ARTICLE IN PRESS

Int. J. Production Economics 113 (2008) 575586 www.elsevier.com/locate/ijpe

On the benets of CPFR and VMI: A comparative simulation study

Kazim Sari
Department of International Logistics and Transportation, Faculty of Economics and Administrative Sciences, Beykent University, Sisli Ayazaga Mah., Hadimkoru Yolu Mevkii, 34396 Sisli, Istanbul, Turkey Received 1 December 2006; accepted 4 October 2007 Available online 10 March 2008

Abstract This paper aims to help managers of a supply chain to determine an appropriate level of collaboration according to their specic business conditions. For this purpose, a comprehensive simulation model representing two popular supply chain initiatives, that are collaborative planning, forecasting and replenishment (CPFR) and vendor-managed inventory (VMI), is constructed. In addition, a traditionally managed supply chain (TSS) is also included in the model as a benchmark. The results indicate that benets of CPFR are always higher than VMI. However, we also realize that under certain conditions, the gap between the performances of CPFR and VMI does not rationalize the additional resources required for CPFR. Especially, when the lead time is short and/or when available manufacturing capacity is tight, a careful consideration has to be given on the selection of an appropriate collaboration mode. r 2008 Elsevier B.V. All rights reserved.
Keywords: Supply chain collaboration; CPFR; VMI; Simulation

1. Introduction A supply chain, consisting of several organizations with different and sometimes conicting objectives, is a complex network of facilities designed to produce and distribute products according to customers demands. By coordinating different enterprises along the logistics network or establishing business partnerships, supply chain management (SCM) is concerned with nding the best strategy for the whole supply chain (Simchi-Levi et al., 2003, p. 2). Nevertheless, nding the best strategy in this complex network of facilities is not an easy task. It
Tel.: +90 212 444 1997; fax: +90 212 867 5060.

E-mail address: kazimsari@msn.com

requires intensive communication and coordination among trading partners so that material ow along the supply chain is optimized as well as information ow. Fortunately, with the emergence of new management paradigms at the beginning of 1980s, e.g. Lean Thinking, Total Quality Management and Partnership Sourcing Programme, much progress has been made in the coordination of material ow (Mason-Jones and Towill, 2000; Simchi-Levi et al., 2003, p. 5). However, an equal attention has not been paid to the optimization of information ow. This ignorance of the information ow has contributed to one important problem in supply chain literature, which is called bullwhip effect (Lee et al., 1997a, b). The bullwhip effect represents the phenomenon where orders to supplier tend to have

0925-5273/$ - see front matter r 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpe.2007.10.021

ARTICLE IN PRESS
576 K. Sari / Int. J. Production Economics 113 (2008) 575586

a larger variance than sales to the buyer (Lee et al., 1997a, b). In return, high inventory levels and poor customer service rates are typical symptoms of the bullwhip effect (Metters, 1997; Chopra and Meindl, 2001, p. 1363). Today, SCM researchers indicate that elimination the bullwhip effect plays a vital role for supply chain enterprises to gain competitive advantage. Most of the researchers focusing on remedies for coping with the bullwhip effect dictate that sharing retail-level information (i.e. point of sales (pos) data) between supply chain members is a prerequisite for elimination of the bullwhip effect, see e.g. Lee et al. (1997a), Chen et al. (2000a, b), McCullen and Towill (2002), Dejonckheere et al. (2004), Ouyang (2006) and Li et al. (2006). Nevertheless, retailers, most of the time, do not desire to engage in information sharing because it provides ignorable levels of benets for them, see e.g. Lee et al. (2000), Yu et al. (2001, 2002), Zhao et al. (2002a, b). Therefore, this requires upstream members (e.g. suppliers or manufacturers) to offer incentives for retailers in return for information sharing. Vendormanaged inventory (VMI) and collaborative planning, forecasting and replenishment (CPFR) are the partnership programs primarily developed to encourage retailers to share information, see e.g. Lee et al. (1997b) and Disney and Towill (2003a, b). VMI, also known as continuous replenishment or supplier-managed inventory, is one of the most widely discussed partnering initiatives for encouraging collaboration and information sharing among trading partners (Angulo et al., 2004). Popularized in the late 1980s by Wal-Mart and Procter & Gamble (Waller et al., 1999), it was subsequently implemented by many other leading companies from different industries, such as Glaxosmithkline (Danese, 2004), Electrolux Italia (De Toni and Zamolo, 2005), Nestle and Tesco (Watson, 2005), Boeing and Alcoa (Micheau, 2005), etc. It is a supply chain initiative where the vendor decides on the appropriate inventory levels of each of the products and the appropriate inventory policies to maintain those levels. The retailer provides the vendor with access to its real-time inventory level. In this partnership program, the retailer may set certain service level and/or self-space requirements, which are then taken into consideration by the vendor. That is, in a VMI system, the retailers role shifts from managing inventory to simply renting retailing space (Simchi-Levi et al., 2003, p. 154; Mishra and Raghunathan, 2004).

