In this paper we present a lossless compression algorithm for colour video sequence which exploits the spatial, the spectral and the temporal correlations of a colour video sequence in the RGB colour space using the well-known optimal prediction theory. The main idea is to construct the optimal prediction coefficients estimating an autocorrelation matrix which exploits all these correlations. No colour transformation or motion compensation are applied because reversible colour trans- formations are not able to fully decorrelate the three bands of each frame, and motion compensation remarkably increases the complexity of the updating step of the auto- correlation matrix estimate. Furthermore, the fact that our algorithm is not based on motion compensation lead it to be robust to scene changes.