342 Economic Cycles
Heckman and Singer (1984), Bover, Arellano and Bentolila (2002) then extend
their logit model to account for unobserved heterogeneity.
In more recent macroeconometrics literature, Mudambi and Taylor (1995), Pagan
(1998), and Ohn, Taylor and Pagan (2004) propose discrete-time tests for duration
dependence and use bootstrap methods for finite-sample inference. Harding and
Pagan (2002, 2006) propose new measures and tests for cycle asymmetries and
synchronization. The practical importance of duration analysis for aggregate series
is best illustrated by the March 28, 2007, testimony of Federal Reserve Chairman
Ben Bernanke to the Joint Economic Committee of Congress. In response to his
predecessor Alan Greenspan, who warned that the current expansion could be
fizzling out, Bernanke responded:
I would make a point, I think, which is important, which is there seems to be a
sense that expansions die of old age, that after they reach a certain point, then
they naturally begin to end. I don’t think the evidence really supports that. If we
look historically, we see that the periods of expansions have varied considerably.
Some have been quite long.Bernanke thus discounts the notion that expansions exhibit positive duration
dependence. This view concurs with that of Ohn, Taylor and Pagan (2004), who
fail to reject the constant-hazard assumption for post-World War II expansions but
who do find statistically significant evidence of positive duration dependence in
pre-World War II expansions. On the other hand, consider that the lack of support
for positive duration dependence in the post-war period may be due to the small
sample size. The mean duration of post-war expansions is about 50 months with a
standard deviation of about 30 months. A mean larger than the standard deviation
suggests positive duration dependence.
Finally, in this chapter we have stressed the advantage of a separate analy-
sis of upswings and downswings in unemployment. Indeed, given a downswing
in unemployment, aggregate output is always rising, but given an upswing in
unemployment, the behavior of output is a coin toss. For a young spell of rising
unemployment, an increase in capacity utilization of 1 percentage point increases
by sevenfold the probability of a turning point from upswing to downswing. In
contrast, for downswings the interest rate spread appears to affect the termination
probability, and capacity utilization does not appear to matter.
7.10 Appendix: LIMDEP 7.0 program for jackknifing duration data
/LIMDEP 7.0 PROGRAM FOR JACKKNIFING DURATION DATA, June 2007/
READ ; FILE = LUexp.TXT ;? DATA IN ASCII FORMAT
NVAR = 6;? NUMBER OF VARIABLES
NOBS = 129;? NUMBER OF OBSERVATIONS
NAMES = SB,D1,D2,D3,BUS,PHASE $? VARIABLE NAMES/* ADD “;TEMP = TFILE” FOR LARGE DATA SETS.