Web-based simulation portal for investigating impacts of ... - CiteSeerX

Web-based simulation portal for investigating impacts of sharing production information on supply chain dynamics from the perspective of inventory allocation Jason S.K. Lau Department of Industrial and Manufacturing Systems Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong George Q. Huang Department of Industrial and Manufacturing Systems Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong K.L. Mak Department of Industrial and Manufacturing Systems Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong

Keywords Supply chain, Performance, Information , Inventory control

Abstract Informatio n sharing and coordinatio n between buyer and vendor have been considered as useful strategies to improve supply chain performance . The debate is about what informatio n to share and how to share most costeffectively to maximize the mutual benefits of the supply chain as a whole and the individual busines s players. Proposes a systematic framework for investigatin g the impacts of sharing production informatio n on the supply chain dynamic performance . This framework supports supply chain researche s to study impacts of informatio n sharing under various scenarios. Examines , under the framework, an inventory allocatio n problem in an arborescen t distributio n supply chain with two distributio n channels competing for the same source of supply. Finds that the levels of benefits by sharing informatio n vary with different players involved in the supply chain. Suggests some guidelines to balance the benefits in a supply chain in order to motivate informatio n sharing.

Received August 2001 Revised February 2002 Accepted February 2002

Integrated Manufacturing Systems 13/5 [2002 ] 345±358 # MCB UP Limited [ISSN 0957-6061] [DOI 10.1108 /0957606021042980 1

Introduction Supply chain dynamics has been studied for more than three decades. Since Forrester (1961) discovered the fluctuation and amplification of demand from the downstream to upstream of the supply chain, there has been a lot of literature analysing this phenomenon (e.g. Towill, 1991; Wikner et al., 1991; Towill et al., 1992; Wu and Meixell, 1998; Helo, 2000). This effect can be readily illustrated in the well-known ``beer game’’ in which one can observe the amplification of demand signal and fluctuation of inventory level along a supply chain which consists of customer, retailer, wholesaler, distributor and factory (Sterman, 1989; Simchi-Levi et al., 2000). This effect is obviously undesirable as it exacerbates the supply chain costs (e.g. stock holding, backlog, late delivery, under/over resource utilization etc.). The source of such fluctuation and amplification of order and inventory is mainly due to the lack of sharing of production information between enterprises in the supply chain. (e.g. Lee et al., 1997; Simchi-Levi et al., 2000). These factors lead to distortion of actual demand information and cause unnecessary wastes. Lee et al. (1997) have studied this phenomenon extensively and termed it as ``bullwhip effect’’. From the studies of bullwhip effect (e.g. Lee et al., 1997; Metters, 1997), one of the remedies is to share information along the supply chain. It has been reported that the benefit of information sharing is significant, especially in reducing the bullwhip effect (e.g. Lee et al., 1997; Cachon and Fisher, 2000; Lee at al., 2000) The current issue and full text archive of this journal is available at http://www.emeraldinsight.com/0957-6061.htm

and supply chain costs (e.g. Swaminathan et al., 1997; Gavirneni et al., 1996; Tan, 1999). By using the shared information, each supply chain entity can make better decisions on ordering, capacity allocation and production/material planning so that the supply chain dynamics can be optimized. Information sharing, however, may not be beneficial to some supply chain entities due to high adoption cost of joining the interorganizational information system, unreliable and imprecise information (e.g. Swaminathan et al., 1997; Cohen, 2000), and different operational condition of each firm (e.g. Dong and Xu, 2001). Zhao and Xie (2002) have recently found that, by using simulation study, sharing information may hurt some supply chain members under most conditions. National Research Council (2000) reported that there are some barriers (e.g. expensive technology investment, personnel training, lack of mutual trust etc.) which hinder small and medium sized enterprises (SMEs). Singer (1999) also reported that, based on an industrial case, sharing information may lead to loss of business. It is, therefore, necessary to investigate the benefits of information sharing that can be gained by each firm and the whole supply chain before implementing. Regarding technology of implementing information sharing, electronic data interchange (EDI) has been employed as a major tool of information sharing for many years (e.g. Davis and O’Sullivan, 1998; Strader et al., 1998; Lee at al., 2000; Bhatt, 2001; Furst and Schmidt, 2001; Warkentin et al., 2001). As Internet and e-commerce technology continue to evolve, there has been much literature studying how such technology can improve supply chain performance, especially on information

[ 345 ]

