Abstract: How to effectively obtain more valuable information from trading data is of vital importance to the financial risk management. However, in the existing studies, the risk measurement method based on low-frequency volatility model has almost reached the limit, but the predictive effect is not robust and the research on high-frequency volatility model is relatively scarce. The question is whether the high-frequency models can extract more valuable information from high-frequency data, which is useful for risk management. To carry out the rolling forecast of out of sample dynamic VaR based on the SSE Composite Index, we establish 12 low-frequency and 9 high-frequency volatility models. The empirical results show that high-frequency volatility models are more robust than low-frequency models and high-frequency models are superior to low-frequency in the most cases. What's more, the risk prediction effects of long and short position are extremely different. As for the long position, high frequency models do well under high risk situations but do poorly in the case of low risk situations, while the short position performs well in all cases.

Key words: low-frequency volatility models, high-frequency volatility models, dynamic VaR in the Out-of-sample, rolling forecasting

摘要： 如何充分挖掘交易数据中有价值的信息对金融风险管理极其重要,现有研究中基于低频波动模型的风险测度方法几乎已经做到了极致,而能达到的预测效果却并不稳健,对高频波动模型的研究相对比较匮乏。那么高频模型能否从高频数据中挖掘出更有价值的信息以便用于风险管理之中呢？本研究通过建立12个低频和9个高频波动模型对上证综指进行样本外动态VaR的滚动预测发现,高频模型相对于低频模型具有更好的稳定性,并且在多数情况下高频模型优于低频模型;多头与空头的风险预测效果具有显著差异,多头风险在高风险情况下高频模型表现出色,低风险情况下并不理想,空头风险则在所有情况下都表现较好。

关键词: 低频波动模型, 高频波动模型, 样本外动态VaR, 滚动预测