Pranav Garg

Lenders have long believed that personality and economic preferences matter for repayment, yet these characteristics remain difficult to measure at scale. We link LinkedIn profile images with credit records for over 800,000 U.S. borrowers and show that facial embeddings extracted using machine learning predict delinquency beyond credit scores, income, and education. Facial information raises the predictive power of credit score by one-third, with the largest gains among younger, lower income, less educated, and thin-file borrowers. These gains remain large within a demographically homogeneous subsample and after ranking predictions within sex-by-ethnicity groups. To understand why faces carry predictive power, we survey over 10,000 participants and train models to predict self-reported personality and economic preferences from their photographs. The resulting measures help account for facial embeddings’ predictive power for repayment: conscientiousness and patience predict lower delinquency, while risk-taking predicts higher delinquency. Our approach offers a scalable way to study the role of personality and preferences in credit markets. Incorporating facial information into credit screening could also expand credit access: in lending simulations, access for thin-file borrowers rises from 9% to 42% at a fixed portfolio delinquency rate. However, the growing use of facial information in industry screening raises ethical concerns about statistical discrimination and individual autonomy.

• Presented at (including upcoming): AFA 2027, ABS Household Finance Conference 2026, SFS Cavalcade 2026, 17th Annual FSU Truist Beach Conference 2026, Adam Smith Workshop 2026, FRA Early Ideas Session 2025, Boulder Summer Conference on Consumer Financial Decision Making 2025.

We show that heterogeneous equity flows across financial intermediaries amplify monetary policy transmission to asset prices. We find that mutual funds with performance-sensitive retail investors experience outflows after contractionary monetary shocks, triggering forced equity sales. Banks have funding that is insensitive to recent performance and therefore absorb these sales and provide liquidity. This intermediary-level heterogeneity aggregates meaningfully: stocks held by more performance-sensitive investors exhibit larger price declines following monetary tightening. To explain these patterns, we develop an intermediary asset pricing model where heterogeneous intermediaries manage portfolios for investors with performance-sensitive flows, highlighting a novel mechanism of monetary transmission to equities. 

• Presented at: SGF Conference Zurich 2026, MFA Annual Meeting 2026, 19th International Behavioral Finance Conference at the University of Chicago 2025, AFA Posters 2025, Midwest Macro Meetings at Purdue University 2024.

• Award: Doctoral Symposium Outstanding Paper, MFA Annual Meeting 2026.

Firms with high industry market shares are insulated from monetary policy. We show that following a contractionary shock, leaders’ cost of capital rises by less than that of lower-share competitors in the same industry: 13 basis points less per 100 basis points of shock for the top 2%, plus their investment and R&D fall by less. We outline a mechanism of earnings cyclicality, which leads to policy insulation and is independent of financial frictions. In our model, firms spend on customer relationships and long-term contracts to lock in a share of revenue that does not move with the cycle, more productive firms lock in more, and they are less exposed to the business cycle. Consistent with this, we show leaders’ revenue betas are about a third of those of bottom-half firms and that earnings-call language about recurring and contracted revenue tends to rise with market position. Within firms, more such language predicts also lower future cyclicality. Tightening thus shifts competitive advantage toward leaders, perhaps permanently.

• Presented at: AFA Posters 2026, Midwest Macro Meetings at Federal Reserve Bank of Cleveland 2025.