Learned Monotone Recurrent Features in Governed Credit Scoring: The Price of the Frame and the Necessity of Macro Conditioning
The research developed a specialized recurrent model for credit scoring that guarantees scores remain non-decreasing with every input variable. The key finding is that the feature's value depends on how strictly you govern its use, and during economic crises, it requires external macroeconomic data to function effectively. This means governance and macro context are essential for using learned temporal features in regulated settings.