Early Warning Indicators of Economic Crises: Evidence from a Panel of 40 Developed Countries

Using a panel of 40 EU and OECD countries for the period 1970-2010 we construct an early warning system. The system consists of a discrete and a continuous model. In the discrete model, we collect an extensive database of various types of economic crises called CDEC 40-40 and examine potential leading indicators. In the continuous model, we construct an index of real crisis incidence as the response variable. We determine the optimal lead employing panel vector autoregression for each potential indicator, and then select useful indicators employing Bayesian model averaging. We re-estimate the resulting specification by system GMM and, to allow for country heterogeneity, additionally evaluate the random coefficients estimator and divide countries into clusters. Our results suggest that global variables are among the most useful early warning indicators. In addition, housing prices emerge consistently as an important source of risk. Finally, we simulate the past effectiveness of several policy instruments and conclude that some central bank tools (for example, reserves) could be useful in mitigating crisis incidence.Bayesian model averaging, dynamic panel, early warning indicators, macroprudential policies, panel VAR.