Thursday, July 19, 2012

Microfoundations would be nice if we had them


Matt writes:
But while reduction is something scientists inquire into, it's never been the case that reduction is the sine qua non of useful scientific inquiry. To the best of my understanding, Newton's account of gravity didn't have any microfoundations at all. It was a mathematical formalization of observations about how the world functions... 
And crucially the key test a purported effort to provide microfoundations for gravity must pass is that it has to explain gravity well. The trend in economics since the Lucas Critique seems to be the reverse. If a theory lacks adequate microfoundations, it's rejected out of hand while you get a lot of wriggle room in terms of accounting for the data properly... 
People who seem very bothered by the fact that the observed reality of nominal wage stickiness is not well-microfounded don't appear to have the same difficulty relying on the observed reality of gravity. This is particularly strange since microeconomics itself is not particularly microfounded in psychology or neurology... 
The right answer is that productive inquiry can happen at many different levels of understanding and the appropriate test of a theory is whether it gives some kind of useful account of the thing it's supposed to explain.
I have a few responses to this, which don't fit into a coherent whole, so I'll just make a list:

1. Gravity is different from macroeconomics in several ways. One way is that we don't actually know if there is any phenomenon "underlying" gravity. Gravity might not be explainable in terms of any broader, more general phenomenon. But we know for a fact that macroeconomics is the result of a whole bunch of little economic decisions by individuals and companies. This, not the Lucas Critique, is the real reason we look for microfoundations for macroeconomics. Now, maybe some macroeconomic phenomena don't have microfoundations; maybe they are emergent. But we don't know that yet.

2. Another way in which gravity is different than macroeconomics is that you can test gravity in a lab. With macro, our empirical data come from economic time series, which are basically a very poor window into the phenomena we're trying to understand. Time-series econometrics, currently our only tool for matching macro theories to data, is pretty inadequate. Microeconomics, however, often can be tested in a lab, or with reasonably abundant "natural experiments" that don't exist in macro. So finding microfoundations for macroeconomics would allow us to be more scientific about the whole thing.

3. Matt says "the appropriate test of a theory is whether it gives some kind of useful account of the thing it's supposed to explain". This is a paraphrase of Milton Friedman's quote that "theory is to be judged by its predictive power for the class of phenomena which it is intended to 'explain.'" But "useful account" and "predictive power" are not precisely defined terms. For example, people once believed that psychological stress caused ulcers. This explanation had good predictive power - people who were under a lot of stress started having ulcer pain. And it was a useful account, since you could reduce pain by reducing stress. But eventually we found out that bacteria play a big role in ulcer formation, almost certainly bigger than the role played by stress. In this case, a decent non-microfounded theory was vastly improved by a successful search for microfoundations.

4. Having said all this, I agree with Matt that there is too little respect for data in macro today (probably because if there were more respect for data, macroeconomists would have to say "We don't really know whats going on" much of the time). And I agree that productive theoretical analysis can in principle be conducted at any level of complexity. If this idea is used to argue that we shouldn't require macro theories to be microfounded, then I agree. If this idea is used to argue that looking for microfoundations in macro is a waste of time and effort, then I disagree. Having microfoundations is better than not having them. 

5. Once again, let me reiterate that the big problem with "microfounded" macro, as I see it, is that the "microfoundations" are bad: not credible, and generally not consistent with anything microeconomists have actually found. Bad microfoundations are worse than none at all.


Update: Here's Peter Dorman with some specifics about bad microfoundations. Actually, having read a bazillion experimental papers, I am of the opinion that some neoclassical micro matches reality pretty well and some doesn't. See here for a partial survey.


Update 2: Here's Tyler Cowen with some thoughts on microfoundations and wage flexibility. He also makes the excellent point that "microfoundations" is not synonymous with a certain policy attitude; in other words, we shouldn't make the mistake of thinking "microfoundations = hard money" and then conclude "Let's ditch microfoundations"...

68 comments:

  1. Anonymous12:56 AM

    Incredibly bright people have written extensively on "micro foundations" in the various special science, eg David Hull, Harold Kincaid, Alex Rosenberg, etc.

    A good place to start if anyone wants to talk intelligently on the topic.

    ReplyDelete
  2. Anonymous12:58 AM

    I read Matt's piece as I was very tempted to write something that said, in a nutshell, that even as a non-fan of DSGE and other "micro-founded" models, I still think the Lucas Critique is valid and worthwhile. And, if data contradicts a model with microfoundations, that doesn't mean microfoundations as an idea is bad, it's just that those particular micro foundations aren't right. That doesn't mean you give up on the idea of microfoundations, it just means you have to fix the micro foundations.

    The flip-side of that is if you really need a macro prediction, it may very well be better to use a non-microfounded model right now, but that doesn't mean you should stop trying to come up with microfounded models altogether.

    ReplyDelete
  3. Anonymous1:05 AM

    Darwin out it plain -- his causal mechanism had tremendous explanatory power but predicted no particular event at any particular place or particular time, and no law following measurable, testable, relation between one variable and another.

    The picture of 'science' they teach to 2nd graders & spouted by economists is false.

