Forecasting downside risk in China's stock market based on high-frequency data

In order to forecast the downside risk of China's stock, we use high-frequency data to calculate downside realized semivariance, and use the downside realized semivariance to measure downside risk. Then, according to the "heterogeneous market hypothesis", we develop the HAR-DR, HAR-DR-J, HAR-DR-SJ, LHAR-DR, LHAR-DR-J and LHAR-DR-SJ models. Finally, we use the above six models to predict downside risk in the Chinese stock market. The results indicate that downside risk in Chinese stock market has long memory and leverage effect. And the downside risk, signed jump and leverage can be used to in-sample predict the future downside risk, while the discontinuous jump variation is poor at its prediction accuracy. Besides, the HAR-DR model shows better out-of-sample performance than the other models on forecasting downside risk. The discontinuous jump variation, signed jump and leverage do not contain out-of-sample information for forecasting the downside risk of China's stock. (C) 2018 Elsevier B.V. All rights reserved.