Like it or not, DSGE models are here to stay. I made the following argument in the First Edition of The Macroeconomics of Self-Fulfilling Prophecies in 1993.
In this book I take a point of view that is becoming less controversial but is by no means universally accepted. I will argue that the future of macroeconomics is as a branch of applied general equilibrium theory.Believe it or not; twenty one years ago, that was a controversial statement. I argued then that the problem with DSGE models is not the assumption that the economy is in equilibrium. The problem with DSGE models is the implication of some of these models that the equilibrium is optimal. Since then, I have consistently argued that the way forward is to reformulate Keynesian ideas with modern mathematics; that is what the DSGE agenda is all about.
For example, I have constructed a DSGE model where 25% unemployment is an equilibrium. That model does not use, or need, the concept of sticky wages or sticky prices to explain why high unemployment persists; persistence is implicit in the notion of an equilibrium. Unlike classical or new-Keynesian DSGE models, my work explains why 25% unemployment is a very bad thing from the point of view of society.
The use of multiple equilibrium models to understand Keynesian economics is part of a research agenda that began at the University of Pennsylvania in the 1980s. That agenda has accelerated recently and the use of multiple equilibrium models to understand data has become mainstream. Jim Bullard uses the work of Benhabib-Schmitt-Grohé and Uribe to understand the liquidity trap. Narayana Kocherlakota applies my work on incomplete factor markets to understand unemployment and the top economics journals are routinely publishing research on the importance of animal spirits as a driver of economic activity. We have moved past the IS-LM model as the true guardian of Keynesian thought.
So what’s wrong with middle brow theorizing? The IS-LM model made the best use of techniques available in 1936 when Hicks introduced it as a way of making logical sense of the General Theory. We’ve moved on since then and we now have tools for bringing dynamics into the picture and for understanding how expectations interact with realized outcomes in ways that respect the methods that have proven successful in so many other branches of economics.
The IS-LM model says nothing about inflation. It says nothing about the passage of time and it does not account for the inability of firms and workers to engage in apparently mutually beneficial trades. We now have the tools to put all of those pieces together and, despite Paul’s claims to the contrary, the result is not a simple regurgitation of 1950s macroeconomics. If a smart theorist like Krugman struggles with formalizing his intuition the problem is not with the mathematics; the problem is with the intuition.
Mathematical formalism is an indispensable tool that has been with us since the late nineteenth century. There was a major leap forward in 1947 with Samuelson’s Foundations of Economic Analysis and a further methodological surge in 1989, when Stokey-Lucas released Recursive Methods in Economic Dynamics. With the publication of Stokey-Lucas, the bar for becoming a practitioner of economics became significantly higher than it was when Adam Smith wrote The Wealth of Nations.
Some in the blogging community hearken for the days when an economist could slap together a verbal argument and publish the result in the Quarterly Journal of Economics. Paul Krugman for example, wants his…
ad hockery back — not as an exclusive approach, but as a permissible one. And that’s not a small thing, given the almost total exclusion of middlebrow modeling from academic macro for the past three decades.The use of ‘ad hockery’ has not been acceptable in economics for quite a while. And for good reason. As Marshall argued in his 1906 letter to Bowley, mathematics is a language; nothing more. I drew attention to Marshall’s instructions in an earlier post but they are worth repeating;
- Use mathematics as shorthand language, rather than as an engine of inquiry.
- Keep to them till you have done.
- Translate into English.
- Then illustrate by examples that are important in real life.
- Burn the mathematics.
- If you can’t succeed in 4, burn 3. This I do often
The research community ignored points (3) and (4). Paul would have us ignore points (1) and (2) and that is at least as bad.
The IS-LM model is static. It cannot explain inflation and it has no well developed theory of expectations. DSGE models are a huge methodological advance that gives us logical tools to integrate all of these pieces. There is simply no substitute for the use of mathematics to make sure that an argument hangs together.