VMI offers a competitive advantage for retailers because it results in higher product availability and service level as well as lower inventory monitoring and ordering cost (Waller et al., 1999; Achabal et al., 2000). For vendors, on the other hand, it results in reduced bullwhip effect (Lee et al., 1997b; Disney and Towill, 2003a, b) and better utilization of manufacturing capacity (Waller et al., 1999), as well as better synchronization of replenishment planning (Waller et al., 1999; C etinkaya and Lee, 2000). While many benets have been identied in the literature, there are also a number of challenges that may exist in practice and that can potentially reduce the benets obtained from VMI or lead to failures in VMI programs. For instance, Spartan Stores, a grocery chain, shut down its VMI effort about 1 year after due in part VMI vendors inability to deal with product promotions (Simchi-Levi et al., 2003, p. 161). Similarly, Kmart cut a substantial amount of VMI contracts because Kmart is not satised with the forecasting ability of VMI vendors (Fiddis, 1997). Consequently, many studies have been carried out to investigate the effectiveness of VMI programs under different conditions. For instance, Kuk (2004) empirically tested the acclaimed benets of VMI programs in electronics industry. Similarly, Sari (2007) used a simulation model to evaluate the benets of VMI under different market conditions. Dong and Xu (2002), on the other hand, evaluated the value of VMI programs both for suppliers and buyers. Most of these studies show that ineffective usage of retail-level information is one major limitation of VMI programs (see e.g. Aviv, 2002; Ovalle and Marquez, 2003; Angulo et al., 2004; Yao et al., 2007). That is, since retailers are closer to the marketplace, they may have better knowledge about customer behaviors, products and marketplace. However, in most, if not all, VMI programs, this unique knowledge of the retailers cannot be joined into inventory decisions. This is because in a typical VMI program, retailers are excluded from demand forecasting process. Indeed, in a VMI system, the responsibilities of the retailers are noting more than sharing sales and inventory data. CPFR, on the other hand, can solve majority of the problems that are encountered in adaptation of VMI because it requires all members of a supply chain to jointly develop demand forecasts, production and purchasing plans, and inventory replenishments (Aviv, 2002). It is a business practice that combines the intelligence of multiple trading partners in the planning and fullment of customer

ARTICLE IN PRESS
K. Sari / Int. J. Production Economics 113 (2008) 575586 577

demand (CPFR Workgroup, 2002). CPFR adds value to the supply chain in the form of reduced inventory and increased customer service level by achieving better match of demand and supply (Foote and Krishnamurthi, 2001; Aghazadeh, 2003; Aichlmayr, 2003; Fliedner, 2003). Nonetheless, successful implementation of CPFR is not an easy task. It requires more intensive organizational resources than VMI as well as mutual trust of multiple trading partners (Barratt and Oliveria, 2001; Fliedner, 2003). Furthermore, dramatic changes are also required in usual ways of doing business for CPFR implementation. Consequently, examination of the previous literature reveals the fact that adaptation of higher levels of collaboration among members of a supply chain creates greater benets for the supply chain. On the other hand, we also see that development and operational costs of a highly integrated collaboration is also higher. That is, while CPFR eliminates most of the problems encountered in VMI programs; investment and operation costs of CPFR are substantially higher along with greater implementation difculties. Indeed, these difculties might explain why many of the CPFR programs have not moved beyond a limited number of product categories or a small set of trading partners (see e.g. Baird, 2003; Program may build CPFR momentum, 2005). Therefore, this trade-off between benets and costs of supply chain collaborations creates an urgent need for SCM practitioners to determine the right collaboration level for their supply chains. Today, many SCM practitioners try to determine the appropriate level of collaboration for their supply chains. Here, the following two questions play a critical role in determining the right collaboration level:

Does it is required to invest in CPFR if an earlier supply chain initiative such as VMI, had already been adopted? In other words, does the gap between the performances of CPFR and VMI compensate the cost of investing in CPFR? Which factors are inuential in answering the question described above? Do capacity of the manufacturing facility, lead times, or uncertainty in customer demand inuence the desire for CPFR?

Raghunathan (1999), Aviv (2001, 2002, 2007), Ovalle and Marquez (2003) and Disney et al. (2004) represent most of the developments in this area. Our paper is different from these previous studies in three ways. First, most of the models have tended to mainly analytical with some very restrictive assumptions (e.g. two-stage supply chain, normally distributed or correlated market demands) for the sake of mathematical tractability (e.g. Raghunathan, 1999; Aviv, 2001, 2002, 2007; Disney et al., 2004). Second, some of the models have been developed so far are only concentrated on forecasting part of CPFR (e.g. Aviv, 2001, 2007). Third, as far as we know, none of the models has explored CPFR and VMI comparatively in capacitated multi-stage supply chains under both stationary and non-stationary customer demands (e.g. Ovalle and Marquez, 2003) as we have done in this paper. Therefore, this paper contributes to the current literature by extending the results of previous research studies in a way that managers in a supply chain enterprise can determine an appropriate level of collaboration for their supply chains. Unlike many prior analytical studies which have very restrictive assumptions for the sake of mathematical tractability (e.g. Mishra and Raghunathan, 2004; Lee and Chu, 2005; Yao et al., 2007), we have used a simulation model in this study to investigate the benets of CPFR and VMI under more realistic circumstances. The simulation approach has been used extensively in the literature for analyzing supply chain systems (e.g. Waller et al., 1999; Zhao et al., 2002a, b; Angulo et al., 2004; Lau et al., 2004; Sari, 2007; Zhang and Zhang, 2007). In this study, we considered a four-stage supply chain, which consists of four echelons: a manufacturing plant, a warehouse, a distributor and a retailer. The plant has limited manufacturing capacity and produces a single product. Each enterprise replenishes its inventory from its immediate upstream enterprise. The remainder of this study is organized as follows. Section 2 claries the methodology and development of the simulation model. Setting of experimental design is identied in Section 3, followed by simulation output analysis in Section 4. Conclusions are presented in Section 5. 2. The simulation model At the initial stages of this research, we intended to use Microsoft Excel in constructing the simulation model; however, research conducted by Keeling

To the best of our knowledge, there have been very few research studies aiming to explore these questions. That is, a few research studies e.g.

