Forecast-Augmented Credit-to-GDP Gap as an Early Warning Indicator of Banking Crises

This paper explores whether augmenting the credit-to-GDP series with a forecast improves the early warning property of the credit-to-GDP gap – a frequently used indicator of excessive credit expansions calculated using the one-sided Hodrick-Prescott filter. Improving the early warning property of this indicator is extremely important as policymakers frequently rely on it when deciding about macroprudential policy interventions such as when calibrating the Basel III countercyclical capital buffer or other macroprudential instruments. Using data for 56 countries over 1950-2016, we simulate in a quasi-real-time setting how different types of forecasts would have changed the gap. We build simple statistical forecasts, more complex economic forecasts based on regression models estimated in real-time with IMF country-specific WEO macro projections used as inputs, and plausible credit cycle corrections. The early warning power of alternative credit-to-GDP gaps is tested within the ROC/AUROC framework. Our results indicate that for advanced markets, none of the adjustments can beat the simple one-sided filter, but for emerging markets, the correction-adjusted gaps could improve the signalling power.