Believe those who are seeking the truth. Doubt those who find it. Andre Gide


Sunday, April 12, 2015

In defense of modern macro theory

The first economist
The 2008 financial crisis was a traumatic event. Like all social trauma, it invoked a variety of emotional responses, including the natural (if unbecoming) human desire to find someone or something to blame. Some of the blame has been directed at segments of the economic profession. It is the nature of some of these criticisms that I'd like to talk about today.

One of the first questions macroeconomists get asked is: How could you possibly not have predicted the crisis? We all remember when the Queen of England asked this (supposedly embarrassing) question. Put on the spot, I might have replied that the same question could have been asked of her ancestor(*) predecessor King Charles I, whose death in 1649 also came about under rather unexpected circumstances. Or, I might have replied that many economists did in fact predict this crisis...along with the many other crises that failed to materialize (recall the old joke about the economist who successfully predicted 10 out of the past 2 recessions).

But seriously, the delivery of precise time-dated forecasts of events is a mug's game. If this is your goal, then you probably can't beat theory-free statistical forecasting techniques. But this is not what economics is about. The goal, instead, is to develop theories that can be used to organize our thinking about various aspects of the way an economy functions. Most of these theories are "partial" in nature, designed to address a specific set of phenomena (there is no "grand unifying theory" so many theories coexist). These theories can also be used to make conditional forecasts: IF a set of circumstances hold, THEN a number of events are likely to follow. The models based on these theories can be used as laboratories to test and measure the effect, and desirability, of alternative hypothetical policy interventions (something not possible with purely statistical forecasting models).

There is a sense in which making predictions is very easy. Here's one for you: Mt. Vesuvius will experience another major eruption on the scale of AD 79, when it buried the city of Pompeii, tragically killing thousands of people (among them, the famous naturalist Pliny the Elder). While volcanologists are getting progressively better at predicting eruptions, it remains very difficult to forecast their size. So when an event like this arrives, it always comes as a bit of a shock. In any case, like I said, making predictions (unconditional forecasts) that will eventually come true is easy. There are thousands of people predicting that the world will end in 2015, 2016, 2016, etc. Some of these prognosticators will one day be proven correct. Those making predictions that fail to come true hide in the shadows for a while, but then re-emerge bolder than ever. I don't blame these soothsayers: there seems to be an insatiable demand for the likes of Nostradamus, and this is clearly a case of demand creating its own supply. In this spirit then, permit me to deliver my own forecast (remember, you heard it here first): there will be another major financial crisis on the scale experienced in 2008.

While we can't predict when the next major crisis will occur (I hope the Queen can forgive us), it is reasonable to expect experts to make good conditional forecasts. IF Vesuvius blows, THEN a lot of people are going to die. This type of conditional forecast should lead policymakers to think of ways in which the potential death toll can be avoided or reduced. Perhaps citizens should be prohibited from inhabiting dangerous areas. At the very least, an emergency evacuation procedure should be put in place. The same is true for financial crises. Perhaps restrictions should be placed on the exchange of some types of financial products. At the very least, an emergency response strategy should be put in place. Actually, there is an emergency response strategy--the Fed's emergency lending facility--which essentially worked according to plan in 2008-09. Now, maybe you don't like various aspects of the Fed's liquidity facility and that's fine (even if it did make a healthy profit for the U.S. taxpayer). But you can't say that economists hadn't predicted the possible need for such a facility. Indeed, the Fed was set up on the premise that financial crises would continue to afflict modern economies (by the way, financial crises were a common part of the economic landscape well before the founding of the Fed in 1913, so think carefully before you accuse the Fed as the source of market instability).

Alright, so much for blaming economists and their less-than-crystal balls (hmm, a part of me says I should have edited that last sentence.) What else? Well, I notice a lot of blame also being heaped on modern macroeconomic theory and the professors of such theory. "What's Wrong With Macro?" the headlines wail (roll eyes here). Things have become so bad that we now see students telling professors how macro should be taught. Next we'll have teenagers telling their parents how to raise children. Well, we already have that of course. But the point is that while parents patiently hear out these protestations (having been young for much longer than the youth in question), they do not generally capitulate to them. I'm sorry, but you're only 16, I love you, and no, you can't have the keys to the car!

