Berry-Esseen and central limit theorems for serial rank statistics via graphs (Q1769775)

Let \(X_1,\ldots,X_n\) be independent and identically distributed random variables with common continuous distribution function, and let \(R(i)\) denote the rank of \(X_i\) among \(X_1,\ldots,X_n\). For any integer \(0<k<n\) set \({\mathcal N}_k=\{(i_0,\ldots,i_k)\in\{1,\ldots,n\}^{k+1}:i_j\not=i_\ell \text{ for }j\not=\ell\}\), and for \(0<r<n\), let \(A=\{a(I):I\in{\mathcal N}_r\}\) be an array of real constants. Moreover, for any \(-n+1\leq M\leq n-1\) set \(M_{\text{ mod}\;n}=M+n\) if \(M\leq0\) and \(M_{\text{ mod}\;n}=M\) if \(M>0\). Then the generalized serial rank statistic pertaining to \(A\) is defined by \[ W_A=\sum_{i=1}^na(R(i),R((i-1)_{\bmod n}), \ldots,R((i-r)_{\bmod n})). \] The first main result of the paper is a Berry-Esséen theorem for \(W_A\), under appropriate conditions on \(A\). The proof utilizes Stein's method and exploits a graph structure underlying \(W_A\) which is particularly useful to compute moments. The second main result is a central limit theorem for a sequence \(W_{A_n},n\geq1\), under appropriate regularity conditions on the sequence \(A_n,n\geq1\) (here \(r\geq1\) is fixed). Again Stein's method is used. This result extends results of \textit{D. M. Mason} and \textit{T. S. Turova} [J. Stat. Plann. Inference 91, 427--440 (2000; Zbl 0970.62028)], where the case \(r=1\) was considered.