As frequent readers are aware, over the past three months (here, here and here) we had been tracking one rather disturbing divergence in one set of BLS data with another: namely the monthly Nonfarm Payroll change, and drift, compared to the monthly Net Turnover number (Hires less Separations) as reported by the monthly JOLTS survey. The reason for this is that historically the two data series have had a nearly perfect correlation in estimating the number of jobs created by the US economy as shown by the following long-term chart (using latest revised data):

However, beginning in May something odd started to happen: there was a major divergence in the data sets. We showed this last month using the pre-revised NFP data, which indicated that while the average NFP monthly average job gain in 2013 was 196K, the JOLTS number suggested a far lower average monthly "gain" of only 140K. It was shown as follows:

The ongoing divergence led us to reach the following conclusion:

This means that either the JOLTS survey is substantially under-representing the net turnover of workers, or that once the part-time frenzy in the NFP data normalizes, the monthly job gains will plunge to just over 100K per month to "normalize" for what has been a very peculiar upward "drift" in the NFP "data."

And just like last month we will conclude with the same advice to the BLS: when manipulating data series across dimensions, make sure the manipulations foot across, and not just in 1 dimension.

Even the "Bond King" read our post and promptly tweeted the following:

Gross: JOLTS data do NOT validate 200K #payroll prints. More like 125K. #Taper may be delayed if #Yellen has a big vote.

— PIMCO (@PIMCO) August 6, 2013

What happened next is precisely as we expected: when last Friday's jobs report came out, the first thing that stood out, aside from the miss in the lead, August number (which came at 169K), is that the previous number for the July NFP print (which last month was the current one), was revised from 162K to 104K or "just over 100K per month."

Still, while we had caught the NFP in "massaging" data for one month, it meant that there was still a whole lot of pent up divergence, which meant that something quite surreal had to happen to the next month's JOLTS survey to make up for the ongoing divergence which still indicated a massive over-representation in the NFP numbers.

Well, when the JOLTS number came out moments ago, surreal is precisely what we got.

Because while everyone else was focused on the far less imporant Openings print which is rather meaningless (printing at 3.900MM or below expectations of 3.936MM), it was the Hires and the Separations and specifically the difference between the two - the Net Turnover - that had our attention.

It was here that we saw what happens when the BLS is caught in a flat out lie, or rather, what happens when one manipulated, made up data set does not comply with another manipulated, made up data set. To wit: hires in July rose from 4318K to 4419K, while separations dropped from 4228K to 4109K. In summary, this amounted to a Net Turnover of 310,000 jobs. Amusingly, the matching NFP print that went with this July JOLTS-implied job surge is what we now know is the downward revised July Non-Farm Payroll print of just 104K - the lowest since June 2012.

To get a better grasp of just how manipulated the JOLTS Survey results had to be for the BLS, which we had exposed in outright making up data, to make the two series congruent again, is the following chart showing the monthly difference between the NFP jobs number and the JOLTS Net Turnover. The outlier is conveniently highlighted.

This was the biggest differnece between the monthly JOLTS and NFP data on record.

Why did the BLS do this massive data adjustment? Recall what we said in August:

... just like last month we will conclude with the same advice to the BLS: when manipulating data series across dimensions, make sure the manipulations foot across, and not just in 1 dimension

It appears, after ignoring our advice for months, the BLS finally scrambled and crammed all the revisions to the reconcile the two non-matching data series in one month.

Sadly, in the process it merely confirmed that both data sets are very much suspect and likely completely made up.

But there is a silver lining: net of the this real-time data fabrication the made up JOLTS series and the made up NFP series, have finally realigned.