Unexpected Utility: Experimental Tests of Five Key Questions about Preferences over Risk

Experimental work on preferences over risk has typically considered choices over a small number of discrete options, some of which involve no risk. Such experiments often demonstrate contradictions of standard expected utility theory. We reconsider this literature with a new preference elicitation device that allows a continuous choice space over only risky options. Our analysis assumes only that preferences depend on the probability p and prize x; U = u(p; x): We then allow subjects to choose p and x continuously on a linear budget constraint, r1p + r2x = m, so that all prospects with a nonzero expected value are risky. We test …ve of the most importantly debated questions about risk preferences: rationality, prospect theory asymmetry, the independence axiom, probability weighting, and constant relative risk aversion. Overall, we …nd that the expected utility model does unexpectedly well. We would like to thank Michele Cohen, Shachar Kariv, Justin Rao, Charles Sprenger, and JeanChrisotphe Vargnaud for helpful comments, and Daniel Gelman and Marion Price expert computer programming and research assistance. This work was supported by NSF grants SES-0551296 (Andreoni) and SES-0112157 (Harbaugh).