WP 2013-09 Identifying Time Preferences With Experiments: Comment
AUTHORS: Glenn W. Harrison, Morten I. Lau & E. Elisabet Rutström
ABSTRACT. Identifying time preferences with laboratory experiments demands attention to theoretical, experimental and econometrics issues. Andreoni and Sprenger [2012a] propose a single choice task and several econometric methods that seek to address these issues. The choice task requires subjects to make portfolio allocations between sooner and later payments of money. All theories they examine imply that subjects pick one boundary or the other, or that they pick strictly interior allocations. Their econometric methods seek to explain the average portfolio choice, but ignore the bald fact that 70% of the responses by the subjects were choices at one boundary allocation or the other. The average portfolio choice implied by the modes at either boundary is chosen by virtually none of their subjects. Their ad hoc econometric attempts to model the truncation of choices at the boundaries fail to account for the economics of the observed behavior. A systematic analysis of their data generates a priori implausible estimates of significantly convex utility functions. Andreoni and Sprenger [2012b] inherits all of the problems of the basic design and econometrics from Andreoni and Sprenger [2012a], and adds one: their findings are immediately confounded by non-additivity of the intertemporal utility function. Apart from this theoretical confound, there is experimental evidence of just this type of non-additivity, leading to an aversion to correlated payoffs over time. The evidence in favor of correlation aversion predicts the qualitative pattern they observe perfectly, without claiming that the utility function for stochastic outcomes is somehow different from the utility function for non-stochastic outcomes.