Comparative Analysis of LSTM, ARIMA, and Hybrid Models for Forecasting Future GDP

Abstract

Gross domestic product (GDP) is an effective indicator of economic development, and GDP forecasts provide a better understanding of future economic trends. This article investigates three methods of forecasting GDP: LSTM, ARIMA and a hybrid approach that combines two models. The principal aim is to compare the performance of these methods by computing mean square error (MSE), Root Mean Square Error (RMSE), Mean Average Error (MAE), the coefficient of determination (R2) and to determine which model provides the most accurate and reliable forecasts. The study collected quarterly GDP data from the Federal Reserve Economic Data, covering a period of 75 years from 1947 to 2022. The LSTM model, using the HE initialization technique to initialize the weights, was trained and tested using the GDP data. the ARIMA model was configured with parameters (1,2,1), and the hybrid (ARIMA-LSTM) model combined the strengths of both approaches. It was found that LSTM (MSE=0.010, RMSE=0.104, MAE=0.077, R2=0.96), ARIMA (MSE=0.095, RMSE=0.309, MAE=0.286, R2=0.75) and Hybrid (MSE=0.0018, RMSE=0.043, MAE=0.028, R2=0.99) and the hybrid model achieves better prediction accuracy than the individual models in predicting GDP.