Robust information design

Department of Economics and Decision Sciences

Speaker:  V. Bhaskar (UT Austin)

Room S-020

Abstract :

We study information design when agents play a  supermodular game and the designer prefers higher agent actions. We assume that the designer is slightly uncertain about the exact preferences of the  agents.

Specifically, we perturb the payoffs of each agent via a continuous payoff shock, as in Harsanyi (IJGT 1973), and focus on purifiable equilibria. These are limits of some sequence of equilibria of perturbed games as the shocks vanish. When the designer has no private information, the standard solution is generically purifiable.  However, this is no longer true when designer has some private information about the state, and can design experiments that can be arbitrarily correlated with her private information.   With binary states or with a perfectly informed designer,   in any purifiable equilibrium, the agent learns the private information of the designer, thereby lowering the designer's ex ante payoff.  We also introduce the notion of the experiment-proofness: an experiment is experiment-proof if, after observing its outcome,  the designer does not want to conduct further experiments. With multiple agents, any experiment-proof purifiable equilibrium must be public, so that all agents share the same information. Finally, if the designer is able to conduct private experiments before interacting with the agent,  this results in complete unravelling -- purifiable experiments are perfectly informative.