    Read some Thomas Kuhn or the book _Chaos_ or some David Hull for hints as to why.

    ReplyDelete
    Replies
    1. [Darwin's] causal mechanism had tremendous explanatory power but predicted no particular event at any particular place or particular time, and no law following measurable, testable, relation between one variable and another.

      That seems false to me. For example, "over time, new species will emerge" seems like a testable, measurable prediction to me.

      Delete
    2. Anonymous2:09 AM

      sorry but the person with the 2nd grader view of science is mr anon above. Darwin's theory explicitly required that the then standard model of inheritence be wrong. We now know Darwin was right - genes are discrete factors - and are not simply an average of our parents.

      also nonlinear dynamics have very very explicit and precise numerical predictions about the behaviour of said systems

      Danny

      Delete
    3. More generally, Darwin's theory says some things are impossible. A new species of pigs with wings cannot emerge *unless* those wings develop from some now-existing skeletal structure (and at every stage of development, confer a reproductive advantage).

      Likewise, Newton's theory (a formalisation of Kepler's and Copernicus's) said certain things were not possible. The old theory of epicycles and deferents allowed any celestial body to orbit in any apparent way, with no possibilities ruled out.

      This point is perhaps a bit subtle, so I'll repeat: the most important function of scientific theories is to *exclude* proposed observations from the realm of possibility.

      This is my criticism of the "microfoundations programme". Having fit the model (DSGE, whatever) to approximate reality, science can start. The scientist's task is to explain why the model parameters can only have the values they do have, *and not any other.* Macroeconomists are too often content to leave off before starting to practise any science.

      Delete
    4. Anonymous1:16 AM

      Sorry, Danny. You are talking out of the back of your pants when it comes to Darwin and what his theory "requires".

      Delete
    5. Anonymous1:17 AM

      Greg is giving us a revised version of Popper.

      Popper, however, conceded that Darwinian biology and other essentially complex sciences refuted Popper's falsifiability criterion of science.

      Delete
    6. The point is, Darwin's ideas are perfectly testable.

      Delete
    7. Anonymous2:11 PM

      If you read the "origin of species" you will find that:

      1. Darwin was very, very good at combining the most minute micro trivia with the most macro geological and ecological processes. This capacity made him also mention that on several points science has not yet yielded the information needed to be sure about his theory - he knew the limits of his knowledge. Which was a good thing and which should be an example to economists.

      Also, he predicted, when confronted with a unusual orchid, that there also had to be an usual insect which pollinated it: http://en.wikipedia.org/wiki/Xanthopan_morgani

      And as far as I know, he also stated that the geological time scale (In those days geologists thought that the earth was about 300 millions of years old) was not long enough for his theories to work out. He did connect the dots.

      Delete
  4. Gravity was perhaps not the best example to use. Thermodynamics is an entire field based on macro behavior with little to be gained by reductionism to micro. Certainly many physicists beginning with Boltzmann tried to show that the basic thermodynamic laws were consistent with their understanding of the micro world.

    But no serious physicist was going to throw out the notion of entropy simply because Newton's laws (the micro) seem to have no preference for entropy going forward in time or backward in time.

    Thermodynamic properties like pressure and temperature even when attempting to reduce it to micro-foundations actually say very little about the specific properties of the individual actors (the molecules of the gas being measured). Some can be very energetic and others less so. The macro measurements are about averages over very large populations.

    I get the sense that those preoccupied with micro explanations for macro-economics are doing something very different than Boltzmann. I get a sense they are interested in defending a philosophical position related to efficient markets, fairness of open markets, and other value propositions. To do so, they tend to replace the randomness and capriciousness of the micro and want to say EACH actor is acting efficiently to maximize their utility.

    This may sound like a small nuance but to me it is a fundamental difference and could yield very different results to the resulting macro theory. Assuming each individual actor actually has the average value of the micro values (same wealth, same knowledge of markets) will trend towards having a very stable and linear macro theory.

    Allowing for each actor to vary over a wide distribution of the average allows for much more complex systems. System that may have multiple points of stability (liquidity trap and poverty trap), or levels of non-linear unpredictability (market crashes). Things that the micro people can't seem to get their head around.

    Perhaps this could also explain why in general those arguing for micro foundations are also less inclined to be concerned with things such as income distribution among the middle class and the wealthy. They just don't see the instability in having actors with widely varying properties.

    ReplyDelete
    Replies
    1. This is key. You get a Maxwell-Boltzmann velocity distribution whether you use neon or water vapour. Dramatically different micro-foundations give identical macro results in a statistical mechanical system. And it is obvious with any ones with eyes to see that economics is a statistical mechanical system.

      You can get all the main outputs of an economy if you model it properly and simply as a stat-mech system, see for example Ian Wright:

      http://arxiv.org/abs/cond-mat/0401053

      or google 'why money trickles up':

      http://www.econodynamics.org/sitebuildercontent/sitebuilderfiles/bullets.pdf

      By reducing multi-body systems to a single representative agent, DGSE and similar throw away the statistical mechanics so that they can get back to the field theory maths that is the only maths they are comfortable working with.

