Tuesday, April 17, 2012

DSGE vs. Weather Forecasting


You often hear defenders of the DSGE modeling framework say something like this: "You can't attack DSGE. Its just a framework. You can use it to model anything." But this isn't true. If it were true, we wouldn't have the label "DSGE", we'd just call it "modeling".

The "E" in DSGE stands for "Equilibrium." DSGE models define an equilibrium in which markets clear. The economy is then assumed to move toward that equilibrium; the motion is called "transition dynamics", and is usually left unspecified. 

Some economists have argued that transition dynamics are not that important. For example, in "Econometric Policy Evaluation: A Critique" (1976), Robert Lucas writes:
[O]ne hears talk of a "disequilibrium dynamics" which will somehow...[go] beyond the sterility of dp/dt = k(p - p_e)...[but this] will fail...
In other words, prices will move smoothly and steadily toward the equilibrium, so we should just focus on finding the equilibrium.

But this is not necessarily true. Why would "disequilibrium dynamics" be important? I can think of several reasons.

Reason 1: The equilibrium may shift at about the same speed that convergence happens. If the economy is trying to hit a moving target, chaos will result. See this paper by Andrew Lo for a semi-technical explanation of why this true. (In math-speak, this happens when the rate of convergence, k, may be of the same order as the parameters governing the time-scale of the shock process.)

Reason 2: The equilibrium may not be stable. See this blog post by Jonathan Schlefer (whose book I just ordered off of Amazon):
In 1960 Herbert Scarf of Yale showed that [even the most ideal kind of] economy can cycle unstably. The picture steadily darkened. Seminal papers in the 1970s, one authored by [General Equilibrium inventor] Debreu, eliminated "any last forlorn hope," as the MIT theorist Franklin Fisher says, of proving that markets would move an economy toward equilibrium. Frank Hahn, a prominent Cambridge University theorist, sums up the matter: "We have no good reason to suppose that there are forces which lead the economy to equilibrium."
In other words, the smooth convergence equation that Lucas wrote down may simply not be true.

Reason 3 (the biggie): There may be multiple equilibria. You rarely see famous and influential DSGE papers with multiple equilibria, and when you do see them, there are usually only two equilibria. But I know of absolutely no reason why the real economy should have a unique equilibrium. And I know of absolutely no reason why the number of equilibrium in the economy should be small! But there seems to be a huge publication bias in favor of smaller number of equilibria (Roger Farmer's efforts notwithstanding). This annoys me. 

So in a Robert Lucas DSGE world, we have one slow-changing equilibrium toward which the economy smoothly and rapidly converges. But the real world may be one of multiple, rapidly-shiftng equilibria toward which the economy does not move smoothly.

Now, you could still formally model that sort of world with a DSGE model, but your results would be useless gibberish. You'd have classical chaos.

What do you do when you have classical chaos? One thing you can do is to use a weather-forecasting approach. Basically, you identify the economy's microfoundations, and then you use big powerful computers to make very short-term predictions using those microfoundations. Here's a Bloomberg article in which theoretical physicist Mark Buchanan suggests exactly this kind of approach:
No mathematician can “solve” the complex equations for air in the atmosphere...It’s natural to wonder if a similar mechanism might be driving the financial crises and business cycles that typify the economic “weather” we’ve experienced over the centuries. Unfortunately, today’s equilibrium theories refuse to entertain the possibility... 
While an appreciation of instability propelled atmospheric science forward, economics was binding itself into a rigid framework...when studies in the 1970s found that [economic] equilibrium is generally unstable - and so should tend to fall apart...[economic] theorists for the most part simply ignored this inconvenient fact and went on as before. Most still do. 
American economist Milton Friedman set the tone. “The study of the stability of general equilibrium is unimportant” because “it is obvious that the economy is stable,” he was quoted as saying. Friedman was notoriously mischievous and slippery in his argument, so I’m not sure he really believed what he said. But most economists today act as if they do. 
This is too bad, because a focus on the origins of instability just might help financial economics achieve a conceptual liberation akin to that which atmospheric scientists achieved in the 1950s. Economists might come to accept that equilibrium doesn’t describe everything, or even very much, and that natural elements of instability and turbulence drive the outcomes that matter most.
This is an alternative to DSGE. Go directly from microfoundations to some variety of agent-based modeling. Ignore the equilibrium and focus entirely on the micro-level "transition" dynamics. Incidentally, I think this kind of approach could have other advantages - for example, because you wouldn't have to solve for a simple equilibrium, it might be easier to add large numbers of frictions to the model. 

