Epstein and Schneider (2007) develop a framework of learning under ambiguity, generalizing maxmin preferences of Gilboa and Schmeidler (1989) to intertemporal settings. The specific belief dynamics in Epstein and Schneider (2007) rely on the rejection of initial priors that have become implausible over the learning process. I demonstrate that this feature of ex-post rejection of theories gives rise to choices that are in sharp contradiction with ambiguity aversion. Concrete, the intertemporal maxmin decision-maker equipped with such belief dynamics prefers, under prevalent conditions, a bet in an ambiguous urn over the same bet in a risky urn. I offer two modifications of their framework, each of which is capable of avoiding this anomaly.
learning under ambiguity multiple prior maxmin ambiguity aversion