Data watchers beware, huge seasonal adjustments distort the numbers
This comes from Alan Abelson at Barron’s talking about seasonal adjustments for February data:
The misleading figures cut across a wide swath of the economy, encompassing housing, manufacturing, employment — you name it. The leading agent of deception, unintentional or otherwise, has been that old sly villain, seasonal adjustment. As it turns out, the seasons don’t need adjustment as much as the adjustors need seasoning.
As Merrill Lynch’s David Rosenberg (who, incidentally, is planning to do a bit of adjusting himself and moving back to his native Canada; our loss, Canada’s gain) points out in a recent commentary, the official keepers of the books have been unusually aggressive in constructing seasonal adjustments for February’s economic data.
To illustrate, the seasonal adjustment for new-home sales was the strongest since 1982; for durable-goods orders, the strongest since they were first released in 1992; the retail-sales figures for February were flat (or, as David says, flattering) after such adjustment, but unadjusted fell 3%, the biggest drop on record. He also notes dryly that the 40,000 raw non-seasonally adjusted housing-start total for February “all of a sudden becomes a headline-adjusted annual rate figure of 583,000.”
Which makes David think that come the inevitably sharp downward revisions of such distorted data, first-quarter real GDP is likely to suffer a 7.2% drop. Which, together with the 6.3% skid in the fourth quarter of 2008, would be the worst back-to-back contraction in the economy in 50 years.
Obviously, we should be wary in looking at any of the government statistics now being released without digging deeper. This little blurb helps demonstrate why.
Source
In Dante’s Footsteps – Alan Abelson, Barron’s
The interesting thing about these adjustments is how it obscures what’s really going on in the economy. Unfortunately for homebuilders, they only anticipate the revenues of selling 40,000 new homes (if they’re lucky), not the revenues from 583,000 “seasonally adjusted” homes. Same with the other numbers Abelson cites.
The only real value of SA numbers is to compare sales with other months within the last year—and not even the year ago number where the YOY NSA number provides a more realistic picture of economic activity.
I also find comparing the change in numbers from this period’s “preliminary” number with last period’s “revised” number is very much an apples and oranges comparison. Still, it is the inevitable “headline” comparison. Compare preliminary with preliminary OR revised with revised–they are measuring the same thing whether SA or NSA. (This takes on a whole new dimension with the persistent revisions of GDP numbers.) The statistically consistent result of these PvR comparisons is that it UNDERSTATES the change period to period. This is usually because the preliminary data and the adjustments to it are both incomplete. The revised data is usually nearly complete.
In short, it takes significant digging to find out the true state of the national economy. One rule: NEVER accept the headline number and comparison at face value!
Terry, you read my mind. Month-on-month comparisons are meaningless as it is – that much more when you have significant revisions. What does have validity as you say is the year-on-year comparison.
John Mauldin says this well in his article from this week, “Why Bother With Bonds?“”
I am certainly looking for bullish data points, but I hardly taking headline dat at face value when we are suffering the deepest recession in livin memory.
Calculated Risk gently corrects Mauldin on Housing?
stilettoheels,
Thanks for the CR post on Mauldin. I hadn’t seen it. While Mauldin’s data analysis may show flaws, his conclusion which I quoted regarding the need to parse the data is very much correct.
CR says one can – and should – use month-to-month data, albeit in the context of a larger picture and with caution. I would agree here as well.
What makes this downturn, especially data-tricky is its magnitude, and that makes data adjustments and data revision very significant.