An Evaluation of Risk Preferences Using Cross Validation

This paper studies which models of risk preference can best describe individual choices. We perform model selection using cross validation techniques that are common in machine learning. We compare the models of expected utility, disappointment aversion, rank dependent utility, stochastic reference dependence, quadratic utility, cautious expected utility, and heuristic decision rules. The heuristics we examine model a subject who targets receiving a fixed probability of a certain prize. We find that heuristics describe choices better than economic models approximately 39% of the time. Of the economic models considered, we find that stochastic reference dependence describes approximately 26% of individuals and quadratic utility describes approximately 22% of individuals