Jason S.K. Lau, George Q. Huang, and K.L. Mak Web-based simulation portal for investigating impacts of sharing production information on supply chain dynamics from the perspective of inventory allocation Integrated Manufacturing Systems 13/5 [2002] 345±358

sharing (e.g. Davis and O’Sullivan, 1998; Strader et al., 1998; Conway, 2000; Graham and Hardaker, 2000; Croom, 2001; Kehoe and Boughton, 2001; Warkentin et al., 2001). Given a wide spectrum of information technologies (e.g. Internet, extensible mark-up language (XML), common object request broker architecture (CORBA)), it is unclear which technology is most suitable for enabling the share of production information in the supply chain. The above brief review of the related literature has highlighted the necessity and significance of sharing production information in order to remedy the so-called bullwhip effect and its associated impacts. However, the literature generally does not address adequately what exactly production information the enterprises in the supply chain should share, and to what extent they should share. Having recognized these limitations in the literature, a research project has been initiated to address these research questions. The overall objective of this project is to develop a framework for investigating the impacts of sharing production information on the supply chain dynamic performance. The proposed framework will be used to carry out a series of simulations from several perspectives such as resource allocation, material requirement planning and rescheduling. Techniques of experimental design will be applied to analyse the results to derive insightful findings. Guidelines will be established to assist industrialists to set up efficient and effective supply chain management systems. The research will take full advantage of the emerging information and telecommunication technologies, and take into consideration industrial needs. This paper reports on a Web-based implementation of a conceptual framework proposed from this research project (Huang et al., 2001) and some early findings from initial investigations.

Web-based simulation portal A conceptual systematic framework has been developed from our early research for investigating the impacts of sharing production information on the supply chain dynamics (Huang et al., 2001). Based on this proposed framework, further development has been accomplished to establish a simulation portal on the World Wide Web (Web or WWW). The resulting prototype simulation portal is diagrammatically shown in Figure 1.

[ 346 ]

Generally speaking, this simulation portal provides a suite of online facilities in a step-by-step fashion for analysts to design and conduct simulation experiments and afterwards to analyse the results and draw implications. The key steps of using the simulation portal follow those of the conceptual framework. They will be explained in detail in the subsequent section one by one. The purposes are threefold: 1 to serve as the test-bed for implementing and executing the experimental simulations; 2 to provide a Web-based test-bed or portal open to the entire research community to share the experimental data and results; and 3 to provide some practical insights to industrial practitioners regarding how the production information can be made shareable between enterprises in the supply chain with the support of the Internet and Web technology. The development and implementation of a Web site for just carrying out the experimental simulations specifically designed in this project is relatively straightforward. Web pages are designed to follow the format of worksheets used at different stages. Input and output data are stored into and retrieved from the backend database. A typical three-tiered architecture of Web applications is adopted in an active server pages (ASP) programming environment. Java/VB (visual basic) scripts have been extensively used, in addition to SQL (structured query language) for database programming. However, significant challenges are expected if the site is intended for the second and third purposes. On the one hand, this Web site is not intended to become a Web-based general-purpose simulation system. On the other hand, some degree of customisation should be provided so that the users are able to change the settings of experimental simulations. A set of supply chain structures, typical decision levels, the production information hierarchy, the supply chain dynamic index hierarchy, supply chain dynamic models will be parametrically pre-defined and thereafter maintained in the backend database. Facilities will be provided for customizing the parameters. The building up of the model of a subject supply chain structure and the definition of simulation logics in the simulation model are some functions generally available from commercial simulation packages. Therefore, this research would avoid repeating the efforts.

Jason S.K. Lau, George Q. Huang, and K.L. Mak Web-based simulation portal for investigating impacts of sharing production information on supply chain dynamics from the perspective of inventory allocation

Figure 1 Specify supply chain structure

Integrated Manufacturing Systems 13/5 [2002] 345±358

Instead, emphasis will be placed upon the modeling itself. As a result, the simulation portal would not be able to act in a self-contained fashion. However, investigations are being carried out to devise new methods of modeling the supply chain structures and their dynamics. In this respect, it would become worthwhile to implement these new models as online Web-based systems. In addition, this simulation portal should also reflect the potential methods of exchanging production information between the supply chain players. The concept of agents will be introduced to represent players or echelons in the supply chain. The production information and supply chain dynamic performance measurements will be captured by the properties of the agents. The data exchange and communication between the echelons become the flow of data between the corresponding agents in the form of messages, either directly between them or through a commonly shared area called blackboard. The technologies (e.g. agents, blackboard, message passing, workflow management ± dataflow management) used for implementing this simulation portal will provide some insights to the industrial practitioners on how they can take advantage

of the technologies to share production information with each other. Once opened for the research community to share, this portal will prove an invaluable source of resources. Experimental data can be reused in different simulations where items of production information and their value levels are changed, or where a different supply chain structure is used at another decision level. This would maintain a high degree of integrity and consistency among the findings from these simulations by different researchers, or allow the researchers to verify their findings with each other.

