Monday, March 10, 2014

Macro muddles


Noah Smith has a new post which displays his great flexibility as a blogger. He's a fairly stern critic of modern macroeconomics, usually. This post is a more or less conscious effort to do mental gymnastics by suspending relevant sense data and seeing just how far it is possible, if one tries really hard, to offer a coherent defense of current practices in mainstream macroeconomics (read: DSGE modelling). One big positive: there was a financial crisis, and DSGE modellers deserve great credit as they have now noticed the importance of the financial system and made strides in including it (in some form at least) in their models. Another: macroeconomists, apparently, are trying to make their models simpler so they can be more useful to policy makers. I say apparently.

Because that's not what I heard at a recent conference from policy makers themselves. My latest Bloomberg column explores a little of what I heard from some of them about what they would like to see more of from macroeconomists.  More realistic finance and more simplicity were certainly on their list, alongside a lot more humility, more acceptance of uncertainty, and more models based on realistic human behavior. They said they haven't seen much of this yet:

... are the top journals in macroeconomics publishing useful work? Judging from a recent conference that included regulators from the U.S. Federal Reserve, the Bank of England, the International Monetary Fund and the Organization for Economic Cooperation and Development, the answer is not encouraging: Six years after a financial crisis that exposed fundamental flaws in the dominant economic theories and models, the profession has made little to no progress in correcting itself.

The policy makers said that top economics journals still aren't publishing research or providing tools that they can apply in their work. The dominant culture favors mathematically complex, intellectually stimulating theories over simpler, more useful ones. "We're in for a long siege,” as one of them put it, “during which useful economics that really works will remain out of the top theoretical journals.” [Read more]

That last quote really struck me. Useful work will stay out of the top journals. How then can they be considered top journals? What am I missing?

At that conference, I gave a short presentation during a panel session. I (cautiously) presented my impression that one reason economists don't want to move away from their models based on rationality and equilibrium is that, in doing so, they will inevitably have to give up some of their favorite theorems and hence be unable to make statements about the possible welfare implications of policies. Taking steps toward realism will invalidate the mathematical basis of their thinking. I see that as a good thing; many economists I believe find it downright frightening.

That is my impression as an outsider, but am pleased to see that economist Peter Dorman believes something similar, a point he made in commenting on Noah Smith's post (h/t Lars Syll):

“….macroeconomists are definitely thinking about heterogeneity.” Come on, you must surely see that one can have both heterogeneity and representative agents. If there are 300 million agents in an economy and you model them as two or three decision-makers, on balance you are doing a lot more homogenizing than heterogenizing. This matters because just about everything we know about complex systems tells us that the density of interaction effects is central. An economy of you, me and a few other people simply isn’t going to have the same dynamics as an economy of millions of interacting agents. This is true even if agent-based modeling turns out to be unproductive. It’s enough to know that the model people are using is systematically giving bad advice. Microfounding macro is a choice, and if the there aren’t any good microfoundations at hand, you don’t have to do it.

“….there’s no clear alternative [to rational expectations].” The previous paragraph applies here as well. If the only microfoundations you can find are empirically disconfirmed, regularly and broadly, then you may just have to postpone this microfounding business until you can come up with better models. Beyond that, I think the core problem is that the models are structured to permit the solution of equilibrium conditions, and that this imposes a restrictive framework for thinking about rationality, optimization. If the point were to model adjustment, we could use a much looser but more empirically defensible conception of rationality. Of course, that would also mean severing economic analysis from welfarism: we’d have to give up trying to answer questions like“what’s the welfare cost of this situation compared to the optimum?” In the end, the attachment of economists, micro and macro alike, to equilibrium models with rational agents is that they want to be able make definitive judgments about what society should do. I prefer Keynes’ dentists: they don’t tell you whether you have an optimal dental structure, but they can help you get the structure you tell them you want.

“….[macro] looks like a vigorous, energetic field full of excited young true believers and respected older figures who are still blazing new trails.” The more accurate criticism of macro is not that it is simply an ideological smokescreen or an unthinking herd, but that it operates on a tilted playing field. It is openly acknowledged that the “leading” (i.e. career-determining) journals have engaged in tendentious selection practices for the past generation. Lots of shoddy research (which in my book includes calibration exercises promoted as “testing” theories) has gotten the star treatment, alongside a stream of genuinely significant macro work. Ideologically loaded assumptions, such as those typically used in public choice, are dropped in without any justification. Not all macro is bad! But the problem is that (a) the bad stuff of a certain ideological orientation gets an extra push that the good stuff doesn’t get, and (b) there isn’t a clear process by which the bad stuff is weeded out over time as its badness becomes evident. We refight the same damned macro battles year after year.

Again, I think economists like to tell stories of a certain kind, stories that fit into a certain framework they find familiar, a framework that links up to all the nice (?) welfare theorems they learned as students. The only problem is that these theorems only apply to a fictitious world that is not at all like our world. So we have the top journals filled with useless theory. Wonderful.

