Research On The Impact Of Internet Search Behavior On Macroeconomic Boom Index Prediction

In today's society,whether it is the individual's economic behavior,the operation of the enterprise or the government's policy formulation,it is closely related to the operation of the macro economy.Therefore,it is very necessary and meaningful to predict the development trend of the macro economy.The macroeconomic prosperity index is to use the economic cycle theory to comprehensively and objectively judge whether the economic operation is too cold,overheated or stable,expansion or contraction.On the one hand,it can help people intuitively understand and analyze the economic growth cycle or cycle,on the other hand,it can help scholars to predict economic or economic warnings.Today,when big data is widely used,whether and how to use macro data to predict the macro economy has become a hot spot for scholars.This article will use Internet search behavior as a representative of online big data to help predict the macroeconomic sentiment index.In order to study whether big data can help predict the macroeconomic sentiment index,this paper discusses the role of structured data and unstructured information in the macroeconomic sentiment index.In the aspect of index selection and data processing,the government statistical indicators are selected from the structured data.The total time is 62 categories of 62 indicators with a time span of 2007 to 2017.Unstructured information selected 56 Baidu search indices based on five categories:investment,consumption,employment,import and export,and government purchase.In the empirical data,the data from 2007 to 2015 was selected as the training set,and the data from 2016 to 2017 was used as the test set..In this paper,five autoregressive fractional lag models(ADL)are established.The fitting results of the model show that the equal treatment of structured data and non-structural information can not optimize the traditional prediction method,and the "two-step method" is adopted to fully use the structure.The addition of unstructured information can reduce the fitting error of the economic sentiment index to MSE=0.046158,which is much better than the traditional model MSE=0.10015.At the same time,due to the large number of selected indicators,the variable selection method was used to select the indicators,and the variable selection method with LASSO,SCAD and Elastic-net penalty functions was compared.From the prediction effect,the prediction effect of adding unstructured information to the Elastic-ADL model is MSE=0.11413,which is better than the traditional prediction effect of the structured data model only MSE=0.14207.It can be seen that joining the Internet search behavior can indeed improve the prediction effect on the macroeconomic sentiment index.