Testing ambiguity theories with a mean-preserving design

Prominent models such as maxmin expected utility/alpha-multiprior (MEU/ a -MP) and Klibanoff, Marinacci, and Mukerji (KMM) interpret ambiguity aversion as aversion against second-order risks associated with ambiguous acts. We design an experiment where the decision maker draws twice with replacement in the typical Ellsberg two-color urns, but with a different color winning each time. Given this set of mean-preserving prospects, MEU/α‐MP, KMM, and Savage's subjective expected utility all predict unequivocally that risk-averse decision makers (DMs) will avoid the 50 - 50 urn that exhibits the highest risk conceivable, while risk-seeking DMs do the opposite. However, we observe a substantial number of violations in the experiments. It appears that the ambiguity premium is partially paid to avoid the ambiguity issue per se, which is distinct from notions of second- order risk. This finding is robust even when there is only partial ambiguity, and is applicable to all models that satisfy a monotonicity condition.