From Veneroso Associates’ US Economy October Employment Report, ” Huge Discrepancy Between the Payroll and Household Surveys:
Executive Summary
1. According to BLS, payrolls fell at a 188,000 a month rate over the last three months. But their own household survey says employment fell at a 589,000 a month rate.
2. Why the discrepancy?
3. Chris Manning of the BLS told us last month that payrolls were overestimated in the twelve months ending March by 824,000. The source of this error was the birth/death model. BLS used “plug” numbers for the number of births and deaths. These “plug” numbers were wrong. They led to estimated positive contributions to employment that were too high. Most of the error (675,000 out of a total 824,000 jobs) occurred in the first quarter of this year. The birth/death model was adding significantly to payrolls when all other payrolls were falling. In reality the contribution from net births and deaths was in fact negative.
4. Manning told us that the faulty birth/death model was still being used for the months after March of this year. The implication was that the faulty birth/death model would continue to overstate payrolls and understate the payroll job losses in the months since March.
5. And, in fact, the BLS is doing just that. For the last three months they are assuming net birth/deaths have added 18,000 jobs a week. Last year over the same period they assumed it added 17,000 a week, the year before 18,000 a week, and the year before smack in the middle of the economic boom 18,000 a week.
6. It is obvious what BLS is doing. They are simply plugging in an extrapolated figure with zero adjustment for the most severe labor market contraction in three generations. And, worse yet, they know the birth/death number they are using is pure baloney.
7. NUTS!
8. Therefore, reality probably lies somewhere between the payroll survey monthly rate of job loss of 188,000 and the noisy household survey rate of job loss of almost 589,000. A best guess would be that jobs continue to be lost at a rate of 300,000 a month or more.
Payrolls were down 190,000. A slightly larger decline than the consensus. But prior payrolls were revised to show a lesser decline in August and September combined of 91,000. Payrolls with revisions declined only 99,000.
From a payroll survey perspective employment conditions are improving significantly. Not so from a household survey perspective.
The unemployment rate rose by .4%. I expected a rise, but only because I expected the sharp drop in the labor force in recent months to be partly reversed. In fact the labor force fell further by 31,000. The increase in the unemployment rate came entirely from another huge decline in the household measure of employment of 589,000. This followed declines of 785,000 in September and 292,000 in August. That is an average monthly rate of decline in employment of 589,000. That is as bad as it has been for the entire recession adjusted for population discontinuities.
The household survey of employment is a very noisy series. I was absolutely certain that, after the huge declines of August and September, we would see a much lesser decline in household survey employment in October. I thought that a decline of 200,000-300,000 would still signal serious employment weakness because of the huge declines in the prior two months.
No matter how noisy we think the household survey is, we have to take these household survey employment declines seriously. The three month decline may not be close to 1.8 million; it may be half that. It does not matter. A 300,000 a month rate of employment decline is very serious.
How can there be such a huge divergence between the household survey which now shows almost 600,000 job losses a month and the payroll survey which now shows average job losses of under 200,000 a month? Part of it, of course, is data noise. But part of it must be a continued overestimation of net positive job creation arising from the notorious birth/death model….
Therefore, reality probably lies somewhere between the payroll survey monthly rate of job loss of 188,000 and the noisy household survey rate of job loss of almost 589,000. A best guess would be that jobs continue to be lost at a rate of 300,000 a month or more.
Is this consistent with anything else? Yes. Though the manufacturing ISM showed a huge increased in its employment index, the non-manufacturing ISM showed a significant decrease to a low level. The vast majority of employment is in the non-manufacturing sector.
Also, if the rate of job loss was seriously contracting the work week should be rising. A move to a longer work week is often the first move by employers when labor conditions start to improve. The payroll survey shows a decline in the work week over the last three months and no improvement in the last month.
The latest initial and continuing claims suggest that there is some recent abatement in job losses. But they have probably continued at a significant rate and income destruction probably continues at a rapid pace….
As for the markets, they are so clueless at reading the fundamentals I have no idea how they will react to this data.
17 comments
Comments are closed.
Rosenberg picked up on this too. Says the household survey leads the payroll survey.
Yves, your article needs editing. The last third is garbage (or I had too much home made rakia).
Thanks, weird I had corrected it earlier, a saved version somehow got posted.
I couldn’t find an email for you, so…
Two fine animal pics from some russian site:
http://pixdaus.com/single.php?id=19788
http://pixdaus.com/single.php?id=72578
ouch!
“As for the markets, they are so clueless at reading the fundamentals I have no idea how they will react to this data.”
I don’t think the major market participant are clueless; they’re again back to their old games and hoping that when everyone realizes that this emperor is stark naked, they’ll be able to get out first.
I was doing some unrelated reading on the Fed MBS purchases, when I came across this Bloomberg article dated October 14, 2009. The unrelated part that caught my eye:
“Economists forecast the unemployment rate to peak at 10.1 percent in the first quarter.”
Oops. In the immortal words of Grandpa Simpson, “Toooooooo Laaaaaaaaate.”
The inability of the economics profession to forecast unemployment in the short, medium, or long run would be downright comical if not for the human tragedy involved.
I am firmly of the opinion that, in hidsight, the period from March 2009 to the present (or slightly beyond) will turn out to be the sort of temporary lull in the crisis that was experienced between the fall of Bear and the fall of Lehman. The next leg down is coming sooner than anyone thinks.