But the main reason to contemplate using a weather-forecasting approach is that the real economy may just be too chaotic for DSGE to be very useful in modeling it.

55 comments:

  1. Anonymous3:51 AM

    perfect amount of wonkishness

    ReplyDelete
  2. Hey, I could almost have written this - but not as well. Your stealing my idea!

    But there is another problem with equilibrium. While single markets may move towards a local equilibrium, there is no reason to think that markets in general at any particular time will be all moving towards the global equilibrium (assuming it is knowable and stable) - in fact we know from experience that they often don't.

    ReplyDelete
  3. If you want an example of the above to assist in thinking about it - think about the term "flight to quality" and what it implies. (I could talk about topology and building lakes that drain eventually via waterfalls - but it just confuses).

    ReplyDelete
    Replies
    1. Or even more obviously the housing bubble.

      Delete
  4. Noah, agree with much of what you're saying, but there are some of your conclusions that I'm not sure about.

    First of all I agree with you that by and large, DSGE does not necessarily lend itself to "disequilibrium" theory since it assumes the economy is in equilibrium and agents are always behaving optimally given information, other agents actions etc. But there's nothing in DSGE that says this equilibrium is a social optimal. There are models (I don't have the references to hand so sorry) where responses to shocks may lead to optimal private outcomes but sub optimal social outcomes, which a policy maker can effect. Is this a form of stabilisation? Kind of I guess.

    I do completely agree with you on the transition dynamics though. Are there no DSGE papers that discriminate between convergence speeds over discrete time intervals?

    ReplyDelete
  5. I presume this is analogous to the evolutionary fitness landscape idea, where there could be multiple local optima. In evolution, one of the main purposes of the landscape is to explain why so many things do not occur in evolution. Evolution is a blind process with no foresight: that may well be true of macroeconomic optimization.

    ReplyDelete
  6. I think the known possibility of chaotic dynamics undermines the rationale for assuming rational expectations. See Agliari et al. (2006, JEBO, V. 60: pp. 526-552.)

    I also don't see why macroeconomists must formally model microfoundations.

    ReplyDelete
  7. Good post.

    Also, that Friedman quote is utterly ridiculous.

    ReplyDelete
  8. Larry Headlund9:16 AM

    How micro based is weather forecasting?

    Weather forecasters don't start with models of molecules interacting and rigorously work their way up. They don't even work from pure fluid dynamics on up. Their modeling starts on a grosser level. It is true that there models don't contradict finer grained analysis but that is because that finer grained analysis is ambiguous.

    It is appealing and would be wonderful if one could build up solidly from first principals but in economics it may be as sterile an approach as the General Semantics idea of constructing psychology up for colloidal chemistry,

    GGSE makes the opposite mistake of assuming away every thing really interesting but because it is cold in the Arctic doesn't imply it is comfortable in the Antarctic..

    ReplyDelete
  9. It's a problem of horizon. The equilibrium exists and DSGE works... in the long term, the very very long term. 150 years maybe.
    For the short term, DSGE is neither right nor wrong, it is irrelevant. Keynes had said it before Lucas' critique. In the long run, we are all dead.
    DSGE is like mid-term weather forecasting (January is likely to be colder than August) or a broken clock (that's got the time right twice a day).

    ReplyDelete
    Replies
    1. No, I would suggest that that kind of time horizon is governed by technology, the environment, international politics, and other things generally outside the scope of economics. For the medium term, if we are in a multiple equilibria situation then equilibrium economics could still be very, very wrong.

      Delete
    2. "It's a problem of horizon. The equilibrium exists and DSGE works... in the long term, the very very long term. 150 years maybe."

      How do you know this?

      Delete
    3. Anonymous5:25 AM

      You are thinking about a CGE model. A DSGE-model models the short run.

      Delete
  10. Anonymous9:44 AM

    Great post. My one quibble is (like Robert Vienneau above) to ask how much microfoundations really matter here. It is true that DSGE models are a purely equilibrium approach and that this is a weakness. It's not true that economists stick only to pure equilibrium processes.

