CEMLA Course: Modern Machine Learning for Macroeconomic Forecasting

The Modern Machine Learning for Macroeconomic Forecasting course, taught by Pablo A. Guerrón-Quintana (Boston College) via videoconference from March 30 to April 1, 2026, bridged classical econometric forecasting with modern machine learning (ML) approaches, focusing on practical applications for central banking. As a starting point, the course showed how foundation models and automated ML have revolutionized forecasting workflows: models like Chronos-2 enable zero-shot forecasting on new datasets without training, while TabPFN achieves state-of-the-art performance on small tabular datasets in seconds. These advances allow central banks to improve forecast accuracy, reduce modeling time, and better quantify uncertainty for policy decisions.

Throughout this intensive course, the emphasis was placed on understanding the fundamental tradeoffs between classical and ML methods, on hands-on implementation with state-of-the-art tools (AutoGluon, Chronos-2, and TabPFN), and on production deployment considerations.