Title - Rationality of Decisions That Avoid Predictably Regrettable Consequences (in working progress)

Abstract - A novel characterization of rationality is offered which relies on the hypothesis that under no predictable circumstances should behaviour in any finite decision tree ever lead to a consequence which, relative to the predicted feasible set F, belongs to a specified subset R(F) of regrettable consequences. The hypothesis is applied to behaviour that is defined on an unrestricted domain of finite decision trees, including continuation subtrees, with:

(i) decision nodes where the decision maker must make a move;

(ii) chance nodes at which a “roulette lottery” with exogenously specified strictly positive probabilities is resolved;

(iii) event nodes at which a “horse lottery” is resolved.

Building on earlier discussions of consequentialist behaviour, the hypothesis is shown to imply that behaviour must maximize a complete and transitive preference relation over consequence lotteries, with preferences that satisfy the independence axiom of expected utility theory, as well as a strict form of Anscombe and Aumann's extension of Savage's sure thing principle. Assuming continuity, non-trivial consequence domains, and a generalized form of state independence, the hypothesis is equivalent to a refined form of Bayesian rationality that excludes zero probabilities.