基于支持向量分位数回归多期VaR测度

Evaluating multiperiod VaR via support vector quantile regression

为实现VaR风险准确测度,考虑到波动聚集、厚尾与非对称等金融市场典型特征,基于支持向量分位数回归模型,研究了VaR及其影响因素之间的线性及非线性依赖关系,给出了多期VaR风险测度方法,并将其与传统VaR风险测度方法进行了比较.选取上证指数、香港恒生指数和标准普尔500指数进行实证研究,VaR回测检验结果表明基于支持向量分位数回归模型的多期VaR风险测度在样本内与样本外都有良好的表现.

Based on the support vector quantile regression（SVQR） model, this article discussed the linear or nonlinear relationship between VaR and its predictors, and proposed a new method ofmultiperiod VaR measure. The new method can give an accurate evaluation of VaR by considering the financial stylized facts, including volatility clustering, heavy tail, and asymmetry. For the empirical illustration, Three securities price indices： SSEC, HSI, and S＆P500 from financial markets are selected. The VaR backtest results show that the new method is superior to the traditional VaR measure both in sample and out of sample.