______________________________________
1. Don’t get me wrong; mathematics for mathematics sake does play a role in economics journals. Sometimes, the real world examples come later. A good example of this process is Lloyd Shapley’s work on stable matches that was used by Al Roth to create markets for kidney exchanges. But the best and most enduring economics papers use the mathematics to explain real world phenomena.
I don't think anyone, least of all Paul Krugman, is suggesting that mathematical formalism is not useful. But mathematical formalism requires abstraction, and the issue is what and how much to abstract. Simplification makes models easier to understand, use, and translate into English, and it makes it possible to add other features to a model that would have made it intractable had not the simplification been made. Comparative statics, determinism, partial equilibrium, and ad hoc assumptions about behavior are among the ways of simplifying a model; obviously they have costs that have to be weighed against the benefits in any particular case, but how is it that they have become unacceptable forms of simplification?
ReplyDeleteKrugman is surely not suggesting "a regurgitation of 1950s macroeconomics." He's suggesting that we build on the foundation provided by late 1970s macroeconomics (as indeed he points out that international economics has largely done, out of a necessity of simplification). You may not think that the expectations-augmented vertical-in-the-long-run Phillips curve is the best way to model inflation, but it does a pretty good job of plugging the hole in the IS-LM model. (I think in the light or recent empirical research, Krugman would argue that "vertical in the long run" is often not the best assumption: a long-run Phillips curve that is downward-sloping at low inflation rates would be an example of how one might build on late-70s macroeconomics.)
It would be great if we could have DSGE models as a tool but not as a prerequisite for being considered serious theory. Unfortunately, that doesn't seem to be an option today. If the price of having DSGE models available is that macroeconomics must renounce involvement with any other type of modeling, I suggest it is a price that has not been worth paying.
Speaking as a decision theorist, I think the DSGE approach to expectations has been a predictable failure. As Andy says, we already had the crucial step with the expectations-augmented Phillips curve (which only needs adaptive expectations). What happened in DSGE was that macro theorists latched on to the simplest version of time-separable discounted expected utility theory at precisely the time when decision theorists were abandoning it. There has been some work on looking at more realistic theories (eg Hansen and Sargent drawing on multiple priors) but EU remains as the starting point for most of the DSGE stuff I see. And, in my view, a useful theory of expectations needs to take account of the fact that people can't possibly consider every contingency that might take place. At some point enough surprises pile up that they have to dump their existing model and start with a new one, which is impossible in DSGE settings
ReplyDeleteJohn
DeleteYour view of DSGE models is, in my view, too narrow. Not all DSGE models impose rational expectations; the work of Evans and Honkapohja is a good example to the contrary.
I agree with you completely on the role of incomplete choice sets as a fruitful avenue of research in macroeconomics. I disagree with you on the usefulness of the expectations augmented Phillips curve. That, in my view, was a huge mistake.
My definition of DSGE includes the temporary equilibrium models that Hicks pushed in Value and Capital and that seems to be perfectly consistent with your view of inconsistent plans.
Andy
ReplyDeleteI am not suggesting that every published economic paper must be a DSGE model. What I AM claiming is a verbal argument must hang together in a logically coherent way. The use of mathematics ensures that that is the case.
Macroeconomics is about the economy as a whole (hence the insistence on general equilibrium); the questions we ask are typically about probabilistic outcomes (hence the insistence on stochastic models) and they all involve the passage of time; hence the dynamic element. You wouldn't use a screwdriver to hammer in a nail. You shouldn't use a static partial equilibrium model to explain the effects of monetary policy on inflation and unemployment.
As for your claim that the expectations augmented Phillips curve "does a pretty good job of pugging the hole in the IS-LM mode". I disagree: it does an awful job. And thats where macroeconomics took the wrong path.
Roger,
ReplyDeleteYour view of DSGE models is more expansive than that of most of us, and I suspect of many of those who insist on DSGE models. Most insist on assuming rational expectations, which is simply empirically false, and thus to me looks like the ultimate in ad hoc assumptions. I think many not include learning models a la Evans and Honkapohja or Hicks-type temporary equilibrium models as true DSGE models because they would not be viewed as GE models.
