A new hidden Markov-switching volatility model

Abstract

The thesis proposes and applies a two-state hidden Markov-switching model for financial time series featured with periodic structure breaks in volatility. The expected return, volatility and state transition probability are determined by three link functions respectively, whose coefficients are further governed by the hidden state. The proposed model particularly emphasizes on the parallel structure of the two states. The parallel structure separates the INTER-state and INTRA-state dynamics, enhances greater transparency, balances the memory of both recent and distant history, provides more consistent economic implication, and greatly simplifies and stabilizes the EM algorithm. We further discuss its estimation, inference, standard errors of the parameter estimate, forecasting, model selection and implementation, especially our innovations in those issues. The Monte Carlo experiments suggest that the proposed estimation method is accurate and reliable, the choice of the initial state probability has little effect on proposed model, and the information matrix calculated numerically is stable and reliable.