Central Bank of Luxembourg study finds climate data and machine learning improve regional growth forecasts mainly for agriculture

The Central Bank of Luxembourg has published a study on how extreme heat and drought affect short term regional economic activity across the European Union, finding that climate augmented machine learning models materially improve forecasting accuracy for climate sensitive sectors, especially agriculture. The paper covers gross value added per capita growth at NUTS-3 level across 1,117 European Union regions from 2002 to 2022 for agriculture, industry and manufacturing, and concludes that adding climate information improves predictions only in non-linear models such as Random Forest and XGBoost, not in linear specifications. The analysis combines annual economic data with high frequency climate indicators from the Copernicus European Drought Observatory, including heatwaves, meteorological, hydrological and agricultural drought, as well as compound heatwave-drought events and spillovers from adjacent regions. Forecast gains are strongest in agriculture, more moderate in industry and largely absent in manufacturing. Using XGBoost to simulate a 2022-like compound heatwave and drought scenario, the study estimates average losses in real GDP per capita growth of 4.54 percentage points in agriculture, with the largest impacts in Eastern Europe, and average reductions of 0.75 percentage points in industrial real gross value added per capita growth, while manufacturing remains broadly insensitive at minus 0.11 percentage points on average. The paper states that the views expressed are those of the authors and not necessarily those of the Central Bank of Luxembourg or the Eurosystem.