Description
The report that investigates the value of various information exchange mechanisms in a four-echelon supply chain under a material requirements planning framework.
International Journal of Production Research, Vol. 45, No. 21, 1 November 2007, 5057–5074
Value of information exchange and synchronization in a multi-tier supply chain
S. VISWANATHAN*y, HANDIK WIDIARTAy and RAJESH PIPLANIz
yNanyang Business School, Nanyang Technological University, Singapore 639798, Republic of Singapore zSchool of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore 639798, Republic of Singapore
(Revision received November 2005) We investigate the value of various information exchange mechanisms in a four-echelon supply chain under a material requirements planning framework. In the absence of any information sharing, each echelon would develop its own forecasts and plan its inventories based on the history of actual demand from its downstream customer (or echelon). Through a simulation study, we compare this policy with policies where each echelon has access to (i) the end-user demand history and (ii) the planned order schedule of the downstream echelon. Among all the demand information exchange mechanisms, planning inventories based on the planned downstream order schedules resulted in the lowest average inventory level for the entire supply chain. However, use of end-user demand history to forecast and plan inventories at all echelons resulted in the lowest total cost. In addition to the information exchange mechanisms, a simple synchronized replenishment system was considered and evaluated in the study. In the synchronized system, the retailer determines a fixed order interval and the upper echelons replenish only at integer multiples of this interval. The study found that synchronized inventory replenishments among the echelons, even without any exchange of demand information, can bring about more benefits and cost reduction than any of the information exchange mechanisms. Keywords: Inventory management; Simulation; Supply chain management; Materials requirements planning; Bullwhip; Value of demand information; Supply chain synchronization
1. Introduction There has been a growing realization among researchers as well as practitioners that sharing demand information with other members in the supply chain can lead to a reduction in inventory-related costs. The advances in information and Web technologies have made this easily possible through software solutions and concepts such as Web-based Enterprise Resource Planning and Collaborative Planning,
*Corresponding author. Email: [email protected]
International Journal of Production Research ISSN 0020–7543 print/ISSN 1366–588X online ß 2007 Taylor & Francis http://www.tandf.co.uk/journals DOI: 10.1080/00207540600930057
5058
S. Viswanathan et al.
Forecasting and Replenishment (CPFR). This has led to initiatives for collaboration and greater visibility of demand information in the supply chain. Although it is clear that visibility of information is beneficial to the entire supply chain (Mason-Jones and Towill 1997, Gavirneni et al. 1999, Mitra and Chatterjee 2004), how the information is to be used to improve inventory planning and reduce supply chain costs is an issue that is worthy of further research. While there have been a number of studies addressing the issue of the value of information sharing, very few have studied the issue under a material requirements planning (MRP) framework or for more than two echelons. Also, in a multi-tier supply chain, information would be available from multiple sources. For example, each echelon may have access to the end-user demand data, as well as to the planned order schedule of its downstream echelon (Lee and Whang 2000). The question then would be to decide what information to use for inventory planning at each echelon. In this paper we investigate through a simulation study the value of various demand information exchange mechanisms in a four-echelon supply chain under a MRP framework. Numerous studies, for example McCullen and Towill (2002) and Disney et al. (2004), have considered a similar form of supply chain, which is composed of several independent organizations representing manufacturer, distributor, wholesaler, and retailer. In the absence of any information sharing, each echelon would develop its own forecasts and plan its inventories based on the history of actual demand from its downstream customer (or echelon). We compare this ‘no-information’ policy or ‘echelon demand history’ mechanism with mechanisms where each echelon has access to (i) end-user demand history and (ii) the planned order schedule of the downstream echelon. Among the information exchange mechanisms, our study found that planning inventories based on the planned downstream order schedules led to the lowest average inventory level for the supply chain. However, use of end-user demand history to forecast and plan inventories at all echelons resulted in the lowest total cost. In addition to the information exchange mechanisms, a simple synchronized replenishment system was also considered and evaluated in this study. Similar to the ‘no-information’ policy, in this synchronized system, no information on end-user demand or planned order schedule is shared within the supply chain. The difference, however, is that instead of allowing the echelons to place an order at any time, this system restricts them to place an order only at fixed time intervals. Assuming that mj is the time interval corresponding to the economic order interval for echelon j, this policy will automatically lead the upper echelon, j þ 1, to place an order only once every (njþ1 Â mj) periods, where njþ1 is a positive integer. This replenishment mechanism is motivated by several cases in reality where sharing demand information in a supply chain may not be feasible for some companies due to various issues such as trust and technology compatibility (Scheraga 2002, D’Avanzo et al. 2003). Our experiments found that synchronized inventory replenishments among the echelons, even without any exchange of demand information, can bring about more benefits and cost reduction than any of the information exchange mechanisms. The paper is organized as follows. We first provide a review of the related research. We then describe the simulation model in detail including the parameters used in the experiment. The findings of the study are discussed thereafter.
Value of information exchange and synchronization in a multi-tier supply chain
5059
Finally, a few concluding remarks and possible directions for future research are provided.
2. Related research When no information is shared across the supply chain and each echelon acts to optimize its own costs, the demand variability in the supply chain becomes amplified as it moves upstream in the supply chain. This phenomenon, known as the bullwhip effect, was first observed by Forrester (1961), and has been studied further by Lee et al. (1997) and later by Chen et al. (2000). One of the solutions proposed to counter the bullwhip effect is to have information sharing across the supply chain. The value of information sharing in the supply chain is an issue that has been studied by many researchers in the recent past. Most have used an inventorytheoretic framework and typically study a two-echelon system. Without information sharing, only the local information on inventory levels (or installation stock information) is used in the replenishment planning process. With information sharing, global inventory information (or echelon stock information) is used for planning the inventory replenishments. The information being shared is normally the inventory information. The probability distribution of the end-user demand is either assumed to be known globally or each echelon just responds only to the demand history or pattern from its immediate downstream echelon. Papers that have adopted this framework include Bourland et al. (1996), Chen (1998), Gavirneni et al. (1999), Cachon and Fisher (2000), Lee et al. (2000), and Raghunathan (1999, 2001, 2003). Most papers, with the possible exception of Cachon and Fisher (2000) and Raghunathan (2001), have concluded that the supply chain inventory costs can be reduced significantly through information sharing. Cachon and Fisher (2000) found that the benefit of information sharing is small compared with the benefit from reduction in lead-time and batch sizes due to information technology. Raghunathan (2001) found that the history of the retailer’s orders itself was a surrogate for its demand and inventory information, and hence the value of information was quite limited. A second stream of related research is coordination (or integration or synchronization) of the inventory management/production schedule across the supply chain. When information alone is shared, the channel members can still act independently but use the additional information to optimize their own costs. Coordination, on the other hand, requires the supply chain members to take ordering decisions in a coordinated fashion. This may not necessarily be the best from the point of view of optimizing the costs of a particular echelon, but it would enhance the profits and reduce the costs for the supply chain as a whole. Typically, in a coordinated supply chain, the supply chain as a whole gains, but some members might see an increase in their costs. Therefore, it requires agreements on how the members who are forced to take locally suboptimal decisions are compensated. A major issue therefore in coordination of supply chains is how the extra benefits are shared among the supply chain members. In the literature on supply chain coordination, the papers again typically assume a two-level structure, mostly with deterministic (but possibly price-elastic) demand.
5060
S. Viswanathan et al.
Typically, a game theory approach is employed by most papers to study mechanisms for achieving coordination between different players in the supply chain and for sharing of the benefits. Some of the papers in this area include Weng (1995), Wei and Krajewski (2000), Chen et al. (2001), Krajeswski and Wei (2001), Viswanathan and Piplani (2001), and Klastorin et al. (2002). There has also been significant interest in supply chain collaboration (which includes exchange of demand information as well as coordination) among practitioners and industry associations (Simatupang et al. 2002, VICS Guide 2002, Larsen et al. 2003, Danese 2006). Though industry efforts such as Vendor Managed Inventory (VMI), and Collaborative Planning Forecasting and Replenishment (CPFR) promoted by bodies such as VICS, ECR committees and consulting companies have propounded the whole umbrella of supply chain collaboration, in reality the focus has been mostly on information exchange (Larsen et al. 2003). Moreover, as mentioned by Danese (2006), there seem to be different ways in which strategies such as CPFR are implemented in practice by different firms. Companies such as Dell and Seagate and Logistics Service Providers such as Fedex claim to provide visibility of demand information to all the supply chain partners. However, it is not very clear as to how the entities in the supply chain use this information to better manage their production and inventories. This is perhaps why there have only been limited success stories beyond pilot projects for these efforts (Larsen et al. 2003). One reason for the apparent lack of success of such efforts to exchange demand information is the reluctance of companies to share what is perceived as sensitive data (Scheraga 2002, Danese 2006). In addition to several information exchange mechanisms, this paper studies a synchronized replenishment scheme that does not require exchange of any demand information.
3. Simulation model We consider a four-echelon supply chain. The most downstream echelon (retailer) faces the demand for the item in each period from the end-user. Each echelon purchases the item from its upstream echelon. When an order in placed, the items are delivered by the upstream echelon provided they have it in stock. Unsatisfied orders are backlogged. The most upstream echelon (manufacturer) is assumed to replenish its inventory from a source with unlimited supply. Therefore, its orders are satisfied without any backlogging. Items supplied by an echelon reach its downstream echelon only after a shipment (or replenishment) lead-time. Each echelon incurs a fixed ordering cost for every order. Holding cost is assessed based on the inventory on hand at the end of each period. Each echelon uses a material requirements planning (MRP) framework for planning its inventories. A dynamic lot-sizing model using a rolling time horizon is used to plan the order schedule. The length of the rolling horizon, N, used in the simulation is 32 weeks. The ordering decisions made at time period T will impact the echelon’s inventory only after T þ L, (where L is the shipment lead-time). Therefore, the dynamic lot-sizing model will use the time frame [T þ L, T þ L þ N] for cost calculations when planning the order schedule at time T. The gross requirement schedule (or projected demand) used by the lot-sizing model will depend on the information exchange mechanism that is adopted. The retailer
Value of information exchange and synchronization in a multi-tier supply chain
5061
faces the actual end-user demand. Therefore, irrespective of the information exchange mechanism used, the gross requirement for the retailer would be the demand forecast created using end-user demand history. For other echelons, when no information is shared, the gross requirement would be the forecast created using the history of demand from its downstream echelon. We refer to the mechanism where no information is shared as the ‘Echelon Demand History’. When information is shared across the supply chain, there are two other possible ways in which the gross requirement schedule for an echelon can be generated for the lot-sizing model. One is to use the projected order schedule of the downstream echelon (which itself is generated by the lot-sizing model at this echelon). This would be the case if the MRP systems are integrated across the four echelons (as if they all belonged to a single organization). We refer to this information exchange mechanism as the ‘Downstream Order Schedule’. The other possible way to generate a gross requirement schedule would be to use a common forecast based on the history of end-user demand for all echelons. We refer to this information exchange mechanism as ‘End-User Demand’. The net requirement for the periods [T þ L, T þ L þ N] is what is used in the dynamic lot-sizing model for generating the projected order schedules at period T. For the periods [T þ L þ 1, T þ L þ N], the net requirement is the same as the gross requirement generated by the forecast or the particular information exchange mechanism. For the period T þ L, the net requirement is the desired safety stock þ gross requirement for the period T þ L minus projected on hand inventory at the end of period T þ L À 1. To be more specific, NRTþL ¼ GRTþL þ SS À ITþLÀ1 , where NRTþL is the net requirement for period T þ L, GRTþL is the gross requirement for period T þ L, SS is the desired safety stock, and ITþLÀ1 is the projected on hand inventory at the end of period T þ L À 1 and ITþLÀ1 ¼ ITÀ1 þ BOTÀ1 þ
TÀ1 X t¼TÀL
Qt À
TþLÀ1 X t¼Tþ1
GRt À DT ,
where Qt is the order quantity placed by the echelon in period t (that is expected to be received in period j þ L), BOT–1 is orders outstanding at T À 1 (i.e. orders placed until T À L À 1, but not shipped by the upstream echelon, i.e. not received until T À 1), and DT is the downstream demand for period T. The algorithm used for solving the N-horizon, dynamic lot-sizing problem in each period for each echelon is a modification to the Silver–Meal (1973) heuristic suggested by Silver and Miltenburg (1984). In the upstream echelons, demand/orders would be lumpy or intermittent (i.e. positive demand interspersed with several zero demands) due to the order batching effect created by the lot-sizing model. The modified Silver-Meal heuristic by Silver and Miltenburg (1984) handles intermittent demand well. When the demand/gross requirement at an echelon is forecasted using the end-user demand history, the exponential smoothing (ES) method is used. Exponential smoothing is used because it is simple to implement (Dekker et al. 2004) and widely adopted in MRP systems (Klassen and Flores 2001). When the echelon demand history is used to forecast the gross requirement, ES may not be
