Sunday, March 25, 2012

Guest Post 2: Blue Aurora on Econophysics

This is Blue Aurora's second installment. It's worth pointing out that his first post elicited some other commentary in the blogosphere. Gene has a humorous treatment of uppity physicists here. Jonathan goes into some Austrian critiques of economics as a predictive science, and the relevance of this point for econophysics . Regular readers know I agree on this point - that I like to think of economics primarily as an explanatory science, just like any study of a complex system.

There was some comment section discussion of statistical distributions that I was unfortunately not able to participate in. I think the point that a lot of people who criticize assumptions of normal distributions miss is that a lot of what they're talking about are assumptions about normally distributed parameter estimators. As far as I am aware, we can assume these estimators are normally distributed - they are random variables themselves, after all. That's a very different proposition from assuming that whatever distribution these parameters contribute to are normally distributed. An estimator for a mean is going to be normally distributed regardless of whether the distribution that generated that mean is normally distributed or not. Anyway - on to Blue Aurora
*****
MOVE OVER, MANKIW, KRUGMAN, AND BERNANKE


Why do I list this cast of economics characters? Because whatever your strand of economic thought, be it Austrian, Feminist, Old Institutionalist, New Institutionalist, Market Monetarist, Neoliberal, Post-Keynesian—do call yourself what you like—but the bottom line’s the same: There’s a new show in town, and it’s dangerous.

Picture, if you will, the orthodox citadel under assault by a bunch of physicists armed with a concentration in statistical mechanics (not to mention a knack for “non-stationary increments”) and Mandelbrot’s fractal geometry. Robert Lucas attempts to draw out his eponymous Critique. It proves futile, as the econophysicists point out that Lucas had erred in his conclusion from a “scientific standpoint” (see page 224 of “Dynamics of Markets: The New Financial Economics” by Joseph L. McCauley). Bernanke doesn’t take long to bow thereafter. Why hasn’t this scenario happened yet? Two simple reasons—the lack of widespread knowledge, and the fact that the econophysics project hasn’t grown that strong just yet.

So far as I have been able to tell, the econoblogosphere hasn’t commented extensively on the so-called “econophysicists”. This is in spite of the fact there is a dedicated blog on econophysics at an econophysics forum, econophysicists who are based at respectable institutions (for example, Jean-Philippe Bouchaud teaches at École Polytechnique), and even the occasional media mention in the movement’s fifteen years—for instance, in a 2005 article on The New York Times and in the Sept/Oct 2009 issue of Adbusters. Okay, there’s Cosma Shalizi’s little rant at Three Toed Sloth, the Gallegati-Keen-Lux-Ormerod “Worrying Trends” paper in Physica A, and Jean-Philippe Bouchaud’s article in Nature. Leading econophysicists Imre Kondor and Rosario N. Mantegna have received grants from the Institute for New Economic Thinking. The Austrian School-sympathizing scholar of quantitative finance, Nassim Nicholas Taleb, has referred to leading econophysicists Dr. Didier Sornette and Dr. Jean-Philippe Bouchaud, in his popular works. Of course, with the instrument that is the World Wide Web, in due course the econophysics project will receive more attention—but for the moment, they have not.

Here and there, they have been referred to on the blogosphere—see the March 3, 2011 entry on online blogger Lord Keynes’s website, “Social Democracy for the 21st Century: A Post Keynesian Perspective”, for instance, and Robert Vienneau’s July 13, 2006 entry. I’ve also encountered some comments on the forum of the Ludwig von Mises Institute. Why is the commentary so dispersed and not so frequent, despite the World Wide Web? A culture clash between the econophysics project and the economics profession is one reason. So would the youthfulness of the econophysics project (only being fifteen years old!), the fact that the natural sciences and social sciences have been interacting for the past two centuries aside—Jan Tinbergen studied physics at the undergraduate level, and so did Paul Samuelson, but there is something different from these previous translations of phenomena in the natural sciences into the social sciences. I figured there was something different to them once I read the article referenced in the bibliography of a post written by Lord Keynes of Social Democracy for the 21st Century.

