基于分位数回归森林+POT的极端VaR风险测度

Research on Extreme VaR Measurement based on QRF + POT

非参数形式的分位数回归方法在测度VaR风险方面已经取得了长足的进展,但是在测度极端VaR风险方面仍存在精度不高的问题。因此,文章结合分位数回归森林(QRF)和极值理论方法(POT)的优势,提出了QRF+POT方法来测度极端VaR风险。一方面,采用分位数回归森林来测度VaR风险的非线性结构,另一方面,使用POT方法来处理极端尾部数据。以上证综指等四支股票指数为研究对象,比较分析了QRF+POT方法与其他方法测度极端VaR风险的效果,结果表明:第一,QRF方法能够精确测度正常分位点的VaR风险,但难以精确测度极端VaR风险;第二,QRF+POT方法能够有效刻画股市暴跌期间的极端风险,获得极端VaR风险的精确测度。

The non-parametric quantile regression method has made great progress in measuring VaR, but there is still a problem of low accuracy in measuring extreme VaR. Therefore, combining the advantages of Quantile Regression Forest(QRF) and Extreme Value Theory(POT),the QRF+POT method is proposed to measure extreme VaR. On the one hand, quantile regression forest is used to measure the non-linear structure of VaR. On the other hand, the POT method is used to process extreme tail data. We selected Shanghai Composite Index and other three stocks to test the performance of the QRF+POT method and other methods. The results show that: Firstly, the QRF method can accurately measure the VaR at the normal quantile, but it is difficult to accurately measure the extreme VaR;Secondly, the QRF+POT method can effectively describe the extreme risk during the stock market crash and obtain an accurate measurement of the extreme VaR.