Wednesday, January 29, 2014

What if preferences are unstable?



A good rule of thumb, though not always true, is this: Modern ("neoclassical") economic theory has been pretty successful at describing people's decision-making patterns in static settings, but not so good at describing how people make decisions in dynamic settings. When people have to peer into the future to make their decisions, they do stuff that we don't really understand yet.

The key way that economists model behavior is by assuming that people have preferences about things. Often, but not always, these preferences are expressed in the form of a utility function. But there are some things that could happen that could seriously mess with this model. Most frightening are "framing effects". This is when what you want depends on how it's presented to you.

For example, suppose I asked you to choose between Skippy and Peter Pan peanut butter, and you chose Skippy. But suppose I  instead asked you to choose between Skippy, Peter Pan, and Jif, and you chose Peter Pan! This is really weird, since the first choice would imply that you preferred Skippy over Peter Pan, and the second choice would imply that you preferred Peter Pan over Skippy. The only thing that changed was the fact that there was Jif sitting nearby.

If this kind of thing happens often, then modeling human preferences is probably hopeless, because there's no way we can predict all the little factors that could affect a given decision. In other words, with framing effects, preferences will tend to be unstable.

Now one of the most important tools we have to describe people's behavior over time is the notion of time preference, also called "discounting". This means that we assume that people care about the future less than they care about the present. Makes sense, right? But while certain kinds of discounting cause people's choices to be inconsistent, other kinds would cause people to make inconsistent decisions. For example, some people might choose not to study hard in college, even though they realize that someday they'll wake up and say "Man, if I could go back in time I would have studied more in college!". This kind of thing is called hyperbolic discounting. It would make it a lot harder to model human behavior. But the models would still be possible to make.

But what would be really bad news is if people's time preferences switched depending on framing effects! If that happened, then it would be very, very hard to model individual decision-making over time.

Unfortunately, that is exactly what experimental economist David Eil of George Mason University has found in a new experiment:
Experimental and field research has shown that individuals often exhibit time inconsistent preferences. Often this is in the direction of “hyperbolic” or quasi-hyperbolic discounting. That is, individuals have a steeper discount rate for a given delay length when that delay comes sooner. This paper presents an experiment that tests this hypothesis with a novel choice task. Instead of being asked how much money today makes them indifferent between some amount later, subjects are asked how long they would be willing to wait to be indifferent between some amount sooner and some larger amount later. In this new task, many subjects exhibit “hypobolic” discounting, the opposite of the standard finding. This result does not appear to be a consequence of payout risk. The result suggests that hyperbolic discounting may be subject to the framing of choices, and therefore not purely an aspect of preferences. (emphasis mine)
Rut-roh.

Now, here's the thing...it gets worse. Eil, though a very careful and expert experimentalist, is not the only person to do experiments on time-discounting; it is a very common research topic. And I've heard whispers that a number of researchers have done experiments in which choices can be re-framed in order to obtain the dreaded negative time preferences, where people actually care more about the future than the present! Negative time preferences would cause most of our economic models to explode, and if these preferences can be created with simple re-framing, then it bodes ill for the entire project of trying to model individuals' choices over time.

This matters a lot for finance research. One of the big questions facing finance researchers is why asset prices bounce around so much. The two most common answers are A) time-varying risk premia, and B) behavioral "sentiment". But Eil's result, and other results like it, could be bad news for both efficient-market theory and behavioral finance. Because if aggregate preferences themselves are unstable due to a host of different framing effects, then time-varying risk premia can't be modeled in any intelligible way, nor can behavioral sentiment be measured. In other words, the behavior of asset prices may truly be inexplicable (since we can't observe all the multitude of things that might cause framing effects).

It's a scary thought to contemplate, but to dismiss the results of experiments like Eil's would be a mistake! It may turn out that the whole way modern economics models human behavior is good only in some situations, and not in others.

