Wednesday, October 5, 2011

Simple-Minded Pseudo-Macroeconomists

Tyler Cowen does not like the IS/LM model. Excellent. Indeed, when Hicks wrote his 1937 paper "Mr. Keynes and the Classics," which interpreted Keynes's General Theory as the now-familiar IS-LM construct, he finished with this:
The General Theory of Employment is a useful book. But it is neither the beginning nor the end of Dynamic Economics.
Thus, Hicks himself is telling you: "This is my effort to figure out what the heck Keynes was trying to get across. Don't take it too seriously though. I'm sure there is plenty of good research to come that will put all of this into perspective, and indeed may replace it." Hicks would probably have been surprised at what happened to IS-LM. Generations of textbook writers found IS-LM a very convenient model to use in getting basic Keynesian ideas across to undergraduate students. However, frontier macroeconomic researchers did not take IS-LM seriously after the early 1970s. By about 1980, IS-LM had essentially disappeared from the top economics journals and from the top PhD programs in economics. But one could still find some version of IS-LM in undergraduate textbooks.

How is IS-LM used today? You do not see it in published macroeconomic research, as a framework for discussion among policymakers, or in PhD programs in economics. It is certainly not necessary to use it in teaching Keynesian economics to undergraduates. In the third edition of my intermediate macro textbook, you will not find an IS-LM model. I have found what I think are more straightforward and instructive ways to get Keynesian economics across, and to get it across in line with what modern Keynesian researchers actually do. For example, I do a version of a Keynesian coordination failure model that looks like what Roger Farmer did in the early 1990s, and an undergraduate version of a Woodford sticky-price model.

So, given that IS-LM is not used by any serious macroeconomic researchers or practitioners, and that we want to represent in an accessible way for undergraduates what macroeconomic researchers and practitioners are actually up to, why would anyone care about IS-LM? Why indeed? But Brad DeLong and Paul Krugman do. In fact, they are quite passionate about it. Well, the Amish are passionate about what they do as well. While DeLong and Krugman might like to freeze the profession at its state in 1937, the rest of us have moved on. In the words of the great bard:
Your old road is rapidly aging. Please get out of the new one if you can't lend your hand, for the times they are a-changing.


  1. The New Keynesian intermediate texts like Chad Jones have dispensed with IS/LM and replaced with the 3-equation IS-PC-MR (monetary rule) model.

  2. Yes, similar idea. I like to show the students that you have to have "money" working in the background in order to hit the interest rate target.

  3. WRT BD:

  4. That's quite funny. His book is a good read. Never thought of it as Nobel prize material though.

  5. Hi Stephen,

    You write that "frontier macroeconomic researchers did not take IS-LM seriously after the early 1970s," but you don't explain why. If you want to reach a wider audience (and maybe you don't), then I'd suggest that you reduce your appeals to "frontier macroeconomic researchers," "top economic journals," "top PhD programs in economics," etc., and explain why IS-LM, aggregate demand, etc., are faulty.

    I gather that you believe great progress has been made in economics by the application of relatively high-powered math, but if you can't translate your arguments into words, it's not clear that the math bears on the world in the way you might think it does. (This is a common problem in philosophy where arguments that flow from a stipulated definition, e.g., of freedom, often bear little or no relation to what people mean when they talk about freedom.

    I should add that your citing J.R. Hicks saying that the General Theory "is neither the beginning nor the end of Dynamic Economics" might be more revealing if you'd pointed out that Sir John Hicks eventually repudiated IS-LM and did so on the basis of a deeper understanding of Keynes's own theory; and that the beginning of Keynesian "dynamic economics" began with Roy Harrod (Keynes's theory being set forth in static terms).

  6. Steve, your assertion that IS/LM is not used by policymakers is simply false. What is the 3 equation DSGE model but a dynamic version of IS/LM? That is probably by far the model most used by policymakers.

  7. "Steve, your assertion that IS/LM is not used by policymakers is simply false."

    Well, it is and it isn't. IS/LM means different things to different people. With many, many models it's possible to group together two or more equations representing the supply and demand for money, show that they imply a relationship between the return on some asset and income or consumption, and call that relationship an LM curve. Similarly one can often generate something not entirely unlike an IS curve from the other equations in the model. Some writers like to do this, either in order to help students to get their bearings or to encourage them to take the older literature seriously. Some writers like to do the opposite.

