John Lounsbury

Employment Data ABCs

Mike - - -

Good thoughts. The sampling process to assure a random sample is not a trivial thing. That is one detail to be dealt with using the internet.

Requiring companies to submit their payroll information would be attacked on the political front as a "government intrusion" and a "1984 Big Brother" tactic. It would be "another government regulation impeding the progress of American business".

So, as attractive as the possibilities you suggest are, they do have problems.

Goldman Pushed AIG to the Edge

Mark - - -

Sorry for the title typo - "I" have corrected. Maybe I was thinking an Attorney General should be involved?

Employment Data ABCs

enigmaman - - -

The use of the number of 7.2 million jobs lost (the most widely mentioned number in the press and by politicians) has been repeatedly questioned by me over the past few months. The number employed peaked in November 2007 and that number has remained stable for over two years. In recent months, each new report has increased the number of jobs lost, reaching a maximum of 8.8 million in December. The current number, 8.4, using the January data, is actually not the high point for this recession.

I have not researched it, but I expect that the number 7.2 million was correct a number of months ago and continued to be used even though new numbers would have been more correct each month.

As I have asked before, don't media people and politicians bother to remain current with the data? I see no conspiracy. I see laziness. The DOL reports the numbers of employed and unemployed every month and it is almost as if they are looked at only a couple of times a year.

You ask: "Why doesn't the Fed government report the numbers to reflect reality?"

My answer is they do report the numbers and the problem is that the numbers are not looked at and used. The DOL data bases have been expanded this month to provide even more detailed data. That is a good thing. The fact that many who should use the data won't look at it (and analyze it) is a bad thing.

Birth/Death Adjustment Hubbub: Just a Side Point

Sorry for an error in the above comment. The sample size ratio is 240,000 / 60,000 or 4/1. The square root is 2. Thus a 4-week moving average reduces an uncertainty of +/-300,000 to about +/-150,000.

The +/-125,000 given above is in error.

I have a new article just posted that discusses the employment data methodologies in great detail.

Goldman Pushed AIG to the Edge

Mike - - -

I really did intend to point the finger at AIG in the article. Maybe I should raise rather than point to get more attention?

Thanks for your comment. I will accept that GS was just doing a good capitalist job getting the first $7 billion claims payout. I am not so sanguine about the $13 billion plus a cut from Société Générale's "bailout".

Birth/Death Adjustment Hubbub: Just a Side Point

Ted - - -

Thanks for the link. I like the Minneapolis Fed graphs. If I use data from there in the future, I'll try to remember to give you a hat tip (with hyperlink) because I have not followed the Minneapolis Fed before now.

Markets: The January Effect

Buzzer - - -

Here are the results for 61 years from my data:

72% up years
66% up Januaries

When January was up, 90% of the years were up
When January was down, 65% of the years were down.

When I leave out 1949 to get 60 years, I get the same 89% that you quote.

I am using index value only. Did your reference include dividends? That could account for the difference between 65% and 48% down years when January was down.

Goldman Pushed AIG to the Edge

Jack - - -

I should never forget to use the Taibbi term: giant bloodsucking squid attached to the face of humanity.

Birth/Death Adjustment Hubbub: Just a Side Point

The measurement error uncertainty should be proportional to the standard deviation, which varies as the square root of the sample size. This means that the measurement error is reduced by the factor square root (360 / 60) when we use a 4-month moving average of data which is randomly sampled 60,000 times each month.

The reduction factor in uncertainty is
square root (360 / 60) = 18.97 / 7.75 = 2.45

Thus, an uncertainty of +/-300,000 becomes 300,000 / 2.45 = 122,538.

An approximation for the measurement error uncertainty for the 4-month moving average of the household survey data is +/-125,000.

Birth/Death Adjustment Hubbub: Just a Side Point

Panda - - -

I should repeat what I have written in previous articles because many readers have not followed me through the many articles I have written on unemployment. That then leaves them understandably confused about my statements. I apologize to you and anyone else who is a relatively new reader.

The birth/death adjustment is made to the non-farms payroll employment number, which is obtained from a survey of some 300,000 businesses every month. This is the so-called "Establishment Survey". It does not include any count of self-employed, 1099 sub-contract workers and many mom and pop establishment workers. This survey is valuable because it separates employment into various sectors, such as retail, construction, manufacturing, etc. It produces a non-farms payroll number every month, which is widely headlined but has questionable value. It is not the number used to determine the unemployment rate, the size of the labor force or the total number of people employed, part-time and full-time.

