Romoney raised an interesting point in response to the previous post:

But aren't economic theories at least theoretically falsifiable? The problem is that they are so very hard to falsify.
My answer is that by Popper's criteria, neoclassical economic theory is not falsifiable. I disagree with Popper fundamentally, and my memory is somewhat hazy, so let me apologize ahead of time for what is surely a crude bastardization of Popper's theory of falsifiability (at least early Popper).

There's an idea in the philosophy of science that comes out of logic called the Duhem-Quine problem. It basically says that certain scientific hypotheses cannot be disproved because in order to test them empirically you need to add extra auxiliary assumptions, and if the test fails you can simply reject the auxiliary assumptions rather than the original hypothesis.

Popper was well aware of this logical difficulty and argued that a discipline could only be considered a science insofar as it structured its tests in a manner that all hypotheses, not only auxiliary assumptions, can be falsified.


A cursory glance at economics reveals that there are certain sacred cows, termed the "hard core", that cannot be disproved. Take general equilibrium (DSGE) models, which forms the basis for a large amount of macro work. In DSGE models, a set of "hard core" assumptions in consumer theory allow hundreds of millions of people to be aggregated into a single "representative consumer."

These assumptions are extremely restrictive. Since the late seventies, behavioral economists have found in the lab, in the field, and with people from all walks of life that these assumptions are consistently violated. People simply do not behave in the manner assumed by neoclassical theory.

There are a plethora of responses by neoclassical economists: these tests have been performed in a lab so they don't apply in the real world; they may have been applied in the real world, but only to a few hundred people etc. These responses are normally followed by the claim that, through the magic of aggregation, inconsistencies on the individual level are miraculously smoothed out.

This argument might make sense if DSGE models actually worked, but they don't. Neoclassical economists like to say that their models are true because they "fit" a set of data that has been fed into them. However, this criteria of "fitness" hardly constitutes a test of their verity. This is because the coefficients in these models are simply adjusted (the technical term is "calibrated") until the best fit is reached. So these models fit by definition!

A better test of a model is whether, once calibrated for one set of data, it works as well with a different set of data: if we feed a new set of data (for which we already know the outcomes) into the model, to what degree do the predictions of the model match the actual outcomes. Unsurprisingly most models fail miserably at this point.

The true test, of course, is whether a model can predict outcomes in the real world. Unfortunately, as we have seen over the past year, they've seriously dropped to ball on this one.

Thus not only have neo-classical macroeconomists known for the past twenty-five years that their assumptions are wrong on the micro level, but their models don't work at the macro level either. Rather than questions these assumptions, because it would involve completely refashioning macroeconomics, they have, as predicted by Quine, simply rejected the auxiliary hypotheses particular to each distinct model. At this point there isn't any evidence left that could get them to change their minds; the "hard core" assumptions of macro are unfalsifiable.

If you've managed to make it this far in my post, kudos to you. I'd like to reiterate that I disagree fundamentally with Popper, but I do think that neoclassical macro is in a precarious state. The only thing that has kept it alive for so long is the absence of a competing research program. There's something called agent based modeling that could be promising, but it has been marginalized up until now by the neoclassics who control all the major journals. Given the myriad shortcomings of neoclassical macro, it seems worthwhile to at least give these guys a shot. If some major breakthrough were to come out of agent based modeling, it could reinvent macroeconomics on a far stronger microfoundation.

5 comments:

  1. Yo' momma on 8 February 2009 at 15:41

    I ain't gonna disagree over-much wi' you, but you're skipping th' 'stimation aspect o' DSGE.

    Drag your arse into a numerical methods course and you'll see the data estimate the parameters for you, and wi' mutha f***** sexy-assed bayesian estimatin' too.

    Want bitty!

     
  2. Yo' momma on 8 February 2009 at 15:42

    Word.

     
  3. Gimpei on 8 February 2009 at 16:19

    But they don't really estimate jack. If DSGE models were actually successful in terms of their forecasts, economists would be making billions.

    Also, tell me with a straight face that there aren't parameters that are set to certain values just because they work.

     
  4. Anonymous on 9 February 2009 at 10:15

    We can predict the economy with much better precision than we can the weather: ie. I can tell you roughly what the economy in China will look in 6 months far more accurately than I could the weather -- even taking into account seasonal variations! So there goes meteorology.

     
  5. Gimpei on 9 February 2009 at 11:51

    I don't think your analogy with meteorology works especially well: you can't predict the state of the economy in China in six months with more accuracy than you can predict the weather. In six months time, we have a pretty good idea that the temperature will be in a certain narrow band in China.

    If only business cycles were as easy to predict as seasonal changes. China just posted a 20 million drop in employment! In six months, they could be sitting pretty or tanking for all I know. If you think you know better go out and make a billion.

    Also, I think you've missed my greater point: I disagree wholeheartedly with Popper. I think the idea of falsificationism is insanity. However, I do believe that macroeconomists should be paying a bit more attention to the microevidence coming out of behavioral economics.

    Meteorologists aren't in the same position as macroeconomists since the science of meteorology on the micro-level is pretty sound. Last time I checked, we didn't have 30 years of experiments undermining the laws of thermodynamics!