CPI Forecast Based On Baidu Index

CPI is a major index to reflect a country's economic level and economic fluctuations.It can be used to reflect the level of the price,inflation and deflation of a country,and provide an important basis for the country to formulate relevant economic policies.However,there is a certain lag which is the CPI data will be promulgated in the middle of the next month.Therefore,more and more attention has been paid to research the prediction of CPI.At present,the data of CPI was usually predicted according to the statistics or the indicators by the government which is not accurate.Meanwhile,the Internet search data is also used by a few scholars.The shortage of the latter is the scholars selected the key words by Online Expansion Algorithm of Long Tail Queries(OEALTQ)or the subjective keywords selection which is subjective and not comprehensive enough.The way of mining key words in this paper are both micro point and macro point.The micro way to select key words is based on the classification standard of compiling specifications of the CPI,and the macro way to select key words is crawling the news websites or the papers related to CPI on the main network and expanding the thesaurus by QEALTQ or Baidu Index.After crawling the related Baidu Index,we can adjust it seasonally,and screen Baidu Index more associated with CPI which are leading or synchronize by the time difference correlation analysis method.There may be over-fitting or poor-fitting problems when the Baidu Index was applied to Machine Learning Model directly because of the relationship of the selected key words of the Baidu Index.In order to solve this problem,we made a further data filtering by the way of Stepwise Regression Analysis.Firstly,we screened the key words by One-stage Method and Two-stage Method.Moreover,we analyzed the selected key words by using Machine Learning Model and Multivariate Regression Model,and divided them into training set and predictive set,analyzed different stepwise regression analysis ways and different machine learning models to draw the best model.Finally,we could conclude that we could fit the data of the training set effectively by the way of XGB model and Two-stage Method,and Improve the forecasting effect of CPI.