Persuasion with Correlation Neglect

We consider an information design problem in which a sender tries to persuade a receiver that has "correlation neglect", i.e., fails to understand that signals might be correlated. We highlight a tradeoff that emerges between the desire of the sender to sometimes use negative correlation while constraints imposed by Bayesian persuasion favour positive correlation. We illustrate this trade-off by showing that a sender whose utility increases in the posterior of the receiver will use positive correlation when the prior is against her interests, and will use negative correlation when the prior is in her favour. We then provide general results characterising the constraints on correlation imposed by Bayesian persuasion. Finally, we show that when the sender has a large number of signals at her disposal, she can achieve a utility close to her first best and can do so using full positive correlation.

Joint with Gilat Levy and Ronny Razin

Recording: To view the seminar recording, click here.