We present a framework for designing optimal allocation strategies for large stock portfolios using dynamic factor models and multivariate volatility parametrisations. We attempt to elaborate on the fundamental structure of the Fama and French (FF) factor model with a special focus on the time variation in risk and correlation between stocks returns and systematic factors. For this reason, variants of the multivariate GARCH models are employed to capture the dynamics in means, variances and covariances of the FF factors structure. Based on these models, we derive optimal capital allocation strategies in the framework of Markowitz's mean-variance portfolio theory. We outline and compare the out-of-sample performance of these mean-variance allocations with those obtained using simpler techniques, such as sample historical and exponentially weighted moving average (EWMA) estimates.