Outliers of moderate magnitude cause large changes in financial time series of prices and returns and affect both the estimation of parameters and volatilities when fitting a GARCH-type model. The multivariate setting is still to be studied, but similar biases and impacts on correlation dynamics are believed to exist. The accurate estimation of the correlation structure is crucial in many applications, such as portfolio allocation and risk management. This paper ocuses on these issues by studding the impact of additive outliers (isolated, patches and volatility outliers) on the estimation of correlations when fitting well known multivariate GARCH models and by proposing a general detection algorithm based on wavelets that can be applied to a large class of multivariate volatility models. This procedure can be also interpreted as a model miss-specification test since it is based on residual diagnostics. The effectiveness of the new proposal is evaluated by an intensive Monte Carlo study before it is applied to daily stock market indices. The simulation studies show that correlations are highly affected by the presence of outliers and that the new method is both effective and reliable, since it detects very few false outliers.