Discrete time periodically correlated Markov processes (Q2772043)

The authors characterize the covariance function and also the spectral density function of a real-valued periodically correlated Markov process \(\{X_n,\;n\in \mathbb{Z}\}\), \(Z\) stands for the set of integers. Periodically correlated (PC) processes are in general nonstationary. The authors consider the discrete time Markov processes [see \textit{J. L. Doob}, ``Stochastic processes'' (1953; Zbl 0053.26802) and others]. Let \(\{X_n,\;n\in\mathbb{Z}\}\) be a second-order process of centered random variables, i.e. \(EX_n=0\), \(|X_n|^2< \infty\), \(n\in\mathbb{Z}\). The process \(\{X_n,\;n\in\mathbb{Z}\}\) is said periodically correlated if there is a positive integer \(T\) for which the covariance function \(R(n,m)=EX_nX_m\) satisfies \(R(n,m)=R\) \((n+T,m+T)\) for all \(n,m\in \mathbb{Z}\). The smallest \(T\) is the period.NEWLINENEWLINENEWLINEIn Section 2 preliminaries on PC processes and Markov processes are presented. Section 3 is devoted to the structure of PCM, where a closed formula for the covariance function \(R(n,m)\) of PCM process of period \(T\) is presented (Theorems 3.1 and 3.2). The authors observe that the covariance can be specified only by the values \(R(i,j)\), \(R(i,j+1)\), \(j=0,1,\dots,T-1\). The spectral density of a PCM process is presented in Section 4 and its general form is derived (Theorem 4.1). For other details see the authors references.