Testing and estimating in the change-point problem of the spectral function (Q1324835)

The change-point problem in the spectrum of a stationary sequence can be divided into the following parts: 1) testing of the zero hypothesis \(H_ 0\) that \(X(1), \dots, X(N)\) is a sample from a stationary sequence with zero mean against the alternative hypothesis \(H_ 1\) that \[ X(t)=X_ 1(t) \text{ for } 1 \leq t \leq [N \theta], \quad \text{and} \quad =X_ 2(t) \text{ for } [N \theta]<t \leq N, \tag{1} \] where \(X_ i(t)\), \(t \in \mathbb{Z}\) \((i=1,2)\), are stationary sequences having spectral functions \(F_ i (\cdot)\), \(i=1,2\) \((F_ 1 \neq F_ 2)\), respectively; \(0<\theta<1\) is an unknown parameter (the change-point); 2) estimation of \(\theta\) when a decision about the existence of the change-point \(\theta \in (0,1)\) is made; 3) estimation of the magnitude of jump \(\rho_ F=\sup_{\lambda \in [0,\pi]} | F_ 1(\lambda)-F_ 2 (\lambda) |\) in the spectrum. The mostly investigated models of observations are autoregressive-moving average (ARMA) sequences of finite order, Gaussian sequences, and mixing sequences. We discuss the case of linear (or moving-average) observations \(X(1), \dots, X(N)\), i.e., when in (1): \[ X_ i(t)= \sum_{s \in \mathbb{Z}} a_ i (t-s) \xi_ s, \quad t \in \mathbb{Z}\;(i=1,2), \] where \(a_ i(t)\), \(t \in \mathbb{Z}\), are real weights, \(\sum_ t a^ 2_ i(t)<\infty\) \((i=1,2)\), and \(\xi_ s\), \(s \in \mathbb{Z}\), is an iid sequence with mean 0, variance 1, and all finite moments.