Empirical Analysis Of CPI Forecast Using Web Search Data

As the consumer price index(CPI)is an important macroeconomic indicator reflecting the change of consumer goods and service prices related to people's lives,accurate and timely forecasting of major macroeconomic indicators is a essential prerequisite for the government to make correct decisions,and it is also the basis for guiding the related enterprises to carry out long-term investment development and strategic decision making.In consideration of the time lag of macroeconomic policy in the process of formulating and playing its role,and the macroeconomic time series data held by the economic policy makers have been continuously adjusted,the timeliness and effectiveness of the final data is hard to guarantee.Therefore,searching for timely and massive data sources to assist us to predict the consumer price index more accurately is a supplement,amendment,and improvement to our traditional CPI forecasting method in China.It can be see that research on the use of network big data to assist in predicting CPI is more meaningful.In this paper,I first combed the traditional consumer price index prediction model,and briefly introduced the limitations of the traditional model.Then,analysis of factors affecting price fluctuation from the perspective of money market and commodity market.Based on the equilibrium price theory,we find that there is an inevitable connection between the network search data and CPI through a variety of measurement methods.Then,according to the Baidu index search,related keywords are found to precede CPI fluctuations.Secondly,in order to simplify the index set and improve the practicability of the model,on the basis of the seasonal adjustment of CPI monthly year-on-year data,the Bayesian network is applied to reduce the dimension of keywords,and the macro index and micro index of the forecasting effect is constructed.The multivariate mixing regression model and Beta-weight function was used to fully exploited the data information related to the fluctuation law of CPI index,and predicted the CPI.The innovative points of this paper are mainly shown in the following two aspects:First,the Bayesian network method has been used to preprocess the prediction of CPIFs leading indicators,then the macro and micro index are synthesized respectively.Secondly,high-frequency search volume data has been introduced from two perspectives of macroeconomic situation and micro demand,and the M-MIDAS-AR model containing keyword search volume synthesis index is applied to the CPI forecast analysis.It is found that compared with the traditional CPI regression model,the prediction model with high-frequency network search data has comparative advantages in both the fitting effect and the prediction accuracy.This method can make real-time forecast and timely forecast of China's consumer price index.