An Allais paradox for generalized Expected Utility Theories ?
Résumé
This article reports the results of an experiment which aims at providing a test of ordinal independence, a necessary property of Generalized Expected Utility theories such as Rank-Dependent Expected Utility theory (RDEU). Our experiment is based on a modified version of the Allais paradox proposed by Machina, which allows testing ordinal independence restricted to simple lotteries, i.e. the tail-separability property. The results tend to support RDEU models since tail-separability is not violated by 71% of subjects while 73% violate the independence condition of classic Allais paradox. This confirms the relative theoretical soundness of RDEU models over Expected Utility model for the particular context of risk.