Tuesday, July 03, 2007

Economic predictions 

William Polley points out something in a discussion of whether economists are any good at prediction.
[T]he textbook model is not perfect but still useful. In my discussion of the Schwartz op-ed, I conclude that monetary incentives can have nonstandard effects in certain circumstances. We don't completely understand those circumstances and therefore more work is needed. I'm not ashamed to admit that perhaps a little more humility is also needed when considering the possibility of these non-standard effects. But it still remains that the incentive story that drives most economic models is mostly right.
Let me add a couple points. All textbook examples, save those doing general equilibrium (which almost never show up in an op-ed in some newspaper) involve invocation of ceteris paribus, an assumption that all other things that might influence the relationship under study are held constant. This includes time, usually. Indeed, a supply and demand graph is best understood as instantaneous: If you could theoretically draw it exactly right, when you turn your head and look back, it will be wrong because something, somewhere, will have changed.

As Polley states, though, that doesn't make supply and demand useless. The question is whether or not the supply and demand apparatus gets you better predictions of human behavior than some alternative model that explains the allocation of goods and services. If you engage economics as a predictive science, the validity of any model is relative to other models in terms of predictive accuracy. Does the supply and demand model get more answers right than wrong, relative to other models? I think the evidence on that is clear, even if we don't bat 1.000.

Put another way, if you are in a forecasting business you will always make errors. Ifyou don't have a logically consistent model, you will not be able to explain your errors. That's likely to influence your job tenure as a forecaster.