Events

Professor Shimon Kogan
Associate Professor of Finance
Arison School of Business
Reichman University

Lenders have long believed that personality matters for repayment, but personality is difficult to measure at scale. We merge LinkedIn profile images with credit records for over 800,000 U.S. borrowers and show that facial embeddings extracted using machine learning algorithms predict delinquency beyond credit scores, demographics, income, and education. Adding facial information to a credit-score-only model raises predictive power by 33% and by over 100% among borrowers with thin credit files. To explore mechanisms, we survey over 10,000 people about their personality and preferences and train models to predict these self-reported traits from their photographs. Applied to borrowers, the resulting photo-derived measures of Big 5 personality traits, patience, and risk-taking retain over 70% of the facial embeddings’ power to explain delinquency. The gains from facial information are largest where credit scores are least informative: among younger, lower-income, and thin-file borrowers. In simulated approval decisions, adding facial information to credit scores expands credit access for these underserved borrowers without increasing delinquency among those approved. The scalability of our approach enables new academic insights into the role of personality and preferences in credit markets, but its growing use in industry screening raises ethical concerns about statistical discrimination and individual autonomy.