The impact of supply chain structures on performance.

The impact of supply chain structures on performance 03/December/2014 2014 Phd Candidate: Roberto Domínguez Cañizares Supervisor: Dr. José M. Framiñán Torres Co-supervisor: Dr. Salvatore Cannella ACKNOWLEDGEMENT First of all I want to express my sincere gratitude to my supervisor Prof. Jose M. Framiñán, who gave me the opportunity of research in the exciting field of the Supply Chain Management. He introduced me in a very interesting modeling paradigm that has been demonstrated to be very useful in this field. His insightful advices and continuous and valuable reviews of my work have been very precious for me. A warm gratitude goes to my co-supervisor Salvatore Cannella, who I met in a research stay in Lisbon. Since the beginning he shared with me his enthusiasm for research, and we started a good collaboration. Even he was very time researching in other centres, his availability, dedication, support and motivation, have been very important for me during my PhD. A special thank is for my family, who supported me during my academic life, since the beginning of my bachelor degree until the end of this PhD. Very special thanks go to Beatriz, who has built it up with me during the last year of my PhD. Her empathy, patience, optimism, understanding, and her constant support have been more important for me than she actually believes. Finally, to those members of the DOIGE I who shared with me very valuable knowledge, advices, and coffees during this period, thank you. Sincerely, Roberto Domínguez Sevilla, December 2013 i Table of Contents PART I: INTRODUCTION CHAPTER 1: INTRODUCTION 1 1.1 CONTEXT AND PROBLEM STATEMENT 1 1.2. RESEARCH OBJECTIVES AND OUTLINE OF THE THESIS 3 CHAPTER 2: SUPPLY CHAIN NETWORKS AND BULLWHIP EFFECT 5 2.1. INTRODUCTION 5 2.2. SUPPLY CHAIN NETWORKS 5 2.3. BULLWHIP EFFECT 7 2.3.1. Introduction 7 2.3.2. Causes 8 2.3.3. Bullwhip effect and the structure of the SCN 9 2.3.4. Bullwhip avoidance strategies and the structure of the SCN 12 2.4. BULLWHIP EFFECT ANALYSIS 15 2.5. METRICS FOR THE BULLWHIP EFFECT ANALYSIS 16 2.3. SUPPLY CHAIN OPERATIONS REFERENCE (SCOR) MODEL 17 PART II: ON MODELING COMPLEX SUPPLY CHAIN NETWORKS BY MAS CHAPTER 3: LITERATURE REVIEW ON MAS APPLICATIONS ON SCM 21 3.1. INTRODUCTION 21 3.2. MAS APPLICATIONS IN SCM 23 3.3. FRAMEWORKS 26 3.4. CONCLUSIONS 28 ii CHAPTER 4: DESIGN OF A MAS-BASED FRAMEWORK FOR SCN MODELING 31 4.1. INTRODUCTION 31 4.2. FUNCTIONAL AGENTS 32 4.3. ENTERPRISE AGENT 35 4.4. SUMMARY AND CONCLUSIONS 37 CHAPTER 5: FRAMEWORK IMPLEMENTATION. A SCN SIMULATION TOOL: SCOPE 39 5.1. INTRODUCTION 39 5.2. MAS SIMULATION TOOLS 39 5.3. SWARM MODELING TOOL 41 5.4. FRAMEWORK IMPLEMENTATION BY SWARM 42 5.4.1 Components 43 5.4.1.1. Objects 43 5.4.1.2. Agents 43 5.4.1.3. Swarms 44 5.4.1.4. Configuration files 45 5.4.2. Overall 45 5.5. VALIDATION OF SCOPE 49 5.5.1. Scenario 1: Chen et al. (2000) 51 5.5.2. Scenario 2: Dejonckheere et al. (2003a) 52 5.5.3. Scenario 3: Chatfield et al. (2004) 52 5.6. SUMMARY AND CONCLUSIONS 55 PART III: SUPPLY CHAIN'S STRUCTURE AND BULLWHIP EFFECT CHAPTER 6: EXPLORING THE BULLWHIP EFFECT IN DIVERGENT SCNS 59 6.1. INTRODUCTION 59 6.2. SUPPLY CHAIN NETWORK EMPLOYED AS TESTBED 59 6.3. EXPERIMENTS DESIGN 63 6.4. NUMERICAL RESULTS 65 6.4.1 Stationary response set 65 6.4.2 Dynamic response set 67 6.5. FINDINGS AND MANAGERIAL IMPLICATIONS 69 6.6 SUMMARY AND CONCLUSIONS 72 iii CHAPTER 7: ON BULLWHIP LIMITING STRATEGIES IN DIVERGENT SCNS 75 7.1. INTRODUCTION 75 7.2. SIMULATED SCENARIOS 75 7.2.1. Traditional SCN with classical OUT policy 75 7.2.2. Traditional SCN with smoothing replenishment rule 76 7.2.3. Information sharing SCN with classical OUT policy 77 7.2.4. Information sharing SCN with smoothing order policy 78 7.3. EXPERIMENTS DESIGN 78 7.4. RESULTS AND DISCUSSION 80 7.4.1. Traditional SCN with classical OUT policy 82 7.4.2. Traditional SCN with smoothing replenishment rule 82 7.4.3. Information sharing SCN with classical OUT policy 82 7.4.4. Information sharing SCN with smoothing order rule 83 7.5. FINDINGS AND MANAGERIAL IMPLICATIONS 83 7.6. SUMMARY AND CONCLUSIONS 85 CHAPTER 8: A SYSTEMATIC ANALYSIS OF THE STRUCTURE OF DIVERGENT SCNS AND BULLWHIP 87 8.1. INTRODUCTION 87 8.2. THE DIVERGENT SCN CONFIGURATION 87 8.3. DESIGN OF EXPERIMENTS 91 8.4. RESULTS AND NUMERICAL ANALYSIS 93 8.4.1. Variance Lens 94 8.4.2. Shock Lens 98 8.4.3. A comparison between the variance lens scenario and the shock lens scenario 100 8.5. SUMMARY AND CONCLUSIONS 103 PART IV: CONCLUSIONS AND FURTHER RESEARCH CHAPTER 9: CONCLUSIONS AND FURTHER RESEARCH 109 9.1. INTRODUCTION 109 9.2. MAIN CONCLUSIONS 110 9.3. RESEARCH PRODUCTION 111 9.3.1. Journals 111 9.3.2. Conferences 112 9.4. FURTHER RESEARCH LINES 112 1 Chapter 1: Introduction 1.1 CONTEXT AND PROBLEM STATEMENT A Supply Chain Network (SCN) can be defined as a network of autonomous or semiautonomous business entities collectively responsible for procurement, manufacturing and distribution activities associated with one or more families of related products (Swaminathan et al., 1998). Since the 1980s, there has been tremendous interest for both researchers and practitioners in the area