Economists are pretty bad at predicting the future (how fast will the economy grow next quarter?). They’re also pretty bad at predicting the present (how fast is the economy growing this quarter?). And, it turns out, they’re equally bad at predicting the past (how fast did the economy grow last quarter?)
In January, economists reckoned that fourth-quarter GDP growth would be 3.0%, but in fact the advance figure came in at 3.5%. In February, that number was revised down to 2.2% — it seems that even in January we really didn’t know what had happened between October and December. And then today, the final number came out. The markets were expecting 2.2% again, but in fact the number is 2.5%. Notes Barry Ritholtz:
Markets rallied on the news that 4 – 6 months ago, GDP was bad, but not quite as bad as previously believed. (So much for the market as a forward looking discounter).
Want to get even more confused? Read Kash Mansori explaining that the upward revision is actually a bad thing: a lot of the upward revision can be attributed to the fact that inventories rose at an annual rate of more than $22 billion. So companies were making things, which boosted GDP; they just weren’t selling them.
Still, all these revisions do put me in mind of Goldilocks. The first report was too high, the second report was too low… could it be that 2.5% is just right?
For what it’s worth, we’re to come in at 2.2% (at the final revision – not the advance or preliminary figure!)
My understanding of the prediction literature is that economists are actually good at making predictions when there is no turning point. Predicting trends is not the problem, getting hammered by turning points is. I might be wrong.
You’re exactly right, and that turning point problem is what’s driving our experiment with what we’re calling the “modified-limo” method, because of the wide discrepancy we observed in the 2006-Q4 number.
What we’re finding interesting is that the raw Climbing Limo method does seem to be able to look far enough ahead to capture the turning points.