Asymptotic expansion for the distribution of the discriminant function in the first order autoregressive processes (Q1820533)

This paper is concerned with the problem of classifying a series of T observations coming from one of two first order autoregressive Gaussian processes. The processes are determined by the stochastic equations \(y_ t=\alpha_ jy_{t-1}+u_ t\) \((t=...-1,0,1,...)\) where \(\alpha_ j\) \((| \alpha_ j| <1)\) is known, and \(u_ t's\) are independent identically distributed as \(N(0,\sigma^ 2_ j)\) with known variance \(\sigma^ 2_ j\) \((j=1,2)\). We consider the discriminant function based on the Bayes method, which is a quadratic function of observations. We derive an asymptotic expansion of the distributions of the discriminant function and give an approximation for the probabilities of misclassifications.