Forecasting Inflation in Nicaragua: Comparative Evidence from Machine Learning and Traditional Econometrics

Abstract

The application of Machine Learning (ML) techniques in econometric analysis has expanded considerably in recent years, deepening the discussion on how to forecast key macroeconomic variables such as inflation. In the case of Nicaragua, there is currently no evidence that ML methods have been integrated into inflation forecasting frameworks, despite the variable’s central role in monetary policy. This study evaluates the predictive performance of Traditional Econometric (TE) and ML models in forecasting inflation in Nicaragua over the 2006–2024 period. A dataset comprising 20 explanatory variables was constructed, and six models were trained: Multiple Linear regression (LM), ARIMAX, and VAR (TE), together with Elastic Net, Random Forest (RF), and XGBoost; (ML). The results indicate no statistically significant differences in predictive accuracy between both approaches in the short and long term. However, ML models identified meaningful relationships with oil and import prices that were not captured by traditional techniques. These findings suggest that combining TE and ML approaches can enhance the robustness and interpretability of inflation forecasting methodologies.