NFP: Retiring the Over/Under Bet

A little over two years ago, I mentioned on Squawk Box that I was taking the "Under" on Non-Farm Payrolls.

Since that appearance, you may have heard Steve Liesman or Mark Haines
or Larry Kudlow talk about the Over or the Under. It was a cute device
that made fun of the dismal scientists in a good natured way.

My reasons for tweaking them were simple: Economists, using the typical post WW2 recession/recovery cycle, were over-estimating employment and growth, and under-estimating inflation. By betting they were overly optimistic, I was making a point about their methodologies, as well as the current economic environment.

My basis for this that we were not in a typical recession/recovery
cycle, but rather, were in a post-crash environment, which has very
different connotations and risk factors.

Indeed, even the Fed admitted (in today’s WSJ) "that because of faulty inflation data, the Fed kept interest rates too low for too long earlier this decade, fueling speculative housing activity." Yes, they too under-estimated inflation.

Unfortunately, the time has come to retire the over/under. Given the magnitude of the subsequent revisions, adjustments, birth/death factors, benchmarkings and other statistical sleights of hand, the initial number has become totally marginalized.

This does not at all change my views about jobs, the economy or inflation. It is just that NFP has become, in a word, meaningless.

The most recent offense was the absurdly enormo BLS revision to their benchmark. It turns out that for the year ending March 2006, BLS now claims their monthly data understated payroll growth by 45 percent. Over that period, we are now told the data was off by 67,500 per month.

Based upon this self-admitted numerical bastardization, it seems foolish to put too much weight onto whether a monthly data point is 90k or 150k or 230k. If it turns out  this single number is plus or minus a million per year, than the monthly data becomes worthless.

So as much as I am tempted to take the Under, I am retiring not just the bet but the entire process.


Today’s consensus is for 130,000  new jobs, with a range of 75,000 to 180,000.

But I feel like Woody Allen (Alvy Singer) as a 9year old kid in Annie Hall, who discovered thermodynamic entropy:

Doctor in Brooklyn: Why are you depressed, Alvy?
[Young Alvy sits, his head down – his mother answers for him]
Alvy’s Mom: Tell Dr. Flicker . . . It’s something he read. 
Doc: Something he read, huh?
Alvy: [his head still down] The universe is expanding.
Doctor in Brooklyn: The universe is expanding?
Alvy: Well, the universe is everything, and if it’s expanding, someday it will break apart and that would be the end of everything!
Alvy’s Mom: What is that your business?  [she turns back to the doctor] He stopped doing his homework!
Alvy: What’s the point?


What’s the point of dissecting a number which may or may not be accurate to within plus or minus 45%?

More on this later . . .


UPDATE:  November 3, 2006 7:47 am

Do not interpret this to mean that the market will ignore the number; Too hot a data point — warts and all — will end speculation of a rate cut anytime soon; Too soft a number further dents the goldilocks scenario.


Fed Official Says Bad Data Helped Fuel Rate Cuts,
Housing Speculation

WSJ, November 3, 2006; Page A6

Print Friendly, PDF & Email

What's been said:

Discussions found on the web:
  1. libertas commented on Nov 3

    The point is that it is something to bet on. It doesn’t matter whether or not it has any economic significance. Even if the number were accurate, employment is such a lagging indicator that it isn’t useful for anything in particular.

  2. Paul Davis commented on Nov 3

    question: does the statistical insignificance of the nfp report extend to average hourly earnings?

    isn’t this the important number, at least in the short term?

    and do you take the over or under?

  3. Eclectic commented on Nov 3

    It’s still before the number is released, so I’ll take a stab.

    My guess is that anything above 200k would be a dramatic surprise and anything less than 25k probably would.

    My guess (purely a guess) is that we may lose jobs this time, maybe 25k or more.

  4. tick tock commented on Nov 3

    51K revised to 148K ?!?!? I guess now the enormo revisions will come monthly instead of annually. Ridiculous. Barry, your comments couldn’t have been more timely.

