We present a general class of nonlinear time-series Markov regime-switching models for seasonal data which may exhibit periodic features in the hidden Markov process as well as in the laws of motion in each of the regimes. This class of models allows for non-trivial dependencies between seasonal, cyclical and long-term patterns in the data. To overcome the computational burden we adopt a Bayesian approach to estimation and inference. This paper contains two empirical examples as illustration, one uses housing starts data while the other employs US post-Second World War industrial production.
