MIDAS模型估计的偏误、有效性及其与同频率模型的区别研究

The Bias and Validity of MIDAS Model′s Estimation and Its Differences with the Same Frequency Model

通过分解高频回归元,探寻出MIDAS类模型及同频率MAR模型之间联系的桥梁,从模型形式、估计量偏误、估计量方差等诸多方面呈现出两类模型的区别。理论推导结果表明:遗漏高频样本数据的传统MAR模型存在偏误,但在一定条件下MIDAS类模型与MAR模型具有等价性;MAR-LS的有效性与频率倍差存在正向相关性,当高频变量与低频变量的数据频率相差迥异时,MIDAS类模型的估计量较MAR模型有效。将此理论应用于具体实际经济中,以研究中国高频资产价格对低频GDP作用机制及预测能力。

Through the decomposition of the high-frequency regression element,it explores the bridge between the MIDAS models and the traditional moving average regression （MAR） models with the same frequency,and shows the difference between the two models from the aspects of the models,the estimation error and the variance of the estimator.The theoretical results show that the MAR models of missing high frequency regression sample data have been biased,but under certain conditions,the MIDAS models is equivalent to the traditional MAR models.There is a positive correlation between the effectiveness of MAR-LS and frequency difference.When the variables of frequency difference between high frequency is different,the estimation of MIDAS models is more effective than MAR model.Then the theory is applied to the real economy,the mechanism and the predicting ability of the high frequency asset price effect on the low frequency GDP is studied.