Case study The case study to be presented in this section has two purposes. The first purpose is to demonstrate the procedure of applying the simulation portal step by step. The second purpose is of course to investigate the effects of sharing production information on the supply chain dynamics from an inventory allocation perspective. This inventory allocation problem is based on an arborescent distribution supply chain in which all companies are independent of each other. There are two distribution channels in

[ 347 ]


this supply chain. Each retailer is supplied by one distributor and both distributors are supplied by a single manufacturer which operates on a make-to-order basis with limited production capacity. The details of this case study will unfold with the progress within the simulation portal from one step to the next, as shown in the following subsections.

Step 1: defining objectives of study The purpose of this study is to analyse the impacts of different levels of information sharing on operating costs of the whole supply chain and each individual company under various supply chain scenarios. Demand and inventory information are considered for sharing among companies in a supply chain. The mechanism of different levels of information sharing will be discussed in Step 4. Operating costs of the supply chain and each company will be defined in Step 5. The results can be used as a guideline for a company which is considering information sharing.

Step 2: selecting/specifying decision level There are basically three levels of decision in a supply chain: strategic, tactical and operational. Huang et al. (2001) discussed the three levels related to this project. The decisions considered in this paper are inventory allocation and replenishment, which belong to operational level.

Step 3: selecting/defining supply chain structure A supply chain structure must be defined for the experimental simulation. At present, only four basic SC structure templates are pre-defined in the database, as shown in Figure 1, and the analyst needs to select one of them. More sophisticated facilities should be provided in this simulation portal for defining more complicated SC structures and models that are more relevant and realistic. A supply chain with two distribution channels is selected in this case study. Each channel consists of one retailer and one distributor. Each retailer, R1 and R2, is supplied by only one distributor, D1 and D2. Both distributors are supplied by a single manufacturer M. This represents a typical distribution supply chain in the real world in which each distribution channel serves a region of the market.

Step 4: specifying production information model (PIM) This step is essential. It defines the factors that exactly will be studied in the experimental simulation, and the modes in

[ 348 ]

which these factors are shared between players. The use of the simulation portal in this step is straightforward, just a matter of filling in forms in relevant Web pages, as shown in Figure 2. The analyst must be clear about the contents being entered into the form.

Step 4.1: selecting information sharing mode Since level of information sharing is the main focus of this paper, it should be clarified first before introducing other production information used in this case study. Four modes of information sharing are identified in this study to represent four levels of sharing, as indicated in Figure 3. The dotted line indicates information flow and the solid line indicates material flow. The first mode (BC) represents the traditional information flow in the supply chain in which each company shares information based on its order (Figure 3a). An order-up-to (s, S) installation stock policy is used by retailers and distributors to replenish their inventory. This policy specifies that when the inventory position IP, which is sum of on-hand and on-order stock, falls below the reorder point s, an order with quantity Q is placed to the supplier in order to raise the inventory position to S. The (s, S) policy of each retailer and distributor is defined by equations (1)-(4). Equation (1) specifies the reorder point for average demand per period, standard deviation of demand , supply lead time L and safety factor . The optimal value of is given by equation (3), which is the solution for standard newsvendor problem (see Silver et al., 1998), where k is ordering cost, b is backlog cost and h is holding cost. The optimal value of Q is given by equation (2) which is the standard EOQ solution, as Axsater (1996) suggested that the standard EOQ solution is a good heuristic in a stochastic environment. It is, therefore, convenient to express S as equation (4): p sˆ L‡ L …1† Qˆ

r 2 k…b ‡ h† bh

ˆ ©¡1

³

b b‡h

S ˆ s ‡ Q:

´

…2† …3† …4†

In the second mode (RD) the retailer shares its demand (i.e. market demand) and inventory information with its distributor (Figure 3b). The retailer places the order according to its own policy while the distributor, by taking


Figure 2 Specify parameters and sharing modes


Figure 3 Four modes of information sharing

the retailer’s demand and inventory information into account, orders from the manufacturer by using a modified (s0 S0 ) policy which is specified by equations (5) and (6) (S0 is obtained as equation (4)). Note that subscript r and d indicate retailer and distributor information respectively:

s0d ˆ Qd0

r …Ld

‡ Lr † ‡

r

p …Ld ‡ Lr †

s 2 r …kd ‡ kr †…bd ‡ hd † ˆ : b d hd

…5† …6†

In the third mode (DM) the distributor shares its demand and inventory information with