of Supply Chain Management (SCM), due to the increasing importance of designing and maintaining lean, agile, flexible and efficient SCNs. SCM may be viewed as an integrated approach to increasing the effectiveness of the SCN through improved coordinated efforts between the upstream and downstream organizations in the system (Hwarng et al., 2005). SCM has to deal with increasingly complex SCNs due to rapid technological advances, the increase of customer expectation, the enlargement and geographically diverse sourcing arrangements as well as the globalization of trade (Modrak et al., 2012). As manufacturing practice shifts toward the outsourcing paradigm, the SCM takes place throughout a high number of entities, generating a highly interdependence between them and making SCM a complex process and demanding a high coordination between partners. This coordination is rarely found (enterprises are not willing to share private information) and it has been recognized as a root cause of one of the most devastating phenomenon in SCM: the bullwhip effect (BWE) (Bhattacharya and Bandyopadhyay, 2011). BWE refers to a progressive increase in order (demand) variance as order information passes upstream in a SCN, from the customer back to the supplier level. Essentially, orders placed by upstream SCN nodes show increased variability, i.e. variance amplification (Chatfield and Pritchard, 2013; Strozzi et al., 2012). This is known to inevitably lead to excessive inventory investment, poor customer service, lost revenues, misguided capacity plans, ineffective transportation, and missed production schedules (Chen et al. 2012). As a consequence, this effect increases the cost of operating the SCN, producing inefficiencies that lets costs increasing up to 25 per cent, deteriorating profitability of 15-30 per cent, increasing annual inventory holding costs Roberto Domínguez Chapter 1 2 of 33 per cent, and cost of capital of 13 per cent (Turrisi et al., 2013). Nowadays, about two-thirds of firms are affected by the BWE (see e.g. Shan et al., 2013; Bray and Mendelson, 2012). Thus, BWE is one of the most widely investigated phenomena in the modern day SCN management research (Nepal et al., 2012; Zotteri, 2013). In order to analyze the BWE under real business world conditions, increasingly complex mathematical representations of SCNs (such as multi-product scenarios, stochastic lead times, production/distribution capacity constraints, reverse logistic and so on) have been developed. However, several assumptions are commonly made to simplify the analysis (Chatfield, 2013), being one of the most relevant what it can be labeled as a “serial structure model”, i.e. each echelon in the system has a single successor and a single predecessor. Undoubtedly, the serial SCN system analysis represents a powerful technique for studying the dynamics of the BWE, but this assumption is seldom verified in real SCNs (Bhattacharya and Bandyopadhyay, 2011). Essentially, due to the complexity and mathematical intractability of multi-echelon systems, most of the scientific works dealing with the study of the BWE are confined to the classical single-echelon, dyadic or the serially-linked structure (Sucky, 2009; Bhattacharya and Bandyopadhyay, 2011; Giard and Sali, 2013). However, modern SCNs, due to the globalization of trade and the outsourcing paradigm are characterized by a high multiplicity (high number of elements and interactions), often presenting configurations that differ from the simplistic seriallylinked or dyadic topologies (characterized by a low multiplicity). Hence, the research on SCNs should use complex SCN models that allow modeling modern SCN configurations in order to obtain results as closer as possible to the dynamics of real SCNs. In other words, there is a need of modeling other SCN configurations found in real business enterprises such as convergent, divergent or conjoined (Strader et al., 1998; Lin and Shaw, 1998; Beamon and Chen, 2001; Giard and Sali, 2013). Convergent SCNs are assembly-types configurations in which a wide range of materials and subcomponents provided by suppliers converges through a series of manufacturing stages until the final product is assembled at one location. Automobile and aerospace industries are associated with this configuration. On the contrary, divergent SCNs are distribution-like configurations, in which a relatively small number of suppliers provide materials and subcomponents that are used to produce a number of generic product models. Appliances, electronics, and computer industries can be The impact of supply chain structures on performance Chapter 1 3 classified as divergent SCNs. Finally, the conjoined configuration is a combination of the convergent and divergent configurations: first a convergent phase where finished products are manufactured, and a second phase for distributing the finished products. Since distribution phase is present in most of SCNs, this Thesis is focused on the divergent configuration. Modeling such SCNs by classic methods presents many problems since SCNs are representative Complex Adaptive Systems (CAS) (Chen, 2012; Li et al., 2010a,b; Bozarth et al., 2009; Wang et al., 2008; Pathak et