    Really, I think we should have a rule that says you don’t say ‘the’ IS/LM model is good or bad, fashionable or unfashionable, deep or superficial, until you have provided your readers with a link to tell them just what you mean by ‘the’ IS/LM model in the present context.

  8. How is recognizing that it has useful insights the same as wanting to "freeze the profession at its state in 1937"??? They still teach and use Newton's laws. Would you say that people want to "freeze physics at its state in 1687? I think that would be a little unfair.

  9. Greg,

    This isn't about mathematics. In 1975, when I was taught IS-LM (Branson version) you needed to know calculus and algebra. To do modern macro, what basic mathematics do you need? Calculus and algebra. That's for the analytical part. To construct, estimate, and solve numerically a full-blown empirical IS-LM model, for example the Fed/Penn/MIT model constructed in the 60s and 70s, the econometrics and numerical techniques are the same, or on the same level of sophistication, as the methods used by modern quantitative macroeconomists. Obviously you missed the whole 1970s discussion about why IS-LM is faulty, or you would not be asking me to explain this. The key problems are:

    1. Lucas critique. The "behavioral" relations in the IS-LM model are not structurally invariant to policy changes we want to consider. The model is therefore going to give you wrong answers.
    2. The "theory" backing up the "behavioral" relations (consumption function, investment function, money demand function, principally) are piecemeal and not internally consistent.
    3. The monetary theory in it stinks.
    4. The model is not dynamic.

    Basically, the thing is a piece of shit.

  10. "Steve, your assertion that IS/LM is not used by policymakers is simply false. What is the 3 equation DSGE model but a dynamic version of IS/LM? That is probably by far the model most used by policymakers."

    That 3-equation model is a marketing ploy, to convince Old Keynesian IS-LM types to buy into New Keynesianism. But New Keynes is not Old Keynes at all.

  11. "How is recognizing that it has useful insights..."

    That's just it. It does not have useful insights.

  12. "They still teach and use Newton's laws."

    Because Newton's laws are approximately correct. ISLM isn't even in the ballpark.

  13. Scott Sumner has an extremely useful blog on IS-LM today.

    But really, at this point, the Fed needs to target nominal 7 percent growth of GDP, and release a photo of Ben Bernanke with his hand on the lever of a printing press. The caption: "You think I can't do it? Make my day."

    I cannot fathom the prissy obsession with models, and hysteria over 1.4 percent inflation rates, while unemployment is at 9 percent, and trend GDP down by 10-15 percent. Economists sound like interior decorators on the deck of the Titanic in a snit over the shade of the chaises.

    Back in the real world, unit labor cost are down to flat for the last three years, while real estate is well down--that's about 70 percent of business costs right there. The Chicken Inflation Littles are scared of shadows in the recessionary twilight.

    Models work until they don't. People fall in love with ideas of their own creation, and then shut the doors to reality.

    This one is simple. Print a lot more money.

    The purpose of an economy is prosperity, growth, innovation, opportunity, commercial freedom. An extremely distant second is perfect price stability, if such a thing could even be measured.

  14. Jordi Gali asks in QJE 1992: How Well does ISLM Fit the Postwar Data ?
    Postwar U.S. time series for money, interest rates, prices, and GNP are characterized by a multivariate process driven by four exogenous disturbances. Those disturbances are identified so that they can be interpreted as the four main sources of fluctuations found in the IS-LM-Phillips curve model: money supply, money demand, IS, and aggregate supply shocks. The dynamic properties of the estimated model are analyzed and shown to match most of the stylized predictions of the model.

  15. IS-LM will fit any time series. That does not make it a good model.

  16. ISLM will fit any time series....

    Is that not also true of estimated DSGE models ? The work of Christiano or Jesus Villaverde or Smet and Wouters seems to just throw in as many things as necessary to get a good fit. Where's the discipline?

  17. Benjamin: Why not sacrifice a kid and the God will pull us out of this mess? I don't see any difference from the superstition about the inflation-unemployment tradeoff

  18. IS-LM models don't fit the data. Gali's New Keynesian book claims that they do too, but if you look at the metrics (comparisons of impulse response functions from the NK model and VARs), you will notice that the NK cannot deliver the persistence that we observe in reality. I don't care if you can show that there is a positive response of real GDP to a negative shock to the interest rate -- we know that the model predicts that without seeing the impulse response function. I can even calibrate the parameters so that the magnitude is the same. Who cares? Show me that the model can explain the persistent effect of a monetary policy shock as we observe in the data. NK models can't do that. Sticky prices only get you so far.