      Delete
    2. But no serious physicist was going to throw out the notion of entropy simply because Newton's laws (the micro) seem to have no preference for entropy going forward in time or backward in time.

      That's because thermodynamic entropy is emergent! With one particle, thermodynamic entropy is constant in time, because the multiplicity of each state is always one.

      If we could identify some emergent properties of macroeconomies, that would be awesome.

      As a side note, I personally wonder if the number of people in a macroeconomy is just too small to have thermodynamic-style phase changes. The number of people in a macroeconomy is not more than 1e10, while the number of particles needed for things like water freezing into ice is orders of magnitude bigger than that.

      Delete
    3. "while the number of particles needed for things like water freezing into ice is orders of magnitude bigger than that."

      How small is the smallest drop of water that can freeze? I suspect that under the right circumstances you could "freeze" as few as a 1e3 water molecules.

      Delete
    4. My first thought is to say a property is emergent is simply saying that our micro models do not adequately explain the phenomenon. But that is exactly what the micro-heads (and I mean that in the nicest of ways) refuse to accept.

      But I see your point, Noah. Macro does not have similar well defined emergent measurable properties such as pressure or concepts such as entropy. Macro typically is dealing with summations: M2, GDP, unemployment.

      Noah's example of phase changes may also be not the best example, but his point that economies have nothing near a mole of actors is important and that perhaps this in itself makes any emergent properties hard to identity as they may appear to fluctuate randomly.

      Delete
    5. Mark

      You do not need very many actors to get complex behavior out of simple rules. The three body problem with point masses and without relativity cannot be solved in closed form.

      Delete
    6. Emergent != complex. Emergence is when you get simple properties out of a complex system. A three-body system doesn't have any emergent properties that we know of.

      Delete
    7. And there may be emergent macro phenomena. Bubbles and crashes in financial markets could be emergent, for example.

      Delete
    8. Anonymous11:20 AM

      @Noah - I'm confused by your terminology.

      Emergence isn't usually defined as "where you get simple properties out of a complex system." It's where you get properties at a macro scale that are not present at the micro scale, and critically, can't be predicted from the properties of the micro-scale components.

      Delete
    9. Emergence isn't usually defined as "where you get simple properties out of a complex system." It's where you get properties at a macro scale that are not present at the micro scale, and critically, can't be predicted from the properties of the micro-scale components.

      Yes, that's right.

      Delete
    10. Not sure how meaningful this is, but the thermo analogy brings something else to my mind: ideal gas law and other gas dynamics rely not just on large numbers but on very simple interactions between molecules. They're assumed to just elastically bounce off each other, and happily this is pretty close to true no matter the specific gas molecule. Thus it works.

      If you have stickier interactions, then you don't have a gas, you have a liquid, with lots more complexity: viscosity, Newtonian/non-Newtonian fluids, mixing (or not). Obviously liquids have their own simple properties -- flowing with gravity, and you can boil them by adding heat, and both liquids and gases have other complexities like turbulence, but still.

      Another angle: a common physics approach is assume something simple and ideal, then add perturbations to it. Sometimes that works well. But sometimes the less ideal detail means something like having a liquid instead of a gas.

      Obviously I'm seeing these as raising whether macro-econ is more like simple gas interactions, maybe with perturbations, vs. having stickiness and entirely different behavior.

      Delete
    11. "If we could identify some emergent properties of macroeconomies, that would be awesome."

      We can. Restricting the supply of money to the poor / middle classes causes less productive economic activity and fewer bubbles. Increasing the supply of money to the poor / middle classes causes more productive economic activity and more bubbles.

      These are emergent properties *discovered by Keynes*. They don't hold at the micro level; in a group of only a few people, printing more money doesn't change the amount of productive activity.

      This is perhaps because at a small scale people can see through the "nominal value illusion" but when it happens at a large scale they *can't*? Or perhaps that's not what it's from. In our current understanding, it is most certainly an emergent property.

      Delete
    12. The number of people in a macroeconomy is not more than 1e10,
      Hari Seldon would agree, I think; Asimov too.

      Bubbles and bubble pops ARE the emergent properties -- extremely poorly predicted by "macro" not-quite science, with respect to timing. But Greenspan end of 1996 "predicted" the 2000 dot.com bubble pop in his irrational exuberance. The Economist, among others, was predicting the house price bubble pop from around 2003.

      Being "right" too soon is better than being late, maybe, but is not so useful. Perhaps too soon is worse. After the predicted "drop" fails to happen, it might be that more think there is no drop possible. Roubini, like Ron Paul, has been wrong far more than correct.


      Remember that physical science is about non-moral conscious agents, so once understood the rock or motor or gas remains understood. Free will humans are able to, and do, respond differently to the "same" stimulus. If, for most of the last 40 years, middle income Americans spent X% of any gov't benefit, that is a good reason to believe a new gov't benefit of B would result in B*X amount of spending.
      But this time, contrary to this prior "relation" (as usually modeled), a lot less spending might happen (because of Net Worth decrease, for instance).

      Further, insofar as most Macro is oriented at helping gov't technocrats make policies that might be best for the gov't, but might well not be best for the given individuals, it is no surprise that individuals attempt to maximize their own benefit, rather than be controlled by the gov't bureaucrats.

