Markov Switching Models

Abstract

This thesis displays a presentation of the Hamilton's Markov Switching model both in simple and State Space form. Moreover, the model is applied in the India's GDP and DJIA Index using R. This thesis is based on three chapters of Markov Switching models. First chapter covers the Classical approach, the parameters of which are estimated taking into consideration only the data sample and inferences are made conditional to that data. This presentation consists of two parts. The first part refers to the simple form of Markov Switching model which can be estimated by the EM-algorithm. The second part refers to the State Space form of Markov Switching model which can be estimated by Kalman Filter. The second chapter presents the Bayesian approach, according to which we treat the parameters as individual random variables with their own prior distributions which are determined by researcher beliefs or randomly by a Dirichlet process before the posterior distribution is determined taking into consideration the sample data. Similar to the first chapter both forms of the model are presented. In the Bayesian approach the parameters are estimated with Markov Chain Monte Carlo methods such as the Gibbs Sampling. In the last chapter a two-state Markov Switching model is applied in India's real GDP and the DJIA Index. The results of the implementation show that the Markov Switching model can fit well financial data and can detect the regimes with effectiveness.