Saturday, April 4, 2015

NK Models

Brad DeLong has a post on New Keynesian (NK) models. Things are certainly looking up, as I'm finding DeLong more agreeable by the day. As far as I can tell, we both think there is a deficiency of safe assets in the world, and we're not big fans of NK. Soon Brad will be doing monetary theory and quoting Neil Wallace, I have no doubt.

It's not like Brad has everything right though. He says:
The baseline New Keynesian model was not, originally, intended to become a workhorse. It was intended as a proof-of-concept...
That's definitely not correct. Mike Woodford can correct me on this, but my impression is that he came out of graduate school with a specific goal in mind, which was creating a version of Keynesian economics that would fit into modern macro. Ed Prescott's project left central bankers scratching their heads about what they were supposed to be doing, and Woodford and others stepped into the void. Interest and Prices is, I think, intended as a handbook for central bankers. There was a lot of effort put into marketing the whole NK project to the world's central banks. This is ongoing, and has been institutionalized, for example here.

But then, Brad gets to his criticisms:
But the extraordinary shortcuts needed for tractability were and are a straitjacket that makes it extremely hazardous for policy analysis. It cannot fit the time series. And when it does fit the time series, it does so for the wrong reasons.
He's not specific enough, but I agree with him. And other people do too. For example Chari/Kehoe/McGrattan.

But then we part ways again:
So why require everything to fit in this Procrustean Box? This is a serious question–closely related to the question of why models that are microfounded in ways we know to be wrong are preferable in the discourse to models that try to get the aggregate emergent properties right.
Again, Brad needs to expand on this to make clear what he's thinking, but my understanding is that the "Procrustean Box" of preferences/endowments/technology/information/equilibrium concept is too confining for him. He also might think that post-1970s macroeconomics opened a Pandora's box - setting loose evil forces that ruined much of the profession. Just guessing of course.

In any case, we're left wondering who the "we" is that knows microfounded models to be wrong. What are those microfounded models that are wrong, and how are they wrong in ways that mislead us? What specific "aggregate emergent properties" are there that some other models are trying to get right, and what are those models anyway?

Several points:

1. Saying a macroeconomic model is wrong misses the point. These models are all wrong, in the sense that, with sufficiently good data in sufficiently large quantities, which has in some sense performed the right natural experiments for us, we can reject any model. But that doesn't make these models useless - they can indeed be useful in carrying out the purpose for which they were intended. You could also stick to the letter of the law, and come up with a crappy model, of course.

2. We can address Lucas critique issues if our models are properly "micro-founded" - i.e they use the best available theory. But being explicit also keeps people honest. If you're specific about all the forces at work, and formalize them - as we're expected to do when we publish papers in serious academic journals - it's easier to understand the ideas, and to check them for accuracy and consistency. This is just good science.

3. It takes a model to beat a model. You can say that you don't like NK, but what's your model? Show me how it works. If you've got a model of monetary policy, show it to central bankers. Try to get them to buy into your ideas. Argue about it in public. Publish papers. Go to conferences.

40 comments:

  1. "It takes a model to beat a model. You can say that you don't like NK, but what's your model."

    That's like saying," the world is flat is the best we have, so lets stay with that."

    What is important is that we have used the all information we do have and properly investigate it. That means, for example, engaging widely with people and accumulated knowledge outside the profession. Often we can't get all the information, even if we make genuine efforts to engage with people out on the field, and go out on the field ourselves. But if this is done, there is a good chance we have enough information to inform the right policy decisions, as Keynes did in 1936. We do not need internally consistent models to make informed policy decisions. We just need to have as much of the facts as we can get to get as close to the truth as we can . Then we can then try and put the story together as best we can. If you mean by micro-founded using all the evidence and knowledge about the target system available, I do not think anybody has an issue. If you mean by using "the best available theory", well, I think that is going to be problematic for many people if the best available theory is from the beginning flawed, controversial, or if it is a one-size-fits-all model which has little relevance to the particular issue at hand or makes us look to narrowly at it. We can get something internally consistent, and miss the most interesting and important thing going on.

    ReplyDelete
    Replies
    1. "That's like saying," the world is flat is the best we have, so lets stay with that.""

