基于标准残差的极值风险模型准确性研究  

The Accuracy of Risk Measurement model Base on Standard Residuals

作　　者：林宇[1] 魏宇[1] 黄登仕[1] 

机构地区：[1]西南交通大学经济管理学院,成都610031

出　　处：《管理评论》2006年第12期8-14,共7页Management Review

基　　金：国家自然科学基金(70501025;70572089);国家杰出青年科学基金(70229001)。

摘　　要：本文使用ARMA(1,1)与GARCH(1,1)、GJR(1,1)模型结合构造出标准残差序列,然后分别与条件EVT、条件正态分布、条件t分布和非条件EVT结合,形成8个风险测度模型,并分别用这些模型估计沪、深股市在置信水平为95%、97.5%、99%、99.5%的动态VaR(Value-at-Risk),然后用Kupiec(1995)和Christoffersen(1998)的返回测试(Back-testing)方法,判定沪、深股市对模型的准确性。研究结果表明,条件EVT风险模型能准确测度沪、深股市的风险,而非条件EVT模型缺乏准确性,对其它模型的准确性因置信水平不同而表现出差异性。在所有模型中,最能准确测度沪、深股市风险的模型分别是,置信水平为99.5%时的条件EVT-GARCH模型和置信水平为97.5%时的条件EVT-GJR模型。本文的研究结果为测度中国大陆沪、深股市风险在模型和置信水平的选择上提供了实证依据。This paper uses ARMA(1,1) and GARCH(1,1) and GJR(1,1) to construct a standard residuals series, and then combines it with conditional EVT, normal, student-t distribution and unconditional EVT respectively, so that 8 risk measurement models are formed. Next, we use these models to measure VaR(Value-at-Risk) of ShangHai and ShenZhen stock markets under confidence level 95%, 97.5%, 99%and 99.5%. And then we also apply Kupiec(1995)and Christoffersen(1998) back-testing methods to test the accuracy of these risk measurement models in Chinese stock markets. Our results show that conditional EVT model can accurately measure the risk of both ShangHai and ShenZhen stock markets, but unconditional EVT model can not. As for other models, there are differences under different confidence levels for different stock markets. Of all models, conditional EVT-GARCH (99.5% confidence level) and conditional EVT-GJR (97.5% confidence level) are the best models for ShangHai and ShenZhen stock markets, respectively. Our results provide an empirical basis for selecting model and confidence level and for estimating dynamics VaR of Chinese stock markets.

关 键 词：EVT 标准残差 风险模型 返回测试 准确性 

分 类 号：F224[经济管理—国民经济]