Forsyth quotes John Williams from Shadow Government Statistics (www.shadowstats.com) who says that seasonal adjustment "has fallen apart in the last several years due to the effects of the extraordinarily protracted and severe economic contraction. Simply put, the severe decline in economic activity has overwhelmed traditional patterns of seasonal activity, destabilizing the calculation of seasonal-adjustment factors using the traditional mathematical models that are based on a number of years of activity, with the greatest weighting given to the most recent period's patterns."
Forsyth then says, "That means the worse the recent experience, the greater the seasonal adjustment to boost the unadjusted result."
However, what both Forsyth and Williams aren't taking into account is that the seasonal adjustment procedure used to produce the employment/unemployment numbers takes unusual events into account. Depending on the filters used and the settings for the extreme value procedure, the values in recent years may NOT be getting the largest weights. The current seasonal adjustment procedure also takes the length of the reporting period into account, something that can sometimes have a significant effect on the numbers.
I was recently looking at some specific manufacturing series for a client. In some sectors, their production dropped significantly in 2008, and when production began to pick up in 2009, the seasonal pattern was different than it had been earlier in the decade. For a series like this, the seasonal adjustment could be quite unstable for several years as the software tries to discover what the new seasonal pattern will be. However, for the employment series I've looked at over the past two or three years, the actual seasonal pattern (and the pattern related to the length of the reporting period or the weekday composition of the month) has remained quite steady. It doesn't matter if the overall level of the series drops --- if the seasonal pattern stays the same, the seasonal adjustment procedure can estimate the pattern and remove the pattern.
As often happens in articles about seasonal adjustment, the "solution" that is often suggested is to look at year-over-year changes. This, in fact, was suggested in the comments by Stephen Wilson, who wrote, "As to seasonal adjustment, I can't imagine why anyone cares. It snows every winter and it's hot in the summer. If you look at year-over-year rates of change the only distortions are from truly unusual events. Both the household and payroll survey agree that jobs are growing slowly and the rate of growth is slowly rising."
The problem with year-over-year changes is that it takes too long to see any meaning changes in the economy. When will we come out of this slump? Isn't that really what everyone is talking about, what everyone wants to know? Year-over-year changes won't let you see a turning point in the economy until it is well past.
Seasonal adjustment does result in loss of information. It is critically important for analysts to look at the original numbers and the year-to-year changes. It is also important to remember that the seasonal adjustment procedures in use now are much more sophisticated that simple exponential weighting of past values.
By the way, the subheading of the article is "Past years' bad economy ironically help flatter current data. Bob Dylan knew better." And he begins the article by saying, "Everybody talks about the weather, but government statisticians actually do something about it: they seasonally adjust it." If only that were true about the weather. I'm reminded of a favorite quote of mine:
The role of the economist is to make weathermen look good. --Stephen Gallogly
No comments:
Post a Comment