Do Peaks in Weekly Unemployment Numbers Mean a Recession is Ending?

INITIAL UNEMPLOYMENT CLAIMS AND RECESSIONS

Many have stated in Seeking Alpha articles and elsewhere that employment is a lagging economic indicator. That may be true from some employment metrics, but is not true for all. Recently the Weekly Initial Unemployment Claims (WIUC) has been much in the news. Here we will examine how this number relates to the business cycle and recessions in particular.

Economic Indicators

Various employment measurements are included in the U.S. Dept. of Commerce economic indicators. Specifically:

I. There are two labor statistics in the Index of Leading Economic indicators:

1. Average weekly hours of production workers (manufacturing)
2. Average weekly initial claims for unemployment insurance (inverted)

II. There are two labor statistics in the Index of Coincident Economic Indicators:

1. Employees on nonagricultural payrolls
2. Personal income minus transfer payments

III. There are three labor related statistics in the Index of Lagging Economic Indicators:

1. Average duration of unemployment (inverted)
2. Change in index of labor cost per unit of output
3. Ratio of consumer installment credit outstanding to personal income

Economic Cycles

Economic cycles are determined by the National Bureau of Economic Research (NBER). Their process and historical data can be obtained here.

The NBER defines a recession to be a period of sustained economic contraction. A recession starts at the end of a period of expansion (a business cycle peak) and ends when the economic contraction ends (a business cycle trough). The measurements they track include some employment factors and Gross Domestic Product (GDP), among others. They do not include Weekly Initial Unemployment Claims. Their stated reason for this is that this metric is too noisy (too much week-to-week variation).

Announcements by NBER usually lag the actual event by 12 – 16 months. Most recently it was announced in December, 2008 that a recession had started in December, 2007.

Weekly Initial Unemployment Claims

This data is reported weekly by the U.S. Dept. of Labor.
The following graphs display the behavior of WIUC over the past 40 years. Information marked on each graph is as follows:

• Vertical green lines define the official beginning and end of each recession.
• Vertical dotted red lines (after 1978) indicate when the start of the recession was officially announced.
• Blue traces are the 4-week averages (seasonally adjusted)
• Red traces are the 52-week moving average of the 4-week averages.

The 40 years has been divided into ten graphs (approximately four years each) so that detail is visible.

Discussion of the data continues after the graphs.
click to enlarge










Discussion

It is clear that there is a significant “bubble” of weekly initial unemployment claims associated with reach recession. For all seven recessions, the bubble has always started well before the recession officially starts. Also, there has always been a peak in WIUC near the end of the recession. However, a peak in WIUC does not always indicate the end of a recession is near. For the 1969-70 recession there were two WIUC peaks. The first was nearly nine months before the end of the recession. Therefore, the pattern for the seven recessions has been that the last WIUC peak has been close to the end of the recession. So when you have established a peak in WIUC (as appears to be the case in the current recession), it is necessary to wait until the recession ends to be sure that there will not be another WIUC peak.

The data for the seven recessions of the past 40 years is summarized in the following table:


I have arbitrarily assigned some signal names to the crossovers of the 4-week average WIUC values with the 52-week moving average of the 4-week average values. (The 4-week average values are used by the U.S. Labor Dept. to smooth the very noisy weekly numbers.) One ground rule is that no signal is generated until the 4-week average (blue line) has been above the 52-week moving average (red line) for 5 weeks. The signal names I have assigned are:

• Recession Alert – Blue line has been above the red line for five weeks without a total of five intervening weeks below the red line.
• Recession Watch – Blue line has been above the red line for 15 weeks without any intervening block of five consecutive weeks below the red line.
• Recession Warning – Blue line has been above the red line for 25 weeks without any intervening block of five consecutive weeks below the red line.

A WIUC bubble is ended on the last week before five consecutive weeks with the blue line below the red line.

In the past 40 years there have been four recession warnings in WIUC bubbles that ended without a recession. These are listed in the following table:



Is there any way to tell early in a WIUC bubble warning whether it has the potential to be followed by recession? One thing we note is that the 30 week and 31 week bubbles had much shorter duration than the shortest warning that was actually fulfilled (69 weeks, 8/1/1981 to 11/27/1982). Until proven otherwise in the future, bubbles this short should be considered unlikely valid warnings.

Another indicator might be found in measuring the “strength” of the bubble in the early weeks. I have chosen to use a summation of the differences obtained each week from (blue value minus red value). This is a measure of how high each week the 4-week average is above its 52-week moving average. I have selected 30 week and 50 week intervals to make the summations. The results are shown in the following table for all seven fulfilled warnings and the four failures:


After 30 weeks, there is no basis to distinguish between legitimate warnings and false warnings. The two false warnings that lasted for 50 weeks, however, have significantly smaller summations than the seven legitimate warnings. Thus, for data to date, we have a basis for screening out false warnings.

In addition to the eleven warnings, there were nine additional alerts. Of these, six did not survive for 15 weeks to become watches. The three that became watches lasted for 16, 20 and 23 weeks. So, in approximately 40 years we have had 20 alerts, 14 of which became watches, 11 became warnings and 7 were followed by recessions. This sounds something like the record of weather predictions by the National Weather Service.

Notes

There are a number of arbitrary decisions made in setting up this analysis. Among these:

1. I selected a 52-week moving average trigger reference. I tried two others (26 weeks and 70 weeks) which did not work as well, but this search for the best trigger line was not exhaustive.
2. It is possible to hand draw horizontal “resistance” lines to get bubble triggers much closer to the start of recessions. This appears to be a promising approach, but for this initial study I felt defining these lines could have a possibility of prejudice, at least as perceived by the reader. The long term moving average removes the possibility of such prejudice.
3. This is a small sample: seven valid warnings and four false warnings. One can not be sure that the next eleven warnings may find one or more that do not fit the patterns established in this study.
4. The intervals for defining alerts could be refined. It is possible that a workable system could be designed with fewer alerts and fewer failed warnings.

Summary

It is proposed that a useful early indicator for the start of recessions can be derived from the Weekly Initial Unemployment Claims reported by the U.S. Dept. of Labor. This is not currently used by the National Bureau of Economic Research, the widely accepted authority for defining the start and end of recessions.

Significant conclusions:

1. Definable patterns of increasing WIUC precede the official start of all recessions studied.
2. A peak in WIUC has occurred within a few weeks of the official end of all recessions studied.
3. To date, increases in WIUC during the past 40 years that have not preceded official recessions have been distinguishable no later than 50 weeks after they have started, most much sooner.

This is a preliminary study. Hopefully, this can be developed further and add something to the ability to forecast the probability of recession well before an official pronouncement by the NBER.