评估Value-at-Risk模型——条件矩检验方法

提出一种基于条件矩检验的Value-at-Risk （VaR）模型评估方法. 其基本思想是如果VaR模型无设定误差，则观察到的“突破”事件是鞅过程.因此可以通过检验“突破”事件是否为鞅对VaR模型进行设定检验. 采用条件矩检验方法对风险管理行业普遍使用的八种VaR 模型进行回测检验，作者发现条件历史模拟法表现最好，通过了各种检验，而无条件正态分布VaR模型表现最差，没有通过任何一种检验. 在风险管理实务中，作者推荐使用条件历史模拟法度量和管理金融风险.

Evaluating Value-at-Risk models——A conditional moment approach

The author proposes a conditional moment test to evaluate Value-at-Risk (VaR) models. The key idea behind the development of the test statistic is rooted in the "violations" being martingale process when the underlying VaR model is specified correctly. Specification test of VaR model is equivalent to test whether the corresponding "violations" being martingale. Through evaluating eight popular VaR models in financial markets by conditional moment test, the author finds that filtered historical simulation model performs best and normal-VaR model performs worst. In practice, filtered historical simulation method is appropriate to measure and manage financial risk.