SHARC: SHAP-Based Interpretability in Machine Learning Risk Models for Regulatory Capital under ICAAP and CCAR
arXiv:2607.05484v1 Announce Type: cross Abstract: The adoption of non-parametric machine learning models for regulatory capital estimation introduces a fundamental governance challenge: the inability to explain model outputs in a manner auditable by supervisory bodies. This 'black box' problem remains a major barrier to the adoption of Gaussian Process Regression (GPR) and related ML architectures in ICAAP and CCAR workflows despite their predictive advantages over traditional parametric approaches. This paper addresses this barrier through SHARC (SHAP for Regulatory Capital), an explainabilit
Published in 2026, this paper addresses the immediate challenge of integrating advanced ML models into highly regulated financial sectors, reflecting ongoing efforts to balance innovation with oversight.
Regulatory bodies and financial institutions face a dilemma: leverage AI's predictive power or adhere to stringent explainability requirements. This research offers a pathway for adoption, potentially unlocking significant efficiency and accuracy gains in risk management.
The ability to explain 'black box' AI models like GPR through SHAP-based methods allows financial institutions to adopt more sophisticated risk models for regulatory capital estimation, meeting both performance and auditability demands.
- · Financial institutions (banks, insurers)
- · AI/ML providers to finance sector
- · Regulatory technology (RegTech) firms
- · Data scientists and ML engineers
- · Traditional parametric modeling approaches
- · Audit firms without ML expertise
- · Institutions slow to adopt explainable AI
- · Legacy risk management software vendors
Increased adoption of explainable AI in regulated financial industries for risk modeling.
Enhanced accuracy and efficiency in regulatory capital estimation, potentially optimizing capital allocation.
Broader acceptance and integration of advanced AI across other highly regulated sectors due to proven explainability frameworks.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG