Does Decomposing Realized Volatility Help in Risk Measure Prediction: Evidence from Chinese Mainland Stock Market

This paper examines jump dynamic patterns in three Chinese medical stocks. It also compares the Value-at-Risk (VaR) forecasting performance of a newly proposed realized volatility model allowing for jumps with that of two commonly used realized volatility models, which do not account for jumps. Using the Heterogeneous Autoregressive Realized Volatility model that allows for jumps (HAR-CJN), we nd that relative to stocks in the developed US market, jumps occur in this emerging market stock more frequently but with a smaller size. Meanwhile, jumps and in particular jump size are more predictable. Two-step VaR backtesting shows that compared with the Autoregressive Moving Average (ARMA) model and the Heterogeneous Autoregressive (HAR) model, the HAR-CJN model isn't always able to provide a better performance in downside risk prediction, but it does when stock return has a relatively large jump size. JEL Classi cation: C13, C32, C52, C53, G17, G32