我国黄金现货市场的动态VaR预测模型研究

Dynamic VaR Predicting Models for Chinese Spot Gold Market

以我国黄金现货市场主力品种——Au99.95的日数据为研究对象,运用滚动时间窗法进行了多种波动率模型和不同条件收益分布假定下的样本外动态VaR预测。进一步,运用更加严谨和稳健的KupicLR检验以及动态分位数回归检验法,对不同模型得到的VaR预测精度进行了深入的后验分析。主要实证结果显示,我国黄金现货市场的波动不存在显著的杠杆效应,但却同时具有明显的条件"有偏"和"尖峰胖尾"特征。另外,假定条件收益服从有偏学生分布的EGARCH模型具有最好的样本外极端风险预测精度。

Based on the dataset of a main product in Chinese spot gold market,Au99.95,this paper carries out dynamic out-of-sample VaR predicting for various volatility models and conditional return distributions by methods of rolling time windows.Furthermore,two robust backtesting methodologies,Kupic LR test and dynamic quantile regression test,are introduced to estimate the accuracy for VaR predictions produced by different models.The main results show that there is no significant leverage effect in Chinese spot gold market but clear conditional skewness and fat-tail are observed.In addition,EGARCH model with conditional skewed student distribution produces the most accurate out-of-sample VaR predictions for extreme market risk.