Lord Keynes refers to a July 2009 article published in Elsevier-owned “Physica A: Statistical Mechanics and its Applications” called “Economic Uncertainty and Econophysics”, written by one Christophe Schinckus. Three renowned economists that touched upon uncertainty covered in the subject of the article were none other than John Maynard Keynes, Friedrich A. Hayek, and Frank H. Knight. Schinckus claims in his article abstract: “By presenting econophysics as a Knightian method, and a complementary approach to a Hayekian framework, this paper shows that econophysics can be methodologically justified from an economic point of view.” I strongly feel that the econoblogosphere needs to discuss this issue in much greater frequency.

Thus far, it has been rather dispersed and all over the place. However, in my quest to learn how my fellow bloggers would make of econophysics, I prodded around and linked to the econophysics forum. One blogger had an initial reaction that indicated some amusement and flippancy: “What does economics have to do with physics? Sounds like a revival of Comte.”

Not taking it seriously initially, I then linked to Jean-Philippe Bouchaud’s article in Nature. He
proclaimed that “In short, [Bouchaud] doesn’t know what he’s talking about.” Then he cited “The Epistemological Problem in Economics”, an essay by Ludwig von Mises, to counter Jean-Philippe Bouchaud’s essay. I countered that the econophysicists have listened to outside criticism referring to the Gallegati-Keen-Lux-Ormerod paper) and have made an impact on finance. As of this writing Jean-Philippe Bouchaud is still the editor-in-chief of the journal Quantitative Finance. Dr. Bouchaud’s methods have set a precedent. Future financiers will have to listen—and
following the example of the econophysics project, adopt an empirically-based approach that rejects a priori theorizing, but stand up in the face of a financial crisis.

It was at this point In the following interactions I had with him, the Austrian School disciple, Jonathan Finegold Catalan, pointed to the “lack of controlled experiments” that “distinguishes physics from economics”. But there is a lack of controlled experiments in astrophysics and geophysics, and they are considered empirical sciences! Observation, as Catalan should know, is a part of the scientific method. Going by Catalan’s logic, astrophysics and geophysics would be unscientific. What makes econophysics scientific? The techniques of the econophysics project have yielded solid results. While predictive power is obviously secondary to explanatory power, what Dr. Bouchaud, Dr. Stanley, and their colleagues have been doing seem to hold up! The techniques of statistical mechanics are designed to analyze and describe non-linear, random processes. Human decision-making processes are non-linear and non-additive, as demonstrated
by the works of cognitive psychologists and countless other scholars. The techniques of statistical mechanics so far, hold up. The explanatory power of the econophysics project may well prove more fruitful than current methods. But enough with the turf wars over method. I suspect that economists could use an even bigger toolbox if they are to avoid being completely falsified by the
econophysics project.

Heterodox economists Mauro Gallegati, Steve Keen, Thomas Lux, and Paul Ormerod are a few economists who have engaged the econophysics project. All four of them have published in Physica A. But engagement of the econophysics project doesn’t have to begin at the fringe (see the special issue in the Journal of Economic Dynamics and Control mentioned above). John Sutton, a scholar at the London School of Economics and Political Science, is one of the few orthodox economists aware of the econophysics literature. So there is still time for the economics
profession to cooperate with the econophysics project and ultimately, learn. Unless of course, Dr. McCauley and Dr. Bouchaud have their way, Brad DeLong and Scott Sumner wouldn’t be debating estimates about the multiplier effect anymore. There wouldn’t be debates between the New Classicists and the New Keynesians. Instead, according to the world of Dr. McCauley and Dr. Bouchaud, both Sumner and DeLong shall be relegated to the role of Ptolemaic astronomer. Dr. McCauley accuses Keynesianism and Monetarism of becoming “ideologies”.) And it seems that the econophysicists—with their Mandelbrotian-Osbornian-statistical mechanical analysis—have the tools to potentially falsify virtually all of economics. One way to deal with this problem would be to take all of the Great Moderation Consensus to aggressively critique and attack the econophysics project. But I feel that reaching out to the econophysics project en masse will have a far more positive effect, and prevent economics from being falsified completely.

For somewhere beneath the acerbic stance taken by the more outspoken econophysicists like Dr. Bouchaud and Dr. McCauley, there just might be a point lying somewhere from which a new economics can evolve.

