This paper compares two specifications of the Capital Asset Pricing Model for a sample of German stocks. The specifications generate time-varying first and second moments by conditioning on past information. This explicit modelling of the time series behaviour of risk allows us to characterize the driving factors of variances and covariances of returns. In addition to a variety of diagnostic tests we evaluate the validity of the one-factor restriction in the CAPM. The main findings are that risk is time dependent and very variable and also that more than one factor is needed to fit the data set.