The convergence properties for randomly weighted sums of the arrays of rowwise <i>m</i> -AANA random variables with related statistical application (Q6925489)

A sequence of random variables \((X_n)\) is called \(m\)-AANA (asymptotically almost negatively associated) if there exists \(q(j)\downarrow0\) such that\N\begin{align*}\N& \mathrm{Cov}(f(X_j),g(X_{j+m},\ldots,X_{j+k}))\\\N& \quad\leq q(j)\big[\mathrm{Var}(f(X_j))\mathrm{Var}(g(X_{j+m},\ldots,X_{j+k}))\big]^{1/2}\N\end{align*}\Nfor all \(j\in\mathbb N\), \(k\geq m\) and all continuous functions \(f,g\) that are non-decreasing in each coordinate and guarantee existence of variance. For \(m=1\) the sequence is called AANA and includes negatively associated random variables for \(q(j)=0\). The authors extend a result in [\textit{X. Wang} et al., Discrete Dyn. Nat. Soc. 2011, Article ID 717126, 11 p. (2011; Zbl 1235.60026)] on sufficient conditions for complete convergence of weighted sums of arrays \((X_{nj})\) of rowwise AANA variables of the form \N\[\N\sum_{n=1}^{\infty}n^{\alpha p-2}\mathbb{P}\left(\max_{i\leq n}\left|\sum_{j=1}^i A_{nj}X_{nj}\right|>b_n\varepsilon\right)<\infty\N\]\N for any \(\varepsilon>0\) to the case of rowwise \(m\)-AANA variables with random weights \((A_{nj})\) instead of constants and with a certain relaxation of assumptions. Furthermore, in the same setting the authors also provide sufficient conditions for complete moment convergence and complete \(f\)-moment convergence when replacing \(\mathbb{P}(\cdots>b_n\varepsilon)\) by \(\mathbb{E}[(b_n^{-1}\cdots-\varepsilon)_+^\alpha]\), respectively \(\mathbb{E}[f((b_n^{-1}\cdots-\varepsilon)_+)]\) for increasing functions \(f\) with \(f(0)=0\). As an application, complete consistency of the estimator in a non-parametric regression model with \(m\)-AANA errors is shown. The theoretical results are accompanied by a simulation study and an application to monthly S\&P500 data.