What Can We Rationally Value?

There are many responses to Allais’s (1953) Paradox, some reconciling it with expected utility (EU) theory, others holding that the two cannot be reconciled (for better or worse for the theory). I here focus on a particular aspect of the debate—the extent to which EU axioms like Independence constrain the content of (not just the relationship between) our preferences. Many accept that an agent’s feelings towards risk/regret can legitimately be considered part of an act’s outcomes. This is nonetheless contentious, because risk and regret do not apparently reside in any particular outcome, but rather arise from the way in which potential act outcomes are related to one another, i.e. they are “global properties” of the outcome space or decision problem. I argue that in the normative context, even if “global properties” can be located in individual outcomes, the appeal to risk/regret attitudes to distinguish outcomes must answer to some kind of consistency considerations. Otherwise there is the danger of EU theory being irrefutable or lacking content. Furthermore, more permissive ways of including risk/regret in outcomes have implications for sequential decision scenarios that are analogous to the kind associated with violating Independence. So if avoiding these kinds of consequences is the main motivation for not violating Independence, then allowing the content of outcomes to be such that the same problems arise even if Independence is respected is not going to be any better a response to Allais’s Paradox.