Revealed preferences over risk and uncertainty

Consider a finite data set where each observation consists of a bundle of contingent consumption chosen by an agent from a constraint set of such bundles. We develop a general procedure for testing the consistency of this data set with a broad class of models of choice under risk and under uncertainty. Unlike previous work, we do not require that the agent has a convex preference, so we allow for risk loving and elation seeking behavior. Our procedure can also be extended to calculate the magnitude of violations from a particular model of choice, using an index first suggested by Afriat (1972, 1973). We then apply this index to evaluate different models (including expected utility and disappointment aversion) in the data collected by Choi et al. (2007). We show that among those subjects exhibiting choice behavior consistent with the maximization of some increasing utility function, more than half are consistent with models of expected utility and disappointment aversion.