al., 2007; Sun and Wu, 2005; Surana et al., 2005; Choi et al., 2001). A CAS is a dynamic network where many agents act simultaneously and continuously react to the actions of the other agents. These systems are characterized by a non-predetermined order (emergent order), an irreversible system history, and an unpredictable future (Mitchell, 1994). Moreover, as Bonabeau (2002) claims, the only way to analyze and understand emergent phenomena is to model the system from the bottom up. In order to cope with this complexity, simulation is often selected as one of the best solutions to model SCNs (Yoo and Glardon, 2009). A simulation framework that is able to both view a complex SCN and examine various causes and their effects at the same time, would provide new insight to the various forces and dynamics in a SCN (Alony and Munoz, 2007). In this regard, a modeling and simulation approach influenced by the complexity paradigm is MAS (Multi-Agent System), derived partly from objectoriented programming and distributed artificial intelligence, and partly from insights from the science of complexity. A MAS is an adequate approach for modeling CAS and it is useful in creating understandable results for managers (Nilsson and Darley, 2006). 1.2. RESEARCH OBJECTIVES AND OUTLINE OF THE THESIS The research objectives of this Thesis are: i. Developing a simulation software tool for modeling and simulating complex SCNs configurations (Part I). The simulation tool takes the name of the project by which this Thesis has been funded: SCOPE (Sistemas COoperativos para la Programación y Ejecución de pedidos). ii. Analyzing and determining the impact of the structure of SCNs on the BWE (Part II). Roberto Domínguez Chapter 1 4 The Thesis is organized in four parts. Part I introduces the context, problem statement and the main objectives of the Thesis (present Chapter 1). Then, Chapter 2 provides a description of SCNs and SCM, along with a literature review on BWE. Part II describes the first objective of the Thesis. It is structured in three chapters: Chapter 3 is a literature review of MAS applications, modeling frameworks, and simulation platforms on SCM problems; Chapter 4 describes the design of a MAS-based framework for modeling complex SCNs; and Chapter 5 describes the implementation of the framework on a MAS software platform (Swarm) and its validation. Part III describes the second objective of the Thesis. It is structured in three chapters as well: Chapter 6 is a comparison analysis of the BWE between a divergent SCN and a serial SCN; Chapter 7 is a comparison analysis of two BWE avoiding strategies (information sharing and smoothing order-up-to policies) between a divergent SCN and a serial SCN; in Chapter 8 a structured design of experiments is performed, by which the divergent SCN configuration is exhaustively varied according to its structural factors, in order to identify a relation between the structure of the network and the BWE. Finally, Part IV is the conclusions of the Thesis (Chapter 9). 5 Chapter 2: Supply Chain Networks and Bullwhip Effect 2.1. INTRODUCTION This is an introductory chapter on SCNs and BWE. First, SCN and SCM are defined, describing why SCNs are considered complex systems and highlighting the problems faced by researchers in order to model the SCN system. Then, it follows a literature review on BWE, introducing the phenomenon and its main causes, and identifying the research gaps related to the structure of the SCNs, which is the focus of this Thesis. Next, a framework for BWE analysis proposed by Towill et al. (2007), which is extensively used in the last chapters of the Thesis, is shortly introduced. After that, a collection of metrics for measuring the BWE are described. Finally, a SCN reference model (SCOR) used as reference in the design of SCOPE (see Chapter 4) is briefly introduced. 2.2. SUPPLY CHAIN NETWORKS A SCN is referred to as a complex network of organizations that synchronizes a series of inter-related business processes, such as procurement, manufacturing and distribution, to create values to final customers in the form of one or more families of related products or services (Li et al., 2009). SCM involves the systemic and strategic coordination of products/services, finances and information flows within and across companies in the SCN with the aim of reducing costs, improving customer satisfaction and gaining competitive advantage for both independent companies and the SCN as a whole (Serdarasan, 2013). It involves complex interactions among suppliers, manufacturers, distributors, third-party logistics providers, retailers, and customers. These entities operate subject to different sets of constraints and objectives. However, they are highly interdependent when it comes to improving performance of the SCN. As a result, decision of any entity in a SCN depends on the performance of others, and their willingness and ability to coordinate activities within the SCN (Wen et al., 2012). The numerous interactions between entities as well as the characteristics of nonlinearity, Roberto Domínguez Chapter 2 6 dynamics etc. in SCNs make it challenging to analyze and predict their responses over time. So far, considerable