Optimizing composite early warning indicators

Research on predicting financial crises has produced various composite early warning indicators (EWIs) using macroeconomic and financial time-series. Much of the focus has been on identifying the best leading indicators for financial crises (e.g., credit-to-GDP ratios, financial asset prices, etc.). This paper instead focuses on how to optimally extract and combine signals from multiple cyclical indicators. We find that when combining multiple indicators into a composite EWI, jointly optimizing the indicators improves performance relative to optimizing individually and combining their signals. The performance of our jointly optimized EWIs is robust to the key modelling choices inherent in their design including the trend-cycle decomposition method and the preference for false positives over false negatives.