  19. Oh, and the New Keynesian Phillips curve doesn't fit the data either:

  20. Krugman:

    "In macro — or at least macro that tries to get at monetary and fiscal issues — what you need, at minimum, is to understand an economy in which there are three goods: money, bonds, and economic output."

    This is not necessarily true, but this raises a question, what model used to talk about monetary policy doesn't have this feature or can't be amended to have this feature? (Bonus points if you said the NK model, ironically.) Has Krugman implicitly accepted that search models (with bonds) are adequate? Can it be he has seen the light? Nah. He is talking about IS-LM again.

  21. Well, if you have money, then we have to be dealing with an infinite horizon (unless we want to use some trick to go with the finite-horizon case). So now there is an infinity of time-dated consumption goods (or an infinite continuum of goods in continuous time). You can address some fiscal policy issues with two periods, in which case I need two consumption goods (current and future periods). Some monetary issues I can address without thinking about the bonds. Yes, you're right, he's just thinking about IS-LM.

  22. Prof Williamson will get his wish. Look at the age of these people,Solow, de Long, Glasner, Krugman, Thoma are all pretty old in terms of intellectual capital. I cannot imagine anyone graduating with a PhD in macro now is going to be teaching this ISLM stuff. Science progresses one funeral at a time said Planck. Sad to say that a few more generations of students at Berkeley are not going to be as well educated as say the grads of St.Louis

  23. Stephen,

    Thanks for supplying your substantive critique of IS-LM. One benefit, for me at least, is that your list of four IS-LM shortcomings bears an interesting relationship to Hicks’s own reasons for abandoning IS-LM.

    “Lucas critique. The ‘behavioral’ relations in the IS-LM model are not structurally invariant to policy changes we want to consider. The model is therefore going to give you wrong answers.”

    While Hicks probably would not have affirmed the “Lucas critique” in all its details, he did come to appreciate the fact that the IS-LM diagram doesn’t capture the role of expectations in decision making and, therefore, in outcomes. But unlike Lucas, who implicitly assumed that market participants could form probability distributions over future outcomes on the basis of historical market data (the ergodic axiom), Hicks eventually came to adopt Keynes’s view (elaborated in greater detail by G.L.S. Shackle) that a) such probability distributions are chimera, and b) it’s important to take account of the varying degrees of confidence people attach to their views of the future.

    “The ‘theory’ backing up the ‘behavioral’ relations (consumption function, investment function, money demand function, principally) are piecemeal and not internally consistent.”

    Even if one insists on micro (optimizing) foundations for the “behavioral relations” implicit in IS-LM, it’s not clear what these micro foundations would look like if the future can’t be tamed by a well-formed probability distribution. Don’t think of the IS curve as a fixed line; think of it as a something that moves like a leaf in the wind as firms adapt their estimate of returns to new projects to changes in “the news.” Similarly, don’t think of the LM curve as fixed line, but as something that moves about as our degree of confidence (and demand for liquidity) ebbs and flows.

    “The monetary theory in it stinks.”

    You don’t say why it stinks, but one shortcoming is that money is exogenous in IS-LM, whereas, in fact, the supply of money depends, in part at least, on the demand for credit. In addition, Keynes gave good reasons for doubting whether the IS and LM curves are independent of one another, e.g., liquidity preference and the expected return to investment will be influenced by similar considerations.

    “The model is not dynamic.”

    Let me simply conclude by saying that Hicks came to see Keynes’s theory as belonging to the realm of history rather than to equilibrium.

  24. "That's just it. It does not have useful insights. "

    Is that why Krugman has been right for the last 3 years, regarding the size of the stimulus (too low), panicked inflation expectations (he has scoffed, correctly for years), and the austerity bug creating MORE difficulties, rather than less?

    Also, your objections substitute math for reality. Angels on the head of a pin.

    So you describe exactly what Krugman makes fun of - people wedded to their model, stuck in a small abstract circle of math, not much related to the real world.

