基于广义帕累托分布稳健估计法的沪市VaR预测

VaR Forecast in Shanghai Stock Market Based on Robust Estimation of GPD

针对金融收益序列的"高峰、厚尾"特征,本文将ARMA-GARCH模型和POT模型结合起来度量上证综指的VaR,用广义帕累托分布(GPD)对POT模型的超额阈值进行拟合得到VaR。考虑到GPD参数的极大似然估计非稳健性,本文使用了GPD参数的三种稳健估计法:最小密度功效散度、中位数和似然矩估计。动态回溯检验结果表明,使用稳健方法拟合GPD,可以得到更为稳健、精准的VaR度量,并得到GPD稳健估计优劣性的比较结果。

For stylized characteristics such as leptokurtic and fat tail for financial return series, this paper u- ses ARMA - GARCH to model conditional return and volatility, and then uses POT in extreme value theory to mod- el standard return and fit excess by GPD to get VaR. Because of non - robustness of GPD parameters of MLE, The paper uses three methods of robust estimation ： minimum density power divergence, median and likelihood moment to fit GPD. The dynamic back inspection shows that using robust methods to estimate GPD parameters can get more robust and more precise VaR estimation.