The control of the trade-off between encoding rate and quality of compressed video is a challenging task. This control requires efficient rate and distortion (R-D) models, able to describe accurately the behavior of the compressed video. R-D models should account for the dependencies induced by the choice of frame-varying quantization parameters (QP). This paper proposes a dependent R-D model involving two independent regime-switching autoregressive (IRSAR) models. Experimental results show that the proposed model is able to represent accurately the distortion dependency between frames. For the part of the rate due to the texture, the model fits experimental curves reasonably well. The model has to be completed to account also for the part of the rate due to motion vectors and signalization.