Law of the iterated logarithm for perturbed empirical distribution functions evaluated at a random point for nonstationary random variables (Q1337953): Difference between revisions

Smooth nonparametric estimations of distribution functions are considered. Let \((X_ i)\) be a sequence of absolutely regular random variables with distribution functions \((F_ i)\) such that \((F_ i)\) converges to a smooth distribution function \(F\). Consider smooth estimations of \(F\): \[ F_ n(x) = n^{-1} \sum^ n_{i=1} G_ n (x- X_ i), \] where \[ G_ n(x) = \int^ x_{-\infty} g_ n (t)dt, \quad g_ n(t) = g(t/ \alpha_ n)/ \alpha_ n, \quad \alpha_ n \to \infty, \] as \(n \to \infty\) and \(g\) is a probability density function. Let \(h(x_ 1, \dots, x_ m)\), symmetric in its arguments, be a measurable kernel and let \(U_ n\) be the corresponding \(U\)-statistics. Then \(\widehat F_ n (U_ n)\) is useful in estimating \(\xi = F(\theta (F))\), where \(\theta (F) = \int_{\mathbb{R}^ m} h(x_ 1, \dots, x_ m) \prod^ m_{i=1} dF(x_ i)\). The almost sure representation and the law of iterated logarithm for \(\widehat F_ n (U_ n)\) when the random variables are nonstationary and absolutely regular are derived.