中文摘要：

本文使用高频数据研究了沪深300指数的尾部风险。基于Skewed-t分布的Realized GARCH模型比基于t分布和正态分布的Realized GARCH模型拟合要好。这说明对于具有聚类波动,尖峰厚尾和偏斜特征的金融数据来说,同时考虑了偏斜和厚尾特征的Skewed-t分布的Realized GARCH模型是一个合适的模型。本文使用两种方法对在险值VaR进行预测,一种是基于样本外的多步预测,另一种是基于滚动窗口的一步预测。两种方法下,Realized GARCH模型基本上都要比GARCH和EGARCH模型预测表现要好,并且,厚尾分布比正态分布能更好地描述风险,尤其是尾部风险。基于Skewed-t分布的Realized GARCH模型,在各个置信水平下,都表现很好且稳定,这说明同时考虑了偏斜和厚尾特征的Skewed-t分布的Realized GARCH模型能很好地预测风险值。

英文摘要：

This article analyzes HS 300 Index tail risk with high frequency data. Realized GARCH model with the skewed student＇s t distribution performs better than that with the normal and standard t distributions. That suggests the realized GARCH model with the skewed student＇s t distribution which models the fat tail and skewness is a suitable model for financial data which have characteristics like volatility clustering, leptokurtosis and fat tail. and skewed. This article applies two methods to forecast value-at-risk. One is out-of-sample multistep forecast, and the other is one step forecast based on rolling windows. Main results from two methods show that realized GARCH model is better at forecast than GARCH and EGARCH model. Moreover, the model with fat tail distribution characterizes the risk, especially the tail risk, better than that with normal distribution. At three confidence levels, realized GARCH model with skewed student＇s t distribution performs well and stably. This result supports that realized GARCH model forecasts VaR very well.