Point process and partial sum convergence for weakly dependent random variables with infinite variance (Q1897166)

Let \(\{\xi_j\}\) be a strictly stationary sequence of random variables such that the distribution of \(\xi_1\) belongs to the domain of attraction of a stable law with index \(\alpha\), \(0< \alpha< 2\). The main results of the paper show that under a mild mixing condition, the asymptotic stability of \(S_n:= \xi_1\cdots \xi_n\) and the large deviations property \(P (S_n> t_n)/nP(|\xi_1|> t_n)\to\text{constant}\) for \(\{t_n\}\) such that \(nP(|\xi_1|> t_n)\to 0\), are related to the asymptotic behavior of the extreme order statistics. In the study of the limit distribution of \(S_n\), this relation was made clear in the independent case by \textit{R. LePage}, \textit{M. Woodroofe} and \textit{J. Zinn} [ibid. 9, 624-632 (1981; Zbl 0465.60031)] and subsequently established by the first named author [ibid. 11, 262-269 (1983; Zbl 0511.60021)] in a dependent setting. Here, the link is given in terms of the weak convergence (as random elements of a suitable space of measures) of the point processes on \(\mathbb{R}\backslash \{0\}\), \(N_n:= \sum^n_{j= 1} \delta_{\xi_j/a_n}\), where \(a_n\) satisfies \(nP(|\xi_1|> a_n)\to 1\) (\(\delta_x\) is the probability measure concentrated at \(x\)); moreover the weak convergence of \(N_n\) is characterized. [\textit{A. Jakubowski} and \textit{M. Kobus}, J. Multivariate Anal. 29, No. 2, 219-251 (1989; Zbl 0687.60025), also used point processes methods to obtain nonnormal limit distributions for sums of weakly dependent random vectors.] For large deviations results this approach appears to be new even in the independent case. Several interesting applications are presented.