Dynamically Consistent Preferences Under Imprecise Probabilistic Information

This paper extends decision theory under imprecise probabilistic information to dynamic settings. We explore the relationship between the given objective probabilistic information, an agent’s subjective multiple priors, and updating. Dynamic consistency implies rectangular sets of priors at the subjective level. As the objective probabilistic information need not be consistent with rectangularity at the subjective level, agents might select priors outside the objective probabilistic information while respecting the support of the given set of priors. Under suitable additional axioms, the subjective set of priors belongs to the rectangular hull of the objective probabilistic information.