Mark Dean
G6493 Behavioral Economics
a

A behavioral economics course covering bounded rationality, temptation and self control and reference dependent preferences. Please email me if you spot any typos

Course Details and Syllabus

UTILITY MAXIMIZATION:

Lecture 1: In this first lecture we did a very rapid tour of revealed preference theory and the testable implications of utility maximization. We began with a discussion of why representation theorems can be useful for testing models with latent variables (i.e. variables which are not directly observable, such as utility, beliefs, information costs etc). We illustrated this point with a reminder of how to test for utility maximizing behavior (notes on which are here). We then started to provide extensions to this result which cover the cases of incomplete data, choice functions and infinite choice sets. We will talk more about these in lecture two, but they are also are covered here. Some useful mathematical background on order theory can be found in these notes. One issue that I mentioned, but which we did not get time to discuss was the problem that representation theorems provide a very 'sharp' test of a model: either the data satisfies a set of axioms or it does not (usually the latter). Some ways of dealing with this issue are discussed here.

I finished the lecture by (starting to) go through some of the well documented violations of utility mazimization which will motivate the section of the course on bounded rationality. The slides for that presentation are here.

For further details on these topics you can check out the readings discussed in the syllabus. For more details on utility maximization and preferences, chapters 2 and 3 of the Kreps book are great. The book Lecture Notes in Microeconomic Theory by Ariel Rubenstein is freely available for download and also has some great stuff.

For next week's class, I want you to prepare a 15 minute presentation on the paper "Who is (More) Rational? by Choi et. al. In particular, I want you to look at Appendix II: testing for consistency, to get an idea about how consistency measures work.

BOUNDED RATIONALITY:

Lecture 1: In this lecture, I provided a quick introduction to the study of bounded rationality, and introduced the concept of consideration sets. The slides from the lecture are available here.

The key theme of this lecture was that marketers think that people do not seriously think about all the available options when making a choice: instead they focus their attention on some 'consideration set', from which they make their choice (seminal marketing articles on this idea are here and here). While this seems like an intuitively plausible idea, testing a model of choice with consideration sets is tricky, requiring either more structure or additional data. We followed one approach, which links consideration set formation to sequential search and satisficing - based on materiel from papers available here and here (another interesting paper on sequential search can be found here).

Other interesting approaches to studying choice with consideration sets are provided by Masatlioglou et al. [2012], De Los Santos et al. [2012] and Manzini and Marrioti [2014]. You can see another recent approach by me here

For next week's class, I want you to prepare a 15 minute presentation on the paper "Competing for Consumer Inattention" by de Clippel et al.

Lecture 2: Next, we considered the broader class of `Rational Inattention' models, in which the decision maker wants to learn about some underlying state of the world which will affect the value of subsequent decisions. We modelled the choice of information in an abstract way, through the choice of information structures. A big free parameter in this class of models is the nature of information costs. In this lecture, we took a 'revealed preference' approach, asking under what circumstances could behavior be explained by any underlying information cost function. We also introduced the concept of state dependent stochastic choice data as a useful tool for testing models of limited information.

The slides for this lecture are available here. Most of the materiel for this lecture was drawn from this paper and the earlier working paper which includes the experiments. An alternative approach we discussed, based on choice from menus, is available here.

For next week's class, please prepare a 15 minute presentation on the paper "Attention Discrimination: Theory and Field Experiments with Monitoring Information Acquisition" by Bartos et al.

Lecture 3: This lecture covered the specific case of rational inattention when costs are the based on the Shannon Mutual Information between state and signal. Shannon costs have been immensely popular since they were introduced into the literature by Sims (see for example here and here for early Sims articles). They have been justified on axiomatic and information theoretic grounds (although, as discussed in class the latter is a bit dubious). We can think of the Shannon model as akin to the 'Cobb Douglas' approach to information costs, rather than the 'revealed preference' approach taken in the previous lecture. Like Cobb Douglas utility, Shannon costs can make life very easy, and much is know (as can be determined from this excellent text book.)

The notes from today's lecture are here here. The key paper is Matejka and Mackay [2015]. Details of the posterior based approach can be found here. For those who are interested, a recent paper by Steiner, Stewart and Matejka looks at a dynamic variant of the problem.

For next week's class, please prepare a 15 minute presentation on the paper "A Sparsity Based Model of Bounded Rationality" by Xavier Gabaix.

Lecture 4: Here we covered the Sequential Sampling Models (SSMs), which have proved extremely popular in psychology and neuroscience as a description of binary choice. These models have increasingly garnered attention in the economics literature.

The slides from the lecture are here. The literature on sequential sampling models is intimidatingly vast. A useful introduction and taxonomy of the various types of model can be found in Ratcliff and Smith [2004]. For the basic maths that we went through to identify the implications of the drift diffusion model (DDM), Shadlen et al. [2007] is invaluable. For a discussion of the optimality of SSMs, see Bogacz et al. [2006], Satohiro et al. [2016], and (for a more decision theoretic view) Fudenberg et al. [2015].