The more important number is obtained from the "Household Survey", which is a completely independent process involving a random sampling of 60,000 households each month. The Household Survey includes the added categories of employment as well as the payroll jobs. This data is used to determine the total number of people employed, unemployed, looking for work, discouraged from looking for work, the number of people officially "in the civilian labor force" and the number of people officially "not in the labor force".

There can be lots of debates about just where the cut-off should be between people not working but still in the labor force and those that are no longer in the labor force. However, the criterion has been consistent for decades. Not working and having sought work during the past four weeks is counted as unemployed and in the labor force; not working and not having sought work in the past four weeks is not counted as unemployed and is counted as not in the labor force.

The measurement uncertainty for the entire population when collecting data from 60,000 households is approximately +/-300,000 every month. That uncertainty in the count of the employed and the labor force is why the monthly calculation of the unemployment rate has no meaningful change unless it is more than 0.4%. When the change takes place smoothly over four months, as is the case from October (10.1%) to January (9.7%), that has more significance than an up and down fluctuation between the two numbers.

When I make trend projections using the household data, I use 4-month moving averages. This reduces the impact of the sampling error uncertainty. I need to consult a statistician (or torture myself with a statistics text book) to determine how much the uncertainty is reduced, but my first estimate is that it might be to a little as +/-100,000 (or even possibly +/-75,000) when 4-month moving averages are used.

I hope this helps you (and others) understand why I made the statements I did.

The Coming of Real, Weak, Imperceptible Economic Growth

Steve - - -

I was interested to read your jobs analysis. It agrees with analysis I published yesterday and today using a completely different process. In my article at TheStreet.com I revised my previous data extrapolation to a crossing of the zero line to start a net gain in employment in the next two months (February or March). The previous extrapolation had indicated starting to gain jobs in the second or third quarter.

However, we can not be sure that we will stay above the zero line for jobs gained. There will be up to a million temporary workers hired by the census bureau that will again be unemployed within about 12 months. Stimulus money that is supporting some state and local government positions will be diminishing as the year progresses. Stimulus dollars flowing into construction jobs will be largely used up by the end of the year.

And even if we do stay above the zero line, how many jobs can actually be added in 2010? The problem with just returning to net jobs growth of 50,000 or 100,000 or even 150,000 per month is that we are in a deep hole. For example, we need more than 4 million added jobs this year just to accommodate the return of the people who have dropped out of the labor force plus new people that are expected to be added in an average year due to population growth. Adding 150,000 per month, which seems an unlikely large average for the year, would accommodate less than half of those 4 million. And that provides nothing to help find jobs for the 15 million unemployed that we start this year with.

I have to disagree with your assertion that the Dept. of Labor data is crap. The data is quite good - it's the headlines that are generated from the data that are crap. You very correctly point out that the change from 10.1% unemployment rate in October to 9.7% in January is not meaningful. That change is in the noise level of the data. The survey data uncertainty is much larger than 0.4% change in unemployment. However, having four months with no rise in the unemployment rate is significant within the sampling error of the DOL surveys. It supports your assertion that a minimum in the number employed could be near.

Steve, your discussion of employment is as complete as any I have seen. Great job.

The Coming of Real, Weak, Imperceptible Economic Growth

bbro - - -

Good data. Recognizing the sampling error uncertainty, the magnitude of the decline from 10.1% to 9.7% is of questionable significance, especially considering the decline in the civilian labor force during that time. I consider the fact that the rate has not gone up to be more significant, and that would tie in to your correlation.

BTW, I question why your factual comment rates any thumbs down. Bear with the negativity and keep commenting with facts. Thanks.

Markets: The January Effect

Steve - - -

In the article at TheStreet.com I showed that there was no correlation of the size of the return for January with the size of the return for the year. I did not break out the detail you suggest might be interesting.

Doug Short has reviewed some other relationships for the average return for the year when January was up and also when January was down. The link is in the comment by dshort (yes, that is Doug), the first for this article.

The Birth/Death Adjustment Hubbub is a Side Show

Mark - - -

I am projecting from 4-month moving averages to get the February - March crossing of the zero line to positive employment changes. A couple of bad months would change the 4-month moving averages significantly.

I have to agree that there is a risk of positive job growth and a later dip. One cause of this could come from up to a million temporary additions to employment for the census and then seeing these people ending their temp positions near the end of the year.

Markets: The January Effect

To all readers - - -

My original title for this article (when I started writing it, not when I posted it) was "The January Effect Myth". I had the thought that the historical correlations would not turn out to show as much significant difference for some cases as it turned out.

I took the word myth out of the title for both the article at TheStreet.com and the post here. Many of my readers would have liked both articles better if I had stayed with my original theme, in spite of the data.