Monday, September 16, 2013

Are Humans Rational Utility Maximisers?

Paul Krugman's recent post on Wynne Godley involves a discussion that Yichuan Wang touched on this week — the widely-used assumption that people are rational utility maximisers. Krugman notes that the alternative approach — the sectoral rules of thumb approach embraced by Wynne Godley and many other post-Keynesians — was abandoned by mainstream macroeconomists in favour of other approaches:
So why did hydraulic macro get driven out? Partly because economists like to think of agents as maximizers — it’s at the core of what we’re supposed to know — so that other things equal, an analysis in terms of rational behavior always trumps rules of thumb. But there were also some notable predictive failures of hydraulic macro, failures that it seemed could have been avoided by thinking more in maximizing terms. 
First involved consumption spending. Conventional Keynesian consumption functions suggested that the savings rate would rise as incomes rose — and this wasn’t just the Keynesian interpreters, Keynes himself made the same claim. This, in turn, led to predictions of rising savings rates after World War II, and hence a persistent shortage of demand — hence the secular stagnation theory briefly prominent. (There was even an early Heinlein novel built in part around the secular stagnation theory. As I recall, it was pretty bad.)

In fact, however, savings rates don’t seem to follow the naive consumption function at all; they rise in booms, and are higher for the wealthy, but exhibit no secular trend. And Milton Friedman appeared to explain this paradox by arguing that people are more less rational: they base consumption on “permanent income”, a reasonable estimate of long-run income, and save temporary fluctuations in income.

The second big problem involved inflation. We can argue how many economists really believed in a stable tradeoff between inflation and unemployment, but that’s certainly what got taught to many students. In came Friedman and Phelps to argue that rational price-setters would build expected inflation into their choices, so that sustained low unemployment would produce accelerating inflation. And the stagflation of the 70s seemed to vindicate their argument.

You could argue, and I would, that the rebellion against hydraulic macro went much too far. It’s not at all clear how much good the whole apparatus of maximizing behavior in New Keynesian models really does, and to the extent that such models do seem more or less to work, it’s only by making some ad hoc behavioral assumptions that are grafted on to the rest of the structure. 
Krugman's points regarding the specific approaches of post-Keynesians of the day are reasonable. There emerged empirical observations that their models did not explain, that could be explained by alternative theories. On the other hand, that doesn't necessarily say anything about a rule of thumb or a sectoral approach to macroeconomics. Many approaches can be used to create a model consistent with data, and similar methodological approaches can lead to wildly divergent models with wildly divergent predictions through varying assumptions.

Post-Keynesian ideas — and those of other heterodox schools that made noise about the possibility of the crisis in the early and mid 2000s — have regained a certain amount of public spotlight in recent years, (and in some cases, almost immediately lost the spotlight by making subsequent bad predictions). On the other hand, they do not appear to have gained much academic credibility, perhaps because claims to have "predicted the crisis" are not as convincing to an academic audience as they are to a general one.

The key difference between post-Keynesian sectoral theories, and neoclassical micro-founded ones is in their assumptions. Post-Keynesians using Godley-style models make behavioural assumptions about group behaviours in different sectors; workers, the financial sector, etc. Neoclassical theories make assumptions about individual behaviour — individual utility maximisation, rational preferences and action based on full and relevant information. In both cases, these are all simple heuristics and should probably not really be taken seriously as an attempt to model reality in a literal sense; we should as Milton Friedman famously noted be concerned most by a theory's predictive power. On the other hand, it does not seem desirable to start off with a framework that is obviously wrong. That is, a good theory should have predictive power at both ends — it should start with an accurate (-ish) depiction of reality, and end with successful predictions of reality.

It is arguable that the notion of utility maximisation falls into the category of obvious wrongness. Krugman's conclusion that "It’s not at all clear how much good the whole apparatus of maximizing behavior in New Keynesian models really does" suggests that he may be cautiously leaning in that direction too.  There are two reasons for this — evidence from behavioural economic experiments, and results from neurophysiology. 