endeavors have been made to construct models and predict their performance. The conventional models of SCNs in the literatures mainly focus on the issue from the three levels: strategic level, which includes location/allocation decisions, demand planning, distribution channel planning etc; tactical level, which covers inventory control, production distribution coordination, order/freight consolidation etc. and operational level, vehicle routing/scheduling, workforce scheduling, record keeping, and packaging belong to this level (Li et al., 2010a). These previous researches provide a beneficial insight on SCNs and they address problems mainly from the microscopic view i.e., focusing on either the focal entity or relations between two entities in the SCNs. However, there are few literatures to describe and analyze the whole performance of the SCN (Wen et al., 2012). How to establish the whole model of the SCN and analysis its characteristics is a challenge of research. SCNs have often been conceptualized as simple linear systems represented by an event dependent series of firms interacting through dyadic relationships (Cox et al., 2006). However, this linear conception of sequential dyadic relationships, while appealing, grossly oversimplifies and distorts the realities of modern SCNs (Hearnshaw and Wilson, 2013), such as those mentioned in Butner (2010), Christopher and Holweg (2011) and Stank et al. (2011), and fails to adequately account for the interdependence between a large number of heterogeneous firms present in SCNs (Choi et al., 2001; Kim et al., 2011). Nowadays, the current tendency to features more tailored to customers’ individual needs−wider product variety, smaller production lot sizes, more echelons and different actors to co-ordinate within each SCN, etc.−(Perona and Miragliotta, 2004), the increase of customer expectation, the enlargement of outsourcing as well as the globalization of trade have led to SCNs much complicated. Most researchers come to realize that SCNs are representative complex systems (Bozarth et al., 2009; Li et al., 2010a; Zhu and Xu, 2012; Modrak et al., 2012; Serdarasan, 2013), in which a large number of firms operate simultaneously with many supply partners and interact through a variety of information and material flows in an uncertain way (Sivadasan et al., 2006). Furthermore, its overall behavior cannot be described exhaustive, although there is comprehensive knowledge of its components and their interaction (Pratt et al. 2005). These characteristics of complex systems are particularly well modeled by modern The impact of supply chain structures on performance Chapter 2 7 modeling approaches such as MAS (see Chapter 3). Summarizing, the complexity of SCNs requires SCNs to be analyzed on the network level, which adds more interrelations, dynamics, and complexity as compared to the more basic and linear chain level (Moser et al., 2011, Xuan et al., 2011; Ma et al., 2013). In order to manage such complex systems and respond appropriately to exigencies, SCNs managers are required to have an understanding of the underlying structure of their system and how their firms interact (Hearnshaw and Wilson, 2013). Indeed, Choi and Hong (2002) acknowledge that if we are to truly practice the management of SCNs, we need to understand the structure of SCNs. 2.3. BULLWHIP EFFECT 2.3.1. Introduction The BWE is one of the most widely investigated phenomena in the modern day SCN management research (Nepal et al., 2012), since it has been recognized as one of the main obstacles for improving SCN performance. In the presence of this phenomenon, orders placed by upstream nodes exhibit a higher variability as compared to that of orders placed by their downstream partners (Chatfield and Pritchard, 2013), having many undesirable effects such as increasing stock and generating stock-outs (Adenso- Diaz et al., 2012). The BWE is relevant both for individual companies that face an unnecessarily variable demand as well as for the entire SCN (Zotteri, 2012). Moreover, the most recent economic downturn has no doubt created a lot of bullwhips around the world (Lee, 2010). For instance, the electronics manufacturing sector has experienced something akin to the BWE in terms of larger sales declines occurring further upstream (Dvorak, 2009). More specifically, in the last quarter of 2008, consumer demand had declined 8 percent, while product shipments fell 10 percent and chip sales fell 20 percent. These data suggest that electronics retailers, wholesalers and manufacturers responded differently to the decline in consumer demand (Dooley et al., 2010). The investigation on this phenomenon has passed through diverse phases, from empirical and ad hoc studies on BWE causes to mathematical approaches to infer on demand amplification solutions. Bullwhip Avoidance Phase is the term coined by Holweg and Disney (2005) to identify the current phase of the studies devoted to the Roberto Domínguez Chapter 2 8 demand amplification phenomenon. One distinctive feature of this phase is the focus on the efficacy of BWE solving approaches (Cannella and Ciancimino, 2010). To accomplish this aim, increasingly complex mathematical