45 comments:

  1. You said: "For example, some people might choose not to study hard in college, even though they realize that someday they'll wake up and say "Man, I wish I had studied more in college!". This kind of thing is called hyperbolic discounting."

    Sorry, but this is not hyperbolic discounting. I may have perfectly consistent preferences and still wish I had studied harder in college or exercised more in the past etc. An individual with time inconsistent preferences will make plans concerning the future and then, when that future comes, choose different plans. Then, we must decide whether such an individual is naive (she thinks she will follow her original plan, i.e. she thinks she will act like an exponential discounter) or is sophisticated (she knows she will not follow the original plan because she knows she will act like a hyperbolic discounter). If the latter is the case, then she may use commitment devices to lock in her desired future plan.

    ReplyDelete
    Replies
    1. I think I just used clumsy phrasing. I changed the phrasing to make it clearer. Thanks!

      Delete
    2. I still don't think it's right, particularly because you are talking about a specific type of time inconsistent preferences due to hyperbolic discounting. I'm not sure why a hyperbolic discounter would have different feelings about the past than an exponential discounter. As I described above, the time inconsistency comes from making plans about tradeoffs between tomorrow (t=1) and the day after (t=2) and then, when tomorrow arrives, weighting t=1 over t=2 by more than when she was at t=0, which causes plans to change.

      Delete
    3. I think this time I've got the terminology right. Though thanks for your comments!

      Delete
    4. "Man, I wish I had studied more in college!"

      Personally, I wish I had spent more time chasing girls when I was in college.

      Delete
    5. Hypobolic discounting!!

      Delete
  2. Lots of researchers (though probably not enough) are going to look at this and realize it messes up their models or theories. How many are going to look at it and realize it causes problems for the philosophy that underlies economics? If utility doesn't work, I can't see how utilitarianism does.

    ReplyDelete
    Replies
    1. That's a good point, but utility still works well to describe some phenomena. For some applications you can just average utilities over time and not worry about the ups and downs. But I agree that it creates an interesting philosophical problem...

      Delete
    2. Anonymous7:46 AM

      Might want to check out some of Larry Temkin's work that goes into why comparative notions of 'better than' lead to intransitivity.

      Delete
  3. Hyperbolic discounting always makes me think of this:

    http://www.theonion.com/articles/amazing-new-hyperbolic-chamber-greatest-invention,1321/

    ReplyDelete
  4. Anonymous10:35 PM

    Gobs of new paper topics shouldn't really frighten young academics...

    ReplyDelete
    Replies
    1. Anonymous10:23 PM

      I had the same thought. Indeed, it seems like an easy way to gain notoriety and plum appointments would be to write a paper deriving the proposition "the behavior of asset prices may truly be inexplicable (since we can't observe all the multitude of things that might cause framing effects)" from some set of reasonable assumptions. Or better yet, figure out a trading strategy based on that proposition and start an investment management company.

      Delete
  5. Everyone's rational as long as you understand what they're optimizing, right?

    ReplyDelete
  6. I think others were trying to get at this, but the bottom line is that the issue of hyperbolic discounting is not the same as that of unstable preferences, both of which are important issues, but not identical. Hyperbolic discounting is a problem for standard theory as it implies time inconsistency, quite aside from the fact that they appear to be very widespread among many people empirically. But they can be stable over time, so not the same thing as "unstable preferences."

    In fact, stable hyperbolic discounting preferences, while undermining certain standard textbook stories, are more easily dealt with analytically than just plain unstable preferences, or worse, incomplete preferences. While we can turn into marketers trying to figure out how to manipulate peoples' preferences, the endogeneity of preferences to who knows what is lala land for economists. We are out of our league on this stuff, with even the most out there behavioral economists struggling mightily to say anything substantial.

    Marginally returning to hardcore micro theory, the assumption of completeness of preferences is the one that most people teaching them gloss over as obviously correct and not worth talking about further, when in fact it is the classic axiom of preference that is the least true of them all. Most people really do not know their own preferences, which tend to be very susceptible to sudden changes for all kinds of silly reasons, even if for most people their deeper underlying values may only slowly change.

