This original model free toolbox evaluates the forecasting abilities of an Early Warning System (EWS) or those of two competing EWS. First, it finds the optimal cut-off, the one that best discriminates between crisis and calm periods. Second, it computes several evaluation criteria for the predictive abilities of the EWS. If two models are considered, it evaluates each model and then uses comparison tests to identify the outperforming one. The code relies on 2 series (optionally, 3 series and the choice nested/non nested models). The first one is the observed crisis (binary, taking the values of 0 and 1). The second one is a series of probabilities issued from any type of EWS. Another series of probabilities (a 2nd EWS) must be provided if 2 EWS are to be compared.
Working Paper (2021)
Couperier Ophélie and Leymarie Jérémy