This proposal contains two related research projects on dynamic choice. Both of these projects will be undertaken jointly with Wolfgang Pesendorfer of Princeton University. Often, observed choice behavior will not be consistent with a single, deterministic choice criterion (i.e., utility function). That is, the decision-maker will make different choices in virtually identical situations. Such behavior is best modeled as random choice. The first project entitled "Random Utility, Uncertain Moods and Bounded Rationality," addresses the following question: when can a random choice be interpreted as the consequence of maximizing a randomly chosen criterion for ranking the alternatives. Hence, we investigate the consequences of behavior that is derived from a fixed, probabilistic rule for selecting a utility function. After the utility function is selected, the agent chooses the alternative that maximizes it among the available options. This way of interpreting behavior is important for the analysis of aggregate data, boundedly rational behavior and situations with asymmetric information. In our work, we study choice over risk prospects and randomly selected expected utility functions.
The second project entitled "The Revealed Preference Theory of Changing Tastes" studies dynamic inconsistency. Dynamic inconsistency refers to the tendency of agents to make different choices from the same set of alternatives depending on when they are asked to choose. For example, an individual who prefers a larger later reward to a smaller earlier reward may reverse his preference as the date of the smaller earlier reward gets closer. Many decision problems require the agent to make choices at various moments in time. In such problems, the key is how the decision-maker forms expectations about his own future behavior. This research takes the position that the agent's beliefs regarding his future behavior can best be analyzed through his current behavior and seeks to identify these beliefs through choice experiments. That is, we wish to identify methods for determining the agent's expectations about his future behavior from actual choices made in the current period. The advantage of the revealed preference approach is that preferences over choice problems are - at least in principal - observable.