    I'm referring, of course, to AS-AD models which strictly speaking aren't equilibrium models at all, but rather "equiblibrium disequilibrium" models similar to the types of models you encounter all the time in thermodynamics. The disequilibrium process (price adjustment) is presumed to be "slow" (or slower at least) and the equilibrium process (market clearing) is presumed to be "fast".

    This is almost certainly a good assumption (except during financial panics when prices are adjusting faster than the market is clearing at current prices...) most of the time. If the speed of the adjustment processes get too close, that's when you get turbulence and chaos.

    The problem, I think, is that economists really don't have a model for tatonnement (the process of price adjustment, and ,yes, I probably spelled that wrong) which means that we really don't have any way of handling the disequilibrium process except to assume it away.

    (BTW, I seem to be having issues posting this comment)

    ReplyDelete
  11. My one quibble is (like Robert Vienneau above) to ask how much microfoundations really matter here.

    I think the better question to ask is why they wouldn't matter. Unless either 1) you have a very good method of aggregating individual agents, or 2) most macro phenomena are emergent rather than aggregate, then you want microfoundations in there.

    (1) may be true but it may not. And if (2) is true, you want microfoundations anyway, so that you can verify that the phenomena you're observing are truly emergent!

    ReplyDelete
    Replies
    1. Anonymous9:48 AM

      Weather forecasting does not use microfoundations. It does not look at agents satisfying Navier Stokes or other fundamental rules. It is solely modeling on macro phenomena (fronts, jet streams, etc.), and it does not care in what way those phenomena emerge. And why should you care?

      Delete
    2. Anonymous9:57 AM

      One continuing issue, however, is that I don't think the necessary emergent phenomena have been identified. There has been so much focus on identifying macrophenomena that look like micro-foundations (GDP, AS, LM, etc.) that the true macrophenomena are not there. One exception is perhaps the velocity of money. So I think the problem is not a lack of micro-foundations, but too much micro-foundations already there.

      Delete
  12. We'll chalk this up to my point:

    "Those that argue sticky wages are a fact, and thus we must have x,y,z... have given up any right to promote policies that support sticky wages."

    It is one thing to argue there are problems with moving towards "E," but when you make the argument...

    IF you are also the guy arguing for policies that making clearing the market HARD, well then the only problem is you.

    We don't have to assume that Noah exists, and that his policy preferences matter and then listen to him bitch about the market when his policy preferences are screwing the market up.

    The likely outcome is we just don't listen to Noah at all, and the market clears far more quickly, and he is simply unhappy.

    ReplyDelete
    Replies
    1. Doesn't this rather depend on WHY prices are sticky? Prices may well be sticky for perfectly good micro-economic reasons (it is costly to change them or the information about what to change them to is costly or unavailable), and just making markets clear is not necessarily only thing of value in the world.

      Delete
  13. Anonymous11:56 AM

    I think that a little reading of economic history is in order. Way back in the '50's, Phillips described economic stabilization as a PID (Proportional, integral,derivative) feedback loop, as well as its effects on policy. Take a look at this IEEE Control System magazine article on Phillips and his hydromechanical simulator. http://oro.open.ac.uk/7942/1/04064850.pdf

    Cambridge has a fully restored working model that they let first year econ students play with. Given that it will solve non-linear equations based only on cutting the right slot in an easy-to-change plastic graph (see article for further description) I'd love to see a real time Moniac mediated debate, say between Krugman and Cochrane. They could sell tickets to that one.

    ReplyDelete
  14. Anonymous12:03 PM

    "Friedman was notoriously mischievous and slippery in his argument, so I’m not sure he really believed what he said."

    Haha. "Frictionless" argument. I like that.

    Gives new meaning to his phrase "Free To Choose" ....

    ReplyDelete
  15. "But the main reason to contemplate using a weather-forecasting approach is that the real economy may just be too chaotic for DSGE to be very useful in modeling it."

    I feel it's necessary to point out that the main reason why many people so often hang desperately onto simple, yet largely useless, models is that simple models are easier to comprehend. If you just assume that they work rather than worry about when or how they might work, life becomes much easier. If you can create a situation, whether realistic or not, where your model works, then regardless of any assumptions you may have made, it can be safely assumed to work universally. If it appears not to work, there must be something out of whack with your data. Essentially, this turns (as so much of human effort does) into a form of religion, where doubts are a sign of lack of faith, and challenges are heresy if not actual sacrilege. If I see this in science, then it doesn't surprise me to see it in the philosophy we call economics.

