Berry-Esseen theorems under weak dependence (Q726800): Difference between revisions

Let \(\{\varepsilon_k \}\) be a sequence of independent and identically distributed random variables. Let \(X_k = g_k(\varepsilon_k, \varepsilon_{k-1}, \ldots)\) for real valued functions \(\{ g_k \}\), so that \(\{X_k \}\) is a zero mean, stationary, causal process. This class of processes includes many widely used linear and nonlinear processes such as \(m\)-dependent processes, ARMA and GARCH time series models and models used for dynamical systems. Assuming moments up to order \(p \in (2,3]\) and mild weak dependence conditions, optimal Berry-Esseen rate results, convergence rates in \(L^q\)-norm, and non-uniform bounds are established relating the distribution function of the normed sum \(S_n = \sum_{k=1}^n \;X_k\) to that of a standard normal variable. As a corollary, results related to the rate of convergence of moments analogous to those for the i.\,i.\,d. case developed by \textit{P. Hall} [Ann. Probab. 10, 1004--1018 (1982; Zbl 0516.60025)] are obtained for sums of dependent random variables. The proofs of the theorems utilize characteristic functions, coupling, conditioning arguments and ideal metrics, and proceed via an \(m\)-dependent approximation method where \(m \rightarrow \infty\).