representations of SCNs have been developed to analyze solving approaches under several scenarios, characterized by reverse logistic, multi-product scenarios, different forecasting techniques, stochastic lead times, collaborative systems, capacity constraints, batching, parameter configuration, pricing and so on. 2.3.2. Causes Among the streams of research dealing with BWE, an important one has focused on demonstrating its existence and on identifying its possible causes (Sucky, 2009), and different root causes have been identified. Lee et al. (1997) provided the seminal work that defined the BWE and identified the well-known five causes (Disney and Lambrecht, 2008; Zotteri, 2012). A further relevant framework was proposed by Geary et al. (2006). The authors identified 10 published causes of BWE, based on the works of Mitchell (1924), Wikner et al. (1992), and Lee et al. (1997). Recently Bhattacharya and Bandyopadhyay (2011) identifies 19 causes, 16 operational and 3 behavioral. Operational causes include demand forecasting (Syntetos et al., 2009; Trapero et al., 2012), order batching (Potter and Disney, 2006), price fluctuation (Ma et al., 2013; Lu et al., 2012), rationing and shortage gaming, lead time, inventory policy, replenishment policy, improper control system (Disney and Towill, 2003; Syntetos et al., 2011), lack of transparency (Cannella et al., 2011; Hussain et al., 2012), number of echelons (Disney et al., 2004; Paik and Bagchi, 2007), multiplier effect, lack of synchronization (Ciancimino et al., 2012), misperception of feedback (Gonçalvez et al., 2005), local optimization without global vision (Disney and Lambrecht, 2008), company processes (Holweg et al, 2005) and capacity limits (Cannella et al., 2008; Crespo-Marquez, 2010). The behavioral causes cover neglecting time delays in making ordering decisions (Wu and Katok, 2006), lack of learning and/or training (Akkermans and Voss, 2013) and fear of empty stock (Croson and Donohue, 2006). The impact of supply chain structures on performance Chapter 2 9 2.3.3. Bullwhip effect and the structure of the SCN The potential relation between the SCN structure−defined as the arrangement of the various SCN nodes (Giard and Sali, 2013)−and the BWE is almost unknown. The related published works have only explicitly investigated the impact of the number of echelons in the BWE. Probably, this is mainly because, in SCN literature most of the scientific works dealing with the study of the BWE are confined to the classical singleechelon, dyadic or the serially-linked configuration (Disney et al., 2004a; Sucky, 2009; Bhattacharya and Bandyopadhyay, 2011; Giard and Sali, 2013). In fact the seriallylinked assumption, i.e. each echelon k in the system has a single successor k+1 and a single predecessor k-1, is commonly made in order to simplify the analysis of the BWE (Chatfield and Pritchard, 2013). The number of echelons or ‘number of channel intermediaries’ (Disney and Lambrecth, 2008) is a root cause of the BWE that explicitly depends on the structure of the SCN. In fact, there is a common agreement on the existence of a positive correlation between the reduction of the intermediate stages in the SCN and the reduction of the BWE (Disney et al., 2004; Paik and Bagchi, 2007; Disney and Lambrecth, 2008; Bottani and Montanari, 2010; Yang et al., 2011; Sodhi and Tang, 2011). However, the number of echelons only represents an indicator of the structure of the SCN. Rarely the BWE has been investigated in more complex configurations (Sucky, 2009), assuming that the SCN structure has an influence on the BWE phenomena (Giard and Sali, 2013). There are only a few anecdotic evidences on the relation between key structural factors of SCNs and the BWE which, still, do not provide information on the impact of the different factors in the BWE (see e.g. Sodhi and Tang, 2011). Table 2.1 reports an overview of relevant contributions published during the Bullwhip Avoidance Phase. Articles are classified according to the focus on the parameters and factors investigated (e.g., information sharing, lead time, order policy and demand forecasting), and the typology of SCN structure (e.g. serial and non-serial). All aforementioned papers have largely contributed to better understand the causes, economics consequences and remedies to BWE. Regardless the adopted methodological approaches, the modeled SCN structure and the metrics used to evaluate the SCN performance, the results have shown how factors such as lead time, the adoption of Roberto Domínguez Chapter 2 16 the system performance under an intense and violent solicitation test to determine the resilience of a given SCN structure (Cannella and Ciancimino, 2010). 