A new framework, Conformal Risk Sharing, provides a mathematically certified method for allocating financial burdens resulting from rare adverse events across multiple participants, ensuring that no individual is made materially worse off while bounding aggregate harm.
This research addresses the challenge of fairly sharing the financial impact of rare adverse events. The authors formalize this as the Certified Allocation Problem: finding a redistribution rule and obligation caps for all agents without distributional assumptions, while verifying no participant is negatively affected. The proposed solution, Conformal Risk Sharing, achieves this by combining an interpretable sharing policy with split conformal calibration. The framework tunes the sharing intensity on training data and uses held-out calibration data to produce distribution-free per-agent guarantees. Experiments confirm that this method effectively reduces extreme obligations for high-risk agents while controlling the harm experienced by others, validated across synthetic and real-world datasets.