Weak convergence of randomly weighted dependent residual empiricals with applications to autoregression (Q1327857)

For each integer \(n \geq 1\), let \((\eta_{ni},\;\xi_{ni},\;\gamma_{ni})\), \(1 \leq i \leq n\), be an array of random variables (r.v.'s) such that \(\{\eta_{ni},\;1 \leq i \leq n\}\) are independent and identically distributed r.v.'s with distribution function \(H\) and the r.v.'s \((\xi_{ni},\;\gamma_{n i})\), \(1 \leq i \leq n\), satisfy some independence and measurability conditions. For real \(x\) define \[ U_n (x) = n^{1/2} \bigl( V_n (x) - J_n (x) \bigr), \quad U^*_n (x) = n^{1/2} \bigl( V^*_n (x) - J^*_n (x) \bigr), \] where \[ V_n (x) = n^{-1} \sum^n_{i = 1} \gamma_{ni} I(\eta_{ni} \leq x + \xi_{ni}), \quad V_n^* (x) = n^{-1} \sum^n_{i = 1} \gamma_{ni} I(\eta_{ni} \leq x), \] \[ J_n (x) = n^{-1} \sum^n_{i = 1} \gamma_{ni} H(x + \xi_{ni}), \quad J^*_n (x) = n^{-1} \sum^n_{i = 1} \gamma_{ni} H(x). \] Let \(o_p (1)\) denote a sequence of r.v.'s that converges to zero in probability. The main result of the paper is the following Theorem: Assume that the following holds: \[ \left( n^{-1} \sum^n_{i = 1} \gamma^2_{ni} \right)^{1/2} = \gamma + o_p1, \quad \gamma \text{ a positive r.v.,} \] \[ n^{1 /2} \max_{1 \leq i \leq n} |\gamma_{ni} |= o_p(1), \quad\max_{1 \leq i \leq n} |\xi_{ni} |= o_p(1), \] \(H\) is continuous and there exists a \(b_0 > 0\) such that \({\mathcal I} (b_0) < \infty\), where \({\mathcal I}\) is the entropy integral. Then the processes \(\{U_n\}\) and \(\{U^*_n\}\) are eventually tight in the uniform metric and \(|U_n - U^*_n |_\infty = o_p (1)\). If, additionally, for each \(n \geq 1\), \(\{\gamma_{ni},\;1 \leq i \leq n\}\) is square integrable, then \(\{U_n\}\) and \(\{U^*_n\}\) converge weakly in distribution to a process \(\gamma B(H)\), where \(B\) is a Brownian bridge in \(C[0,1]\), independent of \(\gamma\). Applications of this theorem to randomly weighted residual empirical processes are given. The weights need not be independent, bounded or square integrable. The main result yields the asymptotic uniform linearity of a class of rank statistics in \(p\) th-order autoregression models. The authors obtain the asymptotic distributions of some robust and Jaeckel-type estimators and of certain minimum distance estimators of the autoregression parameter vector as well.