基于g-h分布的操作风险损失强度分布拟合及风险度量  

Loss Severity Distribution Fitting and Risk Measurement of Operational Risk Based on G-H Distribution

作　　者：陈倩[1] 李金林[2] 

机构地区：[1]北京第二外国语学院国际商学院,北京100024 [2]北京理工大学管理与经济学院,北京100081

出　　处：《数理统计与管理》2018年第1期64-73,共10页Journal of Applied Statistics and Management

基　　金：教育部人文社科青年基金项目(12YJC790013);北京市社会科学基金(14JGC088)

摘　　要：操作风险损失强度分布呈现的“右偏性、厚尾性”特点,使操作风险损失强度的拟合面临着许多困难。为了解决传统损失分布难以拟合损失分布尾部的问题,同时又为了克服极值理论的弊端,将四参数的g-h分布用于操作风险损失强度拟合中,设计当损失强度分布为g-h分布时使用Monte Caro模拟的基本步骤。并以我国的517个风险损失数据为基础,以实证分析的形式验证所提出的方法,将拟合结果与尾部风险及其他4种常用分布进行比较。实证结果显示,在损失强度的拟合中,g-h分布能较好的捕获操作风险损失分布的“厚尾”特性,对操作风险损失分布的拟合效果最好,尾部风险计算较为合理。Loss severity distribution of operational risk is characterized as right-skewed and heavy-tailed, and numbers of challenges and pertinent issues still exist in molding operational risk. In order to solve the problem that it is too difficult to capture the heavy-tailed trait in tail of operational risk loss distribution, as well as to overcome the drawbacks of the Extreme Value Theory, a new commere g-h distribution is used for severity distribution. The steps of convolutions by using Monte Carlo simulation based on g-h distribution are designed and Var and ES is calculated. At the same time, empirical analysis with 517 hand-collected operational losses is carried out, and results of g-h, Lognormal, Weibull, Gamma and Pareto distribution are compared. Empirical results shows that g-h distribution provides a better fit effect than other distributions, and has excellent performance in describing the heavy-tailed nature and VaR.

关 键 词：操作风险 损失分布法 G-H分布 VAR 

分 类 号：F831[经济管理—金融学] O212[理学—概率论与数理统计]