      Delete
  5. I agree with Mark Friesen that perhaps gravity wasn't the best comparison. Statistical mechanics/statistical physics a better comparison.

    But speaking of statistical mechanics, Noah Smith, have you actually blogged on econophysics? I guest blogged on econophysics before over at Daniel Kuehn's Facts and Other Stubborn Things. While they were edited by Daniel Kuehn, I think you might want to read them. It's been a few months since they were posted, but feel free to make a comment on them via your website if you wish.

    http://factsandotherstubbornthings.blogspot.com/2012/03/guest-post-blue-aurora-on-econophysics.html

    http://factsandotherstubbornthings.blogspot.com/2012/03/guest-post-2-blue-aurora-on.html

    Out of curiosity Noah, do you think macroeconomics could benefit further from research in decision theory/decision science? Perhaps better microfoundations can better developed if Itzhak Gilboa's Choquet Expected Utility models become mainstream. However, we'd need to test if Choquet integrals are the more general case.

    The non-linearity and non-additivity of decision-making were long proven by George Boole in An Investigation of the Laws of Thought and J.M. Keynes in A Treatise on Probability, according to the following paper by Dr. Michael Emmett Brady.

    http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1546726

    P.S. Noah Smith - can you please respond to my e-mail?

    ReplyDelete
    Replies
    1. ivansml7:23 AM

      Not Noah, but I'll try to reply:

      What do you mean by "becoming mainstream"? There are many mainstream papers that apply non-expected utility in finance ([1], [2]) or in macroeconomics ([3], [4]). But, as is usual in economics, the goal is not to find some grand unified theory of preferences. Deviations from expected utility may be important in some settings, and unimportant in others, so correct modelling of preferences will always depend on individual problem under investigation.

      Regarding econophysics, I would be interested about its contributions outside of finance. Because frankly, studying patterns in high-frequency financial data is not something most economists really care about, yet that's what seems to occupy most of econophysicist's time.

      Delete
    2. Ivansml: By that, I meant that Choquet utility replaces Subjective Expected Utility. S.E.U. is really a special case that applies in certain situations. Thanks for supplying the links, but my knowledge of decision theory comes from Dr. Michael Emmett Brady. I need to do a deeper reading of the academic literature myself, but it seems to me that Dr. Brady has a good grasp of the academic literature and seems to know what he is talking about. If you would like to correspond with Dr. Brady, you can get his e-mail address on his SSRN account.

      http://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=1033456

      I do agree with you that the econophysicists need to expand beyond the area of finance. They are working on that, however. There are papers by econophysicists about income and wealth distributions, for example. There are also a few on unemployment. Check the database for Physica A: Statistical Mechanics and Its Applications for reference.

      Delete
    3. ivansml9:24 AM

      Looking at titles of Brady's papers, they all seem to be concerned with history-of-thought topics, like what did Keynes really mean by his theory of probability. There is of course nothing wrong with writing papers on HET, I just don't see how they are relevant to applications of non-expected utility today.

      Regarding econophysics of wealth distribution, you mean those models where agents are modeled as gas molecules, randomly bumping into each other and exchanging money? I read some of those, and I wasn't impressed. We actually have models in macro with agent heterogeneity and endogenous wealth distribution (Bewley / Aiyagari / Krusell & Smith -like models). They are not perfect and you could certainly argue about their assumptions. But how is replacing optimizing agents with mechanistic particles any better? You cannot use such models to evalaute policy counterfactuals (as agents do not react to incentives), you cannot say anything about welfare (as there are no preferences). And generally, what insights do you get when you simply take model from physics, relabel variables and show that it roughly matches some economic data?

      Delete
    4. Ivansml: I can supply you links to Dr. Michael Emmett Brady's better papers. His best papers, in my opinion, were published in The British Journal for the Philosophy of Science and International Studies in the Philosophy of Science.

      http://bjps.oxfordjournals.org/content/44/2/357.full.pdf+html

      http://www.tandfonline.com/doi/abs/10.1080/02698599408573487

      Also, if you can access the database for the journal Psychological Reports, do read this article by Dr. Michael Emmett Brady and his colleague, Howard B. Lee.

      http://www.amsciepub.com/doi/abs/10.2466/pr0.1989.64.1.91

      Finally, I recommend reading this paper by Dr. Michael Emmett Brady, as in my opinion, it is the best out of his most recent scholarship.

      http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1546726

      The econophysics of wealth distribution reveals power laws at certain parts of a distribution, if I'm not wrong. Granted, they aren't perfect either, but you could make the argument that you can at least "tag" the individual unlike the situation in statistical mechanics. The econophysicists do use agent-based models, if I'm not wrong, so they too recognise agent heterogeneity.

      Nevertheless, while I do find econophysics interesting, I do have criticisms of it. While some of them are willing to work with economists themselves, the fact that some of them want to burn all of economics isn't necessarily a good thing in my view.

      Delete
    5. Sorry, I am not interested in reading any of the works of Dr. Michael Emmett Brady at this time.

