|
Summary /
Abstract
|
We introduce a new Early Warning System Indicator (EWSI) of banking crises based on a non-linear (Gomperzt curve) panel data model of credit deepening. The capability of our estimated credit-gap measure relative to alternative credit gaps computed through ad hoc procedures (linear or Hodrick Prescott trends) is tested in the context of alternative univariate and multivariate Early Warning Systems (EWS). Our new EWSI proves to be an outperforming and reliable leading indicator of banking crises while overcoming most of the problems of linear and stochastic procedures. The estimated credit gap outperforms the rest of indicators in both in-sample and out–of-sample forecasting accuracy (e.g. AUROC statistics) in a univariate comparison. Furthermore, we also test the importance of our EWSI in a multivariate framework through a Bayesian Model Average (BMA) technique, confirming our initial positive results. Finally, we estimate an Early Warning System for 68 developed and emerging countries based on our credit gap which can be used to compute Banking Crises Probabilities and to estimate dynamic thresholds for the EWSI indicators. |