Academic journal article
By Abbring, Jaap H.; Van Den Berg, Gerard J.; Van Ours, Jan C.
Journal of Business & Economic Statistics , Vol. 19, No. 4
In this article, we study U.S. unemployment dynamics using grouped unemployment data from the Current Population Survey over the period 1968-1992. We estimate a model that traces variation in these unemployment data, both over time and between demographic groups, back to the underlying variation in the inflow and the outflow. In turn, we model the outflow as a transition process in which we allow the exit probabilities to depend on calendar time, duration, and demographic group. We use the estimation results to provide a decomposition of aggregate U.S. unemployment dynamics in various incidence and duration components.
KEY WORDS: Business cycles; Seasonality; Unemployment composition; Unemployment duration; Unemployment incidence.
Over the last decades, macroeconomists have increasingly recognized the relevance of gross worker and job flows to the understanding of aggregate fluctuations in labor markets. Large gross flows between labor-market states coexist at each point in time and dwarf the net changes in the stocks of workers and jobs in each state (Blanchard and Diamond 1990; Davis, Haltiwanger, and Schuh 1996). In particular, this is true for unemployment, which suggests that unemployment dynamics play a central role in the reallocation of labor and therefore in macroeconomic dynamics.
In this article we study U.S. unemployment dynamics by analyzing the variation in the flows into unemployment and the aggregate unemployment duration distribution over time and between groups of workers. We apply a novel methodology to analyze grouped unemployment data from the Current Population Survey (CPS) over the period 1968-1992. These data provide time series of aggregate unemployment by duration class and a limited number of demographic groups. This allows us to calculate, for each demographic group, the aggregate inflow into unemployment and the aggregate outflow from different duration classes at each calendar-time point. Thus, we can estimate models for the inflow, or incidence, and outflow. In particular, we explicitly model the outflow as a transition, or duration, process in which we allow the exit probabilities to depend on calendar time, duration, and demographic group in a flexible way. The results allow for a decomposition of U.S. unemployment dynamics in a variety of incidence and duration components.
Our analysis addresses various issues from the ongoing debate on unemployment dynamics. The first of these is hardly controversial. Unemployment incidence and duration are generally found to be countercyclical, leading to countercyclical unemployment. This is confirmed by our analysis. Moreover, like the existing literature, we find upward trends in both incidence and duration. There is less agreement on the relative contribution of incidence and duration to the countercyclical variation in the unemployment rate. The early literature finds that incidence fluctuations are particularly important in the 60s and the 70s, but the more recent works attribute a dominant role to duration in the 70s and the 80s. It has been argued both that this difference is due to invalidity of the steady-state assumptions used in the early literature and that it reflects actual differences between the time periods considered. Our results are mildly in support of the more recent works.
Another issue that raises controversy is the source of cyclical fluctuations in aggregate duration. Darby, Haltiwanger, and Plant (1985) argued that durations are relatively stable at the disaggregate level and that fluctuations are caused by fluctuations in the shares in the inflow of groups with different exit probabilities. Baker (1992a) dubbed this the "heterogeneity hypothesis." Darby et al. found some evidence for their hypothesis, but the few other works that have pursued this issue rejected it. We find that variation in the shares of the few demographic groups that we distinguish hardly generates any cyclicality of the aggregate exit probabilities. …