انگیختگی ادراکی ایجاد حافظه های جدید اپیزودیک را بهبود می بخشد

This paper studies how dual-self (Fudenberg and Levine, 2006) decision-makers can use commitment technologies to combat temptation and implement long-run optimal actions. I consider three types of commitment technologies: carrot contracts (rewards for ‘good’ behavior financed by borrowing from future consumption), stick contracts (self imposed fines for ‘bad’ behavior) and binding commitment. I compare the welfare implications of these contracts and show that dual-self decision-makers strictly prefer to use carrots instead of either sticks or binding commitments. This is for several reasons: sticks are highly vulnerable to trembles (while carrots are not), sticks and binding commitments create a temptation to cancel them (while carrots do not), and finally carrots allow easy tradeoffs between commitment and flexibility (while sticks and binding commitments do not).

Many of us would like to exercise, work efficiently and stay away from our bad habits yet we often find ourselves skipping a daily run, looking at funny cat videos on the internet, smoking another cigarette or eating another cupcake. Such mismatches between what we would like to do tomorrow and what we actually end up doing create a demand for technologies to help individuals implement their normative goals. Existing literature2 shows that this demand for commitment does exist but there is little theoretical work on what forms optimal commitment technologies would take. This paper begins to bridge this gap. The first step to figuring out optimal commitment technologies is learning what mechanism generates the problem in the first place. Work in psychology and neuroscience tends to focus on decisions as an interplay between an automatic and a controlled process (Kahneman, 2003). Recent work in social psychology makes this model more sharp by positing that the controlled cognitive process uses a costly resource to operate. A standard template in such experiments involves subjects performing a ‘cognitive control task’ (controlling attention, suppressing emotion, solving math problems) or a task that does not require such control followed by a second cognitive control task. Subjects who had to perform the depleting task do worse on the second task than controls (see Muraven and Baumeister, 2000 for a survey). Another set of experiments involve individuals making choices under cognitive load (for example, subjects are asked to remember a 7 digit number), these individuals then act more impulsively than controls (e.g. by choosing more unhealthy foods to eat as in Shiv and Fedorikhin, 1999). In such a model our difficulties at the desert tray come from an interplay of automatic impulses compelling us to eat, and a conscious use of mental resources not to give in.3 This paper uses a particular economic model of this process: the dual-self model of Fudenberg and Levine (2006) (henceforth FL). This model is a specific case of a larger set of costly self-control models.4 In such models decisions are a compromise between a ‘temptation’ ranking and a ‘normative’ ranking, with the DM balancing a desire to choose according to his normative preference with a self-control cost of deviating from the temptation ranking. The fully dynamic version of the model (Fudenberg and Levine, 2012) captures the experimental results above by adding a form of convexity to the control costs: the more self-control has been recently used, the higher the marginal cost of resisting new temptations. The FL model imposes specific restrictions on where the disagreement between the temptation and normative preferences comes from: the DM uses a standard time-separable utility function to evaluate consumption streams (or dynamic plans) but the temptation or automatic process discounts the future at a much sharper rate than the DM would like. Thus the DM is tempted to behave impulsively and must use self-control to choose long-run rewards. Because of this structure FL refer to the automatic impulse as the short-run (SR) self and the cognitive control process as the long-run (LR) self.5 Note that this is a different type of model than those studied in the literature on hyperbolic/quasi-hyperbolic discounting (e.g. Laibson, 1997 and Ainslie and Haslam, 1992). In those models the DM's problems come from the fact that rankings of alternatives change in different periods. In the FL model both the SR and LR self are perfectly time consistent so the tension comes from multiple preferences within a period rather than multiple preferences between periods. The purpose of this paper is not to provide a clean test to differentiate these models, rather it is to look at commitment behavior with the FL model. However, all results that could never hold under a time-inconsistent framework are flagged as such. The FL model generates a huge demand for commitment and this paper considers two types of technologies that could be available to a DM facing temptation. The first are stick contracts that levy a fine on the DM when he gives in to temptation.6 The second are carrots that reward a DM who takes normatively good actions. In this paper, carrots are financed by the intertemporal substitution of future consumption to the present conditional on the DM resisting a temptation. The main results show that both of these types of contracts can only be welfare improving for a DM if they change the nature of the temptation he faces – i.e. the SR self's optimal action. The logic is one of revealed preference combined with the economic idea that self-control is treated as a cost. If a DM gives in to a temptation, this is because the self-control required to resist it was too expensive. If a commitment technology does not physically remove the tempting option, then it must remove the temptation associated with that option because if it does not, its ultimate effect is only to make the DM exert the self-control that he didn’t find optimal to exert in the first place. This means that if the source of the temptation is sharp discounting, both types of contracts must have one of two features: either their effects must be close in time to the choice or they must be particularly large. In fact, as the paper shows later, their size increases exponentially with the delay between action and contract implementation.