[ 349 ]


the manufacturer (Figure 3c). The manufacturer determines the order size by considering the setup cost of production when the inventory position of distributor falls below its reorder point according to equation (1). The modified order size is given by equation (7) where km is production setup cost: s 2 d …kd ‡ km †…bd ‡ hd † 00 Qm ˆ : …7† b d hd The fourth mode (RDM) is the combination of RD and DM modes. It can be regarded as an integration of retailer, distributor and manufacturer. The retailer shares its demand and inventory information to distributor and manufacturer (Figure 3d). The manufacturer uses market demand and inventory information of retailer and distributor to determine the delivery quantity, which is given by equation (8) when the echelon inventory falls below the reorder specified by equation (5): Q00m

s 2 r …kd ‡ km ‡ kr †…bd ‡ br ‡ hd ‡ hr † ˆ …8† …bd ‡ br †…hd ‡ hr †

Under each sharing mode the manufacturer operates in a make-to-order basis and employs a priority rule heuristic to allocate capacity to distributor order. The distributor orders are sequenced first. Production capacity is then allocated to each order according to the sequence. In this case study order size is used to prioritize the distributor orders. Each unit of capacity is occupied for the specified lead time in order to produce one unit of product.

Step 4.2: selecting production information for impact analysis Five types of information are selected in PIM for further study. They are demand variance, holding cost, backlog cost, ordering cost, production setup cost and production capacity. Demand variance indicates the fluctuation of consumer demand at retailer level. It is well known that demand variance is one of the sources of bullwhip effect. Investigating different levels of demand variance is important to enhance our understanding of the values of information sharing. Holding cost is incurred per unit stock in on-hand inventory and on-order inventory. Backlog cost is incurred for each unit short of inventory per period. This cost only applies for retailer and distributor. Ordering cost is incurred when a company places an order. This cost may include administrative cost of ordering and transportation cost. Studying impacts of cost parameters is important because the performance index in this study is operating cost. Moreover, the cost

[ 350 ]

parameters affect the local optimal policy of each company, as presented in equations (1)-(7). Production setup cost is incurred for each order received by the manufacturer when production starts. Production capacity, CP, means the number of products that can be produced concurrently during the production lead time. It is also an influential factor of supply chain performance. Evaluation of performance of capacitated supply chain by using analytical approach (e.g. Gavirneni et al., 1996; Cachon and Fisher, 2000) is, however, more difficult than simulation approach. Therefore it is appealing to include this factor in this study. Figure 5 shows a summary of PIMs that are being studied by using the portal. The analyst can select other information for in-depth study by clicking the check box beside the information node. After selecting information, the analyst needs to input levels of value for each selected information in Step 7.

Step 4.3: specifying values for other parameters Other parameters (e.g. market demand, transport and production lead time) of each firm can be specified consequently. In addition to mean values, variances of some parameters can also be specified in the portal. The values are summarised in Table I.

Step 5: selecting dynamic performance index (DIM) Different performance indicators or indexes must be specified for each firm as well as the entire supply chain in order to assess the impacts of information sharing. Cost is the main performance index considered in this study. The average operating cost of retailers and distributors includes fixed ordering, inventory holding cost and backlog cost. It is defined by: Á ! T X Cˆ …hIPt ‡ bBLt ‡ kmax…Qt ; 0†† =T: …9† tˆTs

IPt is inventory position and BLt is backlog at the end of period t. T is the length of simulation and Ts is the warm-up period. Note that the subscript is omitted as equation (9) applies for both retailers and distributors. The average operating cost of manufacturer consists of production setup cost and process cost. It is defined by: Á ! T X X Cm ˆ …km max…Qdt ; 0† ‡ gQdt † =T: …10† tˆTs

d

Qdt is production size for distributor d at period t and g is process cost per unit. The


Table I Mean values and variances used in simulation Information Demand (units) Production lead time (periods) Transportation lead time between manufacturer and distributor (periods) Transportation lead time between distributor and retailer (periods) total supply chain cost is the sum of operating cost of each firm. The analyst can select total cost of each firm and cost component (e.g. backlog cost) for detail analysis (Figure 4).

Mean

Standard deviation

10 3 5 2

Specified in Step 7 0 1 1

The following are assumed in this simulation model: each retailer faces the same demand

Step 6: establishing simulation model Conceptually, a mathematical model, which represents the operations in a supply chain, is needed to map PIM to DIM. A multi-agent approach is employed to model the supply chain. This multi-agent model provides a practical foundation of developing a distributed decision support system for the whole supply chain. Discussion of multi-agent modeling is out of the scope of this paper. The reader may refer to Swaminathan et al. (1997) and Parunak et al. (1999) for successful application of multiagent modeling in addressing supply chain problems.

distribution; production and transportation setup time do not exist; a setup cost which is independent of production quantity is incurred when a production is started; production lead time is deterministic; market demand and transportation lead time follow normal distribution; unfulfilled order is backlogged in retailers and distributors; supply of material for production is unlimited; holding cost of material in manufacturing site is negligible; production capacity level is constant.

