Managing single echelon inventories through demand aggregation and the feasibility of a correlation matrix
Introduction
The mission of supply chain management is to coordinate the activities of all relevant entities for the purpose of providing goods and services to the end customer in an efficient and effective manner. One function that receives considerable attention in this alignment is the management of inventory. A typical goal is to increase supply chain velocity by minimizing total inventories within the supply chain, while maintaining an appropriate level of customer service. By doing so, products flow faster from raw materials sources to final consumers, leading to lower holding costs, less obsolescence, and reduced storage-space requirements, among other things.
Even in a fully functional supply chain, however, there is justification for holding inventory. No matter how good the demand forecasting system being used is, a firm usually cannot predict end customer orders with complete accuracy. Assuming that the customer is unwilling to wait for the product to be manufactured, inventories must be maintained to meet demand as it occurs. In other situations, it simply may not be economically feasible to produce on a just-in-time basis. Even in the case of Internet-based companies like e-tailers, the expected response time to requests may be such that items must be produced prior to actual customer requirements. As a consequence, methods are needed for efficiently maintaining inventories in general and, of particular interest here, inventories (commonly known as safety stocks) used to meet unexpected customer requirements.
In Tyagi and Das [1], a model and solution procedure for grouping customers in order to minimize total resource requirements such as safety stocks was proposed. Their work stemmed from the notion that demand correlations can be optimally combined in such a way that partial pooling of customers into multiple groups could, in some cases, result in fewer resource requirements than complete pooling of customers into one group. This paper, while not contesting their solution technique, objects to their fundamental contention that partial aggregation of customers can be better than complete aggregation based on demand correlations alone and argues that complete aggregation of demand is always as good as, if not better than, partial aggregation on the basis of minimized total resources. Further, this paper will establish that the results from Tyagi and Das [1] claiming to show partial aggregation to be better were based on mathematically inconsistent correlation matrices. To do so, this paper will demonstrate the existence of feasible and infeasible correlation structures and describe methods that allow researchers to check the validity of a correlation matrix for both numerical and simulation purposes. But before addressing these issues, the notion of pooling and its relationship to inventory management will be explained first.
Section snippets
Statistical economies of scale
Within supply chains, the principle of postponement relies heavily on statistical economies of scale, a term originally coined by Eppen and Schrage [2] and defined as “advantages that result from the pooling of uncertainty [3, p. 51].” Postponement recognizes that when changes in the form, identity, and/or location of a product are delayed, savings accrue because demand is easier to predict [4]. Not only does pushing inventory back up the supply chain lead to lower product valuations since
Partial pooling is never better
To understand why complete pooling can never do worse than partial pooling, consider the following as it relates to physical aggregation. Individual customers are initially classified into n distinct markets on some basis relevant to the firm (say, geographic area). As a starting point, each of these markets is assigned to a different stock-keeping location. Upon pooling, some or all of these markets are coupled together. Thus, pooling can range from one market per location (representing the
Results based on inconsistent correlation matrices
Having shown that partial pooling cannot do better than complete pooling, a logical question arises. How were Tyagi and Das [1] able to claim that partial aggregation was preferred in some cases? In particular, of the four different correlation matrices examined , and R4), they found that partial aggregation dominated in three cases , and R3) while complete aggregation dominated in only one (R4). The source of these contrary findings stems not from their proposed solution
Conclusion
On the topic of pooling, it was shown that complete aggregation never performs worse than partial aggregation on the sole basis of demand correlation. There may, in fact, be cases where partial aggregation of inventory is preferred, but that entails the consideration of other factors not accounted for here, such as lead times or transportation costs. For example, inventory centralization at the wrong location could necessitate higher safety stocks to cover possible increases in average lead
Acknowledgements
Sections of this paper reviewing the risk-pooling literature were presented at the Workshop on Supply Chain Management Practice and Research: Status and Future Directions, in Rockville, MD, April 2001.
Kefeng Xu is an Assistant Professor of Operations Management at the University of Texas at San Antonio. He received his Ph.D. in operations management and logistics/transportation from the University of Maryland. His publications have appeared in journals such as Transportation Research, Journal of Business Logistics, Journal of Transportation Economics and Policy, and International Journal of Physical Distribution and Logistics Management. His research interests are in inventory management;
References (22)
- et al.
Grouping customers for better allocation of resources to serve correlated demands
Computers and Operations Research
(1999) The rediscovery of postponement: a literature review and directions for research
Journal of Operations Management
(2001)Pooling in multi-location periodic inventory distribution systems
Omega
(1999)- et al.
Role of inventory and transportation costs in determining the optimal degree of centralization
Transportation Research E: Logistics and Transportation Review
(1997) - et al.
Centralized ordering policies in a multi-warehouse system with lead times and random demand
The occurrence of statistical economies of scale in intermodal transportation
Transportation Journal
(1994)Marketing behavior and executive action
(1957)Postponement, speculation and the structure of distribution channels
Journal of Marketing Research
(1965)- et al.
Planning physical distribution with the principle of postponement
Journal of Business Logistics
(1988) - et al.
Risk pooling in a two-period, two-echelon inventory stocking and allocation problem
Naval Research Logistics
(1989)
A model for assessing the value of warehouse risk-pooling: risk-pooling over outside-supplier leadtimes
Management Science
Cited by (15)
On the performance of overlapping and non-overlapping temporal demand aggregation approaches
2016, International Journal of Production EconomicsCitation Excerpt :Note that there is another aggregation approach discussed in the literature and often applied in practice that may also lead to the reduction of demand uncertainty and the improvement of stock control performance. This approach is referred to as cross-sectional aggregation, which involves aggregating different time series in order to improve performance across a group of items (Kefeng and Philip, 2003; Zhang and Burke, 2011; Rostami-Tabar et al., 2015). With regards to temporal aggregation, there are two different types of aggregation: non-overlapping and overlapping.
Strategic placement of telemetry units considering customer usage correlation
2019, EURO Journal on Transportation and LogisticsNetworked Base-Stock Inventory Control in Complex Distribution Systems
2019, Mathematical Problems in EngineeringIntelligent Support for Resource Distribution in Logistic Networks Using Continuous-Domain Genetic Algorithms
2018, Proceedings of the 2018 IEEE 2nd International Conference on Data Stream Mining and Processing, DSMP 2018
Kefeng Xu is an Assistant Professor of Operations Management at the University of Texas at San Antonio. He received his Ph.D. in operations management and logistics/transportation from the University of Maryland. His publications have appeared in journals such as Transportation Research, Journal of Business Logistics, Journal of Transportation Economics and Policy, and International Journal of Physical Distribution and Logistics Management. His research interests are in inventory management; logistics, service and manufacturing strategy; supply chain modeling; transportation policy, service, cost and demand studies.
Philip T. Evers is an Associate Professor of Logistics Management at the University of Maryland. He received his Ph.D. in logistics management from the University of Minnesota. His research interests include inventory management, transportation operations management, and intermodal transportation issues. He has published numerous articles in journals such as Transportation Research, Journal of Business Logistics, Transportation Journal, and Production and Inventory Management Journal.