Empirical likelihood estimation of the Markov-switching model

Abstract

The Markov-switching (MS) model is one of the most popular nonlinear time series models in the literature. However, the estimation methods which are normally used to estimate the MS models rely on the assumption of a parametric distribution, which sometimes is considered as a strong assumption. This study, therefore, tries to relax the assumption and develop a more flexible estimator for the MS models that is a maximum empirical likelihood estimation. According to this approach, the parametric likelihood will be replaced by the empirical likelihood function with relatively minor modifications to existing recursive filters. A performance of the suggested estimation method is then evaluated through a Monte Carlo experiment and a real application, the U.S. business cycle. Overall results of both empirical studies indicate that the empirical likelihood could outweigh the classical likelihood estimators.