    To pin this down more precisely, while we call them "indifference curves," in fact stating that two states of the world are on the same indifference curve is not a statement of "I don't care," as many presume, but rather an enormously strong statement that one knows for certain that these two possible states of the world make one exactly equally happy, a very strong statement that most people really are not willing to make most of the time about most realistic comparisons, quite aside from the fact that if they are willing to make such an assertion, it is likely to change within the next 10 seconds.

    Barkley Rosser

    ReplyDelete
    Replies
    1. I think others were trying to get at this, but the bottom line is that the issue of hyperbolic discounting is not the same as that of unstable preferences, both of which are important issues, but not identical...In fact, stable hyperbolic discounting preferences, while undermining certain standard textbook stories, are more easily dealt with analytically than just plain unstable preferences, or worse, incomplete preferences.

      I tried to get this point through in the post, but I might have oversimplified...

      Delete
  7. Schrödinger's cat. Indian or Chinese.

    ReplyDelete
  8. I thought all this time varying preferences was well known for decades:
    Buyer's remorse: I prefer A over B until I've purchased A. Then I decide B has higher utility.

    Not only does no one have perfect information, many people don't have perfect information about their own preferences/utility function! When you through in "which one will I be happier with next year"... rut roh.

    ReplyDelete
  9. Anonymous2:23 AM

    The problem with economic modelling/financial, IMO, is that it wants to uniformisé all the participants, without considering that each participant bears different situations, needs and aims.

    Risk appetite, consumption appetite. And sometimes, the heart wants what the heart wants and decisions are emotionally justified despite rational optimising dictates you to do.

    ReplyDelete
  10. Women especially. They change their minds way too much! Make no sense whatsoever.

    ReplyDelete
  11. Hi Noah, have you had a look at Timothy Wilson and Jonathan Schooler's work on preference instability? Their work was popularised by Malcolm Gladwell in his book 'Bink'.

    Here's two papers of theirs;

    http://www.anderson.ucla.edu/faculty/keith.chen/negot.%20papers/WilsonSchooler_Think2Much91.pdf

    http://personal.stevens.edu/~ysakamot/175/paper/wilson-lisle.pdf

    ReplyDelete
  12. "It may turn out that the whole way modern economics models human behavior is good only in some situations, and not in others"
    Is the purpose of the model: a) to understand? b) to control? c) to make a profit? (higher than in not using the model)

    For true "understanding", it's yet another complex additive. Insofar as understand implies predict, this result might make prediction more precisely less predictable.
    Control is based on predictability, so this result implies less controlability, at least thru peaceful economic non-coercive methods.

    Profit is also based on prediction and control/ influence. But here there's a very predictable response -- if some model of behavior is made that allows the model user to make extra profits, more people will behave like that model user and the modeled behavior relationships will change so that the prior model relationships are invalidated.

    People will, predictably in direction if not in time, change their behavior so that model users won't keep making extra profit.

    ReplyDelete
  13. I wouldn't call myself an expert on financial modelling of any sort, but what role does interdependent preferences play on the whole enterprise? In line with Keynes's beauty contest, many psychologists of the mind are running more and more to the ideals echoed in post-constructivist literature on identity, that how we perceive of ourselves is an act always with an audience in mind. My preferences are never my own in some isolated box, but always dependent on what role, situation, group etc that I'm in. And there are always competing roles, situations and groups for one to "decide" which is the correct reference frame to be aiming at. William Connolly, who has done a lot of work on identity in a political theory realm, talks about how communities are overlapping circles of belonging, think the US melting pot in a global scope. My role as a student overlaps with my role as a son, which overlaps with my role as a friend. Each have their own distinct preference set but they all overlap and conflict with one another at one time or another. I know Akerlof and Kranton wrote a book on identity in economics, but that's more in line with identity as a signaling device and less about how they shift reference frames, or a priori determine preference. How do these sort of role shifts and interdependency between the perceptions of what others think and individual preferences enter these modelling discussions?

