A simulation study is conducted in order to backtest and evaluate one-day ahead Value at Risk estimations when dealing with multivariate portfolios. A commonly practice of banks and other financial institutions is to treat portfolio returns as a univariate series and proceed to estimate VaR for a single return series, i.e., ignoring dependence among the assets that conform this portfolio. Empirical evidence shows that complex patterns of dependence can be found in multivariate data. Copula functions allows us to model these complex patterns of dependence with great flexibility since they can be parameterized to include measures of dependence between the marginal distributions, regardless of the form of the margins. So, the aim of this work is to identify what are the main determinants when estimating portfolio VaR and under which circumstances ignoring dependence among assets can lead to misleading VaR estimates.