In discussing the perceptual evidence, the Ratcliff and Smith and Bogacz et al papers cited above are useful. I also made significant use of Ratcliff and McKoon [2008]. For an application to economic choice, try Milosavljevic et al. [2010]. For relation to the underlying neurobiology, the Shadlen et al. article is good. For more details see Gold and Shadlen [2007] and Bogacz [2007].

One topic that we did not get the chance to discuss in detail is the role of visual attention in sequential sampling. For some intriguing evidence, take a look at Krajbich et al. [2011].

For next week's class, please prepare a 15 minute presentation on the paper "Revealed Indifference" by Konovalov and Krajbich.

Problem Set 1 and necessary Data

TEMPTATION AND SELF CONTROL:

Lecture 1: The introductory lecture gave a summary of why temptation is an important topic for behavioural economists. The relevant slides are here.

As well as providing evidence on the ubiquity of behaviours which we might intuitively think of as relating to temptation (obesity, smoking, drug use, credit card debt etc), we discussed some of the literature linking self control to economic outcomes. On the one hand, the classic marshmallow experiment suggests that self control measured in childhood may be linked to later life outcomes. On the other, recent work has suggested that poverty may impede cognitive function, which in turn has been linked to the ability to exert self control. We also discussed some of the evidence suggesting that self control might be a depletable resource (see eg Galliot et al. [2007]). However, we also noted more recent replicability concerns, discussed here.

I concluded by saying that there were two established ways of behaviourally studying temptation and self control. The first was through preference for commitment, the second was through preference reversals in time preferences. These topics will form the basis of the next two lectures

Lecture 2: This lecture covers in detail the idea that temptation problems might manifest themselves as a 'preference for commitment', or a desire to have tempting alternatives removed from one's choice set. The classic example is that a dieter might prefer to go to a restaurant which has only salad on the menu, rather than one which has both salad and a burger. The slides from the lecture are available here.

My treatment of the subject comes mainly from the excellent survey article by Lipman and Pesendorfer, which is in turn largely based on the seminal article by Gul and Pesendorfer. This is a model of menu preferences in which a preference for commitment can stem from two different possible sources: the fact that the decision maker may succumb to temptation, or the fact they may not succumb, but that they have to exert costly self control in order to avoid doing so.

We considered various extentions and modifications to the Gul Pesendorfer model, including revealed willpower and random self indulgence. We also discussed the potentially offsetting impact of preference uncertainty, which can lead to a 'preference for flexibility', as discussed by (for example) Kreps.

For next week's class, please prepare a 15 minute presentation on a paper that discusses further the interaction between temptation and preference uncertainty: Commitment Vs Flexibility by Amador, Werning and Angeletos.

Lecture 3: Which covered models of time preference. The starting point for this lecture was the observation that 'preference reversals' are often taken as evidence of temptation problems (for example, I will choose to watch an action film rather than a documentary today, but if I am choosing what to watch in a week's time, then I will choose to watch the documentary). The central theme of the lecture is to show the equivalence between non-exponential discounting, preference reversals and demand for commitment. The slides from the lecture are available here.

We began by considering the behavioral implications of different models of discounting for preferences over streams of consumption, following Olea and Strzalecki. This allowed us to show that preference reversals were inconsistent with the standard model of exponential discounting, and in particular violated the assumption of stationarity. We showed how the quasi-hyperbolic (or beta-delta) model popularized by Laibson can be characterized by weakening stationarity in the appropriate way.

While preferences over consumption streams are theoretically useful, often we observe people making dynamic choices, for example amount of consumption today , which then leads to an amount of wealth tomorrow. We considered the implications of the quasi-hyperbolic model for such a data set. Useful in this regard is the hyperbolic Euler equation of Harris and Laibson. In terms of fully characterizing the behavior consistent with such a model, useful (but incomplete) results are provided by Barro [1999], Blow, Browning and Crawford [2015] and Echenique, Imai and Saito [2016].

Lecture 4: Evidence. In this lecture we covered some of the empirical evidence related to the topics we have covered, specifically preference for commitment, preference for flexibility, the measurement of time preferences and present bias, the relationship between preference for commitment and preference reversals and sophistication. The notes are available here here.

We showed that in is possible to generate preference for commitment in lab and field settings. For lab evidence, we looked at a 2010 paper by Hauser et al. For field evidence we looked at Kaur et al. [2015], which shows that workers in a data entry firm opt in to a 'dominated' contract which reduces their pay if they do not hit a pre-specificed target.

