Ambiguity aversion and a decision-theoretic framework using belief functions

This paper introduces a new approach to constructing normative models that exhibit the same ambiguity aversion as human decision makers. The models are constructed using a decision-theoretic framework based on the theory of belief functions interpreted as generalized probability. The level of ambiguity aversion is determined by a subjective parameter in the unit interval so that users have the possibility to fix its strength in the model. We show using three examples that the decisions, which are determined by the optimization of a total subjective reward (corresponding to a reduced expected reward), are consistent with the experimental results observed by Ellsberg and other authors.