Strong consistency and rates for recursive nonparametric conditional probability density estimates under \((\alpha{}, \beta{})\)-mixing conditions (Q1177215)

Let \(\{X_ j: j\geq 1\}\) be a real-valued stationary process on a probability space \(\{\Omega,B,P\}\). For each integer \(d\geq 1\) and integers \(1\leq i_ 1\leq i_ 2\leq\cdots\leq i_ d\) the joint density of the random vector \((X_{i1},X_{i2},\dots,X_{id})\) is denoted by \(f(x)\). For any integer \(p\), \(1<p<d\), the conditional probability density function \(X_ j'=(X_{j+ip+1},\dots,X_{j+id})\) given \(X_ j''=(X_{j+i1},\dots,X_{j+ip})\) is denoted by \(f(x_ 2| x_ 1)\). A class of recursive type kernel estimators of \(f(x)\) and \(f(x_ 2| x_ 1)\), which are more convenient in actual computations, is considered. The strong consistency and the rate of convergence of the estimators are proved under the assumptions that \(\{X_ j: j\geq 1\}\) is \((\alpha,\beta)\)-mixing [\textit{R. C. Bradley} and \textit{W. Bryc}, J. Multivariate Anal. 16, 335-367 (1985; Zbl 0586.62086)] and some other regularity conditions on the bandwidths of kernel estimates. Compared with previous results in this direction obtained by \textit{E. Masry} [Stochastic Processes Appl. 32, 109-127 (1989; Zbl 0692.62034)] and \textit{E. Masry} and \textit{L. Gyoerfi} [J. Multivariate Anal. 22, 79-93 (1987; Zbl 0619.62079)] the regularity conditions in this paper are weaker and the convergence rates of the estimators are faster.