中文摘要：

目前我国期货市场尚不成熟，没有形成完备的上市品种结构，因此，实现一种期货对多种现货的交叉套保在风险管理中具有重要意义．考虑到投资者行为的异质性和波动相关持久性等高频波动经验特征，本文将不同期限的波动因素引入到基于高频数据的CAW（Conditional Autoregressive Wishart）模型中，构建了高频波动率矩阵的动态变化模型HAR（Heterogeneous Autoregressive）-CAW模型．同时，利用矩阵纠偏技术，将该模型与均值模型融合，提出了收益率和波动率的联合动态变化模型一均值HAR-CAW模型．最后，基于该联合模型多期预测值的理论推导，并结合最小方差理论，实现了不同平衡周期下最优套期比的动态估计．针对上证50ETF、中小板ETF和沪深300股指期货的实证分析表明，高频波动率矩阵具有高峰厚尾和长记忆性等经验特征；同时，不同现货资产比例、不同时间区间及不同平衡周期下，本文所提交叉套期保值模型的套保效率显著优于基于日度数据的OLS（Ordinary Least Square）模型和DCC（Dynamic Conditional Correlation）等常用模型，而本文所提的矩阵纠偏技术也显著提高了末纠偏时的套保效率．

英文摘要：

China＇s futures market is not mature and the futures products is not complete now. It is of great significance that using a futures hedging on multiple spot in risk management. In the paper, considering the heterogeneity of investors behavior and volatility persistence, we build the Heterogeneous Autoregressive- Conditional Autoregressive Wishart （HAR-CAW） model, which is a high-frequency dynamic change model of volatility matrices, through introducing the fluctuating factors of different period into the Conditional Autoregressive Wishart （CAW） model based on high-frequency data. Then, using matrix correction tech- nology, we put forward the joint dynamic model of returns and volatility-mean HAR-CAW model-by integrating the HAR-CAW model and average model. Finally, based on the theoretical derivation of the joint modeFs multi-period prediction, and connecting with the minimum variance theory, we achieve the dynamic optimal hedging ratio estimation under different balance cycle. In the empirical research based on the Shanghai 50 ETF, small and medium-sized plate ETF and CSI 300 stock index futures, we compare the models in different spot proportion, different time interval and different balance cycle of hedging efficiency. The empirical results show that our model＇s hedging effect is obviously better than Ordinary Least Squares （OLS） model and Dynamic Conditional Correlation （DCC） model which are based on daily data, and the matrix correction technology proposed in this paper is also significantly improve the efficiency of the hedge compared with no matrix correction model.