      Exactly. Some people did say that, then some others came up with a better model. It takes a model to beat a model.

      Delete
  2. I think one feature that DeLong may bring up is rational expectations. For example, suppose that raising the IROR is welfare-enhancing in a model under RE. That's a useful starting point. But many people may be concerned that we do not know exactly how people will interpret the Fed's move, in which case the policy may prove harmful. One could put together a model with alternative ways of forming expectations, for example sticky expectations as in Mankiw and Weiss and make the point, but I think DeLong would ask why it is not legitimate to raise these concerns verbally. I am not adopting this viewpoint, just trying to guess what he may have in mind.

    However, I do think that people often "stick to the letter of the law" and come up with crappy models (I have some examples in growth theory in mind) and yet quite a few of these models are published in top journals because of their "elegance" or the pedigree of the author. And then when, as a result, their out-of-sample performance ends up being terrible, the same people publish new papers trying to explain why their previous models failed, which paradoxically expands their publication record and enhances their career. Isn't this an incentive problem? And what if policy-makers take these models seriously?

    ReplyDelete
  3. 1. How should we decide which models are "useful", if not in terms of out-of-sample goodness-of-fit (or marginal likelihood)? If the concern is "unrealistic" assumptions, we can always drop any nuisance variables from the validation set.
    2. The vast majority of microeconomic details about preferences, technology, endowments, etc., cannot possibly matter to the macroeconomic phenomena of interest, simply because the dimensionality of the latter is trivial compared with the former. Spilling the microeconomic guts of a macroeconomic model does not keep people honest, since the empirical failure of the microfoundations tells us (almost) nothing about the empirical merits of the macro model. If one could prove that any macro model of type X must incorrectly assume microfoundation Y, that would keep people honest, but such proofs are few and far between. Any particular microfoundation is merely sufficient, not necessary, for its corresponding macro model.
    3. If it is difficult to make a career publishing macro models without microfoundations, then economists will not do a whole lot of that. Then, when critics point out the irrelevance of microfoundations, macroeconomists will say "it takes a model to beat a model". This is an equilibrium configuration, but a sub-optimal one. Maybe it is difficult to make a career like that because microfoundations matter, but I'd like to see that argument--which takes us back to point #2.

    ReplyDelete
    Replies
    1. On 1: A VAR does a great job of fitting the data, but I can't use it to think about monetary policy.

      "the dimensionality of the latter is trivial compared with the former."

      This would imply that we can't get anywhere in economic science. Theory is about reducing the dimensionality of the problem so we can get some understanding of what is going on.

      "...since the empirical failure of the microfoundations..."

      That's just it. You put together the theory so it's consistent with the micro evidence.

      On 3: Or, you could just say: Thank God we got rid of those crappy models.

      Delete
    2. 1. I repeat--what, then, makes one model more "useful" than another?
      2. No, it implies that a microfounded macro model is essentially two independent models--one can be decisively rejected by the data without casting any doubt on the other. Requiring every macro model to be microfounded at best introduces an irrelevant set of additional assumptions, and at worst severely restricts the space of macro models that can be studied. Note that Newtonian physics is not couched in terms of wavefunction collapses, despite being a limiting case of quantum mechanics, since it is almost impossible to think about macroscopic objects in terms of quantum mechanical concepts. Quantum mechanics is the best theory we have ever had of any subject matter, and yet we speak a completely different language when talking about everyday objects which must, ultimately, supervene on quantum mechanical phenomena.

      Delete
    3. 1. It depends what use it was intended for. It it's a model intended to guide central banks, it should do better - by some clearly-defined criteria - than the best alternative, in formulating policy.

      "Requiring every macro model to be microfounded at best introduces an irrelevant set of additional assumptions, and at worst severely restricts the space of macro models that can be studied."

      Not at all. It has to all hang together, which is part of the point. If you're finding yourself restricted when you're doing this, that's probably a good thing.

      "Note that Newtonian physics,,,"

      While we can pick up useful things from the natural sciences, economics is not physics. Physicists are indeed sometimes confused about that.

