It is commonly argued that dynamic consistency, consequentialism and non-expected utility are incompatible. The first aim of this paper is to rebut such arguments, by targeting the implicit assumption that the relevant contingencies correspond to objective resolutions of uncertainty (that is, events in a state space). These are not necessarily the same as the contingencies that the decision maker envisages, and we argue that any reasonable notion of dynamic consistency involves the latter, rather than the former, sort of contingency. We formulate such a version of dynamic consistency and show it to be compatible with consequentialism and non-expected utility. We then analyze the economic consequences of this new perspective. On the one hand, it provides a principled justification for restrictions on non-expected utility models (such as that proposed by Epstein and Schneider (2003)) in applications to dynamic choice problems. On the other hand, it provides a new analysis of the issue of attitude to information; contra standard arguments, the value of information under non-expected utility is non-negative as long as the information offered does not compromise information that the decision maker had otherwise expected to receive. Finally, we give a representation theorem for the contingencies the decision maker envisages, in the case where he uses the maxmin expected utility rule.
