/2)^{n/2}\Gamma(\nu/2) }{\nu^{\frac{\nu+n}{2} } \Gamma
754: (\frac{\nu + n}{2}) }
755: \label{th}\ee
756: for which
757: $b_{1}=\frac{(<{D^\frac{-1}{2}}(s.u_{p}),v>)^2 }
758: {\|{D^\frac{-1}{2}}(s.u_{p})\|^{2}}-\|v\|^2 $,
759: $\frac{b}{a}=\frac{<{D^\frac{-1}{2}}(s.u_{p}),v> }{R\|
760: D^\frac{-1}{2} (s.u_{p})\|^2 } $, $a=R^2
761: \|{D^\frac{-1}{2}}(s.u_{p})\|^{2}$ . In this case, we implicitly
762: assume that $VaR_{\alpha,u_{p}}\leq P(0)/2 $
763: \end{theorem}
764: \begin{remark}
765: \rm{ Note that, Hypergeometric $_2 F _1 $'s have been extensively
766: studies, and numerical software for their evaluation is available
767: in Maple and in Mathematica.}
768: \end{remark}
769:
770:
771:
772: \section{ Quadratic VaR with mixture of elliptic Distributions}
773:
774:
775:
776: Mixture distributions can be used to model situations where the
777: data can be viewed as arising from two or more distinct classes of
778: populations; see also \cite{MX}. For example, in the context of
779: Risk Management, if we divide trading days into two sets, quiet
780: days and hectic days, a mixture model will be based on the fact
781: that returns are moderate on quiet days, but can be unusually
782: large or small on hectic days. Practical applications of mixture
783: models to compute VaR can be found in Zangari (1996), who uses a
784: mixture normal to incorporate fat tails in VaR estimation. In this
785: section, we sketch how to generalize the preceding section to the
786: situation where the joint log-returns follow a mixture of elliptic
787: distributions, that is, a convex linear combination of elliptic
788: distributions.
789:
790: \medskip
791: \begin{defi} \rm{We say that $ (X_{1},\ldots,X_{n})$ has a joint distribution
792: that is the mixture of $q $ elliptic distributions
793: $N(\mu_{j},\Sigma_{j},\phi_{j}) $\footnote{or
794: $N(\mu_{j},\Sigma_{j},g_{j})$ if we parameterize elliptical
795: distributions using $g_{j} $ instead of $\phi_{j} $}, with weights
796: $\{\beta_{j}\}$ (j=1,..,q ; $\beta_{j} > 0$ ; $\sum_{j=1}^q
797: \beta_{j} = 1$), if its cumulative distribution function can be
798: written as
799: $$
800: F_{X_{1},\ldots,X_{n}}(x_{1},\ldots,x_{n}) = \sum_{j=1}^q
801: \beta_{j} F_{j}(x_{1},\ldots,x_{n})
802: $$
803: with $F_{j}(x_{1},\ldots,x_{n}) $ the cdf of
804: $N(\mu_{j},\Sigma_{j},\phi_{j}) $. }
805:
806: \end{defi}
807:
808: \begin{remark} \rm{
809: In practice, one would usually limit oneself to $q = 2 $, due to
810: estimation and identification problems; see \cite{MX}. }
811: \end{remark}
812:
813: We will suppose that all our elliptic distributions
814: $N(\mu_{j},\Sigma_{j},\phi_{j}) $
815: admit a pdf :\\
816: \be f_{j}(x)= |\Sigma _j |^{-1/2 } g_{j}((x-\mu_{j} )
817: {\Sigma_{j}}^{-1} (x-\mu_{j} )^t ) \label{med} \ee
818: for which each $g_{j}$
819: is continuous integrable function over $\mathbb{R}$, and that the
820: $g_{j}$ never vanish jointly in a point of $\mathbb{R}^q $.
