基于峰度法的POT模型对沪深股市极端风险的度量  

POT model based on kurtosis and its empirical study on extreme risk of Chinese stock markets

作　　者：花拥军[1,2] 张宗益[2] 

机构地区：[1]重庆大学贸易与行政学院,重庆400030 [2]重庆大学经济与工商管理学院,重庆400030

出　　处：《系统工程理论与实践》2010年第5期786-796,共11页Systems Engineering-Theory & Practice

基　　金：国家杰出青年基金(70525005)

摘　　要：基于VaR正态性假设导致的尾部风险低估问题,研究极值POT模型,并针对样本平均函数法在某些数据结构下失效的缺陷,利用峰度法定量选取了阈值.沪深股市极端风险实证表明:涨跌停板影响了POT模型的有效性.涨跌停板前,在较高与较低的置信水平下,POT模型均比VaR模型有效;涨跌停板后,POT模型在较高置信水平下优于VaR模型,但在较低置信水平下反而不及VaR模型.研究认为这主要是因为涨跌停板抑制了极值数据的异质性,造成极值密集分布在涨跌停板附近,致使厚尾分界线向内收敛,从而影响了POT模型的有效性.To revise the shortage of VaR model which assumes sequence is normal distribution,this paper employs POT model based on kurtosis method to estimate extreme risk,which amend the limitation of sample mean excess function method.The empirical experiment on Chinese stock markets show that the raising limit influence the result of POT model,POT model is effective to study sequences with fat tail under lower or higher confidence level before raising limit operated,and POT model is more effective under higher confidence level but less effective than VaR model under lower confidence level within raising limit. Because raising limit restrains the heterogeneity of extreme values,result in lots of extreme value gathering near the raising limit,and heavy boundary converges inwardly,which influence the effective of POT model.

关 键 词：阈值模型 广义帕累托分布 在险价值 期望损失 

分 类 号：F830.91[经济管理—金融学]