regime estimates from the unemployment rate series are also closely aligned to the
official business cycle, the exhaustion rate and average duration regimes do not. In
addition, regime estimates from the average duration model are significantly closer to
the regimes of the exhaustion rate model than the jobless claims model. In contrast,
the regimes from the MS model of the unemployment rate align slightly more closely
with the regimes from the jobless claims model. I conclude that the lag in the recovery
of the job finding rate is responsible for the pattern between the unemployment rate
and the average duration of unemployment.
These results add to growing evidence (see Yashiv 2008; Fujita and Ramey 2009;
Elsby 2009) that the inflows into unemployment are counter-cyclical. This is in contrast
to the behavior of inflows in many Pissarides (1991)-type search models which
rely on evidence from Shimer (2005) to model the separation rate as a constant. In
addition, the notion that the exhaustion rate enters a recovery phase later than jobless
claims supports Fujita and Ramey (2009) results, who find that the separation rate
is counter-cyclical and moves contemporaneously with productivity, while the job
finding rate is pro-cyclical, but moves at a lag to productivity.
The nature of labor market recoveries is not only important to our understanding
of the business cycle, but is also critical to policy decisions, such as the extensions
of UI. The lack of job opportunities is often given as the reason for extending UI
and, consequently, it is important for policymakers to understand the source of movements
in labor market statistics. I find that they seem to understand this quite well.
Periods where legislated extension to UI are in place align closely with the regimes
from the exhaustion rate model and are independent from the regimes from the jobless
claims model. This may suggest that Congress pays close attention to the ability to exit
unemployment rather than the total stock of unemployed when determining if longer
benefits are required.
This paper proceeds as follows. Section 2 describes the methodology. Section 3
details the data and Sect. 4 model selection. Then, Sect. 5 reports and evaluates the
parameter and smoothed probability estimates and Sect. 6 concludes.
2 Methodology
It has long been observed that key economic series, including the unemployment rate,
behave asymmetrically over the business cycle, suggesting fundamentally different
processes in each phase of the economy. For instance, the unemployment rate often
rises sharply during a recession and declines at a much slower rate following the
NBER trough. Neftci (1984) was one of the first to present statistical evidence that
such asymmetries exist. While Sichel (1989) shows that some programming issues
led to Neftci (1984) results, subsequent papers (Rotham 1991, 2008) have shown the
unemployment rate to exhibit asymmetry over the business cycle. In order to evaluate
the phases in various labor market series, I employ Hamilton (1989) MS model which
takes the following form:
yt = αst + βst(L)yt−1 + σst t, for st = 0, 1