No label defined (Q4940269)

The authors consider the following problem. Let \(\xi^{n}=(\xi_1, \xi_2,\ldots,\xi_{n})\), \(n\geq 2\), be an observation of the autoregressive process \(\xi_{i}=\theta\xi_{i-1}+w_{i}\), \(i=1,2,\ldots\), where \(\xi_0=0\), \(\theta\in(-\infty,\infty)\) is an unknown parameter and \(w_1,w_2,\ldots\) are independent standard normal variables. Let \(P_{\theta}^{n}\) be a measure which gives a distribution of the observation \(\xi^{n}.\) The problem of testing two simple hypotheses \(H^{n}\) and \({\tilde H}^{n}\) under the observation \(\xi^{n}\) is considered. Here \(H^{n}\) and \({\tilde H}^{n}\) are distributions of the observation \(\xi^{n}\) defined by the measures \(P_{\theta}^{n}\) and \(P_{\tilde\theta}^{n}\), respectively, as \(\theta\not={\tilde\theta}.\) The logarithm of the density of the measure \(P_{\tilde\theta}^{n}\) with respect to the measure \(P_{\theta}^{n}\) has the form \[ \Lambda_{n}=({\tilde\theta}-\theta)\sum\limits_{i=1}^{n} \xi_{i-1}w_{i}-2^{-1}({\tilde\theta}-\theta)^2 \sum\limits_{i=1}^{n}\xi_{i-1}^2. \] The authors prove the large deviation theorems for \(\Lambda_{n}\) as \(n\to\infty\) both under the hypothesis \(H^{n}\) and under the hypothesis \({\tilde H}^{n}.\) On the basis of these theorems the authors establish the relation between exponents of the rates of decrease for the error probabilities \(\alpha_{n}\) and \(\beta_{n}\) of the Neyman-Pearson test \(\delta_{n}\) as \(n\to\infty.\) The investigation is based on the properties of the Hellinger integral \(H_{n}(\varepsilon)\) of order \(\varepsilon\) for the measures \(P_{\theta}^{n}\) and \(P_{\tilde\theta}^{n}\), where \(H_{n}(\varepsilon)= H(\varepsilon,P_{\tilde\theta}^{n},P_{\theta}^{n})= E_{\theta}^{n}\exp(\varepsilon\Lambda_{n}).\)