5062
S. Viswanathan et al.
appropriate as it does not work well with lumpy demands (Willemain et al. 1994). Therefore, a modification of the Croston’s (1972) method is used in this case. Croston’s method is a refinement of ES where the quantum of the positive demand and the length of the number of zero demands are forecasted separately. Of course, when the downstream order schedule is the information exchange mechanism used, no forecasting of the gross requirement is necessary for the wholesaler, distributor, and manufacturer. In addition to the three information exchange mechanisms (echelon demand history (or no information sharing), downstream order schedule, and end-user demand history), we also evaluated the performance of a simple synchronized replenishment system. In this ‘synchronized replenishment’ there is no information sharing and, therefore, the echelon demand history is used to forecast the gross requirement at each echelon. However, instead of carrying out the material requirement planning in every period, order scheduling using the lot-sizing model is carried out only at fixed time intervals (this is staggered for the different echelons to take into account the replenishment lead-time). Within the lot-size model itself, orders are allowed to be placed by echelon j only once every mj periods in the horizon of N periods (see figure 1). The order interval, mj, is the time interval corresponding to the economic order interval (that minimizes the total cost per period) for echelon j, assuming that the demand in each period was equal to the average downstream demand. (Note that the average demand can be estimated using the demand history.) The economic order interval for each echelon can essentially be calculated using the Silver–Meal lot-sizing technique. As a particular echelon j places an order only once every mj periods, clearly the best order interval for the upstream echelon, j þ 1, is an integer multiple of mj or njþ1 Â mj periods, where njþ1 is an integer !1. Consequently, it would be easier for the upstream echelons to forecast the demand using a method such as Croston’s. As we will see later, even though the synchronized replenishment system does not have the benefit of sharing demand information, it still outperforms
Echelon-4 (j = 4) (Manufacturer) m4= n4 x m3= 2 x 4 = 8 Information flow Material flow Echelon-3 (j = 3) (Distributor) Echelon-2 (j = 2) (Wholesaler) Echelon-1 (j = 1) (Retailer) End-User’s demands ... 23
m3 = n3 x m2 = 1 x 4 = 4
Order placement with upper echelon
m2= n2 x m1 = 2 x 2 = 4
m1 = 2 weeks 1 2 3 4 5 6 7 8 9 10
Demand occurrence (Weekly)
Figure 1. Example of ordering pattern for each echelon using the synchronized replenishment mechanism.
Value of information exchange and synchronization in a multi-tier supply chain
5063
all the information sharing mechanisms both in terms of the average inventory as well as total cost to the supply chain. In each time period of the simulation run, the sequence of activities is as follows. First, the shipments from the upstream echelon (shipped L periods ago) are received into the inventory. Next, the demand for the period arises, and it is satisfied if there is adequate stock and backlogged otherwise. Then, the backlogged demand from the previous periods, if any, is satisfied, if there is stock. It is assumed that partial fulfillment of an order is allowed. The forecasts (for demand/gross requirement) are then updated, and followed by computing the net requirement. Finally, the order schedule using the lot-sizing model is calculated and orders corresponding to the current period are placed with the upstream echelon. In each period, the simulation is carried out first for the most downstream echelon (retailer), and then for each upstream echelon, i.e. wholesaler, distributor, and manufacturer in that order. For the upstream echelons, the order placed by the immediate downstream echelon becomes the demand. The end-user demand (or demand for the retailer) is randomly generated using a first-order autoregressive [AR(1)] process. The AR(1) demand process, which has a non-zero lag-1 autocorrelation, has been adopted by numerous authors to study various issues in supply chain management (see, for instance, Chen et al. (2000), Lee et al. (2000), and Disney and Grubbstrom (2004)), as it represents many demand processes in practice (Nahmias 1993, Chopra and Meindl 2001). AR(1) is different from the normally distributed demand process in that the AR(1) process exhibits some non-stationary characteristics that justify the use of a forecasting mechanism (Disney et al. 2003). Its mathematical representation is given as follows (Box and Jenkins 1976, p. 56): d1 ¼ þ " 1 , dt ¼ ð1 À Þ þ dtÀ1 þ "t , where is a non-negative constant, | |51 and is constant at all times, and "t 2 is normally distributed with zero mean and variance " (note that, if ¼ 0, the demand becomes independently and identically distributed (i.i.d.) with mean 2 and variance " ). The simulation test bed is developed using MS Visual Cþþ 6.0. For each set of experimental parameters, the simulation is run for a total of 1300 periods. The initial set of 260 periods (or 20% of the run length) is considered as the warm-up period and the performance statistics/measures are calculated only for the remaining 1040 periods. The parameters used in the simulation experiments are provided in table 1. For a particular set of experimental parameters, and for a particular mode of information exchange mechanism (or for the synchronized replenishment), the safety stock for a particular echelon was kept fixed throughout the simulation run. The simulation was basically run (for a particular set of experimental parameters) under different values of the safety stock until the safety stock that achieves the desired fill rate (e.g. 94% for the base case) for all four echelons is found. The average inventory level and the total cost of the supply chain are calculated corresponding to this level of safety stock. Note that this involves running the
5064 Table 1. Parameter
S. Viswanathan et al. Experimental design parameters. Value 1300 260 32 1000 0.2 2 (basea), 3, 4, 5, 6 200 (base), 400, 600, 800, 1000, 1200 94 (base), 95, 96, 97, 98, 99 À0.6, À0.4, À0.2 (base), 0.2, 0.4, 0.6
Run-length (periods) Warm-up period (periods) Length of rolling horizon (periods) Mean of end-user demand, (units/period) Smoothing constant value used for forecasting in ES and modified Croston’s method Replenishment lead-time, L (periods) Standard deviation of the error term of the demand processb, " (units) Desired fill rate for all echelons, F (%) Serial correlation coefficient used in AR(1) demand process,
a Base indicates the value that is used when the experiment is carried out by varying just one particular parameter. For example, when the performance of all mechanisms is investigated against different values of L, other independent variables such as ", F, and are set to their respective base values of 200, 94, and –0.2. b 2 Standard deviation of the end-user demand ( d) can be obtained by d ¼ ½" =ð1 À 2 Þ?1=2 . Therefore, if ¼ À0.2, we can respectively translate the values of " used in this experiment to d ¼ {204.12 (base), 408.25, 612.37, 816.5, 1020.62, 1224.75}.
simulation repeatedly for a particular parameter set, as the correct safety stock level for each of the four echelons needs to be determined. The cost elements included in the total cost calculation are the ordering and holding costs at each echelon. As the same service level is imposed for all the information exchange mechanisms, no penalty cost is included in the model. The various combinations of holding and ordering cost values used in this study are shown in table 2. The performance of the various information exchange mechanisms and the synchronized replenishment system is measured by the average inventory level and total cost to the supply chain. These two are therefore the dependent variables in the experiments. The independent variables are the various parameters used in the experiments, namely the standard deviation of the error term of the demand process ( "), serial correlation coefficient of the demand process ( ), replenishment lead-time (L), desired fill rate (F), and ordering cost (A). With four different information exchange mechanisms, ten combinations of ordering and holding costs, and the different values for the various parameters, the simulation experiment conducted in this study involved 1000 different experimental scenarios. The system being considered in this study is generic enough to be applied to any distribution channel or manufacturing supply chain. Transformation of the products through the supply chain is assumed implicitly through the different, echelon-specific, holding costs and ordering costs. Of course, there is effectively only a one-for-one transformation or bill of materials. One could of course consider more complex bill of materials and transformation structures, but as the objective of the research was to gain insights into the impact of various information exchange and coordination mechanisms, we did not consider such complexities. The data presented
Value of information exchange and synchronization in a multi-tier supply chain Table 2. Case 1 Holding and ordering cost values used in the experiments. Retailer 0.5 3000 0.5 3000 0.5 2550 0.5 2100 0.5 3000 0.5 3000 0.5 2550 0.5 2100 0.5 3000 0.5 3000 Wholesaler 0.45 3000 0.5 3000 0.5 2700 0.5 2400 0.5 2850 0.5 2700 0.45 2700 0.45 2400 0.45 2850 0.45 2700 Distributor 0.4 3000 0.5 3000 0.5 2850 0.5 2700 0.5 2700 0.5 2400 0.4 2850 0.4 2700 0.4 2700 0.4 2400
5065
Parameter Holding cost, H ($/unit/period)a Ordering cost, A ($/order)b H A H A H A H A H A H A H A H A H A
Manufacturer 0.35 3000 0.5 3000 0.5 3000 0.5 3000 0.5 2550 0.5 2100 0.35 3000 0.35 3000 0.35 2550 0.35 2100
2 3 4 5 6 7 8 9 10
a
Assuming a unit cost of $100 and length of a period to be one week, this works to a holding cost rate of about 26% per year at the retailer’s site. b For every case, six different levels (3000, 4000, 5000, 6000, 7000, and 8000) of ordering cost are evaluated. The ordering costs for the other echelons are adjusted proportionately in each case for the different levels.
in tables 1 and 2 are a typical representation of data found in practice for products such as fast moving consumer goods.
4. Simulation results We now discuss the results of the simulation study and their managerial implications. We first discuss the relative performance of the different information exchange mechanisms and how it is impacted by particular experimental parameters. In the experimental results that are presented in figures 2–5, the total cost (or the average inventory level) for the whole supply chain is plotted against a particular experimental parameter. For these figures, the value of the particular parameter alone is varied and the rest of the parameters are kept fixed at their base values. The figures only report the results for the holding and ordering cost combination under case 1 of table 2. The results were quite similar for other combinations of the holding and ordering costs and hence are not reported. We first report the performance of the different information exchange mechanisms for different values of end-user demand variability in figures 2 and 3. In figure 2, the average supply chain inventory level for the different mechanisms is
5066
Average inventory level (units) 22000 18000 14000 10000 6000 2000 200
S. Viswanathan et al.
400
600 800 Variability of end-user demand (se)
1000
1200
End-user demand Synchronized replenishment
Downstream order schedule Echelon demand history
Figure 2. Plot of the average inventory level for the supply chain against end-user demand variability ( ") for different information exchange mechanisms.