ARTICLE IN PRESS
578 K. Sari / Int. J. Production Economics 113 (2008) 575586

and Pavur (2004) indicated that it might possible for errors to occur in the random numbers generated by Microsoft Excel. Therefore, in order to eliminate this potential problem, we have used Crystal Ball, an Excel add-in published by Decisioneering. It is a popular risk analysis and forecasting program that uses Monte Carlo simulation in a spreadsheet environment. According to different situations of information sharing and ordering information coordination, the partnership between supply chain members can be described as one of the following integration levels (see Fig. 1). The rst structure is a supply chain operated under traditional ways of doing business (TSS) and the second structure is a supply chain model operated under a VMI program. Finally, the third structure is the supply chain operated under a CPFR program. In all three supply chain structures, an (R, S) inventory control policy is used for replenishment decisions. Here, R indicates the review interval and S indicates the order-up-to level. R is chosen as 1 week. Order-up-to level, however, is updated at the beginning of each week to reect changes in demand patterns. Under TSS, each member strives to develop local strategies for optimizing his own organization without considering the impact of his strategies on the performance of other members. Moreover, since

no information is shared between members, upstream stages are unaware of actual demand information at the market place. That is, while creating demand forecasts and inventory plans, supply chain members use only replenishment orders placed by their immediate downstream member. Therefore, each member of the supply chain replenishes his own inventory by following an installation-based (R, S) policy. Under an installation-based (R, S) policy, each member considers his local inventory position. The sequence of events followed by a supply chain member under TSS is outlined as follows: (i) The member receives the delivery from its immediate upstream member, which was ordered L periods ago (the lead time is L periods). If the member is the plant, L is the production lead time. (ii) The member observes the order placed by its immediate downstream member. If the member is the retailer, the order is the market demand. (iii) The member fulls the customer orders (plus backorders if there are any) by on-hand inventory, and any unfullled customer orders are backordered. The member analyzes the historical replenishment orders placed by its immediate downstream member for forecasting. Based on this demand forecast, the member

Fig. 1. Three supply chain structures.

ARTICLE IN PRESS
K. Sari / Int. J. Production Economics 113 (2008) 575586 579

updates its order-up-to point. If the member is the retailer, historical market demand data are analyzed. The order-up-to point of the member at stage k, Sk, estimated from the observed demand is as follows (Nahmias, 1997, p. 278):   bk 1 Sk F k (1) bk hk where Fk (.) is the distribution function of the demand realized by the member at stage k. Similarly, bk and hk are backorder and holding costs of the member at stage k, respectively. Here, parameters of the demand distribution, Fk (.), are updated at the beginning of each week by using the exponential smoothing method (see e.g. Nahmias, 1997, p. 74) to reect changes in demand patterns. (iv) The member decides how many units to order from its immediate upstream member. The quantity of the order is equal to the difference between the order-up-to level and inventory position. If the member is the plant, a production order is placed. Here, the plant, because of its limited manufacturing capacity, cannot always produce enough to bring its inventory position up to the updated value of S. In these cases, the plant makes full capacity production by backordering the remaining requirement. This modication of order-up-to policy for the case of limited production capacity provides an optimal solution for uncertain demands (see e.g. Gavirneni et al., 1999; Federgruen and Zipkin, 1986a, b). Under VMI, on the other hand, the retailer provides the distributor with access to its real-time inventory level as well as its pos data (Fig. 1b). In return, the distributor takes the responsibility of managing the inventories at the retailer. That is, under VMI, the distributor does not only need to take its own inventories into account while making inventory plans, but also the inventories of the retailer. Therefore, under this structure, the distributor follows an echelon-based policy in his replenishment planning. Under the echelon-based policy, the distributor looks at its own inventory position plus the inventory position of the retailer, instead of his local inventory position only. For a discussion of installation and echelon policies, see Clark and Scarf (1960), Axsater and Rosling (1993). All other echelons of the supply chain (the plant and the warehouse), on the other hand, are operated in

