Publications
Publications
Abstract
This paper explores which correlated equilibrium outcomes are robust to players having the option to publicly and verifiably disclose their private information before playing the game. I focus on outcomes that are disclosure-proof: those that can be induced such that no player is incentivized to disclose their signal. Since signals in this environment do not carry payoff-relevant information, disclosing only limits the ability to coordinate agents off-path. I show that an outcome is disclosure-proof if and only if it satisfies obedience and certain interim participation constraints. Each player must expect to earn more after observing any of their signals than they would in the worst correlated equilibria outcome where they play independently of the other players. For two-player games, this is equivalent to the payoff of their worst Nash outcome. When players can disclose over multiple rounds, all implementable outcomes will be disclosure-proof.
Bad Reputation in Repeated Cheap Talk
I examine a class of continuous-state cheap talk games where the receiver is uncertain about the preferences of the sender. Only one of the two sender types (a "good" type) can meaningfully communicate with the receiver; the other will only induce the highest possible action (a "bad" type). I study the two-period version of this game, where the state changes each period but the sender's type is persistent. I show that high levels of reputation are detrimental to communication, but due to a novel mechanism: the good sender chooses to send a revealing message in every state, which causes the bad type to deviate from the pooling message. This causes collapses in communication to be discontinuous with respect to the level of reputation. At lower levels of reputation, communication is improved only if the good sender is misaligned with the receiver; at sufficiently high levels, only babbling outcomes can be sustained. Further, a receiver may prefer to interact with a more biased good sender type, as improved communication will be feasible. I show similar results hold in the finite-horizon case.
Signal Disclosure in Information Design
This paper studies the information design environment where the designer is restricted to using messages that players can publicly and verifiably disclose. I employ a belief-based approach to characterizing the designer's optimal payoff among disclosure-proof outcomes, utilizing the methods of Mathevet, Perego & Taenva (2020). Unlike in the standard information design environment, the designer's optimal outcome may require an indirect information structure. It is without loss for the designer to send messages specifying a player's full beliefs and a menu of contingent action recommendations. I apply these results to a class of games which I call leader-follower games. If a leader-follower game has positive spillovers, then a designer with increasing preferences can do no better than public information. If the game does not have positive spillovers, private information may still be valuable under signal disclosure.