  25. JC, I'm not even sure how I would evaluate each of those claims, let alone know whether or not they were right.

    But hey, since Krugman and you already have all the answers, what's the point of even asking the questions?

  26. "Is that why Krugman has been right for the last 3 years, regarding the size of the stimulus (too low), panicked inflation expectations (he has scoffed, correctly for years), and the austerity bug creating MORE difficulties, rather than less?"

    Whether he was right or not is impossible to say -- what the data says is merely that G went up and employment didn't. At all. ISLM doesn't work that way, so Krugman MUST say it needed to be bigger. JC is definitely in way over his head here. I suspect that he'd be in way over his head in Intermediate Micro as well. Or maybe even Principles.

  27. "But unlike Lucas, who implicitly assumed that market participants could form probability distributions over future outcomes on the basis of historical market data..."

    This is a classic critique of rational expectations, but fails to recognize a key point. Here is Bennett McCallum in the JMCB from 1980:

    "The basic idea of the [rational expectations] hypothesis is simply that economic agents behave purposefully in collecting and using information, just as they do in other activities, an idea that it is hard for an economist to reject without considerable embarrassment. But in practice, of course, this compelling idea usually gets translated into the requirement that expectations are, in the model at hand, formed in a way that is stochastically consistent with the behavior of the realized values of the variables in question. This is clearly a much stronger hypothesis, one that an economist can reasonably dispute...[This assumption, however] has one outstanding strength, namely, the weakness of its competitors. Each alternative expectational hypothesis, that is explicitly or implicitly posits the existence of some particular patter of systematic expectational error. This implication is unattractive, however, because expectational errors are costly. Thus purposeful agents have incentives to weed out all systematic components...

    Adoption of the strong, operational version of the rational expectations hypothesis does not, in my opinion, compel the analyst to believe that there are not detectable patterns in expectational errors of the past."

    After 30 years, people are still arguing that rational expectations is a useless concept because it assumes that economic agents know everything. The main insight of RatEx is the idea that individuals do not make systematic errors. I think that is a fairly reasonable assumption.

    In addition, if the world is really non-ergodic, then this applies to the decisions of policymakers as well. This would seem to rule out fine-tuning policies; something that many advocates of this critique of RatEx seem unable to recognize.

  28. Glasner doesn't like IS-LM:

  29. "The main insight of RatEx is the idea that individuals do not make systematic errors. I think that is a fairly reasonable assumption."

    Rational Expectations, from what I've heard, says that people are paying attention to politics and government economic policy and a great deal of information about the macroeconomy, and so they see the patterns and anticipate and see what's going to happen and change their plans and behavior accordingly in a rational skilled sophisticated way.

    But if the vast majority have little idea of what's going on, have little knowledge of economics to analyze it well, then how are they going to do what RatEx says they will? How? How?? How can you change your plans and actions based on government economic policy when you don't even know what government policy is, and even if you did, you have little economics education to know how to change it in the optimal way that RatEx assumes you will.

    Surveys again and again and again and again show stunning ignorance of the economy and politics, which is understandable given how complicated the world is and how busy people are. And from what I've seen, RatEx doesn't just say people don't make systematic mistakes (which they do anyway, like always blaming the party in the Whitehouse for the current economic cycle no matter what). If people's mistakes are only idiosyncratic, but huge, you will still have RatEx being a very weak factor in economic behavior, a near sunspot.

    Stephen, suppose you have a factor like RatEx, which in reality has strength of just 1 in the economy (with 10 being everyone does it always with 100% skill, education and sophistication). Now you have two models: one ignores RatEx, giving it thus a strength of 0. Another goes to the other extreme and assumes everyone does it always with 100% skill, education and sophistication, thus giving it a strength of 10. If the real strength is only 1, then which model is closer to reality. Is 0 closer to 1, or is 10 closer to 1?

    Now, ideally you're going to, when you make your interpretations to reality and policy, adjust up when you use the zero model and down when you use the 10 model, but at a first unadjusted cut, which model, if all other things are equal, is closer to reality?

  30. Josh,

    If the world, or at least the human part of it, is non-ergodic, does the notion of "systematic errors" still make sense?

    The possible outcomes of a decision don't present themselves ready-made for the decision maker. Rather, agents must imagine possible sequels to the alternative courses of action they're considering. If they can't imagine all possible sequels, would the errors that (are likely to) follow be systematic? Is our inability to imagine all outcomes a source of systematic or only random errors?

