基于GARCH-EVT-COPULA模型的外汇投资组合风险度量研究  

Measuring the Value-At-Risk of Foreign Exchange Portfolio by a Garch-Evt-Copula Based Model

作　　者：苟红军[1] 陈迅[1] 花拥军[1] 

机构地区：[1]重庆大学经济与工商管理学院,重庆400044

出　　处：《管理工程学报》2015年第1期183-193,共11页Journal of Industrial Engineering and Engineering Management

基　　金：教育部人文社会科学基金资助项目(10XJA630003)

摘　　要：基于巴塞尔协议和商业银行风险管理新视角,本文运用GARCH-EVT-COPULA模型构建了GARCHEVT-Gaussian-COPULA和GARCH-EVT-t-COPULA两种方法,对美元、欧元、日元和港元四种人民币汇率的等权重投资组合风险进行了度量,优化了极值分布作为COPULA函数边际分布的多序列超阈值比例确定方法,并与历史模拟法、正态方法、蒙特卡罗方法三种传统方法和单用极值方法度量的VaR进行了比较并作返回检验。结果表明:GARCH-EVT-Gaussian-COPULA和GARCH-EVT-t-COPULA两种方法符合巴塞尔协议的内部模型法(I--)对商业银行风险管理的要求,而且估计结果优于其他四种方法,其中,GARCH-EVT-Gaussian-COPULA方法的估计结果比GARCH-EVT-t-COPULA方法更有利于节省商业银行的经济资本。The foreign exchange portfolio risk is an important element of market risks. The time series of foreign exchange also has characteristics such as non-normal with peak tail,thick tail,extremum,multidimension,and nonlinearity. Therefore,the paper adopts the new optic perspective under Basel Agreement,capture tails via EVT,and use COPULA to build multivariant models. The two methods of GARCH-EVT-Gaussian-COPULA and GARCH-EVT-t-COPULA are constructed to measure equally-weight portfolio risks of foreign exchange,including USD,EUR,JPY,and HKD.This paper optimizes the method of measuring VaR. Firstly,new time series close to independent identically distributed（ I. I.D） are produced by filtering raw time series data based on the FHS method and the GARCH model. Secondly,the kurtosis method is adopted based on the POT model. The way of determining multi-series superthreshold proportion is optimized. Each series of superthreshold proportion is calculated based on the kurtosis method and their maximum value is selected based on multi-series superthreshold proportions. Thirdly,the following six methods of measuring VaR and backtesting are compared： three traditional methods of History Simulation,Normal,and Monte Carlo,the method of Extreme Value Theory（ EVT）,and two methods of GARCHEVT-Gaussian-COPULA and GARCH-EVT-t-COPULA.The following methods are used to compare VaRs with diferent degrees of confidence. Firstly,Monte Carlo has higher probability to underestimate VaR. Secondly,history simulation has higher probability to overestimate VaR,and VaR has low stability with extreme values. Thirdly,the method of Extreme Value Theory has higher probability to overestimate VaR,and also low stability of VaR which is better than History Simulation with high degree of confidence. Fourthly,the two methods of GARCH-EVT-Gaussian-COPULA and GARCH-EVT-t-COPULA have higher probability to correctly estimate VaR because changes are centered and the values are less than that of Normal.The comparison of VaRs by backtesting these method

关 键 词：GARCH-EVT-COPULA 外汇 投资组合 在险价值(VaR) 

分 类 号：F830[经济管理—金融学]