    ReplyDelete
  14. Jamie9:07 AM

    This is interesting but it is also a good example of why academic economists who base their thinking on universal theories and experiments come to different conclusions from the rest of us who base our thinking on real-world examples.

    When real-world businesses and government organisations make procurement decisions, they like to base these decisions on a formal, auditable process. Government organisations, in particular, try to ensure that their decisions are fair and are not open to accusations of bias. Observing these procurement processes gives a good model of human decision making. So how do these processes work?

    First, organisations define a set of criteria for evaluating the offerings from the various suppliers. These will normally cover quality, price and risk. For example, the supplier’s track record may be used as one indicator of risk; the supplier’s longevity and recent profitability as another indicator of risk. These criteria are defined clearly so that they can be measured ‘scientifically’.

    An evaluation using these criteria may show that supplier A has the best quality service, supplier B has the cheapest service and supplier C has the lowest risk. These criteria may also interact. For example, supplier B may quote a price which is far below the other suppliers, and this may suggest a higher risk.

    The final choice of supplier is made using weightings attached to each of the criteria. For example, 50% on quality, 40% on price, 10% on risk.

    Standing back from this, you can see that although the individual criteria are evaluated scientifically, the weightings are decisive in the selection, and the weightings are mostly subjective. In some circumstances, people will seek to change the weightings after the evaluation if they find that the evaluation process has produced the ‘wrong’ result.

    Based on this model, when economists talk about preferences, they appear to be thinking about the scientific evaluation of each criterion but omitting the subjective weightings and the interactions of the criteria.

    Real-world individuals may not use a formal decision making process, particularly for small items. However, observed behaviour is mostly consistent with changes to weightings of fairly stable criteria.

    For example, yesterday I made a series of decisions on my evening meal. I had a choice of cooking for myself or going to a restaurant. I chose a restaurant. I had a choice of cuisine. I chose Indian. I had a choice of three restaurants. I chose one of them. I had a choice of meals from the restaurant’s menu. I chose a specific meal. My tastes in food and my economic circumstances are the same today. However, I am unlikely to make the same decision this evening. Why? First, I’d suggest that I will give different weightings to a number of criteria. What are my favourite foods? How hungry am I? How much time do I have to prepare a meal for myself? Am I in the mood for experimenting with something new? Does my route home take me past a specific restaurant? Second, I may have criteria which promote variety e.g. I may deliberately choose not to have the same meal on consecutive days, and I may also have criteria which promote randomness e.g. I may consider the recommendations of my friends.

    The processes I have described are rational. However, I doubt that they produce predictable results in any useful sense in the real world. There are too many interacting causes and effects, and there is too much randomness.

    ReplyDelete
    Replies
    1. Best comment I've read in many months!

      ", people will seek to change the weightings after the evaluation if they find that the evaluation process has produced the ‘wrong’ result."

      Bingo. "Blink" let's people know what their own, subjective, right result it. It often, in rational folk usually, follows their own known preferences (or is in the range of likely outcomes). But in cases where it doesn't, most folk change their general weightings to accommodate their desired result.

      This seems especially true in dating and marriage-making.

      And haven't you been following the "taste for Indian" and, on your way to the whimsically chosen place, decided "Mexican would be really really good right now" (or maybe it was Pizza?). Of course, if the lady (or any lady?) suggests a place or style, the mood usually changes to support that choice.

      Delete
  15. I do not know much about behavioural economics but am trying to think as I think. Based on my expertise of myself :) I am evaluating long term payments in orders of magnitude i.e. there is not much difference between 100 and 120 dollars but is between 100 and 1000 (obviously). Essentially discount models in the paper operate with real values of money, simple analysis show that if instead of money values you would use their logarithms the difference in experiment would not be observed. Also logarithm is a function for which power expansion (as I read it in the discount functions) will not work best and this is the closest of my feeling how I think about future money. Is there an empirical reason why logarithms are not used in this kinds of experiments?