16 comments:

  1. My argument is not just about controlled experiments, although this is still relevant. My criticism is based on the fact that human action is unpredictable. It's not just about non-linearity, but about the fact that the possible patterns of actin are almost infinite.

    I think you might be able to make very broad generalizations of what will occur, but you can make these broad generalizations by using a methodology that also represents the economy in a more realistic way (or that gives you a better ideal type to model the market process). I don't think statistical mechanics does that. Statistical mechanics seems like the re-birth of the worst branch of positivism.

    ReplyDelete
  2. Statistical mechanics is a sub-branch of physics that has existed long before either of us were born, Jonathan. It's only been in the last twenty or so years that late 20th century statistical mechanics was applied to economics and finance to form "econophysics".

    Econophysics does depict the economy realistically because it uses non-linear approach.

    ReplyDelete
  3. "have the tools to potentially falsify virtually all of economics"

    Well, I can't fault you guys on ambition.

    ReplyDelete
  4. Blue aurora, I'm not sure what the history of statistical mechanics has to do with anything. The problem with probability and human action is that uncertainty cannot be assigned probability, like risk. We are talking about the genuinely unknowable.

    Did you read my post? There are important differences between the limited movements of physical particles and the unlimited and unknowable scope of human action.

    ReplyDelete
  5. Jonathan, I was correcting you on your failure to distinguish between statistical mechanics and econophysics.

    I disagree that human action is "genuinely unknowable". Yes, I did read your post. But statistical mechanics does not deal with a deterministic system. It deals with non-linear systems, which allow for infinite possibilities. Therefore, econophysics is up to task with dealing with human decision-making processes.

    BTW, Dr. Brady would disagree with your assertion that uncertainty cannot be assigned any probability. That is akin to saying that the weight of evidence is zero.

    ReplyDelete
  6. "An estimator for a mean is going to be normally distributed regardless of whether the distribution that generated that mean is normally distributed or not."

    Well, kind of. Even if the underlying data is not normally distributed, the central limit theorem says that the mean will be _asymptotically_ normal (under some assumptions). The rate of convergence (in the sense of "largest difference between cdfs") depends on the distribution but, in the IID case, is bounded by O(n^(-1/2)).

    If the underlying process has a heavier tail than the normal distribution, so will the mean of a sample. For example, this could mean that your 95% confidence interval really only covers 94% (or 94.9%). If what you are doing depends on your 95% CI really being a 95% CI, this could cause problems. I don't know enough about financial math to be sure of my opinion, but I think this is an valid critique (there may be other, more important, critiques). I think it matters much less outside of high finance (because the stakes are lower).

    For linear models fit by OLS, assuming normal errors, the coefficients are linear combinations of normal distributions, so are normal (and under violations to normality, we have asymptotic normality by the CLT). For more sophisticated models, like mixed-effects models, the coefficients are only asymptotically normal even when the error terms are normal. Many people (myself included) view it as a not-very-good approximation (it tends to produce too-small p-values). Though I just use bootstrapping or Bayesian methods, when it really matters.

    ReplyDelete
  7. Daniel,

    According to the CLT, appropriately normalised sample moments are asymptotically normal. The OLS estimator is a function of sample moments so we can appeal to the CLT and the delta method for its distribution.

    Blue Aurora,

    If you want to make alliances and spread your research then it might be wise to adopt a less condescending tone. Economics has been developed over hundreds of years. Who knows--if you are more open minded, it might even be possible for you to learn something from it.

    ReplyDelete
  8. Vimothy: I am an econophysicist, but merely someone with an interest in econophysics. I wasn't intending to come off as condescending, more like polemical. I understand that economics has existed for hundreds of years - but I have a feeling that it's possible that these econophysicists might be right on a number of things.

    ReplyDelete
  9. I meant to say, "not an econophysicist".

    ReplyDelete
  10. Blue, You achieved Polemical. I have already commented on your part 1 in a similar manner.