      Delete
    4. I should add that, before about 1970, a lot of macroeconomics was about various versions of a Hicksian (1937) IS-LM model. That was pretty restrictive. Since 1970, there has been an explosion in the variety of models and questions in macro. So, the modern approach is actually liberating.

      Delete
    5. 1. Suppose it is a model intended to guide central banks. One possible criterion is inflation control: model X is better than model Y just in case policies recommended by model X produce a lower mean squared deviation from the inflation target than policies recommended by model Y. Is this the kind of criterion you have in mind? If so, what kinds of evidence could establish this? Two central banks facing similar macroeconomic environments, but employing two different models?
      2. Suppose I have a model that says that whenever the central bank sets its policy rate target above what is implied by the prices of policy rate futures contracts, inflation goes down. One microfoundation for this is the usual new Keynesian story about unexpected contractionary monetary policy leading to lower inflation. But the policy-making merits of my model have nothing to do with the correctness or incorrectness of this microfoundation, right?

      Delete
    6. 1. Sure, why not? You can see what the problem is though. We can never re-run history. Sometimes we're just stuck with fitting the models to the data, and doing policy experiments in the context of the model to decide what the policy rules should be. We can use those policy rules while they seem to be working OK, in the sense that no one is complaining. Then maybe you get a surprise - the financial crisis, for example. The model doesn't tell you what you should be doing; there's stuff going on in the world your model seems inconsistent with. So you re-calibrate.

      2. "But the policy-making merits of my model have nothing to do with the correctness or incorrectness of this microfoundation, right?"

      Suppose I'm the central banker you're coming to with this idea. If you gave me that line about "the usual new Keynesian story about unexpected contractionary monetary policy leading to lower inflation," I would make you be more precise about that. What I think you mean by "contractionary" is an increase in the nominal interest rate target. Indeed, we have stories - including the NK story - about the short run effects of monetary policy on inflation, and your story is consistent with that. Maybe you want to worry about the long run effects too - i.e. if the central bank increases the nominal interest rate permanently, that would lead to more inflation. Have you thought about that. Then, I'm going to ask you what you have in mind. Do you want a different operating procedure for the central bank? Apparently so. Looks like you want the FOMC to set the policy rate based on some futures price. Now, I want to pick your model apart and see how it works, in light of what I know about the actual mechanics of central bank intervention in financial markets. I want to understand all the bits and pieces of your model and be able to evaluate it. Sorry, if it's just supply and demand curves, or it's only a regression equation, I'm going to be hard to convince.

      Delete
    7. 1. OK, so engineering success is the measure of a macroeconomic model. Machine learning researchers have developed countless models which enjoy remarkable engineering success (think of the models underpinning autonomous cars, etc.), despite being mostly uninterpretable. Would macroeconomists object to the FOMC making policy based on the decisions of a black box developed using machine learning methods? I understand that such a device does not yet exist, but it seems to me that one could be designed with some effort. Why do macroeconomists seem uninterested in this possibility if engineering success is the measure of their output?

      2. "I want to understand all the bits and pieces of your model and be able to evaluate it."

      Do you need to understand all the bits and pieces of your car before you're willing to drive it? If I have a model that is successful in policy making, why does it matter what the bits and pieces in it happen to be?

      Thanks, by the way, for your considered and interesting responses.

      Delete
    8. "Do you need to understand all the bits and pieces of your car before you're willing to drive it?"

      Obviously you don't have to understand all the features of monetary policy in order to use currency, for example, as a means of payment. But if you go to Honda with a prototype for a car, they are going to have to know how it is put together before they decide to build it, don't you think?

      Delete
  4. "There was a lot of effort put into marketing the whole NK project to the world's central banks. This is ongoing, and has been institutionalized, for example here."

    Your link tells me that central bankers who cough up the fee can attend lectures such as Modeling inflation by Thomas Sargent. You know Sargent better than I do (that wouldn't be hard) but I'm a bit surprised to hear that he's a salesman for the NK project.

    Does it get us anywhere to say that models are properly micro-founded if they use the best available theory? That's hopelessly vague. I don't think you and Brad DeLong can be said to have found much common ground merely because you're dissatisfied with the NK model. Who isn't? We might be getting somewhere if you could agree on what makes a theory the best available.

