Linking Climate Extremes to Financial Risk in the Era of Climate Change

Objectif

Climate extremes are becoming more frequent and severe, creating systemic risks for economic stability and financial markets. Current approaches often examine climate prediction and financial risk separately, leaving a critical gap in understanding how physical hazards translate into market impacts. STOCKCLIM addresses this challenge by developing the first integrated framework to detect, project, and quantify the financial impacts of extreme climate events. The project combines recent advances in machine learning with econometric modeling to map the entire pathway from hazard to market response. Capsule Networks will be trained on reanalysis (ERA5), climate simulations (CESM2-LENS2), and disaster records (EM-DAT) to detect heatwaves, floods, and tropical cyclones. The trained models will then be applied to future climate scenarios (SSP3-7.0) to project shifts in event frequency and intensity. Detected events will be aligned with financial datasets, including stock indices, sector-level returns, and volatility measures. Econometric tools, such as event studies, VAR, GARCH, and GARCH-MIDAS, will be used to estimate short- and medium-term market impacts and develop scenario-based climate stress tests for key financial sectors. Expected outputs include open-source machine learning models, catalogs of historical and projected extremes, high-impact publications, and a lightweight visualization dashboard to explore climate-finance interactions. Together these will provide reproducible, event-level evidence of how extremes influence markets and which sectors are most vulnerable. By bridging climate science, machine learning, and financial econometrics, STOCKCLIM delivers both scientific innovation and actionable insights. Its integration of cutting-edge AI, climate modeling, and financial analysis will advance climate resilience, inform risk management, and train a new generation of interdisciplinary expertise, directly supporting the goals of the MSCA program.