Panics and Early Warnings

Abstract

We study optimal adversarial information design in a dynamic regime change game. Agents decide when to attack, if at all. We assume (1) delay incurs a continuous cost and (2) agents doubt the correctness of their actions. The game may end in a “disaster” due to weak fundamentals or panic—agents attacking despite sound fundamentals. We propose a “timely disaster alert” that promptly warns about impending disasters, making waiting for and following the alert the unique rationalizable strategy, thereby eliminating panic. We relate this optimal policy to early-warning systems such as bank stress tests and debt sustainability analysis.