Multivariate Normality of a class of statistics based on extreme observations

Abstract

Let $X_{1},X_{2},...$ be a sequence of independent random variables ($rv$)with common distribution function ($df$) $F$ such that $F(1)=0$ and for each $n\geq 1,$ let $X_{1,n}\leq X_{2,n}\leq ...\leq X_{n,n}$ denote the order statistics based on the n first of these random variables. Lô (\cite{gslod}) introduced a class of statistics aimed at characterizing the asymptotic behavior of the univatiate extremes. This class this estimator of the square of the extremal index of a $df$ lying in the extremal domain of attraction : $$ k^{-1}\sum_{j=\ell +1}^{j=k}\ \sum_{i=j}^{i=k}i(1-δ_{ij}/2)(\log X_{n-i+1,n}-\log X_{n-i,n}) $$ $$ \times (\log X_{n-j+1,n}-\log X_{n-j,n}t), $$ \noindent where $(k,\ell)$ is a couple of integers such that $k\rightarrow +\infty $, $k/n\rightarrow 0$, $\ell ^{2}/k\rightarrow 0$, as $n\rightarrow 0\rightarrow,$ $log$ stands for the natural logarithm and $δ_{ij\text{}}$is the Kronecker symbol. \noindent In total $\mathbb{R}^{8}$-vectors are used in this paper and include the most popular statistics used in the literature. We consider here a multivariate approach and provide the asymptotic laws of such vectors. This allows quickly finding asymptotic laws of functional of new statistics and new estimators of the extremal index such as the Dekkers \textit{et al.} (1989) and Hasofer and Wang (1992) statistics for example as in Hah et al.(2012)