  5. bob commented on Nov 3

    The numbers are completely useless now after the two huge revisions of previos numbers. I’ve got the numbers. I will ignore them.

  6. S commented on Nov 3

    Unit labor costs is my favorite read on how much slack there is in the labor market. And it’s been going in the wrong direction.

  7. Dave L commented on Nov 3

    What Tick Tock said.

    Who can take this stuff seriously? And the result is that we’re left to guess at what employment, inflation and GDP growth actually are! I can’t help wondering if growing official data problems aren’t going to feed into uncertainty and confidence.

  8. Cherry commented on Nov 3

    Election politics. The unemployment rate is useless as they consistantly take the lower margins over the consensus, which is easily seen through and embarrasing. Then of course you have the end of expansion red herrings, see 2000 which makes it equally useless right now.

    Though the job losses can be seen through in October, the tide of unemployment will be rising and maybe .5 jumps at time some months. Revisions down now will commence.

  9. Jason commented on Nov 3

    So, is it a problem with the information that is processed, structure of the process, the process itself, or those who do the processing?

    Is it accident, negligence, or corruption of the process?

  10. The Hube commented on Nov 3

    Its not like the bls is telling people that the numbers mean anything. They are 90% confident that the actual number is between a loss of 343,000 jobs and a gain of 527,000 jobs. bfd.

  11. Josh commented on Nov 3

    With these revisions that just happen to occur before the election….like “Lowest unemployment in 5 1/2 years” makes our government reported figures no better than Venezuela’s.

    What a shame.

  12. Cherry commented on Nov 3

    You also must understand they “revised” the previous months up which caused the consensus to be lowered a bit(which they took the lowest possible number) for October though job cuts really increased.

    This month was by far the worst looking report since 2003. Manufacturing losses and construction losses(from RE bust) starting to climb. They may have to revise downward. Consensus then rises, maybe a lot in November/December.

  13. tick tock commented on Nov 3

    Josh, I think you mistakenly added an “e ” to the end of your last word.

  14. Andy commented on Nov 3

    CNBC’s Steve Liesman ripped you off totally — he said pretty much EXACTLY what you wrote.

    He should have just read your piece word for word.

  15. BDG123 commented on Nov 3

    Payroll is noise that is hard to pin down accurately IMO. The more important statistic is first time jobless claim filers……..That’s also pretty hard to screw up. It’s controlled within one government agency and is an easy count. Or, at least it should be. With the government one never knows.

  16. Eclectic commented on Nov 3

    My guess of -25k was obviously wrong, but it doesn’t feel so wrong.

    Recessions don’t begin from high unemployment, but rather from good times and high employment.

    Employers at first attempt to hold on to employees, hoping their declining business contributions to GDP are short-lived. Once this is proven to be erroneous, then employers have no choice and don’t hesitate to cut jobs.

    Employment then declines even past the point that the economy begins to recover from recession.

    It’s a question of timing.

    Also, let’s put 92,000 jobs in perspective.

    There are approximately about 3,500 counties in the U.S. (can’t remember exactly; but middle 3000s), but I like to think of them as being 5,000 average ‘illustrative’ counties, each with a population of about 60,000. Then these counties have about two cities of 30,000 each. That is a representative picture of much of America.

    What 92,000 means is that: an average county produced approximately 19 jobs last month, and an average city of 30,000 produced less than 10.

    That’s not the powerful economy that’s the “greatest story never told.”

    Roubini has been right on forecasting GDP changes (not sure what he’s said about jobs lately), and if he continues to be correct, the days of even 92,000 new jobs per month may run out soon.

    It’s all a matter of whether we manage to pull a soft landing out of the hat.

  17. Macro Man commented on Nov 3

    So how is the non-manufacturing ISM actually a coded message for impending doom?

