We present the most general model of the type considered by Black and Litterman (1991) after fully clarifying the duality between Black-Litterman optimization and Bayesian regression. Our generalization is itself a special case of a Bayesian network or graphical model. As an example, we work out...