Tuesday, January 15, 2013

Peter Howitt... beyond equilibrium

Mostly on this blog I've been arguing that current economic theory suffers from an obsession with equilibrium concepts, especially in macroeconomics, or in models of financial markets. Most of the physical and natural world is out of equilibrium, driven by forces that are out of balance. Things never happen -- in the oceans or atmosphere, in ecosystems, in the Earth's crust, in the human body -- because of equilibrium balance. Change always comes because of disequilibrium imbalance. If you want to understand the dynamics of almost anything, you need to think outside of equilibrium.

This is actually an obvious point, and in science outside of economics people generally don't even talk about disequilibrium, but about dynamics; it's the same thing. Equilibrium means no dynamics, rest, stasis. It can't teach you about how things change. But we do care very much about how things change in finance and economics, and so we need models exploring economic systems out of equilibrium. Which means models beyond current economics.

The need for disequilibrium economics was actually well accepted back in the 1930s and 40s by economists such as Irving Fisher in the US and Nicholas Kaldor in England. Then in the 1950s, with the Arrow-Debreu results, and later with the whole Rational Expectations hysteria, it seems to have been forgotten. It's curious I think that really good economists, clear thinking people who are trying to address real world issues, often have no choice but to try to understand episodes of dramatic change (bank runs, bubbles, liquidity crises, leverage cycles) by torturing equilibrium models into some form that reflects these things. The famous Diamond-Dybvig model of bank runs is a good example. The model is one with multiple equilibria, one of which is a bank run. This is indeed insightful and useful, essentially showing that some sharp break can occur in the behaviour of the system, and also offering some suggestions about how runs might be avoided with certain kinds of banking contracts. But isn't it at least a little strange to think of a bank run, a dynamic event driven by amplification and contagion of behaviour, as an "equilibrium"?

I'm not alone in thinking that it is a little strange. Indeed, by way of this excellent collection of papers maintained by Leigh Testfatsion, I recently came across an excellent short essay by economist Peter Howitt which makes arguments along similar lines, but in the area of macroeconomics. The whole essay is worth reading, much of it describing in practical terms how, in his view, central banks have in recent decades moved well ahead of macroeconomic theorists in learning how to manage economies, often using tactics with no formal backing in macroeconomic theory. Theory is struggling to keep up, which is probably not surprising. Toward the end, Howitt makes more explicit arguments about the need for disequilibrium in macroeconomics:
The most important task of monetary policy is surely to help avert the worst outcomes of macroeconomic instability – prolonged depression, financial panics and high inflations. And it is here that central banks are most in need of help from modern macroeconomic theory. Central bankers need to understand what are the limits to stability of a modern market economy, under what circumstances is the economy likely to spin out of control without active intervention on the part of the central bank, and what kinds of policies are most useful for restoring macroeconomic stability when financial markets are in disarray.

But it is also here that modern macroeconomic theory has the least to offer. To understand how and when a system might spin out of control we would need first to understand the mechanisms that normally keep it under control. Through what processes does a large complex market economy usually manage to coordinate the activities of millions of independent transactors, none of whom has more than a glimmering of how the overall system works, to such a degree that all but 5% or 6% of them find gainful unemployment, even though this typically requires that the services each transactor performs be compatible with the plans of thousands of others, and even though the system is constantly being disrupted by new technologies and new social arrangements? These are the sorts of questions that one needs to address to offer useful advice to policy makers dealing with systemic instability, because you cannot know what has gone wrong with a system if you do not know how it is supposed to work when things are going well.

Modern macroeconomic theory has turned its back on these questions by embracing the hypothesis of rational expectations. It must be emphasized that rational expectations is not a property of individuals; it is a property of the system as a whole. A rational expectations equilibrium is a fixed point in which the outcomes that people are predicting coincide (in a distributional sense) with the outcomes that are being generated by the system when they are making these predictions. Even blind faith in individual rationality does not guarantee that the system as a whole will find this fixed point, and such faith certainly does not help us to understand what happens when the point is not found. We need to understand something about the systemic mechanisms that help to direct the economy towards a coordinated state and that under normal circumstances help to keep it in the neighborhood of such a state.

Of course the macroeconomic learning literature of Sargent (1999), Evans and Honkapohja (2001) and others goes a long way towards understanding disequilibrium dynamics. But understanding how the system works goes well beyond this. For in order to achieve the kind of coordinated state that general equilibrium analysis presumes, someone has to find the right prices for the myriad of goods and services in the economy, and somehow buyers and sellers have to be matched in all these markets. More generally someone has to create, maintain and operate markets, holding buffer stocks of goods and money to accommodate other transactors’ wishes when supply and demand are not in balance, providing credit to deficit units with good investment prospects, especially those who are maintaining the markets that others depend on for their daily existence, and performing all the other tasks that are needed in order for the machinery of a modern economic system to function.

