REALIZED GARCH: EVIDENCE IN CSI 300 DURING A HIGH-VOLATILITY PERIOD

Numerous studies have suggested the application of GARCH and its extensions to model volatility of stock prices and indices. However, the performance of these models is not well established during the period of unusually high volatility. In this paper, we compare three GARCH specifications namely, standard GARCH, EGARCH, and Realized GARCH, in their ability to model volatility during the recent Chinese stock market debacle. In addition, three models are applied to the quantile forecast of Value-at-Risk (VaR). Normal distribution, student's t distribution as well as skewed student's t distribution are used. While all specifications perform in a similar fashion during normal periods, we document that Realized GARCH model with skewed student's t distribution outperfoms the others during the high-volatility period from January 2015 to October 2015.