    ReplyDelete
  16. Nice post! One of the problems is that as one moves away from simple systems that can come into stochastic equilibrium quickly and easily, you start to see how complexity makes it difficult to do any kind of prediction. Look at the call that was made by the weather service last week in terms of giving everyone a big heads up that there would be some very violent storms with tornadoes. It was a 48 hour advance warning and while it didn't mitigate any property damage, it did probably save some lives.

    I can recall a lot of years ago when I was taking first year physics and we were assigned a task to take some activity we were interested in and analyze the physics behind it. Since I was a baseball player in high school I naively thought that coming up with the equation to predict a home run was in order. Though strictly Newtonian, it quickly ended up being very complicated to analyze.

    One final thought, I'm sure being a good economist you have read Minsky's fine book on economic stabilization.

    ReplyDelete
    Replies
    1. Anonymous2:08 PM

      That is why I brought up the use of control systems theory, re Phillips, in my earlier post in this thread. By attempting to model every little wiggle in markets, GDP, etc.,economists seem to be unable to reach agreement on identifying trends that lead to disaster in a sufficiently robust way as to inform policy makers as to the correct action, which, as we can see with the European crisis, leads them to apply control inputs that will clearly lead to a crash. The most useful thing economists could do now is to come up with a robust equivalent of aviation weather radar and cockpit indicator systems to stop us flying into potentially destructive situations, such as the shadow banking system. The failure of VAR to inform course and velocity lead to disaster. For a chilling analogy of misreading faulty instruments,and making incorrect control inputs, look at the transcript of the last few minutes before the South Atlantic Air France Flt. 447 crash...
      http://www.popularmechanics.com/technology/aviation/crashes/what-really-happened-aboard-air-france-447-6611877 European austerity measures fit all too well with this analogy.

      Delete
    2. "complexity makes it difficult to do any kind of prediction"

      Disagree strongly with the "any kind" part of that statement. I would go so far as to say that *only* chaotic, "unpredictable" models can provide insight into what causes chaotic situations and what makes them more or less likely. You can't ask those kinds of questions with a model that, for simplicity's sake, assumes away those kinds of problems.

      Predicting when and where a tornado will happen is not the only use for modeling them. We'll never be able predict tornados very far in advance, but models of them still provide a lot of explanatory value; they help to show why better understood laws of physics give rise to these phenomena, and they can predict what sorts of conditions are most likely to lead to tornados.

      Delete
  17. Anonymous1:23 PM

    This was the most atrocious quoting of Bob Lucas ever. What a liar! Read what he actually wrote!

    "On the theoretical level, one hears talk of a "disequilibrium dynamics" which will somehow make money illusion respectable while going beyond the sterility of dp/dt=k(p-pe). Without underestimating the ingenuity of either econometricians or theorists, it seems to me appropriate to entertain the possibility that reconciliation along both of these lines will fail, and that one of these traditions is fundamentally in error."

    How can you even pretend to be an economist with a job at a horrible school when you quote like this?

    ReplyDelete
    Replies
    1. Can you stop pretending to be dumb now? Thanks. The full Lucas quote says exactly what I said it said. He is saying that the effort to construct disequilibrium dynamics will fail. Duh.

      Now stop spamming my comments and go get a girlfriend, you will be a happier person for it, and someday you will look back on this period of your life and laugh at how hard you kept trying to lead a lifestyle that didn't make you happy.

      Delete
  18. Wouldn't it be easier to pretend to be an economist at a horrible school than at a terrific school? I mean... right?

    ReplyDelete
  19. Anonymous3:08 PM

    http://www.clevelandfed.org/research/workpaper/2010/wp1016.pdf

    ReplyDelete
  20. Noah -

    In point 2 you state: "The equilibrium may not be stable."

    How about: "The whole idea of equilibrium is a sham and a phony." The contents of the point seem to suggest that, as do reason's comments above.

    Jason Rave's comment says DSGE "assumes the economy is in equilibrium and agents are always behaving optimally given information, other agents actions etc." So the equilibrium idea is simply an underlying assumption of the model; based on what? And the modeling includes other baseless assumption like rational actors and expectations. No wonder it leads to nonsense.

    Via Robert Waldman at Angry Bear we find this NY Fed paper demonstrating that DSGE models utterly failed to predict the Great Recession - though I'm not sure that was the authors' intent.

