Weekly Wrap: Special US Employment Edition
By Gregory Gadzinski
Another special edition of Weekly wrap, with a special focus on the US employment data. After the negative surprise we had last month, we’ve just had a positive one today! The BLS reported today from the Establishment survey that total nonfarm payroll employment declined by 11,000; much less than the consensus that was expecting around -125K. Meanwhile, the unemployment rate edged down to 10.0 percent, better than the status quo of 10.2% expected by a survey of economists.
Establishment Survey
First observation, the largest job losses over the month are again coming from the usual suspects: construction and manufacturing. The manufacturing sector shed 41k jobs whilst the construction sector dropped 27k jobs for a total of -69K jobs for the goods producing industry.
Second, the service industry created 58k jobs, which is a very good surprise, given the ISM non-manufacturing report. Indeed, with a NMI employment number at 41.6, one could have expected job losses (fooled again, just like last month).
So, overall, the 11,000 jobs lost were much better than the ISM numbers would have indicated. The figure also stands in sharp contrast with the ADP report and the weekly initial claims we had earlier this week. So, where is the catch?
First, you have the so-called sampling error, or the fact that the sample estimates differ from the “true” population values they represent. From the BLS website, we can read that only an over-the-month employment change of 107,000 is statistically significant. This means that there is about a 90-percent chance that the “true” over-the-month change in November ranges from -118,000 to 96,000! Given this sampling error, one shouldn’t be surprised then about the revisions that we had today: total nonfarm payroll employment for September was revised from -219,000 to -139,000, and the change for October was revised from -190,000 to -111,000. Practically speaking, it is easy to conclude that those employment numbers are too volatile and I’d say as volatile as the weekly claims (I really start thinking that this is completely a waste of time to try to forecast those numbers!). But, that still don’t explain the discrepancy between the ISM and the BLS numbers.
The answer is related to the notion of non sampling error; or if you want the “impossibility” to obtain information for all respondents in the sample. A well known fix to this issue is the now famous birth/death model. However, the flaws of the birth/death model are notable including the fact that the model has consistently overestimated new job creation during the current employment recession. Fully aware of that fact, the Labor Department said last month that it would plan to revise the job figures by subtracting more than 800,000 jobs that it had wrongly estimated were filled by workers. Here is the catch!
Household survey
The BLS also reported that the unemployment rate decreased from 10.2 to 10%. But, please, do not conclude too hastily that the employment numbers have reached a tipping point, announcing the reversal of the trend (unemployment rates decreased from 9.5% in June to 9.4% in July before shooting up again).
The decrease in the unemployment rate was not that unpredictable given the large increase last month (from 9.8 to 10.2%). Digging into the data, the number of jobs actually rose by 8,000 last month according to the non-seasonally adjusted data. But, for some reason, October is the month with the largest seasonal adjustment down in jobs. The increase in the unemployment rate was only due to the removal of jobs created by seasonal factors. Interestingly, in November, the non-seasonally number of jobs also increased (by 44,000), but this time the seasonal adjustment did not push the official figures into negative territory. However, the threshold for a statistically significant change (sampling error) in the household survey is about 400,000 ! And don’t forget the upcoming revised figures!
Conclusion
All in all, we know now that employment data may be very volatile, which explains the inconsistencies between the unemployment rate and the nonfarm payrolls (as a reminder the correlation is only 50%). But, the bottom line is: do not revise your forecast about the future trend. The arrow still points towards the north. Want more evidence?
The Labor Force Participation Rate fell to 65.0% (the percentage of the working age population in the labor force). This is the lowest since the mid-80s. When the job market starts to recover, a lot of those people will reenter the labor force and look for employment. I fear that this will keep the unemployment rate elevated for some time.
As I’m writing, the US markets are figuring out the reality behind those numbers…
December 5th, 2009 at 11:58 am
Hi GG
Great post, thanks a lot – saves a lot of time knowing where to come for this analysis!.
