Tails of higher-order moments with dominatedly varying summands (Q2010121)

Given dependent, heavy-tailed random variables \(\xi_1,\ldots,\xi_n\) and their sum \(S_n=\xi_1+\cdots+\xi_n\), the authors study the limiting behaviour (as \(x\rightarrow\infty\)) of the tail moments \(\mathbf{E}(S_n^m\mathbf{1}_{\{S_n>x\}})\) for non-negative integers \(m\). This is done under some (relatively mild) assumptions on the dependence structure and tails of the underlying random variables. The main results are established under the condition that the distribution functions of each of the \(\xi_k\) is dominatedly varying, though the authors also note that more precise asymptotics can be established in the special case of consistently varying distributions. The dependence structure assumed between the random variables is motivated by quasiasymptotic independence and related notions. Particular attention is paid to the case where \(\xi_k=\theta_kX_k\), where the \(\theta_k\) are non-negative random weights; the corresponding randomly weighted sum \(S_n\) has applications in many areas of finance and actuarial science.