    If black swans of one kind or another pop up from time to time, but people continue to ignore the tails of the distribution, is that a systematic error or a source of systematic errors?

    Can rational investors always make money by exploiting systematic errors (assuming this notion still makes sense)?

    The problem is not so much that decision makers aren't rational (though you'll find very few people who understand even the most rudimentary principles of probability), but that the subject matter of our judgments may be an inappropriate object of a probability distribution.

    And, yes, the same problem faces policymakers.

  31. Economists need to show some professional ethics. In physics they don't tell students any longer, like they did in the 1950s, that electrons orbit around the nucleus. In psychology they no longer tell students, like they also did in the 1950s, that babies are born blind. In medicine they no longer believe that antibiotics cure colds and flu. So stop teaching students that the IS/LM model describes the world.

  32. Personally, I find this piece to be pretty convincing. Any thoughts?

  33. There is a reason that Christian Eichenbaum and Evans' model is the foundation of modern quantitative policy models and is used by every central bank in the world: it gives great insights, it provides an excellent account of the data and is being extended to allow for financial market frictions.
    What quantitative model is the alternative? Please don't about scream bells and whistles: every model including Lucas' Jet piece makes strong assumptions.

  34. From Minnesota with loveOctober 8, 2011 at 7:27 PM

    The Northwestern approach is to reverse engineer the model until it fits the data. It has little value because a C&E model will always fit the data they are analyzing. It will be useless for out-of-sample events, that deviate from the data set that C&E used.

  35. I guess you think maximum likelihood is a obviously a bad idea. In fact there is a long tradition of using MLE in economics: even in Minneosta - see Hansen, Sargent and Sims. In any event, CEE type models has loads of over identifying restrictions so it is incorrect to say that it has to fit the data.
    As for out of sample predictions: the FED, the ECB and the IMF use variants of the model all the time to analyze out of sample scenarios and find that the exercises are useful.
    Finally, what concrete alternative do you have in mind?

  36. The most honest assessment of DSGE comes from Caballero, who says they are not yet ready for policy analysis. The CEE type models mentioned by anonymous above are of course useless for this purpose since they totally fail the Lucas critique, just like the old Penn-MIT models of the 1960s.

  37. I can't imagine what you mean. The cowles commission defined a parameter as being structural relative to a class of interventions. Can any model be invariant to all interventions? Of course not. Lagos and Wright make reasonable but very strong assumptions about day time and night time markets. Surly there are policies which would cause agents to re-organize. But that doesn't mean Lagos and Wright is useless because it fails the Lucas critique. The Lucas critique needs to be applied with more nuance or you Williams to reject all models that you or I have seen. In the end Potugal judgements must e made.

  38. Sorry for typos. I meant to conclude by saying `The Lucas critique needs to be applied with more nuance or you will reject all models that you or I will ever see. In the end empirical judgements must be made.' It may be correct that you don't like CEE. But what data are you referring to and what model do you like more?

  39. From Minnesota with loveOctober 9, 2011 at 2:17 PM

    C&E do not estimate their model via MLE. Furthermore, i do not have to be able to find a cure for the flu before I can rule out echinesa as an effective remedy. So i do not have to have a better model before I can rule their one out as worthless. Perhaps monetary policy has little effect on the real economy.

  40. CEE estimate their model via GMM which does have over identifying restrictions. But it's clear that we're into religion here so there's no point in pursuing matters. Have fun with playing with technology shocks.

  41. From Minnesota with lovdOctober 9, 2011 at 8:32 PM

    Really! Most of the output fluctustions in a C&E model are from trchnological shocks.

  42. Not true - there are no technology shocks in CEE. Important detail. In Christiano's newest work, he does allow for technology shocks but finds that financial market type disturbances are the ky to business cycles. One more thing Chari, McGratten and Kehoe were wrong about.

  43. Is there anything that can be saved in the Minnesota approach to macroeconomic fluctuations?

  44. @8:55 Anonymous: CKM show that the investment wedge doesn't play a large role in fluctuations. However, that is not at all the same thing as saying that financial frictions do not play a large role. They show that financial frictions usually map to labor or TFP wedges, both of which play a large role in business cycles.