Uncertainty Equivalents: Testing the Limits of the Independence Axiom

Abstract

There is convincing experimental evidence that Expected Utility fails, but when does it fail, how severely, and for what fraction of subjects? We explore these questions using a novel measure we call the uncertainty equivalent. We find Expected Utility performs well away from certainty, but fails near certainty for about 40% of subjects. Comparing non-Expected Utility theories, we strongly reject Prospect Theory probability weighting, we support disappointment aversion if amended to allow violations of stochastic dominance, but find the u-v model of a direct preference for certainty the most parsimonious approach.