      Delete
    6. Drat. Why would that be the case? Just curious. What would get you to be interested in the future?

      Delete
    7. Anonymous12:10 PM

      I suspect Blue Aurora is Dr. Brady.

      Delete
    8. Actually, Anonymous, I'm not Dr. Michael Emmett Brady, but I have corresponded with him frequently via his e-mail address, which you can find on his Social Science Research Network page. Dr. Michael Emmett Brady is quite receptive to correspondence.

      Delete
  6. From Matt's original post:

    This is particularly strange since microeconomics itself is not particularly microfounded in psychology or neurology...

    On that note, it is often assumed that advances in neuro- and experimental economics have only served to undermine the standard micro concepts taught in undergrad classes. (i.e. People aren't hyper-rational, homo economicus types, etc etc...)

    However, many of the canonical micro theories have actually been corroborated by experimental economics. For instance, see this paper by Choi et al. (2011) on individual rationality, which confirms the basic ideas laid out by Paul Samuelson and his notion of "revealed preferences". The take home message is that the experimental results are very much in agreement with the standard axioms of RP theory... People are economically rational and consistent in their choices when maximising utility. (Although, the level of this consistency does vary with factors like wealth.)

    ReplyDelete
    Replies
    1. Yes, I think this is an important point. Experiments have been very supportive of some classical micro theories, and not so much of others.

      Essentially, what I'm finding from reading the literature is that people are very good at maximizing utility, pretty good at predicting future time series from past data, very good at seeing small opportunities for profit, pretty bad at processing large amounts of information quickly, pretty bad at dealing with probabilities, and very bad at dealing with uncertainty. They seem to care a lot about norms and values and to not pay much attention to what other people are thinking unless someone points them in that direction.

      Delete
    2. Commercial: Paul Glimcher's Foundations of Neuroeconomic Analysis contains an account of what is currently known about the neurological foundations of economic decision making. A good deal is in fact known, much of it recently found. I've just finished reading it, and thought it wonderful.

      It's what you read if you really want to know the current status of real, as opposed to theoretical, microfoundations.

      Delete
    3. You'll also find that the literature shows that people's utility functions are quite a bit weirder than "homo economicus"'s utility function.

      Delete
  7. Noah, I really appreciate your candor when you write, "if there were more respect for data, macroeconomists would have to say "We don't really know whats going on" much of the time."

    I am a 1978 era student of macro (and urban and location economics).

    Present events have caused me to make a detailed reading of the literature to find out why the Lesser Depression was allowed to happen. What I have found is beyond appalling.

    You mention gravity. The economic factors in urban and location economics have properties like gravity in that they either attract or repeal economic activity. We know how they work, but we have no models for measuring their power over time, albeit we can all observe the results when we drive through any major urban or semi-urban area.

    One with knowledge of location economics would know, in their gut, that the Euro could never be made to work, long term, for Greece. Greece could be heaven on Earth, only populated by Saints and governed by God, but as a part of the current Euro, over time economic activity would both never come and what is there will leave, regardless of how often John Cochrane mimics Glen Beck.

    One with knowledge of location economics also knows that free trade and globalization likely don't workout well, over the long run. The "gravity" of having the World's largest economic creates location forces in which the end result may be that the largest economy works like a black hole, sucking in economic activity from everywhere else. Economists should be honest, telling people, we don't know whether free trade and globalization are good, over the long run. No one has models measuring location forces over time.

    I don't understand why you are being asked to write a dissertation. Your posts, here, show you more than amply qualified.

    ReplyDelete
    Replies
    1. I don't understand why you are being asked to write a dissertation. Your posts, here, show you more than amply qualified.

      Flattery will get you nowhere! ;-)

      Delete
    2. Miles says you passed. Wonderful news, Doc Smith

      And, your dissertation covered interesting points. You have been reading your Soros and Munger.

      Delete
  8. Major_Freedom7:26 AM

    The core problem with both macroeconomics and microeconomics in "mainstream" economics today is that the profession has been almost completely infiltrated by anti-rationalists. On the one hand we have hermeneutics, rhetoric, and epistemological anarchism, on the other we have empiricism, positivism, and epistemological skepticism.

    I don't know how to make this sound non-cliché, but since at least the 1950s, a war has been waged against economic science. Most philosophers and economists are (mistakenly, IMO) approaching the social sciences the way a physicist approaches matter.

    ReplyDelete
    Replies
    1. The way a *bad* physicist approaches matter.

      Not the way a materials scientist approaches matter!

      Delete
  9. Orange148:15 AM

    Very intriguing post and comments! As a non-economist (and I don't know whether this is good or bad; chemist by training) the chief difficulties as I see it is that the field is very good at developing lots of models that explain past behavior but precious few that can explain future behavior. Even then when they have tried to extrapolate them to the field of finance there have been major disasters along the way. Part of the difficulty is the zero sum game phenomenon (Long Term Capital Management and abuse of CDOs and CDSs more recently).

    When one looks at physico-chemical theories/laws one sees a totally different picture. Models/equations are an excellent predictor of future results (I know if I drive my Honda at 60 mph into a brick retaining wall that the outcome will be less than optimal and of course there are lots of other examples as well). Even in evolutionary theory we can predict certain things these days (development of antibiotic resistance in bacteria).