Roger,
ReplyDeleteI am going to expand a bit on my comment above. You follow in the Azariadix-Cass tradition of multiple equilibria, which are different in real terms from each other, driven by different sets of sefl-fulfilling prophetic expectation/levels of animal spirits/whatever. Obviously there are all sorts of issues your approach faces, most notoriously the matter of how these exogenously determined levels of animal spirits get selected and then coordinated on by everybody. Once that coordination happens, these then can become rational expectations general equilibria, or in the deterministic models of Azariadis and Cass, perfect foresight models, again, just as with multiple equilibria in game theory, facing the problem of focal points or how it is that people fix on and coordinate on one of the equilibrium paths.
With Evans and Honkapohja, they are looking at a different set of kinds of equilibria, Now, I stand to be corrected if I am wrong on this, and I may be, but the last time I looked at their stuff their multiple equilibria were of the sort that Sargent and Lucas were aware of from the early 70s, essentially multiple sets of rationally expected price bubble paths, but all sharing a single real path. The problem became how to make all those bubble paths, each of them essentially self-fulfilling prophetic equilibria, "go away" so that the one true Chicago equilibrium would emerge. So, in this regard the learning models of Evans and Honkapohja, a literature also followed and studied by Sargent, was really a study of the stability of these equilibria, each of which was in its own sense a perfect foresight path equilibrium. Their big story was to find instability among these other paths that were not the One True one. Of course what Sargent found was that this convergence to the One True Rational Expectations equilibrium could be asymptotic so that in the real world what one observes is a learning process in which people are using some sort of adaptive expectations that is converging on ratex.
Have I misrepresented the state of things?
Barkley Rosser
Barkley
DeleteYou are not misrepresenting standard approaches. My work is a little different. I have not just one "True" equilibrium; there is a continuum of them, selected by beliefs. And my more recent work is different from the Penn indeterminacy models of the 1980s. Back then it was all about multiple convergent paths. My recent work is about multiple steady states.
That makes a huge difference. I see Evans-Honkapohja learning rules as fundamentals; they are another way of thinking about my 'belief function'. And in a stationary environment; the learning rule will select the steady state that the economy converges to. Typically, it will be path dependent.
I am happy using DSGE models because, in my environments, the equilibria of the models can capture behaviors that Chicago style models cannot.
I'm someone that defends equilibrium theory from its skeptics, but I'm also someone that argues against economist's tendency to blindly use it. And as I've mentioned before, I'm also someone that appreciates your specific contributions to the field.
ReplyDeleteCoordination requires signals and, as I like to put it; signals require realizations. That is to say that in order to coordinate on an equilibrium, real economic agents need to see the consequences of deviating and the equilibrium concept that you use'd better reflect that. I like the "self-confirming" equilibrium concept in particular because agents don't know where in the tree they are, but they've observed where they can't be.
Most equilibrium concepts require, absolutely require, common knowledge of which equilibrium is the correct one. That includes Bayes Nash, which is the concept in DSGE. That's ludicrous. That's my problem with equilibrium as economists use it. Not to say that you are personally at fault, obviously... You're clearly more on the side of the angels than not--I think, in particular, that your belief functions make the particular equilibrium concept that you use similar to the self-confirming equilibrium concept... maybe not in a completely rigorous sense, but close enough for intuition.
Ultimately, the issue is that your boosterism of DSGE and equilibrium macro does more harm than good. The mainstream is DSGE with Bayes Nash equilibrium, and that is the side of the discussion which is strengthened by your advocacy. I know it violates academic norms, but I'd love to hear you say, correctly I might add, that the solution concept used in most papers is a cartoonish caricature of the real world that presumes that everyone knows, probabilistically, how the whole rest of the world is going to behave.
As Krugman might argue as well, too much false-rigor in the form of difficult-to-solve-and-understand models just serves to hide the fact that the microfoundations that macroeconomists are employing are a little ridiculous.
bseconomist
DeleteI'm not sure how to respond to this, although I'm certainly happy to hear that I'm with the heavenly host.
