(Universidad Icesi, 2006-08-01) Alonso Cifuentes, Julio César
This paper evaluates the performance of different parametric and semiparametric
methods, as well as the historical simulation method, to estimate the nexttrading-
day VaR of 7 representative portfolios for 7 different Latin American countries. It is found that there is not a single model that outperforms the others. For a 95% confident
level, parametric models with EWMA and TGARCH specification to update the volatility
outperforms the others. On the other hand, those models over-estimate the “true” VaR for a 99% confidence level.
