Friday, March 19, 2010

Why criticize modern macro when you do not follow modern macro?

Economics and in particular macroeconomics have come under assault lately because they supposedly were not able to identify the housing bubble and warn about the dangers in the banking sector. It has thus become fashionable to bash the field. It perfectly fine to review the basic tenets of research, this is part of the scientific process. Still, this needs to be done with a good understanding of the current state of the field.

Richard Holt, Barkley Rosser and David Colander claim that neoclassical economics is now dead and a new era has now begun, which they call "Economics of Complexity." I think we all agree that an economy is a complex object, none the least because you have many different people interacting. Like in other sciences, we use models that are simplified abstractions of the reality in order to understand it. One could argue that one simplifies too much, but often the resulting basic intuition is enough to understand what is going on. Sometimes more complexity is necessary, and can certainly no accuse macroeconomics of shying away from complexity. The days of analytically solvable models, even representative agent models are long gone, there are frictions galore and market imperfections are frequent modeling features. But this is still done within models that are clearly of a mainstream neo-classical tradition.

So why do Holt, Rosser and Colander claim that the future economics "is a vision that sees the economy as so complicated that simple analytical models of the aggregate economy—models that can be specified in a set of analytically solvable equations—are not likely to be helpful in understanding many of the issues that economists want to address. Thus, the Walrasian neo-classical vision of a set of solvable equations capturing the full interrelationships of the economy that can be used for planning and analysis is not going to work"? While I can this in say, international trade and public economics, macroeconomics, which is the most under fire, has in fact made that step long ago.

What do they concretely see coming in economics? "Instead, we have to go into the trenches, and base our analysis on experimental and empirical data. From there we build up, using whatever analytic tools we have available. This is different from the old vision where economists mostly did the opposite of starting at the top and then built down." That is not economics they are talking about, this is statistics. Also, when talking more about theory, they say "combined, these changes can be summarized as a movement in economics from a textbook economics of rationality, selfishness, and equilibrium to a new economics of purposeful behavior, enlightened self-interest, and sustainability. What they are criticizing here is the Max U paradigm, but when you think about it, "purposeful behavior" is rationality with information problem, "enlightened self-interest" is selfishness with a some altruism, and "sustainability" is about multiple equilibria, which are common in heterogenous agent models. There is nothing new here, and such modeling features are routinely used.

Does this mean the "Economics of Complexity" of the future is already happening? Probably. It is just that Holt, Rosser and Colander have not noticed it yet, despite the fact that it has been pursed for to decades already. Consider this gem: "the thought that one could develop a micro foundation of macro without considering the feedback of the macro system on the individual is beyond belief." Did they read any macroeconomics paper published in the last twenty years?

27 comments:

VIlfredo said...

The same could be said about Krugman, with the difference that he has done much more damage.

rosserjb@jmu.edu said...

Economic Logician is barking up the wrong tree. If one googles "Macrofoundations of Microeconomics" one will find a paper written by one of us, David Colander to be precise, that was published in 1994 in the Eastern Economic Journal. Looks like E.L. is the one who has not been reading the literature thoroughly enough in the last 20 years.

E.L. could also check out my 1991 _From Catastrophe to Chaos: A General Theory of Economic Discontinuities_, Kluwer, or, better yet, the 2000 second edition.

BTW, anybody interested in economic logic might find my paper "On the Foundations of Mathematical Economics" of interest, available on my website at http:/cob.jmu.edu/rosserjb

Barkley Rosser

VIlfredo said...

Here is the link to the article: The Macrofoundations of Micro, EEJ 1993. No word about heterogeneous agent models in macro either, which is the point of EL I believe.

Agent Continuum said...

Here's a better alternative *snarf* just kidding, but I suspect some of you remember the episode: Economists, Incentives, Judgment, and the European CVAR Approach to Macroeconometrics.

HET people are supposed to be lagging, but that puts them into a disadvantage re: forecasting or making claims about the future of the profession.

Kansan said...

As much as I admire Colander's work on the Economics profession and especially the status of graduate teaching, I have to admit he is way off the mark here, and that is a huge disappointment to me. A paper he should not have written.

Barkley Rosser said...

Kansan,

How or why is he "way off the mark"? BTW, anybody who goes to my website will find papers dating well back into the 90s, as well as the book I mentioned, that talk about and even do (in some cases) agent-based modeling of macro. E.L. and some of the rest of you are just out to lunch.

Dan in Euroland said...

I am a little puzzled why they just don't write papers in the vein that they are talking about ("complexity"), and stop worrying about what the rest profession is doing.

My 6th grade English teacher imparted some valuable wisdom that these authors seemed to have missed. "Show, don't tell."

If Agent based models are better then build one and compare it to a DSGE model. People will listen if you are getting better results. If their method is better then eventually people will pick up on it.

rosserjb@jmu.edu said...

Dan,

Are you talking about me? Go to my website. I have and do.

Sheesh. Whoever is "Economic Logician" invited me to observe his posting here, and all I can say is that it was full of ignorance, and a large number of the subsequent comments have not been much better.
What kind of a blog is this? Gag!