    The key problem is that economics is centered on human activity which in the overall results in a great deal of uncertainty. We can use Occam's razor to simplify things considerably (current economic situation is a result of a lack of demand) and use tools to address that base assumption. The problem is that this is too simple and doesn't rely on complex math to solve the problem! :-)

    ReplyDelete
  10. I like you conclusion "bad microfoundations are worse than none at all".
    I fear however that even "good" microfoundations may have a perverse effect: they require too much time for too little result. They might be relevant, but not efficient (very ironical for economics).
    What I mean here is that economists who engage in micro-founded models need to heavily specialize in their field of research. It may even induce a selection of economists who are only comfortable with models, but pay little attention to what's going on in the real world.

    There is a post by Chris Dillow (link below) about historical unemployment in the UK. I believe his graph invalidates a lot of fancy theories. Just like the study of financial crises in the US and the UK in the 19th century may be the best guide for central banking today, the study of historical facts and data may be the most efficient way of making advances in economic thinking.

    http://stumblingandmumbling.typepad.com/stumbling_and_mumbling/2012/07/unemployment-a-brief-history.html

    ReplyDelete
    Replies
    1. Good point. Good microfoundations are nice, but do we have the luxury of a 50-year research program before making further macroeconomic predictions? I don't think so.

      Delete
  11. Orange1412:23 PM

    Congratulations appear to be in order for you. I see on Miles Kimball's site that he is (and I hope not against his better judgement) drawing up the papers with the thesis committee to grant your degree!!! Welcome to the Piled Higher and Deeper crowd.

    ReplyDelete
  12. "the big problem with "microfounded" macro, as I see it, is that the "microfoundations" are bad: not credible, and generally not consistent with anything microeconomists have actually found. Bad microfoundations are worse than none at all."

    Of course, I could play devil's advocate and say is that if all the microfoundations are bad, how can we be so sure there are any god ones.

    I'm not saying that,just I-or any skeptic. Don't get me wrong, I'm not saying they aren't important, you seem convincing and you know better about some of this than I would.

    However, isn't a big part of the desire for microfoundations by people like Lucas and Prescott, just that they largely don't like macroeconomics very much.

    Even Sumner has suggested that if only we could do his thing with NGDP Macro would once and for all be solved and we'd have nothing left but Micro?

    When I hear about microfoundations I wonder if what this really means-out of some people's mouth, I don't think yours-that what we're really talking about is a return to the pre-Keynesian world? Macro of course being what Keynes founded.

    ReplyDelete
    Replies
    1. Of course, I could play devil's advocate and say is that if all the microfoundations are bad, how can we be so sure there are any god ones.

      Experiments and natural experiments.

      Delete
    2. The fact is that we all experience the micro-foundations whenever we make an economic decision, and we know that our combined micro-economic decisions are what drive macro-economics. We are the micro, and we comprise the macro. We may lack the necessary insight, but we understand that we make economic decisions for reasons, and that other people may make inscrutable decisions, but that they have their own reasons as we do.

      Imagine we all had experience living in a quantum mechanical world of atoms, photons and electrons and wanted to understand how efficient a steam engine we could build. The gap is statistical and mathematical. To understand steam engines, we'd have to start in the relatively unpredictable quantum world and apply our understanding of statistics and probability to get to classical physics and then to statistical mechanics and thermodynamics.

      Can the gap be bridged? The fact that we can measure macro-economics and the efficiency of steam engines suggests that it is not in either domain.

      Delete
    3. If you were to really base macroeconomics on "microfoundations", you'd spend a great deal of time studying "tastes" and "preferences" of the general population, given that those are key to determining practically everything. Then you'd spend a lot of time studying how advertising affected those. Only then would you start talking about money...

      This isn't done. In transportation economics, consider how hard it has been to get the "professionals" to admit that there is a generic preference towards rail service and against rubber-tires-on-asphalt bus service among the vast majority of the population, and that this can't be changed even with advertising. We don't exactly know why (though there are some theories), but it's an observed fact. Yet, *massive resistance* to incorporating this in models until very recently.

      Delete
  13. Anonymous5:32 PM

    A very nice and thought-provoking post. Thanks.

    ReplyDelete
  14. 2slugbaits7:05 PM

    Noah,

    Isn't the idea of microfoundations in physics subject to the critique that there are microfoundations, micro-microfoundations, and micro-micro-microfoundations? For most practical problems Newtonian physics is just fine...good enough to get you to the moon and back. But at some level Newtonian physics fails at the extreme micro level (quantum physics) and at the extreme macro level (gravity). In practice don't both economists and physicists arbitrarily (but not unreasonably) decide how micro a microfoundation needs to be? And isn't that decision based on the nature of the problem?

    ReplyDelete
    Replies
    1. In practice don't both economists and physicists arbitrarily (but not unreasonably) decide how micro a microfoundation needs to be? And isn't that decision based on the nature of the problem?

      Yep, exactly right!

      How is that a "critique"? It means that you use the foundations that work, however micro they might be...but they have to work...

