Empirical likelihood approach toward discriminant analysis for dynamics of stable processes

Discriminant analysis for time series models has been studied by many authors in these few decades, but many of them deal with second order stationary processes. In this paper, we introduce an empirical likelihood statistic based on a Whittle likelihood as a classification statistic, and consider problems of classifying an α-stable linear process into one of two categories described by pivotal quantities _θ1 and _θ2 of time series models. It is shown that misclassification probabilities by the empirical likelihood criterion converge to 0 asymptotically without assuming that the true model is known. We also evaluate misclassification probabilities when _θ2 is contiguous to _θ1, and carry out simulation studies to make a comparison between goodness of the empirical likelihood classification statistic and that of an existing method. We observed that the empirical likelihood ratio discriminant statistic performs better than the existing method in some cases even if a family of score functions does not contain the true model. Since the stable processes do not have the finite second moment, this extension is not straightforward, and contains a lot of innovative aspects.