    ReplyDelete
  5. "Your link tells me that central bankers who cough up the fee can attend lectures such as Modeling inflation by Thomas Sargent."

    Heck does nothing change. Central bankers paying Sargent for lectures about how to run the post-crisis economy is even worse than going to Summers for advice.

    ReplyDelete
    Replies
    1. I have never been to the Northwestern workshop, but it looks like this is a about teaching central bankers about the intricacies of macro modeling. There are parts I think, that are quite technical. Tom Sargent has been engaged for his entire career in producing research - indeed, pushing the frontier of macroeconomics. He's also been steadily occupied as a teacher and supervisor of graduate students in economics. Larry Summers indeed had an active research career, but he was a dabbler in macro. Mainly his interests were in public finance and labor. Summers left Harvard 24 years ago for a career in running stuff.

      Summers is very good at what he does (most of the time). He can speak extemporaneously and make sense to politicians and people who have no background in economics. Ask Tom Sargent to do that, and you might be disappointed in the results.

      But, if I had to choose someone to fill in for a graduate macro class for a semester, and the choice was between Sargent and Summers, with no budget constraint, Summers would be a really bad choice. The students might be entertained, but they wouldn't learn a damn thing about how to do macroeconomics.

      Delete
    2. OK, you are the expert here, we respect your opinion and that is why we are all read this blog, even if it surprises us sometimes. But how much did the ideas of Sargent et al really have an impact when it counted - eg in the decision making processes to led to the macro policy response of 2008, when credit channels and the flow of funds seized up, finance ministers, HOGs and central bankers around the world immediately met up in Scotland and decided that something had to be done, and fast. I do not know what you call what they did, but if it is macro-economics, it is eighty years old. The lesson of that whole episode leading up to that was that a few people had their eyes off the ball for a long time, and there is a widespread suspicion that the academic macro-economic establishment and what it was doing was likely to have been a big part of it. If the profession had not improved on theories to deal with the times when the world most needs them, what was it doing? And by improved theories surely we mean something that leads to a more effective policy response to events like 2008, not eg whether it can be dressed up in symbols and conform to micro-economic theory (eg, rational agents, limited resources and unlimited wants, efficiency-equity tradeoffs - all highly political and very controversial stuff).

      Delete
    3. " I do not know what you call what they did, but if it is macro-economics, it is eighty years old."

      Absolutely not. Go back and read the 2007-2008 FOMC transcripts. You think Bernanke didn't learn modern macro?

      Delete
    4. There is nothing controversial or political about rationality, limited resources, or equity-efficiency tradeoffs. The only people who find these ideas political are those who are losing the intellectual war and therefore want to paint the winners as evil and heartless. Rationality is empty -- since you don't know what my preferences are, how can you tell if I'm not acting in my own best interest? If you try to choose for me, I'll tell you to piss off. Limited resources -- how is this political or controversial? Equity-efficiency tradeoffs arise in some models and not in others.

      Do you people ever even read economics beyond what some nutsack Austrian or bitter Post-Keynesian says? Paul Davidson knows nothing about the way macroeconomics is done now, so we ignore him, just like we ignore you.

      Delete
  6. "What specific "aggregate emergent properties" are there that some other models are trying to get right, and what are those models anyway?"

    The whole is more than the sum of its parts. In physics a proton is more than just 3 quarks, in chemics adding a proton and a neutron leads to something totally new, in biology organic molecules can combine to form complex lifeforms. The textbook definition of society is that it is neither just the sum of all its individual members nor an organic whole, it is both and none, something in between.

    It is quite strange to witness social scientists denying this very basic feature of their object of study.

    Back to econ, of course there are benefits to microfounded macro. But as there are aggregation problems in econ there are also benefits to using non-microfounded, aggregate models. It is no coincidence that central banks use both kind of models (often the core is a DSGE and the periphery is a structural model).
    Extremists on both sides want to use only one type of model: Post Keynesians and other heterodox economists want only structural models while New Classical want only microfounded models.

    ReplyDelete
    Replies
    1. "It is no coincidence that central banks use both kind of models (often the core is a DSGE and the periphery is a structural model)."

