| Summary: | 碩士 === 國立嘉義大學 === 管理研究所 === 95 === The empirical results in exchange rate changes are exhibit leptokurtosis, skewness and volatility clustering. In traditional GARCH models, which always imply spurious high GARCH persistence and that is a distinct error in the estimation. Markov Switching GARCH (MS-GARCH) model which allowed the mean and volatility in different level is used to solve the nonlinear problem. In many literatures, this method is applied with assuming the Gaussian and non-central t distribution in error term. In this paper,we introduce a MS-GARCH model with more flexible parametric error distribution based on the exponential generalized beta two (EGB2) distribution and make the shape parameters change with time. The results in this paper, we find the evidence that the estimation of the high-order moments with EGB2 distribution fits the real value much better, and the time-varying shape parameters are not constant significant in each regime. In particular, the MS ARCD-EGB$_2$ model provides better performance in forecasting performance relative other traditional models.
|
|---|
