Banking crisis in Asia and Latin America – A single pattern for emerging market economies?

Abstract

Most extant work on prediction of banking crises has utilised global samples, which are in turn dominated by observations from middle-income countries, and rely on a single estimator, while a range of specifications is desirable to check robustness. However, economic and financial structure as well as the pattern of shocks may differ substantially across regions. Accordingly, in this paper we test the implicit pooling assumption in earlier work on Early Warning Systems using the widest range of models, by estimating logit, signal extraction and binary recursive tree specifications separately for crises in Asia and Latin America, as well as the pooled sample. Results suggest markedly different crisis determinants across regions, implying global samples are inappropriate.