Refining Underlying Demand Functions Using Information from Related Time Series
Glenn E. Maples, Ronald B. Heady, Zhiwei Zhu · Academy of Information and Management Sciences journal · 2008
ABSTRACT This research focused on improving estimations of demand functions using information from a large number of related many-period time series. The key premise is that by improving the data's internal consistency random variation will be reduced in the component time series. The research starts with the assumption that some of the demand time series are significantly correlated, some only slightly correlated, and some completely uncorrelated. A process is then developed for improving the demand function of any distinct part by using the correlated demands of the other parts. Metrics of effectiveness are proposed and used to test the methodology. The resulting tests demonstrate the effectiveness of the method, referred to as an internally consistent correlations (ICC) method. Importantly, the research makes no of a clustering algorithm and no assumption about the end use of the data set. Thus, it is expected that the results will apply to a wide array of domains, including forecasting, data mining, signal processing, and process monitoring. KEY WORDS: Preprocessing Time Series Data, Reducing Time series Noise, Inventory Control, Demand Forecasting, Data Mining, Internally Consistent Correlations. (ProQuest: ... denotes formulae omitted.) INTRODUCTION Modern organizations are using information technology to form new business models by tying customers (CRM), suppliers (SCM), and processes (ERP) to traditional business practices. This expansion of business models offers manifest benefits, including increased efficiencies and opportunities to develop new relationships. However, these advantages come at the price of increased complexity in the decision and operational domains. Inventory processes are among those most affected by these changes. Alternatives for sourcing, new production methodologies, the increasing of remanufacturing-all add to the challenges of managing inventory. Efficient administration of inventory, aimed at both improved availability and cost reduction, has become critical to the success of many organizations (Timme, 2003). One common way to attempt to reduce the complexity of managing inventory is using assemblies. Rather than stocking elemental parts, manufacturers may inventory pre-assembled subcomponents. These assemblies, the basis for modular manufacturing, confer benefits of process flexibility, decreased manufacturing time, increased customization, and adaptive design. Modularization has become an important thread in recent research. At the firm level, modularization requires different managerial competencies and perhaps requires those people charged with determining standards play a more important role in companies (Baldwin and Clark, 1997). From a strategy perspective, Montreuil and Poulin (2005) have coined the term varietizing to combinatorial strategy of personalization of products requiring the standardization of product design interfaces and modularity. The of assemblies permits the ability to postpone activities in the manufacturing process, which is a key point in mass customization (Piller, Moeslein, and Stotko, 2004; Su et al, 2005). Return policies, which can be critical to retail competitiveness (Yao, Wu, and Lai, 2005), can be impacted by modularization, as it is a primary determinant of the reuse economics of returned parts. This paper seeks to add to this stream of research in the area of inventory planning and control. The ultimate goal of this paper is to demonstrate an improved method for determining demand for individual parts by relying on information derived from the demand matrix for all parts. It begins by introducing a conceptual model that shows how important and even unexpected correlations among the demands for parts can occur in a variety of normal manufacturing environments. Then it proposes a process to these inter-part correlations to extract heretofore-unused information from the historical time series demand data matrix to identify a more correct demand curve of parts. …