Study Finds Screening Tech Errors by Banks and Fintechs Can Reshape Credit Competition Through “Pool Externality”

Maryam Farboodi, Péter Kondor, and Pablo Kurlat examine how competition among lenders changes when banks and fintech firms adopt new tools to screen borrowers—and when those tools make mistakes in similar or different ways. Their focus is not simply on whether screening gets “better,” but on how the pattern of errors and the ability to invest in screening technology reshape the entire structure of credit markets.

At the center of the paper is a dynamic that economists call a “pool externality.” In many credit markets, applicants who are denied by one lender do not disappear; they apply again somewhere else. That means each lender’s screening rules and pricing decisions alter the mix of borrowers that competitors will see later. When a lender tightens screening and rejects more people, those rejected borrowers flow into other lenders’ applicant pools—changing the risk profile those competitors face. Conversely, when a lender expands credit, it may siphon off some of the easier-to-approve applicants and leave behind a tougher pool for others.

The authors build a model in which lenders compete while choosing their screening technologies endogenously. The key distinction is whether screening errors are independent across lenders or correlated. Independent errors mean one lender’s misclassification of a borrower is largely unrelated to another lender’s misclassification. Correlated errors mean lenders tend to be wrong about the same people—an increasingly relevant possibility if firms rely on similar AI models or draw from overlapping datasets used throughout the financial system.

When screening errors are independent, the model delivers a relatively simple market outcome: credit is organized into a single segment. In that environment, interest rates move gradually as lenders adopt better screening tools. Counterintuitively, lenders with more accurate technologies may end up serving borrowers who are more adversely selected, pushing interest rates upward in a smooth way rather than creating sharp divides across products.

The picture changes when errors are correlated. If lenders’ models tend to fail in the same direction, the market can split even when firms look identical at the outset. Instead of converging to one segment, lenders specialize into distinct parts of the market. The resulting interest-rate pattern resembles a “hockey stick”: some segments feature nearly uniform interest rates even though borrower quality varies widely within the segment, while other segments contain borrowers of similar quality but face a broad range of interest rates. The authors argue that these segments mirror recognizable real-world categories: traditional relationship-style banking, fintech-style lending, and high-rate lenders that extend credit more indiscriminately.

The paper also explores how changes in technology and regulation that affect the cost of screening can reshape access to credit. Lower screening costs—driven, for example, by frontier AI advances—or policy reforms such as forms of open banking that expand data access can have uneven effects across the market’s segments. Depending on how these changes alter lenders’ incentives and the degree of correlation in screening errors, they may expand credit for some groups while making outcomes worse for borrowers who are already financially excluded.

Farboodi, Kondor, and Kurlat acknowledge comments and input from a wide range of prominent economists, and they note that the views in the paper are their own and not necessarily those of the National Bureau of Economic Research.

Why it matters

As lenders increasingly rely on similar AI-driven screening systems and shared data sources, the risk is not only biased decisions but also market-wide shifts in who gets credit and at what price. The research suggests that policies promoting data sharing or cheaper screening could either improve inclusion or deepen exclusion, depending on how they change competitive dynamics. The next key question for regulators and firms is whether new technologies diversify screening judgments—or make them more correlated, pushing markets toward sharper segmentation.