And yet, amazingly, we have to read things like this (source):
Wendy Carlin, professor of economics at University College London, who is directing a project at the Institute for New Economics Thinking, a think-tank set up by billionaire financier George Soros, said at a conference last year that students had become “disenchanted” and lecturers “embarrassed” by the way economics is taught.
Lecturers at UCL are "embarrassed" by the way economics is taught? What does this mean? Are they embarrassed about the way they personally teach their economics classes? Then they should be fired for incompetence. Are they embarrassed by the current state of macroeconomic theory? Then they should be fired and sent back to grad school (or the Russian front, if you're a Hogan's Heroes fan).

The dynamic general equilibrium (DGE) approach is the dominant methodology in macro today. I think this is so because of its power to organize thinking in a logically consistent manner, its ability to generate reasonable conditional forecasts, as well as its great flexibility--a property that permits economists of all political persuasions to make use of the apparatus.

For the uninitiated, let me describe in words what the DGE approach entails. First, it provides an explicit description of what motivates and constrains individual actors. This property of the model reflects a belief that individuals are incentivized--in particular, they are likely to respond in more or less predictable ways to changes in the economic environment to protect or further their interests. Second, it provides an explicit description of government policy. While this latter property sounds straightforward, it is in fact a rather delicate and important exercise. To begin, in a dynamic model, a "policy" does not correspond to a given action at a point in time. Rather, it corresponds to a full specification of (possibly state-contingent) actions over time. Moreover, there is no logical way in which to separate (say) "monetary" policy from "fiscal" policy. The policies of different government agencies are inextricably linked through a consolidated government budget constraint (see A Dirty Little Secret).  Thus, any statement concerning (say) the conduct of monetary policy must explicitly (or implicitly) contain a statement stipulating a consistent fiscal policy. The exercise is delicate in the sense that model predictions can depend sensitively on the exact details of how policies are designed and how they interact with each other. The exercise is important because the aforementioned sensitivity is quite likely present in real-world policy environments. Finally, the DGE approach insists that the policies adopted by private and public sector actors are in some sense "consistent" with each other. Notions of consistency are imposed through the use of solution concepts, like competitive equilibrium, Nash equilibrium, search and bargaining equilibrium, etc. Among other things, consistency requires that economic outcomes respect resource feasibility and budget constraints.

Now, what part of the above manifesto do you not like? The idea that people respond to incentives? Fine, go ahead and toss that assumption away. What do you replace it with? People behave like robots? Fine, go ahead and build your theory. What else? Are you going to argue against having to describe the exact nature of government policy? Do you want to do away with consistency requirements, like the respect for resource feasibility. Sure, go ahead. Maybe your theory explains some things a lot better than mine when you dispense with resource constraints. But do you really want to hang your hat on that interpretation of the world? An internally inconsistent theory that happens to be consistent with some properties of the data is not what I would call deep understanding. (Nor is an internally consistent theory inconsistent with the data something to be happy about, but that's the purpose of continued research.)

The point I want to make here is not that the DGE approach is the only way to go. I am not saying this at all. In fact, I personally believe in the coexistence of many different methodologies. The science of economics is not settled, after all. The point I am trying to make is that the DGE approach is not insensible (despite the claims of many critics who, I think, are sometimes driven by non-scientific concerns).

I should make clear too that by "the DGE approach," I do not limit the phrase to New Keynesian DGSE models or RBC models. The approach is much more general. While one might legitimately observe that these latter sets of models largely downplay the role of financial frictions and that practioners should therefore not have relied so heavily on them, it would not be correct to say that DGE theory cannot account for financial crises. If you don't believe me, go read this (free) book by Franklin Allen and Douglas Gale: Understanding Financial Crises. While this book was published in 2007, it reflects a lifetime of work on the part of the authors. And if you take a look at the references, you'll discover a large and lively literature on financial crises well before 2007. In my view, this constitutes evidence that "mainstream" economists were thinking about episodes like 2008-09. If central bank economists were not paying too much attention to that branch of the literature, it is at most an indictment on them and not on the body of tools that were available to address the questions that needed to be answered. (In any case, as I mentioned above, I think the Fed did act according to the way theory generally suggests during the crisis).