Kansan said...

Barkley, you seem to be confusing agent-based computing with heterogeneous agent DSGE models. The first have brainless automatons that have been imparted with particular rules by the researcher. The second have optimizing agents who react to their environment. This is what is currently done in macro and you have seemingly grossly overlooked.

rosserjb@jmu.edu said...

Kansan,

I would suggest that you and the rest of the commenters here should get your facts straight before you make too many remarks about who is doing what or should do what. So, I have been involved in agent-based modeling where learning occurs. Good enough for you?

I would suggest that anybody who is interested go check out previews of papers about to come out at Macroeconomic Dynamics, where there is a paper up by me and Mauro Gallegati and Antonio Palestrini, which is the first example (since my 1991 bookd, but better) that models the most common form of speculative bubble crashes, namely those that peak, decline, and then crash, the "period of financial distress" model observed in the vast majorityh of major historical speculative bubbles and initially anaylyzed by my old friend Hyman Minsky, and discussed in his classic book, Manias, Panics, and Crashes, by my old friend, Charles Kindleberger.

OK?

Kansan said...

Barkley,

we are not disputing the fact that there are agent-based models and that they may be useful. What we are criticizing is that your paper paints a picture of macro that is plain wrong. There is a rich literature with heterogeneous agent DSGE models that you completely ignore while writing that macro is only using representative agent models.

Anonymous said...

cat fight!

AfricaForEver said...

Isn't Colander a macroeconomist? Shouldn't he know the literature enough to figure this out?

rosserjb@jmu.edu said...

Have had trouble posting here from my home computer.

Kansan,

You guys are badly misrepresenting things. There have been het agent DSGE models around, but not for "decades" or even from before the Colander EEJ paper. Many, such as Krusill-Smith, are just homog rat agent with a distribution being dragged along. No interactions among agents beyond the GE calculation. I will grant that current models have more bells and whistles.

So, there is an emerging competition in the research shops of the central banks between ACE and DSGE, with neither having done well on the recent events. However, Kansan is wrong that ACE models have "automatons." Many ACE models have learning, such as my forthcoming one in MD.

I would also dispute E.L.'s contention that using empirically research as bases for agent behavior in ACE is "just statistics." This is silliness in the extreme.

I am aware that there is the possibilitiy of a sort of synethesis. An example of this might recent work out of CeNDEF by Hommes and crew. These guys are doing ACE models, but they have taken to having them move off a benchmark of a DSGE model. That is rather interesting, but in the end it is an ACE approach that is going on.

Anonymous said...

I see Colander is an author on this paper as well as on the one that generated a similar outrage in the blog linked by Agent Continuum. Colander was also heading the publication committee of the American Economic Association that decided on launching the American Economic Journals. His misunderstanding of macroeconomics may explain why the editorial board of AEJ-Macro was staffed and why this journal has had such difficulties attracting papers, at least compared to the others.

Kansan said...

Krusell-Smith has heterogeneous agents, but in a form that can be summarized by a moment or two, which allows to use some methods used by representative agent models.

But models without such restrictions to the distribution have existed in the early 90's already. See the work of Ayiagari, Huggett, Rios-Rull, Imhrohoroglu, among others. Some have been published after Colander's 1993 piece, but that does not excuse that you current knowledge about this class of models should be updated.

Barkley Rosser said...

Anon,

The heavies on the pub committee of the AEA were Bob Hall and Peter Diamond, not Colander. The macro journal changed editors, from Blanchard to Davis. I do not think that is the fault of any of those three.

Frankly, I see almost zero citations to any papers in any of the AEJs. My current take on that project is that it is a miserable flop, although that could change.

Kansan,

I do not see much in the way of serious interactions between agents in any of the papers you are talking about, Huggett, Aiyagari, or Krusell-Smith. In all of them, there are distributions of agents, but the aggregation problem shown by Jerison and publicized by Kirman is not addressed, and they are all subject to it. Just garbage. Furthermore, Huggett and Aiyagari are not full GE papers. They are incomplete markets papers. Maybe it is "you knowledge" of these models that needs some fixing.

I would say that there have been some papers in the last few years that are moving the direction of incorporating such elements, with the one I mentioned from last year by (my former coauthor) Cars Hommes and Domenico Massano being a good example. We did not make all the generalizations in the paper that some here think we did, but I am perfectly willing to accept that some of these should probably have been noted, although I would continue to maintain that this one is more of an ACE one that is connected to a DSGE benchmark, which is one way to go.

I note that in E.L.'s original post he scarfs over and simply gets wrong what we (and anyone who knows the lit on this) means by "complexity." It is not just "complicated" or having lots of agents, and it is most certainly not at all about "frictions," although as a practical matter for macro we see the ACE-type models as the obvious complexity competitor with the DSGE ones. The discussion of what complexity is draws mostly on my work, which is readily available.

BTW, I notice that not only is "Econmic Logician" ultimately anonymous, but he or she encourages those commenting here to use phoney monikers. This seems to encourage a particularly nasty and insulting tone that is reminiscent of the nauseating and embarrassing econjobrumors blog. It is all the more annoying when those leveling all kinds of insults make flaming errors in their remarks.

