分位数回归的金融风险度量理论及实证

The Theory and Empirical Research of Quantile Regression in Financial Risk Measurement

金融风险度量指标VaR是风险管理的重要内容。本文研究了不同的分位数回归模型在估计该指标时的表现,并选取标准普尔500指数、日经225指数、上证综指、深证成指进行了实证检验。在递归形式的分位数回归CAViaR模型中,本文的研究发现,SAV-CAViaR模型被拒绝的概率最高。而后验测试表明,IGARCH-CAViaR模型更适合刻画相对成熟的美国和日本金融市场风险的演化过程,而分位数回归的GARCH模型(QGARCH)在国内金融市场有良好表现,尤其是用于估计市场指数收益的1%VaR。

VaR, as the most commonly used practical financial risk measure, is an important part in risk management. This paper investigates the performance of different quantile regression models in estimating VaR, using indexes S＆P 500, Nikkei 225, SSEC and SZSC. The paper estimates the parameters in recursive quantile regression model CAViaR, Among these models, SAV-CAViaR model is rejected by DQ test in high frequency. Meanwhile, the paper compares CAViaR with the quantile regression in GARCH model, and demonstrates the evolutionary pattern of market risk by backtesting. Finding of this paper are as following. The relatively mature financial markets such as American and Japan can be well depicted by IGARCH-CAViaR, while in Chinese stock market the QGARCH model shows good ability, especially for estimating the 1% quantile.