      It depends what you mean by "use." People do all kinds of things in central banks. Some of those things show keen insight; some are wrongheaded. Take the FRB/US model for example. In what sense is that "core DSGE" and "peripheral structural?" What could that mean?

      "Post Keynesians and other heterodox economists want only structural models while New Classical want only microfounded models."

      My impression is that Post-Keynesians and heterodox types want no models - they basically want to complain about what everyone else does. I don't know what "new classical" is supposed to be. People used to talk about that in days of yore I think. Part of what "microfounded" is about is indeed "structural."

      Delete
    2. "My impression is that Post-Keynesians and heterodox types want no models - they basically want to complain about what everyone else does."

      That is a bit unfair. In this category you are including the likes of Angus Maddison, Gunnar Myrdal, Joan Robinson and many others, and I think their contribution to understanding how the world works is unquestioned. Certainly many would say Angus Maddison is far and away the best economic historian ever. What they do not like is things like rational expectations, everything being submitted to rational choice theory or ahistoricism. Opposing these things is not like opposing the Law of Gravity. Keynes was a heterodox economist, and invented macro. When it comes to models, I think there is just a difference of opinion on what models actually are and what they are supposed to be used for.

      Delete
    3. "...I think there is just a difference of opinion..."

      Some people say there is a difference of opinion about global warming.

      Delete
    4. but surely the Godley and Lavoie book on monetary economics is model-driven, being based on Tobin's work.

      Delete
    5. "Some people say there is a difference of opinion about global warming."

      I would take the side that looked at the evidence - all of it - even if there were things they could not tie up and admit there are unknowables . I would want to make sure they did rigorous empirical work - measuring the glaciers compared with 100 years ago, and had consulted everyone from experts on the industrial revolution to soil experts, not those who get a whole lot of differential and Markov equations, a certain type of micro-foundation which they think is universally applicable (most have never really understood why economics has taken this micro-foundation as the basis of its analysis - which by the way came from a certain time and place, which was not a coincidence) and just boast - "we can explain what goes on with an internally valid model".

      Delete
    6. I should say, you probably find us a bit annoying, but we appreciate your responses. But I think in many ways our concerns represent those of many academics outside the discipline, those whose main game is not economics - but have to engage with it, and much of the general public at large.

      Delete
    7. "My impression is that Post-Keynesians and heterodox types want no models - they basically want to complain about what everyone else does."

      Well there is a role for these people, even if they do not have a better "model", they can at least stop people getting too comfortable with the idea that the world is flat.

      Delete
    8. People in this thread do not know what structural means -- it does not mean an ISLM system of equations, because those are not policy-invariant equations. Structural means microfounded.

      And microfoundations are indispensable. What is the macro demand for consumption -- it is the sum of the micro demands. You can argue with how the microfoundation is done, but it is logically nonsensical to argue against microfoundations themselves.

      And for the guy a couple of posts above here -- when your concerns start making sense we'll start listening to them.

      Delete
    9. Post Keynesians came up with one smart idea, that debt matters, just like New Classicals came up with one smart idea, that we should use intertemporal models.

      That's really all you should incorporate from the radical fringes of the discipline into a decent macro model. If you disregard the very basics, i.e what Mill, Wicksell, Keynes and Friedman wrote, you are bound to end up with a crappy model that not coincidentally does not match the facts.

      Delete
    10. The basics of modern economics are in Samuelson, Arrow-Debreu, Solow, Cass-Koopmans, Townsend, Myerson, Hurwicz, to name a few. What's a "new classical" anyway?

      Delete
    11. "What's a "new classical" anyway?"

      It is your school of thought. Looks like you are even ignorant about what you are.

      Delete
    12. Well, since you know so much, and like to label "schools of thought," as if we're all slaves to higher powers, why don't you tell me exactly what a "new classical" thinks and does? Feel free to go on at length, and be specific.

      Delete
    13. There seems to be a lot of anonymous's on here on all sides of the debate. My understanding is that the debate is two fold. One is do you need micro-foundations when you look at a social system. This is not a debate that is restricted to economics. For example international relations asks this question - do you look at individuals when you look at the relationship between states? If you do not it means you think they are homogenous and cannot make a difference. But sometimes they arguably do. But you could argue that WWII would have happened with or without Hitler.

