Wednesday, February 19, 2014
The usual metodological rant
I managed to resist commenting over at Krugman's blog but not here I think Krugman didn't go far enough. I think the heuristic explanation of why the model gives the result it does is often the useful product of the whole effort. One use of models is to clarify thought. The clearer thought is the heuristic explanation of what drove the model. Even if the details of the model aren't true, that informal story can be true. I add that I think that this is the only way in which economic theory has ever been useful. Obviously abstruse theories in the natural sciences have been useful aside from the intuitive explanations including some which remain incomprehensible. But they are used because they yield astonishing but accurate predictions and are ruthlessly tested (they are hypotheses or theories not models). I think it just so happens that, so far, economic theory doesn't include anything like that. I think many economists try to convince themselves to fall for a very elementary logical error. The fact that it is not true that models which are rejected by the data are necessarily useless doesn't mean that they are useful. I wasn't an economist in 1982, but by 1985 the argument that models were false by definition and could still be useful wasn't controversial. Furthermore, it has been an absolutely standard argument made by economists at least for well over a century. I think Williamson set up a straw 1982 economist. I am often very irritated by economists' use of made up intellectual history. If it matters enough to mention, it matters enough to get right. Assertions about what people wrote should be supported by citations. Also I'm trying to get to a clear statement of a sense I have about a catch 22 use of hypothesis testing. The point is that the argument that rejection of a model doesn't tell us anything interesting because it could still be useful is not followed by any discussion of anything which might tell us that it is useless. You note the argument that it doesn't matter if the models are not useful for forecasting, because they are for something else. I think an effort to see if the world looks roughly like the model is not going to be taken seriously at all, because it involves statistics whose distribution under the null is not known. I think that economic theory defends itself by arguing that hypothesis tests are irrelevant and also that anything which isn't a hypothesis test is invalid econometrics. I guess I have two questions for straw orthodox macroeconomist . I will make them narrow. 1) do you think it is conceivable that the Euler equation for optimal intertemporal consumption choices is no good -- not useful -- something else should be used in macroeconomics ? 2) If the answer to question 1 is yes, what would convince you that this conceivable possibility is the case ? I don't think that he will argue that we know a priori that good models must contain such an Euler equation. I don't think anything could possible convince him that a model without such an Euler equation is better than a model with such an Euler equation. I think that he has, and incorrectly denies that he has, absolute a priori faith.