A Structural Time Series Model with Markov Switching
Roland G. Shami and Catherine S. Forbes
We propose an innovations form of the structural model underlying exponential
smoothing that is further augmented by a latent Markov switching process.
A particular case of the new model is the local level model with a switching
drift, where the switching component describes the change between high and
low growth rate periods. This new model is used to analyse the US business
cycle using US Quarterly real GNP data. Model parameters are estimated using
a Gibbs sampling algorithm and subsequently used for forecasting purposes.
In addition, the stability of the new model is tested against Hamiltonís model
over a range of observation periods.
Keywords: Structural model; Markov switching regime; Gibbs sampling;