Stephen Williamson on the employment picture:
Stephen Williamson: New Monetarist Economics: let's not be too hasty.... I looked at the 12-month growth rates in unadjusted establishment employment.... [T]his looks more innocuous. Employment growth is still negative (just barely) year-over-year, but it is coming back strongly, and faster than was the case during the last recession. Conclusion: How you filter the data matters a lot. There is no reason you should trust my year-over-year filter more than what the BLS uses (X11, X12, or some variant) to do seasonal adjustment, but at least the year-over-year filter is simple enough for a simpleton like me to understand exactly what it is doing...
But, I think, the problem is precisely that Stephen Williamson does not know what he is doing with his year-over-year filter. He thinks it "is simple enough for even a simpleton like me to understand exactly what it is doing." But he is wrong.
The X11 and X12 filters deseasonalize employment: the level of the filtered series then tells you (the filter's guess at) the amount of employment, the slope about the change in employment.
A twelve-month year-over-year change filter both deseasonalizes employment and takes a derivative: the level of the filtered series then tells you whether things have gotten better or worse over the past year, and the slope tells you whether things are getting better or worse more rapidly or more slowly.
When Williamson eyeballs his series, he finds it "innocuous" because it is telling him that things are getting worse more slowly.
When the rest of us eyeball the BLS series, our hair is on fire because it tells us that things are bad and that things are not getting better.
Williamson's claim that "how you filter the data matters a lot" is, to my mind, completely wrong--or, rather, right only to the extent that how you filter the data stands as a proxy for what question you are asking. And, indeed, what question you are asking does matter a lot.
There's no point in pretending that the answer to "are things getting worse more slowly?" has much to do with the answer to "are things good, or at least getting better?"
Yet Stephen Williamson thinks the X11 and the year-over-year filters are both producing data series that answer the same question. And the reason he thinks that is precisely that he does not understand that year-over-year is not only deseasonalizing but also taking a derivative.
I remember being very proud of my seventeen-year-old last January when she was looking over my shoulder at some Obama Administration economic charts and said they appeared to "place excessive reliance on the second derivative." Seventeen is no longer a babe, but still: out of the mouths...