Economic Logician said...

Wow, it had to be on a superb week-end like the last one that a passionate discussion erupted. I missed all the fun.

I see the point that complexity is agent interaction and not heterogeneity. But you need heterogeneity to get interaction. And you get this, in the class of DSGE models, with the myriads of search models out there. Also, one could argue that agents do not need one-on-one interaction to actually feel the impact of others, after all there is a price system that does that. And let us not forget all the models with bargaining.
And you can have general equilibrium under incomplete markets, of course. It is just harder, as you need to specify all markets.

All in all, I still think you grossly misrepresented the macroeconomics literature by pretending it was still only working with representative agent models.

rosserjb@jmu.edu said...

Economic Logician,

Ah, so you were off and not following all this. I note that I am usually accused of providing too many citations in my papers, and that coauthors tend to reign me in. The section of the paper that had you most upset was rather short, and as I have agreed above, probably some of the recent papers could have been mentioned, although I continue to be unimpressed by the ones from the 1990s that some of your folks think are suitable examples.

I would note that there is a definitional matter here as well. Once one starts introducing lots of frictions and incomplete markets and so on, one is arguably not in a strict general equilibrium mode, although one is certainly in a contingent one. There is also the matter that once one abandons ratex for adaptex, one is no longer in a position to claim that one is not subject to the Lucas Critique, which was the supposed original reason for going the DSGE route. As it is, I consider the ratex assumption in particular to be the worst part about the DSGE models, at least the ones that use it.

I can't resist making a more humorous poke at Agent Continuum, presumably a strong defender of the view that such assumptions constitute serious use of het agents in DSGE models. So, a continuum is an uncountable infinity, but such can have no weight, as the Cantor set has a continuum of members, but Lebesgue measure zero, which seems appropriate for "Agent Continuum," :-).

Vilfredo said...

Why would an economy with frictions and incomplete markets, yet balanced markets for all goods, not be general equilibrium? General equilibrium is not necessarily "perfect."

Barkley Rosser said...

Vilfredo,

If you had read carefully what I wrote, you would have noticed that I put modifiers into what I wrote that somehow you seem to have overlooked.

E.L.,

BTW, it is ludicrous to claim that people responding to prices is an example of a macro foundation of micro. This is the rank nonsense.

VIlfredo said...

Barkley,

relax. Debating is part of the scientific process. This is what we are doing here. And, don't you appreciate it that for once it is about research and not some politics?

AlabamaFor You said...

Barkley,

before spreading false rumors on other forums, the comment you claim EL censored is on the Social Security thread. Keep track of where you post...

Barkley Rosser said...

Sorry Alabama, you are full of shit. What is on the social security thread is not what I attempted to post that did not appear. Now maybe there was a technical glitch that is responsible, but given E.L.'s dishonorable record of third rate trolling, I would not put it past him to censor my post here.

As it is, I wonder how it is that you presume to know what has been posted and what has not? And why do all of you follow E.L in a lockjaw lockstep, even referring to yourselves as "we" and repeating his most inane assertions?

Frankly, I suspect that at least some of you, if not the whole lot, are sock puppets of E.L. I offer all of you the opportunity to prove to me that you are not sock puppets by communicating with me privately at rosserjb@jmu.edu your identity, preferably with a link to proof that you are who you say you are. I offer the same I offered E.L. on this matter, complete anonymity.

If I see at least two of you (E.L could fake one) who are real people, I shall apologize here for making the accusation. But the conduct of this blog is so poor and E.L.'s public reputation so low that I will not be surprised if I do not hear from any of you supposedly real people.

Anonymous said...

A 62-year-old economist starting a flame on a blog! Talk about "low public reputation"...

Anonymous said...

I came late to this post. I must say that it has disappointed me. I thought EL knew what he was talking about.

"I see the point that complexity is agent interaction and not heterogeneity."

So, you didn't know that??

"But you need heterogeneity to get interaction."

No, actually you don't.

"And you get this, in the class of DSGE models, with the myriads of search models out there. Also, one could argue that agents do not need one-on-one interaction to actually feel the impact of others, after all there is a price system that does that."

And your point is?

Christos Koulovatianos said...

I find resistance to the overall idea of agent-based modeling as being potentially constructive. One problem that I and others into dynamic games have is the replacement of explicit forward-lookingness of players by rule-of-thumb heuristics in agent-based models.

Think about a city with driverless cars, like the scientific project indicated here:

http://en.wikipedia.org/wiki/Driverless_car

At the moment it is true that such a city of driverless cars driven by robots would result in more accidents than those driven by humans, i.e. human error is lower than error due to poor driving modeling.

Yes, it is a fact that cars in the world of human drivers are driven through human heuristics as opposed to the trillions of calculations made by robots. But does this mean that the project of driverless cars should seek heuristics in order to be more successful?

Moreover, research done to simplify a complex system, always within the spirit of scientific reductionism is potentially more promising, at least in terms of usefulness through policy prescription.

But open-minded we should always be.