基于GARCH类模型的我国创业板市场风险度量比较

摘要

As the research object,this article takes our country the GEM index in normal distribution,t-distribution and GED distribution assumption that comparative analysis of the different types of GARCH model,predict the VaR of yield sequence under different confidence level,and carries on the comparison to the results and inspection,the following conclusion:although PARCH and EGARCH model to predict the VaR than GARCH and TGARCH model is more accurate,but not VaR measure-ment model selection of the key factors,the key factors influencing the VaR measurement is distribution hypothesis and the level of significance. Under the normal distribution,when the confidence level is low,can well depict earnings estimates of VaR of the end of the sequence characteristics,but when significance level is very high (99%),estimates of VaR has the phenomenon of underpriced risk;Under the distribution of t-estimates of VaR value overestimate risk phenomenon;Under the GED distribution,regardless of the significance level of high and low,GARCH kind of model can well describe benefits of the end of the sequence features.%以我国创业板指数为研究对象，在正态分布、t分布和GED分布假设下比较分析不同类型的GARCH模型，预测不同置信水平下收益率序列的VaR值，并对结果进行比较和检验，得出如下结论：尽管EGARCH和PARCH模型预测的VaR值比GARCH和TGARCH模型更加精确，但模型种类的选择并非VaR值度量的关键，而分布假设与显著性水平则是影响VaR值精确度的关键因素。在正态分布下，当置信水平较低时，估计的VaR值能较好地刻画收益序列的尾部特征，但当显著性水平很高（如99%）时，估计的VaR值存在低估风险的现象；在t-分布下估计的VaR值存在高估风险的现象；在GED分布下，无论显著性水平的高低，GARCH类模型均能很好地刻画收益序列的尾部特征。