12000 Total supply chain cost ($)
10000
8000
6000
4000 200
400
600
800
1000
1200
Variability of end-user demand (se) End-user demand Synchronized replenishment Downstream order schedule Echelon demand history
Figure 3. Plot of the total supply chain cost (inventory holding þ ordering) against end-user demand variability ( ") for different information exchange mechanisms.
plotted against ", the standard deviation of the error term in the end-user demand process (as mentioned in table 1, " is directly related to the variability or standard deviation of the end-user demand). As can be seen, no information sharing (or echelon demand history) resulted in the highest average supply chain inventory level for most values of ".The use of downstream order schedule resulted in a lower average inventory than using a common single forecast based on the history of enduser demand. As discussed earlier, if the MRP systems for the different echelons were integrated (as if it were a single organization), then the gross requirement for the
Value of information exchange and synchronization in a multi-tier supply chain
12000 Total supply chain cost ($)
5067
10000
8000
6000
4000 2 3 4 Replenishment lead-time (weeks) 5 6
End-user demand Synchronized replenishment
Downstream order schedule Echelon demand history
Figure 4. Plot of the total supply chain cost against the replenishment lead-time for different information exchange mechanisms.
16000 Total supply chain cost ($) 14000 12000 10000 8000 6000 4000 3000
4000
5000 6000 Ordering cost at each echelon ($)
7000
8000
End-user demand Synchronized replenishment
Downstream order schedule Echelon demand history
Figure 5. Plot of the total supply chain cost against the echelon ordering cost for different information exchange mechanisms.
upstream echelons would be based on the downstream order schedules. Figure 2 suggests that the supply chain inventory is minimized by integrating the MRP systems of the different echelons and using the downstream order schedule as the gross requirement for each echelon. This finding justifies the interest among companies to provide information visibility and integrate the information systems across the supply chain partners.
5068
S. Viswanathan et al.
However, when minimizing the total cost per period (holding cost þ ordering cost) of the supply chain is the objective, the downstream order schedule does not do that well as an information exchange mechanism. This fact is demonstrated in figure 3, which plots the total supply chain cost per period against " for the various information exchange mechanisms. From this figure it is clear that, for most values of ", the use of a common forecast based on history of end-user demand resulted in a lower total supply chain cost than use of downstream order schedules. In fact, for low values of the end-user demand variability, the total supply chain cost using the downstream order schedule mechanism was nearly identical to that using the echelon demand history mechanism or no information sharing. Only when the demand variability was very high (or, in other words, end-user demand information was not very reliable) did the total cost using downstream order schedule information do better than the total cost using end-user demand history. One reason for the poor (total cost) performance of the downstream order schedule mechanism could be due to the so-called MRP nervousness. MRP nervousness is the situation where there is frequent change to the planned replenishment orders, due to the use of lot-sizing algorithms to determine order schedules in every period under a rolling horizon framework. There is a large amount of literature that has addressed the issue of MRP nervousness. Researchers who have studied this include Kropp et al. (1983), Blackburn et al. (1986), Sridharan et al. (1987), Sridharan and Berry (1990), and Vollman et al. (1997). Even though the average inventory level was lower for the downstream order schedule mechanism, the total costs were higher than that for the mechanism using end-user demand history except when the demand variability was extremely high. Clearly, the ordering cost or the average number of orders under downstream order schedule is much higher than that for end-user demand history. One way to measure the impact of MRP nervousness on the total cost is to develop a nervousness index. We define the nervousness index as the ratio of the average number of orders per period calculated by the lot-sizing method in each planning period to the actual number of orders per period observed in the simulation. Table 3 reports the average nervousness index based on the simulation results for the different echelons for the various information exchange mechanisms. As can be seen from the table, for mechanisms other than the downstream order schedule, the index value was close to one, implying that the actual number of orders per period was close to the planned orders as per the lot-sizing heuristic. For the downstream
Table 3.
Nervousness index observed in the simulation experiments. Average for the supply chain 0.988 0.993 1.663 0.980
Mechanisma EDH EUD DOS SR
Retailer 1.13 1.13 1.13 0.99
Wholesaler 0.96 1.02 1.71 0.99
Distributor 0.98 0.93 1.81 0.98
Manufacturer 0.88 0.89 2.00 0.96
a EDH, echelon demand history; EUD, end-user demand; DOS, downstream order schedule; SR, synchronized replenishment.
Value of information exchange and synchronization in a multi-tier supply chain
5069
order schedule mechanism alone, the nervousness index value was much higher, implying that the actual number of orders per period was much higher than the planned orders. One common suggestion to alleviate MRP nervousness is to freeze the master production schedule for a certain period. In the context of our research, it appears that the total supply chain cost can be reduced further by use of a common forecast using end-user demand history or through a synchronized replenishment scheme (without any exchange of demand information) rather than by integrating the MRP systems of the different echelons. Figure 4 plots the total cost of the whole supply chain for the various information exchange mechanisms against the replenishment lead-time. The pattern in this figure was consistent with those in figure 3. That is, the total supply chain cost through the use of (forecast based on) end-user demand was consistently lower than that through the use of downstream schedules. The total costs for the various information exchange mechanisms plotted against the desired fill rate (F) and the serial correlation coefficient of the end-user demand ( ) were also similar to the earlier figures; therefore, we do not present them in the paper. Figure 5 plots the total supply chain cost for the various mechanisms against the ordering cost for the echelons. In this case again, the total supply chain cost through the use of forecasts based on end-user demand was lower than that through the use of downstream order schedules. However, for higher values of the ordering cost, the total cost through the use of echelon demand history was lower than the cost through the use of either of the other two information exchange mechanisms. This is probably because of the fact that the effect of MRP nervousness becomes more prominent as ordering cost increases. In other words, information sharing had no value when the echelon ordering costs were higher. In all the four figures, the synchronized replenishment scheme (without any exchange of demand information) consistently resulted in a lower supply chain cost than any of the information exchange mechanisms. As explained earlier, the synchronized replenishment system is a simple coordination scheme in which all echelons commit to reviewing and placing orders only at fixed time intervals, but do not share any end-user demand or planned order schedule information. The simulation results therefore seem to suggest that there is substantially more value in attempting to synchronize the replenishment epochs rather than just to share demand information. The superiority of the synchronized replenishment scheme in the simulation study also seems to tie in with the research findings in the MRP nervousness literature, where freezing the master production schedule (MPS) is offered as one of the solutions to the problem. By committing to review and place orders only at fixed time intervals (in the synchronized replenishment scheme), the orders are essentially frozen for that period. However, the significant difference between the synchronized replenishment scheme and the freezing of MPS is that, in the MRP nervousness literature, all the different product levels belong to the same MRP system, and the gross requirement at a level is determined by the downstream order schedules. In the synchronized replenishment scheme that was used in our simulation study, there was only synchronization, but no sharing of demand information. The echelon demands were forecast using the echelon demand history rather than the downstream order schedules.
5070
S. Viswanathan et al.
One question that would arise is what would be the best or optimal ordering policy for an integrated supply chain system in this context. When the end-user demands are known with certainty, an optimal algorithm has been proposed by Graves (1981). However, this method is computationally prohibitive, especially as the length of the planning horizon increases. Besides, the algorithm has to be run in every period of the simulation to calculate the schedule in a rolling horizon context. Heuristics have been proposed by McLaren and Whybark (1976) and Blackburn and Millen (1982). We had implemented these two heuristics in our experimental study, but their total costs were higher than most other information exchange mechanisms for most cases. As mentioned, these methods were developed for finding the optimal order schedules at a particular point in time for deterministic demands. Apart from the fact that the demands are varying in a stochastic manner, the schedules are calculated in every period in a rolling horizon framework, which can cause MRP nervousness. This is probably why these heuristics performed poorly in our study (the average nervousness index for lot-sizing based on these heuristics was 2.18, which is much higher than that for all other mechanisms, including downstream order schedule). As the performance of the two heuristics was inferior to even the echelon demand history mechanism, we have chosen not to report these in the paper. When the end-user demands are stochastic and stationary, optimal policies for an integrated supply chain have not been found to date. The best policy in this case seems to be the (R, nQ) policy suggested by Chen and Zheng (1998). In this policy, the order quantity for a higher echelon is an integer multiple
of the order quantity at the lower echelon (Q). However, their algorithm can be applied only for Poisson or compound Poisson demands. Moreover, this algorithm may not be very appropriate in a MRP context where the demand forecasts evolve periodically. The synchronized replenishment mechanism used in our study in a sense mimics the (R, nQ) policy. Orders are placed by the retailer only at fixed intervals and the order intervals for the other echelons are at integer multiples of this interval (after making the time shift for the lead-time). One could possibly try to improve the synchronized replenishment scheme by trying to use downstream order schedule instead of forecasts based on echelon order history. However, as discussed in the Introduction, the motivation for this scheme was based on the fact that sharing of demand information is difficult in many supply chains. The average of the inventory level and total cost incurred at each echelon for all the experiments are presented in table 4. As can be seen from table 4, with the exception of the retailer, the cost at all the other echelons decreased substantially under synchronized replenishment. Table 5 reveals the average savings in the total cost and inventory level generated at each echelon using different information exchange mechanisms, compared with the use of just echelon demand history (where there is no information exchange or coordination). Compared with the case where echelon demand history is used, use of downstream order schedules reduced the average supply chain inventory level by as much as 29.41%, whereas use of a common forecast based on end-user demand history reduced the average supply chain inventory level by only 17.69%. However, use of end-user demand history resulted in a 15.02% reduction in the total supply chain cost, whereas use of downstream order schedules resulted only in a 1.52% reduction in the total cost.
Value of information exchange and synchronization in a multi-tier supply chain Table 4.
5071
Summarized results of the simulation experiments: total cost and average inventory at the four echelons for different information exchange mechanisms. Total for the supply chain 12024.05 9897.01 8488.19 4829.33 9058.46 7698.05 8920.94 5448.83
Dependent variable Average inventory level (units) Total cost ($)
Mechanisma EDH EUD DOS SR EDH EUD DOS SR
Retailer 1170.35 1187.39 1171.19 1800.17 1587.43 1596.03 1589.04 1701.66
Wholesaler 2122.04 2171.21 2084.91 637.86 1943.89 1881.58 2409.84 1106.24
Distributor 3594.90 2928.53 2643.85 949.54 2544.56 2041.96 2528.85 1231.21
Manufacturer 5136.77 3609.88 2588.24 1441.76 2982.59 2178.49 2393.22 1409.72
a EDH, echelon demand history; EUD, end-user demand; DOS, downstream order schedule; SR, synchronized replenishment.
Table 5.
