Research

While prior studies find that automated underwriting outperforms manual underwriting, I show that there is significant heterogeneity in the adoption of automated underwriting both within and across lenders. To explain this heterogeneity, I examine the performance of automated underwriting systems under conditions of heightened data uncertainty caused by the COVID-19 pandemic. Using a combination of difference-in-differences and regression discontinuity designs, I estimate the impact of this unprecedented shock on the performance of automated underwriting in the auto loan market.

My findings show that the performance of automated underwriting, as measured by ex-post default rates, deteriorated substantially relative to human underwriters during the pandemic period. The effect is particularly pronounced among higher-risk segments of borrowers, whose income and employment were more likely to be disrupted by the pandemic. Together, these results highlight the limitations of automated underwriting systems when faced with unprecedented shocks outside the scope of their historical training datasets, underscoring the continued relevance of human underwriters in the auto lending industry.

Using a novel dataset on auto leases and a tax policy change by the state of Georgia, we estimate how tax savings on financial products are passed through to consumers and study the determinants of the pass-through rate. We find that (1) auto dealers (not lenders) capture a substantial portion of this tax subsidy and (2) consumers spend about 50% of their subsidy to upgrade and lease a more expensive vehicle. In contrast to prior literature on consumer credit markets, we find no evidence that demand factors including credit score and past experience affect this pass through rate. Our findings suggest that the market structure of the auto lease market is the main driver of the heterogeneity in the pass-through rate.

How does the interest rate on a loan affect its likelihood of default after disbursement? What economic forces drive such a relationship? We examine these questions in the context of the indirect auto loan market. Using lender-specific discontinuities in auto loan interest rates as sources of quasi-exogenous variation, we find that a 100 basis point increase in interest rates is associated with a 41 basis point increase in the auto loan default rate. The increase in the default rate is concentrated among liquidity-constrained borrowers. We find no evidence of higher default sensitivities among borrowers with lower default costs or higher strategic default motives. Our results suggest that the ability to pay — and not the incentive — is the predominant driver of the ex-post relation between default and the auto loan interest rate.