I propose a class of hybrid models to describe and predict the dynamics of a multivariate stationary random vector, e.g. a vector of stock returns. These models combine essential features of the multivariate mixture normal distribution and the conditional correlation models. I describe in detail...