根据证券投资基金收益率序列的尖峰厚尾特征,建立估计基金风险的VaR-GARCH模型。在正态分布、t分布及GED分布三种不同的分布假设下,对基金的VaR值进行估计,并应用Kupiec失败频率检验方法对VaR模型的准确性进行了返回检验。研究结果表明,相比之下,基于GED分布的GARCH模型计算的VaR值最能真实地反映基金风险。 According to the peak-fat tail characteristic of securities investment fund yield series, a VaR-GARCH model for estimating fund risk is established. Under normal distribution, t distribution and GED distribution under the assumption of three different distributions, the VaR value of the fund is estimated, and the accuracy of the VaR model is tested by Kupiec’s failure frequency test. The results show that, in contrast, the GARCH model based on GED distribution calculated VaR most truly reflect the fund risk.