University of Washington and Korea University - Department of Economics
In some recent Bayesian approaches to structural break models, hierarchical priors are employed to allow for dependence of parameters across different regimes (Koop and Potter (2008), Pesaran et. al. (2006), Giordani and Kohn (2008), etc). In this paper, we extend these approaches to deal with structural breaks in the regime-specific parameters of Markov-switching models. For example, within a two-state Markov-switching model of the business cycle characterized by boom and recession regimes, we consider the possibility that different episodes of booms (or recessions) may be characterized by different mean growth rates of real GDP. The model consists of three features: i) specification of the Markov-switching latent variable that determines the regime; ii) specification of the evolving regime-specific parameters; and iii) specification of the time series within each regime. Once the model is cast into the state-space representation, a Markov Chain Monte Carlo procedure can be easily developed based on those proposed by Carter and Kohn (1993) and Albert and Chib (1993). However, one potential difficulty is that, conditional on a specific regime, the parameters for the other alternative regimes are not defined. We propose to resolve this difficulty by employing pseudo priors that are implied by the hierarchical priors for the regime-specific parameters.
Joint work with Yunjong Eo, University of Sydney