Coalition-Proof Disclosure

Abstract

We study generalized verifiable disclosure by an informed sender with type-independent preferences. We first characterize all perfect Bayesian equilibria as individually rational partitions of the sender's types into coalitions ordered by payoff. An equilibrium survives our coalition-proofness criterion if no group of types can profitably coordinate an announcement that the receiver will interpret as coming precisely from the types willing to join it. Coalition-proof equilibria are partitions that are greedy: at each step, a coalition of the remaining types attains the highest payoff consistent with equilibrium. Given the generality of our setting, coalition-proof equilibria may fail to exist. We establish existence under restrictions on preferences and evidence: quasiconcavity with message completeness, betweenness, cheap talk with message completeness, and payoff degradation. For a rich disclosure game with stochastic evidence, we show that the sender's ex-ante payoff in a coalition-proof equilibrium is a piecewise linear tent over the belief simplex, analogous to the concave closure in Bayesian persuasion and the quasiconcave closure in cheap talk.