2.5. METRICS FOR THE BULLWHIP EFFECT ANALYSIS First proposed by Chen et al. (2000), the Order Rate Variance Ratio (𝛷) is the most widely used indicator to detect the BWE (Cannella et al., 2013), measuring the internal process efficiency and showing the performance of each node in the SCN. It is a demand-independent measure, allowing the comparison between different SCNs. Nevertheless, measuring the internal process efficiency at the individual level (single echelon) is insufficient as it only accounts for the individual performance of each link in the chain separately (Cannella et al., 2013). Therefore, a network measure has to be used as a complementary measure of Φ. The Bullwhip Slope (BwSl) summarizes all the ratios obtained for each stage in a single measure, allowing a complete comparison between different SCNs at the network level (Ciancimino et al., 2012; Cannella et al., 2013). The procedure to calculate this metric is to perform a linear regression on the values of Φ using the echelon position as independent variable (equation 2.2). A high value of the slope means a fast propagation of the BWE through the SCN, while a low value means a smooth propagation. Since BwSl is a synthesis of Φ, there are similar costs associated to this metric (procurement, overtime and subcontracting) but at the network level. Below, these two metrics are summarized. - Order Rate Variance Ratio of a node i (𝛷𝑖): computed as the ratio of the order variance in a generic node (𝜎𝑂𝑖 2, estimated by 𝑠𝑂𝑖 2) to the order variance of the end customer demand (𝜎𝑑2, estimated by 𝑠𝑑2). 𝛷𝑖=𝑠𝑂𝑖 2 𝑠𝑑2 (2.1) - 𝐵𝑤𝑆𝑙: computed as the slope of the linear regression of the Φ curve. 𝐵𝑤𝑆𝑙=𝐾∑𝑝𝑖𝛷𝑖− 𝐾 𝑖=1 ∑𝑝𝑖 𝐾 𝑖=1 ∑𝛷𝑖 𝐾 𝑖=1 𝐾∑𝑝𝑖2 𝐾 𝑖=1 −(∑𝑝𝑖 𝐾 𝑖=1 )2 (2.2) The impact of supply chain structures on performance Chapter 2 17 Being 𝐾 the total number of echelons and 𝑝𝑖 the position of the ith echelon. The above mentioned metrics have been conceptualized for a serial SCN. Since this Thesis focuses on more complex SCNs, particularly in divergent SCNs, the metrics need to be modified in order to be used on such SCNs. The reason is that divergent SCNs contain, in general, more than one node in each stage. In the serial SCN, the parameter required to compute the different metrics on each stage (i.e. the order variance) is taken from the only node in the stage. In the divergent SCN, it is necessary to find an aggregate measure for the whole stage. To obtain this measure, the orders of every node j in stage i (𝑂𝑖𝑗) are considered at the same time and added, resulting in an aggregate order pattern for the stage i: 𝐴𝑂𝑖=∑𝑂𝑖𝑗 𝑛𝑖 𝑗=1 , being 𝑛𝑖 the number of nodes in the stage i. Following the same procedure, the aggregate end customer demand pattern can be obtained as 𝐴𝑑=∑𝑂𝐶𝑗 𝑛𝐶 𝑗=1 , being 𝑛𝐶 the number of customers. Then, the aggregate variance of each stage (𝜎𝐴𝑂𝑖 2, 𝜎𝐴𝑑 2) can be estimated (𝑠𝐴𝑂𝑖 2, 𝑠𝐴𝑑 2), and 𝛷𝑖 is: 𝛷𝑖=𝑠𝐴𝑂𝑖 2 𝑠𝐴𝑑 2 (2.3) If all the customer demands are assumed to be independent and each node in the SCN places orders independently, the aggregate variance in each stage i is the sum of the variances of orders of each node j in the stage i (𝜎𝑂𝑖𝑗 2, 𝜎𝑂𝐶𝑗 2), estimated by (𝑠𝑂𝑖𝑗 2, 𝑠𝑂𝐶𝑗 2), and thus, the calculation of 𝛷𝑖: 𝛷𝑖=∑𝑠𝑂𝑖𝑗 2 𝑛𝑖 𝑗=1 ∑𝑠𝑂𝐶𝑗 2 𝑛𝐶 𝑗=1 (2.4) 2.3. SUPPLY CHAIN OPERATIONS REFERENCE (SCOR) MODEL The SCOR model was developed by the Supply Chain Council (SCC, 2006), founded by a group of 70 companies in 1996. It has been developed to describe the business activities associated with all phases of satisfying a customer’s demand. The Roberto Domínguez Chapter 2 18 model itself contains several sections and is organized around the five primary management processes or activities. These activities are divided in two groups: physical activities (Source, Make, Deliver and Return) to manage the physical resources of the enterprise, and planning activities (Plan) to make decisions. These five management processes are represented in Figure 2.1. By describing SCN using these process building blocks, the model can be used to describe simple SCNs as well as very complex enterprise networks using a common set of definitions. Figure 2.1. The five major management processes of SCOR-model (font: Röder and Tibken, 2006). PART II On modeling complex supply chain networks by MAS 21 Chapter 3: Literature review on MAS applications on SCM 3.1. INTRODUCTION This chapter introduces the most relevant literature of the last 15 years on the development of models, frameworks and software based on MAS technology, and its application on SCM. Since one of the objectives of the Thesis is the development of a MAS-based framework for modeling and simulating complex SCNs, the literature review has been focused on generic frameworks that allow modeling and simulating a wide variety of SCN configurations (in terms of structure, inventory policies, forecasting, order fulfillment, etc.). However, since there are many others of MAS applications designed to analyze specific problems within the SCM, a general overview of these applications is performed first in order to get an idea of what particular topics have been analyzed by this methodology. Due to the complexity of SCM, it is very difficult for managers and decision-makers to predict the effects of implementing new management policies and to decide the best strategies to improve the performance of real SCNs. Hence, the existence of SCN modeling tools is very helpful to managers and of great benefit for enterprises. Traditional methods, like analytic models, classical operational research methods, continuous time differential equation models, and discrete time difference equation models are not able to cope with the inherent complexity of