On policymakers’ loss functions and the evaluation of early warning systems: Comment

Sarlin (2013) suggests that if a loss function approach is chosen to derive the optimal threshold for financial crisis early warning indicators, the loss function specification should explicitly take into account the unconditional sample crisis probability. In this comment we argue that this approach is not robust to small perturbations of the preference parameter and is not easy to use for policy purposes. We suggest therefore to continue using a simpler loss function specification.