Forecasting Euro Area Recessions in real-time with a mixed-frequency Bayesian VAR

In this paper I use the predictive distribution of the back-, now- and forecasts obtained with a mixed-frequency Bayesian VAR (MF-BVAR) to provide a real-time assessment of the probability of a recession in the euro area for the period from 2003 until 2013. Using a dataset that consists of 135 monthly data vintages and covers 11 soft and hard monthly indicators as well as quarterly real GDP, I show that the MF-BVAR is able to capture current economic conditions extremely well. For both recession periods in the sample, the Great Recession of 2008/2009 and the European debt crisis 2011/2013, the MF-BVAR real-time recession probabilities soar right at the onset of the pending slump of GDP growth. By contrast a BVAR estimated on quarterly data detects both recessions with a substantial delay. While typically non-linear discrete-choice or regime switching models have to be used to predict rare events such as recessions, my results indicate that the MF-BVAR can not only compete with other nowcasting approaches in terms of the accuracy of point forecasts, but also reliably detect rare events through the corresponding predictive distribution which is easily available as a by-product of the estimation procedure.