A reader calls my attention to this article arguing that the large upside surprise in employment growth reported for January 2022 is due to seasonal adjustment. It takes 10 seconds to find the requisite not-seasonally-adjusted data on FRED, and another 10 seconds to load it into a decent software package as simple as Excel, and another 10 seconds (at most) to type in the command to take a 12 month log difference to see seasonal adjustment issues are not the reason for the big job growth number (there might very well be other reasons, but that ain’t it).
If seasonal adjustment were the issue, as David Goldman* argues, then one would expect the 12 month growth rate in not-seasonally adjusted data to diverge from the 12 month growth rate in seasonally adjusted data. At the aggregate level (nonfarm payroll employment), it doesn’t.
Figure 1: 12 month growth rate in nonfarm payroll employment, seasonally adjusted, FRED series PAYEMS (black), not seasonally adjusted, FRED series PAYNSA (red), calculated as log differences. Source: BLS via FRED, and author’s calculations.
I provide the FRED series so people can do their own calculations without accusing me of manipulating the data and/or hiding the “raw” data, as in this case. (To be explicit, these FRED series are unmodified coming from BLS, so they are “raw” data in the sense that CoRev meant “raw”.) Here is the change in number of jobs rather than growth rates.
Figure 2: 12 month change in nonfarm payroll employment, seasonally adjusted, FRED series PAYEMS (black), not seasonally adjusted, FRED series PAYNSA (red), both in thousands. Source: BLS via FRED, and author’s calculations.
This is not to say at the sectoral level, where samples are smaller, one might not see issues. In manufacturing (less than 10% of total NFP) and leisure and hospitality services (slightly more than 10%), one sees the following
Figure 3: 12 month growth rate in manufacturing employment, seasonally adjusted, FRED series PAYEMS (black), not seasonally adjusted, FRED series PAYNSA (red), calculated as log differences. Source: BLS via FRED, and author’s calculations.
Figure 4: 12 month growth rate in leisure and hospitality services employment, seasonally adjusted, FRED series PAYEMS (black), not seasonally adjusted, FRED series PAYNSA (red), calculated as log differences. Source: BLS via FRED, and author’s calculations.
But these differences wash out at the aggregate level.
So, if Mr. David Goldman has spent the 30 seconds to download the data into Excel, he might’ve found out that his long essay on seasonal adjustment for the aggregate number was for naught. But of course, that was not his purpose; rather it was to cast doubt on the validity of the data itself (and I would be the last one to day there aren’t problems with tabulating the data during a time of pandemic, but I don’t think the one he is pointing to is the right one).
To see how a real economist deals with issues surrounding seasonal adjustment, when one cares about levels and might not have a seasonally unadjusted series to compare against, see Jonathan Wright’s BPEA article (in other words, I do take seasonal adjustment issues seriously — I just want to do it in a serious way).
By the way, about 20 years ago, various conservative commentators argued the establishment series undercounted the amount of employment; the BLS developed a research series which adjusted the household survey data to an NFP concept. I plot the level, and the 12 month growth rates of the official and research series below:
Figure 5: Nonfarm payroll employment series (black), and household survey adjusted to NFP concept (red), both in 000’s, s.a., on log scale. NBER defined recession dates peak-to-trough shaded gray. Source: BLS via FRED, BLS, and NBER.
Figure 6: 12 month growth rate in nonfarm payroll employment series (black), and household survey adjusted to NFP concept (red), both s.a. NBER defined recession dates peak-to-trough shaded gray. Source: BLS via FRED, BLS, and NBER.
The official series on NFP is growing 6.5% y/y, while the research series is growing 6.6% (both calculated as log differences).
For more on data paranoia, see [1], [2], [3], [4].
* David P. Goldman, “American economist, music critic, and author”, BA, Columbia (1973), MA Music Theory, CUNY (see Wikipedia).
” It takes 10 seconds to find the requisite not-seasonally-adjusted data on FRED, and another 10 seconds to load it into a decent software package as simple as Excel, and another 10 seconds (at most) to type in the command to take a 12 month log difference to see seasonal adjustment issues are not the reason for the big job growth number (there might very well be other reasons, but that ain’t it).”
You have to understand that Kelly Anne Conway has restricted Bruce Hall’s computer to do what only she commands he do. Bruce is not allowed to search FRED. Now that is an interesting bio on David P. Goldman. One might suggest Bruce do a little research on the backgrounds of the nut jobs he cites as alleged experts. But then remember – Kelly Anne Conway has total control over Bruce’s computer.
So, if I were an innumerate hack and I looked at the non-seasonally adjusted factory and leisure and hospitality series, I might be inclined to holler about BLS under-counting job growth?
Hang on a minute. Let me dig out my throat lozenges and charge up my megaphone!
And I need to dial up the megaphone to 11 when I yell about the household job “undercount” due to the BLS seasonal adjustment conspiracy.
Bruce Hall is following in the foot steps of Jack Welch!
https://foreignpolicy.com/2012/10/08/jack-welch-owes-us-an-apology/#:~:text=Former%20GE%20Chairman%20and%20CEO%20Jack%20Welch%20was,it%20had%20fallen%20below%208%20percent%20since%202009.
macroduck: They could, although they could then check to see how “big” the difference is (less than 1% over the year for leisure and hospitality, less than 0.07% for manufacturing…), and if they were really thinking about it, might conclude that the difference was inessential.
Can you name even a single topic where Bruce Hall has provided us with a reliable report or analysis? I cannot. Especially when it comes to the issue of COVID-19.
Regarding the nonfarm FRED series, PAYEMS, it is interesting to me that at least since 2020m12 PAYEMS has been linear trend stationary. It seems that PAYEMS has “powered” through the ebbs and flows of Covid for the past year. If this trend stationary relationship remains through February 2022, the change in PAYEMS for February 2022 could be 526k plus or minus 107K assuming one standard error.
You would be hard pressed to argue that “seasonally adjusted” data would have a systematic divergence from the 12 month change data. The 12 month data is by its nature “seasonally adjusted”. The purpose of doing a seasonal adjustment is to allow comparisons from month to month.
Ivan: Yes, but in practice, when we apply estimated seasonal components, to data that are not used to calculate the seasonal components, then one can come up with some mysterious results. In other words, decomposing any series into trend, cycle, and seasonal components involves some judgment.
You would be hard pressed to argue that “seasonally adjusted” data would have a systematic divergence from the 12 month change data. The 12 month data is by its nature “seasonally adjusted”. The purpose of doing a seasonal adjustment is to allow comparisons from month to month.