A decomposition for someU-type statistics

Abstract

Define ,$S_{k,n} = \Sigma _{1 \leqslant i_1< \cdot \cdot \cdot< l_k \leqslant n} X_{i_1 } \cdot \cdot \cdot X_{i_k } ,n \geqslant k \geqslant {\text{1}}$ where {X, X _n ,n≥1} are i.i.d. random variables withEX=0,EX ²=1 and letH _k (·) denote the Hermite polynomial of degreek. By establishing an LIL for products of correlated sums of i.i.d. random variables, the a.s. decomposition

$$\begin{gathered} k!S_{k,n} = n^{k/2} H_k (S_{1n} /n^{{1 \mathord{\left/ {\vphantom {1 2}} \right. \kern-\nulldelimiterspace} 2}} ) - \left( {\begin{array}{*{20}c} k \\ 2 \\ \end{array} } \right)S_{1.n}^{k - 2} \sum\limits_{i = 1}^n {(X_i^2 - 1)} \hfill \\ + O(n^{(k - 1)/2} (\log \log n)^{(k - 3/2} ) \hfill \\ \end{gathered} $$

valid whenEX ⁴<∞, elicits an LIL forη _k,n =k!S _k,n −n ^k/2 H _k (S _1n /n ^1/2) under a reduced normalization. Moreover, whenE|X| ^p<∞ for somep in [2, 4], a Marcinkiewicz-Zygmund type strong law forη _k,n is obtained, likewise under a reduced normalization.