Epistemological Mechanism Design (Revised version of CARF-F-496)

This study demonstrates a novel epistemological approach to mechanism design. We consider a type space in which agents are either selfish or honest, and show that a slight possibility of honesty in higher-order beliefs motivates all selfish agents to behave sincerely. Specifically, in our model, a central planner attempts to elicit correct information through mutual monitoring. We assume severe restrictions on incentive device availability: neither public monitoring nor allocation rules are available. Thus, the central planner uses only monetary payment rules. It is well-known that if "all agents are selfish" is common knowledge, eliciting correct information as unique equilibrium behavior is impossible. Nevertheless, we show a very permissive result: the central planner can elicit correct information from all agents as unique Bayes Nash equilibrium behavior if "all agents are selfish" is not common knowledge. This result holds even if honest agents are mostly motivated by monetary interests.