Tuesday, July 14, 2009

Jobless Claims: Comparing the current data with "False Dawns"

- by New Deal democrat
Partly in response to my previous post, New Jobless Claims at the End of Recessions, my co-blogger Invictus has raised the issue of "False Dawns". That's a fair question, and one that I deal with in detail in this post.

The original idea that under certain circumstances (all met here) the failure to make a new high in initial jobless claims during a recession only 3 weeks afterward, meant that the Recession was within a month or two of ending, was espoused by Robert J. Gordon at Vox EU, and was subsequently examined by Dr. James Hamilton of the UCSD at Econbrowser. The discussion by Gordon featured a nice comparative graph, which I have updated to show the subsequent course of new jobless claims since the time he prepared the graph (the data for the current recession is shown in deep red):

Gordon's article is lengthy and I encourage you to click through and read the whole note. In any event, Gordon identifies 6 "false dawns" during the recessions from 1970 to the present, analyzes each one, and purports to show that none of the conditions apply to the new high in the 4 week moving average for initial jobless claims that was recorded on April 4. Since we are now 13 weeks from that high, I have revisited and expanded his analysis to take into account behavior of initial jobless claims more than 3 weeks after a new high. Using a definition of at least a 3% decline in new claims over at least 2 weeks after a high, I have identified a further 3 candidates for "false dawns" in the 7 recessions - including the present one - since initial jobless claims have been reported starting in 1964.

Those calculations result in the chart below, identifying the year(s) of the observation of the period of "false dawn", the raw numbers of claims at both the false dawn and the subsequent false low, the percentage decline in claims, and the week of the subsequent new high in claims. Finally I give the number of weeks from false dawn to false low, and false dawn to new high.

YearweekFalse HighweekFalse Low% declineWeeks to new highWeeks to false lowWeeks to next High
1974 2/09321 5/11 290.5(-9.4%)8/1013 wks.26 wks.
1981 12/26 551 1/16521 (-5.4%)2/13 3 wks. 7 wks.
1982 2/13 552.5 3/6534 (-3.4%)3/27 3 wks. 6 wks.
1982 4/24 587 5/22583 (-0.7%)6/5 4 wks. 6 wks.
1982 6/19 601 7/31570 (-5%)8/21 6 wks. 9 wks.
1990-1 12/29 456 1/19438 (-4%)2/2 3 wks. 5 wks.
2000 8/26 313 10/14297 (-5%)11/18 8 wks. 13 wks.
2001 1/6 352 1/20338 (-4%)2/10 2 wks. 5 wks.
2008-9 12/20 546 1/10524 (-6%)1/24 3 wks. 5 wks.
2009? 4/4?659? 7/4?606? (-8%) --- 13 wks.? ---

In summary, only two of the 9 possible previous false dawns last more than 4 weeks to the false low, and only two lasted more than 9 weeks until a new high. Of those two, only one "false dawn" lasted 13 weeks until its false low, and only that same one featured a percentage decline in new claims as high as the current situation: the false dawn of February 1974. Gordon distinguishes the 1974 false peak by noting that "a distinguishing feature of global peaks is a preceding period of relatively rapid increases in new claims. The weekly change in new claims (as before, the four-week moving average) is always greater than 3% prior to true peaks" - unlike the +0.2% change in the 8 weeks just prior to the 1974 false peak.

As stated in the previous post, our current situation has featured a higher percentage decline from the high in new claims than most prior actual highs since the series' inception in 1964. That means, while the odds are not perfect, there is an extremely high probability, based on past patterns, that the April 4 high was the actual high for this recession and not a false dawn.