    By 'economists' I assume you mean those implicated in 'the Great Moderation scandal’. My own candidate for a new economics would be to build on Keynes’, which would naturally include many of the analytic ideas of econophysics. From a trading perspective, this new economics could look like econophysics, but policy would need the full version, drawing on the tools of econophysics but applying its own (broader) contextualisation.

    If economists were complicit in the Great Moderation’s digging of a hole, heedless of the potential for a collapse, then many would regard econophysics as about digging quicker, and prefer the older, slower, ways. But what we need, perhaps, is not just ‘new’ or ‘more efficient’ but more effective economics.

    ReplyDelete
  11. A challenge for any theory of economics is to establish an appropriate notion of uncertainty. Keynes' criticisms of statistical mechanics and Laplace's criticisms of the general approach need addressing.

    Suppose that I have a model with agents who adapt their actions (e.g. trading) to maximize some notion of value, taking account of ‘news’. If the agents have a set state-space and programme then this is an econophysics type model.

    Statistical mechanics is concerned with the emergence of overall behaviours from the characteristics of agents and the space that they are in. I can work with, for example, traders or policy makers, seeking to develop models that explain their behaviour. It may be that in the case of traders I could be successful in developing a programmed trader that is always successful. Even if not, I may develop an algorithm that turns a profit in some market for a useful period. Thus even if the assumptions of econophysics are wrong, they may be ‘pragmatic’. At the other extreme, policy is not something that I would like to see turned over to a robot, but even here some ‘what-if’ analysis – using the tools of econophysics as tools – is useful: just don’t forget its limitations.

    As Blue noted, statistical mechanics actually applies classical probability theory, which is a little odd as the models can generate behaviours, such as those in economics, that display a broader type of probability, such as Keynes’. But even if reformed, one would still need to treat the outputs of such models with caution.

    ReplyDelete
  12. Thanks for commenting on both posts, Dr. Marsay! Yes, I was faulting the Great Moderation Consensus.

    Your comments on econophysics remind me of the Lucas Critique. Was that what you had in mind?

    ReplyDelete
  13. @ Blue Aurora

    Are you forgetting that there are some nonlinear systems whose behavior can not be predicted despite all of the methods for solving for those variables? Plus, to make a satisfactory equation, don't you have to know all the relevant variables? I'd be willing to bet that human action is one of those situations where not all variables are known.

    ReplyDelete
  14. Jonathon Hunt: You're right - human action contains variables that are not known. But the econophysics project has designed statistical ensembles to find regularities out of the noise. What is your opinion of econophysics, anyway?

    ReplyDelete
    Replies
    1. I'm very, very interested in how econophysics develops, and the predictions it will makes. If it does turn out to falsify many of the tenants of economics, then everyone has no choice but to endorse that new method. If that day comes, then I will surely do it. I'm not an economics major to defend ideologies; I'm here simply for knowledge. Ignoring my rambling, at this point I'm very skeptical and would have similar views to the other Jonathan that posted here. The difference between particles and humans, which I believe Jonathan was referring to, is that there is no causality in praxeology. Particles react in the same way, every time, yet humans simply don't. Conveniently enough for me, there was a daily article over this same subject, where the author explains the point better than I.

      "Economic life, unlike natural life, is not subject to the law of causality. In natural life the same causes always produce the same effects. In economic life the same causes may produce different effects. Why? Because man's volition interferes between causes and effects. And man's volition is not ruled by causes (past), but by aims (future). Events in human life are not ruled by the law of causality, but by that of finality."

      This implies that the methods being touted by the econophysicists would not be appropriate. As I said above, if you tried to model human action mathematically, then it would surely fall into indeterminate nonlinear systems. This leads one to ask what is the point of these methods if it confirms nothing but what we already know? That is, we know human action is unpredictable, and mathematically confirming this does not offer any predictive or explanatory powers.

      Delete
  15. You make good points. Are you an Austrian? Just curious.

    However, I disagree that we would be confirming what we already know. Look at how committed neoclassical economists are to Subjective Expected Utility despite overwhelming evidence from psychologists and decision theorists over the last half-century. Sure, modelling human decision-making processes mathematically would make it one of those "indeterminate nonlinear systems", but it's better to have a model than not.

    ReplyDelete

All anonymous comments will be deleted. Consistent pseudonyms are fine.