Uniform strong estimation under \(\alpha\)-mixing, with rates (Q1199868): Difference between revisions

The authors consider a strictly stationary \(\alpha\)-mixing sequence \(\{X_ n\}\), \(n\geq 1\), of random variables with a stationary distribution function \(F(x)\) and a mixing coefficient \(\alpha(n)\) satisfying the summability condition \[ \sum^ \infty_{n=1}n^{- 1}(\log n)(\log\log n)^{1+\delta}\alpha(n)<\infty\quad\text{for some } \delta>0. \] The function \(F\) is estimated by the empirical distribution function \(F_ n\). Almost sure convergence of \(F_ n\) to \(F\) uniformly in \(x\in R^ 1\) has been shown. The theorem is a consequence of two laws of large numbers under \(\alpha\)-mixing which are also stated and are of independent interest. By adopting a Kiefer process approximation technique [s. \textit{I. Berkes} and \textit{W. Philipp}, Z. Wahrscheinlichkeitstheorie Verw. Gebiete 41, 115-137 (1977; Zbl 0349.60026)], the law of the iterated logarithm for \(\sup_{x\in R}| F_ n(x)-F(x)|\) is also obtained. For the case when a smoothed version \(\hat F_ n\) instead of \(F_ n\) is used, a similar result is derived. The latter is then used to obtain rates of convergence of a usual kernel estimate \(f_ n\) of the Lipschitz continuous density \(f\) of the \(X_ i\)'s. These results are applied in studying uniform almost sure convergence and rates of convergence in hazard rate estimation and in estimating higher order derivatives of the density \(f\). The rates obtained are the best for the problems considered under \(\alpha\)-mixing.