Figure 4 Select indexes for performance measurement DIM

[ 351 ]


Step 7: designing experiments A total of 15 items of production information have been specified in this case study as the control factors (parameters). One factor is assigned to demand variance; 12 factors are assigned to holding, backlog and ordering cost of each retailer and distributor. The remaining two factors are assigned to production setup cost and capacity of manufacturer. All the 15 control factors have two levels of value. For all sharing modes, an orthogonal array L16(215) of the 15 control factors is constructed to find out the impact of these factors under each sharing mode. Advantages of using orthogonal array over full factorial combination to generate experiment set are discussed in Roy (1990). Figure 5 shows a summary of selected PIM and allows the analyst select the type of array. The analyst needs to specify the value of each level. A column number is assigned to control factors according to the sequence of selection. The analyst can change column assignment by moving the selected control factor up and down. The complete orthogonal array is generated as shown in Figure 6.

Step 8: running simulation After confirming the PIM, DIM and experiments, the investigator can run the simulation. Length of each simulation run is

Figure 5 Summary of selected PIM

[ 352 ]

600 periods. The results of first 100 periods are discarded in order to eliminate the effect of transient response. The number of replications of each experiment is 20. The analyst can also specify the simulation parameter of each experiment defined in Step 7, as shown in Figure 7.

Step 9: analysing dynamic performance measurements Once the results from the simulation experiments are obtained, the analyst can study the dynamic behaviour of each firm (as represented by an agent in the agent-based simulation model) by selecting different dynamic characteristics like inventory, demand, order size and backlog. The analyst may gain additional insights (e.g. fluctuation of inventory and order size) which cannot be obtained by using average performance measures (Towill et al., 1992; Parunak et al., 1998). Figure 7 shows the dynamics of inventory of one distributor.

Step 10: analysing impacts of PIM and sharing modes The effects of the sharing mode and the 15 control factors (items of production information) are calculated for each of their level, and presented in Figures 8 and 9


Figure 6 DOE orthogonal array


Figure 7 Dynamics of simulation results of the chosen DIM (costs of the total supply chain and individual agents)

[ 353 ]


respectively. Based on these charts, the analyst is able to draw some implications.

Step 10.1: analysing impacts of sharing modes Let us examine Figure 8 first. This chart is obtained by averaging the cost measures (i.e. DIM) of each experiment. It shows the varying impacts of different levels of information sharing mode on the total supply chain cost and the breakdown of cost performance of each agent (firm). It becomes apparent from the chart that the RD mode results in the highest total supply chain cost while the RDM mode results in the lowest. This is perhaps because the distributor orders less from the modified policy, leading to higher backlog in both retailer and distributor. The cost for the distributor is reduced due to reduction in inventory while the cost for retailers is increased due to increase in backlog. The smaller size of distributor orders leads to higher production cost for the manufacturer. In the DM mode, the inventory holding cost for distributors is higher due to a larger order size which is determined by the manufacturer. The manufacturer benefits because of the optimal production size. The cost for retailers in the DM mode is very close to that in the BC mode because sufficient supply from the distributor is guaranteed. But the overall cost is still higher than that in the BC mode. In the RDM mode extra backlog of distributors and retailers introduced by the RD mode is compensated by extra inventory due to larger order size in the DM mode. The overall result is that the manufacturer can produce in optimal batch size, the distributors reduce the inventory level and the retailers can maintain the service level. Hence the cost performance is superior to other modes.

Figure 8 Effects of sharing mode on the total supply chain and individual costs

[ 354 ]

Although the ordering policies are not global optimal as the reorder point and order size are determined `òptimally’’ from each local agent perspective, they can be easily implemented in a real supply chain. The investigator can test different feasible policies, instead of the local optimum one, in the simulation model as control factors to seek `òptimal’’ policy in a realistic sense. Since not all sharing modes (except BC) are feasible in a real situation (e.g. retailer is not willing to share information to manufacturer due to sensitivity of the information or lack of information technology), this system can support investigation of partial information sharing in which only one distribution channel is engaged in information sharing while the other still operates in traditional mode (i.e. BC). Furthermore, it is clear that from the discussion of this case study, not all firms in the supply chain benefit by information sharing. The investigator can introduce some financial incentives in the simulation model in order to balance the benefits among the companies. For example, different levels of price discount together with different levels of information sharing (i.e. sharing modes) may be tested in the model to find out which combination is `òptimal’’ from both a local and global perspective in the supply chain.

Step 10.2: analysing impacts of PIM control factors Next, let us move to discuss the interaction between the 15 PIM control factors and sharing modes. Table II shows the F-value which indicates the significance of the effect of each PIM control factor on the supply chain and individual operating costs. For analysing cost of individual firm, R1, D1 and M are selected. As distribution channels formed by R1, D1 and R2, D2 are symmetric to each other, the results of one channel is very similar to the other.


