Improving supply chain management with statistical quality methods

Supply chain management has become critical for global companies to ensure profitable growth under competition. An important supply chain research problem is the bullwhip effect caused by information distortion and variation amplification along a supply chain, which can lead to tremendous inefficiencies, such as excessive inventory investment and lost revenues. This thesis establishes the fundamental connection between the well developed statistical quality improvement methods and the bullwhip effect problem. The bullwhip effect is also studied in view of the application of forecasting models, statistical model identification, and parameter estimation. Engineering process control method is one of statistical quality control methods. It motivates the construction of a class of order-up-to policies and the developing of a nearly optimal policy developed to reduce the bullwhip effect. The proposed policy can significantly reduce the order variance while keeping the expected cost nearly optimal. According to numerical studies, the order variance of the nearly optimal policy can be reduced by more than 50% while the expected cost is only slightly greater than that of the optimal policy derived in Lee, Padmanabhan and Whang (1997). Since forecasting models play an important role in the inventory control, Exponentially Weighted Moving Average (EWMA) forecasting model and the Minimum Mean Square Error (MMSE) forecasting model based on an Autoregressive (AR), which are commonly used, are analyzed so that the bullwhip effect can be reduced by suitable design of a forecasting model. The analysis shows that (a) the EWMA forecast is robust to an AR(l) demand within a wide range; (b) the EWMA forecast contributes the bullwhip effect to the supply chain; (c) for an AR(l) demand, suitable forecasting models may be designed to trade-off the expected cost and order variance in a supply chain; (d) for an Integrated Moving Average (IMA) demand, if the IMA demand is mis-identified as a stationary AR(l) demand, the forecasting model based on the AR(l) model is applied. This kind of model mis-identification generates very high inventory cost and contributes to significant bullwhip effect. In order to reduce the bullwhip effect, the measure of the process variability should be defined to explicitly express inventory loss. In a constant lead time supply chain, it is shown that the supplier's loss depends on the uncertainty of the lead time demand and the mechanism of the supplier's order policy. In order to measure and reduce the variability of the demand process to a supplier, a two-stage supply chain model is constructed. Linear Gaussian state space models are used to describe the demand process to provide a clear measure of the process variability. A kind of optimal forecasting model is put forward to reduce the variability of the order process to the upstream supplier. In the phase of demand model identification and parameter estimation, smoothness priors Bayesian modeling is applied to reduce the order variability in the retailer-supplier relationship supply chain. Compared to general demand Bayesian modeling in which smoothness priors is not considered, the proposed demand model estimation considers the smoothing factor of the retailer's order process. Based on this demand model obtained by the smoothness priors Bayesian modeling, the variability of the optimal order to the supplier can be smaller than that of the optimal order based on the demand model obtained by general Bayesian modeling