基于QRNN-GARCH-POT模型沪深指数收益率风险度量的研究

Research on Risk Measurement of Shanghai and Shenzhen Index Returns Based on QRNN-GARCH-POT Model

本文以沪市和深市综合指数收益率为研究对象.首先,考虑到收益率序列存在波动聚集性、非线性以及非对称等特征,因此本文一方面利用了GARCH模型在准确刻画波动聚集性方面的优势,另一方面引入神经网络分位数回归(QRNN)模型来解决非线性和非对称方面的问题,建立了QRNN-GARCH模型对收益率风险特征进行度量.但QRNN-GARCH模型对收益率序列极端尾部数据处理能力不足,为此用极值理论的POT方法对此模型进行改进,构建了QRNN-GARCH-POT模型,将其应用于极端VaR风险测度.此外利用失败率、似然比检验与相对误差率,对比QRNN-GARCH-POT模型与其他模型在VaR风险测度中的表现.结果表明:第一,基于QRNN-GARCH模型的VaR风险测度取得了比GARCH模型更好的效果,仅是极端(99%)VaR风险测度精度不高.第二,QRNN-GARCH-POT模型改善了极端VaR风险测度效果.

The Shanghai and Shenzhen comprehensive index returns are taken as the research object.Firstly,with volatility clustering,nonlinear and asymmetric characteristics existing in return sequence,this article uses the GARCH model to describe volatility clustering and introduces quantile regression neural network(QRNN)model to solve the nonlinear and asymmetric problems.The QRNN-GARCH model is set up to measure characteristics of return risk.But in view of the insufficient processing capacity of the QRNN-GARCH model for the extreme tail data of return rate series,the POT method of the extreme value theory is used to improve the model,and the QRNN-GARCH-POT model is set up and applied to measure the extreme VaR risk.In addition,the failure rate,likelihood ratio test and relative error rate are used to compare the performance of QRNN-GARCH-POT model and other models in VaR risk measurement.The results show that the VaR risk measure based on the QRNN-GARCH model achieves a better effect than the GARCH model,the extreme(99%)VaR risk measure has only low precision.The QRNN-GARCH-POT model also improves the effect of extreme VaR risk measure.