基于AEPD分布和ALD分布的VaR模型  

VaR Models Based on AEPD and ALD

作　　者：李孝华[1] 宋敏[1] 

机构地区：[1]南开大学经济学院

出　　处：《数量经济技术经济研究》2013年第1期135-149,共15页Journal of Quantitative & Technological Economics

基　　金：国家自然科学基金项目(11101225);教育部人文社会科学研究项目(10YJC910006);中央高校基本科研业务费专项资金项目(NKZXB10052)的资助

摘　　要：本文以成熟市场和新兴市场的六个主要的市场指数为例,将更精确反映金融资产收益率典型事实的AEPD分布和ALD分布运用于股票市场VaR的度量。并与其他常见的非参、半参和参数法VaR模型进行全面比较。实证表明,对于参数法模型,误差项服从ALD分布和正态分布的GARCH族模型分别当且仅当在度量低分位数和高分位数水平下的VaR值时表现优异;而误差项服从AEPD分布的GARCH族模型在度量各种分位数水平下的VaR值时均取得不错的效果。另外对于CAViaR模型,它们在度量VaR时与参数法中表现最好的AR-GJR-GARCH-AEPD(ALD)两个模型效果相当。The paper takes the six major market indices in mature markets and emerging markets for example, applying the Asymmetric Exponential Power Distribution （AEPD） and Asymmetric Laplace Distribution （ALD） which can more accurately reflect the stylized {acts of the return on financial assets to measure Value at Risk （VaR） in the stock market. Then we compare these models with other common non-parametric, semi-parametric and parametric VaR models. The empirical results show that for the parametric models, the GARCH models which the error term obeys the ALD and normal distribution do well if and only if we measure low and high quantile VaR respectively. The GARCH models which the error term obeys the AEPD perform well both in the high and low quantile VaR measure. In addition, for the CAViaR models, they are as good as the two best parametric models that are AR-GJR-GARCH-AEPD model and AR-GJR-GARCH-ALD model respectively.

关 键 词：VAR GARCH模型 AEPD分布 ALD分布 CAVIAR模型 

分 类 号：F830.91[经济管理—金融学]