Persuasion without ex-post commitment

Abstract

We study a persuasion model with limited commitment in which the sender can commit to a signal ex ante but not to truthful reporting ex post. Sender-optimal signals exploit a key trade-off: although better-quality information allows the sender to communicate more precisely, it adversely affects the credibility of his messages. Building on this observation, we provide sufficient and necessary conditions for the sender to strictly benefit from obtaining an imperfect signal. Under linear preferences, we show that sender-optimal signals can take on a bi-pooling structure; if preferences are also convex, bi-pooling becomes necessary. Finally, we provide a full characterization of sender-optimal signals for the popular uniform-quadratic specification, which we use to quantify and compare the value of commitments.