Extremes of aggregated Dirichlet risks (Q476250)

The paper deals with the asymptotic behaviour of the weighted aggregated risk \(S_p:=\sum_{i=1}^d\lambda_iX_i^p\) (where \(\lambda_i\geq 0\) and \(p>0\)) corresponding to a \({d}\)-dimensional Dirichlet random vector \(X=\left(X_1,\dots,X_d\right)\) with the parameter \(\alpha=\left(\alpha_1,\dots,\alpha_d\right)\in (0,\infty)^d\) and the radial component \(R\). Under the assumption that the distribution function of the positive random variable \(R\) belongs to the Gumbel max-domain of attraction, the author displays the exact asymptotics of the tail of \(S_p\), for \(p\in(0,1)\), \(p=1\) and \(p>1\), respectively. A converse statement is, then formulated. Similar results are obtained for the case of the Weibull max-domain of attraction. Some important consequences of these asymptotic theorems are carefully discussed. This study requires refined mathematical tools.