Abstract
In this paper, we apply the empirical likelihood approach to discriminant analysis
of non-Gaussian vector stationary processes. We propose a classification statistic based on the
empirical likelihood ratio function, and develop the discriminant procedure without assuming
that the true spectral density matrix is known. Even if the true structure of the process is
unknown, it is shown that the empirical likelihood classification criterion is consistent in the
sense that the misclassification probabilities converge to 0 as sample size tends to infinity. A
noteworthy point of the procedure is that the asymptotics of the empirical likelihood discriminant
statistic for scalar processes are always independent of non-Gaussianity of the process
under contiguous conditions.