One of the most famous tests of Paul Samuelson's definition of utility maximisation (completeness, transitivity, non-satiation, convexity) was undertaken by Reinhard Sippel in 1997. He gave his student test subjects a budget, and a set of eight priced commodities to spend their budget on:

This was repeated ten times, with ten different budget and price combinations. Sippel found that 11 out of 12 of his test subjects’ behaviour failed to meet Samuelson’s criteria for rational utility maximisation. Sippel repeated the experiment later with thirty test subjects, finding that 22 out of 30 did not meet Samuelson’s criteria. Sippel concluded:
We conclude that the evidence for the utility maximisation hypothesis is at best mixed. While there are subjects who appear to be optimising, the majority of them do not.
The evidence emerging in neurophysiology against utility maximisation is more damning. As Daniel McFadden summaried this year in The Atlantic:
Our brains seem to operate like committees, assigning some tasks to the limbic system, others to the frontal system. The “switchboard” does not seem to achieve complete, consistent communication between different parts of the brain. Pleasure and pain are experienced in the limbic system, but not on one fixed “utility” or “self-interest” scale. Pleasure and pain have distinct neural pathways, and these pathways adapt quickly to homeostasis, with sensation coming from changes rather than levels. Overall, presumably as a product of evolution, our brains are organized well enough to keep us alive, fed, reproducing, and responsive to but not overwhelmed by sensation, but they are not hedonometers.
None of this, of course, should be treated as a conclusive blow for any other extant modelling approach or set of assumptions. One reason why neoclassical economics has remained dominant in academia is that other heterodox schools have not yet convincingly envisaged anything conclusively superior Yet, I think, it should be treated as a conclusive blow for the idea that a lot more work is required in coming to a better understanding of the patterns and systems underpinning human economic behaviour. And I think the answers will emerge not from theory but from experimental economics and psychology on the interpersonal level, and from neurology and neurophysiology on the individual level.

This is something that Krugman (and all wannabe Hari Seldons) may eventually be able to get very, very excited about. The only road, I think, to anything approaching psychohistory is mechanically understanding the underlying drives and dynamics of human behaviour, rather than making crude black box assumptions and hoping for the best. 


  1. In one-shot decisions where you can easily quantify what people want, rational utility maximization is a great model. This is why auction theory works so well.

    In one-shot decisions where preferences are likely to be vague and unstable and subject to framing effects, such as the Samuelson example you gave, rational utility maximization is an OK approximation. It fails for a lot of reasons, but those reasons are hard to predict, and there are lots and lots of them, so even though rational utility maximization only describes some of what's going on, it's tough to consistently do better.

    In dynamic situations where forward-looking behavior under uncertainty is required, rational utility maximization seems to explain only a very little bit of what we see happening in the world.

    So I think the useful question is: Can we know when rational utility maximization is a useful model, and when it's not? I am hopeful that we can.

    1. Noah — I agree with your designations, and in the post probably should have discussed the difference between experiments and the macroeconomy, and noted that it adds a whole other layer of problems.

      If we agree to consider the macroeconomy as a dynamic situation where forward-looking behaviour under uncertainty is required, I think the answer to your question is that in models of the macroeconomy rational utility maximisation is not a useful heuristic, and different heuristics are needed.

    2. Yep. I think in macro, utility maximization rarely adds to any model.

    3. I've been reading a couple of blog-posts and articles on 'economic thought' today (Tyler Cowen's etc) and I am genuinely curious.

      Is there so many unknown thing in Macro? I know of some practical curiosities such as the relatively low unemployment in the UK (mostly a question around the quality of our data and how/what we actually measure, I think) but things like economic activity, inflation, deflation etc seem relatively settled.

      It's mostly a matter of political practicality, no?

  2. Anonymous10:45 AM

    Sippel is hardly the only paper that has tested revealed preference axioms. See the second section of for a nice, short summary of the literature.

    1. "More recent work by Burghart, Glimcher, and Lazzaro (2012) explores the degree to which alcohol impairs an individual’s ability to choose consistently with GARP. Surprisingly, even highly inebriated subjects’ choice behavior appears to be fairly consistent with axioms of revealed preference."

      Doing that experiment must have been fun :)

      Anyway, reading through the lit review, it sounds like even when violations of revealed preference are found, they are not very quantitatively significant. And especially with lab experiments, there is always the question of external validity.