  18. andiron commented on Nov 3

    150 k /month is old stuff..fed has adjusted it to 100k/month..90 k , if true, is not a bad number abd infact iss inflationary…
    hence the 10 yr yield jumped

  19. S commented on Nov 3

    How can the market not demand a higher equity risk premium when the economic data are totally inconsistent and unreliable causing the bond market to sell off hard? Instead, they buy equites. VIX and VXN decline. Bizzaro world.

  20. jmf commented on Nov 3

    73.000 from the 92.000 (80%!!!) are just a creation of the bls from their black box “birth death model”. this gets more curios when you compare the 06 numbers with the 05 oktober number wich assumes that 57.000 were somehow createt. in 05 the economy seems to be way stronger then today. (maybe this is relatet to the katrina).

    have a nice weekend

  21. Cherry commented on Nov 3

    Also to add, a big chunk of the jobs revised upwards were government jobs. Big Government Bush at it again lol………..

  22. Hockeyman77 commented on Nov 3

    I have posted here several times in the past about just how useless this number was. With large up and down revisions on a regular basis. Glad to see others are now agreeing

  23. Neal commented on Nov 3

    Data was captured in the field, sent to a secret prison, was shocked and waterboarded. Days later data said the economy is great. Is it surprising that data is tortured by this administration especially when there are no data rights organizations. The stakes are too high to be restricted by outdated conventions.

  24. Kevin Rooney commented on Nov 3

    I wonder if it might not unnerve the market a bit to realize that it does not have reliable jobs data to use. That might be the most significant “data” in this data, the fact that there is no usable data. Which could be scary if jobs data seems inherently impossible to gather; scarier if jobs data is seen to be made unavailable.

  25. alexd commented on Nov 3

    I think Kevin Rooney gets the elephant in the room prize.

    The elephant that no one mentions.

    Usually if someone is selling me something and presents information that is suspect I am going to think at least one of two things.

    I am dealing with someone who does not know what they are talking about or they do and they are trying to decieve me for their own gain.

    Actually the worst thing might be when we take bad information and deceive ourselves.

    It might be a matter of convienant oversight. I would think that if a government had really great figures based on the truth the information would be as clean and direct as can be. But if things were corrupt then they would either want to not present the figures or distort those figures through distortion or distraction. (Ignore that man behind the curtain!)

    It is like watching Kudlow and having him talk about how much businesses ( or ceo’s) are earning without mentioning that for the average woking person or even worse unemployed are not experiancing such gains. It is not lying but it manipulates attention and tries to present a part of the truth as the whole truth or the most important part of that truth.

    Just to be fair to Mr. Kudlow, anyone doing this is just as incomplete. Also he does not always do this, (I might differ politicaly, but he is smart) but the political tail has been wagging the doggy for a while.

  26. alex commented on Nov 3

    It is just soo bad…and it’s getting worse by the day…everything is being manipulated by the cronies…GDP data, employment data…it is sickening…if manufacturing workers (with beneifist, etc.) are cut off the income and their whole family geoes burger flipping for $8/hour (without any benefits) we celebrate that as an increase in employment. It’s time to talk about what kind of employment…the data point,s indeed are meaningless.

  27. tt commented on Nov 3

    the data has always been manipulated …… all of a sudden it’s a surprise !!!! …. what’s the surprise 1) that you didn’t know this has been going on the last 50 years ..or …. 2) that you are so naive ?????

  28. Macro Man commented on Nov 3


    The unemployment rate amongst workers with a bachelor’s degree or higher is 1.9%. The unemployment rate amongst workers with some college education is 3.4%. The unemployment rate amongst those with a high school education is 4.1%. The unemployment rate amongst those with less than a high school education is 5.8%.

    Wage and salary income for the population as a whole is running at a pace 7.6% higher y/y. Adjusting this for the increase in employment over the least twelve months lowers the rate of incerase to 5.6% per worker. This is higher than the y/y change in average hourly earnings and has been for a number of years.

    The household data suggests that year to date, employment amongst those with at least some college education has risen by 1.2 million, while employment amongst those with none has risen by 1.0 million.

