The Selection Of Forecasting Models Based On Consumer Price Index

The consumer price index(CPI),as the relative number to evaluate the price changes of consumers’ purchase of living goods and expenditure on service items,mainly reflects the trend and range of price changes.The change of consumer price index also reflects whether the country is in a state of inflation or deflation.The prediction of consumer price index has important practical significance for the country and the government to formulate macroeconomic policies and control macroeconomic.This paper takes China’s monthly CPI series from January 2007 to December 2021 as the research object,uses R language software to establish multiple models to fit and predict China’s CPI data from January 2022 to June 2022,and compares and evaluates the prediction results of multiple models to select the best model.Firstly,this article uses ARIMA model to fit and predict CPI sequences.To achieve the goal of improving prediction accuracy,this article selects 9 economic indicators with strong correlation with CPI,and uses LASSO regression method to screen important variables(industrial producer purchase price index,fiscal revenue,and fiscal expenditure).These three variables are used as input variables for the ARIMAX model,and an ARIMAX model is constructed to fit and predict the CPI sequence.Secondly,based on the theoretical methods of support vector machines in regression problems,this article constructs a support vector regression(SVR)model for CPI sequences and predicts CPI.Finally,based on the principles and advantages of combination models,this article believes that the correct selection of combination models and methods can improve prediction performance.Therefore,the ARIMAX-SVR model is proposed and established,and series combination models and parallel combination models are also constructed.The specific idea of the ARIMAX-SVR model proposed in this article is to select an ARIMAX model that has better predictive performance compared to the ARIMA model to characterize linear features in China’s CPI sequence.Then,support vector regression(SVR)model is used to extract non linear features in the residual sequence of the ARIMAX model.The prediction results of the two models are superimposed to obtain the predicted values of the ARIMAX-SVR combination model.In this paper,three single models,ARIMA model,ARIMAX model and SVR model,as well as series combination model,parallel combination model and ARIMAX-SVR combination model of ARIMAX model and SVR model,are respectively established for CPI sequence.And this article compares and analyzes the prediction results of the above several model methods,and draws the following conclusions:Firstly,in terms of time series models,the ARIMAX model performs better than the ARIMA model,indicating that the variables extracted through the LASSO regression method(industrial producer purchase price index,fiscal revenue,and fiscal expenditure)are conducive to fitting the trend of CPI series changes.Secondly,the models established in this article have relatively good predictive performance for CPI sequences,indicating that the models established in this article can be applied to the prediction of CPI sequences.Thirdly,through comparative analysis,it can be seen that the ARIMAX SVR model proposed in this article has the best prediction performance.This indicates that the effective combination of ARIMAX model and support vector regression(SVR)model can better characterize the characteristics of CPI and grasp the dynamic laws of CPI,which has advantages in short-term CPI prediction.A reasonable combination model can combine the advantages of multiple single models,make up for the defects of a single model,and effectively improve the prediction accuracy.