No label defined (Q4356547)

Let \(\{X_t; t\in \mathbb{Z}\}\) be a strictly stationary process with \(E(X_1) =0\) and \(E (X_1X_{1+ |v|}) =\gamma_v\), \(v\in \mathbb{Z}\), and let \(X_1, \dots, X_n\) be a sample from \(\{X_t\}\). The serial covariance \(\widehat \gamma_v\) of order \(v\) is given by \[ \widehat \gamma_v =n^{-1} \sum^{n- |v|}_{t=1} X_t X_{t+ |v|}. \] It is well known that under mild regularity conditions on \(\{X_t\}\), the distribution of \(n^{1/2} (\gamma_{v_j} -\gamma_{v_j})\) \((1\leq j\leq l)\), where \(l,v_1, \dots, v_l\) are fixed positive integers, is asymptotically (as \(n\to \infty)\) normal with mean zero and a finite covariance matrix. We shall discuss the asymptotic normality of serial covariances \(\widehat\gamma_v\) where \(v\to \infty\) but \(v/n \to 0\) as \(n\to\infty\) and state regularity conditions under which this property holds. For the entire collection see [Zbl 0846.00010].