Cyclical variations in the duration of unemployment spells
by Miles Corak
Business and Labour Market Analysis Division
Analytical Studies Branch research paper series, No. 054
It is often accepted as a truism that the higher the economy-wide unemployment
rate, the longer are the durations of individual unemployment spells.
Aggregate data on the duration of interrupted unemployment spell lengths
certainly display this pattern. Economic theory, however, does not offer
unambiguous implications concerning this relationship, and recently
its empirical validity has come into question. Darby, Haltiwanger, and
Plant (1986) have argued that the economy-wide average duration of unemployment
rises during recessions because the composition of the unemployed becomes
more heavily weighted with individuals that "normally" experience
long unemployment spells, not because individual spells are any longer
than they normally would be.
The purpose of the research reported in this paper is to examine one
implication of this view, namely that controlling for individual characteristics
the probability of leaving unemployment is invariant over the course
of the business cycle. While there are many existing Canadian studies
that examine the duration of unemployment spells, few have adopted a
long enough time horizon to shed any light on this issue. The data used
are derived from the Annual Work Patterns Survey for the years 1978
to 1980 and 1982 to 1985. Accelerated life-time models under a variety
of distributional assumptions are used to examine unemployment spells
from each year of data.
Three major results are obtained. First, the hypothesis that individual
exit probabilities are constant over the cycle is rejected, but the
direction in which they mover is not clear-cut. There is strong evidence
to suggest that the duration of individual unemployment spells increased
between 1980 and 1982 as the economy moved into recession, and then
decreased as recovery took hold. However, the results also reveal increases
in spell durations when the unemployment rate was falling, most notably
between 1984 and 1985.
Second, the pattern of change varies across individual characteristics,
most notably across age. The results document an important and long-lasting
deterioration in the relative position of older unemployed individuals.
The average spell duration of the old not only increased during the
onset of the 1981-82 recession, but it continued to increase during
each year of the subsequent recovery. This contrasts sharply with the
experience of individuals in the prime-aged, and younger age groups.
Finally, from a methodological stand point, it is found that the estimation
results are robust across functional forms, but not over time. Researchers
employing parametric methods to analyze the duration of unemployment
spells should be aware that the magnitude and signs of their estimated
coefficients may be very sensitive to the cyclical state of the economy
at the time their data was collected.
These results permit some inferences as to the nature of the adjustment
process in the labour sector. Adjustment to severe negative shocks might
reasonably be characterized as an evolutionary process in which older
workers are, in one way or another, moved into semi-permanent or permanent
retirement, while younger workers are rehired at increasing rates. The
results also call for continued analysis of spell durations for more
recent data, and with more extensive co-variates.
Not available electronically.