Summarized results of the simulation experiments: percentage savings obtained by different mechanisms. Retailer (%) 0 À1.46 À0.07 À53.82 0 À0.54 À0.10 À7.20 Wholesaler (%) 0 À2.32 1.75 69.94 0 3.21 À23.97 43.09 Distributor (%) 0 18.54 26.46 73.59 0 19.75 0.62 51.61 Manufacturer (%) 0 29.73 49.61 71.93 0 26.96 19.76 52.74 Total savings (%) 0 17.69 29.41 59.84 0 15.02 1.52 39.85
Dependent variable Average inventory level (units) Total cost ($)
Mechanisma EDH EUD DOS SR EDH EUD DOS SR
a EDH, echelon demand history; EUD, end-user demand; DOS, downstream order schedule; SR, synchronized replenishment.
Finally, the synchronized replenishment mechanism resulted in a 59.84% reduction in average supply chain inventory level and a 39.85% reduction in the total cost.
5. Summary and conclusions We have investigated through a simulation study the value of various information exchange mechanisms in a multi-tier supply chain under a MRP framework. In the absence of any sharing of demand information, each echelon develops its own forecasts and plans its inventories based on the history of actual demand from its downstream customer (or echelon). We compared this policy with policies where each echelon had access to (i) end-user demand history and (ii) the planned order
5072
S. Viswanathan et al.
schedule of the downstream echelon. Among the information exchange mechanisms, integrating the MRP systems of the various echelons (or replenishing inventories based on the planned downstream order schedules) resulted in the lowest average inventory level for the supply chain. However, use of end-user demand history to forecast and plan inventories at all echelons resulted in the lowest total cost. In addition to the mechanisms for exchange of demand information, a simple synchronized replenishment system was considered and evaluated in the study. In this synchronized replenishment system, each echelon is restricted to place an order only after a fixed time interval. The study found that synchronized inventory replenishments among the echelons, even without any exchange of demand information, can bring about more benefits and cost reduction than any of the information exchange mechanisms. While the research literature seems to have had equal amounts of focus on both information sharing and coordination in supply chains, practitioners have tended to focus more on information sharing and information visibility. One of the deterrents to sharing information freely has been the lack of trust and willingness among entities in the supply chain. Our results from the simulation study suggest that, to enjoy the full benefits of collaboration, practitioners should focus more on synchronization (or coordination arrangements) rather than just on information visibility or sharing information. Of course, attempting synchronization has its own sets of deterrents, such as how to share the benefits, and uncertainty regarding any short-term negative cost impact; however, these can possibly be overcome more easily than the uneasiness involved in sharing sensitive demand information. The findings of this study are applicable only for the MRP framework and structure of the system considered in the simulation and cannot easily be generalized across other supply chain structures. However, such limitations would exist for most previous studies on supply chain coordination found in the literature. It should also be pointed out that the multi-echelon supply chain structure with a MRP framework used in this study is very commonly found in practice. The proposed synchronization scheme also requires some information exchange. All entities need to indicate their ordering interval and stick to it. However, they need not reveal their downstream demand information to their upstream supplier. As already discussed, the demand information would be perceived as sensitive, especially if the upstream entity is also supplying the goods to the downstream entity’s competitor. Revealing the ordering interval would not be perceived as that sensitive. Most of the research on information sharing and supply chain coordination has been from an inventory theoretic point of view rather than under a rolling horizon, MRP framework. This paper has attempted to fill this gap. However, one issue that has not been addressed in the paper is the development of mechanisms or rules for sharing the benefits of synchronization and information sharing under a MRP framework. For example, even though synchronized replenishment resulted in cost savings of 39.85% for the entire supply chain, the total cost for the retailer actually increased by 7.20% (see table 5). Clearly, appropriate mechanisms are needed to allocate the benefits from coordination and information sharing. This could be a possible issue for future research.
Value of information exchange and synchronization in a multi-tier supply chain
5073
References
Blackburn, J.D., Kropp, D.H. and Millen, R.A., A comparison of strategies to dampen nervousness in MRP systems. Mgmt. Sci., 1986, 32, 413–429. Blackburn, J.D. and Millen, R.A., Improved heuristics for multi-stage requirements planning systems. Mgmt. Sci., 1982, 28, 44–56. Bourland, K.E., Powell, S.G. and Pyke, D.F., Exploiting timely demand information to reduce inventories. Eur. J. Oper. Res., 1996, 92, 239–253. Box, G.E.P. and Jenkins, G.M., Time Series Analysis: Forecasting and Control, 2nd ed., 1976 (Holden-Day: San Francisco). Cachon, G.P. and Fisher, M., Supply chain inventory management and the value of shared information. Mgmt. Sci., 2000, 46, 1032–1048. Chen, F., Echelon reorder points, installation reorder points, and the value of centralized demand information. Mgmt. Sci., 1998, 44, 221–234. Chen, F., Drezner, Z., Ryan, J.K. and Simchi-Levi, D., Quantifying the bullwhip effect in a simple supply chain: the impact of forecasting, lead times, and information. Mgmt. Sci., 2000, 46, 436–443. Chen, F., Federgruen, A. and Zheng, Y., Coordination mechanisms for distribution system with one supplier and multiple retailers. Mgmt. Sci., 2001, 47, 693–708. Chen, F. and Zheng, Y., Near optimal echelon stock (R, nQ) policies in multi-stage serial systems. Oper. Res., 1998, 46, 592–602. Chopra, S. and Meindl, P., Supply Chain Management, 2001 (Prentice-Hall: Engelwood Cliffs, NJ). Croston, J.D., Forecasting and stock control for intermittent demand. Oper. Res. Q., 1972, 23, 289–303. Danese, P., Collaboration forms, information and communication technologies and coordination mechanisms in CPFR. Int. J. Prod. Res., 2006, 44, 3207–3226. D’Avanzo, R., von Lewinski, H. and Van Wassenhove, L.N., The link between supply chain and financial performance. Suppl. Chain Mgmt. Rev., 2003, Nov/Dec, 40–47. Dekker, M., van Donselaar, K.H. and Ouwehand, P., How to use aggregation and combined forecasting to improve seasonal demand forecasts. Int. J. Prod. Econ., 2004, 90, 151–167. Disney, S.M., Farasyn, I., Lambrecht, M., Towill, D.R. and Van de Velde, W., Taming the bullwhip whilst watching customer service, in EURO/INFORMS Joint International Meeting, 2003 (Istanbul). Disney, S.M. and Grubbstrom, R.W., Economic consequences of a production and inventory control policy. Int. J. Prod. Res., 2004, 42, 3419–3431. Disney, S.M., Naim, M.M. and Potter, A., Assessing the impact of e-business on supply chain dynamics. Int. J. Prod. Econ., 2004, 89, 109–118. Forrester, J.W., Industrial Dynamics, 1961 (MIT Press: Cambridge, MA). Gavirneni, S., Kapuscinski, R. and Tayur, S., Value of information in capacitated supply chains. Mgmt. Sci., 1999, 45, 16–24. Graves, S.C., Multistage lot-sizing: An iterative procedure. In TIMS Studies in Management Science. Multi-Level Production/Inventory Systems: Theory and Practice, edited by L.B. Schwarz, Vol. 6, pp. 95–109, 1981 (Amsterdam). Klassen, R.D. and Flores, B.E., Forecasting practices of Canadian firms: survey results and comparisons. Int. J. Prod. Econ., 2001, 70, 163–174. Klastorin, T.D., Moinzadeh, K. and Son, J., Coordinating orders in supply chains through price discounts. IIE Trans., 2002, 34, 679–689. Krajewski, L. and Wei, J.C., The value of production schedule integration in supply chains. Decis. Sci., 2001, 32, 601–634. Kropp, D.H., Carlson, R.C. and Jucker, J.V., Heuristic lot-sizing approaches for dealing with MRP system nervousness. Decis. Sci., 1983, 14, 156–169. Larsen, T.S., Thernøe, C. and Andresen, C., Supply chain collaboration: theoretical perspective and empirical evidence. Int. J. Phys. Distr. Log. Mgmt., 2003, 33, 531–549.
5074
S. Viswanathan et al.
Lee, H.L., Padmanabhan, P. and Whang, S., Information distortion in a supply chain: the bullwhip effect. Mgmt. Sci., 1997, 43, 546–558. Lee, H.L., So, K.C. and Tang, C.S., The value of information sharing in a two-level supply chain. Mgmt. Sci., 2000, 46, 626–664. Lee, H.L. and Whang, S., Information sharing in a supply chain. Int. J. Technol. Mgmt., 2000, 20, 373–387. Mason-Jones, R. and Towill, D.R., Information enrichment: designing the supply chain for competitive advantage. Suppl. Chain Mgmt., 1997, 2, 137–148. McCullen, P. and Towill, D., Diagnosis and reduction of bullwhip in supply chains. Int. J. Suppl. Chain Mgmt., 2002, 7, 164–179. McLaren, B.J. and Whybark, D.C., Multi-level lot-sizing procedures in a material requirements planning environment. Discussion Paper No. 64, Indiana University, 1976. Mitra, S. and Chatterjee, A.K., Leveraging information in multi-echelon inventory systems. Eur. J. Oper. Res., 2004, 152, 263–280. Nahmias, S., Production and Operations Analysis, 1993 (Irwin: Boston). Raghunathan, S., Interorganizational collaborative forecasting and replenishment systems and supply chain implications. Decis. Sci., 1999, 30, 1053–1071. Raghunathan, S., Information sharing in a supply chain: a note on its value when demand is nonstationary. Mgmt. Sci., 2001, 47, 605–610. Raghunathan, S., Impact of demand correlation on the value of and incentives for information sharing in a supply chain. Eur. J. Oper. Res., 2003, 146, 634–649. Scheraga, D., Disappointment reigns. Chain Store Age, 2002, August, 83–85. Silver, E.A. and Meal, H., A heuristic for selecting lot size quantities for the case of a deterministic time-varying demand rate and discrete opportunities for replenishment. Prod. Invent. Mgmt., 1973, 14, 64–74. Silver, E.A. and Miltenburg, J., Two modifications of the silver–meal lot sizing heuristic. INFOR, 1984, 22, 56–69. Simatupang, T.M., Wright, A.C. and Sridharan, R., The knowledge of coordination for supply chain integration. Bus. Process Mgmt J., 2002, 8, 289–308. Sridharan, V. and Berry, W.L., Freezing the master production schedule under demand uncertainty. Decis. Sci., 1990, 21, 97–120. Sridharan, V., Berry, W.L. and Udayabhanu, V., Freezing the master production schedule under rolling planning horizons. Mgmt. Sci., 1987, 33, 1137–1149. Viswanathan, S. and Piplani, R., Coordinating supply chain inventories through common replenishment epochs. Eur. J. Oper. Res., 2001, 129, 277–286. Vollman, T.E., Berry, W.L. and Whybark, D.C., Manufacturing Planning and Control Systems, 4th ed., 1997 (Irwin: Homewood). Voluntary Interindustry Commerce Standards (VICS), Collaborative planning, forecasting and replenishment guidelines, version 2.0. Available online at: http://www.cpfr.org, 2002. Wei, J.C. and Krajewski, L., A model for comparing supply chain schedule integration approaches. Int. J. Prod. Res., 2000, 38, 2099–2123. Weng, Z., Channel coordination and quantity discounts. Mgmt. Sci., 1995, 41, 1509–1522. Willemain, T.R., Smart, C.N., Shockor, J.H. and DeSautels, P.A., Forecasting intermittent demand in manufacturing: a comparative evaluation of Croston. Int. J. Forecast., 1994, 10, 529–538.
doc_372405476.pdf
The report that investigates the value of various information exchange mechanisms in a four-echelon supply chain under a material requirements planning framework.