the same way as in TSS. Here, in order to compute the echelon order-up-to levels of the retailer and the distributor, the heuristic developed by Shang and Song (2003) is used. Again, in this supply chain, the exponential smoothing method is used to update the order-up-to level at each week. Finally, under CPFR, inventory levels, pos data, promotion plans, sales forecasts and all other information that may be inuential on the market demand are shared between supply chain members (Fig. 1c). Consequently, a single joint demand forecast is created by the contribution of each member. Here, there is no doubt that demand forecasts created with the joint contributions of all supply chain members are more accurate than the ones created by the individual organizations (e.g. demand forecasts created by the supply chains operated under TSS or VMI). Indeed, it is very possible that the parameters of the demand distribution can be predicted under a CPFR program. Therefore, in the simulation model, it is assumed that distribution parameters of the market demand are predicted under CPFR by contribution of each member. This assumption does not mean that at the end of the collaborative forecasting process, the members can know exactly what the customer demand is, but rather they can know what the parameters of the underlying demand distribution are (i.e. if the customer demand is normally distributed, mean and standard deviation of the customer demand is known only). Indeed, this assumption makes it sure that the promise of CPFR is realized. That is, the large amounts of information available with CPFR are effectively used to minimize the uncertainty along the supply chain. Of course, in practice, as the model of Disney et al. (2004) also indicates, it might possible that bulk of information available with CPFR result in confusion of supply chain managers, which leading low levels of supply chain performance. Therefore, the benets of CPFR obtained in this study are valid only if CPFR is properly implemented. Moreover, under this collaboration mode, an echelon-based (R, S) inventory policy is used for entire supply chain. That is, inventory positions and inventory costs of all four supply chain members are taken into account in replenishment decisions. The cost structures for the supply chain members in the simulation model are assumed to be as follows; the unit backorder costs per week for the plant, the warehouse, the distributor and the retailer are $5, $11, $18 and $25, respectively.

ARTICLE IN PRESS
580 K. Sari / Int. J. Production Economics 113 (2008) 575586

The unit inventory costs per week for the plant, the warehouse, the distributor and the retailer are $0.25, $0.50, $0.75 and $1.00, respectively. 2.1. Retailers demand structure Although normal distribution is more widely used in supply chain research studies, the g-distribution is used here to represent the customer demand realized by the retailer. This is due to the fact that there are some limitations of normal distribution in representing demand structures. For example, normal distribution allows the occurrence of negative customer demand. Therefore, in order to avoid this unrealistic situation, some restrictive assumptions have to be included in the model (e.g. Waller et al., 1999; Zhao et al. 2002a, b; Lau et al., 2004). The g-distribution, on the other hand, does not have such problems because it allows only non-negative values. Moreover, the g-distribution is exible in that it can represent a wide variety of demand structures. Keaton (1995), for instance, states that choosing g-distribution is an effective choice to represent the demand patterns. There are two parameters of the g-distribution. These are shape (a) and scale (b) parameters. The mean and the variance of the distribution can be expressed as ab and ab2, respectively. In the simulation model, we assume that the shape parameter of the demand distribution is 15 (a 15). The scale parameter (b), on the other hand, is assumed to be a stochastic variable in the form of Eq. (2).   2p t (2) bt 20 season sin 52 In Eq. (2), b(t) is the scale parameter of the g-distribution in week t. The variability in the scale parameter of the demand distribution allows us to generate both seasonal and non-seasonal customer demands. For example, while assigning zero to the season constant produces non-seasonal demand pattern, assigning non-zero values results in seasonality in customer demand. A representative histogram of the market demand for the selected parameters is generated in Fig. 2 to clarify the distribution of the market demand to the readers. Three demand structures representing different combinations of seasonality are used in this study. These are customer demand with no seasonality (SDV), customer demand with medium level of seasonality (MDV) and customer demand with high level of seasonality (HDV). The values of the season

Fig. 2. Histogram of the market demand when a 15 and b(t) 20.

constant for each demand structure are determined as 0, 2 and 4, respectively. The values of the season constant are selected in such a way that both nonseasonal and seasonal customer demands with different strengths are generated. For example, while SDV represent the non-seasonal customer demand, MDV and HDV represent the demand structures with seasonal swings of the size of approximately 10% and 20% of average demand, respectively. 2.2. Verication and validation of the simulation In order to verify that the simulation program performs as intended, the conceptual model is divided into three parts: demand generation and determination of total manufacturing capacity, forecasting and production/inventory planning, and order fulllment and reporting. Each part is designed separately so that more efcient and effective debugging is made possible. Moreover, the combined simulation model is also traced and tested with the results calculated manually. Later, in order to validate the simulation output, the random demand variables generated in the simulation model are plotted on a scatter diagram. Then, it is validated that the intended demand structure is generated. The supply chain model above is simulated for 1128 weeks. The initial parameters of the forecasting model are estimated with the rst 400 weeks of simulation run, which are removed later from the output analysis to eliminate the worm-up period effect. Therefore, the rest of the data are used for effective simulation output analysis. In order to reduce the impact of random variations, the same random numbers are used to simulate all three systems. That is, same customer demand is generated for all types of supply chain systems. In addition to this variance reduction

ARTICLE IN PRESS
K. Sari / Int. J. Production Economics 113 (2008) 575586 581

technique, 15 replications for each combination of the independent variables are conducted. 3. Experimental design Four independent factors are considered in the experimental design. These are; type of supply chain (SCTYPE), available production capacity of plant (CAP), uncertainty in customer demand (DV) and replenishment lead times (L). The number of levels of these variables and their values are listed in Table 1. Factor SCTYPE refers to the way the supply chain is operated. Specically, this factor indicates whether the supply chain is operated under TSS, VMI, or CPFR. The factor CAP is expressed as the ratio of the plants total capacity to the total market demand. Total capacity of the plant is distributed to each week, equally. The factor L denotes the replenishment lead times between each member of the supply chain. Finally, the factor DV indicates the level of uncertainty seen in market demand. Two factors are used as dependent variables in the experimental design in order to evaluate benets of CPFR and VMI. These factors are total cost for the entire supply chain (TSC) and customer service level of the retailer (CSL). TSC is the sum of the inventory holding costs of all members in the supply chain and backorder cost of the retailer. Here, we include the backorder cost of the retailer only, because all other backorder costs are internal costs within the entire supply chain and they are not actually incurred. Factor CSL is the percentage of customer demand satised by the retailer through the available inventory. 4. Simulation output analysis The output from the simulation experiments are analyzed using MANOVA procedure of the SPSS. MANOVA analysis is chosen because it is more
Table 1 Independent factors of the experimental design Independent factors Levels 1 SCTYPE CAP DV L TSS 1.10 SDV 1 2 VMI 1.30 MDV 4 3 CPFR 1.50 HDV

