With Doug Bernheim
American Economic Review, 107(2), 2017
Although economists have made substantial progress toward formulating theories of collusion in industrial cartels that account for a variety of fact patterns, important puzzles remain. Standard models of repeated interaction formalize the observation that cartels keep participants in line through the threat of punishment, but they fail to explain two important factual observations: first, apparently deliberate cheating actually occurs; second, it frequently goes unpunished even when it is detected. We propose a theory of equilibrium price cutting and business stealing in cartels to bridge this gap between theory and observation.
With Annie Liang
Many organizations, such as banks and insurers, determine what services to offer based on a perceived quality of the recipient, e.g. their creditworthiness. With new access to detailed data on individual consumers, organizations are increasingly estimating quality not only from a given consumer's interactions with the organization, but also from interactions with comparable individuals. What are the consequences for consumer incentives to exert effort in their interactions with the firm, e.g. to maintain a good credit rating? To answer this question, we study a multiple-agent career concerns model in which agents choose whether to interact with a principal, who provides a service and aggregates data across all participating agents. Individuals' interactions create an informational externality on others, shaping participation rates and effort provision in equilibrium. We show that whether data sharing is welfare-improving depends crucially on how the actions of individuals affect inferences about related consumers, specifically on whether information across consumers is "complementary" or "substitutable."
With Aaron Kolb
We study how an organization dynamically screens an agent of uncertain loyalty whom it suspects of committing damaging acts of undermining, for instance leaking sensitive information or sabotaging production. The organization's screening tool is the agent's access to sensitive information, i.e. the stakes of the relationship, governing both productivity and the harm from undermining. A disloyal agent strategically chooses when and how intensively to undermine, with undermining stochastically detected at a rate proportional to its intensity. When the organization can commit, it optimally guides stakes by holding them constant for a time, then gradually escalating them, and finally jumping them to their maximal level. This stakes path is also the unique equilibrium outcome when the organization cannot commit, and the disloyal agent's unique equilibrium undermining path exhibits variable, non-monotonic intensity. In an information design microfoundation, the optimal stakes path is implementable via an inconclusive contradictory news process.
With Rishabh Kirpalani
Motivated by stylized facts in the market for entrepreneurial fundraising, we build a two-firm model of investment timing with endogenous information acquisition to study the interaction between investment delay and free-riding. There are three perfect Bayesian equilibria of the model. In the unique symmetric equilibrium, both firms investigate the project simultaneously, while in the remaining asymmetric equilibria one firm leads the other in obtaining information and investing. Investment delay and free-riding arise only in the asymmetric equilibria, but nonetheless these equilibria improve aggregate welfare over or even Pareto-dominate the symmetric equilibrium when firms are patient.
I analyze how a firm should elicit advice from an expert on terminating a project with a stochastic lifespan. The firm cannot directly observe the project's lifespan, but imperfectly monitors its state by observing incremental output. The expert directly observes the state of the project, but collects on-the-job benefits and so prefers to prolong operation as much as possible. He possesses no capital and enjoys limited liability, preventing efficient trade even in case the expert has no initial private information. The optimal long-term contract involves a stochastic project deadline and a completion bonus to the expert which declines as the deadline approaches. The deadline is responsive to good and bad runs of output, and exhibits variable output sensitivity over the lifetime of the project, in particular becoming more sensitive the closer the project is to termination. Elicitation of expert advice increases the ex post operational efficiency of the project, but asymmetrically - late terminations are completely resolved, while early terminations are mitigated but not entirely eliminated. These features are robust to extensions in which expert has limited initial capital, can be replaced, or can have his on-the-job benefits dissipated by busywork.