Prediction of Consumer Price Index based on Long Short-Term Memory Model

Abstract

Consumer Price Index(CPI) is the main standard to identify inflation or deflation. Accurate prediction of CPI will help the government to implement macro-control and formulate price stabilization policies, so as to achieve the goal of building a moderately prosperous society. CPI is a non-stationary and non-linear time series, and has relevance in time dimension. In order to fully mine the correlation of CPI sequence in long and short time span, a method of predicting CPI using Long Short-Term Memory(LSTM) model is proposed. Taking historical CPI data of Anhui Province as the empirical analysis object, modeling and predicting are carried out. The prediction effect of LSTM is compared with classic time series model-Autoregressive Integrated Moving Average(ARIMA). According to the predicted results, LSTM model has significantly improvement in RMSE and MAE indicators compared with ARIMA model, indicating the LSTM model has higher prediction accuracy. The SDAE indicator of LSTM model is smaller than ARIMA model, indicating the LSTM model has better prediction stability.