Needless to say, the functioning of markets is not the subject of modern macroeconomics, which instead focuses on the interaction between a small number of aggregate variables under the assumption that all markets clear somehow, that matching buyers and sellers is never a problem, that markets never disappear because of the failure of the firms that were maintaining them, and (until the recent reaction to the financial crisis) that intertemporal budget constraints are enforced costlessly. By focusing on equilibrium allocations, whether under rational or some other form of expectations, DSGE models ignore the possibility that the economy can somehow spin out of control. In particular, they ignore the unstable dynamics of leverage and deleverage that have devastated so many economies in recent years.

In short, as several commentators have recognized, modern macroeconomics involves a new ‘‘neoclassical synthesis,’’ based on what Clower and I (1998) once called the ‘‘classical stability hypothesis.’’ It is a faith-based system in which a mysterious unspecified and unquestioned mechanism guides the economy without fail to an equilibrium at all points in time no matter what happens. Is there any wonder that such a system is incapable of guiding policy when the actual mechanisms of the economy cease to function properly as credit markets did in 2007 and 2008?
Right on, in my opinion, although I think Peter is perhaps being rather too kind to the macroeconomic learning work, which it seems to me takes a rather narrow and overly restricted perspective on learning, as I've mentioned before. At least it is a small step in the right direction. We need bigger steps, and more people taking them. And perhaps a radical and abrupt defunding of traditional macroeconomic research (theory, not data, of course, and certainly not history) right across the board. The response of most economists to critiques of this kind is to say, well, ok, we can tweak our rational expectations equilibrium models to include some of this stuff. But this isn't nearly enough.

Peter's essay finishes with an argument as to why computational agent based models offer a much more flexible way to explore economic coordination mechanisms in macroeconomics on a far more extensive basis. I cannot see how this approach won't be a huge part of the future of macroeconomics, once the brainwashing of rational expectations and equilibrium finally loses its effect. 

8 comments:

  1. "Equilibrium means no dynamics, rest, stasis."

    No, in economics it doesn't. Lots of similar criticisms of economics are based on semantic differences between fields. The term, as used by economists, might mean slightly different things in different contexts, but the defining feature of equilibrium is that plans of individuals are somehow mutually consistent. In Arrow-Debreu model, this means that markets clear; in Diamond-Dybvig model, it means that people's beliefs about actions of others are correct.

    Equilibrium defined in this sense can be (and is) used to model dynamics. For example, in Arrow-Debreu model the time dimension can be simply incorporated by indexing goods by time - and an outcome (equilibrium) of the model is then a sequence of quantities and prices, sequence which doesn't have to be constant (and in principle could be even chaotic).

    When macroeconomists say that their models can be tweaked, it' because equilibrium framework allows to incorporate a lot of things and modern research goes far beyond simple textbook models (for example, work on leverage cycles by John Geanakoplos you wrote about previously, is based on equilibrium models). That being sad, I think Howitt is right in emphasizing that we don't have a good theory of how such an equilibrium could be attained (although this is a more fundamental issue, not related to rational expectations in particular), and that for some phenomena understanding the disequilibrium dynamics might be crucial.

    Maybe agent-based models could be helpful in this regard. On the other hand, an ABM model is only as good as are its underlying assumptions about agents' behavior, which could be largely arbitrary. Neoclassical economics tries to impose "discipline" by deriving individual behavior as an outcome of a more parsimonious subproblem (utility maximization), and of course there are problems with such reductionism - but you have to replace it with something.

    ReplyDelete
    Replies
    1. "Equilibrium means no dynamics, rest, stasis."

      No. Not in statistical physics as well. Equilibrium means there is not energy flow into system. System can be dynamical and still be in equilibrium.

      Delete
  2. Hi Ivan,

    Yes, criticism accepted! Thanks for pointing out the error. I do know (especially from the nice paper "Virtues and Vices of Equilibrium in Economics" by Farmer and Geanakoplos, cowles.econ.yale.edu/~gean/art/p1274.pdf) that the term equilibrium is used differently in the two fields, and it seems that it is often used in economics in the sense you mention -- a consistent meeting of plans of different agents. One thing that is not clear to me is how authentic (I can't think of a better word) is the manner in which time is introduced in the Arrow-Debreu types models that you mention. Somehow it is still the case in these models that one *assumes* that people come to this state of consistent plans through time; there is no exploration of how this happens, or how it might NOT ever happen, as people face uncertainty, take actions, learn, err and so forth. So the dynamics seems to be a pale version of what you might think of as real world dynamics. Am I wrong about this?

