On the generalizability of MRP simulation results

Abstract

Much of the reported production and inventory control research of the last two decades has addressed the relative performance of decision rules for specific PIC problems. The usual procedure involves testing the decision rules in simulated environments. Highly simplified simulation models are normally utilized because of the size and computational burden of additional realism. Although the results of such experiments can not be legitimately generalized to any substantial degree, the nature of the citation process is such that results are emphasized far more than the accompanying qualifications. The danger is that there may exist unknown interactions between the decision areas under study and those accepted as given in the original experiments. Certain features must exist simultaneously in simulated Material Requirement Planning environments to facilitate generalizability of the results. This paper identifies those features and describes the development of a simulation model which incorporates them. Results of simulation experiments with the model are used to demonstrate how seemingly straightforward results must be reinterpreted as additional experimental factors are introduced. Specifically, simulation runs were used to establish the effects of lot sizing and firm planned orders on five different measures of performance under specified conditions. Additional repetitions of the simulation runs under different conditions yielded comparable performance data sets. Comparison of the performance of selected lot sizing rules and firm planned order horizons, as additional experimental factors were introduced, established the importance of interactions in determining system performance.