Simulation estimation of dynamic switching regression and dynamic disequilibrium models — some Monte Carlo results

Abstract

This article considers the estimation of dynamic exogenous switching regression models and dynamic endogenous switching models. With autocorrelation in disturbances or latent lagged-dependent variables, likelihood functions of such models involve high-dimensional integrals and a huge number of summations over unobserved regime paths. Simulated likelihood methods and simulated methods of moments are available. These approaches simulate both continuous and discrete latent-dependent variables. By Monte Carlo experiments, it is found that the performances of various approaches depend crucially on how discrete state variables are simulated. The valuable approach is to simulate regime paths with regime probabilities based on the current and past sample information.