Job Market Paper: Learning about the Self in a Discrete Environment
Abstract: This paper studies a Bayesian agent who learns about the precision of his own private signal rather than an external fundamental. Each period, he observes a private signal that is informative about an external state and decides whether to invest. The precision is unknown, and the agent's belief regarding it is his confidence: a more confident agent believes that his signal is more likely to be precise. The state is drawn i.i.d. in each period according to a fundamental parameter referred to as optimism, which captures how likely the drawn state is to favor investment. The learning process is determined by actions: investment reveals the state and the agent actively learns from whether the signal matched it, whereas without investment, he can still learn passively from the signal realization alone. Passive learning provides less information about signal precision than active learning and is therefore slower. I show that whenever confidence is pivotal for future investment decisions, the agent may rationally experiment by investing at an expected current loss. Under a correctly specified prior, learning converges to the truth almost surely and welfare achieves the first best in the limit. Experimentation accelerates this process. Under a misspecified prior, however, learning can lead to a suboptimal policy and result in persistent welfare loss. Experimentation can push confidence toward the wrong policy and exacerbate the welfare loss.
Working Papers:
Costly Information and the Elimination of Strategic Delay in Observational Learning
Abstract: This paper studies a two-player, two-period model in which the players choose both whether to acquire private information at a cost and when to act. At the beginning of the game, each player decides whether to buy a private signal that is informative about the state then observes the signal if he purchases. In the first period, each player decides whether to make an irreversible, fixed-size investment. If a player does not invest in the first period, he has another opportunity to invest
in the second period. I analyze a perfect Bayesian equilibrium where players optimally choose their information acquisition and timing of actions. In the unique mixed-strategy equilibrium, the players randomize over information acquisition. A player who buys the signal acts in the first period and never delays the investment choice, whereas a player who does not buy the signal always delays the investment choice. Furthermore, I compare this equilibrium outcome with that of a model whose information is generated exogenously. I show that the endogenous information structure decreases overall investment likelihood in the equilibrium, which increases the players’ joint surplus when the state is bad.
Work in Progress:
“Social Learning with Endogenous Timing and Payoff Externality” (joint with Ayush Gupta)
“Optimal Delegation in a Social Learning Context”
“Pricing and Competition in a Semi-Credence Goods Market”