跳跃行为、杠杆效应及长期记忆下的VaR预测

摘要

基于高频数据构建了一个简单的收益率模型用于预测风险价值。该模型使用Corsi&Reno(2010)的带杠杆效应的异质自回归模型刻画已实现方差，考虑了跳跃行为、杠杆效应和长期记忆对条件波动率的影响；刻画了收益与风险的权衡关系；而且结合极值理论，无须假设具体收益率分布，便可得到风险价值的预测值。使用该方法预测了沪深股票市场的市场风险，而且基于Christoffersen(1998)方法检验表明，收益率模型的风险价值预测是有效率的。%We construct a simple return model for forecasting Value at Risk based on high frequency data. We use the LHAR-CJ model of Corsi&Reno (2010) to model the realized variance, taking jump behavior, leverage effect and long memory into account. We capture the relationship between return and risk, and use Extreme Value Theory to forecast VaR without any specific assumption about the distribution of return. Empiric researches on Chinese stock market risk show that, the VaR forecasting using our method is efficient.