    My comment:

    So, if I understand correctly, the DSGE model is, in realistic terms, absolutely useless; and adding a bit of ad-hocery makes it somewhat less useless, but still not at all useful.

    Is that about right?

    JzB

    ReplyDelete
    Replies
    1. Anonymous2:26 AM

      'How about: "The whole idea of equilibrium is a sham and a phony." The contents of the point seem to suggest that, as do reason's comments above.

      That's it! Correct. The problems are obvious to all.

      Delete
  21. Not just multiple, there can be a continuum of equilibria and sunspots can matter.

    I think the key point of this excellent post is that you note that eg Lucas assume that, in the long run, the economy must go to a unique equilibriumevance (ok be stationary around it). This is just assumed with no motivation. The policy irrelevance ( or with frictions near irrelevance) result is not a result but a methodological a priori.

    Note this can't be true if there is endogenous growth (I think my student Alessandra Pelloni may have been the first to note this in the literature). Some rarios might be stable, but GDP isn't. Note that when he decided to
    work on endogenous growth, Lucas ceased to discuss learning dynamics at all. Maybe because rational expectations had become the orthodoxy or maybe because he knew that assuming no need for learning matters in that new context. By the way, I keep trying to interest people in a paper on how learning vs magically knowing the joint probability distrubution of everything must make a permanent difference if there is hysteresis (as in all endogenous groth models). I can't tell if the point is too obvious or not obvious enough. Also I can never tell how many times something has been noted in the literature.

    This is all off topic. On topic comment later.

    ReplyDelete
  22. Bravo from a layman follower. I even read the wonkish ones although I understand little.

    It looks to me that this post supports the tack being taken by such as Steve Keen and Dirk Bezemer. (Se the Inet movies) Perhaps you didn't want to mention Keen, since he is being ostracized by the establishment for attacking the basis of DSGE and more with his aggressive "Debunking Economics". Be that as it may, to this layman, this pure dynamic line of model development seems much more promising than fiddling with DSGE.

    Keen claims that most economists avoid dynamic modeling because they can't do the math. What do you think?

    ReplyDelete
    Replies
    1. Anonymous7:00 PM

      Keen's claim about other economists is incorrect. Many economists are quite capable of the math: some PhD economists, especially American ones, did their undergraduate studies in math or engineering.

      Dr Keen doesn't know this because he came to economics late in life, did not do any undergraduate studies in economics, did his PhD at an Australian university where the faculty does not have that math-y profile, does not talk to or engage with other economists, and assumes he knows more than the rest of the profession. He thinks we haven't studied dynamical systems because it wasn't offered as part of the graduate program at his alma mater, and he has presumably not read Varian's book or Leonard and Van Long's book or any of the other standard texts on dynamical systems in economics that are regularly used in graduate programs. (I used the latter in my Masters degree at a different Australian university some years before Keen even started grad school.)

      And he is not being "ostracised" by the profession for attacking DSGE. Krugman took him on because Keen showed he didn't understand (or want to understand) how the money multiplier works, and because he was representing modern research as being simplistic "neoclassical" models that assume market clearing and perfect competition. This is not true, and even those mainstream economists like Krugman and Farmer, who are not supporters of the canonical DSGE model, do not appreciate the profession being misrepresented the way Keen does.

      Delete
  23. Great post.

    "I think this kind of approach could have other advantages - for example, because you wouldn't have to solve for a simple equilibrium, it might be easier to add large numbers of frictions to the model."

    I've been thinking this kind of thing for a long time:

    http://noahpinionblog.blogspot.com/2012/02/are-macroeconomic-methods-politically.html?showComment=1330026945646#c544200474217644614

    ReplyDelete
  24. Anonymous7:47 AM

    I think it's even plausible that in reality there is a continuum of equilibriums. Like in a valley, one rolls back into the valley when moving in a perpendicular direction, but one can move unimpeded and without friction along the valley. This by the way describes both 3 and 2, as there are an infinite number of equilibriums, but they are all only marginally stable (so they're not stable.) The position in the valley, i.e. the equilibrium you will reach, is determined by policy (fiscal and monetary.) However it is important to realize how the various instruments restrict you in your movement. When the instruments (like interest rate setting) restrict you in your movement along the valley, this is an effective instrument. However when the instrument restricts your movement mostly in directions perpendicular to the valley, this will have virtually no effect, and the instrument is not effective, because it allows you to keep moving chaotically along the valley of equilibriums, leading to unstable economies. I believe that this is exactly what happens with traditional monetary policy at the zero bound. I have seen this explained in terms of IS-LM type aggregate models, but isn't anybody able to model this using DSGE?

