Simple Newsvendor Heuristics for Multiechelon Distribution Networks
Ferguson, Mark E.
Lystad, Erik D.
MetadataShow full item record
We consider the problem of determining optimal stocking levels in a multi-echelon distribution network consisting of m echelons and n non-identical terminal locations. Lead-times are deterministic, there are no fixed ordering costs, and unmet demand is backlogged. Both Clark and Scarf (1960) and Federgruen and Zipkin (1984b) propose heuristic solutions for such a problem based on a stochastic dynamic programming formulation. The disadvantage of their formulations lies in the very large state space needed for its solution. For serial supply chains, Shang and Song (2003) provide single period newsvendor problems that bound the optimal stocking levels determined by the Clark and Scarf (1960) serial supply chain model. Newsvendor bounds have a number of valuable qualities; they are considerably less computationally intensive, allow for ready parametric analysis, and facilitate the development of intuition. In this paper, we extend the newsvendor bounds technique to distribution systems, thus providing a simple and surprisingly accurate heuristic. Through a simulation study, we show that our heuristic significantly outperforms other common heuristics over a wide range of parameter values. The closed form solutions provided by the newsvendor bounds also allow insights into the system behavior of a distribution network that was not previously possible through alternative solution techniques.