      Delete
  15. Anonymous10:48 PM

    What if a economic phenomenon is observable and predictable, but doesn't have micro foundations?

    Then a requirement for microfoundations in order to be academically acceptable would be perverse.
    The search for the holy grail of microfoundations would be futile. Where would that leave modern macro?

    jonny bakho

    ReplyDelete
  16. "Now, maybe some macroeconomic phenomena don't have microfoundations"

    It's not that there aren't microfoundations for every macro phenomena -- macro phenomena are the sum of micro unit (i.e. people) behavior.

    It's that the true microfoundations are, to an extent, ones that the field is very resistant to accept, partly because they make modeling less mathematically pretty and tractable, partly because they sound unsavory, partly because they make the dominant paradigm, that lots of economists spent years learning and make a rich living with, less marketable and prestigious, and partly because they make the libertarian ideology many economists live by a lot less attractive sounding to the vast majority.

    For example, one might say that there are no microfoundations for extraordinary nominal wage cut resistance, even though it’s super empirically observed. But there are microfoundations for this, just not ones the field likes, microfoundations like a psychological reaction to nominal losses, over and above the real losses; microfoundations like widespread ignorance of many things in a supercomplicated world, and perhaps microfoundations that don’t necessarily fit into a nice tractable formula for aggregating up – the Haugen critique.

    ReplyDelete
    Replies
    1. It's not that there aren't microfoundations for every macro phenomena -- macro phenomena are the sum of micro unit (i.e. people) behavior.

      Ahh, but if there is classical chaos (can't predict macro outcome from micro data) plus emergence (can predict a macro outcome from macro data), then a phenomenon effectively doesn't have microfoundations!

      Delete
  17. It is a historical fact that the lack of micro-foundations for gravity was a matter of serious concern for decades at least (and I think centuries). The preceding micro founded theory which survived (barely) from Democritus through DesCartes was that the first law of physics is that there can't be two objects in the same place at the same time (note Newton accepted this but modern physicists don't). From this they argue that particles must rebound when they collide (this doesn't follow usefully as they can and do stick together some times). Then forces were assumed to be mediated by particles (an odd and ironic anticipation of Feynman by a couple of millenia). The action at a distance in Newton's theory was a matter of concern, basically forever.

    But people advising on say how to aim a cannon didn't ignore the'tequations because they weren't micro founded.

    I see a gap between Yglesias (with whom I agree) and Friedman. Friedman's key phrase is "for the class of phenomena which it is intended to 'explain.'" This means that theories which are inconsistent with available data are to be judged favorably, because Friedman says the data are not in "the class of phenomena which it is intended to 'explain.'" It means that a theory which fits some data but not other data is OK.

    This is not how Physics works. The whole of theoretical physics was turned upside down, because of obscure phenomena observed in Vienna mostly. There was no reason to be confident that the new physics would be useful in explaining things we care about. The atomic bomb came as a surprise.

    Friedman is in the tradition of those who denounced Gilbert's study of magnetism because who cares about lode stones (they include Francis Bacon on an off day) and those who ignored Mendel on the grounds of who cares about peas (they include Mendel on an off decade after he stopped studying peas). Friedman's proposal that we ignore inconvenient data which don't fit our model but which we don't care about for their own sakes (for him everything but inflation and occasionally real GDP) is an attack on almost all useful scientific research in human history.

    I mean who cares where the planets are ? We can navigate by the stars. Friedman's proposal that we ignore inconvenient data made him an enemy of sciene. Yglesias makes no such proposal.

    ReplyDelete
    Replies
    1. Friedman's key phrase is "for the class of phenomena which it is intended to 'explain.'" This means that theories which are inconsistent with available data are to be judged favorably, because Friedman says the data are not in "the class of phenomena which it is intended to 'explain.'" It means that a theory which fits some data but not other data is OK.

      Really?!?!?!?!?!?!

      That sounds nuts. I am very skeptical that this was what Friedman was actually saying...

      Delete
  18. Oh and about ulcers, there are two common kinds of ulcers gastric (stomach) ulcers and duoidenal (sp??? means first loop of the small intestines) ulcers. The pattern in the data was that stress is related to duidenal (sp ????) ulcers. This is the current theory.

    Helicobacteria Pilori is associated with Gastric ulcers. Notably, gastric ulcers are rare in countries with good sanitation. Also gastric ulcers tend to become stomach cancers. Notably the raw data shows a dramatic decline in stomach cancers in developed countries (as dramatic as the increase in lung cancer following the invention of the cigarette rolling machine).

    ReplyDelete
  19. My point (if any) is that there was a huge anomaly in the data suggesting an important not at all understood change over time. It isn't as if the theory fit the available data then there was a huge advance. It is more that, since stomach cancer was already rare in developed countries, few people cared. Well before the helicobacter research I asked my mom (who has an MD) what the hell was going on with stomach cancer.

