Hits and runs: Determinants of the cross-country variation in the severity of impact from the 2008–09 financial crisis

Recent empirical work on the 2008–09 financial crisis has found mixed results on the usefulness of indicators to explain the cross-country variation in the incidence of the crisis in non-originating countries. While some authors have found success with various indicators, Rose and Spiegel (2009a,b) find that almost no indicators are robust. We employ Bayesian model averaging (BMA) to verify Rose–Spiegel’s conclusions under model uncertainty, confirming their findings. We then employ latent class models (LCM) to check the data for parameter heterogeneity. We find that there is substantial evidence of heterogeneity in the relationship between various indicators and crisis impact, both across individual indicators as well as across financial crisis episodes. In particular, when using de-trended growth rates, a similar model fits the 1997 Asian financial crisis, although the coefficients change qualitatively in some cases. These results highlight the difficulty in employing simple linear models for early warning purposes, but demonstrate that there are robust indicators of cross-country variation in crisis impact across episodes, such as the pre-crisis growth in banking credit. A 2-class model explains the variation in crisis impact, where pre-crisis level of per-capita income assists in the prediction of membership in a particular class.