Hypothesis testing for nearly nonstationary AR(1) model with Gaussian autoregressive innovation (Q2638701)

Let \(x_ n\) be generated by the following first-order (generalized) autoregressive (AR) equation with AR residuals: \[ x_ n+ax_{n- 1}=\epsilon_ n,\quad | a| \leq 1, \] \[ where\quad \epsilon_ n+c_ 1\epsilon_{n-1}+...+c_ p\epsilon_{n-p}=w_ n, \] and where \(w_ n\) is a Gaussian white noise. The goal of the paper is to construct and study the likelihood ratio statistic for testing the null hypothesis \(H_ 0:\) \(a=-1\) against the alternative hypothesis \(H_ A:\) \(a=- 1+\lambda /N\) (\(\lambda\neq 0)\). (Here N denotes the number of available samples). In other words, the purpose is to test whether \(\{x_ n\}\) is a (generalized) random walk.