Are credit scores gender-neutral? Evidence of mis-calibration from alternative and traditional borrowing data

Abstract

This study investigates whether credit scoring systems inherently disadvantage women within the subprime borrowing context, where alternative credit data is frequently used. While recent advancements in machine learning and alternative data usage promise greater fairness and accuracy in lending, our findings highlight systemic biases embedded within current credit scoring models. Using a comprehensive sample of alternative borrowers, our analysis reveals that women consistently receive lower credit scores than men, despite exhibiting lower default rates and controlling for extensive credit risk variables. Furthermore, credit scores demonstrate systematically reduced predictive accuracy for women compared to men, underscoring gender biases embedded within these scoring systems. These findings emphasize the urgent need to recalibrate credit scoring models to enhance fairness, accuracy, and financial inclusivity.