Wednesday, August 07, 2013

What Killed Theory

Noah Smith notes that the status of theory within economics has declined dramatically.  He gives the credit to psychologists.

Krugman has thoughts.  He finds two problems. First modern theory has no implications.  It turns out that the implications of old theory came from the simplifying assumptions made to achieve closed form solutions and that, with new mathematical tricks, it is possible to make a consistent model (with fully rational agents) which does anything one wants it to do.  This should have been obvious to everyone to begin with (in the 80s it certainly seemed obvious to me).

He writes "After a while, the new approaches came to seem tooliberating; by the early 90s the joke was that a smart graduate student could devise a model to justify any policy."  

The joke predates the early 90s.  Some time before 1989 in a chinese restaurant at around 1050 Mass avenue Robert Barro said, and I quote, "isn't there an existence theorem somewhere that there is a second best argument for anything"

Second Krugman notes that modern macro theory, in particular, has great difficulty avoiding plainly false implications.  I think these two cases are really similar and commented.

In macro anything is possible too.  One bit of macro theory which never became popular is sunspot macro, in which fluctuations occur for no good reason.  In contemporary macro assumptions are made so that equilibrium is determinate.   I think the cases of macro and trade are more similar than you imagine.

I would place Lucasian macro about like Samuelsonian trade theory.  Models with solutions with implications which are grossly false. Macro was liberated by imperfect competition and spillovers too (I'm thinking mostly Jess Benhabib and Roger Farmer but really a lot of stuff which hardly anyone ever read).  It's just that you are one of that very vast majority of people who never found liberated macro the slightest bit interesting.

Another thing, what about numerical methods ?  Macro was totally taken over by computer simulations.  This liberated it (so that anything could happen) but also ruined the fun. When computers were new and scary, simulation based macro was scary and high status.  When everyone can do it, setting up a model and simulating just doesn't demonstrate brains as effectively as finding one of the two or three special cases with closed form solutions and then presenting them.  Also simulating unrealistic models is really pointless.  People end up staring at the comuter output and trying to think up stories which explain what went on in the computer.  If one is reduced to that, one might as well look at real data.  Models which can't be solved don't clarify thought.  Since they also don't fit the data, they are really truely madly useless.


Luis Enrique said...

I think - I hope - you're overdoing the nihilism here. For example, the statement models which can't be solved don't clarify thought looks obviously wrong - you just need a numerically solved model in which what's going on can be seen sufficiently clearly.

But I would be interested to hear what you've got to say on a sort-of-related question.

Do you know what happened to the coordination literature? The sort of stuff done by Russell Cooper. From afar, coordination looks to me like an obviously important idea (it's a very old idea too) but the research seems to have petered out in the eighties. If you know why, I'd love to hear.

reason said...

"Models which can't be solved don't clarify thought."

I don't think this is in any way true. A model may not be solvable, but it none the less restricts the solution space. It allows you to specify ways in which in might be disproved. How is this not clarifying thought?

Robert said...

@Luis I should have Made clear what economists mean by "models". It includes the statement "models are false by definition. One can (and we do) test the implicati one of models, but rejection of the testable implicazioni has no effect. I would restate your argument as hypotheses whose implications can be' approximately determinerà numerically, but not as closed form soluzione are useful. As in, for ex ample quantum electrodynamics with approximately solutions which fit the data to 10 decimal placet.

But economic models aren't like that. They aren't testable hypotheses because we are Already sure they are false. Note I said the models don't fit the data. They aren't the sort of things which are useful even if you can't figure out how they work. This is true of all existing macro models not of all theory with only approximately solutions.

@reason I considered the list Benhabib Farmer and Cooper for liberated macro. I don't know why it petered out. My guess is models which imply we can't forecast well(and we sure can't) do not interest policy makes Who want forecasts (in which they don't believe and which they ignore). The works orse macro models can have indeterminate equilibria and sunspots can have real effects for some parametro avulse. Therefore it is assumed that parametro don't fall in that range. Trying to understand economies which fluttuate unpredictably, economists rule out models in which the economay fluvpctuates unpredictably. I really don't know why.

reason said...

Russel Cooper is a name from my past. He was mentioned as I guy who was a bit like me (worked away alone on stuff nobody understood or appreciated). Maybe like me, he didn't have a lasting impact because he wasn't an effective heirachy climber and didn't build coalitions. Economics really is such a political subject (and I don't just mean the subject matter).

Luis Enrique said...

Hi reason,

well Russell Cooper still seems to be going strong (his CV is still full of AER, ReStud, etc. see here


but the emphasis on coordination seems to have fallen away.

I was wondering whether coordination hasn't become part of the computable agent based modelling movement, which is still only on the fringes of the mainstream (although starting to appear in top rank journals)