CEA Chair Miran asserts that some of the surprise in the employment numbers is due issues of seasonality. Can we see that? I can’t…
Here’s y/y growth rate of the official series and the seasonally adjusted series.
Figure 1: Year-on-year growth rate in seasonally adjusted NFP (blue), and in not seasonally adjusted NFP (tan). Growth rates calculated using log differences. Source: BLS via FRED, and author’s calculations.
The seasonally adjusted series is actually above the y/y not seasonally adjusted series!
What about the level?
Figure 2: Seasonally adjusted NFP (blue), not seasonally adjusted NFP (tan), not seasonally adjusted NFP adjusted by author using X-13 (in logs, 2021M07-2025M07) (green). Source: BLS via FRED, and author’s calculations.
Another reason to think the BLS numbers have not been “massaged”: the July release numbers better track ADP, on private NFP.
Figure 3: Private nonfarm payroll employment, July release (bold black), Jun release (purple), ADP July release (green), all s.a., in logs, 2025M01=0. Source: BLS, ADP via FRED, and author’s calculations.
In any case, there are usually big NFP misses around turning points. Civilian employment — which is not revised month to month — peaked in April.
A decline in response rates probably contributes to the latest big revisions. The first report of any series will be less reliable when response rates fall.
If there is a seasonal adjustment “problem”, it would be nice to know what it is. There is always some possibility that seasonal factors are poorly matched to actual seasonality, but to honestly blame seasonal factors, that mismatch should be larger than normal. Is there an explanation attached to the “bad seasonals” story? I haven’t found one.
There is a longstanding market folk wisdom which holds revisions to be predictive; downward revisions mean future data will be weak, while upward revisions mean future data will be strong. I don’t know whether there is actual evidence to support this idea – looked briefly and couldn’t find any. Even if it was once true, would it still be true when response rates fall? Dunno about that, either. Any stats guys want to chime in?