SHARC: SHAP interpretability in ICAAP and CCAR regulatory capital

In the field of banking supervision, regulatory frameworks such as ICAAP and CCAR require full transparency in the estimation of regulatory capital. However, the adoption of advanced machine learning models, such as Gaussian process regression (GPR), encounters the well-known black box problem: although their predictive accuracy surpasses traditional parametric methods, their lack of interpretability hinders auditing by supervisors. To address this challenge, SHARC emerges, an explainability framework based on SHAP values that decomposes Stressed Value at Risk (SVaR) into components linked to mean and volatility, enabling an auditable link between macro scenarios —such as wars, climate risks, or technology bubbles— and capital requirements. This approach not only satisfies the criteria of local accuracy, consistency, and efficiency from cooperative game theory but also demonstrates that, under extreme conditions, the directional magnitude of losses (mean return component) weighs more than baseline volatility, thus guiding hedging strategies and capital limit setting. For entities seeking to implement such solutions, having a technology partner that offers artificial intelligence for businesses with an ethical and auditable approach is key. At Q2BSTUDIO we develop custom software that integrates explainable models, also leveraging our AWS and Azure cloud services to scale intensive computing infrastructures. Our offering also includes customized AI agents, business intelligence services with Power BI to visualize risk factors, and cybersecurity to protect regulatory data. Thus, we combine custom applications, artificial intelligence, and cloud expertise so that financial institutions meet FRTB, ICAAP Pillar 2, and CCAR standards without sacrificing innovation.