Earthquakes, Financial and Geological

As Paul Krugman’s diagnosis of the state of economics bounces around the blogosphere, we are being treated once again to the old debate over whether economics can ever be as scientific as the “real” sciences, the hard ones that have big budgets for lab gear.

If you want to talk about hard sciences, talk about geology. They even have hardness tests, which I remember from my freshman course many long years ago. (I did pretty well in it, but would have done better if I could have ID’ed coal on a midterm exam.) Yet when it comes to a crucial task like predicting earthquakes, geologists are not much better than the pseudo-scientists who populate economics departments. They do even worse, actually, when you consider that their margins of error are hundreds of miles, hundreds of years, and an order of magnitude or two. This makes Alan Greenspan look like Nostradamus.

But of course, that’s not what does or does not qualify geology as a science. Rather, scientists are researchers who subject themselves to rigorous falsification testing that, in the long run, weeds out error. What’s important about the study of earthquakes, for instance, is not its predictive power but that it’s based on an understanding derived from plate tectonics. Yet this was not always the case. That freshman course I mentioned was taught by some of the last holdouts against plate theory; we learned about igneous intrusions and isostatic rebound, but not subduction faults.

You can’t take a course like that any more, because the reward system for geologists gives lots of points for devising critical tests and takes away even more for failure to pass them. Suppose there were a parallel universe in which geologists acted like economists. You might have a popular school of thought that holds that earthquakes are caused by the differential pressure exerted by topography; that’s why there are more quakes in California than Kansas. You could show that the coefficient on topographical variation is significant at a very low p-value in a pooled cross-section. You could calibrate a model showing how much variation is associated with how many earthquakes. Hell, you could generate hundreds of models with different functional forms and control variables and estimators and build whole careers around nuanced discussions of which one “performs” better.

But none of this has anything to do with science. In the universe we actually live in, geologists looked for critical tests: evidence that decisively discriminated between plate tectonics and competing theories.

If you want to tell me that people formulate lifetime spending plans based on the present value of their expected future stream of real after-tax income, and that because of this any dollar of additional government spending is a dollar less of private spending, fine. But don’t calibrate it or show me results “consistent” with it. Think of a critical test, a real-world situation in which we would get a particular result only if this story is correct, and then live or die by the result. Yes, I know this can be difficult in economics, just as it is in geology, but it’s even harder if you don’t try.