Monday, April 11, 2011

De divinationibus orbium oeconomiae


















Except for showing off my knowledge of Latin, I have nothing to add to the following four items.

Rochelle Edge and Refet Gurkaynak on the failure of Dynamic Stochastic General Equilibrium models to help us predict the economy:
A good forecast should have a zero intercept and unit slope as well as a high R-squared. Table 1 shows the efficiency tests for DSGE model forecasts of inflation at different maturities and demonstrates clearly that the forecasts are very poor. R-squareds at all horizons are essentially zero, implying no forecasting ability.
Larry Summers on the future prospects of these Dynamic Stochastic General Equilibrium models:
Summers says Ptolomy model outperformed Copernicun model for 50 years after discovered. Same for IS-LM vs. DSGE... 
Robert Waldmann responds:
Is it true that Ptolomaic models gave better predictions than Copernican models until 50 years after the publication of "De revolutionibus orbium coelestium" in 1543 ?

I have recently noticed that this claim is sometimes made by economists. The appeal is simple. The fact that decades of effort by macroeconomists have not yielded models which give better predictions than an IS(with accelerator) LM adaptive expectations augmented Phillips curve model is frustrating. If, indeed, the same was true of the Copernican revolution, we can believe that we have made great progress...

A serious problem with the claim of fact is that the Copernican model was not improved in the 50 years which followed 1543. In fact, it wasn't improved at all. Rather it was replaced by Kepler's model (which yields excellent predictions)...

I am actually interested in whether non economists (other than Van Jacobson) make the claim that Ptolomaic predictions were better than Copernican predictions for 50 years. I suspect not. The belief seems to follow from the conviction that we have achieved a scientific revolution combined with the fact that our theories have had no empirical success.
And finally, Brad DeLong on the state of graduate economics education:
Rather, the educational problems [in economics] appear to be concentrated at the graduate level...we appear to have produced a professoriate--trained in graduate school--that appears remarkably ill-equipped to survive in the wild...

I suppose that I am still astonished at the failure of the financial crisis and the Great Recession to bring about a sea-change in the teaching of graduate macro. I expected people to say: we need to train our stunts to know--we need to learn--what Reinhart and Rogoff know. There is no point in turning out students who know the models of Prescott who do not know the models of Say, Mill, Bagehot, Wicksell, Fisher, Hicks, Metzler, Friedman, Tobin--Keynes.
In my introductory graduate macro class I learned "Real Business Cycle" models and Ricardian Equivalence, but nothing with money or demand shocks. In my macro field class I learned New Keynesian models. Having been a physics major in undergrad (and thus never having taken an econ class before grad school), I had to learn IS-LM by reading Wikipedia.

Thus marches onward the progress of science...

Update: Krugman and Thoma weigh in.

Update 2: More from Waldmann on Ptolemaic vs. Copernican models. It turns out that Ptolemaic models in their original, unmodified form were very useful for forecasting planetary motion, even over 1000 years after Ptolemy. Never knew that! Waldmann presents this as "Score one for Ptolemy over DSGE", and I sort of agree, though I don't think the comparison is 100% apt. We don't have the telescopes to see the economy the way we see the Sun and planets, and humans adapt to policy changes, so economic models will likely never have the forecasting performance that astronomical models do. But one would hope that someday we'd get an R-squared bigger than zero. And in the meantime, one would think we'd be a little bit more humble about our (lack of) explanatory power regarding the macroeconomy.

3 comments:

  1. Robert Waldmann8:31 AM

    thanks for the link

    ReplyDelete
  2. Anonymous8:58 AM

    i thought the benchmark for predictions was the (theory-free) BVAR?

    ReplyDelete
  3. Actually I think Brad's post could be improved by a simple edit. I think it should read

    "There is no point in turning out students who know the models of Prescott." period full stop.

    ReplyDelete