Consumer Price Index Forecast For Joining Internet Search Data

The consumer price index(CPI)is an important economic indicator that reflects the inflation or contraction of a country.It is usually used to reflect the general price level of the economy and provides an important basis for the country to formulate corresponding macroeconomic policies.However,the time for government agencies to issue CPI exists.Half-month lag,so predicting CPI in advance becomes a hot spot in the academic world.Currently,the forecasting models of CPI mainly include regression models,time series models,machine learning models,and combination forecasting models.The data used by forecasting models are almost all statistical data released by government departments,and some scholars use Internet search data to CPI does make predictions,but there is very little research that uses government statistics and web search data to predict CPI.Therefore,this paper uses these two kinds of data to build models and compares "one-step" and "two-step" methods.The difference in the prediction effect of the constructed model shows that adding network search data based on government statistical data prediction has practical and practical help in improving the prediction effect of CPI.This paper is based on the CPI to compile standard product classification criteria and text mining methods.It obtains the initial web search keywords,and then expands the keyword library through the long tail keyword expansion method and the Baidu exponent demand pattern expansion method,and uses the time difference correlation coefficient method to screen out Strong correlation with CPI,and leading or synchronizing web search keywords.Because there are correlations between some of the selected web search keywords,if the direct model establishment is prone to problems of overfitting and poor performance of the model,this paper uses Stepwise Regression Analysis,Adaptive-Lasso Algorithm,and Principal Component Analysis,respectively.The three dimensionality reduction methods reduce the dimensionality of government statistical indicators and screened web search keywords,and compare and analyze the dimensionality reduction effects of different methods.Finally,a stepwise regression method is selected for variable selection,and the key with no relevance is selected.Based on words,and on this basis,the data sets are divided into training sets and test sets,and neural network models are fitted with training set data.The prediction results of neural network models constructed by different data types or different modeling methods are compared in test sets.Conclusions: Using the “two-step method” modeling approach,adding network search data to the CPI prediction model can effectively improve the CPI forecasting effect.