    Does this sound like all the new jobs are of the burger flipping variety? Without wishing to put word’s in anyone else’s mouth, the general tone on this board sounds as if everyone has already made their mind. A ‘strong’ number is evidence of statistical noise at best or perhaps a government conspiracy, whereas a ‘weak’ number is just proof that the housing market is killing the economy.

    While it is important to stick to one’s intellectual guns, it is also important to remain open-minded in evaluating new information. Viewing data through ‘bull goggles’ or ‘bear goggles’ is, in the end, a self-defeating exercise. Challenging one’s own views is likely to prove more rewarding (jntellectually and financially) than writing off contrary data as irrelevant (though I would concur that the monthly NFP figure SHOULD fall into that category) or a conspiracy.

    Just my $0.02.

  29. Eclectic commented on Nov 3


    My Son, my Son… my, but you do have your BVDs in a wad about this BLS report today! Are you going to take your sit-n-spin and go home?

    Time for a little recess, and you all can gather around as well.

    I wonder if all the posters today realize what they’re saying about the jobs report.

    Liesman, if you’re listening, you should be ashamed of yourself.

    First of all, the BLS statisticians are seasoned professionals at what they do and they’re not about to let politics govern their reports. You’re all confusing the comments of pundits about what they report with the actual report.

    Silly, silly bond market today… selling off so big when there’s no evidence from this report that changes anything from even yesterday.

    There are approximately 137 million jobs that are the basis for this very complex study that the BLS conducts almost continuously, since they not only have to survey the current month, but also have to revise previous months.

    In their own literature, they tell you that the report is merely assumed to be 90% accurate to plus or minus 430k jobs. Do you realize that your complaints of this type of accuracy are tantamount to asking for an accuracy greater than .003?; that is, you’re asking for some count of 1,000 observations to be more accurate than 3 observations in 1,000.

    430k/137mil = .00314

    The fact that they can claim 90% accuracy to within 3 observations per thousand is pretty remarkable if you think about it.

    Follow along with me in this wee parable:

    Assume you’re standing near a large elementary school and the school lets all the children out for recess at the same time. There are something like 1,000 children milling about in a large playground, and your task is to count them. But, the rules are that you can’t corral them; you can’t meter them through a gate; you can’t make them sit or stand still; and you can only count them once.

    These children have all sorts of aspirations about what they want to do. Some want to play… some want to sit… some want to fight… some want to group together, and some want to be alone. In places there is great activity and movement and in other places, they’re quiet and not moving. But, in all these places you have to count them with equal intentions of accuracy… all while they’re in a state of flux.

    Who among you could claim that you could count these children, as you stand beside the playground (even if you were allowed to mill about yourself)… and be assured of being accurate to within plus or minus just 3 children… and this if you were only allowed to count them once… and even if you were allowed to tally recorded marks as you proceeded with the count?

    Could you, Tick Tock?…

    Jason, you really think there’s corruption in a report with a 90% probability of having just plus or minus .003 error?

    Hube has it figured out, but he’s (or she’s) just not taking the time to give all of you a spankin’ like I am.

    Hockeyman, if you were a sheep farmer, would you think one of your shephards was a responsible employee or not if he counted a flock of yours in the field and he was accurate to within just 3 sheep in a thousand… while they were moving and grazing?

    Puts a new perspective on this whole BLS thing, huh?

    Even my earlier picture of what 92,000 jobs would look like makes the assumption that the 92,000 is itself accurate, and it may be far from it.

    The boob tube and the online financial print media are full of what you, BR, call… innumeracy. You don’t want to also be an innumate, do you?

    If we have a significant upturn or downturn in employment, it’ll only show up as statistically significant amounts that are greater than plus or minus 430k jobs. Until that happens, an economy in flux has no more accurate means of measuring employment than this BLS study.

    Hat’s off to the BLS. They do a good job. I’ve seen some of their people interviewed and read some things they’ve written. They’re crack jack professionals at what they do. They would probably think all of you are as silly as I think you are.