821: The pdf of the mixture
822: will then simply be $\sum _{j = 1 } ^q \beta _j f_j (x) . $
823: \medskip
824:
825: Let
826: $$\Sigma_{j} = A_{j} ^t \; A_{j} $$
827:
828: So, following (\ref{eq0}),we introduce $J_{k}(r,R)$ such that
829: \be
830: \alpha R^{-n} = \sum_{k=1}^{q} \int_{1}^{\infty } r^{n-1} \Big{[}
831: \int_{S_{n-1}} g_{k}(\| rR{D_{k}^\frac{-1}{2}}\xi -v_{k}\|^2 )
832: d\sigma (\xi ) \Big{]} \frac{dr}{\sqrt{det(D_{k})}}
833: =\sum_{k=1}^{q} \int_{1}^{\infty } r^{n-1} J_{k}(r,R) dr
834: \label{ellip1} \ee
835:
836: Next
837: following (\ref{note3}), we introduce the function
838: \be G_{u_{p},s,k}^{j,g}
839: (R)=R^{n}\int_{1-\frac{b_{k}}{a_{k}}}^{\infty } z^j \cdot
840: g_{k}(a_{k} z^2 -b_{1k} ) dz \label{note3mix}\ee with
841: $a_{k}=\|R{D^\frac{-1}{2}}(s.u_{pk})\|^{2}$,
842: $b_{k}=R<{D_{k}^\frac{-1}{2}}(s.u_{pk}),v>$,$
843: c_{k}=\|v_{k}\|^2$,$b_{1k} =\frac{b_{k}^2 -a_{k}c_{k}}{a_{k}}$,
844: then we have the following corollary
845:
846: \begin{theorem}
847: If we have a portfolio of securities (i.e equities)
848: such that the Profit \& Loss function over the time
849: window of interest is, to good approximation, given by $\Delta \Pi
850: \approx \sum_{i=1}^{n} S_{i}(0)
851: \cdot\theta_{i} ( \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} ) $,
852: with portfolio weights $\theta_{i}$. Suppose moreover that the
853: joint log-returns is a random vector $(\eta_{1},\ldots,\eta_{n})$
854: is a mixture of $q$ elliptic distributions, with
855: density
856: $$
857: h(x)=\sum_{j=1}^q \beta _j {|\Sigma_{j}|}^{-1/2 }
858: g_{j}((x-\mu_{j} )\Sigma_{j}^{-1}(x-\mu_{j} )^{t})
859: $$
860: where $\mu_{j}$ is the vector mean, and $\Sigma_{j} $ the
861: variance-covariance matrix of the $j $-th component of the
862: mixture. We suppose that each $g_j $ is integrable function over
863: $\mathbb{R}$, and that the $g_j $ never vanish jointly in a point
864: of $\mathbb{R }^m $. Then the value-at-Risk, or {\em Quadratic
865: mixture-elliptic VaR}, at confidence $1 - \alpha $ is given as the
866: solution of the transcendental equation
867: \be
868: \sum_{k=1}^{q}\sum_{|p|=m} w_{p} \sum_{s} \sum_{j=0}^{n-1}
869: \binom{n-1}{j} {(b_{k}/a_{k})}^{n-1-j}
870: \frac{G_{u_{p},s,k}^{j,g}\Big{(}(\frac{P(0)}{2}-VaR_{\alpha,u_{p}}^{g})^{1/2}
871: \Big{)}}{ \sqrt{det(\Sigma_{k})\prod_{i=1}^n {\theta_{i}\cdot
872: S_{i}}(0)}} = \frac{(1-\alpha)}{2^{n/2}} \label{corm1} \ee
873: for which $G_{u_{p},s,k}^{j,g}$ is
874: defined in (\ref{note3mix}). In this case, we assume that our
875: losses will not be greater than half-price of the portfolio at
876: time 0.
877: \end{theorem}
878: \begin{remark} \rm{
879: One might, in certain situations, try to model with a mixture of
880: elliptic distributions which all have the same variance-covariance
881: and the same mean, and obtain for example a mixture of different
882: tail behaviors by playing with the $g_j $'s.}
883: \end{remark}
884:
885: The preceding can immediately be specialized to a mixture of
886: normal distributions: the details will be left to the reader.
