David Weigel reports:
I just talked to Bill Beach at Heritage’s Center for Data Analysis, who tells me that the team re-ran part of its model for the Path to Prosperity. The unemployment projections, which predicted sub-7 percent unemployment in 2012 and sub-3 percent unemployment in 2020, were too low.
“We adjusted the full employment unemployment variable,” said Beach. “Nothing else changes as a result of that, but the employment number changes.”
The new numbers — 7.89 percent unemployment in 2012, and 4.27 percent unemployment in 2020.
The good news is that they seem to have followed good statistical practice (as outlined in my previous entry on the topic): when you see an implausible forecast, go back and figure out what went wrong with your model.
The bad news is that I don’t believe their new number either! Their forecast of “4.27%” (which I’ll do the favor of rounding to 4.3%, on the theory that it’s a bit much to be forecasting the second decimal place of the unemployment rate ten years ahead of time!) still seems to low.
If you take an (optimistic) upper bound of 7% and put 4.3% in the middle, you get a range of 1.6%. Still too low. Again I understand that the interval might be asymmetric but in any plausible distribution this asymmetry won’t get you much. Given the generally-agreed very low probability of an unemployment rate below 3%, I don’t see 4.3% as a plausible point estimate.
You can see this from the other direction by setting a lower bound on the forecast of 3.0%. If the center is 4.3%, you get an upper bound of 5.6%. That’s a bit optimistic for an upper bound forecast of unemployment, ten years into the future, no? Again, playing with asymmetric distributions might move this up a bit, but not much. No matter how you slice it, if your point forecast is 4.3%, you’re being mighty optimistic.
I’m no macroeconomist and am willing to be convinced on this point–and I certainly agree that 4.3% is an improvement over 2.8%–but a basic crunching of the numbers makes me skeptical. And the implausibility of this number makes me wonder about the rest of the forecast. It would be good to get a sense of what’s inside the model, driving the forecast to be so optimistic.
In any case, I’m looking forward to including this example in the forecasting and calibration chapter of the introductory statistics textbook we’re writing. It will fit in well with Paul Samuelson’s notorious forecast of GNP growth in the USSR:
