🤖 AI Summary
This study addresses the challenge of accurately attributing prediction discrepancies between CCAR and CECL forecasting processes to specific input changes, without relying on the order of input substitutions. Framing attribution as a cooperative game-theoretic problem, it presents the first systematic evaluation of multiple attribution methods—including Exact Shapley values, Hierarchical Shapley values, Integrated Gradients, Gradient SHAP, Permutation SHAP, and Kernel SHAP—in real-world production settings. By analyzing these methods across dimensions such as allocation properties, computational cost, implementation requirements, and inherent limitations, the work proposes a practical selection framework that balances interpretability, reproducibility, and engineering feasibility. This framework offers financial institutions actionable guidance for choosing attribution techniques that support transparency and governance in regulatory forecasting systems.
📝 Abstract
Forecasting systems used in the Comprehensive Capital Analysis and Review (CCAR) and Current Expected Credit Losses (CECL) processes combine portfolio data, macroeconomic scenarios, model specifications, business assump- tions, and management adjustments. When the forecast changes from one run to the next, practitioners need an attribu- tion that reconciles to the total change without depending on an arbitrary sequence of input replacements. This paper formulates forecast-gap attribution as a cooperative game and examines several approaches: the exact Shapley value, hierarchical or nested Shapley values, Integrated Gradients, Gradient SHAP, Permutation SHAP, and Kernel SHAP. We compare their allocation rules, computational costs, implementation requirements, and limitations in production forecasting systems. The analysis provides a practical framework for choosing an attribution method according to the number and type of inputs, the feasibility of hybrid forecast runs, and the need for interpretability, reproducibility, and governance.