« Social safety nets, opportunity cost, and the importance of understanding economics (Part II) | Main | Bush announces nominees for two open Board of Governors positions »
January 26, 2006
Forecasting GDP
4th quarter GDP figures come out in the morning. Talk is that it will be under 3% for the first time in almost 3 years. That might well be. I'd put my chips on the "under 3%", but I'd never go "all in" on advance quarterly GDP. In the latest Econoblog, Kash (of Angry Bear) explains why:
In October 2005 the Congressional Budget Office published "The CBO's Economic Forecasting Record," in which they took at look at the track record of both the CBO and "Blue Chip consensus" forecasts for various macroeconomic variables. Doing a quick check on the accuracy of the Blue Chip consensus forecast for GDP growth shows that, on average, the consensus forecast was off by 1% (in absolute value). Forecast errors were particularly large during large changes in GDP. In other words, economists are particularly bad at identifying turns in the business cycle.
Now, I'm not sure whether this average error of 1% is larger or smaller than I would have expected, compared to an average GDP growth rate of 3.1%. But I am quite sure that it's disappointing to see that a forecast that simply extrapolated last year's GDP growth into next year's GDP growth would have had a slightly smaller average error. In other words, the average of our best macroeconomic models does no better at predicting GDP growth than a person who simply always guesses that next year's GDP growth will be the same as this year's!
To me, that's a pretty bad testament to the state of the economics profession's understanding of the macro economy. Are economists leaving out important information? Is the importance of psychology and other inherently difficult-to-model factors making hay of our economic models? I'm not sure … but add it all up and I find myself regularly disappointed with our ability to make good macroeconomic forecasts.
Kash and James Hamilton (Econbrowser) bat it around a bit, and it's well worth the read--and timely given the upcoming release of the advance GDP data.
Hamilton relates an interesting anecdote:
I spoke recently with the manager of a fund with one of the best forecasting records of anybody in the business, and was very interested in his description of how they worked. At the research stage, they use all variety of sophisticated econometric techniques to look for relations in the data. But when it gets to the point of actually making the investment call, they throw out the econometric estimates of all the coefficients, and replace them with values that make sense from the point of view of an understanding of the basic forces that are operating, values that are hopefully consistent with the econometric estimates, but not identical. They are thus deliberately coming up with a statistical model that does a worse job of fitting the data than something else that is available, but hopefully is more robust about predicting what may come next in a world which, as we've both observed, is constantly changing. That's certainly consistent with my advice for any forecasters -- don't try to do too much with overfitting the data, but settle for a simple model that gets the broad brush correct.
Sound advice. Turning points, when they happen, can look like outliers. Overfitting the data is bound to cause you to miss the turning. There is much that we do not understand.
UPDATE: The folks at Davos aren't doing much better.
Posted by William Polley at January 26, 2006 7:53 PM
Trackback Pings
TrackBack URL for this entry:
http://www.williampolley.com/cgi-bin/mt-tb.cgi/468