Strong invariance principles for sequential Bahadur-Kiefer and Vervaat error processes of long-range dependent sequences (Q2497191): Difference between revisions

The authors study the asymptotic behavior of the Bahadur-Kiefer, the Vervaat and the Vervaat error process in the case when the underlying observations are long-range dependent. Let \(X_n= G(\eta_n)\) be a subordinate process, where \(G\) is some measurable real-valued function and \((\eta_n)_{n\geq 1}\) a long-range dependent Gaussian process with mean zero and covariance function \(E(\eta_1\eta_{k+1})= k^{-D} L(k)\), \(0< D< 1\), and \(L\) some slowly varying function. The authors assume that the marginal distribution function \(F(x)= P\) \((X_1\leq x)\) is continuous and define by \(U_n= F(X_n)\) a new subordinate process with marginals uniformly distributed on \([0,1]\). Introduce the sequential uniform empirical distribution function \(\widehat{E_{[nt]}}(y)= {1\over[nt]} \sum^{[nt]}_{i=1} I(U_i\leq y)\) and the uniform empirical quantile function \(\widehat{U_{[nt]}}(y)= \widehat{E_{[nt]}}^{-1}(y)\), and the corresponding uniform empirical process and quantile process \(\alpha_n(y,t)= d^{-1}_n[nt](\widehat{E_{[nt]}}(y)- y)\) and \(u_n(y,t)= d^{-1}_n(y- \widehat{U_{[nt]}}(y))\), where \(d_n\) is some normalizing sequence. The uniform Bahadur-Kiefer process, the uniform Vervaat process and the uniform Vervaat error process are then defined by \[ \begin{aligned} R^*_n(y, t)&= d_n(\alpha_n(y, t)- u_n(y, t)),\\ V_n(y, t)&= d^{-2}_n[nt] \int^y_0 R^*_n(s,t)\,ds,\\ Q_n(y,t)&= V_n(y, t)- \alpha^2_n(y, t). \end{aligned} \] The main results of the present paper are almost sure invariance principles for each of these processes, as well as for the general, i.e. non-uniform, counterparts. The limit process in each of the cases has a representation as a product \(J(t)Z(y)\), where \(J(t)\) is a deterministic function and \(Z(y)\) is some Hermite process. This type of processes was first encountered by \textit{H. Dehling} and \textit{M. S. Taqqu} [Ann. Stat. 17, No. 4, 1767--1783 (1989; Zbl 0696.60032)] in their investigations of the empirical process of long-range dependent observations. A remarkable aspect of the present paper is the observation that the results about the Bahadur-Kiefer and the Vervaat error process are in sharp contrast with those obtained in the classical case of i.i.d. observations. For i.i.d. observations neither the Bahadur-Kiefer process nor the Vervaat error process converge, nor can they be renormalized in such a way that they converge to a non-trivial limit.