ECONOMIC PREDICTION MODELS USING MACHINE LEARNING

Abstract

This study analyzes the predictive capacity of econometric and machine learning models to estimate Ecuador’s real GDP, using quarterly data for 2008-2022 and considering inflation and the interest rate as explanatory variables. Multiple linear regression models were estimated through Generalized Least Squares, along with distributed lag models, Random Forest, and Long Short-Term Memory (LSTM) neural networks. Predictive performance was assessed using out-of-sample error metrics, namely RMSE and MAE. The results show that, in the highly volatile context of 2022, the optimized distributed lag model achieved the lowest prediction error, whereas in a more stable macroeconomic scenario the LSTM model reached higher relative accuracy. The study concludes that predictive effectiveness depends on both the model specification and the economic context, and that traditional econometric approaches and machine learning techniques may be complementary for macroeconomic forecasting in dollarized economies.