Prediction Of CPI Based On EEMD-SARIMA-LSTM Hybrid Model

The consumer price index(CPI)is an extremely important statistic index in the macro economy,it is of great significance in measuring the degree of inflation and excluding the influence of prices in the calculation of GNP,the relationship with exchange rate fluctuations and PPI.At the same time,its rate of change can also effectively reflect the degree of inflation.The study of the change range and fluctuation trend of CPI can provide a practical basis for government departments to grasp the economic market situation and formulate relevant intervention poli-cies.While CPI needs to collect price data from each survey site for calculation,so it is a set of time series data with hysteresis quality.As the CPI calculation formula contains the final prices of social products and services,these two factors are affected by a large number of factors,which have properties such as uncer-tainty and non-linearity,so it is difficult to improve the accuracy of prediction.Traditional time series models,such as ES,ARIMA,GM and BP neural network model,etc,have some advantages in predicting based on certain data character-istics of CPI,but there are also inevitable disadvantages.For example,when ARIMA and SARIMA predict the time series with significant nonlinear factor ratio,the prediction will have large errors and limitations.Therefore,this article adopts the method of ordered combination of various models to predict China's CPI,which makes up for the shortcomings and improves the accuracy of predic-tion.In addition,CPI,which is the non-stationary and non-linear financial time series,it is an intelligent method to break through the bottleneck of prediction accuracy to input model prediction after extracting the characteristics of the se-quence.At present,it is also a cutting-edge method for forecasting the trend of complex financial time series.Therefore,Therefore,The article will appropriately improve the existing CPI prediction methods,and adopt the method of orderly combination of various models to predict China's CPI.EEMD decomposition method is used to extract the characters of CPI sequence by applying the idea of”decomposition before combination".For the decomposed subsequence,the SARIMA and the neural network LSTM models are combined to effectively pre-dict the linear and nonlinear effects of the subsequence,and the hybrid model is constructed to make up for the shortcomings of the single model.This paper first uses the CPI data from January 2001 to November 2019 to construct and predict by the single model,like SARIMA and LSTM neural net-work.Then use the EEMD decomposition method to decompose the original CPI sequence to obtain several eigenmode functions(multiple high-frequency and low-frequency sequences)and residual sequences,and apply multiple sequences to SARIMA,LSTM neural network for fiting prediction respectively.The pre-diction results of each model are added to the final CPI prediction result,and The quality of the model are comprehensively compared through the two values of root mean square error and mean absolute percentage error,finding the best CPI prediction model and giving the CPI results predicted by the model in the next six months.The empirical results show that compared with the prediction of a single model,the prediction accuracy of a hybrid model-EEMD-SARIMA-LSTM constructed using the idea of "decomposing first and then grouping" has been improved,which confirms to some extent that the hybrid model is practical in the field of analysis and prediction of CPI.Meanwhile,it can guide the government to grasp the economic market situation and formulate relevant intervention policies.