Research

The views in these papers do not necessarily reflect the views of the Office of the Comptroller of the Currency, the US Department of the Treasury, or any federal agency and do not establish supervisory policy, requirements, or expectations.

Published

Voting and Social Pressure Under Imperfect Information (with Alexander Clark)
International Economic Review, 2019
We develop a model in which costly voting in a large two-party election is a sequentially rational choice of strategic, self-interested players who can reward fellow voters by forming stronger ties in a network formation coordination game. The predictions match a variety of stylized facts, including explaining why an individual’s voting behavior may depend on what she knows about her friends’ actions. Players have imperfect information about others’ voting behavior, and we find that some degree of privacy may actually be necessary for voting in equilibrium, enabling hypocritical but useful social pressure. Our framework applies to any costly prosocial behavior.

Working papers

De-Biasing Models of Biased Decisions: A Comparison of Methods Using Mortgage Application Data
latest draft July 2023 (link coming soon)
Prediction models can improve efficiency by automating decisions such as the approval of loan applications. However, they may inherit bias against protected groups from the data they are trained on. This paper adds counterfactual (simulated) ethnic bias to real data on mortgage application decisions, and shows that this bias is replicated by a machine learning model (XGBoost) even when ethnicity is not used as a predictive variable. Next, several other de-biasing methods are compared: averaging over prohibited variables, taking the most favorable prediction over prohibited variables (a novel method), and jointly minimizing errors as well as the association between predictions and prohibited variables. De-biasing can recover some of the original decisions, but the results are sensitive to whether the bias is effected through a proxy.

Optimal Echo Chambers (with Gabriel Martinez)
latest draft December 2022
When learning from others, people often focus their attention on those with similar views. This is often attributed to flawed reasoning, and thought to slow learning and polarize beliefs. However, we show that echo chambers are a rational response to uncertain information quality, and can improve learning and reduce disagreement. Furthermore, broadening the range of views someone is exposed to can backfire, slowing their learning by making them less responsive to the views of others. A Bayesian decision maker learns about the world by first selecting a set of information sources and then observing a signal from one of them. If some sources are more accurate than others but it’s not clear which, sampling the signals close to one’s prior expectation is more informative, as they are more likely high quality. The optimal echo chamber balances the credibility of views similar to one’s own against the usefulness of those further away.

Social Connections and Racial Wage Inequality
latest draft May 2020 (previously titled Social Capital and Racial Inequality)
How much of the wage gap between black workers and others in the US owes to differences in jobs found through social connections? Panel data from the NLSY79 are used to estimate a job search model in which individual human capital is distinguished from social capital by comparing the wages and frequency of jobs found directly with those of jobs found through friends. Jobs found through friends tend to pay more, but this premium is lower for black workers; the difference can account for 10% of the racial wage gap.

Coordinated Shirking
latest draft November 2018
In the financial crisis of 2008, losses on popular new securitized products far exceeded predictions. This paper studies this episode with a model of technology adoption: a principal tries to induce costly effort from a group of agents charged with vetting new technology. The principal is unwilling to simultaneously punish large groups of agents, so they shirk when coordination is possible. Widely applicable technology expands productive possibilities but also provides an opportunity for coordinated shirking, and can thus lead to widespread production failure. Furthermore, even agents who learn that they are using flawed technology may continue to do so.