12.07.2015 Views

Copulas: a Review and Recent Developments (2007)

Copulas: a Review and Recent Developments (2007)

Copulas: a Review and Recent Developments (2007)

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.

(i) à 1 (x 1 ;::: ;x n )=x 1 + :::+ x n ;(ii) à 2 (x 1 ;::: ;x n )= P ni=1 (x i ¡ m) + ,wherem>0<strong>and</strong>a + = max(a; 0);(iii) à 3 (x 1 ;::: ;x n )=( P ni=1 x i ¡ m) +, m>0.The case (i) can be found in the context of Insurance when one would be interestedin some quantile of the join position X 1 + :::+ X n . Such a case occurs when consideringaggregate claims of an insurance portfolio in a given reference period, or whenobserving discount payments associated to a policy at di®erent future moments oftime, see Kaas et al. (2003). The case (ii) corresponds to analyzing an excess-of-losstreaty in reinsurance where the X i 's could be individual claims or insurance lossesdue to di®erent lines of business; The case (iii) has an interpretation to ¯nancialderivatives (e.g. Asian options) or stop-loss reinsurance, see Dhaene et al. (2002b).In practice, ¯rst we work with the marginals <strong>and</strong> afterwards with the dependencestructure choosing the underlying (more suitable) copula. Once the dependence structureof the risks is speci¯ed, the calculation of quantities related to distribution functionof Ã(X 1 ;::: ;X n ) becomes a computational issue. In many situations only partialor no information about the copula corresponding to (X 1 ;::: ;X n )isknown. Insuchcases it is useful to evaluate the bounds of the distribution function of Ã(X 1 ;::: ;X n ).Such bounds have been found ¯rstly by Makarov (1981) when n =2<strong>and</strong>Ã(x 1 ;x 2 )=x 1 + x 2 . Later Frank et al. (1987) use a copula approach extending Makarov's results(except for the optimality of the bounds) to include arbitrary increasing continuousfunctions à as follow. Let C L (u; v) =max(u + v ¡ 1; 0) <strong>and</strong> z 2 (¡1; 1). Then,P (X + Y · z) ¸ sup C L (F (x);G(y)) = Ã(z);x+y=zà ¡1 (p) =inf fF ¡1 (u)+G ¡1 (v)g; p2 (0; 1)C L (u;v)=p<strong>and</strong> VaR p (X + Y ) · à ¡1 (p). Denuit et al. (1999) extended these results for the casen ¸ 3, when the marginals are the same.Let us consider the multivariate case. Let (X 1 ;::: ;X n )ber<strong>and</strong>omvariableswithdistributions F X1 ;:::;F Xn <strong>and</strong> associate copula C. Let à :(¡1; 1) n ! (¡1; 1)be increasing <strong>and</strong> left continuous in the last argument. Denote by à k the function Ãwith the ¯xed arguments x i1 ;:::;x ik for 1 · i 1

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

Saved successfully!

Ooh no, something went wrong!