The asymptotic distribution of the likelihood ratio for autoregressive time series with a regression trend (Q1057607)

Let \(\{x_ i\}\) be given \(q\times 1\) vectors and let \(\eta_ n=\epsilon_ n-\theta_ 1\eta_{n-1}-...-\theta_ p\eta_{n-p}\) be a stationary autoregressive process, where \(\{\epsilon_ n\}\) is a white noise. The author investigates the process \(Y_ n=x^ T_ n\beta +\eta_ n\), where \(\beta =(\beta_ 1,...,\beta_ q)^ T\) is a vector of parameters. It is proved that the asymptotic form of the likelihood ratio is locally asymptotic normal under some mild assumptions. The general results are applied to the power of a test for positive dependence based on the Durbin-Watson statistic under local first order autoregressive alternatives.