Working Papers

Black Box Credit Scoring and Data Sharing [Paper]
Abstract: Should credit scoring algorithms be transparent or opaque? I study this question in a model where a lender uses data shared by borrowers for allocating and pricing credit, and is privately informed about how the algorithm maps data to allocations. I show that revealing the algorithm's parameters exposes it to gaming in the form of strategic withholding of unfavorable information. Under opacity, data withholding emerges as a prudent strategy against the unpredictability of the black box and the risk of credit rationing. The lender's optimal transparency regime maximizes data collection and is socially inefficient as it results in excessive credit rationing.  Algorithmic opacity can be welfare-improving as it reduces the stigma around data withholding, thereby expanding credit access for privacy-concerned borrowers. I analyze the distributional impacts of recent algorithmic transparency regulations and offer policy recommendations.

Market Information in Banking Supervision: the Role of Stress Test Design [Paper]
(with Haina Ding and Alexander Guembel)
Abstract: The Basel Committee views market discipline as complementing banking supervision. We study how a supervisor should design a bank stress test when aiming to foster information production by stock markets and show that full disclosure is suboptimal. The optimal test pools banks around the funding threshold into at most two coarse grades and deliberately misclassifies some of them. It is lenient when the funding threshold is low and conservative when it is high. The results are robust to trading in alternative securities. Unlike existing theories, ours does not require a conflict of interest between the supervisor and investors.

Preemptive Competition, Risk Governance, and Artificial Intelligence [Paper]
(with Matthieu Bouvard and Samuel Lee)
Abstract: Preemptive competition over time-sensitive opportunities creates a contractual externality that undermines risk governance. Firms delegate search and vetting to agents; faster execution by rivals increases preemption risk, raising the agency rents required to incentivize compliance. This interaction leads to a deal-driven front-office culture where internal controls are bypassed, even if firms fully internalize their losses. Compensation regulation promotes risk governance by slowing the competitive race. While artificial intelligence can reduce human agency frictions, its acceleration of search efficiency may paradoxically exacerbate externalities. Consequently, laissez-faire outcomes under AI can remain inferior to regulated regimes with human agents.