  3. Anonymous10:53 AM

    Also, the neuro evidence is also not as damning towards the standard model as you might expect. Check out the work of Glimcher, a neuroscientist at NYU.

    For example, in this paper (, he asserts that "Economists have long argued on theoretical grounds that humans typically make these choices as ifthe values of the options they consider have been mapped to a single common scale for comparison. Neuroimaging studies in humans have recently begun to suggest the existence of a small group of specific brain sites that appear to encode the subjective values of different types of rewards on a neural common scale, almost exactly as predicted by theory."

    Also, it might be fun to check out his paper "An expected utility maximizer walks into a bar..." located here:

  4. Anonymous10:57 AM

    Neuroscientists love to talk about behavior being 'optimal' (usually in a Bayes sense, though often in an information-theoretic or control sense as well). Typically, deviations from optimality are accounted for by the constraints upon the system: oh, this behavior really is optimal if you assume it has to do xyz as well. I think utility-maximization falls into that category for us as well. Individuals maximize utility, assuming that the individual is also taking into account xyz expectations that you haven't thought of yet.

    I'm not sure how damning the 'multiple systems' problem is. After all, there may be many different systems for pain and pleasure, but in the end they have to come together to produce a single action, right? Which represents some form of utility.

    Actually, I think the bigger problem is the definition of utility is so vague and meaningless that I can't even figure out how to put it into a neural context.

    1. Anonymous9:53 PM

      Progress has been made on the neural basis of utility. Glimcher takes about it at length in his provocatively titled "Foundations of Neuroeconomic Analysis." It's not a light read but it's a very interesting one.

      Real progress is being made on giving a firm basis to concepts that have been stuck in the mud for a long time. Exciting stuff.

  5. Maybe the utility function is wrong.
    Is 30 minutes of unknown reading material equivalent to 30 minutes of interactive play.
    How does 100 ml of sweet liquid compare to 10 minutes or reading, or 100 gm of chocolate?

    My experience is that the ranking of the 'goods' will change with time, depending on my mental
    state and eating history. Sequential choices will be based on differing valuations. I would
    probably start with a shot of liquid followed by food followed by a game followed by reading.
    But my sequence would change depending on my initial state.

    Of course the goal is not to predict individual behaviour, but aggregate behaviour of a large
    population. A different problem. And the words used to describe the behaviour may not
    be appropriate. The real driver of the system could be an advertising program attempting to
    define the 'utility' function.

    When making a model based on known incorrect assumption with probably incorrect
    descriptors, you should not be too surprised if the 'atoms' do not behave.

    Remember phlogistone and the flat earth.

  6. Hydraulic Keynesian is "Old" Keynesianism. Contra Krugman, Post Keynesianism is something else.

    Mainstream macroeconomic models are not micro-founded. See the literature on the Sonnenschein-Mantel-Debreu theorem. Alan Kirman, in that literature, recommends making assumptions about groups of agents, just as Post Keynesians have been doing for more than half a century. See, for example, macro-economic theories of distribution put forward in the late 1950s and early 1960s by Kaldor, Kahn, and Robinson. Apparently Frank Hahn's doctoral thesis belongs in this literature too. (I am not claiming that these models can be directly applied, empirically, to any economies today. Social norms have changed quite a bit, and the theories must be updated to reflect those changes.)

    The use of heuristics in decision-making is also a question of microeconomics, not solely a question of how to micro-found macroeconomics. Quite a bit of literature, both mainstream and not, has been recently generated in economics, has been generated on divided minds or Faustian agents, which might be of interest here.

    I do not think Noah was fair to Keen in the linked post.

  7. Anonymous11:34 AM

    Behavioral economics brings Kahneman and Tversky to mind. A lot of experimental evidence by them and others shows that the "Independence Axiom" is what gets violated (a fundamental assumption of Von-Neumann Morgenstern Expected Utility Theorem) because of heuristic biases, the framing effect and so on. Fortunately or unfortunately, the Probability Space of Expected Utility that these two designed is still largely used in conventional microeconomic theory. So what we really need is an alternative to this way of modelling decision theory/uncertainty.

