基于门槛效应的Realized MAT-HAR GARCH模型波动预测

Volatility Forecasting of the Realized MAT-HAR GARCH Model Based on Threshold Effects

基于门槛效应视角,结合MAT-HAR模型能够捕捉门槛效应的特点与Realized GARCH模型在高频波动率预测的优势,文章建立了Realized MAT-HAR GARCH模型,旨在研究高频金融市场的门槛效应.与Realized HAR GARCH模型不同的是,该模型在波动方程中引入MAT-HAR结构,将已实现波动率序列的移动平均项作为门槛,并基于AIC信息准则选取最优的波动方程结构.数值模拟结果显示,Realized MAT-HAR GARCH模型有着较好的参数估计和波动预测效果.基于沪深300指数的实证研究表明,在残差服从正态分布的假设下,沪深300指数的周和月波动信息存在显著的门槛效应,且基于AIC准则的Realized MAT-HAR GARCH模型比Realized HAR GARCH模型有着更好的样本内拟合、波动预测和VaR度量效果.最后,在不同误差分布假设下,研究该模型基于AIC信息准则选取模型结构的合理性和波动预测的稳健性.基于MCS检验的稳健性检验结果表明,在不同厚尾分布的假设下,Realized MAT-HAR GARCH模型的样本外表现效果均优于Realized HAR GARCH模型,其效果不依赖于滚动预测窗口长度的选取.

Based on the perspective of threshold efects,combining with the characteristic of the MAT-HAR model capturing the threshold effects and the advantage of Realized GARCH model in high-frequency volatility prediction,this paper establishes the Realized MAT-HAR GARCH model to study the threshold effects of high-frequency financial market.Different from the Realized HAR GARCH model,this model introduces MAT-HAR structure into the volatility equation,takes a moving average of the lagged realized volatility series as the threshold,and selects the optimal structure of volatility equation based on AIC information criterion.The numerical simulation results show that the Realized MAT-HAR GARCH model has the good effects on parameter estimation and volatility prediction.The empirical study based on the CSI 300 index shows that under the assumption that the residuals obey normal distribution,significant threshold effects exist at the weekly and monthly level,and the Realized MAT-HAR GARCH model based on AIC criterion performs better on in-sample fitting,volatility forecasting and VaR measurement than Realized HAR GARCH model.Finally,the rationality of the model structure selected based on AIC information criterion and the robustness of volatility prediction are studied under the assumption of different error distributions.The results of robustness test based on MCS test show that under the assumption of different heavy-tailed distributions,the out-of-sample performance of Realized MAT-HAR GARCH model performs better than that of Realized HAR GARCH model,which does’t depend on the length of rolling windows.