Inflation Forecasting in Spain: Combining Machine Learning with Alternative Data Sources

Details

Supervisors

Monti, Francesca

Faculty

Faculté des sciences économiques, sociales, politiques et de communication

Degree label

Master [120] en sciences économiques, orientation économétrie, à finalité approfondie

Abstract

enThis thesis presents a machine learning framework that forecasts Spanish inflation using multiple measures, including monthly, quarterly, annual, and core indices. This approach integrates traditional macroeconomic indicators with alternative data sources, such as real estate transactions, commodity prices, internet search intensity, and sentiment indices. A systematic feature engineering pipeline generates lagged predictors, and model interpretability is ensured through SHAP values, providing a transparent understanding of inflation drivers. LightGBM consistently emerges as the strongest performer among the competing algorithms, with results highlighting the central importance of lagged values in capturing inflation dynamics. Benchmarking exercises are carried out against the Bank of Spain’s macroeconomic projections and random walk models. While the proposed framework outperforms institutional forecasts at certain time horizons, the random walk benchmark remains difficult to surpass, highlighting the challenges of inflation forecasting. Overall, the results show that combining conventional and unconventional information enhances the predictive ability of inflation models. In addition to improving predictive accuracy, using interpretable machine learning methods provides valuable insights into the role of various economic and financial drivers. This dual focus on accuracy and interpretability renders the framework relevant for both academic research and policymakers responsible for anticipating and responding to changes in price dynamics.