On the self-normalized Cramér-type large deviation (Q2576805): Difference between revisions

Let \(X_1, X_2,\dots\) be a sequence of i.i.d. random variables with mean zero and distribution in the domain of attraction of the normal law. Let \(T_n= (\sum^n_{i=1} X_i)/(\sum^n_{i=1} X^2_i)^{1/2}\) be the self-normalized sum. A Cramér-type large deviation for \(T_n\) is investigated. Namely, it is proved that \(|\log P(T_n\geq x)- \log(1- \Phi(x))|\leq A\Delta_{n,x}\) for some absolute constant \(A\) and all nonegative \(x\)'s satisfying \(A\Delta_{n,x}\leq (1+x)^2\), where \(\Delta_{n,x}\) is explicitely expressed by the first three truncated moments of \(X_1\) while \(\Phi\) is the standard normal distribution function. A non-uniform Berry-Esseen exponential bound is derived as a corollary. Under finite variance assumption similar results were obtained by \textit{B.-Y. Jing}, \textit{Q.-M. Shao} and \textit{Q. Wang} [Ann. Probab. 31, No.~4, 2167--2215 (2003; Zbl 1051.60031)].