  8. scepticus1:13 PM

    Has anyone come across the 'Maximum Entropy Production Principle'. It has emerged as a leading theory in ecology and geology and is one of a number of 'extremal principles'. The claim is that planetary systems (with or without life) self organise to maximise their rate of entropy production. Note that the whole theory is a branch of non linear thermodynamics which is very different to the classical equilibrium thermo of clausius et al. Also, it is not the same thing as entropy maximisation as a statistical tool. It certainly is the mother-of-all-heuristics though.

    Anyway, proponents of the theory claim that it extends to all complex systems and to all life and ecology and would thus also apply to human macroeconomic evolution. Deriving models of individual behaviour using this approach is of course unlikely to be successful, but then that would not matter because accepting this approach requires one to accept that what is being optimised for cannot be located within the individual. While this particular theory may be wrong or may be rejected on various grounds, the possibility the the locus of optimization is some way above the individual seems reasonable.

    Of course mentioning these kinds of words on economics blogs is deeply unfashionable and likely to have readers reaching for the ignore function - yet these concepts are at the forefront of geological, ecological and astrophysical research (not to mention having related theories in statistics) and are represented by leading figures in these areas.

  9. So, you are saying that Neoclassical Econonomics must dominate the University because they are more brilliant, or more fashionable modelling, in spite of its absolute no eficiency?

    1. No, the opposite. If there is one particular school I am closest to, it is probably post-Keynesianism. I don't think rational utility maximisation is a good way of framing human action in the context of the macroeconomy. But there's still a lot more work to be done in terms of building something better.

  10. Zathras2:29 PM

    I'm as much of a critic of utility maximization as anyone, but I cringe whenever I read something like "neurophysiology shows...." The quote above is nonsense. Whenever you read a pop psychology statement like above and compare it to what was actually seen in the data, you get to appreciate the mostly-fictional account of these pseudo-scientific statements.

  11. Anonymous7:27 PM

    According to rational expectations, sales and marketing makes no impact.

    Most industries spend billions on sales and marketing, and zero on economists.

    Who's wrong?

    1. I like this comment.

    2. "According to rational expectations, sales and marketing makes no impact." Why? I can come up with plenty of models with rational expectations in which sales and marketing make an impact. Rational expectations does not preclude informational frictions and does not preclude changing preferences. People tend to believe (incorrectly) that rational expectations means that everybody knows everything about everything.

      By the way, which industries spend zero on economists?

    3. Carola — You're right that Anonymous is taking rational expectations to a silly conclusion (I found it quite amusing) but if the frictions that you're introducing into the model are the thing that make the model begin to resemble reality, then I'd say that makes the underlying assumptions quite vulnerable to revision.

      Most empirical papers I've read on rational expectations suggest it is a pretty unrealistic assumption. For instance:

    4. ... what is their alternative, more realistic assumption?

    5. Bill Ellis1:59 PM

      Great point.
      Some Advertising supplies information needed to make rational decisions, but most of it is designed to work on our irrational/emotional needs. It is very effective.

      If for some bizarre reason there were two huge year long ad campaigns on the topic of which is better, the number 43 or the number 29, we all would end up with an opinion on it...even though there is no rational reason to favor one over the other.

    6. Anonymous12:56 AM

      @Carola: that one can come up with "plenty of models" where sales & marketing provides rational actors with informational "lube" for optimal decision-making makes me giggle.

      Q: Which industries spend zero on economists?
      A: Competitive ones that require customer acquisition.

      @Tom Grey: Most marketing practitioners do segmentation and conduct field testing of real consumer response models to measure the coefficients, instead of assuming a homogenous backwards-looking utility optimization function and stroking your beard for further insights.

  12. Insofar as the key value economic / scientific theory has is the ability to predict events, all are right to question the "science" that failed to predict the House price bubble pop (2006) and the resulting financial crisis. [Dr. Mike Burry did predict it, and bet/ invested millions on it coming in his Big Short.]

    "One reason why neoclassical economics has remained dominant in academia is that other heterodox schools have not yet convincingly envisaged anything conclusively superior"
    This can be quantified, and possibly should be.

