Testing axiomatizations of ambiguity aversion

The study of choice under uncertainty has advanced through key “paradoxes,” such as the Ellsberg paradox. We implement Machina’s (2014) three-outcome extension, in which four major ambiguity-aversion theories (multiple priors, rank-dependent, smooth ambiguity, variational) all predict indifference between two ambiguous acts. Contrary to these predictions, we find most participants do not express indifference. Our design elicits each subject’s certainty equivalent (CE) for an embedded 50–50 lottery and uses that CE in the Machina acts. Under lottery independence—i.e., if individuals apply standard (von Neumann–Morgenstern) expected utility to each objective lottery—these acts map to the same distribution of payoffs and thus should be evaluated identically. Yet we document a robust preference for one act over the other. This preference is associated with violations of lottery independence (e.g., Allais inconsistencies), as well as with disappointment aversion. Our results highlight that Machina’s three-outcome paradox is at least as much about failing independence over lotteries as it is about ambiguity aversion.