    ReplyDelete
    Replies
    1. Actually this is a pretty well-known fact (people think logarithmically). But for some reason this fact has not made it into experimental economics yet.

      Delete
    2. Other then that, one thing lacking in the paper (which I just skipped) is the financial power of people in the experiment. I am not sure that one can use the same discounting curves for budget constrained individuals. I would definitely use different logic with money that solves nothing the money that solves something in my life. For example if the offer is 50.000 at hand or 100.000 in 3 years I will wait as this solves my long term housing problem in just 3 years. If the offer is 5000 now or 10000 in three years I will take 5000 as it immediately solves my transportation problems. If the offer is 50 now 100 in 3 years I am not sure what I will decide - neither sum solves anything so it is more the question if there is some small reward I want now and I take the money or if nothing pops I wait (which is more rational choice anyway).

      Delete
  16. Bill Ellis12:13 PM

    What if preferences are unstable? As a non-economist I just assumed that it was obvious that preferences are unstable and Malleable.

    ReplyDelete
  17. "It may turn out that the whole way modern economics models human behaviour is good only in some situations, and not in others."

    don't most economists think this already?

    ReplyDelete
  18. My wife retorted that economists know that her preferences do not in fact change over time.

    ReplyDelete
  19. Dang Noah, sometimes I just don't know. You've got these big artillery guns...and I'm your forward observer. I try and tell you how to adjust your fire...but I really don't think you're listening. It's like we're speaking a different language. But even when I put it in other people's words...your shots don't seem to get any closer to the target. For example...

    "Individuals do not act so as to maximize utilities described in independently existing functions. They confront genuine choices, and the sequence of decisions taken may be conceptualized, ex post (after the choices), in terms of "as if" functions that are maximized. But these "as if" functions are, themselves, generated in the choosing process, not separately from such process." - James M. Buchanan, Order Defined in the Process of its Emergence

    See also...

    Behavioral Law and Economics: Its Origins, Fatal Flaws, and Implications for Liberty

    The Knowledge Problem of New Paternalism

    Does Consumer Irrationality Trump Consumer Sovereignty?

    It's like you haven't even read Buchanan or Hayek. Even worse, it's like you haven't even read me! LOL

    Not reading Buchanan or Hayek is a mistake...just like worrying about rationality is a mistake. We all make mistakes...some more than others. Look, I even created a graph just for you...Evaluating Mistakes on an XY Graph. We're making a huge mistake (-10,-10) by not allowing taxpayers to choose where their taxes go.

    So keep firing away with your big guns! Hopefully you'll eventually hit the target! When that happens I'm sure you'll like the Tax Choice Party on facebook and encourage everybody else to do the same. Because there's no bigger mistake than not knowing what the actual demand for war truly is.

    ReplyDelete
  20. Anonymous5:44 PM

    There might be a way to avoid throwing the baby out with the bathwater. See Hayden & Ellis, "Law and Economics After Behavioral Economics," KULR (2007)

    ReplyDelete
    Replies
    1. Anonymous5:45 PM

      Online at http://www.law.ku.edu/sites/law.drupal.ku.edu/files/docs/law_review/v55/Hayden.pdf

      Delete
  21. Anonymous1:00 AM

    "It may turn out that the whole way modern economics models human behavior is good only in some situations, and not in others."

    Is this really not obvious?

    ReplyDelete
  22. I dunno... my immediate response whenever I read papers like these is to think "hmm, interesting, now add two or three zeros to each payoff and let's see if it still happens".