    ReplyDelete
    Replies
    1. Orange148:32 AM

      Another set of good examples comes from baseball. The amount of statistical data from the start of the two major leagues in the early 1900s to the present day is immense and one can use a variety of statistical tools to analyze past performance and predict future performance. Still there are a couple of microfoundations that can be readily identified (changing the baseball around 1920 to make it more lively which increased home runs and batting averages; decreasing the height of the pitcher's mound in 1968; and the use of performance enhancing drugs more recently). One needs to be quite cautious when looking at big data sets so as not to ignore some of the underlying foundation.

      In addition, it points out that some things are just hard to model (improved nutrition and training, leading to bigger and faster athletes; changes in the size of gloves that enhanced fielding, etc.).

      Delete
  20. I wrote about the need for DSGE adherents to provide the "Microfoundations of the crisis" a few months back:

    http://www.disequilibria.com/blog/?p=213

    ReplyDelete
  21. Anonymous10:18 AM

    The concept of emergence is a bit slippery, as it seems to imply an autonomous identity never the less tethered to some complex causal nexus. An emergence is an incident that is neither incidental nor independent of foundational phenomena, yet radically other than the sum of its parts and thereby irreducible as such. Transcendent perhaps? Seems a bit too metaphysical. In any case, even a strong emergent macroeconomic conjuncture would not be devoid of microfoundations. It seems to me that it would only require that the micro not strictly supervene the macro to make the case for emergence.

    ReplyDelete
  22. Seems to me the difference is stability. The rules of gravity are roughly speaking always the same, so if you can successfully capture the effect with a mathematical model, however non-microfounded, you're done, you can use it whenever you want. The rules of macro are really time-variant, and anyone who designed a successful non-microfounded model in say 1960 would have difficulty explaining anything that's happened in the last five years (I work for an organization that uses purely macro models of that vintage). The reason for microfoundations isn't purely "reductionism FTW," it's that hopefully the microfoundations are time-stable. But even if we ignore the ridiculous assertion that "RRA" or "IES" (whatever they really represent) are time-stable, I have my doubts that anything substantive regarding consumer behavior is truly invariant to culture and time.

    ReplyDelete
    Replies
    1. There is no reason to believe that microfoundations are time-stable.
      In fact, we can be absolutely sure that they are not!

      One of the earliest and most important discoveries in experimental economics is that the concept of "fairness" is crucial to most cultures, but that the value of "fairness" is culturally dependent. Tested in the ultimatum game and a few others, done cross-culturally. The typically accepted, and therefore "fair" according to social standards, offers range from 20-80 to 80-20.

      This means that the microfoundations, the basic behavior which underlies the economy, are massively different from culture to culture. And obviously cultures can change over time.

      So there will never be anything like "universal microfoundations".

      Delete
  23. JohnR2:38 PM

    Welcome back, Noah! Congratulations on the next step towards Becoming Serious. I'd love to explore your ideas a bit from a non-economics/poor-science PoV, but it's Friday and my brain is full so I'm going to ask to be excused. Anyway, one of the things you wrote did trigger a response:
    "..macroeconomists would have to say "We don't really know whats going on" much of the time.."
    My immediate reaction, apropos of nothing in particular, was to wonder how many macroeconomists are women?

    ReplyDelete
  24. Let's take a look from another angle: fluid mechanics. By the late 19th century the Navier-Stokes Equation was known to describe fluid motion perfectly but it could be solved for only a few simple cases. The actual field of hydraulics was entirely empirically based, often using non-physical formulations that were based on what worked in practice. Then Prandtl discovered the boundary layer and the door was opened for actually applying mathematics to solve real fluid problems. Most of the Twentieth Century can be seen as reconciling the practice of fluid mechanics with the mathematics of the Navier-Stokes equation. We still can't solve the NS equation directly for most practical problems - can't determine turbulent flow in a pipe from first principals even - but we do know how to model (approximate) enough small scale motion that we can handle turbulence problems quite accurately. Economics needs its "boundary layer" concept to reconcile the micro with the macro.

    ReplyDelete
  25. Anonymous11:17 AM

    Tyler Cowen makes one-half of an excellent point. This should not be overlooked. If it is wrong to dislike the microfoundation project becasue one dislikes hard money, then it is just as wrong to favor microfoundations because one likes hard money. Go look at who flogs microfoundations the hardest, and enquire as to their views on currency.

    At other times on other topics, Tyler is eager to talk about the perils of "mood affiliation bias," picking your arguments after you've reached your conclusions. That should be sauce for the gander, too.

    ReplyDelete
  26. Probably the key science that should be related to economics is biology.
    Because of the S curve.

    I'm pretty sure there a huge numbers of S curve type functions in (a future, improved) macro. Most of the time, below and above the big rapid (first mistyped raped) change, there is a time before the inflection point where a line relationship is not too far off the curve.

    But the failure of macro, whether micro founded or not, is the failure to know/ predict where the inflection points are. When does the bubble start? When does it pop?


    I support Sumner's call for NGDP targeting, and believe it would have, and will, reduce the pain of massive Net Worth loss from the Great Recession. But I don't believe it would have stopped the debt-financed house price speculation/ investment bubble from starting and growing.

    I mean, most macro models don't consider homeowner equity as part of "money", but almost all $100k/yr income workers with $300k in equity feel and spend like they have more money than those with the same income but only $50k in equity.

    ReplyDelete