887:
888:
889: \subsection{\bf{Application with mixture of Student $t$-Distributions}}
890: We will consider a mixture of q Student $t$-distributions such
891: that, the $k^{th}$ density function $i=1,..,q $ will be given by
892: \be g_{k}(x)=\frac{\Gamma (\frac{\nu_{k} +
893: n}{2})}{\Gamma(\nu_{k}/2). \pi^{n/2} } {\Big{(}1+\frac{x
894: }{\nu_{k}}
895: \Big{)}}^{(\frac{-\nu_{k}-n}{2})}=C(n,\nu_{k}){\Big{(}1+\frac{x
896: }{\nu_{k}} \Big{)}}^{(\frac{-\nu_{k}-n}{2})}\label{pdfm3}
897: \ee
898: and $\Sigma_{k}=A_{k}^t A_{k}$ therefore by replacing $g_{k}$ by g in (\ref{th}) and since integration is a linear operation,
899: we obtain the following theorem
900: \begin{theorem}
901: If we have a portfolio of securities (i.e equities)
902: such that the Profit \& Loss function over the time
903: window of interest is, to good approximation, given by $\Delta \Pi
904: \approx \sum_{i=1}^{n} S_{i}(0)
905: \cdot\theta_{i} ( \eta_{i}(t) + \frac{{\eta_{i}(t)}^{2}}{2} ) $,
906: with portfolio weights $\theta_{i}$. Suppose moreover that the
907: joint log-returns is a random vector $(\eta_{1},\ldots,\eta_{n})$
908: is a mixture of $q$ t-distributions, with
909: density
910: $$
911: h(x)=\sum_{j=1}^q \beta _j {|\Sigma_{j}|}^{-1/2 }\frac{\Gamma
912: (\frac{\nu_{j} + n}{2})}{\Gamma(\nu_{j}/2). \pi^{n/2} }
913: \Big{(}1+\frac{(x-\mu_{j} )\Sigma_{j}^{-1}(x-\mu_{j})^t }{\nu_{j}}
914: \Big{)}^{-\frac{n+\nu_{j}}{2}}
915: $$
916: where $\mu_{j}$ is the vector mean, and $\Sigma_{j} $ the
917: variance-covariance matrix of the $j $-th component of the
918: mixture. We suppose that each $g_j $ is integrable function over
919: $\mathbb{R}$, and that the $g_j $ never vanish jointly in a point
920: of $\mathbb{R }^m $. Then the value-at-Risk, or {\em Quadratic
921: mixture-student VaR}, at confidence $1 - \alpha $ is given by :
922: $$R_{u_{p}}^{2} =-VaR_{\alpha}+ \frac{P(0)}{2} $$
923: for which $R_{u_{p}}$ is the unique
924: positive solution of the following equation:
925: \be \sum_{k=1}^{q}
926: \sum_{|p|=m}\frac{w_{p}\Gamma (\frac{\nu_{k} + n}{2})}{\Gamma(\nu_{k}/2).
927: }
928: \sum_{s}\sum_{j=0}^{n-1}\binom{n-1}{j}\frac{
929: R^{n}(b_{k}/a_{k})^{n-1-j}}{ {((\nu_{k} -b_{1k}^2 )/\nu_{k}
930: )}^\frac{-n-\nu}{2}} \frac{ {_2 F}_1 \left[ \frac{n+\nu_{k} }{2 }
931: , \frac{n +\nu_{k}-j-1 }{2 } ; \frac{n+\nu_{k}-j+1 }{2 } ; \frac{
932: b_{1k}^2
933: -\nu}{a_{k}(1-\frac{b_{k}}{a_{k}})^2}\right]}{(n+\nu_{k}-j-1)
934: \sqrt{|\Sigma_{k}|\prod_{i=1}^n {\theta_{i}\cdot S_{i}}(0)}}
935: =\frac{(1-\alpha) }{(\pi/2)^{\frac{-n}{2}}} \label{th}\ee
936: for which
937: $b_{1k}=\frac{(<{D_{k}^\frac{-1}{2}}(s.u_{pk}),v_{k}>)^2 }
938: {\|{D_{k}^\frac{-1}{2}}(s.u_{pk})\|^{2}}-\|v_{k}\|^2 $,
939: $\frac{b_{k}}{a_{k}}=\frac{<{D_{k}^\frac{-1}{2}}(s.u_{pk}),v_{k}>
940: }{R\| D_{k}^\frac{-1}{2} (s.u_{pk})\|^2 } $, $a_{k}=R^2
941: \|{D_{k}^\frac{-1}{2}}(s.u_{pk})\|^{2}$ and
942: $det(\Sigma_{k})=|\Sigma_{k}|$ . In this case, we implicitly
943: assume that our losses will not be greater than $P(0)/2$.