    ReplyDelete
  25. I've made this point before, and nobody seems very impressed with it, but I'll try again.

    I recall from Kendrick et al Computational Economics that when you use computable agent based modelling to try and simulate price formation in a market created by agents bumping into each other and trading, you find (forgive me if my memory lets me down) that the price in the simulated market meanders around the price that a simple supply and demand equilibrium model would predict (which forms, I think the term is, a "strange attractor")

    So it looks to me like the equilibrium model could be a perfectly acceptable simplification of a more complex reality, and so long as you understand that what's you are dealing with, you can get along perfectly well if all you're looking for is a good point-estimate of the price that will prevail (for example). The fact that in reality the price is likely to potter around in the general vicinity of this point-estimate does not make point-estimates useless.

    So it is just not the case that the belief reality has no equilibria tells us that modelling the economy as a system with an equilibrium is completely misleading. You don't need to believe in stability and smooth transitions etc., you don't need to treat equilibria models as if they are simulcra of reality that are literally true. They could still give us a useful idea of where the economy is headed.

    I am not arguing that disequilibrium, adjustment dynamics etc. aren't important, nor that everything is okay in the world of mainstream macro. As I recall, plenty of people didn't think much of mainstream macro before the crisis.

    ReplyDelete
    Replies
    1. I think all this talk about price, is avoiding the real issue. Everybody knows that the real issue is that qunatities are adjusted first not prices.

      Delete
    2. I don't think I am "avoiding the real issue" I was merely trying to illustrate that when you actually use an alternative modelling approach, with no equilibrium, (like a simulated market of autonomous agents where prices and quantities are emergent and unstable) what you can end up might be equally well described by a standard supply & demand equilibrium model, so long as you understand that model to be simplification of a messier reality (which of course is how you should be thinking of anything that tries to describe reality in two equations)

      I'd need to go back and check, but I'm pretty sure that in the simulated agent-based market both quantities and prices are determined, so I could have written the above in terms of quantities not prices, it wouldn't matter.

      [Could it be that some people are complaining loudly about how stupid equilibrium-based models are, without really knowing anything about the alternatives? ]

      Delete
    3. oh, my memory really has let me down. I just picked up Kendrick et al. Computational Economics and it doesn't seem to contain the agent-based simulated market that I have in mind.

      Delete
    4. Keshav10:49 PM

      Luis, you should check out Herbert Gintis's recent papers on agent based models, available on his website. He finds that in a particular class of models with decentralized trade, there is very rapid convergence to the law of one price and this price almost always eventually converges to within a few percentage points of the Walrasian price. Caveats - as with all agent-based modelling, it's unclear how sensitive these results are to the particular behavioral assumptions made, and it's unclear why this convergence takes place - but the results are certainly interesting.

      Delete
    5. Thanks Keshav. that's just what I had in mind - it may even have been Gintis' work I meant (it's about 3 years since I looked at this stuff).

      I know it's micro and the OP is about macro, but I don't know why those who are convince equilibrium models are so dreadful don't pay more attention to results like that.

      I wonder how many objection to equilibrium-based macro models could be addressed by a richer specification of the equilibrium outcomes, to give more role to path dependence, policy etc.

      Delete
  26. for the sake of example, say you believe that strengthening unions will raise real wages at the lower end of the wage distribution.

    So, first you have some idea connecting a policy change with a direction of response (wages up). And presumably you have some idea of magnitude too. If you advocate strengthening unions, you probably aren't doing so because you think it will increase wages by 0.001% or by 1000%.

    Now, suppose I come along with my old school search frictions style equilibrium model of the labor market in which I have attempted to incorporate the important relevant mechanisms that connect unionisation to wage determination. And based on this model which I attempt to calibrate somehow, I say, I think the policy you are proposing could increase wages by 10%, based on various factors which I have done my best to estimate at the present point in time but which of course could change in the future which would in turn change the predicted equilibrium. But of course this 10% prediction should be regarded as merely my best estimate.

