Version numérique PDF

This paper investigates empirical risk measures that were available during the 2008 financial crisis. We compare traditional market risk computations and some new computations based on Extreme Value Theory. We show that standard computations led to deeply underestimated real risks. We also show that the conditional Generalized Pareto Distribution model is more accurate and reliable for predicting asset risk losses and would have protected bank trading portfolios. The Basel II agreement reveals a perverse effect, namely that banks have an interest in hedging between models. The capital penalties applied for abuse appear to be insufficient to push banks into choosing the most reliable model.Keywords: Market risk; Value at Risk; Extreme Value Theory; GARCH model; Financial Crisis; Basel Requirements.JEL classification: C22; G10; G21.

Auteurs :Kourouma Lanciné
Extrait de la revue BMI 126