AbstractCapital flow surges have become a major source of concern as they have been often followed by disruptive reversals and sudden stops. We introduce F-score methodology which evaluates how well particular capital flow surge method can predict reversals and sudden stops. F-scores consider both type 1 and type 2 errors and provide policy makers a framework to weigh economic costs of false negative and false positive signals. We construct and compare a large number of commonly used surge identification approaches, including several machine-learning methods, to investigate which types of formulations best help explain which surges are more likely to be reversed. While considerable literature has investigated the determinants of capital flow reversals and sudden stops with surges being included as one of the independent variables, so far little research attention has been focused directly on attempting to determine the likelihood that particular surge will result in reversals or sudden stops. This is the most important question for policies toward capital inflows since the optimal responses to capital flow surges would be quite different depending on whether the flows are likely to be reversed or not. Unfortunately, theory does not offer a clear guide to identifying surges other than that they are unusually large inflows. We emphasize that appropriate evaluation should involve not only precision in predicting reversals but also accuracy in not giving false alarms by predicting reversals that do not occur. In other words, attention needs to be paid to both type 1 and type 2 errors.