The empirical process for bivariate sequences with long memory (Q2573222)

Let \((X_t, Y_t)\), \(t=1,2, \ldots\), be a bivariate, strictly stationary process with marginal distribution function \(F(x, y) = P (X_1 \leq x, Y_1 \leq y)\). Given the set of observations \((X_1, Y_1), \ldots, (X_n, Y_n)\), let \[ F_n (x,y) = n^{-1} \sum_{t=1}^n {\mathbf 1} (X_t \leq x, Y_t \leq y) \] be the empirical marginal distribution function. Consider the random processes \[ H(r; p, q) = \int_{R^p}' \int_{R^q}' g(r; \xi_1, \ldots, \xi_q) \prod_{j=1}^p | \xi_j| ^{(\alpha_\varepsilon -1)/2} \prod_{k=1}^q | \xi_k| ^{(\alpha_\eta -1)/2} \times\prod_{j=1}^p W_1 (d \xi_j) \prod_{j=1}^p W_2 (d \xi_k), \] where \(W_1(\cdot)\), \(W_2(\cdot)\) are independent copies of a Gaussian white noise measure on \(R^1\), the integrals \(\int'\) exclude the hyperdiagonals \(\xi_j = \pm \xi_{j'}\), \(j, j' = 1, 2, \ldots, p\), \(j \neq j'\); \(\xi_k = \pm \xi_{k'}\), \(k, k' = 1, 2,\ldots, q\), \(k \neq k'\), and \[ g(r; \xi_1,\ldots, \xi_q) = C(p,q; \alpha_\varepsilon, \alpha_\eta)^{-1/2}[(\exp (i r (\xi_1 + \cdots + \zeta_q)) -1)/i(\xi_1 + \cdots + \zeta_q)]. \] The following Proposition is the main result of this paper. Proposition: Under some conditions, as \(n \to \infty\) \[ (n/d_n(p_1^*, q_1^*)) \{F_n(x,y) -F (x,y)\} \Rightarrow \sum_{i=1}^h\widetilde k_j (J_{p_i^* q_i^*}(x, y)/p_i^*! q_i^*!) H(1;p_i^* q_i^*), \] where \(\Rightarrow \) denotes weak convergence in the Skorohod space \(D[-\infty,\infty]^2\), \({\widetilde k_j}, p_i^*, q_i^*\) are some constants, \(d_n(p_1^*, q_1^*)\) and \(J_{p_i^* q_i^*} (x, y)\) are known nonrandom sequences.