    Yesterday I went to the grocery store. My wife gave me a coupon for Brand X Pizza. I was thinking about economic rationality while shopping, really intensely, and mindlessly bought Brand Y, and my coupon was no good. I didn't notice till I got home. I was out of a dollar or so. And I prefer Brand X to Brand Y! Am I irrational? Am I a walking counter-example to economic theory? Or did I just make a silly mistake cuz the stakes weren't that high? Experimental subjects are (rationally) as lazy as the next person.(yes I know that some of these results have significant coefficients which suggests that the mistakes are not random. But that just means it's not that hard to design an experiment to trick people into making "wrong" choices when the stakes are small)

    I should also add that there are some experiments on certain behavioral phenomenon which I find convincing (for example, I do think regular hyperbolic discounting, or loss aversion, or ambiguity aversion are real). But there's also a publishing incentive to over do it. You know, like finding "puzzles" in macro.

    ReplyDelete
  23. In physics we don't know where any one atom is and its behavior appears random but we can still make useful statements about the behavior of large groups of atoms.

    The examples you use seem to focus on one decision by one decision maker. When there are multiple decision makers and each makes repeated decisions things may become more "rational".

    The question that occurs to me is whether micro-foundations that assume individual decisions have a large random factor lead to different macro consequences.

    I personally try to be rational but: (1) I operate under a belief system where I believe that much of the information I receive is false; (2) I recognize that there is a lot of uncertainty and I cannot assume that the people around me will be rational going forward; and (3) I am not really smart enough to rationally optimize under all that complexity and uncertainty. Given my intellectual limitations, I try to repeat what works, minimize downside risk and use simple rules of thumb.

    ReplyDelete
    Replies
    1. In physics we don't know where any one atom is and its behavior appears random but we can still make useful statements about the behavior of large groups of atoms.

      Yup. Random noise might cancel out across large groups of people. Then we're OK.

      BUT.

      What if noise is random in time, but correlated across individuals?

      Delete
    2. What if noise is random in time, but correlated across individuals?

      Indeed. If rational behavior in the face of uncertainty requires that you "follow the herd", or if you believe that what people do is a more reliable information source than what they say, then a random fluctuation could spread through the population and lead to bubbles and crashes. Answering that question could get you a Noble prize. I suspect that it would be much easier to show those types of effects using an agent based model with large numbers of discrete agents rather than trying to use a continuous distribution function to model the population.

      Delete
    3. Well, DeLong, Shleifer, Summers and Waldmann showed that when some subset of the populace is affected by coordinated noise, then the rational remainder can't stop them from wrecking financial markets!

      And there's a good bit of work on herding too, though not enough lately...

      Delete
    4. coordinated noise

      Noise is just false information. The way to head off that problem is to eliminate false information to the extent possible. That is why bankers and auditors should be going to jail on mass for the real estate bubble and financial crash.

      Delete
    5. Not necessarily! Noise could be ex-ante unpredictable preference changes.

      Delete
  24. What if economics makes sense when agents' behavior doesn't matter?

    http://informationtransfereconomics.blogspot.com/2014/01/it-doesnt-matter-what-people-think.html

    ReplyDelete
    Replies
    1. Should have an "only" before "makes sense".

      Delete
  25. Out of curiosity Noah...where did you get the image from, and who created it? Also, what do you make of Paul A. Samuelson's notion of "revealed preference"?

    ReplyDelete
  26. I agree with your proposition. Time preferences are always framed in terms of individual's expectations of their future wealth. As such expectations change, so do their time preferences. If people fear that they will be poorer in the future (due to job prospects, stock market crash, uncertainty as to the future of social security and medicare and a host of other reasons), their time preference will be negative. Accordingly, they will arbitrage negative ROE by holding cash at 0%.

    I made a similar comment to your post on the Consumption Euler Function. I attempted to extract time preference from widely-available monetary aggregates such as M2, V, Monetary Base, Excess Reserves and Inflation. Using IS/LM framework, I derive a data set going back 55 years which exhibits strong negative correlation to real consumption growth over the period. I am interested in your thoughts, critique... Data set here http://tinyurl.com/qzp4h3f

    ReplyDelete