An Early Warning System for Macro-Prudential Policy in France

We construct an early warning system for detecting banking crises that could be used for the macroprudential policy conduct in France. First, we select macro-financial risk indicators among a large number of candidates by considering their performances both over a panel of ten euro area countries and at the French level, for the 1985:Q1 to 2009:Q4. Second, we run all the logit models including four of these indicators, one being necessarily a measure of credit gap to fit the Basel Committee recommendations. We then retain two sets of models: one including those with all coefficients significant and expected signs, another one, obtained by relaxing these criteria. Third, we aggregate the probabilities estimated by the models, by giving each a weight proportional to its usefulness at predicting crises either at the panel or the country-level. The simulations performed both over and out of the sample show that aggregating more models yields better results than relying on one single model or only a few. Performance is also enhanced by aggregating models' results with country-specific weights relatively to common panel-weightings