On the optimality of the bayes prediction policy in two‐echelon multistation inventory models