    Now, you, my imaginary interlocutor, also propose strengthening unions because you think it will raise wages, but you think equilibria models are all wrong. You say: whoa there! the process by which wages will adjust isn't at all clear, there's no reason to think there is such a thing as a single stable equlibria wage. You're whole approach is all wrong!

    and ... so what? What difference does your rejection of equilibrium approaches to modelling the economy make? You've Sonnenschein-Mantel-Debreu, but your support of a particular policy reveals that don't really think "anything might happen". You think it's terrifically important to reject equilibrium-style thinking, and I'd like to understand what changes when we do.

    ReplyDelete
  27. "But the main reason to contemplate using a weather-forecasting approach is that the real economy may just be too chaotic for DSGE to be very useful in modeling it."

    Absolutely, exactly, right. And, I hate to say, obvious to anyone who isn't an economist.

    Before the recent blogosphere discussion of DSGE, I really thought it sounded interesting. I even had a place at university to do a masters in economics. Now I am glad I didn't take it (thanks to a job offer from a hedge fund). I spent quite a lot of my undergraduate career on pointless, self-sustaining academic disciplines (modern philosophical ethics, philosophy of logic and language, international-relations theory, etc.) and I feel like I have narrowly dodged getting caught up in another one.

    ReplyDelete
  28. Just another thought -- here is one of my favourite graphics:

    http://xkcd.com/808/

    Crazy phenomenon: DSGE modelling.

    If it worked, companies would be using it to make a killing in: Macro trading.

    Are they?: No.

    Macroeconomists are completely useless at predicting the behaviour of the economy. It seems to me that they could get a long way if they could find a way to predict price and volume in various markets (e.g. the labour market).

    But there are already people making a killing predicting price and volume in various markets -- they are called (successful) systematic hedge funds. If you really want to understand what can and cannot be predicted about the future evolution of a market, it strikes me that they would be a good place to start.

    ReplyDelete
  29. I'm not sure I understand your proposal: "ignore the equilibrium and focus entirely on the micro level transition dynamics". OK, the micro level transition dynamics give you an Euler equation for your jump control/jump variables so that t+1 is knowable given t. But normally we can pin down t=0 only because we know t=infinity is at the equilibrium. If we ignore equilibrium then how do we choose values for our control variables?

    ReplyDelete
    Replies
    1. If we ignore equilibrium then how do we choose values for our control variables?

      I think there are a number of different ways to do this.

      As for Euler equations, I don't believe they accurately describe individuals' consumption behavior at all...do you know about this?

      Delete
    2. I'm not claiming consumption euler equations are accurate! But there is a methodology there.

      Can you link to a paper that provides an example for the sort of methodology you think is appropriate.

      Delete
    3. Not sure what I think is appropriate. But in another comments thread, someone suggested Howitt (JPE 1992).

      Delete
    4. Anonymous2:33 AM

      Name dropping Euler.

      Delete
  30. Anonymous4:16 AM

    "Disequilibrium Dynamics" is a very nostalgic word. It used to be a little bit famous approach among Japanese economists due to Professor Katsuhito Iwai's same titled book, http://cowles.econ.yale.edu/P/cm/m27/index.htm , although it was considered as heresy in the storm of rational expectation revolution.

    ReplyDelete
    Replies
    1. Yes, I know Katsuhito (and his research). I had that in mind when I was writing about the possibility that disequilibrium is the norm. Should have cited him...

      Delete
  31. Anonymous5:33 PM

    Maybe you will find my paper interesting:
    http://econpapers.repec.org/paper/cesceswps/_5f4017.htm

    ReplyDelete
  32. Anonymous2:18 AM

    "Reason 2: The equilibrium may not be stable"

    Your statement is self contradictory. If it was an equilibrium it would be stable. You should know that.

    The equilibrium assumption is were theory assumes away reality. So, the assumption cannot be even called and approximation for more than one point in in time.

    Reason 3 statement is debatable also.

    ReplyDelete
  33. Somehow I think this is all a bit too sanguine. Models are good or bad depending on the purpuse for what they are built. I see no problem at all in using the one sector growth model with convergence and a single equilibrium to explain basic patterns of macro variables if that model roughly fits them (which it does). The reason this approach is so popular is because it actually works remarkably well on many dimensions. I also think most good economists in the RBC tradition acknowledge all that is written in this post and are in fact quite humble about its limitations, and that it is others that draw abusive inferences on their work. To me, it is all a non issue. Use the best model for the purpose at hand and define best clearly. It is not that hard.

    ReplyDelete