944: \end{theorem}
945:
946:
947: \section{\bf{Elliptic Quadratic Expected Shortfall for portfolio of securities}}
948: \medskip
949:
950: Expected shortfall is a sub-additive risk statistic that describes
951: how large losses are on average when they exceed the VaR level.
952: Expected shortfall will therefore give an indication of the size
953: of extreme losses when the VaR threshold is breached. We will
954: evaluate the expected shortfall for a quadratic portfolio of
955: securities under the hypothesis of elliptically distributed risk
956: factors. Mathematically, the expected shortfall associated with a
957: given VaR is defined as:
958: $$
959: \mbox{Expected Shortfall } = \mathbb{E } (-\Delta \Pi \vert
960: -\Delta \Pi > VaR ),
961: $$
962: see for example \cite{MX}. Assuming again a multivariate elliptic
963: pdf $f(x) = {|\Sigma|}^{-1} g((x-\mu )\Sigma^{-1}(x-\mu )^{t}) $,
964: the Expected Shortfall at confidence level $1 - \alpha $ is given
965: by
966: \begin{eqnarray*}
967: - ES_{\alpha } &=& \mathbb{E } ( \Delta \Pi \mid \Delta \Pi\leq
968: -VaR_{\alpha } ) \\
969: &=& \frac{1 }{\alpha } \mathbb{E } \left( \Delta \Pi \cdot 1_{\{
970: \Delta
971: \Pi \leq -VaR_{\alpha } \} } \right) \\
972: &=& \frac{1}{\alpha }\int_{\{ (x,\Lambda.x) -P(0)/2 \leq
973: -VaR_{\alpha }\}}
974: ((x,\Lambda.x)-P(0)/2) \ h_{1}(x) \ dx \\
975: &=& \frac{ {|\Sigma|}^{-1/2}}{\alpha} \int_{\{ (x,\Lambda.x) \leq
976: -VaR_{\alpha }+ P(0)/2\}} ((x,\Lambda.x)-P(0)/2) \ g((x-\mu-1
977: )\Sigma^{-1}(x-\mu-1 )^{t}) dx .
978: \end{eqnarray*}
979: Using the definition of $VaR_{\alpha}$ and by replace $\Delta \Pi
980: =(X,\Lambda.X)-\frac{P(0)}{2}$, with random vector X define in
981: section 2,
982:
983: \be ES_{\alpha }=\frac{P(0)}{2} -\frac{ {|\Sigma|}^{-1/2}}{\alpha}
984: \int_{\{ (x,\Lambda.x) \leq -VaR_{\alpha }+ P(0)/2\}}
985: (x,\Lambda.x) \ g((x-\mu-1 )\Sigma^{-1}(x-\mu-1 )^{t}) dx
986: \label{qes1} \ee
987:
988:
989: Let $\Sigma = A^t \; A $, as before.Doing the same linear changes
990: of variables as in section 2 and section 3, we arrive at:
991: \begin{eqnarray*}
992: ES_{\alpha } &=& \frac{P(0)}{2} -\frac{
993: {R^{n+2}|D|}^{-1/2}}{\alpha}\int_{0}^{1} r^{n+1} \Big{[}
994: \int_{S_{n-1}} g(\| rR{D^\frac{-1}{2}}\xi -v\|^2 ) d\sigma (\xi
995: ) \Big{]} dr \\
996: &=& \frac{P(0)}{2}-\frac{ {R^{n+2}|D|}^{-1/2}}{\alpha}\int_{0}^{1
997: } r^{n+1} J(r,R)dr\\
998: &\approx & \frac{P(0)}{2}-\frac{
999: {R^{n+2}|D|}^{-1/2}}{\alpha}\int_{0}^{1 } r^{n+1}
1000: J_{u_{p}}(r,R)dr\\
1001: &=& \frac{P(0)}{2}-\frac{
1002: {R^{n+2}|D|}^{-1/2}}{\alpha}\sum_{|p|=m} \sum_{s} w_{p} \quad
1003: \int_{0}^{1 } r^{n+1} g(\| rR{D^\frac{-1}{2}}(s.u_{p})^t -v\|^2
1004: )\quad dr \quad, \label{qes1}
1005: \end{eqnarray*}
1006: By introducing the function $Q_{u_{p},s}^{g} $ such that
1007: \be
1008: Q_{u_{p},s}^{g}(R) =R^{n+2} \int_{0}^{1 } r^{n+1} g(\| rR{D^\frac{-1}{2}}(s.u_{p})^t -v\|^2