Here’s a link to BLS re this topic:
http://www.bls.gov/web/ces_cps_trends.pdf
The first table and Chart 1 are esp interesting. (I assume that both household surveys use the birth-death adjustment and therefore are currently underestimating unemployment.)
My understanding is that the establishment survey unemp # Mr. Veneroso provided was AFTER subtracting the useless birth-death adjustment (86,000 alleged new unmeasured jobs added to the establishment in October). This subtraction may have a bit of complexity to it, so it may not be as simple as one thinks. But doing simple addition, adding 86,000 to 190,000 almost gets you to Mr. Veneroso’s 300 K “average”.
In any case, it’s unclear why he is averaging two different numbers to obtain a “best guess” number. If you mix a cherry pie with an apple pie you have a mess.
Perhaps it also would have been helpful had he done a bit more than simply asserting that the household survey is “noisy”. How volatile is it? (Chart 1 on the BLS link above doesn’t make it look bad.) Census on behalf of BLS surveys about 60,000 households per month. Sounds statistically signif to me.
Can anyone explain why it is necessary to apply a ‘birth/death’ ratio to the data? Why doesn’t the BLS simply count up what it can count and report that?
Yves where does the Obama job stimulus enter into all this? I understand there are some largish flaws in how job creation has been accounted for in that program?
Siggy @ 11:30 is right. At least the major media (especially the bloggers) should just make this (the number, sans B/D) the standard in the current environment. This isn’t really the BLS fault. When things are moving as quickly as they are (despite the FED’s effort to slow things down) it is going to be difficult to get any reliable stats. If the BLS just started ignoring the birth/death input they would be accused of politicizing the number. To the credit of the blogosphere, most have been all over the inadequacy of the B/D adjustment.
Another post on the discrepancy between payroll data and household data can be found at Jesse’s Cafe
Americain.
It has a very nice chart of the birth/death model as well as a humorous cartoon showing Obama in a sombrero snarling
“Jobs, you don’t need no sinking jobs! You need a loan!”
with Gentle Ben (the enforcer) looking on in the background. To the former (the chart) what is apparent is not only the suspect uniformity of data from this model (as discussed here), but the abrupt adjustments in January and July downward in the in the number of employed persons represented by the model, times at which such correctional downturns least affects perceptions. The article also notes some people suggesting the Bush Dynasty (Georges I and II) deliberately underestimated the Census population to “synchronize” household and payroll data; yet another manipulation of the “facts” to profit a more favorable result.
Related to this topic, I read an article yesterday on the unemployment situation (which I can’t unfortunately find again) which details characteristics of unemployment, not by trade or region, but by ability to find work. Seems as though those that have work are less affected by economic woes (loss of income) then those who have no job and are looking for work; in other words it is harder to find employers hiring even though the numbers of persons being – ah relieved – of employment is relatively low. This trend would indicate caution by firms, particularly in light of evidence an increasing portion of those new hires are temps; not a ringing vote of confidence that the new economic miracle of the past few months has long term focus. What else this suggests, I don’t know, but maybe someone else has thoughts on it.
Both the jobs model and the household survey have problems. With the survey, the data sample is so small that the confidence interval is huge. This is the noise I think that gets referred to. The two are also not based on the same population: non-farm jobs vs. those over 16 with a job (or not). The other thing to keep in mind is that both of these float in a background of what’s happening in the wider economy. And let’s face it we are in uncharted waters. These measures are increasingly dubious. What I do is compare and contrast and add in other measures. The U-3 at 10.2% represents 15.7 million people. This is a much different number than the U-6 which measures everyone who would like a fulltime job but can’t get one. That number is 17.5% and represents a whopping 27 million Americans. And even this last figure doesn’t capture Americans who have permanently left the pool of the working population.
There tends to be a conflation of the U-3 and jobs numbers. I think this is deliberate. The U-3 is twice as large as the jobs number and U-6 number is 70% higher than the U-3. But the U-3 is a percentage (like the U-6) so by reporting the jobs number along with the U-3 the mistaken impression is left that the employment problem is smaller than it really is. However when you look at the numbers behind these measures you begin to see how utterly meaningless the Obama Administration’s job efforts really are.
Methinks the decision to recommend 300k as the right number for jobs lost in October is little more than the “fallacy of the middle ground”. When faced with divergent positions, one simply decides the right answer must be “somewhere in between”. As others have rightly observed, the confidence interval around these figures is wide, even when we are confident in them. For the payroll figure, isn’t the 95% confidence interval something like 180k either way? If we decide that the plug has messed up the payroll survey, we still haven’t figure out how much. Taking out the jobs the plug puts in is no more than a guess at the right adjustment for births and deaths. We might reasonably want to take away a lot of jobs, not just go to zero.
Tossing out 300k is one of those “I’m an expert” bits of nonsense. The right answer is that, if we don’t trust the data, we really don’t know how many jobs were lost.
Now, our confidence should grow over a period of months, because randomization takes over. The margin of error gets smaller. Sadly, over a 3 month period, the relationship between household and payroll job tallies is the same as in October – households reporting 3 times as many jobs lost as does the payroll survey. We can guess that whatever is generating the error is persistent, but we haven’t identified anything that accounts for that sized error. So again, the right answer is “we don’t know”, not “300k).