Sequential Learning under Informational Ambiguity

This paper investigates a sequential social learning problem in which individuals face ambiguity about others' signal structures and have max-min expected utility preferences, thereby exhibiting ambiguity aversion. Unlike previous findings, which suggest that learning outcomes depend on the specifics of the learning environment, this study establishes information cascades as a robust outcome under ambiguity. With sufficient ambiguity, cascades arise almost surely, regardless of the statistical properties of signal structures. Moreover, standard results predicting the absence of cascades can easily break down: Even minimal ambiguity can trigger cascades when signals are bounded and lead to incorrect herding when signals are unbounded.