The purpose of this study is to investigate the function and ranking of multivariate GARCH models in value at risk estimation. To this end, three portfolios having different dimensions, small, medium and large and having different fluctuations are selected, including the return of industrial stock price index in Tehran Stock Exchange from 3/26/2011 to 10/21/2014 in order to lead us picking up the best model in different portfolio conditions. In order to challenge of estimating value at risk in evaluating the variance-covariance matrix in high dimension because of constructing stochastic and undetermined parameters, this study uses the modeling of composite likelihood approach. Following the consideration of statistical efficiency of the models, the models ranking procedure was done based on sener test. The result indicates that for final conclusion focusing on over allocation of resource to hedge the risk plus noticing the gaps in erros, those models taking correlation into account, especially dynamic correlation models, have worked out better in different levels and dimensions. In most cases, the dynamic correlation models with composite likelihood estimations are in first place, as compared to other models and Considering the effects of asymmetric shocks partly improved the results. Also, different conditions with different sizes and the various sampling of data can affect the rating of the model. However, there are not sufficient reasons to conclude that the performance of the t distribution at different levels of statistical error is better than normal, because with decreasing the levels of statistical error at 5% to 1% it works worse.