A new estimator for the constrained regime-switching model

Abstract This article proposes a new method for estimating regime-switching models when some slope parameters are constrained to be non-switching. The constrained parameters are simply found by an appropriate weighted average of the unconstrained parameters. The advantage of this approach is twofold. First, the constrained estimates are obtained by an unconstrained estimation procedure such as the EM algorithm, and hence, the procedure is relatively straightforward to implement. Secondly, as both the constrained and unconstrained estimators are available, testing based on the likelihood ratio and model selection by means of likelihood-based information criteria is particularly simple. The procedure is applied to a three-state Markov-switching variance model.