appropriate for our model because MANOVA considers the correlations between the dependent variables in the experimental design, see Hair et al. (1998, p. 331). Selected MANOVA results are presented in Table 2. MANOVA results in Table 2 show that at 5% signicance level, SCTYPE has signicant impacts on both performance factors, which indicates that CPFR and VMI have substantial inuences on the performance of the supply chain. The performance of each type of supply chain is presented in Table 3. Examination of Table 3 reveals that the reduction in total supply chain cost derived from CPFR signicantly higher than the reduction derived from VMI. For example, while CPFR provides 33.90% cost savings, VMI provides 17.34% on the average. Similarly, the results also indicate that CPFR provides higher level of increase in the customer service level than VMI does. For example, while CPFR leads to an increase of 3.84% in customer service level on the average, VMI results in 1.54% increase in customer service level. Therefore, these results lead us to conclude that CPFR produces substantially higher benets than VMI in terms of total supply chain cost and customer service level. Actually, these ndings are simple and intuitively expected for us, so we will not concentrate on them further. Instead, we will concentrate on how the performance increase gained from CPFR and VMI change in parallel to changes in specic conditions of supply chains. For this purpose, performance of CPFR and VMI under various capacity levels (CAP), demand uncertainty (DV) and lead times (L) are produced in Fig. 3. 4.1. Impact of manufacturing capacity (CAP) on the supply chain collaboration MANOVA results in Table 2 shows that at 5% signicance level, the interaction effect between CAP and SCTYPE has signicant impacts on both dependent variables. This means that manufacturing capacity has a signicant inuence on the performance of CPFR and VMI for all performance measures. Examination of Fig. 3 reveals that both supply chain initiatives better off operating in the environments where larger manufacturing capacities are available. That is, we see that contribution of CPFR and VMI is in its lowest level when the plant has its smallest manufacturing capacity (i.e. CAP 1.10). For example, when CAP is 1.10, the reductions in

ARTICLE IN PRESS
582 Table 2 Selected MANOVA results Source Dependent variables CSL F value SCTYPE CAP L DV SCTYPE CAP SCTYPE L CAP L SCTYPE CAP L SCTYPE DV CAP DV SCTYPE CAP DV L DV SCTYPE L DV CAP L DV SCTYPE CAP L DV
a

K. Sari / Int. J. Production Economics 113 (2008) 575586

TSC (a) Pr4F 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0017 0.0000 F value 424.7569 24.7766 3491.4777 205.2806 12.4347 104.2851 3.8058 8.9866 5.0148 27.2560 6.2094 10.0771 2.6900 5.0067 6.3501 Pr4F 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0227 0.0000 0.0005 0.0000 0.0000 0.0000 0.0302 0.0005 0.0000

881.2724 153.6741 508.1819 627.2433 13.7673 50.8457 13.1954 12.4123 18.3525 59.6543 9.5052 106.4525 6.1199 4.3597 13.3417

Based on residual analysis, log transformation of TSC was made to satisfy the assumptions of MANOVA.

Table 3 Performances of each type of supply chain Performance measures SCTYPE Average 95% condence interval Lower bound CSL (%) TSS VMI CPFR TSS VMI CPFR 94.37 95.91 98.22 699,869 578,451 462,737 94.25 95.78 98.09 677,720 556,301 440,588 Upper bound 94.50 96.04 98.35 722,019 600,600 484,886

TSC ($)

total supply chain cost for VMI and CPFR are realized as 14.6% and 19.2%, respectively. On the other hand, when CAP is increased to 1.50, we see that the reductions for VMI and CPFR dramatically jump to 21.0 and 42.3, respectively. This result, consistently with previous studies (e.g. Gavirneni et al., 1999; Lee et al., 2000; Gavirneni, 2002; Simchi-Levi and Zhao, 2003; Lau et al., 2004), shows that the contributions of supply chain collaborations are signicantly higher when larger levels of manufacturing capacity is available. In addition, examination of Fig. 3 also reveals that the distinction between CPFR and VMI reaches its maximum level when the capacity ratio is in its

highest level (i.e. CAP 1.50). For example, while the difference between the percentage of cost reductions provided by CPFR and VMI is around 22% when CAP is 1.50, it is realized around 4% only when CAP is 1.10. Therefore, this result clearly shows that choosing to implement a CPFR program rather than a VMI program provides substantially higher benets for the supply chain where larger levels of manufacturing capacity are available. In other words, this shows us that under the conditions where available manufacturing capacity is tight, a careful consideration has to be given on the selection of an appropriate collaboration mode. That is, under the conditions where the manufacturing capacity is too tight, additional performance increase provided by CPFR over VMI may not justify the higher implementation and operational cost of CPFR. 4.2. Impact of demand uncertainty (DV) on the supply chain collaboration MANOVA results in Table 2 show that at 5% signicance level, the interaction effect between DV and SCTYPE has signicant impacts on both demand variables. This means that uncertainty in market demand has a signicant inuence on CPFR and VMI for all performance measures. Examination of Fig. 3 reveals that higher levels of demand uncertainty results in substantial decreases

