Date of Graduation
Doctor of Philosophy in Business Administration (PhD)
Supply Chain Management
Social sciences; Applied sciences; Collaboration and integration; Data science; Demand planning; Forecasting; Information sharing; Supply chain management
Data-driven decisions have become an important aspect of supply chain management. Demand planners are tasked with analyzing volumes of data that are being collected at a torrential pace from myriad sources in order to translate them into actionable business intelligence. In particular, demand volatilities and planning are vital for effective and efficient decisions. Yet, the accuracy of these metrics is dependent on the proper specification and parameterization of models and measurements. Thus, demand planners need to step away from a "black box" approach to supply chain data science. Utilizing paired weekly point-of-sale (POS) and order data collected at retail distribution centers, this dissertation attempts to resolve three conflicts in supply chain data science. First, a hierarchical linear model is used to empirically investigate the conflicting observation of the magnitude and prevalence of demand distortion in supply chains. Results corroborate with the theoretical literature and find that data aggregation obscure the true underlying magnitude of demand distortion while seasonality dampens it. Second, a quasi-experiment in forecasting is performed to analyze the effect of temporal aggregation on forecast accuracy using two different sources of demand signals. Results suggest that while temporal aggregation can be used to mitigate demand distortion's harmful effect on forecast accuracy in lieu of shared downstream demand signal, its overall effect is governed by the autocorrelation factor of the forecast input. Lastly, a demand forecast competition is used to investigate the complex interaction among demand distortion, signal and characteristics on seasonal forecasting model selection as well as accuracy. The third essay finds that demand distortion and demand characteristics are important drivers for both signal and model selection. In particular, contrary to conventional wisdom, the multiplicative seasonal model is often outperformed by the additive model. Altogether, this dissertation advances both theory and practice in data science in supply chain management by peeking into the "black box" to identify several levers that managers may control to improve demand planning. Having greater awareness over model and parameter specifications offers greater control over their influence on statistical outcomes and data-driven decisions
Jin, Yao, "Data Science in Supply Chain Management: Data-Related Influences on Demand Planning" (2013). Theses and Dissertations. 898.