Figure 9 Pareto charts: Interaction effects of PIM control factors on total cost of supply chain, Retailer 1, Distributor 1 and Manufacturer


Table II ANOVA results (F-value) for the interaction effects of 15 control factors on total cost of supply chain, Retailer 1 (R1 cost), Distributor 1 (D1 cost) and Manufacturer (M cost) under different sharing modes PIM A B C D E F G H I J K L M N O

Supply chain cost Demand variance R1 Backlog cost R1 Holding cost R2 Backlog cost R2 Holding cost R1 Ordering cost D2 Holding cost D1 Backlog cost R2 Ordering cost Setup cost D2 Ordering cost Capacity D1 Holding cost D1 Ordering cost D2 Backlog cost

a

499.38 0.37 0.23 0.54 1.24 1.85b 174.34a 0.01 2.33b 275.63a 1.89b 65.80a 246.97a 1.95b 0.97

R1 cost a

6.39 2.20b 0.45 1.28 1.04 1.53 0.28 0.054 1.23 0.13 0.34 9.25a 6.18a 4.07a 0.01

D1 cost a

35.80 14.32a 109.21a 0.40 0.22 3.55a 1.14 1.05 0.84 53.38a 0.28 16.30a 37.90a 14.21a 0.80

M cost 9.75a 1.53 0.44 0.99 0.64 0.25 19.61a 0.09 1.50 1,132.47a 20.15a 42.88a 20.54a 25.75a 0.42

Notes: a PIM control factors with significance greater than 95 per cent b PIM control factors with significance ranging from 80 per cent to 95 per cent [ 355 ]


[ 356 ]

For each cost performance measure three sets of PIM are classified based on their F-value. Class I represents a set of PIM control factors (indicated by superscript `à’’ in Table II) of which the significance is greater than 95 per cent. PIM control factors (indicated by superscript ``b’’ in Table II) with significance ranging from 80 per cent to 95 per cent are classified in class II. The remaining insignificant factors are classified in class III. The PIM control factors are ranked by their F-values, from the largest to the smallest, in each operating cost measure in Figure 9. The class labels of PIM control factors are also shown below the horizontal axis. As can be seen from Figure 9, only a few factors, which significantly impact on supply chain performance, need to be addressed carefully. It is obvious that demand variance, holding costs, production setup cost and capacity, which belong to class I, are influential on operating costs. These factors are discussed in this section one by one. Class I. Demand variance significantly affects the benefits of sharing information. The stock-out risk and average backlog is higher under higher demand variance. This outweighs the cost saving due to information sharing in RD mode. Under DM and RDM mode, the operating cost of both retailers and distributors is lower that in RD mode. This is because average order size and on-hand inventory are increased under these modes. On the other hand, the manufacturer gains cost saving under RD, DM and RDM mode as the average order size from distributors is increased. Capacity is another influential factor on the benefits of information sharing. The manufacturer’s holding cost is higher under low capacity. As the production lead time of each order is longer, finished goods inventory of each order is higher. The longer order lead time increases the backlog cost and on-order cost of distributors and retailers. This actually reduces and even cancels out the cost saving caused by sharing information. It is obvious that DM mode is most beneficial to the manufacturer under high setup cost. However it is not the case for the distributor as it needs to hold more inventory. This explains the high interaction between setup cost and sharing modes with respect to distributor and manufacturer operating costs. Class II. Cost parameters of the retailer seem to have little interaction with sharing mode with respect to the retailer’s operating cost. The reason is that the local optimal policy, which adjusts the reorder point and

order size based on different cost parameters, employed by the retailer maintains a consistent performance between different sharing modes. The effect of interaction between holding cost of retailer and sharing modes on distributor operating cost is significant. Since higher retailer’s holding cost leads to smaller retailer order size, the average backlog of the distributor is reduced. The cost saving, when the retailer’s holding cost is high, in RD mode is more than that when the retailer’s holding cost is low. In DM mode, however, high on-hand inventory plus reduced shipment, which is due to high retailer’s holding cost, leads to higher distributor’s cost. This situation still occurs in RDM mode as the distributor needs to hold more on-hand inventory. Like the retailer’s holding cost, its backlog cost also shows significant interaction with sharing mode with respect to the distributor’s operating cost. Higher backlog cost increases the retailer’s order size. The distributor then holds fewer inventory in BC and RD mode. The distributor’s cost in DM mode is smaller when the retailer’s backlog cost is higher. This is because larger demand from the retailer compensates for some of the increased distributor’s inventory. The distributor’s cost performances under high and low retailer’s backlog cost in RDM mode are similar, as holding cost of distributor is reduced in this mode. Holding cost of distributor has high interaction with sharing modes with respect to distributor’s operating cost. As higher holding cost leads to smaller average order size, this increases the risk of backlog in RD and RDM mode and hence, the operating cost is higher. The increase in holding cost actually outweighs the cost saving of sharing information under these two modes. The interaction between distributor’s backlog cost and sharing modes is very small as the saving of reduced backlog, which is caused by higher unit backlog cost, cannot cover the increased holding cost (both on-hand and on-order) due to larger average order size. Hence the operating cost is higher when the backlog cost is higher under all sharing modes. Ordering cost of the distributor, like ordering cost of the retailer, does not have high interaction with sharing modes. The reason is that the high ordering cost compensates for the cost saving in RD and RDM mode. The interaction between distributor’s cost parameters and sharing modes with respect to retailer’s operating cost is similar to that of distributor’s operating cost. This is because the retailer performance is directly affected by the distributor.


