Necessary and sufficient conditions for asymptotic normality of \(L\)- statistics (Q1203656)

Let \(T_ n=n^{-1}\sum^{n-m}_{i=k+1}c_{ni}h(X_{n:i})\) denote a general \(L\)-statistic, where \(X_{n:1}\leq\cdots\leq X_{n:n}\) are the order statistics of an i.i.d. sample of size \(n\) from a distribution function \(F\), where \(h\) is a known function, \(c_{ni}\) are known constants generated by a continuous score function \(J\), and \(k,m\geq 1\) are fixed integers. The basic regularity condition on \(J\) is regular variation at the endpoints of the unit interval. Let the random variable \(Y\) be defined by \(Y=K(\xi)\), where \(\xi\) is a uniform (0,1) random variable and \(dK=Jd(hF^{-1})\). Then \(T_ n\) is shown to be asymptotically normal with \(\sqrt n\) norming and some centering constants if and only if \(0<\text{Var}(Y)<\infty\), whereas \(T_ n\) is asymptotically normal with arbitrary norming and some centering constants if and only if the distribution of \(Y\) is in the domain of attraction of the normal distribution. In order to prove the necessity parts of these very general results, a necessary and sufficient condition for the stochastic compactness of \(T_ n\), appropriately centered and normalized, is established first, together with a representation formula for all possible subsequential limit laws.