1009: )\quad dr \label{esquad2} \ee
1010: we have the following theorem
1011:
1012: \begin{theorem} \label{EllipticES}
1013: Suppose that the portfolio is quadratic in the risk-factors $X =
1014: (X_1 , \cdots , X_n )$: $\Delta \Pi=(X,\Lambda \cdot
1015: X)-\frac{P(0)}{2} $ and that $X \sim N(\mu+1 ,\Sigma ,\phi ) $,
1016: with pdf $f(x)= {|\Sigma|}^{-1} g((x-\mu-1 )\Sigma^{-1}(x-\mu-1
1017: )^{t}) $. If the $VaR_{\alpha}$ is given, then the expected
1018: Shortfall at level $\alpha $ is given by :\\
1019: \be ES_{\alpha }
1020: =\frac{P(0)}{2}-\frac{{|D|}^{-1/2}}{\alpha}\sum_{|p|=m} \sum_{s}
1021: w_{p} \quad Q_{u_{p},s}^{g}\Big{(}(\frac{P(0)}{2}-VaR_{\alpha})
1022: ^{1/2}\Big{)} . \label{EESformula} \ee
1023: \end{theorem}
1024: we introduce $I_{1}^g $ and $I_{2}^g$ such that
1025: $$ R^{-n-2}Q_{u_{p},s}^{g}(R)=\int_{0}^1 r^{n+1} g(a r^2 -2 b r +c )dr =\int_{0}^\infty -\int_{1}^\infty =I_{1,u_{p},s}^g (R) - I_{2,u_{p},s}^g (R) $$
1026: Following the Integral (\ref{note3})
1027:
1028: $$I_{2,u_{p},s}^g (R)=\sum_{j=0}^{n+1} \binom{n+1}{j} (b/a)^{n+1-j}\cdot G_{u_{p},s}^{j,g} (R)$$
1029: for which a,b and c are defined in remark (\ref{rem}) and
1030: $$I_{1,u_{p},s}^g (R)=\int_{0}^\infty r^{n+1} g(a r^2 -2 b r +c)dr$$
1031:
1032: \subsection{Expected Shortfall with normal distribution}
1033: In the case of normal distribution, the pdf is given by
1034: (\ref{pdfnorm}) and the specific g is given as follow
1035: $$ g(x)=(2\pi)^{-\frac{n}{2}} e^{-\frac{x}{2}}=C(n).e^{-\frac{x}{2}}$$
1036: therefore it suffices to replace g in ($\ref{note3}$) then
1037:
1038: \begin{eqnarray*}
1039: G_{u_{p},s}^j (R)&=&(2\pi)^{-\frac{n}{2}} \exp{\Big{(}\frac{b^2 -a c}{2a}\Big{) }}
1040: \int_{1-\frac{b}{a}}^{\infty } u^j \exp{(-\frac{a u^2}{2})} du\\
1041: &=&(2\pi)^{-\frac{n}{2}} \exp{\Big{(}\frac{b^2 -a c}{2a}\Big{) }}
1042: {(2/a)}^{\frac{1+j}{2}}
1043: \Gamma\Big{(}\frac{j+1}{2},\frac{a}{2}(1-\frac{b}{a})^2 \Big{)}
1044: .\label{eqes12}
1045: \end{eqnarray*}
1046: By using the following lemma
1047: \begin{lemma} (cf. \cite{GR}, formula 3.462(1)).
1048: If $Re(\nu) >0 $, and $Re(\beta)>0$ , then \be
1049: \int_{0}^{+\infty}
1050: x^{\nu-1} \exp{(-\beta x^2 - \lambda x)} dx = {(2\beta)}^{-\nu/2}
1051: \Gamma{(\nu)} \exp{\Big{(}\frac{\lambda^2}{8\beta}\Big{)}}
1052: \mathbb{D}_{-\nu}\Big{(}\frac{\lambda}{\sqrt{2\beta}}\Big{)}
1053: \label{rem3} \ee
1054: Here $\mathbb{D}_{-\nu} $ is the parabolic cylinder function with
1055: $$
1056: \mathbb{D}_{-\nu}(z)=2^{\frac{-\nu}{2}} e^{\frac{-z^2}{2}}
1057: \Big{[}\frac{\sqrt{\pi}}{\Gamma(\frac{1+\nu}{2})}\Phi\Big{(}\frac{\nu}{2},\frac{1}{2};\frac{
1058: z^2}{2}\big{)}-\frac{\sqrt{2\pi}z}{\Gamma(\frac{\nu}{2})}\Phi
1059: \Big{(}\frac{1+\nu}{2},\frac{3}{2};\frac{ z^2}{2}\big{)}\Big{]}
1060: $$
1061: where $\Phi$ is the confluent hypergeometric function (for more
1062: details see \cite{GR} page 1018).
