Real Time Probabilities of Recession

Figure 1 shows the real time probabilities of recessions from the Dynamic Factor Model with Regime Switching (DFMS), estimated in Marcelle Chauvet and James Hamilton (published in Nonlinear Time Series Analysis of Business Cycles, 2006, ed. by Costas Milas, Philip Rothman, and Dick van Dijk). These probabilities are recursively estimated using just-in-time information, which includes unrevised and preliminary data. For example, the probability of a recession in March 2001 is obtained using information up to that month. Since these probabilities use real time information, they also reflect the uncertainty about the economy at each month and are, therefore, spikier than the smoothed probabilities. Nevertheless, the estimated real time probabilities from the DFMS model capture all U.S. recessions, including the beginning (and end) of the 2007-2009 recession (click here to read about The Beginning and End of the 2007-2009 Recession).

A fast assessment on the state of the economy can be obtained by observing whether the real time probability of recession has crossed the 50% threshold for a month. This rule maximizes the speed at which a turning point might be identified, but increases the chances of declaring a false positive. See analysis in Marcelle Chauvet (from International Economic Review, Vol. 39, No. 4, 969-96 (IER) and in Chauvet and Hamilton (2006). A more reliable inference can be obtained using more information to verify a turning point. Considering that the probability has decreased below (increased above) 50% for at least two months avoids taking a one-month-only weak (strong) performance as evidence that the economy has exited (entered) a recession. This can happen, for example, due to natural disasters, strikes, short-lasting external shocks, or a deterioration (or improvement) in one or two sectors but not in the economy as a whole.

Chauvet and Hamilton (2006) proposed to use a low-order smoothed probability in addition to the current real time probability to increase accuracy. The information from the readily available real time probabilities is combined with the more precise information obtained from 1-step or 2-step ahead smoothed probabilities in real-time assessment of the business cycle phases. Using this metric improves the quality of the inference in terms of accuracy. Figure 2 shows the real time current probabilities and the 1-month-smoothed probabilities recession by recession.

Chauvet and Hamilton (2006) compare NBER news releases with the performance of the DFMS model in dating and announcing business cycle chronology in real time. One of the great advantages of the model is with respect to the timely announcement of turning points. The algorithm does very well in announcing the beginning and end of downturns compared with statements released by the NBER. The DFMS model would have beaten the NBER in calling the beginning of a recession in two out of four occasions (the start of the 1990 and 2001 recessions, respectively) and would have coincided in two cases (the start of the 1980 and 1982 recessions). The advantage of the dates inferred from the multivariate model is even more significant for dating the end of recessions. The model beats the NBER announcements in all occasions, with leads from three to seventeen months. The model would have announced the end of the 1980 recession 8 months before the NBER’s announcement, the end of the 1982 recession three months earlier than the NBER, the 1990 recession 17 months earlier, and the recession in 2001 would have been declared to have ended 14 months before the announcement by the NBER. For the performance of the model in the most recent recession in 2007-2009, click here.

Figure 1 – Real Time Probabilities of Recession and NBER recessions based on real-time unrevised monthly data after 1977:12 and revised monthly data before 1977:12