International Journal of Production Research, Vol. 45, No. 21, 1 November 2007, 5057–5074
Value of information exchange and synchronization in a multi-tier supply chain
S. VISWANATHAN*y, HANDIK WIDIARTAy and RAJESH PIPLANIz
yNanyang Business School, Nanyang Technological University, Singapore 639798, Republic of Singapore zSchool of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore 639798, Republic of Singapore
(Revision received November 2005) We investigate the value of various information exchange mechanisms in a four-echelon supply chain under a material requirements planning framework. In the absence of any information sharing, each echelon would develop its own forecasts and plan its inventories based on the history of actual demand from its downstream customer (or echelon). Through a simulation study, we compare this policy with policies where each echelon has access to (i) the end-user demand history and (ii) the planned order schedule of the downstream echelon. Among all the demand information exchange mechanisms, planning inventories based on the planned downstream order schedules resulted in the lowest average inventory level for the entire supply chain. However, use of end-user demand history to forecast and plan inventories at all echelons resulted in the lowest total cost. In addition to the information exchange mechanisms, a simple synchronized replenishment system was considered and evaluated in the study. In the synchronized system, the retailer determines a fixed order interval and the upper echelons replenish only at integer multiples of this interval. The study found that synchronized inventory replenishments among the echelons, even without any exchange of demand information, can bring about more benefits and cost reduction than any of the information exchange mechanisms. Keywords: Inventory management; Simulation; Supply chain management; Materials requirements planning; Bullwhip; Value of demand information; Supply chain synchronization
1. Introduction There has been a growing realization among researchers as well as practitioners that sharing demand information with other members in the supply chain can lead to a reduction in inventory-related costs. The advances in information and Web technologies have made this easily possible through software solutions and concepts such as Web-based Enterprise Resource Planning and Collaborative Planning,
*Corresponding author. Email: [email protected]
International Journal of Production Research ISSN 0020–7543 print/ISSN 1366–588X online ß 2007 Taylor & Francis http://www.tandf.co.uk/journals DOI: 10.1080/00207540600930057
5058
S. Viswanathan et al.
Forecasting and Replenishment (CPFR). This has led to initiatives for collaboration and greater visibility of demand information in the supply chain. Although it is clear that visibility of information is beneficial to the entire supply chain (Mason-Jones and Towill 1997, Gavirneni et al. 1999, Mitra and Chatterjee 2004), how the information is to be used to improve inventory planning and reduce supply chain costs is an issue that is worthy of further research. While there have been a number of studies addressing the issue of the value of information sharing, very few have studied the issue under a material requirements planning (MRP) framework or for more than two echelons. Also, in a multi-tier supply chain, information would be available from multiple sources. For example, each echelon may have access to the end-user demand data, as well as to the planned order schedule of its downstream echelon (Lee and Whang 2000). The question then would be to decide what information to use for inventory planning at each echelon. In this paper we investigate through a simulation study the value of various demand information exchange mechanisms in a four-echelon supply chain under a MRP framework. Numerous studies, for example McCullen and Towill (2002) and Disney et al. (2004), have considered a similar form of supply chain, which is composed of several independent organizations representing manufacturer, distributor, wholesaler, and retailer. In the absence of any information sharing, each echelon would develop its own forecasts and plan its inventories based on the history of actual demand from its downstream customer (or echelon). We compare this ‘no-information’ policy or ‘echelon demand history’ mechanism with mechanisms where each echelon has access to (i) end-user demand history and (ii) the planned order schedule of the downstream echelon. Among the information exchange mechanisms, our study found that planning inventories based on the planned downstream order schedules led to the lowest average inventory level for the supply chain. However, use of end-user demand history to forecast and plan inventories at all echelons resulted in the lowest total cost. In addition to the information exchange mechanisms, a simple synchronized replenishment system was also considered and evaluated in this study. Similar to the ‘no-information’ policy, in this synchronized system, no information on end-user demand or planned order schedule is shared within the supply chain. The difference, however, is that instead of allowing the echelons to place an order at any time, this system restricts them to place an order only at fixed time intervals. Assuming that mj is the time interval corresponding to the economic order interval for echelon j, this policy will automatically lead the upper echelon, j þ 1, to place an order only once every (njþ1 Â mj) periods, where njþ1 is a positive integer. This replenishment mechanism is motivated by several cases in reality where sharing demand information in a supply chain may not be feasible for some companies due to various issues such as trust and technology compatibility (Scheraga 2002, D’Avanzo et al. 2003). Our experiments found that synchronized inventory replenishments among the echelons, even without any exchange of demand information, can bring about more benefits and cost reduction than any of the information exchange mechanisms. The paper is organized as follows. We first provide a review of the related research. We then describe the simulation model in detail including the parameters used in the experiment. The findings of the study are discussed thereafter.
Value of information exchange and synchronization in a multi-tier supply chain
5059
Finally, a few concluding remarks and possible directions for future research are provided.
2. Related research When no information is shared across the supply chain and each echelon acts to optimize its own costs, the demand variability in the supply chain becomes amplified as it moves upstream in the supply chain. This phenomenon, known as the bullwhip effect, was first observed by Forrester (1961), and has been studied further by Lee et al. (1997) and later by Chen et al. (2000). One of the solutions proposed to counter the bullwhip effect is to have information sharing across the supply chain. The value of information sharing in the supply chain is an issue that has been studied by many researchers in the recent past. Most have used an inventorytheoretic framework and typically study a two-echelon system. Without information sharing, only the local information on inventory levels (or installation stock information) is used in the replenishment planning process. With information sharing, global inventory information (or echelon stock information) is used for planning the inventory replenishments. The information being shared is normally the inventory information. The probability distribution of the end-user demand is either assumed to be known globally or each echelon just responds only to the demand history or pattern from its immediate downstream echelon. Papers that have adopted this framework include Bourland et al. (1996), Chen (1998), Gavirneni et al. (1999), Cachon and Fisher (2000), Lee et al. (2000), and Raghunathan (1999, 2001, 2003). Most papers, with the possible exception of Cachon and Fisher (2000) and Raghunathan (2001), have concluded that the supply chain inventory costs can be reduced significantly through information sharing. Cachon and Fisher (2000) found that the benefit of information sharing is small compared with the benefit from reduction in lead-time and batch sizes due to information technology. Raghunathan (2001) found that the history of the retailer’s orders itself was a surrogate for its demand and inventory information, and hence the value of information was quite limited. A second stream of related research is coordination (or integration or synchronization) of the inventory management/production schedule across the supply chain. When information alone is shared, the channel members can still act independently but use the additional information to optimize their own costs. Coordination, on the other hand, requires the supply chain members to take ordering decisions in a coordinated fashion. This may not necessarily be the best from the point of view of optimizing the costs of a particular echelon, but it would enhance the profits and reduce the costs for the supply chain as a whole. Typically, in a coordinated supply chain, the supply chain as a whole gains, but some members might see an increase in their costs. Therefore, it requires agreements on how the members who are forced to take locally suboptimal decisions are compensated. A major issue therefore in coordination of supply chains is how the extra benefits are shared among the supply chain members. In the literature on supply chain coordination, the papers again typically assume a two-level structure, mostly with deterministic (but possibly price-elastic) demand.
5060
S. Viswanathan et al.
Typically, a game theory approach is employed by most papers to study mechanisms for achieving coordination between different players in the supply chain and for sharing of the benefits. Some of the papers in this area include Weng (1995), Wei and Krajewski (2000), Chen et al. (2001), Krajeswski and Wei (2001), Viswanathan and Piplani (2001), and Klastorin et al. (2002). There has also been significant interest in supply chain collaboration (which includes exchange of demand information as well as coordination) among practitioners and industry associations (Simatupang et al. 2002, VICS Guide 2002, Larsen et al. 2003, Danese 2006). Though industry efforts such as Vendor Managed Inventory (VMI), and Collaborative Planning Forecasting and Replenishment (CPFR) promoted by bodies such as VICS, ECR committees and consulting companies have propounded the whole umbrella of supply chain collaboration, in reality the focus has been mostly on information exchange (Larsen et al. 2003). Moreover, as mentioned by Danese (2006), there seem to be different ways in which strategies such as CPFR are implemented in practice by different firms. Companies such as Dell and Seagate and Logistics Service Providers such as Fedex claim to provide visibility of demand information to all the supply chain partners. However, it is not very clear as to how the entities in the supply chain use this information to better manage their production and inventories. This is perhaps why there have only been limited success stories beyond pilot projects for these efforts (Larsen et al. 2003). One reason for the apparent lack of success of such efforts to exchange demand information is the reluctance of companies to share what is perceived as sensitive data (Scheraga 2002, Danese 2006). In addition to several information exchange mechanisms, this paper studies a synchronized replenishment scheme that does not require exchange of any demand information.
3. Simulation model We consider a four-echelon supply chain. The most downstream echelon (retailer) faces the demand for the item in each period from the end-user. Each echelon purchases the item from its upstream echelon. When an order in placed, the items are delivered by the upstream echelon provided they have it in stock. Unsatisfied orders are backlogged. The most upstream echelon (manufacturer) is assumed to replenish its inventory from a source with unlimited supply. Therefore, its orders are satisfied without any backlogging. Items supplied by an echelon reach its downstream echelon only after a shipment (or replenishment) lead-time. Each echelon incurs a fixed ordering cost for every order. Holding cost is assessed based on the inventory on hand at the end of each period. Each echelon uses a material requirements planning (MRP) framework for planning its inventories. A dynamic lot-sizing model using a rolling time horizon is used to plan the order schedule. The length of the rolling horizon, N, used in the simulation is 32 weeks. The ordering decisions made at time period T will impact the echelon’s inventory only after T þ L, (where L is the shipment lead-time). Therefore, the dynamic lot-sizing model will use the time frame [T þ L, T þ L þ N] for cost calculations when planning the order schedule at time T. The gross requirement schedule (or projected demand) used by the lot-sizing model will depend on the information exchange mechanism that is adopted. The retailer
Value of information exchange and synchronization in a multi-tier supply chain
5061
faces the actual end-user demand. Therefore, irrespective of the information exchange mechanism used, the gross requirement for the retailer would be the demand forecast created using end-user demand history. For other echelons, when no information is shared, the gross requirement would be the forecast created using the history of demand from its downstream echelon. We refer to the mechanism where no information is shared as the ‘Echelon Demand History’. When information is shared across the supply chain, there are two other possible ways in which the gross requirement schedule for an echelon can be generated for the lot-sizing model. One is to use the projected order schedule of the downstream echelon (which itself is generated by the lot-sizing model at this echelon). This would be the case if the MRP systems are integrated across the four echelons (as if they all belonged to a single organization). We refer to this information exchange mechanism as the ‘Downstream Order Schedule’. The other possible way to generate a gross requirement schedule would be to use a common forecast based on the history of end-user demand for all echelons. We refer to this information exchange mechanism as ‘End-User Demand’. The net requirement for the periods [T þ L, T þ L þ N] is what is used in the dynamic lot-sizing model for generating the projected order schedules at period T. For the periods [T þ L þ 1, T þ L þ N], the net requirement is the same as the gross requirement generated by the forecast or the particular information exchange mechanism. For the period T þ L, the net requirement is the desired safety stock þ gross requirement for the period T þ L minus projected on hand inventory at the end of period T þ L À 1. To be more specific, NRTþL ¼ GRTþL þ SS À ITþLÀ1 , where NRTþL is the net requirement for period T þ L, GRTþL is the gross requirement for period T þ L, SS is the desired safety stock, and ITþLÀ1 is the projected on hand inventory at the end of period T þ L À 1 and ITþLÀ1 ¼ ITÀ1 þ BOTÀ1 þ
TÀ1 X t¼TÀL
Qt À
TþLÀ1 X t¼Tþ1
GRt À DT ,
where Qt is the order quantity placed by the echelon in period t (that is expected to be received in period j þ L), BOT–1 is orders outstanding at T À 1 (i.e. orders placed until T À L À 1, but not shipped by the upstream echelon, i.e. not received until T À 1), and DT is the downstream demand for period T. The algorithm used for solving the N-horizon, dynamic lot-sizing problem in each period for each echelon is a modification to the Silver–Meal (1973) heuristic suggested by Silver and Miltenburg (1984). In the upstream echelons, demand/orders would be lumpy or intermittent (i.e. positive demand interspersed with several zero demands) due to the order batching effect created by the lot-sizing model. The modified Silver-Meal heuristic by Silver and Miltenburg (1984) handles intermittent demand well. When the demand/gross requirement at an echelon is forecasted using the end-user demand history, the exponential smoothing (ES) method is used. Exponential smoothing is used because it is simple to implement (Dekker et al. 2004) and widely adopted in MRP systems (Klassen and Flores 2001). When the echelon demand history is used to forecast the gross requirement, ES may not be