1063: \end{lemma}
1064: we next obtain
1065: $$I_{2,u_{p},s}^g (R)=(2\pi)^{-\frac{n}{2}}\sum_{j=0}^{n+1} \binom{n+1}{j} (b/a)^{n+1-j}\cdot \exp{\Big{(}\frac{b^2 -a c}{4a}\Big{) }} {(2/a)}^{\frac{1+j}{2}}
1066: \Gamma\Big{(}\frac{j+1}{2},\frac{a}{2}(1-\frac{b}{a})^2 \Big{)}$$
1067: and
1068: \begin{eqnarray*}
1069: I_{1,u_{p},s}^g (R)&=& (2\pi)^{-\frac{n}{2}} \exp(-\frac{\|v\|^2}{2}) \int_{0}^1
1070: r^{n+1} \exp{\Big{(}-\frac{a r^2 -2 b r}{2}\Big{)}} dr\\
1071: &=& (2\pi)^{-\frac{n}{2}}\exp(-\frac{\|v\|^2}{2})
1072: {a}^{\frac{n+2}{2}}\Gamma(n+2) \exp{\Big{(}\frac{b^2}{4 a}\Big{)}}
1073: \mathbb{D}_{-n-2}\Big{(}\frac{-b}{\sqrt{a}}\Big{)}
1074: \end{eqnarray*}
1075: for which $\mathbb{D}_{-n-2}$ is the parabolic cylinder function.
1076: We have
1077: therefore prove the following result:\\
1078: \begin{theorem}
1079: Suppose that the portfolio is quadratic in the risk-factors
1080: $X=(X_1 , \cdots , X_n )$: \\
1081: $\Delta \Pi=(X,\Lambda \cdot
1082: X)-\frac{P(0)}{2} $ and that $X$ is a multivariate normal
1083: distribution, If the $VaR_{\alpha}$ is given, then the expected
1084: Shortfall at level $\alpha $ is given by :\\
1085: \be ES_{\alpha }
1086: =\frac{P(0)}{2}-\frac{R^{n+2}{|D|}^{-1/2}}{\alpha}\sum_{|p|=m}
1087: \sum_{s} w_{p}\Big{[}I_{1,u_{p},s}^g (R)-I_{2,u_{p},s}^g
1088: (R)\Big{]} . \label{EEeSformula} \ee
1089: for which
1090: $R=\sqrt{\frac{P(0)}{2}-VaR_{\alpha}}$
1091: \end{theorem}
1092:
1093: The preceding can immediately be specialized to a mixture of
1094: normal distributions. The details will be left to the reader.
1095:
1096: \subsection{Student $t$-distribution Quadratic Expected Shortfall}
1097: Following the precedent section 3,4,5, and particularly the lemma
1098: (\ref{rem13}) , the application can be specialized to a Student
1099: $t$-distribution. The details will be left to the reader.
1100:
1101:
1102:
1103:
1104:
1105:
1106:
1107:
1108:
1109:
1110:
1111: \subsection{\bf{ How to choose an interpolation points $u_{p}$ on hypersphere}}
1112: In order to obtain a good approximation of our integral, one
1113: will choose the points of interpolation $u_{p}$ of our g
1114: function such that our approximation is the best as possible.
1115: In the case where the g function decreases quickly with all its derivatives of all order, in inspiring
1116: of the classical analysis, one will choose the points which
1117: render the maximum function $\|r R D^{-1/2}(s.u_p )- v\|$.
1118: \section{\bf{ Conclusion}}
1119: By following the notion of Delta-Gamma Portfolio that contains derivatives instruments, we have introduced
1120: a Quadratic Portfolios of securities (i.e equities) without the use of Delta and Gamma.