There is immense turmoil right now amongst active researchers in the DSGE community and we are in a phase where lots of new approaches are being tried.
I would guess that its easier now to publish a paper that drops the rational expectations assumption than at any time in the last thirty years. As for the 'cartoonish caricature'; of course that's true. A theory IS a cartoon, a map, a sketch, a simplification: Put in your own adjective.
As for the common knowledge assumption; I've never been a fan of it. Evolutionary game theory is a better approach. Actors use rules of thumb and population proportions evolve to make the outcomes look rational.
I think we're talking past one another...
DeleteLetting myself ruminate a bit over your response, I think I know how to bridge the divide, so here it goes. My problem with what you're saying is not about your economics; it's about your methodology and philosophy of science.
Taken in isolation, your line of research is helpful tweaking the dominant paradigm in various ways... and of course there are others doing the same. The problem is that you dismiss other approaches. What Krugman and others are saying is that "ad hoc-ery" is a compliment, not a substitute for the "more rigorous" approach that you and others are using. The point I was trying to get across--which I think I didn't communicate well--is that there are always more tweaks to do if your approach is to be both rigorous and comprehensive. I used the example of common knowledge because your work with belief functions attacks that particular problem. The point is that tweaking this sort of model can have profound effects on the model's output with the implication being that unless you have ALL the microfoundations truly correct, you really know nothing. Saying that all models are caricatures is true, but an obvious evasion; the so-called rigorous approach is all or nothing because you can't know a priori which pieces of the model will be necessary to get the right answer. The fact that you will never have all the pieces right strengthens my case, not yours.
Now there are no good examples in economics about how ad hoc modelling is a compliment for this sort of work, but there's a great example from physics. Planck discovered quantum mechanics because he was trying to understand blackbody radiation. What he found was that he had to make an obvious BS tweak to his model to make it work: he had to assume that electrons behaved like little oscillators. It worked and gave birth to the most successful theory in the history of science. Planck's model didn't become the standard model of particle physics, but the success of the model helped others to understand the right questions to ask and that's what got physics back on track.
The thing is, ultimately, your so-called rigorous approach is more ad hoc in a sense than the old IS-LM... not because IS-LM dotted all the i's and crossed the t's. We both know it doesn't. It does, however, help to clear our thinking and if we use it--just like Planck's successors used his work--we'll be a little better at asking the right questions.
That's what I was trying to get at in my rambling, imperfect, way.
I think perhaps you've been reading a little too much Feyerabend .
DeleteAs for "tweaking the dominant paradigm". Perhaps. I guess that what's a tweak and what's a leap is in the eye of the beholder. If you don't like my approach; that's fine. Follow your own.
No. I am definitely not a epistemological anarchist, definitely not anything goes. Yet, at the same time, I find it incredible that macroeconomists, of all people, could fall for the fallacy of "putting all your eggs in one basket". I mean, it is in relation to macro that I, personally, first encountered the rules governing optimal portfolio management. The weaknesses of some approaches are covered by the strengths of some others... they can be compliments, just as two assets that anti-correlate compliment one another. This is no more anything goes than portfolio management when done correctly.
DeleteAd hoc-ery and rigor as defined by the dominant paradigm compliment each other because the weaknesses and strengths of each point in opposite directions. That make sense?
BTW, I don't mean "tweak" as an insult to your work, which I've already made clear is some of my favorite out there. I mean it in the sense that you accept the dominant paradigm as given and then build around that paradigm. I absolutely do not mean it in the sense that your contribution is "small".
Deletebseconomist
DeleteI agree that there is room for alternative approaches.... but that doesn't allow for woolly thinking.
By the way -- I rather like Feyerabend. He makes a point that I think you would agree with; that any set of rules that we try to write down to codify the 'right approach' to scientific method, has been violated in the past by some important leap forward. Your example from quantum mechanics fits this exactly.
As a simple time series econometrician i say that Macro without #Econometrics is as irrelevant as Lucas without #Sargent or Sims. No fit, no glory! Central Banks outside the U.S. are learning this fast!
ReplyDelete