基于网络搜索量和混合频率模型的经济变量预测研究

针对大数据背景下利用互联网搜索量数据进行经济预测的问题,提出建立能够充分利用高频变量信息的混合频率模型,并尝试解决建模过程中的关键词选取、数据预处理和降维等问题。在对金融和消费领域预测的实证研究中,经过筛选的关键词搜索量变量与作为预测对象的经济变量是高度相关的,并且混频模型相对于经过频率转换的模型具有更优的估计量性质和更高的样本内外预测精度。同时,根据估计结果得到的权重函数还可以发现月内各日搜索量在预测模型中的贡献度分布具有不同模式,借助该分布模式可以对经济主体行为进行描述和测度,也为搜索量数据的频率转换提供了一些参考。

Research on the Prediction of Some Economic Variables based on the Internet Search Volume and Mixed-f requency Model

This paper has proposed a mixed‐frequency modelling method for economic prediction through the Internet search volume ,which can fully utilize the information contained in the high frequency variables .The paper has also tried to address the problems of the choice of key words ,data preprocessing and dimension reduction during the modelling cycle .In the empirical studies for the prediction of key indicators in the fields of finance and consumption ,the paper has found that the search volumes of some screened key words are highly correlated with the predicted economic variables and the mixed‐frequency models have better estimator properties and in/out‐sample prediction accuracy than the models through frequency transformation .Meanwhile ,the weight function has shown that daily search volume has different contribution mode in a month ,and with this mode we can describe and measure economic body's behavior .