This paper compares the forecasting performance of Markov-switching GARCH (MSGARCH) models and standard GARCH models. MSGARCH models are proposed by Gray (1996), Klaassen (2002) and Haas et al. (2004), and recently there have been many empirical applications of these models in financial time series. This paper analyzes daily TOPIX returns and uses realized volatility as a proxy of true volatility for comparison of models. The results show that MSGARCH models can't improve forecasting performance as a whole, but that the low volatility is captured by these extended models better than standard GARCH models.
