Early warning indicators for banking crises: a conditional moments approach

This paper presents a novel methodology to calculate thresholds in an early warning signalling framework for extracting signals useful to predict the occurrence of banking crises. The conditional moments based methodology does not rely on assumptions on an objective function trading off Type I and Type II errors and leads to the identification of zones corresponding to different intensities of the signal. The signalling performance of these signalling zones is similar to that of the traditional early warning method based on the optimisation of a policymaker’s loss function; our methodology in fact outperforms the latter for a number of indicators. The methodology is then extended to allow for country specificities, which leads to a substantial improvement of the signalling power. On average, across all indicators, the country-specific signalling zones outperform the pooled approach, resulting in a larger average true positive rate and a lower false alarms rate.