As to the market, it did make an (inconclusive) effort to correct its initial euphoria. But, I suspect this had as much to do with closing carry positions. The EUR/USD flipped out on the jobs number, and managed to break the technical resistance that defined its one way trend. The dollar’s sharp reaction most likely explains the market’s response. Sadly, the jobs numbers remain far to complicated for Mr Market to process in real time!
December 6th, 2009 at 5:30 pm
So, if I understand it correctly, we should not even care about the -11,000 November number, because it is not outside the interval [-96,000, 118,000].
But what about the revisions to the previous two months? Aren’t those relevant? “: total nonfarm payroll employment for September was revised from -219,000 to -139,000, and the change for October was revised from -190,000 to -111,000.” So, we’ve gained 80,000 jobs in each of the previous two months. Isn’t that bullish (for stocks)?
Unless of course those gains are negated when the Labor Department will adjust the numbers “by subtracting more than 800,000 jobs that it had wrongly estimated were filled by workers.” But when is it going to do that? How far back will it revise the numbers? You don’t say in your post.
“Interestingly, in November, the non-seasonally number of jobs also increased (by 44,000), but this time the seasonal adjustment did not push the official figures into the negative territory. ”
You mean 44,000 jobs were created? Does that mean that the seasonal adjustment was -11,000 – 44,000 = -55,000?
At any rate, isn’t a loss of -11,000, or a gain of 44,000, better than the loss of 125,000 that was expected, whether it is significant or not? Is a loss of 11,000 the same as a loss of 399,000 jobs, just because bother are insignificant?
December 6th, 2009 at 8:56 pm
Hi James, I fully agree on the carry trade fears and the subsequent technical correction in stocks due to the rise of the dollar. Indeed, “Jobs numbers are far too complicated for Mr Market to process in real time”, they simply relied on a simple rule: dollar up, stocks down, and applied the rule really quickly thanks to automated trading….
December 6th, 2009 at 10:07 pm
[...] tell us? First of all there was a lot of statistical noise, as Gregory pointed out on Friday in his weekly wrap. Noise, one would think, is not enough to keep stocks afloat after a greater-than-60% run since [...]
December 6th, 2009 at 10:08 pm
Dino, thanks for giving me the opportunity to correct the typo and clarify the post. So first, the interval for the nonfarm payroll is of course [-118,000 96,000] and then not significantly different from zero. This is the sampling error, which applies to all numbers (intial and revised). It reflects simply the error that one could get by working with samples (or surveys here) instead of getting information on the whole US population (which is impossible practically speaking).
Now, the revisions are due to nonsampling error, i.e the fact that the Labor Department released initial numbers (-190,000 in October) without all the information collected from their survey. Once they have processed all the data, they publish the revised numbers (-111,000 in October) the following month. Finally, they revised the data one more time to get the final numbers (it is usually an annual revision). According to the NYT article, it seems like they are going to do so next February, something to watch closely obviously.
The 44,000 increase comes from the Household survey (while the -11,000 are from the Establishment survey, do not confuse them) which gives the unemployment rate. Actually, the seasonally adjusted numbers show an increase of 227,000 of the employment population ( the seasonal adjustment helped in November !), which explain the drop from 10.2 to 10%.
So, bottom line, the data (establishment and household survey data) look better and better even on a statistical basis. But, the adjustment next February is going to be crucial. If we have “only” -800,000 jobs, it would make the first three quarters of 2009 more ugly but would mean no additional loss for Q4. Having said that, I have to emphasize again the fact that more people will reenter the labor force in 2010, which will keep the unemployment rate elevated for some time.
December 7th, 2009 at 8:35 am
[...] came from the dollar (see top chart), which sold-off substantially on Friday’s job news (excellently analysed by MM’s GG). EURUSD tripped past its previous high before stabilising, implying that the one-way short-dollar [...]
January 8th, 2010 at 7:40 pm
[...] already explained the importance of the noise embedded in those data and the unreliability of this first release. Moreover, I also pointed earlier that the Labor [...]
February 5th, 2010 at 8:05 pm
[...] recovery periods) put too little weight on the tough conditions faced by smaller companies. As I pointed before, the expectation was 800,000 additional job losses over 2009. The number came out as 620,000. Good [...]