This contribution generalizes the work of Drexl and Haase about the so-called proportional lot sizing and scheduling problem which was published in 1995. While the early paper considered single-level cases only, the paper at hand describes multilevel problems, i.e., items are interconnected via a directed network of acyclic precedence constraints. It provides mixed-integer programs for several important extensions which differ in the allocation of resources. A generic solution method is presented, and following the preceding paper, a randomized regret-based sampling method is tested. A computational study proves that, even for the multilevel case which is far more complex than the single-level problem, promising results are obtained.