Dynamic Preference "Reversals" and Time Inconsistency - and some parts of - Identifying Procrastination from the Timing of Choices

Philipp Strack (Yale University)

March 2, 2021, 17:00–18:30


Economic Theory Seminar


Paper 1 - We study identification of time preferences in environments where an agent first makes an advance commitment, and later can revise her choices. A frequently-discussed empirical finding in such environments---often interpreted as evidence of time inconsistency---is that on average, agents exhibit a stronger preference for time 1 gratification when deciding in time 1, rather than in time 0. We establish a series of theorems that show that in the absence of (untestable) parametric assumptions the only preference patterns that reject time-consistent expected utility maximization are where an agent's time 0 choice is revealed to be strictly dominated at time 1 with probability 1. Imposing quasi- or strict concavity of preferences does not aid with identification. Imposing strong parametric assumptions on the quasi-hyperbolic discounting model, such as constant relative risk aversion and multiplicative taste shocks, can only generate identified sets, which we argue are likely to be large. However, we prove that there is one class of empirical designs that does produce point-identification of time preferences: designs that elicit agents' willingness to pay for different alternatives, and where the marginal utility of money can be assumed to not vary with agents' preferences for the different alternatives. (joint work with Dmitry Taubinsky) ------------------------------------------------------------------------------------------------------------ Paper 2 - A (partially naive) quasi-hyperbolic discounter repeatedly chooses whether to complete a task. Her net benefits of task completion are drawn independently between periods from a time-invariant distribution. We show that the probability of completing the task conditional on not having done so earlier increases towards the deadline. Conversely, we establish non-identiability by proving that for any time-preference parameters and any data set with such (weakly increasing) task-completion probabilities, there exists a stationary payoff distribution that rationalizes the agent's behavior if she is either sophisticated or fully naive. Additionally, we provide sharp partial identification for the case of observable continuation values. (joint work with Paul Heidhues).