1121: By using the assumption that the joint securities log-returns follow a mixture of elliptic distributions,
1122: we have reduced the estimation of VaR of such quadratic portfolio, to the resolution of a
1123: multiple integral equation, that contain a multiple integral over hypersphere.
1124: To approximate a multiple integral over hypersphere, we propose to use a numerical approximation method given by
1125: Alan Genz in\cite{G}. Therefore, the estimation of VaR is reduced to the resolution of
1126: one dimensional integral equation. To illustrate our method, we give special attention to mixture of normal
1127: distribution and mixture of multivariate t-student distribution.
1128: In the case of t-distribution, we need the hypergeometric special
1129: function. For given VaR, we also show how to estimate the expected
1130: shortfall of the Quadratic portfolio without derivatives
1131: instruments, when the risk factors follow an elliptic
1132: distributions and we illustrate our proposition with normal
1133: distribution by using the parabolic cylinder function. Note that
1134: this method will be applicable to capital allocation, if we could
1135: consider an institution as a portfolio of multi-lines businesses.
1136: In the sequel paper, we will dealt with this numerical Quadratic
1137: method, when the Portfolio contains derivatives instruments (i.e
1138: options). A concrete application need the estimation of $w_{p}$
1139: such that $|p|=m$, therefore we send the reader to Alan Genz
1140: \cite{G}.
1141: \appendix
1142:
1143:
1144: \newpage
1145: \begin{thebibliography}{aAaAa} \baselineskip=10pt
1146: \bibitem{BCQS} R. Brummelhuis, A. Cordoba, M. Quintanilla and L. Seco,
1147: Principal Component Value-at-Risk, {\it Mathematical Finance} {\bf
1148: 12}, 2002, pp. 23-43.
1149: \bibitem {CFKM} Cardenas, J., E.Fruchard, E.Koehler, C.Michel, and
1150: I.Thomaseau(1997): VaR: One Step beyond, RISK 10(10), 72-75.
1151: \bibitem{D} Kevin Dowd,(1998) Beyond Value-at-Risk: the new science
1152: of risk Management (Wiley series in Frontiers in Finance).
1153: \bibitem {EMS} P.Embrechts , A.McNeil and D.Strauman .
1154: Correlation and dependance in Risk Management Properties and
1155: Pitfalls. In M.A.H. Dempster, editor, Risk Management: Value at
1156: Risk and Beyond, pages 176-223. Cambridge University Press, 2002.
1157: \bibitem{DP} D.Duffie and J.Pan. An overview of value-at-risk. {\it Journal of Derivatives, 4(3):7-49,1997.}
1158:
1159: \bibitem{G}A. Genz, Fully Symmetric Interpolatory Rules for Multiple
1160: Integrals over Hyper-Spherical Surfaces, {\it J. Comp. Appl.
1161: Math.} {\bf 157} (2003), pp. 187--195.
1162: \bibitem {GR} {I.S Gradshteyn, I.M. Ryzbik Table of integrals, series, and products ( 2000) Editor Alan Jeffrey }
1163: \bibitem{HU} J.C. Hull, {\it Options, Future and Other Derivatives},
1164: $4^{th}$ Edition, , Prentice Hall, Englewood Cliffs, NJ.
1165: \bibitem{MX} Jorge Mina and Jerry Yi Xiao (2001), Return to
1166: Riskmetrics: The Evolution of the standard (www.riskmetrics.com).
1167: 2001 RiskMetrics Group Inc.
1168: \bibitem{MOR}
1169: J.P. Morgan/Reuters, RiskMetrics Technical Document, $4^{th}$
1170: Edition, 1996, J.P. Morgan.
1171: \bibitem{S2} J. Sadefo-Kamdem, Value-at-Risk and Expected
1172: Shortfall for a Linear Portfolios with Elliptically Distributed
1173: Risk Factors (2003), submitted.
1174: \bibitem {SY} P.Sylvester, Symmetric quadrature formulae for
1175: simplexes, Math. Comp. 24 (1970), pp.95-100.
1176: \bibitem {ZA} Zangari, P. (1996). An Improved Methodology for
1177: Measuring VaR, RiskMetrics Monitor, 2nd quarter pp.7-25.
1178: http://www.riskmetrics.com/research/journals
1179: \end{thebibliography}
1180: \end{document}
1181: