Un test de validité de la Value at Risk

This paper proposes a new simple test of market risk models validation or Value at Risk (VaR) accuracy. The test exploits the idea that the sequence of VaR violations verifies the properties of a white noise. More precisely, we use the Multivariate Portmanteau statistic of Hosking [1980] to jointly test the absence of autocorrelation in the vector of violation sequences for various coverage rates considered as relevant for the management of risks. We show that this multivariate dimension appreciably improves the power properties of the VaR validation test for reasonable sample sizes. Classification JEL : C23, C11