On functional limit theorems for multivariate linear processes with applications to sequential estimation (Q1969133)

Let \(X_t\) \((t= 0,\pm 1, \pm 2,\dots)\) be an \(m\)-dimensional linear process of the form \({\mathbf X}_t- \mu= \displaystyle\sum^\infty_{u=0} A_u {\mathbf Z}_{t-u}\), defined on a probability space \((\Omega,{\mathcal F},P)\), where \(\{({\mathbf Z}_t,{\mathcal F}_t)\}\), \(({\mathcal F}_t= \sigma({\mathbf Z}_u: u\leq t))\) is an \(m\)-dimensional martingale difference sequence. Let \(W^m\) denote Wiener measure on \(C^m[0,1]\). Let \(T= (\sum^\infty_{j=0} A_j)\Sigma (\sum^\infty_{j=0} A_j)'\), where \(\Sigma\) is some positive-definite nonrandom matrix, and as an estimator of \(T\), consider \[ \widehat T_n= \widehat\Gamma_n(0)+ \sum^{h_n}_{h= 0} \widehat\Gamma_n(h),\quad \widehat\Gamma_n(h)= n^{-1} \sum^{n-h}_{t= 1} ({\mathbf X}_{t+ h}-\overline {\mathbf X}_n)({\mathbf X}_t- \overline {\mathbf X}_n), \] where \(\{h_n\}\) is some sequence of positive integers and define the stochastic process \(\xi_n\) by \[ \xi_n(u)= n^{-1/2}T^{-1/2}\Biggl[\Biggl( \sum^r_{t=1} {\mathbf X}_t- r\mu\Biggr)+ (nu- r)({\mathbf X}_{r+ 1}- \mu)\Biggr] \] \[ (r\leq nu< r+ 1;\;r=0,1,\dots, n-1). \] Further, let \(\{N_n, n\geq 1\}\) be a sequence of positive integer-valued random variables on the probability space such that as \(n\to\infty\), \(N_n/n@> P>>N\) with \(P(0< N<\infty)= 1\). Under some conditions, the authors prove that \(\xi_n\Rightarrow W^m\) (mixing in the sense of Rényi) and \(\xi_{N_n}\Rightarrow W^m\), and the same conclusions hold if \(T\) is replaced by \(\widehat T_n\). They also show that the results obtained may be used to study the asymptotic properties of some sequential procedures.