  30. Con Spiracy commented on Nov 3

    Eclectic ,

    you’re ruining our fun ….. if there’s no conspiracy , then maybe the numbers are better — and we don’t want that

  31. Alex Khenkin commented on Nov 3

    Eclectic, you’re correct, strictly speaking. However, they DO release their central value of, let’s say, 92,000 as if it matters. Psychologically, it is hard to put that number away and pay attention to the probable range, unless that range is given more prominence. It’s like the average temperature on the Moon – no one in the right state of mind would even mention the average when the range is between -150°C and +100°C, or some such. But they put out the number as if it means something, and then are surprised that people cling to it.
    Small Investor Chronicles

  32. Eclectic commented on Nov 3

    Fate is a cruel taskmaster.

  33. Eclectic commented on Nov 3

    “Eclectic, you’re correct, strictly speaking. However, they DO release their central value of, let’s say, 92,000 as if it matters.”

    No, Alex Khenkin… it doesn’t matter at all. No more than if I elected to have you count sheep and you claimed an accuracy greater than .0006; that’s just 6 sheep in 10,000.

    [92/430 times .003] = .00064

    Who would be the greater fool… me, for believing your accuracy, or you for claiming it?

  34. Eclectic commented on Nov 3

    “But they put out the number as if it means something, and then are surprised that people cling to it.”

    A foolish consistency is the hobgoblin of little minds.

  35. blam commented on Nov 3

    Thanks to Eclectic for putting the whole thing in mathematical perspective. Now could someone please explain it to the hype machine. The jobs number has taken on the import of a real number.

    The agency is at fault for announcing a number as if they can measure whether thre has been a change in jobs from the previous month. By ommission, they infer accuracy and precision of the estimate, when in reality, the number is always zero or no change.

  36. A Dash of Insight commented on Nov 3

    An Open Note to Steve Liesman

    Steve Liesman is a fine journalist. He is not an economist, but he has covered the economic beat long enough to understand the key issues. He can be depended upon to listen carefully and distill the key points of analysis

  37. Eclectic commented on Nov 4

    While no single observation that is small enough to be within the limits of statistical error (and thus of no assured significance) may mean anything at all, cumulative observations that exceed those limits are significant.

    In other words; a single change in the number of jobs that is within plus or minus 430k may not mean much, but a cumulative change of, say, 1 mil jobs does.

    I haven’t said that the economy hasn’t produced jobs or that it won’t continue to do so, because it has so far. There is a statistically significant difference between, say, 134 mil jobs and the current 137 mil as measured by the BLS.

    Those who claimed 6 months ago that the economy would continue to expand and produce jobs have been correct so far (I thought not and I’ve been wrong), because we can now measure a significant change that proves it.

    To be fair to BLS, it’s not the agency itself that takes the estimates and claims accuracy from them… it is the pundits who desire to spin the figures they produce in some light that supports their point of view.

    Consequently, if it had only produced 20,000 jobs instead of the 92k, they’d still attempt to spin it as being evidence the expansion will continue.

    An economy slowing from 3 plus percent GDP to 1.6 is not consistent with employment growth, however if the economy continues to slow (as Roubini is rather poetic about claiming), when employment does begin to decline, we can’t be sure that it will be proven in any upcoming particular monthly observation.

  38. diva commented on Nov 4

    I ‘told you so’.

    (Sorry, couldn’t resist.)

  39. Cherry commented on Nov 4

    E, real growth was strong through the first half of the year. The 3rd quarter(late summer, early fall) was the first “sluggish” indicator that things like employment growth will be falling, but it takes time, though things are hard to restart again when they do come to a complete stop and contract.

  40. Econbrowser commented on Nov 5

    Using those employment numbers

    What do you do when one line of the latest government statistical release says that U.S. employment grew by 92,000 jobs during October, while 4 paragraphs later the same report gives the number at 437,000?

Posted Under