Three essays on sequential decision making.

摘要

In the literature of sequential choices, individuals' behavior is typically analyzed according to the relative distribution of choices over a set of trials. Because this representation of choices neglects how preferences between alternatives change over time, we restore to describe individuals' switching behavior. In the first essay, we characterize switching behavior by creating a statistical measure called the persistence index. To demonstrate the utility of this index, we simulate choices of groups of individuals making decisions with diverse decision rules. While the statistic that measures the overall proportion of choices does not distinguish among the various decision rules, the persistence index captures significant contrasts among them.;In the second essay we focus on the study of learning rules used by boundedly rational individuals. Given a set of standard assumptions, we show that individuals' learning rule consists in the estimation of the observed frequency of rewards in outcomes experienced. Since this rule does not incorporate the fact that subjects are capable of learning sequential features of outcomes, we modify the structure of the utility function. For these modified preferences, individuals estimate a statistic that combines information on the observed frequency of rewards and the length of runs of rewards. Predictions for this second rule entail a partial Positive Recency Effect in subjects' behavior. This implies that for non-random outcomes, individuals using the modified learning rule might choose the best alternative with higher probability than individuals using the basic rule.;In the third essay, we enlarge the equilibrium concept presented by Osborne and Rubinstein (1998), and apply it to the framework of sequential choices in the Prisoner's Dilemma game. We analyze the connection between the Osborne and Rubinstein's equilibrium predictions and the incentives to cooperate implied by the payoffs of the game. We find out that cooperation arises in equilibrium, only for sufficiently high incentives to cooperate and certain amount of information collected. In addition, we introduce a dynamic interpretation to the Osborne and Rubinstein's equilibrium concept. If a unique cooperative equilibrium arises, it is locally stable. If two cooperative equilibria exist, only one is locally stable.