Confusing Unemployment Numbers

I have been digging deeply into the unemployment numbers and trying to decipher what can and can not be learned at this point in the business cycle from various statistics relating to unemployment. The sources of data and analysis that I’ve looked into include the U.S. Dept. of Labor (DOL), the DOL Bureau of Labor Statistics (BLS), John Williams’ at ShadowStats.com, various individual investigations and my own analysis. Hyperlink references are provided in the article.

U.S. Dept. of Labor Data


There is hope that the recession, now approaching 18 months in length may be ending, based on a metric that has a good record of indicating the end of a recession, a peak in weekly initial unemployment claims. Last week we reported a signal from the initial claims data that the recession might have ended, a signal with a historical success record of 86% (six correct out of seven signals).


However, two other signals that have been nearly coincident with the end of recessions have not yet been flashed. These are the average weekly manufacturing hours worked reaching a minimum and the number of insured unemployed reaching a peak. These were discussed last week here. A possible reason for delay in reaching a peak in the number of insured unemployed will be discussed later.


Prof. Jeffrey Frankel of Harvard has also discussed the coincidence of a lengthening work week with the end of recessions. Prof. Frankel directs the program in International Finance and Macroeconomics at the National Bureau of Economic Research, where he is also a member of the Business Cycle Dating Committee, which officially declares recessions.


What Is the Actual Unemployment Rate?



A very good video clip is available, courtesy of Barry Ritholtz’s The Big Picture, which explains some of the current employment numbers that seem arcane to many. That video can compared to a new analysis by the author that develops the idea of implied total unemployment. The video suggests that the true unemployment rate might be as high as 18-20% and the author’s implied total unemployment rate weighs in at 25%.

Another unemployment rate that is watched by many is calculated by John Williams’ at Shadow Government Statistics - Home Page. This rate is currently 20%. These various estimates are all higher than the DOL unemployment rates, currently at 9.4% for U3, the “traditional” rate and 16.4% for U6, the “broadest” measure of unemployment, according to the BLS.

No wonder someone who only has a couple of hours to find out what the unemployment rate is gets confused. Somewhere between 9% and 25% is not a very satisfying answer.


How Reliable is the Official Data?


Last week the author posted a detailed report discussing the processes and uncertainties in the DOL employment data. There are some questions about reliability of some of the models used in analyzing the data, but none has been discussed (and cussed) more recently than the so-called Birth/Death adjustment for businesses.

Barry Ritholtz (The Big Picture) has shown the following graph, source attributed to Mike Panzner (B/D Data series source here).



The big question on the right hand side of the graph refers to the unlikely relationship shown, with the highest net new business formation since 2000 (from a model) occurring while GDP is at a low not seen for 60 years. Many have expressed skepticism that the model used by the DOL is adequate for the extreme economic conditions. If the skeptics are right, then the DOL unemployment rate (U3) should actually be higher.


What Other Factors Are Different for Unemployment in This Recession?


The average length of unemployment is at record levels, as shown in the following graph, form the St. Louis Fed FRED data base. This has recently been discussed by Dirk van Dijk. The peak in this statistic has always come after the end of recessions, more than two years after for the two most recent.

We can expect the record to be much higher than the 22.5 weeks average for May, and probably not reached until at least the middle of 2010. It is possible this peak might not be reached until the second half of 2011. The long average duration also implies a long delay until the peak unemployment rate is reached, as well.


The peak insured unemployment rate may occur later than past recessions because of the unprecedented extension of unemployment benefits. This has been pointed out by Jeff Miller of New Arc Investments.




The following two graphs, from BLS Table A-5, show how the two categories of part-time employment have varied since 1964. Three important things jump out:

• Part-time employment for non-economic reasons (I’ll call this part-time by choice) has generally been increasing steadily over time. There have been four periods when this has declined: the early 1980’s and the early 1990’s had gradual declines; 1965-66 and 2007-09 have had steeper declines.
• Part-time employment for economic reasons (I’ll call this forced part-time) has fluctuated up and down with the economic cycle. It has generally peaked after recessions have ended.
• The level of forced part-time work is approaching 50% higher that the previous highs

There is a fourth noticeable effect in both charts: discontinuity jumps (actually one sharp jump and a corresponding sharp drop) in January, 1994 from December, 1993 levels. I have not researched the reason, but it almost certainly is the result of a change in measurement parameters. The change apparently caused people who had been counted as forced part-timers into the part-time by choice classification. The current high level of forced part-time would probably be double the previous highs if measured under the old rules.






Summary


A number of things are giving mixed signals in the unemployment numbers. Indicating that a recession may be ending is the April 4 peak in weekly initial unemployment claims. With one exception, such a peak has occurred within 9 weeks before the date eventually declared to be the end of the recession.

If that relationship were to hold in the current situation, the recession would eventually be declared to have ended no later than this month (June).

Other signals historically immediately preceding, or coincident with the official end dates of recessions are:

1. a minimum in the average weekly manufacturing hours worked and
2. a maximum in the insured unemployed rate. Neither has yet occurred.

The employment data in this recession shows a number of divergences from other recession of the past 40 years. These divergences include:

• The historic relationship between the Birth/Death adjustment and the GDP is massively out of whack. Either GDP will be corrected drastically upward (as much as 6% annualized for the first quarter 2009); or the Birth/Death adjustment has added too many jobs to the employment rolls (too many by 100,000 or more).
• The average duration of unemployment is well above all previous records (since 1948). Since this statistic peaks well after a recession ends, average duration of unemployment may reach levels much higher, possibly 50% or more above previous post World War II recessions.
• The number of people in forced part-time positions is currently between 50% and 100% higher than previous recessions since 1964. This also generally peaks well after a recession ends and could, therefore, go much higher.

We have a mixed set of signals about the possible end of the recession from historically accurate indicators. The analysis of unemployment data is further compounded by the excursion of several measurements far into uncharted regions. It should be no wonder that the title of this article starts with the word “confusing”.


By John Lounsbury, http://piedmonthudson.wordpress.com
John Lounsbury -- Seeking Alpha