ARTICLE IN PRESS
K. Sari / Int. J. Production Economics 113 (2008) 575586 583

Fig. 3. Overall performance of CPFR and VMI.

in the cost savings derived from VMI. For example, the reduction in total supply chain cost decreases from 24% to 15% when the uncertainty in the customer demand (DV) changes from low level (LDV) to high level (HDV). On the other hand, same level of increase in the demand uncertainty does not produce that much performance reduction in CPFR. That is, it results in approximately 3% of reduction in total supply chain cost. In addition to this, Fig. 3 also shows that the supply chain operated under CPFR produces higher customer service level under all levels of the demand uncertainty. Similarly, the gap between the customer service levels of VMI and CPFR is in its highest level when the uncertainty in market demand is in its highest level. For example, we see that, while CPFR produces 1.52% of higher customer service level under low level of demand uncertainty (DV SDV), it is increased to 2.90% when there is high level of uncertainty in the market demand (DV HDV). The results obtained here lead us to conclude that compared with a VMI system, the value of CPFR is substantially greater under the market conditions where demand variability is high. Therefore, SCM practitioners have to be more eager to implement CPFR programs under more volatile market conditions. 4.3. Impact of lead times (L) on the supply chain collaboration MANOVA results in Table 2 show that at 5% signicance level, the interaction effect between L

and SCTYPE has signicant impacts on both performance factors. This means that lead times have a signicant inuence on CPFR and VMI for all performance measures. Examination of Fig. 3 reveals that CPFR and VMI exhibit different performance levels under different replenishment lead times. That is, while the reduction amount in total supply chain cost under CPFR substantially increases as the replenishment lead times increase, the supply chain cost savings gained from VMI do not change signicantly. For example, when the supply chain operated under CPFR is considered, it is seen that the reduction in total supply chain cost dramatically increases from 20% to 40% as lead times increase from 1 to 4 weeks. On the other hand, in case of VMI, it is apparent that any level of increase or decrease in replenishment lead times does not result in signicant changes in the reduction of total supply chain cost. Furthermore, when the customer service level is considered, we see that under all levels of lead times, CPFR produces higher level of ll rate than VMI does. There is, in addition, one further point to make that the gap between the service levels achieved by CPFR and VMI substantially increase as the replenishment lead times increase. The results obtained here imply that compared with a VMI system, the value of CPFR is substantially greater under the conditions where replenishment lead times are longer. Therefore, SCM practitioners have to be more eager to invest in CPFR instead of VMI in supply chains where lead times are long.

ARTICLE IN PRESS
584 K. Sari / Int. J. Production Economics 113 (2008) 575586

5. Conclusion This paper comparatively investigates the performance increase obtained from VMI and CPFR in a four-stage supply chain under both stationary and non-stationary customer demands viva a comprehensive simulation study. Through comprehensive simulation experiments and subsequent statistical analysis of the simulation outputs, we make the following three important observations. First, we observe that the benets gained from CPFR are always higher than that of VMI under all conditions considered in this study. That is, compared with VMI, CPFR produces lower total supply chain cost as well as higher customer service levels. Therefore, from this study, we may sure and clear about the fact that the managers of a supply chain enterprise better off investing in CPFR. Second, through simulation output analysis, we observe that the performance increase gained from CPFR and VMI signicantly depends on three factors. These are capacity tightness of the plant, replenishment lead times and uncertainty in market demand. As these factors get different levels, the benets obtained from both initiatives also change substantially. Moreover, the gap between the performance improvements produced by CPFR and VMI also changes signicantly. For example, we observe that when the lead times are short and/ or where available manufacturing capacity is tight, the benets of switching from VMI to CPFR are at its lowest value. That is, contribution of switching to CPFR is almost ignorable when we consider the additional resources required for CPFR adaptation. The managerial implication of this nding is great because nowadays, adaptation of every business practice, which is in fashion, is popular without analyzing the suitability of it for specic business conditions. Therefore, this research indicates that it is of highly importance to make careful benet/cost analysis to invest in CPFR under the conditions where lead time is short and/or where available manufacturing capacity is very tight. Finally, we recognize that there are substantial decreases in the performance of VMI as the uncertainty in customer demand increases. On the other hand, we also recognize that there is only a slight decrease in the performance of CPFR under higher variable customer demands. Thus, indicating that highly variable customer demand results in widening the gap between the performances of CPFR and VMI. This is because of the fact that

supply chain members may better manage the uncertainties through joint forecasting and inventory planning under CPFR. Therefore, one other managerial implication that can be drawn from this nding is that the practitioners have to invest in CPFR as soon as possible in the industries in which demand uncertainty is highly variable. The computer industry, for instance, with its very short product life cycle and highly variable customer demand, is a good example to industries where the adaptation of CPFR is an urgent need. Although this study provides important insights into CPFR and its relationship with VMI, we have to state that there are some limitations of this study. First, we consider a serial supply chain structure with one member at each echelon. This supply chain structure is only a simplied case and in future research studies, modeling more realistic supply chain structures may better explain and extend the results obtained from this research. Second, we assume that the members in the supply chain apply order-up to policies to make their production/ inventory decisions; however, there are other types of inventory/production policies that can be included in the model. Third, the cost structure used in the simulation model only represents one special case.