5062
S. Viswanathan et al.
appropriate as it does not work well with lumpy demands (Willemain et al. 1994). Therefore, a modification of the Croston’s (1972) method is used in this case. Croston’s method is a refinement of ES where the quantum of the positive demand and the length of the number of zero demands are forecasted separately. Of course, when the downstream order schedule is the information exchange mechanism used, no forecasting of the gross requirement is necessary for the wholesaler, distributor, and manufacturer. In addition to the three information exchange mechanisms (echelon demand history (or no information sharing), downstream order schedule, and end-user demand history), we also evaluated the performance of a simple synchronized replenishment system. In this ‘synchronized replenishment’ there is no information sharing and, therefore, the echelon demand history is used to forecast the gross requirement at each echelon. However, instead of carrying out the material requirement planning in every period, order scheduling using the lot-sizing model is carried out only at fixed time intervals (this is staggered for the different echelons to take into account the replenishment lead-time). Within the lot-size model itself, orders are allowed to be placed by echelon j only once every mj periods in the horizon of N periods (see figure 1). The order interval, mj, is the time interval corresponding to the economic order interval (that minimizes the total cost per period) for echelon j, assuming that the demand in each period was equal to the average downstream demand. (Note that the average demand can be estimated using the demand history.) The economic order interval for each echelon can essentially be calculated using the Silver–Meal lot-sizing technique. As a particular echelon j places an order only once every mj periods, clearly the best order interval for the upstream echelon, j þ 1, is an integer multiple of mj or njþ1 Â mj periods, where njþ1 is an integer !1. Consequently, it would be easier for the upstream echelons to forecast the demand using a method such as Croston’s. As we will see later, even though the synchronized replenishment system does not have the benefit of sharing demand information, it still outperforms
Echelon-4 (j = 4) (Manufacturer) m4= n4 x m3= 2 x 4 = 8 Information flow Material flow Echelon-3 (j = 3) (Distributor) Echelon-2 (j = 2) (Wholesaler) Echelon-1 (j = 1) (Retailer) End-User’s demands ... 23
m3 = n3 x m2 = 1 x 4 = 4
Order placement with upper echelon
m2= n2 x m1 = 2 x 2 = 4
m1 = 2 weeks 1 2 3 4 5 6 7 8 9 10
Demand occurrence (Weekly)
Figure 1. Example of ordering pattern for each echelon using the synchronized replenishment mechanism.
Value of information exchange and synchronization in a multi-tier supply chain
5063
all the information sharing mechanisms both in terms of the average inventory as well as total cost to the supply chain. In each time period of the simulation run, the sequence of activities is as follows. First, the shipments from the upstream echelon (shipped L periods ago) are received into the inventory. Next, the demand for the period arises, and it is satisfied if there is adequate stock and backlogged otherwise. Then, the backlogged demand from the previous periods, if any, is satisfied, if there is stock. It is assumed that partial fulfillment of an order is allowed. The forecasts (for demand/gross requirement) are then updated, and followed by computing the net requirement. Finally, the order schedule using the lot-sizing model is calculated and orders corresponding to the current period are placed with the upstream echelon. In each period, the simulation is carried out first for the most downstream echelon (retailer), and then for each upstream echelon, i.e. wholesaler, distributor, and manufacturer in that order. For the upstream echelons, the order placed by the immediate downstream echelon becomes the demand. The end-user demand (or demand for the retailer) is randomly generated using a first-order autoregressive [AR(1)] process. The AR(1) demand process, which has a non-zero lag-1 autocorrelation, has been adopted by numerous authors to study various issues in supply chain management (see, for instance, Chen et al. (2000), Lee et al. (2000), and Disney and Grubbstrom (2004)), as it represents many demand processes in practice (Nahmias 1993, Chopra and Meindl 2001). AR(1) is different from the normally distributed demand process in that the AR(1) process exhibits some non-stationary characteristics that justify the use of a forecasting mechanism (Disney et al. 2003). Its mathematical representation is given as follows (Box and Jenkins 1976, p. 56): d1 ¼ þ " 1 , dt ¼ ð1 À Þ þ dtÀ1 þ "t , where is a non-negative constant, | |51 and is constant at all times, and "t 2 is normally distributed with zero mean and variance " (note that, if ¼ 0, the demand becomes independently and identically distributed (i.i.d.) with mean 2 and variance " ). The simulation test bed is developed using MS Visual Cþþ 6.0. For each set of experimental parameters, the simulation is run for a total of 1300 periods. The initial set of 260 periods (or 20% of the run length) is considered as the warm-up period and the performance statistics/measures are calculated only for the remaining 1040 periods. The parameters used in the simulation experiments are provided in table 1. For a particular set of experimental parameters, and for a particular mode of information exchange mechanism (or for the synchronized replenishment), the safety stock for a particular echelon was kept fixed throughout the simulation run. The simulation was basically run (for a particular set of experimental parameters) under different values of the safety stock until the safety stock that achieves the desired fill rate (e.g. 94% for the base case) for all four echelons is found. The average inventory level and the total cost of the supply chain are calculated corresponding to this level of safety stock. Note that this involves running the
5064 Table 1. Parameter
S. Viswanathan et al. Experimental design parameters. Value 1300 260 32 1000 0.2 2 (basea), 3, 4, 5, 6 200 (base), 400, 600, 800, 1000, 1200 94 (base), 95, 96, 97, 98, 99 À0.6, À0.4, À0.2 (base), 0.2, 0.4, 0.6
Run-length (periods) Warm-up period (periods) Length of rolling horizon (periods) Mean of end-user demand, (units/period) Smoothing constant value used for forecasting in ES and modified Croston’s method Replenishment lead-time, L (periods) Standard deviation of the error term of the demand processb, " (units) Desired fill rate for all echelons, F (%) Serial correlation coefficient used in AR(1) demand process,
a Base indicates the value that is used when the experiment is carried out by varying just one particular parameter. For example, when the performance of all mechanisms is investigated against different values of L, other independent variables such as ", F, and are set to their respective base values of 200, 94, and –0.2. b 2 Standard deviation of the end-user demand ( d) can be obtained by d ¼ ½" =ð1 À 2 Þ?1=2 . Therefore, if ¼ À0.2, we can respectively translate the values of " used in this experiment to d ¼ {204.12 (base), 408.25, 612.37, 816.5, 1020.62, 1224.75}.
simulation repeatedly for a particular parameter set, as the correct safety stock level for each of the four echelons needs to be determined. The cost elements included in the total cost calculation are the ordering and holding costs at each echelon. As the same service level is imposed for all the information exchange mechanisms, no penalty cost is included in the model. The various combinations of holding and ordering cost values used in this study are shown in table 2. The performance of the various information exchange mechanisms and the synchronized replenishment system is measured by the average inventory level and total cost to the supply chain. These two are therefore the dependent variables in the experiments. The independent variables are the various parameters used in the experiments, namely the standard deviation of the error term of the demand process ( "), serial correlation coefficient of the demand process ( ), replenishment lead-time (L), desired fill rate (F), and ordering cost (A). With four different information exchange mechanisms, ten combinations of ordering and holding costs, and the different values for the various parameters, the simulation experiment conducted in this study involved 1000 different experimental scenarios. The system being considered in this study is generic enough to be applied to any distribution channel or manufacturing supply chain. Transformation of the products through the supply chain is assumed implicitly through the different, echelon-specific, holding costs and ordering costs. Of course, there is effectively only a one-for-one transformation or bill of materials. One could of course consider more complex bill of materials and transformation structures, but as the objective of the research was to gain insights into the impact of various information exchange and coordination mechanisms, we did not consider such complexities. The data presented
Value of information exchange and synchronization in a multi-tier supply chain Table 2. Case 1 Holding and ordering cost values used in the experiments. Retailer 0.5 3000 0.5 3000 0.5 2550 0.5 2100 0.5 3000 0.5 3000 0.5 2550 0.5 2100 0.5 3000 0.5 3000 Wholesaler 0.45 3000 0.5 3000 0.5 2700 0.5 2400 0.5 2850 0.5 2700 0.45 2700 0.45 2400 0.45 2850 0.45 2700 Distributor 0.4 3000 0.5 3000 0.5 2850 0.5 2700 0.5 2700 0.5 2400 0.4 2850 0.4 2700 0.4 2700 0.4 2400
5065
Parameter Holding cost, H ($/unit/period)a Ordering cost, A ($/order)b H A H A H A H A H A H A H A H A H A
Manufacturer 0.35 3000 0.5 3000 0.5 3000 0.5 3000 0.5 2550 0.5 2100 0.35 3000 0.35 3000 0.35 2550 0.35 2100
2 3 4 5 6 7 8 9 10
a
Assuming a unit cost of $100 and length of a period to be one week, this works to a holding cost rate of about 26% per year at the retailer’s site. b For every case, six different levels (3000, 4000, 5000, 6000, 7000, and 8000) of ordering cost are evaluated. The ordering costs for the other echelons are adjusted proportionately in each case for the different levels.
in tables 1 and 2 are a typical representation of data found in practice for products such as fast moving consumer goods.
4. Simulation results We now discuss the results of the simulation study and their managerial implications. We first discuss the relative performance of the different information exchange mechanisms and how it is impacted by particular experimental parameters. In the experimental results that are presented in figures 2–5, the total cost (or the average inventory level) for the whole supply chain is plotted against a particular experimental parameter. For these figures, the value of the particular parameter alone is varied and the rest of the parameters are kept fixed at their base values. The figures only report the results for the holding and ordering cost combination under case 1 of table 2. The results were quite similar for other combinations of the holding and ordering costs and hence are not reported. We first report the performance of the different information exchange mechanisms for different values of end-user demand variability in figures 2 and 3. In figure 2, the average supply chain inventory level for the different mechanisms is
5066
Average inventory level (units) 22000 18000 14000 10000 6000 2000 200
S. Viswanathan et al.
400
600 800 Variability of end-user demand (se)
1000
1200
End-user demand Synchronized replenishment
Downstream order schedule Echelon demand history
Figure 2. Plot of the average inventory level for the supply chain against end-user demand variability ( ") for different information exchange mechanisms.