Class III. There is little interaction effect of cost parameters of one distributor (R1) on another distributor (R2) under different sharing modes. This is because the capacity of the manufacturer can absorb change in average order size of the distributor due to change in cost parameters. The same argument can be used to explain the little interaction on the other retailer (i.e. R2) performance. The effect of sharing modes on distributor’s operating cost is less significant when retailer’s ordering cost is high. The reason is that high ordering cost outweighs the cost saving gained from information sharing in RD and RDM modes (i.e. from equations (6) and (8) respectively). Moreover, the higher order size from the retailer increases the holding cost of the distributor as indicated in BC and DM mode.

Conclusions In this paper we have analysed the impacts of information sharing on supply chain dynamics by using the proposed framework. A Web-based system is developed to facilitate the analysis. A case study on a distribution supply chain is used to illustrate this framework. The results indicate that sharing information may not be beneficial to all supply chain members. The effects of several factors (PIM) on operating cost are also identified in the case study. The following directions are identified for future research: More complicated supply chains (e.g. more distribution channels, more intermediate tiers and assembly structure) can be investigated by using the facilities provided by this portal. As there is imbalance of benefits due to information sharing, investigation of an incentive system for motivating information sharing in the supply chain is necessary. Different inventory control policies (e.g. periodic review) and capacity allocation policies can be analysed based on this framework. Optimal configuration (e.g. sharing mode, control parameters) of information sharing in a supply chain can be searched by using computational algorithm (e.g. genetic algorithm, simulated annealing). Further research is needed in developing such optimization algorithm based on the simulation model. Based on this Web-based system, a handbook for supply chain design/redesign will be established. Guidelines have played

significant roles in design decision-making process. In the field of product design, axiomatic design theory proposed by Suh at MIT has become widely accepted as a generic design framework (where guidelines are called axioms, theorems and corollaries). In the field of product design for manufacture and assembly, guidelines have also played essential roles. We will investigate the possibility of introducing the Web-based guideline system (Huang et al., 2001) for representing supply chain design guidelines.

References Axsater, S. (1996), `Ùsing the deterministic EOQ formula in stochastic inventory control’’, Management Science, Vol. 42 No. 6, pp. 830-4. Bhatt, G.D. (2001), ``Business process improvement through electronic data interchange (EDI) systems: an empirical study’’, Supply Chain Management: An International Journal, Vol. 6 No. 2, pp. 60-73. Cachon, G.P. and Fisher, M. (2000), ``Supply chain inventory management and the value of shared information’’, Management Science, Vol. 46 No. 8, pp. 1032-48. Cohen, S.L. (2000), `Àsymmetric information in vendor managed inventory systems’’, PhD thesis, Stanford University. Conway, D.G. (2000), ``Supplier affiliated extended supply chain backbones’’ , Information Systems Frontiers, Vol. 2 No. 1, pp. 57-64. Croom, S. (2001), ``Restructuring supply chains through information channel innovation’’, International Journal of Operations and Production Management, Vol. 21 No. 4, pp. 504-15. Davis, M. and O’Sullivan, D. (1998), ``Communications technologies for the extended enterprise’’, Production Planning and Control, Vol. 9 No. 8, pp. 742-53. Dong, Y. and Xu, K. (2001), `À supply chain model of vendor managed inventory’’, Transporatati on Research Part E. Forrester, J.W. (1961), Industrial Dynamics, MIT Press, Cambridge, MA. Furst, K. and Schmidt, T. (2001), ``Turbulent markets need flexible supply chain communication’’, Production Planning and Control, Vol. 12, No. 5, pp. 525-33. Gavirneni, S., Kapuscinski, R. and Tayur, S. (1996), Value of Information in Capacitated Supply Chains, Graduate School of Industrial Administration, Carnegie Mellon University, Pittsburgh, PA. Graham, G. and Hardaker, G. (2000), ``Supplychain management across the Internet’’, International Journal of Physical Distribution and Logistics Management, Vol. 30 No. 3/4, pp. 286-95. Helo, P.T. (2000), ``Dynamic modelling of surge effect and capacity limitation in supply chains’’, International Journal of Production Research, Vol. 38 No. 17, pp. 4521-33.

