Can statistics-based early warning systems detect problem banks before markets?

• Systems designed to identify problematic banks have grown in complexity and accuracy.
• Our research shows they have not become efficient enough to inform markets.
• Accumulating data over time allows models to function during periods with no bank failures.
• We find that the factors determining bank failure are stable over long periods of time.

Statistical early warning systems (EWS) to identify problematic banks have grown in sophistication, complexity, and accuracy, but can they inform markets? We utilize five “archetypical” EWS using a unique dataset which accumulates data from 1986 through 2009. An arbitrage portfolio is formed by shorting problematic banks and going long the remaining banks. We find accumulating data allows the models to function during long periods with few or no bank failures and that the factors used are stable. While all models studied do a good job predicting bank failure, we find that EWS are unable to inform markets.