12000 Total supply chain cost ($)
10000
8000
6000
4000 200
400
600
800
1000
1200
Variability of end-user demand (se) End-user demand Synchronized replenishment Downstream order schedule Echelon demand history
Figure 3. Plot of the total supply chain cost (inventory holding þ ordering) against end-user demand variability ( ") for different information exchange mechanisms.
plotted against ", the standard deviation of the error term in the end-user demand process (as mentioned in table 1, " is directly related to the variability or standard deviation of the end-user demand). As can be seen, no information sharing (or echelon demand history) resulted in the highest average supply chain inventory level for most values of ".The use of downstream order schedule resulted in a lower average inventory than using a common single forecast based on the history of enduser demand. As discussed earlier, if the MRP systems for the different echelons were integrated (as if it were a single organization), then the gross requirement for the
Value of information exchange and synchronization in a multi-tier supply chain
12000 Total supply chain cost ($)
5067
10000
8000
6000
4000 2 3 4 Replenishment lead-time (weeks) 5 6
End-user demand Synchronized replenishment
Downstream order schedule Echelon demand history
Figure 4. Plot of the total supply chain cost against the replenishment lead-time for different information exchange mechanisms.
16000 Total supply chain cost ($) 14000 12000 10000 8000 6000 4000 3000
4000
5000 6000 Ordering cost at each echelon ($)
7000
8000
End-user demand Synchronized replenishment
Downstream order schedule Echelon demand history
Figure 5. Plot of the total supply chain cost against the echelon ordering cost for different information exchange mechanisms.
upstream echelons would be based on the downstream order schedules. Figure 2 suggests that the supply chain inventory is minimized by integrating the MRP systems of the different echelons and using the downstream order schedule as the gross requirement for each echelon. This finding justifies the interest among companies to provide information visibility and integrate the information systems across the supply chain partners.
5068
S. Viswanathan et al.
However, when minimizing the total cost per period (holding cost þ ordering cost) of the supply chain is the objective, the downstream order schedule does not do that well as an information exchange mechanism. This fact is demonstrated in figure 3, which plots the total supply chain cost per period against " for the various information exchange mechanisms. From this figure it is clear that, for most values of ", the use of a common forecast based on history of end-user demand resulted in a lower total supply chain cost than use of downstream order schedules. In fact, for low values of the end-user demand variability, the total supply chain cost using the downstream order schedule mechanism was nearly identical to that using the echelon demand history mechanism or no information sharing. Only when the demand variability was very high (or, in other words, end-user demand information was not very reliable) did the total cost using downstream order schedule information do better than the total cost using end-user demand history. One reason for the poor (total cost) performance of the downstream order schedule mechanism could be due to the so-called MRP nervousness. MRP nervousness is the situation where there is frequent change to the planned replenishment orders, due to the use of lot-sizing algorithms to determine order schedules in every period under a rolling horizon framework. There is a large amount of literature that has addressed the issue of MRP nervousness. Researchers who have studied this include Kropp et al. (1983), Blackburn et al. (1986), Sridharan et al. (1987), Sridharan and Berry (1990), and Vollman et al. (1997). Even though the average inventory level was lower for the downstream order schedule mechanism, the total costs were higher than that for the mechanism using end-user demand history except when the demand variability was extremely high. Clearly, the ordering cost or the average number of orders under downstream order schedule is much higher than that for end-user demand history. One way to measure the impact of MRP nervousness on the total cost is to develop a nervousness index. We define the nervousness index as the ratio of the average number of orders per period calculated by the lot-sizing method in each planning period to the actual number of orders per period observed in the simulation. Table 3 reports the average nervousness index based on the simulation results for the different echelons for the various information exchange mechanisms. As can be seen from the table, for mechanisms other than the downstream order schedule, the index value was close to one, implying that the actual number of orders per period was close to the planned orders as per the lot-sizing heuristic. For the downstream
Table 3.
Nervousness index observed in the simulation experiments. Average for the supply chain 0.988 0.993 1.663 0.980
Mechanisma EDH EUD DOS SR
Retailer 1.13 1.13 1.13 0.99
Wholesaler 0.96 1.02 1.71 0.99
Distributor 0.98 0.93 1.81 0.98
Manufacturer 0.88 0.89 2.00 0.96
a EDH, echelon demand history; EUD, end-user demand; DOS, downstream order schedule; SR, synchronized replenishment.
Value of information exchange and synchronization in a multi-tier supply chain
5069
order schedule mechanism alone, the nervousness index value was much higher, implying that the actual number of orders per period was much higher than the planned orders. One common suggestion to alleviate MRP nervousness is to freeze the master production schedule for a certain period. In the context of our research, it appears that the total supply chain cost can be reduced further by use of a common forecast using end-user demand history or through a synchronized replenishment scheme (without any exchange of demand information) rather than by integrating the MRP systems of the different echelons. Figure 4 plots the total cost of the whole supply chain for the various information exchange mechanisms against the replenishment lead-time. The pattern in this figure was consistent with those in figure 3. That is, the total supply chain cost through the use of (forecast based on) end-user demand was consistently lower than that through the use of downstream schedules. The total costs for the various information exchange mechanisms plotted against the desired fill rate (F) and the serial correlation coefficient of the end-user demand ( ) were also similar to the earlier figures; therefore, we do not present them in the paper. Figure 5 plots the total supply chain cost for the various mechanisms against the ordering cost for the echelons. In this case again, the total supply chain cost through the use of forecasts based on end-user demand was lower than that through the use of downstream order schedules. However, for higher values of the ordering cost, the total cost through the use of echelon demand history was lower than the cost through the use of either of the other two information exchange mechanisms. This is probably because of the fact that the effect of MRP nervousness becomes more prominent as ordering cost increases. In other words, information sharing had no value when the echelon ordering costs were higher. In all the four figures, the synchronized replenishment scheme (without any exchange of demand information) consistently resulted in a lower supply chain cost than any of the information exchange mechanisms. As explained earlier, the synchronized replenishment system is a simple coordination scheme in which all echelons commit to reviewing and placing orders only at fixed time intervals, but do not share any end-user demand or planned order schedule information. The simulation results therefore seem to suggest that there is substantially more value in attempting to synchronize the replenishment epochs rather than just to share demand information. The superiority of the synchronized replenishment scheme in the simulation study also seems to tie in with the research findings in the MRP nervousness literature, where freezing the master production schedule (MPS) is offered as one of the solutions to the problem. By committing to review and place orders only at fixed time intervals (in the synchronized replenishment scheme), the orders are essentially frozen for that period. However, the significant difference between the synchronized replenishment scheme and the freezing of MPS is that, in the MRP nervousness literature, all the different product levels belong to the same MRP system, and the gross requirement at a level is determined by the downstream order schedules. In the synchronized replenishment scheme that was used in our simulation study, there was only synchronization, but no sharing of demand information. The echelon demands were forecast using the echelon demand history rather than the downstream order schedules.
5070
S. Viswanathan et al.
One question that would arise is what would be the best or optimal ordering policy for an integrated supply chain system in this context. When the end-user demands are known with certainty, an optimal algorithm has been proposed by Graves (1981). However, this method is computationally prohibitive, especially as the length of the planning horizon increases. Besides, the algorithm has to be run in every period of the simulation to calculate the schedule in a rolling horizon context. Heuristics have been proposed by McLaren and Whybark (1976) and Blackburn and Millen (1982). We had implemented these two heuristics in our experimental study, but their total costs were higher than most other information exchange mechanisms for most cases. As mentioned, these methods were developed for finding the optimal order schedules at a particular point in time for deterministic demands. Apart from the fact that the demands are varying in a stochastic manner, the schedules are calculated in every period in a rolling horizon framework, which can cause MRP nervousness. This is probably why these heuristics performed poorly in our study (the average nervousness index for lot-sizing based on these heuristics was 2.18, which is much higher than that for all other mechanisms, including downstream order schedule). As the performance of the two heuristics was inferior to even the echelon demand history mechanism, we have chosen not to report these in the paper. When the end-user demands are stochastic and stationary, optimal policies for an integrated supply chain have not been found to date. The best policy in this case seems to be the (R, nQ) policy suggested by Chen and Zheng (1998). In this policy, the order quantity for a higher echelon is an integer multiple
of the order quantity at the lower echelon (Q). However, their algorithm can be applied only for Poisson or compound Poisson demands. Moreover, this algorithm may not be very appropriate in a MRP context where the demand forecasts evolve periodically. The synchronized replenishment mechanism used in our study in a sense mimics the (R, nQ) policy. Orders are placed by the retailer only at fixed intervals and the order intervals for the other echelons are at integer multiples of this interval (after making the time shift for the lead-time). One could possibly try to improve the synchronized replenishment scheme by trying to use downstream order schedule instead of forecasts based on echelon order history. However, as discussed in the Introduction, the motivation for this scheme was based on the fact that sharing of demand information is difficult in many supply chains. The average of the inventory level and total cost incurred at each echelon for all the experiments are presented in table 4. As can be seen from table 4, with the exception of the retailer, the cost at all the other echelons decreased substantially under synchronized replenishment. Table 5 reveals the average savings in the total cost and inventory level generated at each echelon using different information exchange mechanisms, compared with the use of just echelon demand history (where there is no information exchange or coordination). Compared with the case where echelon demand history is used, use of downstream order schedules reduced the average supply chain inventory level by as much as 29.41%, whereas use of a common forecast based on end-user demand history reduced the average supply chain inventory level by only 17.69%. However, use of end-user demand history resulted in a 15.02% reduction in the total supply chain cost, whereas use of downstream order schedules resulted only in a 1.52% reduction in the total cost.Value of information exchange and synchronization in a multi-tier supply chain Table 4.
5071
Summarized results of the simulation experiments: total cost and average inventory at the four echelons for different information exchange mechanisms. Total for the supply chain 12024.05 9897.01 8488.19 4829.33 9058.46 7698.05 8920.94 5448.83
Dependent variable Average inventory level (units) Total cost ($)
Mechanisma EDH EUD DOS SR EDH EUD DOS SR
Retailer 1170.35 1187.39 1171.19 1800.17 1587.43 1596.03 1589.04 1701.66
Wholesaler 2122.04 2171.21 2084.91 637.86 1943.89 1881.58 2409.84 1106.24
Distributor 3594.90 2928.53 2643.85 949.54 2544.56 2041.96 2528.85 1231.21
Manufacturer 5136.77 3609.88 2588.24 1441.76 2982.59 2178.49 2393.22 1409.72
a EDH, echelon demand history; EUD, end-user demand; DOS, downstream order schedule; SR, synchronized replenishment.
Table 5.