      The second questions is what micro-foundations. Economics was created in Britain at a time when a new commercial classes - the bourgeoisie and the middle classes were challenging existing power arrangements. They were fighting for the opening of new markets, at home and abroad, and a new allocation of property rights. The basis of their ideology was limited resources, unlimited resources, and their assumption was that this was basic to human nature, and could be extended to a view of a social system as a whole. They used geometric and algebraic gadgets - budget lines and indifference curves and other such stuff to make their case. There was a lot of other stuff going on at this time in Victorian England, in fact, Darwinian theories emerging at the same time and in the same place were being extended across the natural sciences. In many ways this was a reaction against Enlightenment theories which were more dominant across the Continent. Classical and Neo-classical economics is really a subdivision of neo-liberalism in political science (which many economists today seem to not properly understand, I would say even Paul Krugman judging by his use of the term). A key argument about neo-liberalism is that free and open markets and democracy are linked and the former is crucial to a stable and democratised international political order. This is not necessarily wrong, but highly contentious. There are good reasons (and some bad ones) why the US adopted neo-classical approaches as part of its (I would say on the whole unquestionably successful) agenda for the multilateral political and economic system after 1945. But it is important not to ignore political economy and say that something is scientific when it is not. This is what people mean when the micro-foundations of economics needs need ontological and epistemological foundation. -ie why is it that it has taken the approach it has? OK this is a high school level discussion of the issues, but I am glad that Roger Farmer in his latest blog sees the importance of understanding the history of economic thought and reading the classics - unlike a few others such as Sargent.

      Delete
    14. ^ This post is completely silly and misses the point entirely. Here's a helpful tip -- Austrians are not economists.

      Delete
  7. Excellent point. MM is persuasive (to me) about targeting NGDPL, but I'd like to see it modeled and not just hand waved.

    ReplyDelete
  8. Excellent point. MM is persuasive (to me) about targeting NGDPL, but I'd like to see it modeled and not just hand waved.

    ReplyDelete
  9. Although you are skeptical about the importance of sticky prices, international macro economists are more likely to think they are important because real exchange rates seem to track nominal exchange rates.

    ReplyDelete
    Replies
    1. "...real exchange rates seem to track nominal exchange rates."

      That's an important empirical fact. That doesn't mean that the explanation for this fact is, for example, Calvo pricing.

      Delete
    2. Calvo pricing is a shortcut because doing the Blanchard/Kiyotaki kind of thing is pretty hard in a DSGE.

      Now I agree that if one does microfoundations they should be done properly and sticky prices/wages are most likely not the key inefficiency that leads to underemployment equilibria. Financial market imperfections are most likely the key inefficiency.

      There are a plethora of market failures that aggregate into the giant macro market failure but as we cannot model all of them at the same time we do just one in DSGE.
      I view it as a robustness test, if one micro market failure suffices to create underemployment equilibria the notion is sound (not that we need theory, just opening your eyes during a nasty recessions suffcies).

      Delete
  10. "All models are wrong..."

    Oh please, enough with that fallacious nonsense! Get this into your head: if your model doesn't fit reality well, not even approximately, it is *garbage*. Period. It cannot be "useful" in anyway, not help "clarify"/"think through"/"gain insight" or whatever you economists keep telling yourselves you are doing; if it doesn't have *any* homomorphic connection with reality. And the only way to substantiate that is, yes, verifying how well it fits existing data out-of-sample. Sorry pal but this is how science works: "social" science is no exception.

    Now fine, dismiss this comment with derision and condescension like you always do. After all, this is your life's work and it must really suck to dedicate so many years of your life to an inconsequential game of smoke and mirrors. But that doesn't make it any less illegitimate, or any less wasteful.

    And seriously, for Pete's sake: stop using that quote. Try learn something, for once:

    http://andrewgelman.com/wp-content/uploads/2012/03/tarpey.pdf

    ReplyDelete
    Replies
    1. I urge you to go away and sulk in silence.

      Delete