In this thesis various Value-at-Risk models are compared and evaluated towards finding the optimal model for the bank's trading book. The focus is on linear instruments (stocks, indicies and currency) and real market data, both domestic and foreign, is used for the calculations. I find that the GARCH(1,1) based model outperforms other models in volatility estimation and should thus be a wise choice when volatility estimation is needed, but GARCH based methods become rather complex for a multivariate covariance estimation. Therefore a mixture of simpler models such as EWMA is needed for making rational estimates.