Summarized results of the simulation experiments: percentage savings obtained by different mechanisms. Retailer (%) 0 À1.46 À0.07 À53.82 0 À0.54 À0.10 À7.20 Wholesaler (%) 0 À2.32 1.75 69.94 0 3.21 À23.97 43.09 Distributor (%) 0 18.54 26.46 73.59 0 19.75 0.62 51.61 Manufacturer (%) 0 29.73 49.61 71.93 0 26.96 19.76 52.74 Total savings (%) 0 17.69 29.41 59.84 0 15.02 1.52 39.85
Dependent variable Average inventory level (units) Total cost ($)
Mechanisma EDH EUD DOS SR EDH EUD DOS SR
a EDH, echelon demand history; EUD, end-user demand; DOS, downstream order schedule; SR, synchronized replenishment.
Finally, the synchronized replenishment mechanism resulted in a 59.84% reduction in average supply chain inventory level and a 39.85% reduction in the total cost.
5. Summary and conclusions We have investigated through a simulation study the value of various information exchange mechanisms in a multi-tier supply chain under a MRP framework. In the absence of any sharing of demand information, each echelon develops its own forecasts and plans its inventories based on the history of actual demand from its downstream customer (or echelon). We compared this policy with policies where each echelon had access to (i) end-user demand history and (ii) the planned order
5072
S. Viswanathan et al.
schedule of the downstream echelon. Among the information exchange mechanisms, integrating the MRP systems of the various echelons (or replenishing inventories based on the planned downstream order schedules) resulted in the lowest average inventory level for the supply chain. However, use of end-user demand history to forecast and plan inventories at all echelons resulted in the lowest total cost. In addition to the mechanisms for exchange of demand information, a simple synchronized replenishment system was considered and evaluated in the study. In this synchronized replenishment system, each echelon is restricted to place an order only after a fixed time interval. The study found that synchronized inventory replenishments among the echelons, even without any exchange of demand information, can bring about more benefits and cost reduction than any of the information exchange mechanisms. While the research literature seems to have had equal amounts of focus on both information sharing and coordination in supply chains, practitioners have tended to focus more on information sharing and information visibility. One of the deterrents to sharing information freely has been the lack of trust and willingness among entities in the supply chain. Our results from the simulation study suggest that, to enjoy the full benefits of collaboration, practitioners should focus more on synchronization (or coordination arrangements) rather than just on information visibility or sharing information. Of course, attempting synchronization has its own sets of deterrents, such as how to share the benefits, and uncertainty regarding any short-term negative cost impact; however, these can possibly be overcome more easily than the uneasiness involved in sharing sensitive demand information. The findings of this study are applicable only for the MRP framework and structure of the system considered in the simulation and cannot easily be generalized across other supply chain structures. However, such limitations would exist for most previous studies on supply chain coordination found in the literature. It should also be pointed out that the multi-echelon supply chain structure with a MRP framework used in this study is very commonly found in practice. The proposed synchronization scheme also requires some information exchange. All entities need to indicate their ordering interval and stick to it. However, they need not reveal their downstream demand information to their upstream supplier. As already discussed, the demand information would be perceived as sensitive, especially if the upstream entity is also supplying the goods to the downstream entity’s competitor. Revealing the ordering interval would not be perceived as that sensitive. Most of the research on information sharing and supply chain coordination has been from an inventory theoretic point of view rather than under a rolling horizon, MRP framework. This paper has attempted to fill this gap. However, one issue that has not been addressed in the paper is the development of mechanisms or rules for sharing the benefits of synchronization and information sharing under a MRP framework. For example, even though synchronized replenishment resulted in cost savings of 39.85% for the entire supply chain, the total cost for the retailer actually increased by 7.20% (see table 5). Clearly, appropriate mechanisms are needed to allocate the benefits from coordination and information sharing. This could be a possible issue for future research.
Value of information exchange and synchronization in a multi-tier supply chain
5073
References
Blackburn, J.D., Kropp, D.H. and Millen, R.A., A comparison of strategies to dampen nervousness in MRP systems. Mgmt. Sci., 1986, 32, 413–429. Blackburn, J.D. and Millen, R.A., Improved heuristics for multi-stage requirements planning systems. Mgmt. Sci., 1982, 28, 44–56. Bourland, K.E., Powell, S.G. and Pyke, D.F., Exploiting timely demand information to reduce inventories. Eur. J. Oper. Res., 1996, 92, 239–253. Box, G.E.P. and Jenkins, G.M., Time Series Analysis: Forecasting and Control, 2nd ed., 1976 (Holden-Day: San Francisco). Cachon, G.P. and Fisher, M., Supply chain inventory management and the value of shared information. Mgmt. Sci., 2000, 46, 1032–1048. Chen, F., Echelon reorder points, installation reorder points, and the value of centralized demand information. Mgmt. Sci., 1998, 44, 221–234. Chen, F., Drezner, Z., Ryan, J.K. and Simchi-Levi, D., Quantifying the bullwhip effect in a simple supply chain: the impact of forecasting, lead times, and information. Mgmt. Sci., 2000, 46, 436–443. Chen, F., Federgruen, A. and Zheng, Y., Coordination mechanisms for distribution system with one supplier and multiple retailers. Mgmt. Sci., 2001, 47, 693–708. Chen, F. and Zheng, Y., Near optimal echelon stock (R, nQ) policies in multi-stage serial systems. Oper. Res., 1998, 46, 592–602. Chopra, S. and Meindl, P., Supply Chain Management, 2001 (Prentice-Hall: Engelwood Cliffs, NJ). Croston, J.D., Forecasting and stock control for intermittent demand. Oper. Res. Q., 1972, 23, 289–303. Danese, P., Collaboration forms, information and communication technologies and coordination mechanisms in CPFR. Int. J. Prod. Res., 2006, 44, 3207–3226. D’Avanzo, R., von Lewinski, H. and Van Wassenhove, L.N., The link between supply chain and financial performance. Suppl. Chain Mgmt. Rev., 2003, Nov/Dec, 40–47. Dekker, M., van Donselaar, K.H. and Ouwehand, P., How to use aggregation and combined forecasting to improve seasonal demand forecasts. Int. J. Prod. Econ., 2004, 90, 151–167. Disney, S.M., Farasyn, I., Lambrecht, M., Towill, D.R. and Van de Velde, W., Taming the bullwhip whilst watching customer service, in EURO/INFORMS Joint International Meeting, 2003 (Istanbul). Disney, S.M. and Grubbstrom, R.W., Economic consequences of a production and inventory control policy. Int. J. Prod. Res., 2004, 42, 3419–3431. Disney, S.M., Naim, M.M. and Potter, A., Assessing the impact of e-business on supply chain dynamics. Int. J. Prod. Econ., 2004, 89, 109–118. Forrester, J.W., Industrial Dynamics, 1961 (MIT Press: Cambridge, MA). Gavirneni, S., Kapuscinski, R. and Tayur, S., Value of information in capacitated supply chains. Mgmt. Sci., 1999, 45, 16–24. Graves, S.C., Multistage lot-sizing: An iterative procedure. In TIMS Studies in Management Science. Multi-Level Production/Inventory Systems: Theory and Practice, edited by L.B. Schwarz, Vol. 6, pp. 95–109, 1981 (Amsterdam). Klassen, R.D. and Flores, B.E., Forecasting practices of Canadian firms: survey results and comparisons. Int. J. Prod. Econ., 2001, 70, 163–174. Klastorin, T.D., Moinzadeh, K. and Son, J., Coordinating orders in supply chains through price discounts. IIE Trans., 2002, 34, 679–689. Krajewski, L. and Wei, J.C., The value of production schedule integration in supply chains. Decis. Sci., 2001, 32, 601–634. Kropp, D.H., Carlson, R.C. and Jucker, J.V., Heuristic lot-sizing approaches for dealing with MRP system nervousness. Decis. Sci., 1983, 14, 156–169. Larsen, T.S., Thernøe, C. and Andresen, C., Supply chain collaboration: theoretical perspective and empirical evidence. Int. J. Phys. Distr. Log. Mgmt., 2003, 33, 531–549.
5074
S. Viswanathan et al.
Lee, H.L., Padmanabhan, P. and Whang, S., Information distortion in a supply chain: the bullwhip effect. Mgmt. Sci., 1997, 43, 546–558. Lee, H.L., So, K.C. and Tang, C.S., The value of information sharing in a two-level supply chain. Mgmt. Sci., 2000, 46, 626–664. Lee, H.L. and Whang, S., Information sharing in a supply chain. Int. J. Technol. Mgmt., 2000, 20, 373–387. Mason-Jones, R. and Towill, D.R., Information enrichment: designing the supply chain for competitive advantage. Suppl. Chain Mgmt., 1997, 2, 137–148. McCullen, P. and Towill, D., Diagnosis and reduction of bullwhip in supply chains. Int. J. Suppl. Chain Mgmt., 2002, 7, 164–179. McLaren, B.J. and Whybark, D.C., Multi-level lot-sizing procedures in a material requirements planning environment. Discussion Paper No. 64, Indiana University, 1976. Mitra, S. and Chatterjee, A.K., Leveraging information in multi-echelon inventory systems. Eur. J. Oper. Res., 2004, 152, 263–280. Nahmias, S., Production and Operations Analysis, 1993 (Irwin: Boston). Raghunathan, S., Interorganizational collaborative forecasting and replenishment systems and supply chain implications. Decis. Sci., 1999, 30, 1053–1071. Raghunathan, S., Information sharing in a supply chain: a note on its value when demand is nonstationary. Mgmt. Sci., 2001, 47, 605–610. Raghunathan, S., Impact of demand correlation on the value of and incentives for information sharing in a supply chain. Eur. J. Oper. Res., 2003, 146, 634–649. Scheraga, D., Disappointment reigns. Chain Store Age, 2002, August, 83–85. Silver, E.A. and Meal, H., A heuristic for selecting lot size quantities for the case of a deterministic time-varying demand rate and discrete opportunities for replenishment. Prod. Invent. Mgmt., 1973, 14, 64–74. Silver, E.A. and Miltenburg, J., Two modifications of the silver–meal lot sizing heuristic. INFOR, 1984, 22, 56–69. Simatupang, T.M., Wright, A.C. and Sridharan, R., The knowledge of coordination for supply chain integration. Bus. Process Mgmt J., 2002, 8, 289–308. Sridharan, V. and Berry, W.L., Freezing the master production schedule under demand uncertainty. Decis. Sci., 1990, 21, 97–120. Sridharan, V., Berry, W.L. and Udayabhanu, V., Freezing the master production schedule under rolling planning horizons. Mgmt. Sci., 1987, 33, 1137–1149. Viswanathan, S. and Piplani, R., Coordinating supply chain inventories through common replenishment epochs. Eur. J. Oper. Res., 2001, 129, 277–286. Vollman, T.E., Berry, W.L. and Whybark, D.C., Manufacturing Planning and Control Systems, 4th ed., 1997 (Irwin: Homewood). Voluntary Interindustry Commerce Standards (VICS), Collaborative planning, forecasting and replenishment guidelines, version 2.0. Available online at: http://www.cpfr.org, 2002. Wei, J.C. and Krajewski, L., A model for comparing supply chain schedule integration approaches. Int. J. Prod. Res., 2000, 38, 2099–2123. Weng, Z., Channel coordination and quantity discounts. Mgmt. Sci., 1995, 41, 1509–1522. Willemain, T.R., Smart, C.N., Shockor, J.H. and DeSautels, P.A., Forecasting intermittent demand in manufacturing: a comparative evaluation of Croston. Int. J. Forecast., 1994, 10, 529–538.
doc_372405476.pdf