What people know about their future preferences and how they take this knowledge into account in their decisions are questions of primary importance in formal models of intertemporal choice and in many domains of public policy. I investigate prediction of changes in state-dependent preferences in the case of habit formation, prediction of future self-control problems, and how agents with self knowledge with respect to future self-control problems think about the actions and beliefs of their future selves.
In chapter one I and a coauthor extend the gym-attendance study of Charness and Gneezy (2009) by incentivizing subjects to attend the gym for a month and observing their pre- and post-treatment attendance relative to a control group. In addition we elicit subjects' pre- and post-treatment predictions of their post-treatment attendance. We find a habit formation effect similar to that of Charness and Gneezy in the short-run, but with substantial decay caused by winter vacation. We additionally find that subjects seriously over-predict future attendance, which we interpret as evidence of partial naivete with respect to self-control problems. Subjects also appear to have biased beliefs about their future cost of gym attendance. Our design allows us to estimate the monetary value of habit-formation equivalent to a forty cents per visit subsidy, as well as the welfare cost of naivete.
In chapter two we address whether individuals accurately predict habit-formation, a question of both theoretical and practical interest. Gym-attendance is one domain in which this question is of particular interest to public policy makers. We test for mispredic-tion of habit-formation in gym attendance with a field experiment and find that subjects do form a habit, and do not predict it fully. We develop a simple model that incorporates habit-formation and projection bias in the framework of quasi-hyperbolic discounting and calibrate the parameters of the model.
In chapter three, borrowing from Cognitive Heierarchy Theory, I introduce bounded rationality into the beta{delta model of present-biased preferences. I define a level-two agent, or k-2 sophisticate, as one who is aware that her future selves will have present-
bias, but believes that they will be naive. The k-2 sophisticate does one round of strategic thinking about her future behavior instead of the unlimited number of rounds of the full so-phisticate. In the Doing it Once model of procrastination of O'Donoghue and Rabin (1999) the k-2 sophisticate typically procrastinates and preproperates less than the full sophisticate, and is protected from severe harm from both extreme preproperation and extreme procrastination, though she may suffer from excessive costly preemption due to pessimism about future preemption when costs are immediate.
