Asymptotic properties of \(U\)-processes under long-range dependence (Q638797)

Let \((X_i)_{i\geq 1}\) be a stationary, mean-zero Gaussian process with covariances \[ E(X_1, X_{k+1})= k^{-D}L(k),\;k \geq 1, \] where \(D\) is in \((0,1),\) and \(L\) is slowly varying at infinity. This paper develops asymptotic results for the \(U\)-process \(\{U_n(r),\;r\in I\}\), where \(I\) is an interval included in \(R,\) \[ U_n(r) = {n \choose 2}^{-1} \sum_{1\leq i<j\leq n} I_{\{G(X_i, X_j)\} \leq r}, \] and \(G\) is a symmetric function. The asymptotic behaviour of \(U_n(\cdot)\) depends on the value of \(D\) and on the Hermite rank \(m\) of the class of functions \(\{I_{\{G(\cdot, \cdot) \leq r\}} - U(r),\;r \in I\}.\) The results are applied to obtain the asymptotic behaviour of several estimators based on \(U\)-processes constructed using Gaussian long-range dependent sequences. The examples include the Hodges-Lehmann estimator of location, the sample correlation integral and the Shamos scale estimator. The robust estimators of location and scale are shown to have the same asymptotic distributions as the classical estimators.