Research On CPI Combination Forecast Based On Network Search Data

In recent years,the speed of price increases has accelerated in China.The government has taken a series of measures to stabilize the rise of the Consumer Price Index(CPI),including increasing the reserve ratio,vigorously cracking down on speculative behavior in the capital market,and increasing credit control.However,these economic policies have external lag and the implementation effect is also relatively unstable,which to a large extent cannot solve practical problems.In the era of big data,internet search engines record the behavioral choices of demand and supply sides towards information,condense valuable social activity information,and provide effective data resources for studying user behavior.Network search data,with its advantages of timeliness and convenience,provides new ideas for studying macroeconomic indicators.Compared to traditional data prediction,network search data can accurately and timely reflect economic fluctuations affected by certain factors.Based on this,using network search data to predict and analyze the influencing factors and changing trends of CPI can provide reference value for the country to formulate macroeconomic regulation policies in a timely and effective manner.The paper first organizes and studies traditional CPI prediction models,analyzing their limitations.Then,the factors influencing CPI changes are analyzed from both macro and micro perspectives,and the internal correlation mechanism between network search data and CPI changes are elaborated.Secondly,the correlation between the selected macro and micro keywords and CPI is defined,leading keywords are selected,and principal component analysis is used for dimensionality reduction,ultimately obtaining the synthesized network search data indicators.Finally,a combination prediction method based on the Set Empirical Mode Decomposition(EEMD)method is used to denoise both CPI and network search data indicators.Cluster analysis is used to reconstruct the stationary time series and give it certain economic significance.Based on the reconstructed high-frequency,intermediate frequency,low-frequency,and trend terms,reasonable modeling and prediction are carried out.Support Vector Machine(SVM)is used to integrate the final CPI index.Through analysis and research,the following conclusions are obtained: firstly,based on the influencing factors of CPI,keywords are initially screened from macro and micro perspectives,and time difference correlation analysis is used to screen keywords with high correlation and a leading period.The synthesized network search data indicators can better reflect the fluctuation trend of CPI.Secondly,compared to traditional prediction models,the SVM model with network search data indicators has better prediction performance than the ARIMA model without network search data indicators.Third,compared with the single prediction model,the combined model after EEMD decomposition is significantly better than the single model,and the combined model after the integration of the sub item predictive values is better than the non integrated combined model.The model in this article has significant advantages in predicting CPI trends,and its error rate is the lowest among the comparative models.The innovation of the paper is mainly reflected in the following aspects: at the theoretical level,theoretical analysis is conducted on the influencing factors of CPI,the characteristics of network search data,and the internal correlation mechanism between network search data and CPI.CPI keywords are selected from both macro and micro perspectives.At the methodological level,a CPI combination prediction model based on network search data is constructed by utilizing the timeliness advantage of network search data and the high-precision characteristics of combination prediction methods.At the conclusion level,the combined forecasting model with online search data has comparative advantages in the fitting effect and forecasting accuracy of CPI compared with the single model and the traditional forecasting model.The model proposed in this article can provide timely prediction of macroeconomic indicators and provide certain reference for policy formulation.