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

Background and Context

Research Focus

The study examines how the adoption of machine learning technology in credit screening affects different racial and ethnic groups in the U.S. mortgage market.

Data Source

Analysis uses data from 9.37 million U.S. mortgages originated between 2009 and 2013, combining Home Mortgage Disclosure Act (HMDA) data with McDash mortgage servicing data.

Methodology

Researchers compare traditional logistic regression models with Random Forest machine learning models to predict mortgage default probabilities across different demographic groups.

Higher Default Rates and Credit Characteristics for Minority Borrowers

  • Shows baseline default rates across different racial/ethnic groups from the dataset
  • Black and Hispanic borrowers have significantly higher default rates than Asian and White non-Hispanic borrowers
  • These differences in default rates provide context for understanding the impact of different prediction models

Improved Prediction Accuracy with Machine Learning Models

  • Compares performance metrics between traditional Logit and Random Forest models
  • Random Forest shows consistent improvements across all performance measures
  • Demonstrates the superior predictive power of machine learning approaches

Disproportionate Impact on Minority Borrowers Under Machine Learning

  • Shows how interest rate spreads change when switching from traditional to machine learning models
  • Black and Hispanic borrowers face higher interest rates under machine learning
  • Asian and White non-Hispanic borrowers see slight reductions in rates

Increased Rate Dispersion Under Machine Learning

  • Illustrates how rate dispersion changes across racial groups under different models
  • Machine learning leads to greater rate dispersion for all groups
  • Minority borrowers experience the largest increases in rate dispersion

Impact on Loan Acceptance Rates

  • Shows loan acceptance rates under the machine learning model
  • Significant disparities in acceptance rates persist across racial groups
  • Black borrowers face the lowest acceptance rates despite improved overall prediction accuracy

Contribution and Implications

  • Machine learning improves overall prediction accuracy but may exacerbate existing racial disparities in mortgage lending
  • The technology leads to greater dispersion in interest rates, particularly affecting minority borrowers
  • Findings suggest need for policy consideration regarding fairness in automated lending decisions

Data Sources

  • Default rates and borrower characteristics from Table I
  • Model performance metrics from Table III
  • Interest rate changes and dispersion from Table VII
  • Acceptance rates from Table VII
  • All visualizations use actual data from the published tables in the paper