6 comments:

  1. Most accurate Nifty tips for today
    CALL: BUY BANK NIFTY 12000 CALL ABOVE 360.00 TG- 375.00/400.00/440.00 SL- 330.00 (CMP- 358.00)

    Free intraday tips on mobile

    ReplyDelete
  2. The models can be either simpler, or more realistic, but hardly both. And I don't understand the criticism that top journals do not publish applied work - that's not their purpose. Goal of top journals is to publish leading academic research, which is naturally more technical, abstract and geared toward "intelectual stimulation" than immediate application. Using your analogy, it's not like your average pediatrician routinely reads leading immunology journals either. But, of course, policy makers employ dozens (at Fed or IMF, more like hundreds) of macroeconomists. Surely translating and adapting academic research for policy-making is supposed to be their main job, no?

    As for the claim that journals are closed to certain type of work - sure, there is some truth to it. Economics, like any other science, is conservative, and there is natural tendency to maintain current framework and language. But what exactly is supposed to be the alternative? Marxist, Postkeynesian, Austrian, ABMs, econophysics? What people mean when they say it is of course that journals should pay more attention to THEIR particular research - maybe so, but then that can be hardly counted as objective, dispassionate criticism.

    What Noah's post shows is that macroeconomics is not static. Even within the old framework, there's lot of interesting research going on. Are let's say agent-based simulations better (and in what respect?) in modeling financial frictions than recent DSGE models? We won't know until somebody evaluates and compares empirically both, and such effort would be much more useful than just rehashing same criticisms over and over.

    ReplyDelete
    Replies
    1. Yes, I agree. Rehashing criticisms is not helpful. I think what we need more or, and what the people I met wanted to see more of, is a greater diversity of approaches in macroeconomics. I don't think ABM is THE solution to everything, but it is obviously a promising area and deserves to be explored. I think there is considerable resistance to this idea, however. Are the top journals embracing this as a new technique that really deserves attention? I think they're instead quite dismissive. This is very much NOT the case in other areas of social science such as epidemiology, where people are open to any new ideas that might be useful.

      Anyway, I absolutely agree -- encourage diversity and encourage real competition between modelling methods based on tough comparisons to data. That has to be the way forward. Treat all methods equally, the DGSEs, the ABMs, the XYZs: do they get things right at the aggregate level, also at the intermediate and micro levels? Do the agents in the models fit the data for real agent behavior? Do they interact with one another as data shows agents do in the real world? Etc.

      Delete
    2. Ivan, I'd also be interested to know what you think of the paper by Assenza et al. (Cars Hommes group) I mentioned in a post (http://physicsoffinance.blogspot.fr/2013/12/macroeconomics-illusion-of-learning.html) a while back (the paper is http://ideas.repec.org/p/dgr/uvatin/20130016.html). It takes a laboratory experiment approach to exploring how expectations evolve and interact. This is another different way of testing ideas, clearly promising, but some economists I've encountered have been quite uninterested. Their reaction: "I don't think this is relevant because people, in my experience, have rational expectations." Even though one of the mains points of this paper, backed by empirical evidence, is that people, in a macroeconomic context, do not have rational expectations. To be fair, my sample of reactions is extremely limited. Maybe this is unrepresentative of the profession as a whole.

      Delete
    3. Looks like a nice paper. From a quick reading, their results actually seem to be somewhat in line with rational expectations - at the end of experiment, economy has more or less converged to REE if central bank reacts strongly to inflation, like a standard NK model would predict.

      Results about learning dynamics, switching strategies, etc. sound interesting. But I'm a bit skeptical about experimental economics in general - can we really extrapolate results from a bunch of undergrads clicking on a computer for a few dozen euros to actual economies?

      Delete
    4. Yes, I think the most interesting run was the case 3b (if I recall) where the REE was achieved in the very long run. I think it was like 40 quarters in the experiment. A natural question is whether such long convergence is relevant to any real economy where the nature of the dynamics probably doesn't remain stable over such a period; hence people have to form expectations about a moving target.

      I'm much less skeptical of the experimental approach. Maybe it's because I think the real economic world is made up of people not much sharper than undergrads who are clicking on computers with millions or billions (often of other peopIes' money) at stake; I doubt they act much differently, but certainly think much more needs to be done on this. I'd love to see the experiment redone with say 100,000 people online.

      Also, I think it is really important to remember that, even in this experiment, the experimenters worked hard to give the REE outcome a fighting chance. All the participants had to form their expectations without knowing anything about the expectations of others. In the real world, of course, this is probably THE most important dynamic in expectations formation -- looking around to see what other people are expecting. This is likely to lead to much more variable dynamics, I suspect, as it makes herding possible. But we won't know until experiments are done that include this.

      Delete