    I do agree also that any ABM is only as good as the assumptions put into it. Obviously. Those assumptions need to be tested for plausibility also, as should many other outputs of the model. I still fail to see how the "discipline" of sticking to utility maximization is a sensible discipline, if we don't have lots of evidence that people in real world situations act this way (or do we?). It seems more like the discipline of trying to tie your shoes with one hand behind your back. It certainly constrains your behaviour, but it doesn't make success in the task (tying shoes or understanding economics) any easier.

    ReplyDelete
    Replies
    1. Authenticity of time - yes, to some extent you're right. The original A-D model assumes that all trades are made at the beginning of time, which is of course silly. One can show that this is equivalent to a more realistic situation ("sequential equilibrium") when each period people trade goods and one-period-ahead claims, provided that a) markets are complete, i.e. there is enough different claims so that uncertainty can be hedged, and b) people correctly anticipate future prices conditional on future realizations of uncertainty (so e.g. they don't know with certainty the price of wheat next year, but they do know what the price will be conditional on the realization of weather next year). There is research on models with incomplete markets , but I think b) is pretty much universally assumed.

      This still leaves some room for learning, because to make their decisions, additionally agents must have beliefs about how likely are different realizations of uncertainty (what's the probability that weather will be good?), and those beliefs could evolve over time, or could be heterogeneous across people (rational expectations then assume that subjective beliefs are equal to objective mathematical probabilities, conditional on informations availible). Or maybe there is some unobserved state variable which changes randomly over time, but people observe only a noisy measure of it, and they must solve a filtering problem to update their beliefs each period.

      So from the point of view of an individual in such economy, I think that time is quite authentic, but still these are some strong assumptions (basically, that he correctly understands the structure of the economy), and of course it doesn't explain how prices, which coordinate all individual decisions, are formed in such an economy.

      Discipline: well, the problem is that we don't have anything close to a theory that would describe human behavior with the same precision as physics describes, say, behavior of fluid in a weather simuation. That means we have to make assumptions and somehow judge their appropriateness. Speaking in terms of utility maximization means that we can argue about assumption on preferences, which many economists find more intuitive and relevant than arguing about behavior directly, which sounds a bit arbitrary (and could lead to vacuous explanations). But anyway, I'm not familiar enough with ABM literature to judge how much this affects their results, so I don't want to overstate my case here.

      Then you could argue that utility maximization is unrealistic, that people don't have cognitive capacity to solve complicated maximization problems, etc. True, but in that case, I'll just plead the Friedman defense and say that what matters is whether they behave as if they maximized utility. After all, the underlying economic idea is rather obvious (people act to achieve more preferred outcomes), and utility functions, maximization, Euler equations, etc. is just a clumsy way to capture this fact in a mathematical model.

      Delete
  3. This comment has been removed by the author.

    ReplyDelete
  4. In my quest to figure out what it was that macroeconomics had been up to that has led them so astray, I found that goldmine maintained by Testfatsion. The answer is not pretty. The complete market hypothesis has no actual basis in reality. It seems to have been formulated - consciously or unconsciously, probably a little of both - to allow the stability it posits to provide for solutions to otherwise unsolvable macroeconomic problems and the equations they're written in.
    Some of the attacks against the hypothesis are absolutely ferocious, such as this one by William Buiter:
    http://www.econ.iastate.edu/tesfatsi/UselessnessOfMacro.WBuiter2009.pdf
    The responses are just as telling. such as this one by Robert Lucas, Jr.:
    http://www.econ.iastate.edu/tesfatsi/LucasReplyToKrugman.EconomistAug2009.pdf
    In that one his basic argument is 1) We get it right most of the time and 2) We never promised you we'd always get it right.

    I spent my career working with entomologists. Those scientists dealt with instability and non-linear systems throughout their professional lives. The ideas they used in their work included ideas about instability, about bifurcations in systems, about feedback. They were always aware of the distinct possibility that strange attractors might be embedded in the state space where spruce budworm, mountain pine beetle, larch casebearer, tussock moth, and other insect populations were being modelled.

    To now find out that none of these ideas have made their way very far into macroeconomics is a shocker. There's something seriously wrong with a discipline that won't use computer models and qualitative analysis to help them feel their way through the complex human systems they think they understand.

    ReplyDelete
  5. I do think that is all just the same. It is definitely have same work principles. If there is a proper management of recourses everything is going to stay stable and work they it is expected. If poor team managements combines with corrupt workers is not going to do any good to people. Finances, for example: Mortgages were approved improperly before and what have we had as a result? -A crush of entire financial market. This is where instant payday loans have appeared and again what we received is a huge personal finance debt. Nice, huh? This is what I think

    ReplyDelete