Despite the fact we can generate preference for commitment, there are few commitment devices which are naturally occurring in the market (see here for David Laibson's take on the matter). One possible reason is that preference uncertainty leads to a preference for flexibility. We discussed a paper which provides evidence that preference for flexibility is also important for labor contracts

Next we discuss the measurement of time preferences. Typically, this has been done by comparing the monetary amount at time t that makes the subject indifferent to a different amount at time t+1. Recent advances (for example Andersen et al. [2008] and Andreoni and Sprenger [2012]) have concentrated on controlling for curvature of the utility function. However, as this paper shows there are other problems with the use of monetary payments to measure time preferences. As a result, authors such as Augenblick et al. [2015] have suggested using other measures, for example based on real effort. This paper also provides evidence that present bias is linked to a preference for commitment.

Finally we studied whether people appeared to be sophisticated about their future choices. Broadly speaking the answer seems to be 'no' (or at least not completely). Della Vigna and Malmendier [2006] provide evidence that people may be over optimistic about their gym-going habits, while a recent paper by John [2015] suggests that people may take on commitment contracts which are not good for them due to a lack of sophistication.

For next week's class, please prepare a 15 minute presentation on a paper that discusses the experimental evidence for violations of stationarity and what they mean : Time Consistency: Stationarity and Time Invariance by Yoram Halevy.

Problem Set 2

CONTEXT DEPENDENT PRFERENCES:

Lecture 1: Notes available here. In this section of the course we will look at the way in which `context' can affect the choices that people make. Specifically, we will discuss two different (but possibly related) types of context effects: reference point effects, by which outcomes are judged relative to some reference point, and choice set effects, by which the range of options available affect the assessment of each alternative.

In this first lecture we begin with a fairly rapid discussion of reference dependence by describing three classic demonstrations of the fact that people's choices can be affected by a reference point: the endowment effect (see and here, here and here for a more recent argument on the subject), status quo bias, and reference dependence for risky choices (see Ert and Erev [2013] for an interesting discussion).

Next we moved on to a discussion of some of the standard models of reference dependence, the most famous being the loss aversion model of Kahnemann and Tversky. We discussed how this could explain all three of the phenomena described above, and in the guise of prospect theory has become one of the workhorse models of behavioural economics.

For a more decision theoretic take on reference dependence, and in particular status quo bias, we looked at the work of Masatlioglou and Ok [2005], the ideas in which are expanded upon here.

One of the problems of this class of model is that they assume that, for a fixed status quo, people act in line with the standard model of utility maximization. This means they cannot capture phenomena such as the 'too much choice' effect. Alternative models, based on the concept of decision avoidance, information overload, or attention may be necessary to capture this type of thing. Papers that try to do this are here and here.

One question not so far addressed is 'where do reference points come from'? In order to address this question Koszegi and Rabin [2006] introduced the idea of a 'personal' equilibrium', which demands that the chosen object must be optimal when it is itself the reference point, making beliefs and actions consistent.

We concluded the lecture with a number of applications of reference dependence, to the equity premium puzzle, the effect of evaluation period on risk attitudes, information acquisition, the disposition effect, labor supply and tax rebates.

Please prepare a presentation on this paper, which uses a labor supply experiment to try to determine where reference points come from.

Lecture 2: In this lecture we focussed on some recent work on 'context effects' - an umbrella term we used to describe the fact that the range of options available to a decision maker can create a context which affects the way in which these options are evaluated. This in turn can lead to violations of the independence of irrelevant alternatives, an axiom of standard utility maximization, which states that adding an option to a choice set can only affect choice if that option is itself chosen.

The lecture (the notes for which are here) covers three approaches to context effects. The first, by Louie et al. [2013], uses ideas from neuroscience to build a model of context effects in stochastic choice. The other two papers consider instead the domain in which alternatives are multidimensional, and context effect change the way that one alternative is traded off for another. Bushong et al. [2015] consider a model of "relative thinking", in which the range of values along a given dimention affects the way that absolute differences in value are percieved. Bordalo, Gennaioli and Shleifer consider in two papers (from [2012] and [2013]) a model in which context affects choice by making some dimensions 'salient', which then recieve a higher weight in the decision problem.

Lecture 3: Finally, we looked at rational models of reference dependence, the notes for which are available here. In this section we covered first a model of optimal defaults in the face of transaction costs and hyperbolic discounting by Carroll et al. [2009]. Next we took a quick look at Mike Woodford's magnum opus on rational inattention and reference dependence.

Problem Set 3 (for which you will need to look at this paper and this paper)

NEUROECONOMICSS:

Unfortunately, we didn't get the chance to talk about neuroeconomics, but here are this notes from last year which cover the topic, and in particular why neuroscientists think that there is reference dependence in the brain, in the sense that the dopamine system encodes a 'reward prediction error', or the difference between how good something was, and how good it was expected to be. The seminal neuroscience article on this is Schultz et al. [1993] (see here for a more recent neuroscience review). In the class (notes here), we discussed how axiomatic methods from economics could be used to construct tests for RPE, an idea explored in papers here, here and here.

Department of Economics

Columbia University, Rm 1031, International Affairs Bld, 420 W. 118th St., New York, NY, 10027, USA

mark.dean@columbia.edu

+1 212 854 3669