Preference for mixture and local ambiguity reduction in nonconvex preferences

Ambiguity aversion is often modeled as a preference for mixing different alternatives, but such behavior may not be global, emerging only when the mixture sufficiently reduces ambiguity. This paper provides a utility function characterization that accommodates various patterns of mixture preferences by introducing a novel concept called the core bound. A preference for a specific mixture is characterized by the nonemptiness of the core bound, which indicates substantial overall ambiguity but relatively limited ambiguity for that mixture. Several notable special cases of the main theorem include nonconvex preferences as well as the Maxmin Expected Utility preference. Additionally, core bounds are useful in identifying relative ambiguity aversion between two different preference relations.