References
Achabal, D.D., McIntyre, S.H., Smith, S.A., Kalyanam, K., 2000. A decision support system for vendor managed inventory. Journal of Retailing 76 (4), 430454. Aghazadeh, S.M., 2003. Going toward a better production by CPFR. Journal of Academy of Business and Economics 2 (1), 123131. Aichlmayr, M., 2003. Is CPFR worth the effort? Chief Logistics Ofcer 2, 2730. Angulo, A., Nachtmann, H., Waller, M.A., 2004. Supply chain information sharing in a vendor managed inventory partnership. Journal of Business Logistics 25 (1), 101120. Aviv, Y., 2001. The effect of collaborative forecasting on supply chain performance. Management Science 47 (10), 13261343. Aviv, Y., 2002. Gaining benets from joint forecasting and replenishment process: The case of auto-correlated demand. Manufacturing and Service Operations Management 4 (1), 5574. Aviv, Y., 2007. On the benets of collaborative forecasting partnerships between retailers and manufacturers. Management Science 53 (5), 777794. Axsater, S., Rosling, K., 1993. Notes: Installation vs. echelon stock policies for multilevel inventory control. Management Science 39 (10), 12741280. Baird, N., 2003. Collaboration collusions. Frontline Solutions 4, 1414.

ARTICLE IN PRESS
K. Sari / Int. J. Production Economics 113 (2008) 575586 Barratt, M., Oliveria, A., 2001. Exploring the experiences of collaborative planning initiatives. International Journal of Physical Distribution and Logistics Management 31 (4), 266289. Chen, F., Drezner, Z., Ryan, J.K., Simchi-Levi, D., 2000a. Quantifying the bullwhip effect in a simple supply chain: The impact of forecasting, lead times and information. Management Science 46 (3), 436443. Chen, F., Drezner, Z., Ryan, J.K., Simchi-Levi, D., 2000b. The impact of exponential smoothing forecasts on the bullwhip effect. Naval Research Logistics 47 (4), 269286. Chopra, S., Meindl, P., 2001. Supply Chain Management. Prentice-Hall, New Jersey. Clark, A.J., Scarf, H., 1960. Optimal policies for a multi-echelon inventory problem. Management Science 6 (4), 475490. CPFR Workgroup, 2002. Collaborative Planning Forecasting and Replenishment Voluntary Guidelines Voluntary Interindustry Commerce Standards, available at /http://www.vics. org/committees/cpfr/voluntary_v2/CPFR_Tabs_061802.pdfS (accessed 17 May 2006). C etinkaya, S., Lee, C.Y., 2000. Stock replenishment and shipment for vendor-managed inventory systems. Management Science 46 (2), 217232. Danese, P., 2004. Beyond vendor managed inventory: The Glaxosmithkline case. Supply Chain Forum: An International Journal 5 (2), 3239. Dejonckheere, J., Disney, S.M., Lambrecht, M.R., Towill, D.R., 2004. The impact of information enrichment on the bullwhip effect in supply chains: A control engineering perspective. European Journal of Operational Research 153 (3), 727750. De Toni, A.F., Zamolo, E., 2005. From a traditional replenishment system to vendor-managed inventory: A case study from the household electrical appliances sector. International Journal of Production Economics 96 (1), 6379. Disney, S.M., Towill, D.R., 2003a. Vendor-managed inventory (VMI) and bullwhip reduction in a two level supply chain. International Journal of Operations and Production Management 23 (6), 625651. Disney, S.M., Towill, D.R., 2003b. The effect of vendor managed inventory dynamics on the bullwhip effect in supply chains. International Journal of Production Economics 85 (2), 199215. Disney, S.M., Naim, M.M., Potter, A., 2004. Assessing the impact of e-business on supply chain dynamics. International Journal of Production Economics 89 (2), 109118. Dong, Y., Xu, K., 2002. A supply chain model of vendor managed inventory. Transportation Research Part E: Logistics and Transportation Review 38 (2), 7595. Federgruen, A., Zipkin, P., 1986a. An inventory model with limited production capacity and uncertain demands I: The average-cost criterion. Mathematics of Operations Research 11 (2), 193207. Federgruen, A., Zipkin, P., 1986b. An inventory model with limited production capacity and uncertain demands II: The discounted-cost criterion. Mathematics of Operations Research 11 (2), 208215. Fiddis, C., 1997. ManufacturerRetailer Relationships in the Food and Drink Industry: Strategies and Tactics in the Battle for Power. Financial Times Retail & Customer Publications, London, UK. Fliedner, G., 2003. CPFR: An emerging supply chain tool. Industrial Management & Data Systems 103 (1), 1421. 585 Foote, P.S., Krishnamurthi, M., 2001. Forecasting using data warehousing model: Wal-Marts experience. The Journal of Business Forecasting Methods and Systems 20 (3), 1317. Gavirneni, S., 2002. Information ows in capacitated supply chains with xed ordering costs. Management Science 48 (5), 644651. Gavirneni, S., Kapuscinski, R., Tayur, S., 1999. Value of information in capacitated supply chains. Management Science 45 (1), 1624. Hair, J., Anderson, R.E., Tatham, R.L., Black, W.C., 1998. Multivariate Data Analysis, fth ed. Prentice-Hall, New Jersey. Keaton, M., 1995. Inventory control under gamma demand and stochastic lead time. Journal of Business Logistics 16 (1), 107131. Keeling, K.B., Pavur, R.C., 2004. Numerical accuracy issues in using excel for simulation studies. Proceeding of the 2004 Winter Simulation Conference, 15131518. Kuk, G., 2004. Effectiveness of vendor managed inventory in the electronics industry: Determinants and outcomes. Information and Management 41 (5), 645654. Lau, J.S.K., Huang, G.O., Mak, K.L., 2004. Impact of information sharing on inventory replenishment in divergent supply chains. International Journal of Production Research 42 (5), 919941. Lee, C.C., Chu, W.H.J., 2005. Who should control inventory in a supply chain. European Journal of Operational Research 164 (1), 158172. Lee, H., Padmanabhan, V., Whang, S., 1997a. Information distortion in a supply chain: The bullwhip effect. Management Science 43 (4), 546558. Lee, H., Padmanabhan, V., Whang, S., 1997b. The bullwhip effect in supply chains. Sloan Management Review 38 (3), 93102. Lee, H., So, K.C., Tang, C.S., 2000. The value of information sharing in a two-level supply chain. Management Science 46 (5), 626664. Li, J., Sikora, R., Shaw, M.J., Tan, G.W., 2006. A strategic analysis of inter organizational information sharing. Decision Support Systems 42 (1), 251266. Mason-Jones, R., Towill, D.R., 2000. Coping with uncertainty: Reducing bullwhip behaviour in global supply chains. Supply Chain Forum 1 (1), 4045. McCullen, P., Towill, D.R., 2002. Diagnosis and reduction of bullwhip in supply chains. Supply Chain Management: An International Journal 7 (3), 164179. Metters, R., 1997. Quantifying the bullwhip effect in supply chains. Journal of Operations Management 15 (3), 89100. Micheau, V.A., 2005. How Boeing and Alcoa implemented a successful vendor managed inventory program. Journal of Business Forecasting 24 (1), 1719. Mishra, B.K., Raghunathan, S., 2004. Retailer- vs vendormanaged inventory and brand competition. Management Science 50 (4), 445457. Nahmias, S., 1997. Production and Operations Analysis. Irwin/ McGraw-Hill, Homewood, IL. Ouyang, Y., 2006. The effect of information sharing on supply chain stability and the bullwhip effect. European Journal of Operational Research. Ovalle, O.R., Marquez, A.C., 2003. The effectiveness of using e-collaboration tools in the supply chain: An assessment study

