Research On Chinese Inflation Rate Forecast Based On CEEMDAN-GARCH-LSTM Model

The inflation rate reflects the rising rate of the general price level in a certain period,which can effectively measure the residents’ consumption level and purchasing power.In addition,forward-looking and accurate prediction of inflation rate is helpful for government agencies to formulate reasonable monetary policy,so forward-looking prediction of inflation rate is very important.Due to the complexity of financial time data,it includes linear characteristics and nonlinear characteristics.For the linear part,the traditional model is more advantageous;For the nonlinear part,the use of deep learning model has more advantages.Therefore,this paper combines the traditional linear model GARCH with the deep learning model LSTM to predict the inflation rate.This thesis selects the consumer price index(CPI)and production price index(PPI)from January 1993 to February 2022 as the research objects.In order to eliminate the impact of seasonal changes,monthly year-on-year data are selected.Firstly,according to AIC and BIC criteria,the lag order of the mean value equation is determined,and GARCH model,TGARCH model and EGARCH model are respectively constructed to select the optimal model for prediction;Secondly,the number of neurons,time step,batch size and optimization algorithm of LSTM are adjusted by trial and error method to select the optimal parameters,and then LSTM model is used for modeling and prediction;Thirdly,build CEEMDAN-LSTM model.CEEMDAN method is used to decompose the original sequence,and the decomposed high-frequency components and low-frequency components are added respectively.Then,LSTM model is used to predict the sum of high-frequency components,the sum of low-frequency components,and the trend term respectively,and the predicted values are added to get the final result;Finally,CEEMDAN-GARCH-LSTM model is built.GARCH model is used to predict low-frequency components and trend items,LSTM is used to model high-frequency components,and the final results are reconstructed to obtain the final prediction results.The results show that although CPI and PPI data meet the modeling requirements of GARCH family models,the prediction accuracy of GARCH family models is low;The CEEMDAN algorithm is used to decompose the original sequence,and it is found that the volatility of the sequence is basically consistent with that of the high-frequency component.Therefore,the change of the high-frequency component can be regarded as the change caused by random factors,while the low-frequency component and trend term can be regarded as the influence of long-term stability factors;Among GARCH model,LSTM model,CEEMDAN-LSTM model and CEEMDAN-GARCH-LSTM model,CEEMDANGARCH-LSTM model has the best prediction effect.