Please rotate your device to landscape mode to view the charts.

Predictably Unequal? The Effects of Machine Learning on Credit Markets

Journal: Journal of Finance

Date: 20220201

Author: Fuster, Andreas; Goldsmith-Pinkham, Paul; Ramadorai, Tarun; Walther, Ansgar

Abstract:
Innovations in statistical technology in functions including credit-screening have raised concerns about distributional impacts across categories such as race. Theoretically, distributional effects of better statistical technology can come from greater flexibility to uncover structural relationships or from triangulation of otherwise excluded characteristics. Using data on U.S. mortgages, we predict default using traditional and machine learning models. We find that Black and Hispanic borrowers are disproportionately less likely to gain from the introduction of machine learning. In a simple equilibrium credit market model, machine learning increases disparity in rates between and within groups, with these changes attributable primarily to greater flexibility.

Link: Google Scholar

Key Findings

Machine Learning Improves Default Prediction

Random Forest model outperforms traditional Logit models in predicting mortgage defaults, with a 5.1% improvement in precision score and 14.3% improvement in R-squared value

Unequal Distribution of Benefits

Benefits of improved prediction accuracy disproportionately favor White non-Hispanic and Asian borrowers, while Black and Hispanic borrowers see less favorable outcomes

Higher Rate Dispersion for Minorities

Machine learning technology leads to greater dispersion in interest rates, particularly affecting minority borrowers with increases of 8-10 basis points vs 5-6 for non-minority borrowers

Model Performance Comparison

  • Random Forest model shows superior performance across all metrics
  • Greatest improvement seen in R-squared value (14.3% increase)
  • ROC-AUC shows 0.8% improvement with Random Forest

Acceptance Rates by Race/Ethnicity

  • Clear disparities in acceptance rates across racial groups
  • Asian and White non-Hispanic borrowers have highest acceptance rates
  • Black borrowers face lowest acceptance rates under both models

Interest Rate Dispersion Effects

  • Interest rate dispersion increases progressively across racial groups
  • Black borrowers face 43% higher rate dispersion than Asian borrowers
  • Pattern suggests amplified pricing uncertainty for minority borrowers

Contribution and Implications

  • First comprehensive analysis of distributional effects of machine learning adoption in mortgage lending
  • Demonstrates that technological advancement can amplify existing inequalities despite improving overall prediction accuracy
  • Highlights need for careful consideration of fairness in deployment of new lending technologies

Data Sources

  • Performance comparison chart based on Table III - comparing model performance metrics
  • Acceptance rates visualization derived from Table VII - showing equilibrium outcomes across racial groups
  • Interest rate dispersion chart constructed from Table VII columns 5-6 - standard deviation of SATO data