Once again (lest I be misunderstood, which I'm afraid seems unavoidable these days) I am not claiming that DGE is the be-all and end-all of macroeconomic theory. There is still a lot we do not know and I think it would be a good thing to draw on the insights offered by alternative approaches. I do not, however, buy into the accusation that there "too much math" in modern theory. Math is just a language. Most people do not understand this language and so they have a natural distrust of arguments written in it. Different languages can be used and abused. But this goes as much, if not more, for the vernacular as it does for specialized languages. Complaining that there is "too much math" in a particular theoretical exposition is like complaining that there is too much Hiragana in a haiku poem. Before criticizing, either learn the language or appeal to reliable translations (in the case of haiku poetry, you would not want to rely solely on translations hostile to Japanese culture...would you?).

As for the teaching of macroeconomics, if the crisis has led more professors to pay more attention to financial market frictions, then this is a welcome development. I also fall in the camp that stresses the desirability of teaching more economic history and placing greater emphasis on matching theory with data. However, it's often very hard, if not impossible, to fit everything into a one-semester course. Invariably, a professor must pick and choose. But while a particular course is necessarily limited in what can be presented, the constraint binds less tightly for a program as a whole. Thus, one could reasonably expect a curriculum to be modified to include more history, history of thought, heterodox approaches, etc. But this is a far cry from calling for the abandonment of DGE theory. Do not blame the tools for how they were (or were not) used.


(*) My colleague Doug Allen points out that Elizabeth II did not descend from Charles I. The Stuart line dies out with Queen Anne, at which point George I is brought over from Germany. EII is a member of the house of Hanover/Windsor, and not a Stuart. Many also think that EII is a descendant of Elizabeth I, but of course she had no children, and ended the Tudor line. 

22 comments:

  1. David, the legitimate complaint isn't that economists didn't predict the crisis, but that the models they were using (explicit formal ones, or implicit mental ones) were badly miscalibrated, significantly underestimating the probability of a crisis.
    Here's some evidence from the US and Eurozone SPF's. Most probably none of the forecasters surveyed was using a DSGE model. The intellectual framework behind most of these forecasts is probably some sort of non-optimising IS/LM model, maybe formalised as a large macro SEM.
    See table 5 here,
    http://www.phil.frb.org/research-and-data/real-time-center/survey-of-professional-forecasters/2008/spfq308.pdf?CFID=71298847&CFTOKEN=7c0687510db831cd-F3677CB7-E8F6-9D5E-4B08654C2BC6870B&jsessionid=8430f6afc26f734e61d5107b13146a794d2b
    And here for the Eurozone
    http://www.ecb.europa.eu/stats/prices/indic/forecast/html/table_3_2008q3.en.html

    ReplyDelete
    Replies
    1. Daniels,

      Perhaps. But my own view is more along the following line. A weatherman forecasts that tomorrow will be sunny, with 95% probability. Tomorrow comes and it rains. When it does, people will say that the weatherman significantly underestimated the probability of rain. It's easy to say that the probability of crisis was underestimated after the fact, isn't it?

      The point is that we already know it's going to rain one day. And so we should have contingency plans in place for when that day comes. The fact that some are caught unprepared has nothing to do with forecasting accuracy.

      And again, after the fact, it seems easy to suggest pre-emptive action, but in real time it is much more difficult. Should the Fed have intervened in the late 90s to prevent the dotcom bubble? The stock market crashed without any financial crisis. What would have happened if the Fed *had* intervened to avert a crisis that would never have materialized?

      Delete
    2. Yes, but unless you're maximising some highly conservative ambiguity averse preferences, you'll want to invest much less in the contingency plan if it's against a 2% probability event than if it's a 10% probability event (maybe 20% if you conditioned on the right information in mid 2008?). And contingency plans are rarely free lunches. My impression is that they often involve putting some brakes on economic activity that may reduce steady state output (e.g limiting borrowing in normal times) in order to reduce the intensity of crises. So having a macro model that better captures the tail of the distribution is important. I guess, one answer to this complaint is that 1) the economy is one of the most complex systems to predict for any given percentile of the distribution, and 2) tail behaviour a priori is even more complicated and an area of further research than behaviour within +/- epsilon of validity of the loglinar approximation of a macro model.
      Anyways, I think my point was in part that proponents of old style keynesian macro models are not exactly in the best position to complain about DSGE models not predicting the crisis.

