The Great Recession of 2007-2009 has not only caused a large wealth loss, it was also followed by a sluggish subsequent recovery. Two years after officially emerging from the recession, the economy was still growing at a low pace and payroll employment was far from reaching its previous peak. However, assessment of the employment situation was markedly different across different series. The two most important employment series, payroll employment (ENAP) and civilian employment (TCE), have recently been displaying divergent patterns. This has been a source of great uncertainty regarding labor market conditions. This paper investigates the differences in the cyclical dynamics of these series and the implications for monitoring business cycle on a current basis. Univariate and multivariate Markov switching models are applied to revised and real time unrevised data. We find that the main differences across these series occur around recessions. The employment measures have diverged considerably around the last three recessions in 1990-1991, in 2001, and in 2007-2009, but especially during their subsequent recoveries. In particular, while the probabilities of recession for models that include ENAP depict jobless recoveries, the probabilities of recessions from models with TCE fall right around the trough of the last three recessions, as determined by the NBER. This significantly impacts the identification of turning points in multivariate models in sample and in recursive real time analysis, with models that use TCE being more accurate compared to the NBER dating, and delivering faster call of troughs in real time. Models that include ENAP series, on the other hand, yield delays in signaling business cycle troughs, especially the most recent ones.
