Robustness of Full Revelation in Multisender Cheap Talk

This paper studies information transmission in a two-sender, multidimensional cheap talk setting where there are exogenous restrictions on the feasible set of policies for the receiver. Such restrictions, which are present in most applications, can, by limiting the punishments available to the receiver, prevent the existence of fully revealing equilibria (FRE). We focus on FRE that are i) robust to small mistakes by the senders, in that small differences between the senders’ messages result in only small punishments by the receiver, and ii) independent of the magnitudes of the senders’ bias vectors. For convex policy spaces in an arbitrary number of dimensions, we prove that if there exists a FRE satisfying property ii), then there exists one satisfying both i) and ii). Thus the requirement of robustness is, under these assumptions, not restrictive. For convex policy spaces in two dimensions, we provide a simple geometric condition, the Local Deterrence Condition, on the directions of the senders’ biases relative to the frontier of the policy space, that is necessary and sufficient for the existence of a FRE satisfying i) and ii). We also provide a specific policy rule, the Min Rule, for the receiver that supports a FRE satisfying i) and ii) whenever one exists. The Min Rule is the anonymous rule that punishes incompatible reports in the least severe way, subject to maintaining the senders’ incentives for truthtelling, no matter how large their biases. We characterize necessary and sufficient conditions for collusion-proofness of a FRE supported by the receiver using the Min Rule and show that if such a FRE is not collusion-proof, then no other FRE satisfying ii) can be collusion-proof. We extend our existence results to convex policy spaces in more than two dimensions and to non-convex two-dimensional spaces. Finally, our necessary and sufficient condition, as well as our specific policy rule, can be easily adapted if the receiver is uncertain about the directions of the biases and/or if the biases vary with the state of the world.