      Delete
    3. Well, I agree with your concluding point. However, I'm still curious to know how what you suggest might work in practice. Tell me how your preferred policy might have evolved in lat 1990 and in the early 2000s (not foreseeing, of course, the actual path of the economy).

      Delete
  2. "Alright, so much for blaming economists and their less-than-crystal balls"

    The Italians will be an early warning system on this matter: http://www.theguardian.com/world/2012/oct/23/jailing-italian-seismologists-scientific-community

    You've got a lot packed in here, so I'll just add a couple of my pet peeves. First, regarding the attacks that come from outside the profession. It's unacceptable to blame modern macro theory for things that are not the responsibility of macro theory. If the government fucks up by the numbers, don't blame the messenger that said mandating longer UI benefits would result in lengthening unemployment spells (for example). I can point to any number of macro models that say intervening with markets during a recession is counterproductive.

    Second, within the profession. There's a lot of dismissal going on that is politically motivated. If someone wants a policy job, then it's a good idea to write down a model that says government intervention is a good idea. I don't blame the theorist for doing this, but I do blame the government for creating this incentive as it doesn't advance the science. However, competition for journal space is high, which helps get only the best papers in. Of course, it could keep some very innovative papers out, but there's plenty of journals out there for everyone :).

    I'm not convinced the above is a huge problem, though. I follow several of the Fed periodicals, and if any economist would be expected to be biased towards the Fed, it would be the Fed's own economists. However (and not trying to polish the apple here) I see basically no bias amongst Fed economists.

    Finally, there is perhaps a big divide between academic macro that appears in the journals, and the musings that appear on blogs. Hadn't we better define our terms before attacking macro theory? If that sounds like I'm still waiting for an NGDPLT model, that's because I am.

    ReplyDelete
    Replies
    1. Always good to hear from you, Prof J. :)

      By the way, on a theory of NGDP targeting, look here: http://andolfatto.blogspot.com/2013/09/like-good-neighbor-evan-koenig-on-ngdp.html

      Delete
    2. Thanks! This paper looks like a good start. I'll sit with it for a while.

      Delete
    3. David,

      Koenig's paper was interesting, and a good start. I'm still left in the cold on one important question, and that is the mechanism for NGDPLT. Unless the entire federal reserve system is to be re-engineered, monetary policy still has to work through reserve requirements, fed funds rate setting, and open market operations. I'm not convinced that what was needed from the FRBNY is to buy even more fixed income securities than they did. If you can aid my confusion on this issue, I'd be very obliged.

      Delete
    4. Prof J, well, I'm not sure of the exact mechanism people have in mind. But if one accepts that the Fed has some control over the price-level P through whatever tools it presently has, then surely it must have some control over PY, no?

      Delete
    5. David,

      Yes indeed, I buy your argument. My question is how would targeting PY (which is a far cry from NGDP, but let that go for now) instead of P cause banks to lend out their excess reserves.

      Delete
    6. Prof J, a few things to clear up:

      [1] PY = NGDP, where Y is RGDP.

      [2] Targeting P is just as hard as PY in a world of excess reserves.

      [3] Banks do not "lend out" reserves. They create demand deposit liabilities made redeemable for cash. They reserves always stay in the system, unless the Fed drains them through an open market operation.

      Delete
    7. (1) I'm thinking in terms of the MV = PY equation. PY in that case is spending on all goods & services, not just final goods & services. From an accounting perspective, sure PY=NGDP. From a money demand perspective, I'm not so sure.

      (2) Indeed.

      (3) I guess I'm confused as to why there is so much 'excess reserves' now, but not prior to 2008.

      Thanks, David.

      Delete
    8. Never mind on (3). I found a nice paper from FRBNY that addresses my confusion. If anyone else is confused: http://www.newyorkfed.org/research/staff_reports/sr380.pdf

      Delete
  3. David, you ask, “Do you want to do away with consistency requirements, like the respect for resource feasibility”?

    No, but in your spirit of pluralism, I would like model builders to occasionally drop the premise that market participants share the same expectations regarding the future course of prices, etc., and that all their plans are mutually consistent (or something like this).

