| Supply chain modeling has posed many challenging and longstanding optimization problems, and the need for novel methods of modeling inherently complex supply chains is great. A major complicating factor in developing such methods is the uncertainty involved in supply chains. In this dissertation, we develop a novel approach to modeling a realistically sized supply chain that operates under uncertainty. Our iterative approach combines optimization and simulation methodologies in order to (a) obtain optimal supply chain plans via mathematical modeling and (b) incorporate uncertainty in the execution of these plans via simulation. Our modeling approach provides the basis for developing flexible decision support systems that can be used to analyze various trade-offs in the management of supply chains.; First, we evaluated the relative impact of demand, lead-time, and supply uncertainties on supply chain performance. Four scenarios were used to assess the incremental impact of these uncertainties on supply chain cost and customer service performance. Results showed that demand uncertainty was the most detrimental to supply chain performance, thus underscoring the importance of effective forecasting.; Second, we compared two methods of selecting a forecast method: one based on supply chain performance and the other based on traditional accuracy measures. In brief, we evaluated the performance of four well-established forecasting methods (i.e., simple exponential smoothing, additive trend, damped additive trend, and damped multiplicative trend) in terms of both forecast accuracy measures and supply chain performance. Comparison of the forecasting models in terms of supply chain performance involved running the supply chain models under conditions of demand, lead-time, and supply uncertainty and observing the resulting effects on cost and customer service performance. In addition to using ANOVA and Tukey's test in the analysis of the results, we developed tradeoff curves to allow the investment-based evaluation of forecasting methods. The tradeoff curves provided additional insights into the performance of forecasting methods by incorporating cost and customer service performance simultaneously. The results showed that comparison of the forecasting models in terms of supply chain performance captured the differences among the forecasting methods much better than the traditional accuracy measure-based approach did. Among the four forecasting methods, the damped additive trend method was by far the best in terms of both cost and customer service performance, followed in descending order by additive trend, simple exponential smoothing, and damped multiplicative trend methods, respectively.; Third, we evaluated the forecasting methods on the basis of supply chain performance by simulating the maintenance of varying levels of customer service at various levels of safety stock. Before starting, however, we compared the bootstrapping and traditional analytical approaches to determining safety stock quantities in order to determine which method to use in our supply chain model. Since our comparison revealed no significant differences between the two approaches, we opted for using the bootstrapping method.; Together, our findings suggest a complex relationship between forecast accuracy, supply chain cost, and customer service performance and warrant further research in several directions including supply chain modeling and forecasting. |
