Wednesday, May 26, 2010

Barry Ritholtz falls into the "year-over-year" trap

- by New Deal democrat

One of the things I constantly harp on is the misuse of "year over year" comparisons. Sometimes we have to use them, due to seasonality (housing for example). When I use them, I am always careful to tease out the trend -- i.e., is the YoY comparison getting better or worse. If you don't, you make bad, bad mistakes like Mike Shedlock did when he unintentionally bottom ticked Rail traffic a year ago (when the 13 week trailing YoY average was at its worst, but the more contemporary 4 week YoY average had been improving).

Well, today, over at the Big Picture, in Positive Sign: Worker Mobility, Barry Ritholtz uncharacteristically falls into the trap, with regard to Daily Treasury withholding.

Barry says:

One of the data points that has been getting some attention is the total withholding tax receipts, as reported by the IRS.
According to the table below, it is down year over year. Some are interpreting this to contradict BLS, and likely means that the improving jobs data are bogus. Bill King specifically noted that
“as of May 20, IRS data shows Withheld Income & Employment Taxes declined 3.36% y/y. This means income is still declining and by extension, meaningful jobs are still decreasing

That certainly is a possibility. Might there be a better, higher probability explanation?
I am less sure of that conclusion. While I have been skeptical over the years about BLS model changes, I do not believe they are simply making up numbers to please their political masters. Before jumping to conclusions, I suggest we consider other possibilities: If the Employment data is okay, then what might explain the change in withholding?:

- The 2009 tax cut that impacted the vast majority of salaried workers.
- A shift down the payscale by workers;
- Changes in with holding rules as part of the Stimulus package;
- Start up business founders working for little or no pay;
- Hours worked continues to get cut

[emphasis mine]

There is a much simpler explanation. King's supposition is rubbish. Withholding data in no way supports the position that "jobs are still decreasing."

The "year over year" data that King cites is fiscal year to date. He is comparing October 1, 2008-May 20, 2009 with October 1, 2009-May 20, 2010.

If we want to find out if "jobs are still decreasing" (present progressive tense) or not, why the h*ll do we want to include old data?

Look what happens when we divide up the "fiscal year to date" data along the following dates (all numbers in Billions):

Oct-Dec 2008: $445.8
Oct-Dec 2009: $407.3

Jan-Feb 2009 $294.6
Jan-Feb 2010 $280.0

March-May 2009 (1st 16 days) $391.0
March-May 2010 (1st 16 days) $413.3

In other words, in the first period YoY receipts trailed by about -9%. In the second period, they trailed by -5%. In the final period, 2010 receipts exceeded 2009 receipts by about +5%.

The only reason "fiscal year to date" YoY receipts are still (barely) negative is because they haven't been positive YoY recently (about 3 months) for as long as they were negative YoY previously (about 5 months).

Looking at withholding taxes YoY fiscal year to date is a horribly lagging way to look at the data. The correct reading of the data is: "Jobs were still decreasing until about February 2010. Since then jobs are increasing. Very robustly, thank you very much." King's analysis is garbage and it is surprising that Ritholtz falls into the trap.