    I realize they're "only models," but it seems that you're apt to miss something essential when you ignore (or abstract from) the fact that there are "bulls" and "bears" in every market, that there are Austrians, New Market Monetarists, New Keynesians, RBC types, etc., who advise organizations, public and private, that have massive resources to deploy.
    Wouldn't the results of a change in Fed policy depend on the distribution of these Macro Views across the organizations the decisions of which will determine the policy's result? (Maybe you could call this the “Lucas Critique 2”).

    I’m pretty sure these assumptions don’t rule out the use of mathematics. The small group of economists who explored “disequilibrium macro” – F. Hahn, F. M. Fisher, J.P. Benassy, E. Malinvaud, et al – were/are very good at math. It’s too bad this route from Arrow-Debru has so few travelers.

    ReplyDelete
    Replies
    1. Greg,

      In fact, economists experiment with alternative solution (consistency) concepts all the time. For an example of heterogeneous preferences, see the work of John Geanakoplos of Yale: http://cowles.econ.yale.edu/~gean/crisis/index.htm

      My own boss, Jim Bullard, has worked extensively on models that drop the rational expectations hypothesis. I could go on and on.

      And I don't like the "disequilibrium" economics idea -- these are just "equilibrium" models in the sense they employ solution concepts to close the model (different solution concepts than are normally used, but solution concepts nevertheless).

      And finally, we have to ask the question of *why* would you want to go down the paths you lament are not being sufficiently explored? Precisely *how* would they have generated (ex ante, of course) superior policy advice?

      Delete
    2. David,

      Thanks for taking the time to reply. John Geanakoplos (and J. Doyne Farmer) published a paper in 2008 entitled "The Virtues and Vices of Equilibrium and the Future of Financial Economics," in which the authors, in addition to describing the virtues of equilibrium models, argue that "the majority of economists have become so conditioned to explain everything in terms of equilibrium that they do not appreciate that there are many circumstances in which this is unlikely to be appropriate.” They go on to argue for the use of agent-based modeling in some of these cases.

      I didn't mention "heterogeneous preferences" nor "rational expectations." Rather, I was suggesting that it might be useful to explore the consequences of heterogeneous expectations, which, after all, seem to characterize the world in which we actually live.

      I don't know whether models that assume different expectations and inconsistent plans would generate "superior policy advice." But I think it's fair to say that "modern macro," as you describe it, seems to have generated a lot of conflicting policy advice.

      Delete
    3. The "conflicting policy advice" comes from the uncertainty surrounding conditioning factors and parameters, what forces are to be emphasized, and which are to be assumed away. I guess there really is no "modern macro theory," it is probably better described as a methodology that encompasses a wide range of theories, or hypotheses, many of which offer conflicting policy advice.

      Thanks for the reference to Geanokoplos and Famer, I had not seen that work.

      Delete
  4. I think mist people who did not study economics or just in bachelor do not realize that mathematematics is a language and a way to check internal consistency of a theory, as you say. They do not realize wxhen they criticize the use of dynamic programing (for example), they are actually critizing the hypothesis of rationality or rational expectations.

    Furthermore, I think the analogy with physics some people are using is misleading and make them forget what economics is about: understanding and predicting human behavors. When we speak of markets, GDP, growth etc. I guess lot of people seem to forget that we are speaking of the outcome of intercating agents with endogenous behaviors. If we econ'omists say publicly that there will be a crisis on market X day Y, can we believe that it won't affect the equilibrium? (EMH?). Economics is not weather forecasting.

    ReplyDelete
  5. I have a question: how sensitive are DSGE models to inaccuracies in reported data?

    I ask this because there is a growing underground cash economy in Western nations. When a modern nation with high taxes gets to deflation, the use of cash explodes. We have seen this in Japan and now Europe and now the United States.

    Of course this will mean large amounts of economic activity slip off the radar. The problem will get worse as long as there is deflation and it makes sense to hold lots of cash.

    See my recent post over at Marcus Nunes' place.

    ReplyDelete
    Replies
    1. Ben,

      I think you're suggesting that measured GDP is likely to decline going forward because significant amounts of trade will be moved to the informal sector. If this is true, then we may conclude that we are experiencing a "secular stagnation" when in fact it's just a measurement issue. I'm not sure what the implications are for DGSE models...do you mean, implications for estimating their parameters?

      Delete
  6. Minor quibble: I think you meant "Hiragana" by "Hirigana".

    ReplyDelete