基于半参数模型的中国商品期货尾部风险度量

China commodity futures tail risk measurement Base on semi-parametric models

以我国四种代表性的商品期货指数为例,分别运用CAViaR和条件EVT两类半参数模型对尾部数据进行动态VaR测度,同时在现有CAViaR模型基础上构造了双门限常数-TGARCHT模型和双门限AR-TGARCH模型,最后通过3个似然比检验和1个动态分位数检验对比了各个模型的预测效果以及在不同分位数的适用情况。实证结果发现:两类半参数模型各有优劣,其中CAViaR模型更适合于刻画低分位数的尾部风险,而条件EVT模型对高分位数的极端尾部风险具有更强的预测能力;新建的双门限常数-TGARCH模型显著地提升了原有模型的预测绩效,而双门限AR-TGARCH模型提升的效果不如前者。

By taking four representative commodity futures indexes in China as sample, this paper adopts two semiparametric models （CAViaR model and the conditional EVT model） to measure the dynamic VaR for the tail data. Considering the deficiency of modem models, two new models （dual-threshold constant-TGARCH model and dual-threshold AR-TGARCH model） are put forward. Then three likelihood ratio tests and one dynamic quantile test are used to compare the predictive performance and applicability in different quantiles of different models. The empirical results show that： Both of the two semiparametric models have their advantages and disadvantages respectively. Between which the CAViaR model is more suitable for depicting low-level risk, while the conditional EVT model has stronger predictive power for high-level risk; the new dual-threshold constant-TGARCH model significantly improves the prediction performance of the original models, while the improved efficiency of dual-threshold AR-TGARCH model is no better than the former.