Decentralized Persuasion

Abstract: Information is often distributed across multiple senders. This paper explores the extent to which senders with a common utility function but independent and imperfect information can collaboratively persuade a receiver. We suppose that the senders' combined information (almost) fully reveals the true state, and compare these senders' attainable utility to the maximum achievable by a single, fully informed sender. We show that decentralized senders cannot generally attain this maximum, except in cases where fully revealing a state is optimal. On the positive side, we show that they can guarantee a constant fraction of this maximum, and that the maximum itself can be obtained if senders have access to a public randomization device.   Joint work with Itai Arieli and Yishu Zeng.