Thursday, January 7, 2010
Mirowski On The Market Crash
At the recently completed AEA/ASSA meetings in Atlanta I organized a joint HES/AEA session on "Complexity in the History of Thought." Speakers included me, David Colander, John Davis, and Phil Mirowski of the Notre Dame department that is about to be abolished. He spoke provocatively to an overflow audience on "Inherent Vice: Complexity vs. Behavioral Explanations of the Crisis." In this he dismissed views that emphasized "bad behavior by individuals" including irrationality and corruption as causes of the crash and larger breakdown. He argued that it was a systemic problem, doing so by extending his recent idea of markomata, that markets are fundamentally algorithms that evolve as competing systems with increasing levels of hierarchy in the form of futures and derivatives markets. While conventional theory says such developments should improve efficiency and spread risk, they can lead to increased fragility as the rising complexity can lead to a breakdown of computation as in the halting problem in computer science, implying an inability of the system to set prices, which was exactly what happened in the crash. I was the discussant, and while I had some technical complaints, I find this a very intriguing argument, and it generated considerable and very lively discussion in the session.
"He argued that it was a systemic problem, doing so by extending his recent idea of markomata, that markets are fundamentally algorithms that evolve as competing systems with increasing levels of hierarchy in the form of futures and derivatives markets. While conventional theory says such developments should improve efficiency and spread risk, they can lead to increased fragility as the rising complexity can lead to a breakdown of computation as in the halting problem in computer science, implying an inability of the system to set prices, which was exactly what happened in the crash."
ReplyDeleteNot to argue with the idea that an emerging financial superstructure can lead to increasing fragility, but how much is necessary for crashes? Aren't positive feedback loops enough?
Barkley Rosser,
ReplyDeleteIf this subject interests you I recommned this working paper by Markus Brunnermeier and a couple of Computer scientists. The paper integrates derivative markets, asymmetric info ad computational complexity.
Anonymous,
ReplyDeleteThanks for the link.
Min,
I should have mentioned that Mirowski specifically linked his argument to Minsky's (indeed, I should have labeled this "Mirowski on the Minsky Moment"), but claims to move beyond Minsky. There certainly is a large lit on bubbles that basically involves positive feedback loops once one gets speculative purchases driving the market. This is an extension of that and many ways more a model of how the crash comes about, especially a fully systemic widespread collapse.
Doesn't that beg the question concerning the role of greed, moral hazard, etc. in creating the levels of complexity? E.g., in creating such dubious derivatives, in creating and packaging the underlying loans, &c.?
ReplyDeletei might disagree with technicalities. the halting problem basically applies to infinite systems, and real computers are finite, so the issue is more 'NP completeness'.
ReplyDeletealso, i like to take a 'fundamentalist' approach, which can mean i accept General Equilibrium theory, and also take for granted that the SMD is just a part of that (as chaos is for many newtonian systems). So, instability is part of the system, as the GE theory precicts. (Even the austrians at GMU i think have argued that the absence of stable equilibria shows that unregulated markets operate efficiently, just that the 'efficiency frontier' is a chaotic attractor. (won't even the 'end times' have millions of years of cataclysms----Fed policy should try to move in that direction for reasons of virtue).
and again, if one does not really distinguish algorithms from differential equations (an interesting issue) , seeing economies as NP complete problems----meaning they may have a find it hard to get the best price, at walmart or wherever---that can be taken to mean there is imperfect information (a long search) for the perfect deal. which can be seen as a chaotic trajectory. (which is why NP hard problems typically are solved using statistical methods, to get around NP completeness of algorithms and chaos).
in sum it doesnt seem to be trtue to me to say that the economy couldn't set prices----they just werent the prices people like, because price setting is a complex problem on real economies. (also, stuff like greed, prejudice, statistical discrimination, etc. all can make an algorithm long. structural problemns may equally lie in cognition.)
A.J. Sutter,
ReplyDeleteMirowski spent a great deal of time dismissing what he calls "scapegoat" theories, arguing that it is not bad or foolish behavior, but the increasing complexity of the system, with the Minskyan result of Ponzi finance arising as a result. As the discussant I defended Akerlof in particular as seeing a link between the informational problems that are connected with the bad derivatives and so on with the Minsky problems and older Keynesian views.
media,
Technically speaking, yes, although a finite computer can get into an infinite do loop.
Thanks for your explanation. I look forward to reading some write-up by Mirowski of his remarks. I'm a big fan of his work as an historian, but judging by your précis, his analysis in this case sounds very abstracted from reality, or at least (my point in my previous comment) not looking far enough upstream. My wife was an officer at a retail bank in California during years when many subprime loans were being originated. As I understand it from her perspective, as well as from published accounts, there was certainly bad or foolish or greedy behavior in those originations. There was debate in her own institution about whether to offer such loans at all, and many of those in favor were quite explicitly linking this sort of product to their own compensation, for example. Without those loan originations, there wouldn't have been anything on which to build CDOs and related credit-default derivatives that lent complexity to the system (at least, the complexity that precipitated this crisis).
ReplyDelete