ARTICLE IN PRESS
586 K. Sari / Int. J. Production Economics 113 (2008) 575586 Watson, E., 2005. Nestle switches to vendor managed inventory with Tesco. Food Manufacture 80 (8), 2020. Yao, Y., Evers, P.T., Dresner, M.E., 2007. Supply chain integration in vendor-managed inventory. Decision Support Systems 43 (2), 663674. Yu, Z., Yan, H., Cheng, T.C.E., 2001. Benets of information sharing with supply chain partnerships. Industrial Management and Data Systems 101 (3), 114119. Yu, Z., Yan, H., Cheng, T.C.E., 2002. Modelling the benets of information sharing-based partnerships in a two-level supply chain. Journal of Operational Research Society 53 (4), 436446. Zhang, C., Zhang, C., 2007. Design and simulation of demand information sharing in a supply chain. Simulation Modelling Practice and Theory 15 (1), 3246. Zhao, X., Xie, J., Zhang, W.J., 2002a. The impact of information sharing and ordering co-ordination on supply chain performance. Supply Chain Management: An International Journal 7 (1), 2440. Zhao, X., Xie, J., Leung, J., 2002b. The impact of forecasting model selection on the value of information sharing in a supply chain. European Journal of Operational Research 142 (2), 321344. with system dynamics. Journal of Purchasing & Supply Management 9 (4), 151163. Program may build CPFR momentum, 2005. DSN Retailing Today 44 (10), 1414. Raghunathan, S., 1999. Interorganizational collaborative forecasting and replenishment systems and supply chain implications. Decision Sciences 30 (4), 10531071. Sari, K., 2007. Exploring the benets of vendor managed inventory. International Journal of Physical Distribution and Logistics Management 37 (7), 529545. Shang, K.H., Song, J.S., 2003. Newsvendor bounds and heuristic for optimal policies in serial supply chains. Management Science 49 (5), 618638. Simchi-Levi, D., Zhao, Y., 2003. The value of information sharing in a two stage supply chain with production capacity constraints. Naval Research Logistics 50 (8), 888916. Simchi-Levi, D., Kaminsky, P., Simchi-Levi, E., 2003. Designing and Managing the Supply Chain, second ed. McGraw-Hill, IL. Waller, M.A., Johnson, M.E., Davis, T., 1999. Vendor-managed inventory in the retail supply chain. Journal of Business Logistics 20 (1), 183203.