Bernoulli Without Bayes: A Theory of Utility-Sophisticated Preferences under Ambiguity

A decision-maker is utility-sophisticated if he ranks acts according to their expected utility whenever such comparisons are meaningful. We characterize utility sophistication in cases in which probabilistic beliefs are not too imprecise, and show that in these cases utility-sophisticated preferences are completely determined by consequence utilities and event attitudes captured by preferences over bets. The Anscombe-Aumann framework as employed in the classical contributions of Schmeidler (1989) and Gilboa-Schmeidler (1989) can be viewed as an important special case. For the class of utility sophisticated preferences with sufficiently precise beliefs, we also propose a definition of revealed probabilistic beliefs that overcomes the limitations of existing definitions.