Mean-dispersion preferences and constant absolute uncertainty aversion

We axiomatize, in an Anscombe–Aumann framework, the class of preferences that admit a representation of the form V(f)=μ−ρ(d), where μ is the mean utility of the act f with respect to a given probability, d is the vector of state-by-state utility deviations from the mean, and ρ(d) is a measure of (aversion to) dispersion that corresponds to an uncertainty premium. The key feature of these mean-dispersion preferences is that they exhibit constant absolute uncertainty aversion. This class includes many well-known models of preferences from the literature on ambiguity. We show what properties of the dispersion function ρ(⋅) correspond to known models, to probabilistic sophistication, and to some new notions of uncertainty aversion.