Value-at-Risk at Commercial Banks Before, During and After Financial Crisis: An International Perspective

Using actual data of seven banks from 2001 to 2014, this paper investigates the performance of banks’ internal risk models and alternative VaR models. Our empirical analysis shows that banks’ internal models intend to overstate their VaR. Fitting banks’ P/L data to alternative VaR models, we find that the Historical Simulation and GARCH-type models with Gaussian innovation can provide more accurate VaR figures under normal market conditions. During crisis period, the GARCH-type models with Student t assumption are superior. The Extreme Value Theory approach, which was shown to be powerful in estimating VaR in recent financial crises, performs poorly when applying to the actual banks’ data. We conclude that good VaR estimates can be obtained using simple, accessible models rather than complicated parametric methods or banks’ internal VaR models. Thus, the blame for misleading VaR estimates during financial crisis should not be cast upon the mathematical techniques, but simply depends on the choice of VaR models.