Asymptotic behavior of the perturbed empirical distribution functions evaluated at a random point for absolutely regular sequences (Q1321983)

Let \(X_ 1, X_ 2,\dots\) be a strictly stationary sequence of real- valued random variables having marginal distribution function (d.f.) \(F\); it is assumed that the sequence is either absolutely regular or strong mixing. The d.f. \(F\) is estimated by a kernel-type estimate \(\widehat {F}_ n\), and the main result of the paper is that of establishing asymptotic normality of a normalized version of \(\widehat {F}_ n\), evaluated at a certain \(U\)-statistic. More specifically, let \(k\) be a probability density function, and let \(a_ n\) be a sequence of bandwidths. Define \(k_ n\) by \(k_ n(t) = a_ n^{-1} k(ta^{-1}_ n)\), for \(t\) in the real line \(\mathbb{R}\), and \(K_ n\) by \(K_ n(x) = \int^ x _{-\infty} k_ n(y) dy\), \(x \in \mathbb{R}\). On the basis of the segment \(X_ 1, \dots, X_ n\), estimate \(F_ n(x)\) by \[ \widehat {F}_ n(x) = n^{-1} \sum^ n_{i = 1} K_ n(x - X_ i). \] Next, let \(g\) be a real-valued Borel measurable function defined on \(\mathbb{R}^ m\) and being symmetric in its arguments, and let \(\xi = Eg(X_ 1, \dots, X_ m)\). Define \(\mu_ n\) by \(\mu_ n = EK_ n(\xi - X_ 1)\), and \(A_ j\) by \[ A_ j = u(\xi - X_ j) - F(\xi) + mF' (\xi) [g_ 1(X_ j) - \xi)], \] where \(u(t) = 1\) for \(t \geq 0\) and \(u(t) = 0\) otherwise; \(g_ 1\) is the projection of \(g\) on \(\mathbb{R}\). Set \(\sigma^ 2 = E A^ 2_ 1 + 2 \sum^ \infty_{i = 1} E(A_ 1 A_{i + 1})\), and suppose that \(\sigma^ 2 > 0\); finiteness of \(\sigma^ 2\) is implied by the assumed conditions. Finally, let \(U_ n\) be the \(U\)- statistic defined by \(U_ n = \left( \begin{smallmatrix} n\\m \end{smallmatrix} \right)^{-1} \sum_{c_{n,m}} g(X_{i_ 1}, \dots, X_{i_ m})\), where \(c_{n,m}\) denotes the set of all \(\left( \begin{smallmatrix} n\\ m \end{smallmatrix} \right)\) combinations of \(m\) distinct elements \(\{i_ 1, \dots, i_ m\}\) from \(\{1, \dots, n\}\). Then, under suitable regularity conditions, it is shown that \(n^{1/2} [\widehat {F}_ n (U_ n) - \mu_ n]\) converges in distribution, as \(n\) tends to \(\infty\), to the normal distribution \(N(0, \sigma^ 2)\).