On the Optimality of Generalized (s,S) Policies