    " Sippel found that 11 out of 12 of his test subjects’ behaviour failed to meet Samuelson’s criteria for rational utility maximisation." Rather than a 100% yes or a "not met", the behavior of the 11 non-maximizing subjects was probably over 90% as compared to a pure randomized score of 50% (or 0%?).

    Let's say the average of all 12 was 90%.

    Then any alternative theory of individual behavior would have to predict the actual behavior of the subjects with some greater than 90% score.

    And here's the other issue of Not Quite Maximum rationality -- the non-rationality of each individual is likely to be closer to random.

    I propose that decision making people are a combination of about 90%-60% rational and 10%-40% idiosyncratic.

    Better micro-foundation models will occur when modelers learn how to model that 10-40% irrationality.

    Marketers & Sales people are already doing this.

    comment #2: Were the experiment repeated based on a shopping basket of 8 disparate items that each individual had been frequently buying over the last year (as evidenced by itemized shopping lists plus entertainment), so that while each basket had different items, the relevance of those items to the decision makers would be similar. (Maybe one has peanut butter, but another has nutella.)
    I'd guess that over multiple price changes, the rationality % would be higher and more would be at 100%.
    Because people are more rational about things they care about/ have more experience in.
    Which is why "pretty rational" expectations in models will dominate other theories' for prediction.

  13. Bill Ellis1:17 PM

    On topic..."Bad Decisions Don’t Make You Poor. Being Poor Makes for Bad Decisions.
    New research shows that worrying about money causes cognitive impairments."
    By Matthew Yglesias.

    Link to paper Matt talks about....

  14. I guess that the ordinary consumer does not optimize, but big businesses and banks do, because it's what they're supposed to do. Ordinary consumers probably use rules-of-thumb.

    I guess microfoundations in macroeconomy, at the moment, are only of academic interest. Macroeconomic models without microfoundations, but with predictive quality, are more useful in the political sphere. Maybe that will change if microfounded macromodels achieve a decisively increased predictive quality.

    1. Bill Ellis2:12 PM

      Executive compensation has gone through the roof as workers wages have stagnated.
      If businesses actually rationally optimize, than the advantage gained by caping workers wages should also apply to executive pay.

      No mater the level, people's emotional needs will distort rationality.

    2. I guess that the ordinary consumer does not optimize, but big businesses and banks do, because it's what they're supposed to do. Ordinary consumers probably use rules-of-thumb.

      This seems to be true more than with consumers, but not always. Here's an example of an ultra-successful and ultra-profitable (although it wasn't always) company acting, well, strangely.

      Sometimes things that people consider to be acting rationally isn't what an economist would define as rational. And sometimes people and companies just do weird things for weird reasons.

    3. Well, Apple always had, even in their darkest years, the advantage of being a lifestyle choice, the brand of the digital elite, if you will. Turning that advantage into a business model was not weird. Their design was part of the Apple package and worked well together with Jobs' preference for easy handling.
      But yes, the definitions of rationality may differ. The high compensation for executives was also justified by competition for talent. And if a bank managed to attract talents who repeatedly beat other bankers at derivative bets, an extra million spent on salaries might be justified by even higher profits. So the bank would act rational, but the bank on the losing side would gain nothing and the economy has no benefit at all.

  15. Anonymous9:43 PM

    I found Robert Waldmann's comments on the Permant Income Hypothesis here"

    quite interesting. Short summary: he finds it unsupported. To the extent that the case against rules of thumb is based on the PIH it's on shaky ground.

  16. There was a guy and a school of thought that followed trying to base their economic thinking in terms of human action, a priori reasoning and strong Aristotelic deductionism. That was Ludwig von Mises, which you kind of disdain indirectly by making fun of Schiff, but it seems that you're alluding to something very similar, using different words.

    Although I'm no Austrian myself, getting back some of the aprioristic method of science instead of the current ultra-empiricism that seems to govern Economics seems just right. As Kahneman has shown, sometimes we deviate from the utility maximization-rational expectations frame of mind, but the question is whether we correct such biases and adjust behavior accordingly or we continue hopelessly biased (probabilities is a good example).