中文摘要：

混合数据抽样（MIDAS）模型能处理不同频率变量间的动态关系。本文比较了MIDAS波动模型和ABDL模型，发现我国股市对数实现波动呈长期记忆性。在预测波动方面，ABDL模型优于MIDAS模型；利用MIDAS波动模型预测实现波动水平，日绝对值报酬是最好的回归项；利用MIDAS模型预测未来波动，至少应采用一个月的历史数据。

英文摘要：

This paper compares MIDAS （Mixed Data Sampling） regression models and ABDL model in the ability of predicting volatility. MIDAS models can take different specification of regressors, such as realized volatility, realized power, squared returns, absolute returns, and return ranges. Using 5-minute stock return data in Shanghai and Shenzhen Stock Exchange, we find MIDAS models do not excel ABDL model in pre- dicting volatility. However, measured by increments in quadratic variation, daily absolute return is the best predictor in MIDAS models. Furthermore, historical data of one month are sufficient to capture the persistence in volatility.