by Thomas Cooley and Peter Rupert
Data on initial claims for unemployment insurance and continued claims (aka insured unemployment) were released today. Many of the headlines took the announcement and provided articles titled Rise in Weekly Unemployment Claims Points to Faltering Jobs Recovery (WSJ) or Weekly Unemployment Claims Rose to 1.4 Million in U.S.: Live Updates (NYT). While technically true these articles are misleading. The reason is that the headline numbers that come out are almost always seasonally adjusted. Here is a graph of the seasonally adjusted data for initial claims starting at the peak on March 28:
As mentioned by the media, indeed claims increased from 1,307,000 to 1,416,000, an increase of 109,000 people. However, seasonal adjustment alters the raw data by applying seasonal adjustment factors. When times are crazy, when something very odd happens, seasonal adjustment can give misleading information. Here is a graph of not seasonally adjusted initial claims:
The actual number of people applying for initial claims actually fell from 1,512,763 to 1,370,947, a decline of 141,816! Moreover there was an increase in initial claims from July 4 to July 11.
So, what is going on? Think about what happens when, say, Christmas is approaching. We know every year that Christmas is on December 25. Knowing that there will be increased shopping in the preceding months, producers and stores ramp up employment to deal with it. The following graph plots payroll employment both seasonally adjusted and not. Note how the not seasonally adjusted (red) data is so regular. October and November build up employment and every January and February it falls. The seasonally adjusted data (blue), on the other hand is quite smooth. Why? Essentially the statistical adjustment accounts for the fact that every November employment goes up in anticipation of Christmas. What the seasonal adjustment does is to compare, say, this November with last November.
Imagine now the government mandated that Christmas would be on November 25 instead of December 25. In anticipation of that change the employment build up would most likely be shifted by a month. However, the seasonal adjustment factor would apply the old seasonal.
This pandemic has altered the traditional employment unemployment cycles. Stores were closed, schools were closed, auto retooling changed months, and so on. To understand where we are then we should really be looking at the not seasonally adjusted data. This begs the question as to why we look at seasonally adjusted data at all. Back to the Christmas example. Is the holiday season a boom or a bust? The only way to think about that is to compare this season to last season and not this December to this November. That is essentially what seasonal adjustment does.
To be sure, given all of the caveats above, we still ain’t doing so hot.