SCNs such as the high number of enterprises and interactions between them, or the stochasticity and uncertainty present in most of their processes. Classical operational research methods approaches are not always able to handle the characteristics of dynamic SCNs (Riddals et al., 2001; Long et al., 2011). Analogously, continuous time and discrete time difference equation models are not always suitable for analyzing complex SCN structures, given the high order of differential equations (one tier generally gives a 2nd- 4th order system; 2 tiers even 2nd-6th order), which makes analytical analysis difficult (Lee and Kim, 2008; Holweg and Disney, 2005). Thus, new modeling techniques are required. Roberto Domínguez Chapter 3 22 Simulation has rapidly become a significant methodological approach to theory development in the literature focused on strategy, organizations and SCN management, that allows modelers to capture the dynamics of complex systems like SCNs due to its ease for modeling and its capability of handling their dynamics and stochastic behavior, and enables managers to analyze and evaluate the effects of alternative processes or operation modifications (Chan and Prakash, 2012; Stefanovic et al., 2009; Munoz and Clements, 2008; Chatfield et al., 2001). Particularly, there is a great interest in modeling SCNs as MAS (Surana et al., 2005; Pathak et al., 2007), because there is a natural correspondence between SCN participants and agents in a simulation model: SCNs tend to be decentralized systems with the participants acting independently, according to their own interests and policies (Long et al., 2011). MAS have the capacity to consider the interactions between large numbers of heterogeneous firms allowing SCN managers improving their understanding into how interventions in one part of the SCN may affect another part (Hearnshaw and Wilson, 2013). Thus, the use of MAS turns out to be one of the most effective tools to model and analyze SCNs (Long et al., 2011; Chatfield et al., 2007). A key feature of MAS that allows to properly modeling SCNs is the bottom-up methodology by which a MAS model is constructed. This methodology is based on a synthesizing philosophy, where the modeler assumes that he/she cannot understand the whole phenomenon of interest but can observe, on a micro level, specific activities and processes (agents) and tries to understand their behavior. These agents interact and communicate with other agents and they join to form a coherent whole on a macro level, often emerging behaviors that cannot be predicted in advance. On the contrary, top-down methodologies are not able to cope with CAS, since they are based on the assumption that knowledge is outside the system and someone can measure and analyze the observable phenomenon of interest and from that, decompose it correctly into different sub-units, where the sub-problems are solved separately (Nilsson and Darley, 2006). The adoption of MAS has several other benefits: an increasing modeling realism, seeing as individual agents can be made directly comparable to machines, vehicles, products or groups of such, found in a real life context; heterogeneity, because there is no need to aggregate different agents’ behavior into average variables; bounded rationality, since the individuals involved lack perfect information, having their own The impact of supply chain structures on performance Chapter 3 23 goals and their own policies; scalability and flexibility, and finally, low cost, since the software needed is open source. 3.2. MAS APPLICATIONS IN SCM A literature review has been performed in order to study the state of the art on MAS applications in SCM. Table 3.1 summarizes the reviewed literature. It can be seen that MAS has been used in the research of a wide variety of topics in SCM in the last decade, such as scheduling, coordination between enterprises, information sharing, order fulfillment process (OFP), collaborative production planning, and provider selection, among others. Further information can be found in Table 3.1, such as their development degree, the role of the agents involved and the software platform used (if any). The development degree gives an idea of the maturity of works. Three development degrees have been considered: - Low development degree: presents theoretical models which are not implemented in any software platform and hence, do not provide any results yet. - Medium development degree: models have been implemented and simulated, providing some coherent results. - High development degree: models have been implemented in real industry or used to solve a real problem. Most of the revised literature has a medium development degree, with just a few works with high development degree. From this analysis it is possible to get an idea of the maturity of MAS application in SCM: current research is already developing models and software, but it still needs to give a step through in the development of applications for real industry. The role of agents determines the granularity (or level of detail) of the models, from low granularity (agents modeling enterprises) to high granularity (agents modeling machines, trucks and other resources). In