[ 357 ]


The authors wish to thank the Hong Kong Research Grant Council and the Committe on Research and Conference Grants of the Hong Kong University for the financial support.

[ 358 ]

Huang, G.Q., Lau, J.S.K. and Mak, K.L. (2001), A Framework of Investigating the Impacts of Sharing Production Information on Supply Chain Dynamics’’, Technical Report, Department of Industrial and Manufacturing\Systems Engineering, University of Hong Kong. Kehoe, D.F. and Boughton, N.J. (2001), ``New paradigms in planning and control across manufacturing supply chains ± the utilisation of Internet technologies’’, International Journal of Operations & Production Management, Vol. 21 No. 5/6, pp. 582-93. Lee, H., Padmanabhan, V. and Whang, S. (1997), `Ìnformation distortion in a supply chain: the bullwhip effect’’, Management Science, Vol. 43 No. 4, pp. 546-58. Lee, H.L., So, K.C. and Tang, C.S. (2000), ``The value of information sharing in a two-level supply chain’’, INFORMS, Vol. 46 No. 5, pp. 626-43. Metters, R. (1997), ``Quantifying the bullwhip effect in supply chain’’, Journal of Operations Management, Vol. 15, pp. 89-100. National Research Council (2000), Surviving Supply Chain Integration: Strategies for Small Manufacturers, National Academy Press, Washington, DC. Parunak, H.V.D., Savit, R. and Riolo, R.L. (1998), `Àgent-based modeling vs equation-based modeling: a case study and users’ guide’’, Proceeding of Workshop on Multi-agent Systems and Agent-based Simulation, Springer, Berlin. Parunak, H.V.D., Savit, R., Riolo, R.L. and Clark, S.J. (1999), DASCh ± Dynamic Analysis of Supply Chains, Final Report, Center for Electronic Commerce, ERIM, Ann Arbor, MI. Roy, R. (1990), A Primer on the Taguchi Method, Van Nostrand Reinhold, New York, NY. Silver, E.A., Pyke, D.F. and Peterson, R. (1998), Inventory Management and Production Planning and Control, 3rd ed., John Wiley & Sons, New York, NY.

Simchi-Levi, D., Simchi-Levi, E. and Kaminsky, P. (2000), Designing and Managing the Supply Chain, Irwin/McGraw-Hill, Boston, MA. Singer, T. (1999), ``Sharer beware’’, Inc. Magazine, Vol. 21 No. 4. Sterman, J.D. (1989), ``Modeling managerial behavior: misperceptions of feedback in a dynamic decision making experiment’’, Management Science, Vol. 35 No. 3, pp. 321-39. Strader, T.J., Lin, F. and Shaw, M.J. (1998), `Ìnformation infrastructure for electronic virtual organization management’’, Decision Support Systems, Vol. 23, pp. 75-94. Swaminathan, J.M., Sadeh, N.M. and Smith, S.F. (1997), Effect of Sharing Supplier Capacity Information, Haas School of Business, University of California, Berkeley, CA. Tan, G.W. (1999), ``The impact of demand information sharing on supply chain network’’, PhD thesis, University of Illinois at Urbana-Champaign, IL. Towill, D.R. (1991), ``Supply chain dynamics’’, International Journal of Computer Integrated Manufacturing, Vol. 4 No. 4, pp. 197-208. Towill, D.R., Naim, M.M. and Wikner, J. (1992), `Ìndustrial dynamics simulation models in the design of supply chains’’, International Journal of Physical Distribution & Logistics Management, Vol. 22 No. 5, pp. 3-13. Warkentin, M., Sugumaran, V. and Bapna, R. (2001), `È-knowledge networks for interorganizational collaborative e-business’’, Logistics Information Management, Vol. 14 No. 1/2, pp. 149-62. Wikner, J., Towill, D.R. and Naim, M. (1991), ``Smoothing supply chain dynamics’’, International Journal of Production Economics, Vol. 22 No. 3, pp. 231-48. Wu, S.D. and Meixell, M.J. (1998), Relating Demand Behavior and Production Policies in the Manufacturing Supply Chain, Lehigh University, Bethlehem, PA. Zhao, X. and Xie, J. (2002), ``Forecasting errors and the value of information sharing in a supply chain’’, International Journal of Production Research, Vol. 40 No. 2, pp. 311-35.