The Short-term Prediction Of CPI Based On Baidu Search Index

The consumer price index reflects the level of price changes of specific products and services residents consume,playing an important reference in our country's macro-control policy and also reflecting the inflation and people's livelihood.National Bureau of Statistics generally releases CPI of previous month around the 15th of each month.CPI has lag property,So the prediction of CPI has received many researchers's attention.Most traditional forecasting methods of CPI are time series,neural network,combination forecasting method etc.The data used to predict CPI is often historical data of CPI and macroeconomic indicators linked with CPI.However,big data has given new content to CPI predicting.With the development and popularization of the Internet,there is an important stage in the middle of purchasing needs and purchasing decisions whose name is searching information.Consumers query information through the Internet,while the Internet records consumer behavior.Consumer behavior form data,which implies much value,providing foundation for the study of economic phenomena and economic activities.,Baidu search index come into being in Internet age.It records the behavior of data users.If you type the keyword in Baidu search box,you can get search index curve of the keyword?It selects the monthly CPI data from January 2008 to April 2015 as the dependent variable.It introduces consumer behavior theory and shows the feasibility of using Baidu search index predict the short-term of CPI.It explains that how to predict CPI in Bayesian point.It uses Spike-Slab priori select variable based on using correlation coefficient method select keywords.Spike-Slab method can be performed automatically and more robust in variable selection.It uses Markov chain Monte Carlo simulate parameters of CPI short-term forecasting model.It analyze simulated Markov chains and establish the optimal distributed lag model.The results show that the paper predicts CPI from May 2014 to April 2015 based on short-term forecasting model of CPI.Comparing predicted value to real value of CPI,the fluctuations are less than 1.6%.The explanatory variables in the model contain seven items in addition to tobacco and liquor.Food and housing commodities only have the current variable and have affect on CPI in short duration relatively.The other five items contain first order lagged variables or second order lagged variables and have impact on CPI in the long duration relatively.Baidu search index can achieve real-time monitoring of CPI.