Certifiable information: revelation and persuasion.

Summary In many situations, decision-makers have to choose an action or a policy without being fully informed. In addition, it may be difficult or expensive to acquire missing information directly. In such cases, they may seek help from informed institutions or individuals. The latter may try to influence the decision in their favor by hiding or presenting only part of the information. For example, employers rely on information presented by job seekers, financial authorities use company reports to evaluate them, and elected officials consult experts before proposing legislation. In these examples, at least some information is certifiable or verifiable. In other words, the informed party can prove certain statements by presenting evidence, or the policymaker can verify the accuracy of those statements. Because verification can be costly or time-consuming, the decision-maker can often verify only a portion of the information received. These constraints determine the amount of information that can be verified before the decision is made. The first two chapters focus on models adapted to situations where the decision maker must evaluate a statement or respond to a request made by an individual or institution. In the third chapter, I consider a slightly different setting where the decision maker consults informed agents before choosing an action.In the first chapter, I study a model where the preferences of the informed agent are state independent. In one-way communication, only the agent sends a message to the decision maker. In two-way communication, both exchange messages. I compare these two mechanisms by assuming that the same amount of evidence can be presented in both cases. In the canonical two-way communication mechanism, after receiving a statement from the agent, the decision maker asks the agent to present a particular piece of evidence. The decision depends only on his ability to present the requested evidence. The main result of this chapter states that two-way communication improves the outcome if the certification of information is limited in a way that prevents the decision maker from reaching his optimum in one-way communication.The second chapter, which results from joint work with Frédéric Koessler (CNRS, Paris School of Economics), studies the implementation in the presence of ambiguity averse agents. We show that if an allocation rule can be implemented with unlimited certification, it can also be implemented with limited information certification if the decision maker can use ambiguous communication mechanisms and if the agents are ambiguity averse in the sense of maxmin. The opposite implication is true if there is a single agent and a punishing action.In the third chapter, I study a model with two types of informed agents. One type wants to maximize the decision maker's action while the other wants to minimize it. In this case, there may be a need to consult more than one agent. I study sequential consultation and examine its impact on information revelation. In equilibrium, the decision maker continues to consult informed agents as long as her uncertainty is high enough. Minority agents - in terms of preferences - can influence the decision maker by withholding information when it is unfavorable because he rightly anticipates that the majority is more likely to do so. In addition, the threat of sequential consultation can be used to extract more accurate information while consulting with a single agent.