Forecast of Broad Money Supply Based on Long Short-term Memory Neural Network (LSTM-NN)

Abstract

The broad money supply (M2) reflects changes in aggregate demand and future inflationary pressures. Forecasting the broad money supply is of great significance to economic and financial decision-making. The factors affecting the changes in the broad money supply are diverse and unclear. Therefore, this article forecasts the broad money supply based on time series. In this article, we established a broad money supply forecasting model based on LSTM (Long Short-Term Memory). This model has a long-term memory function, which effectively solves the problems of vanishing gradient and exploding gradient in the long sequence training process. In the experiment, we have established the model based on RNN (Recurrent Neural Network) and traditional forecasting models such as exponential smoothing forecasting models and regression forecasting models to compare with the broad money supply forecasting model based on LSTM. The exponential regression forecasting model with the highest prediction accuracy in the traditional model has a goodness of fit greater than 0.99. But experiments show that the prediction accuracy of the broad money supply forecasting model based on LSTM is 2 orders of magnitude higher than the prediction accuracy of the exponential regression forecasting model. This shows that the broad money supply forecasting model based on LSTM has extremely high prediction accuracy. The forecasting model of broad money supply based on LSTM is of great significance in the short-term forecasting of Chinese broad money supply.