This study examines the performance of sixteen Value-at-Risk (VaR) models in the context of institutional portfolio management. We focus on multivariate versus univariate approaches of asset modeling, monthly versus shorter risk horizons, and filtered historical simulation (FHS) versus Monte Carlo simulation (MCS) techniques. Tests on VaR violations show that the best performing models are generally the univariate FHS and MCS models with daily asymmetric GARCH specification.