Supermodular Order and Lundberg Exponents (Q4780927)

This paper deals with a multivariate risk process \(C=(C_{t})_{t\geq 0}\), where \(C_{t}=(C_{t}^{1},\ldots,C_{t}^{n})\) has the form \(C_{t}^{i}=u_{i}+ c_{i}t-\sum_{k=1}^{N_{t}}Z_{k}^{i},\;i=1,\ldots,n\); \(N_t\) is a point process; \((Z_{k})_{k\geq 1}\) is a sequence of \(n\)-dimensional i.i.d. random vectors such that for \(i=1,\ldots,n\) the \(i\)-th component \(Z_{k}^{i}\) of \(Z_{k}\) has the distribution function \(F_{i}\) with \(F_{i}(0-)=0\); \(u_{i}, c_{i}\) are real positive constants. Let \(u^{*}=u_1+\ldots+u_{n}\), \(c^{*}=c_1+\ldots+c_{n}\) and \(Z^{*}_{k}=Z_{k}^1+ \ldots+Z_{k}^{n}\). Then the process \(C_{t}^{*}=u^{*}+ c^{*}t-\sum_{k=1}^{N_{t}}Z_{k}^{*}\) is an one-dimensional risk process such that claims are i.i.d. and independent of \(N\). The author gives the definition of a supermodular order \(\leq_{sm}\) for random vectors and gives the notions of comonotonic random variables and mutually exclusive random variables. He proves that replacing the claim vector sequence \((Z_{k})_{k\geq 1}\) by another sequence \((\widetilde Z_{k})_{k\geq 1}\) such that \(Z_{k}\leq_{sm}\widetilde Z_{k}\) implies \(\widetilde R^{*}\leq R^{*}\), where \(R^{*}\) and \(\widetilde R^{*}\) are the corresponding infinite-time Lundberg exponents of the risk process \(C^{*}\). Bounds for the infinite-time Lundberg exponents in the comonotonic or mutually exclusive claims cases are obtained. Similar results for the finite-time ruin probability are presented. Also some results for directionally convex ordering are proposed.