MARKOV REGIME-SWITCHING MODELS

Abstract

A regime-switching model is a time-series model in which parameters change values according to the regime at present time. While regime-switching models have been very popular in applied work, there is a lack of literature for simulation studies. New methods based on regime-switching models are often proposed with neither a proof of convergence nor simulations to demonstrate their basic properties. In this thesis, a detailed simulation study of regime-switching models is conducted. A strategy to generate initial search values in the parameter estimation of regime-switching models is proposed. It is shown that this method can dramatically reduce the number of restarts of the optimizer. Even in 3-regime models (with 15 unknown parameters), parameters can be estimated reasonably well with only 5 restarts.