The U.S. Department of Labor has two monthly surveys: the household survey and the establishment survey, which also is called the payroll survey. These surveys have two different objectives. The household survey estimates the employment status of the entire population and makes some demographic assessments, while the establishment survey assesses the distributions of employment among various industry groups and governments.
In addition to the two monthly surveys, on a weekly basis initial unemployment claims and continuing insured unemployed are reported. This article examines the relationship of data from all of these surveys to the end of recessions.
The following graph shows how the number of unemployed has varied since the late 1940s. We see that unemployment peaks have occurred at the official end of recessions or after the end.
The following table shows the interval between the end of each recession and the peak in the number of unemployed workers.
Of the last 10 recessions, we have had eight showing unemployment peaks nearly coincident with, but never preceding, the end of the recession. Twice unemployment has continued to rise for more than a year after the recession ended. The last two recessions witnessed a long delay before unemployment started improving, is this a new normal?
However, it isn't new news to most that if you wait for unemployment to decline to make investment changes for the end of a recession you may miss a significant opportunity. However, the ending date for a recession is often officially determined by the National Bureau of Economic Research, or NBER, many months after the fact. Thus, unemployment peaks have often occurred well before the end of a recession has been declared, although not for the last two recessions. A peak in unemployment, therefore, can be a valuable tool to confirm that a recession has ended before an official announcement.
The graph below shows the insured unemployment claims since 1967, when recordkeeping started.
The following table shows the interval between the end of each recession and the peak in the number of unemployed workers. The values obtained from the total unemployment data and from the insured unemployment data are compared with the intervals that occurred before the NBER officially announced a date for the end of a recession.
It is clear that investors make a serious mistake if they do not take advantage of unemployment data, especially insured unemployment, to confirm that a recession has probably ended.
Jeff Miller of New Arc Investments has cautioned that the insured unemployment number may be skewed because of the extended benefits programs during this recession. However, for the last two recessions, the insured unemployment level has been timelier than the total unemployment figure in indicating the end of the recession.
There is a significant sidenote to these observations. The graphs show there were other peaks after the first peak marking the end of a recession -- I call these "aftershocks." There was one aftershock in the total unemployment data for the recession of 1969-1970 and two aftershocks in the insured unemployment data for 2001-2002. The specifics for these are summarized in the following table.
To date, aftershocks have had no significance, other than to indicate a slow employment response to the recovery, commonly called a "jobless recovery."
Weekly Initial Unemployment Claims
The other widely followed employment data item is weekly initial unemployment claims. To smooth short-term volatility, the four-week moving average is commonly used.
The moving average crossings can be used as an indicator of the start of a recession. All seven of the above recessions were forecast with lead times that varied from zero to 15 months. In addition, three other recessions were forecast that did not occur. This was an indicator that forecast 10 of the last seven recessions.
The peak in initial claims (four-week moving average) is a good indicator that a recession is ending. Using the second peak for the recession of 1969 to 1970, all recessions have ended within nine weeks of the claims peak. However, there can be two peaks -- the first peak during the 1969 -1970 recession occurred 30 weeks before the recession ended.
There was a peak in claims (four-week moving average) eight weeks ago (April 4). This has led some to proclaim it is a signal that the recession will be seen to end sometime in May or June.
There are two cautions here. First, we had a head-fake in December and January, when a five-week decline had people looking for the end of the recession. Compare that to the current situation where a three-week drop -- from 659,500 to 638,500 - was followed by five up and down weeks between 632,000 and 624,500.
Second, December/January behavior is not isolated. Examples are the double peak in 1969-1970 and multiple short pullbacks before the peaks in 1981-1982 and 2001-2002. If there is one more week (June 4 report) below the 659,500 peak, nine weeks will have passed and, to be consistent with previous cycles, April 4 should be called a confirmed peak.
Using the Establishment Employment Survey Data
The establishment employment survey is also called the payroll survey. It can be used to assess when various sectors have started growing by looking for bottoms in employment declines by sector coming out of recessions. In addition, this survey also reports hours worked and hourly pay in a few economic sectors. One specific survey result of particular interest is the average weekly hours for the manufacturing sector, shown in the graph below.
The minimum weekly manufacturing hours worked seen in the above graph for each recession are tabulated below. This minimum has never occurred after a recession has ended. There is a possible minimum in place for March 2009; April hours worked was 0.2 higher than March. This may mark the point that the end of the recession was within six months, based on data from the 11 previous recessions. If we look at the most recent eight recessions, the indicated end of the recession would be no later than June (three months).
However, there have been some false minima in this indicator, as shown in the table below. There were false signals in seven of the last 11 recessions. In the recession of 1981-1982, there were two false signals. All of the false signals lasted for only one or two months.
For this recession to have behavior similar to the last eight, one of two things must happen: first, it must end in the March to May time period; or second, the hours worked will turn down again in May or June.
The most timely data from the various employment surveys are the weekly manufacturing hours worked (reported monthly) and the weekly initial unemployment claims (four-week average, reported every week). The weekly manufacturing hours worked had a minimum in March and, if not removed by a lower reading for May or June, will indicate the recession probably ended in May or June. The weekly initial unemployment claims peak (four-week average) apparent peak of April 4 indicates the recession will probably end by June, unless the April 4 peak is taken out with the report this Thursday, June 4.
This recessionary cycle has some characteristics not seen in the 64 years of history covered. Thus, we must always consider the possibility that "this time is different" from the 11 immediately preceding recessions.
At the time of publication, Lounsbury had no positions in stocks mentioned.
John B. Lounsbury is a financial planner and investment adviser, providing comprehensive financial planning and investment advisory services to a select group of families on a fee-only basis. He worked for 34 years with IBM, and spent 25 years in R&D management and corporate staff positions. He also was a Series 6, 7, 63 licensed representative with a major insurance company brokerage for nine years.
Specific interests include political and economic history and investment strategy analysis. He holds degrees from the University of Vermont, Columbia University and the Illinois Institute of Technology, where he studied chemistry, physics and mathematics. He is a contributor to Seeking Alpha and his own blog,