MIDAS模型估计的偏误、有效性及其与同频率模型的区别研究

发布时间：2018-07-12 13:30

本文选题：MIDAS类模型 + MAR模型　；参考：《统计与信息论坛》2017年10期

【摘要】：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。
[Abstract]:By decomposing the high frequency regression element, the bridge between Midas model and the same frequency Mar model is found. The difference between the two models can be found in many aspects, such as model form, estimator bias, variance of estimator and so on. The theoretical results show that there are errors in the traditional Mar model which omits the high frequency sample data, but under certain conditions the Midas model and the Mar model are equivalent. There is a positive correlation between the efficiency of MAR-LS and the frequency doubling. When the data frequency of high frequency variable and low frequency variable is very different, the estimator of Midas model is more effective than Mar model. This theory is applied to the actual economy to study the mechanism and prediction ability of Chinese high-frequency asset prices on low-frequency GDP.
【作者单位】：内蒙古财经大学统计与数学学院;东北财经大学经济学院;
【基金】：国家社会科学基金重大项目《新常态下我国宏观经济监测和预测研究》(15ZDA001);国家社会科学基金青年项目《中国高等教育扩张背景下过度教育特征、效应及政策研究》(13CJY010) 内蒙古自治区自然科学基金项目《结构转换GARCH建模及其在金融资产收益波动中的应用》(2016MS0716)
【分类号】：F124;F224

【相似文献】