Working Papers
Disagreement, Subjective Uncertainty, and the Stock Market (with Jingoo Kwon and Younggeun Yoo)
Abstract: We propose a method to separately quantify cross-sectional disagreement and subjective uncertainty at firm and fiscal quarter levels, and investigate how they jointly affect stock market volatility, returns, and trading volume. While disagreement and subjective uncertainty are often treated interchangeably in empirical research, we find that conflating the two may lead to misleading or even opposite results. Subjective uncertainty is positively correlated with volatility when disagreement is low, but this relationship reverses and the correlation becomes negative when disagreement is high. These results suggest the need for caution when using volatility as a proxy for agents' uncertainty, especially during periods of heightened disagreement. We also find that disagreement and subjective uncertainty have opposite yet interactive effects on returns and trading volume. Stocks with higher disagreement earn lower returns and exhibit higher trading volumes, with these effects amplified 2-3 times when subjective uncertainty is low. Conversely, stocks with higher subjective uncertainty earn higher returns and experience lower trading volumes, with these effects similarly amplified when disagreement is high. We provide a theoretical framework and perform numerical simulations to explain our empirical findings.
Signals from SMS: Alternative Data and Machine Learning in Credit Scoring (with Kenneth Ryu and Jaehyeok Shin)
Abstract: Using data on 26 million loan applications from a FinTech lender in India, we study how alternative data and machine-learning algorithms can improve credit underwriting. We show that transaction records and credit histories can be extracted from text messages and used to assess applicants’ creditworthiness. The alternative credit scoring system predicts consumer default 33% more accurately than traditional credit scores, with gains attributable to both algorithmic improvements and the use of alternative datasets. This system expands credit access by approving loans for over 50,000 new-to-credit applicants with no bureau score and no prior loans. However, relative to the traditional system, the benefits of this technological shift are skewed toward financially advantaged groups.
Publications
(with Minwoo Kang, Suhong Moon, Ayush Raj, Joseph Suh, David Chan, and John Canny), Conference On Language Modeling (COLM) 2025
Abstract: Large language models (LLMs) are increasingly capable of simulating human behavior, offering cost-effective ways to estimate user responses to various surveys and polls. However, the questions in these surveys usually reflect socially understood attitudes: the patterns of attitudes of old/young, liberal/conservative, as understood by both members and non-members of those groups. It is not clear whether the LLM binding is deep, meaning the LLM answers as a member of a particular in-group would, or shallow, meaning the LLM responds as an out-group member believes an in-group member would. To explore this difference, we use questions that expose known in-group/out-group biases. This level of fidelity is critical for applying LLMs to various political science studies, including timely topics on polarization dynamics, inter-group conflict, and democratic backsliding. To this end, we propose a novel methodology for constructing virtual personas with synthetic user "backstories" generated as extended, multi-turn interview transcripts. This approach is justified by the theory of narrative identity which argues that personality at the highest level is constructed from self-narratives. Our generated backstories are longer, rich in detail, and consistent in authentically describing a singular individual, compared to previous methods. We show that virtual personas conditioned on our backstories closely replicate human response distributions (up to an 87% improvement as measured by Wasserstein Distance) and produce effect sizes that closely match those observed in the original studies of in-group/out-group biases. Altogether, our work extends the applicability of LLMs beyond estimating socially understood responses, enabling their use in a broader range of human studies.
A Randomized Trial of Behavioral Nudges Delivered Through Text Messages to Increase Influenza Vaccination Among Patients With an Upcoming Primary Care Visit [Manuscript] [Online Appendix]
(with Mitesh Patel, Katherine Milkman, and 43 co-authors), American Journal of Health Promotion, 2023, 37(3): 324-332
Test-based Accountability and the Effectiveness of School Finance Reforms [Manuscript] [Online Appendix]
(with Christian Buerger and John Singleton), AEA Papers and Proceedings, 2021, 111: 455-59
Abstract: Recent literature provides new evidence that school resources are important for student outcomes. This paper examines whether school accountability systems that incentivize performance (such as No Child Left Behind) raise the efficiency with which additional resources get spent. We leverage the timing of school finance reforms to compare funding impacts on student test scores between states that had accountability in place at the time of the reform and states that did not. The results show that finance-reform-induced increases in student performance are driven by those states where the reform was accompanied by the presence of test-based accountability.
Media coverage: Thomas B. Fordham Institute
Work in Progress
Anticipatory Spending (with Scott Baker, Michael Gelman, and Lorenz Kueng)
Firm Investments and Human Capital Growth (with Bryan Seegmiller, Yufeng Wu, and Miao Ben Zhang)
Access to and Gains from Selective Education (with Ying Shi and John Singleton)