Are individuals typically self-aware to this degree?We are not familiar with denitive evidence on this question and in its absence, we are inclined to feel that full self-awareness is a plausible working hypothesis.4 Even where the opposite extreme of complete An appendix outlines the parallel analysis for changes in risk attitude.
Adapt their model rst so that menus consist of (Anscombe-Aumann) acts over a state space S2 .
In this setting, since preference A rough statement is as follows: if conditional preference at every decision node conforms to subjective expected utility theory and is independent of unrealized parts of the tree, then preferences are dynamically consistent if and only if they have a common v NM index and conditional beliefs at each node are derived by Bayesian updating of the initial prior.
We focus on changes in beliefs because we nd this route to be both intuitive and useful - it leads to a new model of and way of thinking about non-Bayesian updating.
4 Comparable sophistication is assumed by GP and also in the literature on non-geometric discounting where the agent is often modeled as gaming against herself (see Laibson(1997), for33 naivete seems descriptively more accurate, our model may help to clarify which economic consequences are due to non-Bayesian updating per se and which are due to naivete.
In either case, it is as though she retroactively changes her prior and then applies Bayes Rule to the new prior.
The resulting posterior belief diers from what would be implied by Bayesian updating of the original prior and in that sense re?
Temptation refers to experiencing these urges, which here stem from a change in beliefs. As in GP, by assuming that preference is dened over (contingent) menus, we are able to model the agent dynamic behavior via maximization of a stable s (complete and transitive) preference relation.
This is possible because our agent is sophisticated - she is forward-looking and anticipates her exuberance or, more generally, her psyche as it aects her reactions to signals ex post.
We think it worthwhile to explore such modest departures from standard models before discarding the entire framework.
Several systematic deviations from Bayesian updating have been discussed in the psychology literature and some of these have been incorporated into modeling exercises in behavioral nance.5 Our model cannot address these ndings directly because the experimental literature deals with settings where prior probabilities are given objectively, while our model deals with the case where probabilities are subjective (which case we would argue is more relevant for economic modeling).
Finally, it is the agent sophistication that permits updating s behavior to be inferred from her (in principle observable) ranking of contingent menus.