US: in 31 years of 44, unemployment came in below what forecasters expected
US: unemployment, the survey’s next-year forecast and the outcome
Above the forecast 8 times, about as expected 5. But the big misses go up, in crises: 2020: 3.7% expected, 8.1% came; 2009: 7.4% expected, 9.3% came.
More
The Survey of Professional Forecasters: each quarter since 1968 the Federal Reserve Bank of Philadelphia asks several dozen economists at banks, firms and research centres for their forecasts of growth, unemployment, inflation and the probability of a downturn. Here: their median forecast of US unemployment for the next year.
By year: blue, the unemployment rate the fourth-quarter survey of the year before expected for that year; orange, the outcome (annual average). The shading is the error.
- 2020: survey 2019Q4 forecast → outcome
3.65% → 8.10%
- 2009: survey 2008Q4 forecast → outcome
7.40% → 9.28%
- 2021: survey 2020Q4 forecast → outcome
6.26% → 5.35%
- 1984: survey 1983Q4 forecast → outcome
8.35% → 7.51%
One forecast a year: the median of the previous year’s fourth-quarter survey, 1982–2025. Outcome: annual average unemployment, latest BLS data. “About as expected”: within 0.05 percentage points. Sign test: the chance of such a lean at random is p = 0.0003.
The lean is small: in the years unemployment came in below the forecast, by 0.4 percentage points on average. Forecasters almost never expect a sharp rise in unemployment, and it comes with recessions, so the rare upward misses are large.
A question for this picture
Is a forecast that is almost always a little off one way better than one that misses less often but by more?