Two rationality arguments are used to justify the link between conditional and unconditional preferences in decision theory: dynamic consistency and consequentialism. Dynamic consistency requires that ex ante contingent choices are respected by updated preferences. Consequentialism states that only those outcomes which are still possible can matter for updated preferences. We test the descriptive validity of these rationality arguments with a dynamic version of Ellsberg's three color experiment and nd that subjects act more often in line with consequentialism than with dynamic consistency.

This paper shows that, for Choquet expected utility preferences, the axioms consquentialism, state independence and conditional certainty equivalent consistency under updating characterise a family of capacities, which we call Generalized Neo‐Additive Capacities (GNAC). This family contains as special cases, among others, neo‐additive capacities as introduced by Chateauneuf, Eichberger, and Grant, Hurwicz capacities, and ɛ‐contaminations. Moreover, we will show that the convex version of a GNAC is the only capacity for which the core of the full Bayesian updates of a capacity, introduced by Jaffray, equals the set of Bayesian updates of the probability distributions in the core of the original capacity.

This paper explores the relationship between dynamic consistency and the existing notions of unambiguous events for Choquet expected utility preferences. A decision maker is faced with an information structure represented by a filtration. We show that the decision maker's preferences respect dynamic consistency on a fixed filtration if and only if the last stage of the filtration is composed of unambiguous events in the sense of Nehring (1999). Adopting two axioms, conditional certainty equivalence consistency and constrained dynamic consistency to filtration measurable acts, it is shown that the decision maker respects these two axioms on a fixed filtration if and only if the last stage of the filtration is made up of unambiguous events the sense of Zhang (2002).

We characterize prior-by-prior Bayesian updating using a model proposed by Gilboa, Maccheroni, Marinacci, and Schmeidler (2010) that jointly considers objective and subjective rationality. These rationality concepts are subject to the Bewley unanimity rule and maxmin expected utility, respectively, with a common set of priors and the same utility over consequences. We use this setup with two preference relations to develop a novel rationale for full Bayesian updating of maxmin expected utility preferences.