For a class of Markov-switching models, the likelihood function and inferre
d state distributions for a given sample are shown to have closed-form repr
esentations under a set of sufficient conditions. Based on these results, i
t is demonstrated that the closed-form partial derivatives (when exist) of
the likelihood function can be readily found. These results may be used to
improve the efficiency of numerical optimization techniques used for estima
ting the Markov-switching models. (C) 2001 Elsevier Science B.V. All rights
reserved.