Leading indicators of crisis incidence: Evidence from developed countries

We examine which indicators are most useful in explaining the cost of economic crises in EU and OECD countries between 1970 and 2010. To define the dependent variable we combine a measure of costs to the economy, which consists of the output and employment loss and the fiscal deficit, with a database of crisis occurrence designed specifically for this task. We take into account model uncertainty in two steps. First, for each potential leading indicator we select the relevant prediction horizon by using panel vector autoregression. Second, we identify the most useful leading indicators with Bayesian model averaging. Our results suggest that domestic housing prices, share prices, and credit growth, and some global variables, such as private credit, constitute important sources of risk.

► We propose an early warning model of crisis incidence for 40 developed countries.
► Crises are captured by a continuous indicator of the real costs for the economy.
► Prediction horizon is determined for each indicator by panel vector autoregression.
► The set of useful leading indicators is selected using Bayesian model averaging.
► Domestic private credit emerges as the most useful predictor of crisis incidence.