If science tells you you can’t predict something, is it no longer science?

by James Fearon on September 2, 2013 · 13 comments

in Science

The NYT ran an op-ed last week by philosophers of Science Alex Rosenberg and Tyler Curtain that said that (1) economics is not a capital-S Science because it has no “record of improvement in predictive success” and (2) therefore the next Fed chair should be someone who understands that economics is not a capital-S Science and instead has “wisdom” and “a feeling for the economy.”  (Interestingly, experience was not mentioned, which you’d think would be more important the more the job is a matter of craft, as R and C say, than of book learning, which is mainly what Ben Bernanke had.  R and C applaud Bernanke for his craft.)

Harvard Econ theorist Eric Maskin wrote a letter in reply and then NYT did a “Room for Debate” exchange between Maskin and other readers.  Maskin made the point that explanation can be valuable and scientific independent of whether it leads to sharp predictions in some particular domain.  But most of the readers the Times printed were not buying this at all—for them, sufficient evidence that Economics is not a capital-S Science is that it can’t predict stock market movements and crashes reliably.  Rosenberg and Curtain also seem to have this in mind as the key example.

But what you could reasonably call a scientific thought experiment is enough to show that you can’t have a Scientific Theory that reliably predicts stock market movements and crashes:  If such a theory existed that predicted a giant market collapse on date T, the collapse wouldn’t happen on date T.  Ditto for forecasting market movements from publicly available data.  In the long run, or “in equilibrium” in the sense of the developed formal models of this kind of thing, market movements and crashes should be random or unpredictable based on public information.  Yes, that’s an “idealization,” and an idealization that R and C obliquely criticize at the end of their op-ed.  But it’s an idealization that provides an important and valid insight into the kind of system an asset market is.

The reply by reader David Berman raises this issue indirectly, but he concludes that it means that Economics can’t be a “proper” capital-S Science (“if only we could [predict market crashes] without the prediction’s entirely changing the behavior of the markets! That’s the other critical difference between economics and meteorology, or physics, or any of the disciplines we properly call scientific”).  I read what R and C are saying the same way, although it’s less spelled out there.  But can that be right?  I would have thought the argument should be that scientific inquiry has clarified the reasons for why a theory that predicts specific market movements or crashes is difficult, or impossible in the long run, in contrast to theories of simple physical systems.  In other words it’s not so much a case of inappropriate hubris of Economics in trying to be like Physics, but of a scientific effort producing a better and deeper understanding of why this sort of system (an asset market, here) is different from a physical system that doesn’t have agents that condition what they do on expectations about what others will do.

Of course, not all markets—or economic, political, or social interactions—have this specific dynamic that asset markets have, so the difficulties in predicting the stock market from public information do not imply that you can’t develop theory that makes reasonably good predictions in other areas (eg., basic supply and demand analysis for prices and quantities).  R and C were making a reasonable point about choosing Fed chairs.  Sure, I’d like someone who is wise and not rigidly attached to some particular mathematical idealization.  But the broader line of argument seems wrong.


Comments on this entry are closed.

Previous post:

Next post: