Andrew Ellis (Economics, LSE), joint with Michele Piccione: Complexity, Correlation, and Choice

Abstract: Often a profile of actions rather than a single action determines payoffs. The former case is more complex, as the correlations among outcomes across actions are payoff-relevant. Existing choice-theoretic models do not allow this complexity to affect behavior. To construct such a model, we introduce a framework that explicitly models choice of an action profile. We show that the axiom “Monotonicity” precludes a role for complexity and propose a weakening thereof that accounts for the possibility that the agent perceives correlations incorrectly. When the agent satisfies the other subjective expected utility axioms, the axiom ensures she acts as if she assigns probabilities to all potential correlations between actions and then maximizes expected utility. The model accommodates, and provides insight into the behavioral connection between, phenomena such as correlation neglect, cursed behavior, and home bias. We also identify the behavior that characterizes the events and actions the agent understands and relate it to our representation. (joint with Michele Piccione).