Forecasting Inflation in Times of Stability and Crisis: A Machine Learning Approach

T he Bolivian economy is undergoing its most severe crisis since the 1980s, marked by a dramatic transition from low and stable inflation to pronounced inflationary pressures. In this context, the development of reliable forecasting tools has become increasingly critical. This study evaluates the predictive performance of several widely used Machine Learning (ML) models under two distinct macroeconomic conditions: periods of relative stability and periods of crisis. The findings reveal that ML models in general outperform traditional econometric approaches across both conditions, with the XGBoost algorithm demonstrating the best performance. Additionally, it was found that incorporating a broader set of macroeconomic indicators enhances forecast accuracy. These results suggest that ML techniques can serve as valuable complements to econometric models in macroeconomic forecasting, particularly in complex environments such as Bolivia’s.