SHARC: SHAP-Based Interpretability in Machine Learning Risk Models for Regulatory Capital under ICAAP and CCAR

Researchers have developed a framework called SHARC to apply SHAP-based interpretability to machine learning models used for calculating regulatory capital. By providing auditable explanations for these non-parametric models, the method aims to help financial institutions meet strict oversight requirements for risk management under ICAAP and CCAR standards.