基于TVS-MHAR模型金融市场高频多元波动率的预测

本文基于Kalli和Griffin(2011)的时变稀疏模型和多元HAR模型,构建了具有时变稀疏性的多元HAR模型(TVS-MHAR),并利用中国上证综指、沪深300期货和国债期货的五分钟高频数据,对金融市场的已实现波动率矩阵进行预测.本文通过Cholesky转换方法保证预测波动率矩阵的正定性.通过对不同多元波动率模型的预测结果进行数值比较和经济比较,本文发现,本文构建的TVS-MHAR模型无论对于短期预测、中期预测还是长期预测都具有最高的预测精度和最大的投资改善.同时,时变多元波动率模型可以获得比固定参数模型更好的预测效果,高频数据模型比低频数据模型获得更大的投资改善.

Multivariate realized volatility forecasts of financial markets based on TVS-MHAR model

We develop a multivariate HAR model with time-varying sparsity, or TVS-MHAR by combining the multivariate HAR model with the time-varying sparsity structure of Kalli and Griffin (2011). We forecast the covariance matrix of China financial markets by utilizing the high frequency data from Shanghai Stock Exchange (SHSE) Composite index, China stock index 300 (CSI300) futures and China Treasury futures. We employ the Cholesky decomposition approach to guarantee the positive definiteness of the forecast volatility matrices. And then, we compare the forecast performance of the proposed TVS-MHAR model with other multivariate volatility forecast models in literatures based on both the statistical loss functions and the economic evaluation criterions. The results suggest the TVS-MHAR model performs the best for the out-of-sample forecasts and has the greatest economic gains among all the forecast models. In addition, the time-varying multivariate volatility forecast model performs better as compared with the fixed-parameter models and the models based on high frequency data have more economic gains as compared with the models based on the low-frequency data.