VaR若干典型度量方法的比较研究

【作者】 刘丹；

【导师】 杨德权；

【作者基本信息】 大连理工大学 ， 系统工程， 2004， 硕士

【摘要】 目前，VaR的度量方法有很多，而且，由不同方法计算出的VaR值往往相差很大，使得风险管理者面对如此多的模型不知如何选择才能真正达到其风险管理的目的。针对这一点，本文给出了一种将理论分析与实证分析相结合的VaR模型准确性评价方法。本文比较检验了六种典型的风险度量方法，并得出了在不同置信水平下的模型的准确性排序。 VaR表示在一定时段内，在给定的置信水平下预期的潜在最大损失。参数度量方法主要是依赖于对波动率的估计。本文选取了四种参数估计的方法即EQMA，EWMA，GARCH，EGARCH模型，通过对其波动率的估计来估算VaR。另外，还选取了两种非参数方法HS和EVT理论中的POT方法，对这六种方法通过Nasdaq指数的日回报数据进行了比较检验。 本文所用的检验方法主要分两步进行。第一步是统计性检验，第二步是通过损失函数来评价模型的优劣。本文的研究结果表明：在95％置信水平下，不应该使用EGARCH方法，POT方法准确性最佳，接下来依次是：HS，EQMA，EWMA，GARCH；在99％置信水平下，不应该使用EWMA方法，POT方法准确性最佳，接下来依次是：HS，EQMA，EGARCH，GARCH。 对于样本数据很大的情况下，本文的研究结果推荐管理者使用极值理论方法进行VaR风险度量。更多还原

【Abstract】 Currently, we can use a lot of models to estimate value-at-risk, but different models often deduce different VaR values. Risk managers are therefore often left with the daunting task of choosing from a plethora of models. This paper brings about an accuracy evaluate method which combined theories analysis and empirical analysis together. And furthermore this paper compares and tests six typical methods and get out the models’ series in descending order based on superiority.The Value at risk (VaR) is the maximum expected loss over a given horizon period at a given level of confidence. A crucial factor for the accuracy of VaR models that are based on the parametric approach is the measure of volatility. In this paper, 4 parametric models are compared, they are equal weighted moving average model (EQMA); exponentially weighted moving average model (EWMA); GARCH model; exponential GARCH (EGARCH) model. In addition, historical simulation (HS) and extreme value theory based model (POT) are used and compared here. We compare the estimations in an application to daily returns on the Nasdaq index.The test processes are carried out in two steps: the lst step is to test the statistical validity of all the 6 methods and the 2nd step is to use the loss function evaluation to evaluate the performance and testing for superiority of the models. The empirical results of the study show that at 95% confidence level, POT is the best then followed by HS, EQMA, EWMA, GARCH, and EGARCH should be rejected; at 99% confidence level, POT is the best, then followed by HS, EQMA, EGARCH, GARCH, and EWMA should be rejected.The EVT-based method POT is recommended to estimate VaR when samples arelarge.更多还原