models with a medium granularity, agents model at a functional level (e.g. departments in each enterprise). Roberto Domínguez Chapter 3 24 Table 3.1. MAS literature review. Authors Topic Degree of development Role of agents Software Abid et al. (2004) Collaborative production planning High Functional Java/Cplex Ahn and Park (2003) Coordination Information sharing OFP Medium Functional Not described Alpay (2007) Scheduling Medium Resource Not described Alpay and Saricicek (2008) Scheduling Low Mixed N/A Álvarez and de la Calle (2009) Collaborative production planning Low Enterprise N/A Azevedo et al. (2004) OFP High Functional FIPA-OS/Java Bo and Zhiming (2003) Provider selection Order Management Medium Enterprise Not described Caridi et al. (2005) Collaborative production planning Medium Mixed SIMPLE++ Caridi et al. (2006) Collaborative production planning Medium Mixed SIMPLE++ Chatfield et al. (2001,2006,2009) Framework Medium Enterprise Java/XML/Silk ™ Cheeseman et al. (2005) Scheduling High Resource JADE Dong et al. (2006) Framework Medium Mixed Swarm Forget et al. (2008) Multi-Behavior Low Enterprise N/A Forget et al. (2009) Multi-Behavior OFP Medium Enterprise FORAC Framinan (2009) Order Management Medium Enterprise Swarm Fung and Chen (2005) Coordination Provider selection Medium Functional Not described Goh and Gan (2005) OFP Low Mixed N/A Govindu and Chinnam (2010) Framework High Mixed JADE/Java Guo and Zhang (2009) Scheduling Medium Resource Not described Hilletofth et al. (2009) OFP Medium Mixed Anylogic Ito and Abadi (2002) Inventory management Medium Mixed Java Julka et al. (2002) Framework High Mixed Not described Kaihara (2003) Resource allocation Medium Enterprise Not described Kiralp and Venkatadri (2010) Framework Medium Functional Not described Komma et al. (2011) Framework Medium Resource JADE The impact of supply chain structures on performance Chapter 3 25 Table 3.1. MAS literature review (continued). Authors Topic Degree of development Role of agents Software Lin et al. (1998) Information sharing OFP Medium Functional Swarm Lin and Shaw (1998) Information sharing OFP Medium Functional Swarm Lin et al. (2002) Information sharing Provider selection OFP Medium Mixed Swarm Lin et al. (2005) Provider selection OFP Medium Mixed Swarm Liu and Min (2008) Collaborative production planning Low Mixed N/A Long et al. (2011) Framework Medium Mixed JADE Nilsson and Darley (2006) OFP High Mixed Not described Panti et al. (2005) Coordination Low Functional N/A Papadopoulou and Mousavi (2007) Scheduling Medium Resource JACK Si and Lou (2009) Inventory management Provider selection Order Management Medium Enterprise Not described Strader et al. (1998) Information sharing OFP Medium Functional Swarm Swaminathan et al. (1998) Framework Medium Enterprise Not described Yoo and Glardon (2009) Framework High Mixed JADE/Repast Yu and Huang (2001) OFP Medium Functional Not described The role of the agents is chosen depending of the problem to be analyzed. In case of focusing on coordination between enterprises or collaborative planning, agents may play the role of enterprises, while in case of analyzing inventory management policies, order management or OFP, agents may play the functional role. Finally, in the analysis of scheduling problems or resource allocation it might be interesting to choose the resource role for agents. Furthermore, a mixture of roles is possible, allowing the analysis of problems at different levels of details, and increasing the flexibility and realism of models. The literature review reveals that, in fact, mixing different roles of agents is the preferred choice by authors, primarily mixing enterprise and functional roles. Roberto Domínguez Chapter 4 32 Figure 4.1. Two layer MAS framework design. According to this methodology, a bottom-up description of the framework is provided next, starting with the functional agents and then, describing how they can be combined within the Enterprise Agent to adopt different roles. 4.2. FUNCTIONAL AGENTS A good design of the functional layer is crucial because functional agents must capture the internal dynamic of a real enterprise and should model the key aspects of enterprise management. The initial design of the functional layer is based in the level 1 of the SCOR model (SCC, 2006) and the literature revised. Each one of the physical activities of SCOR (with the exception of Return) is modeled by one independent agent, so there are three physical agents: Source Agent, Make Agent and Deliver Agent. The Return activity is implicitly implemented by allowing the Deliver Agent to return products or receive returned products. According to the Supply Chain Planning Matrix in Stadtler (2005), the Plan activity has been divided into six planning functions, being each of these functions carried out by a different functional agent: Demand Fulfilment Agent, Demand Forecast Agent, Master Planning Agent, Production Planning Agent, MRP (Material Resource Planning) Agent and Scheduling Agent. The functional layer of an enterprise can be then modeled by a mix of planning agents and physical agents. Planning agents store management policies and take the main decisions. Physical agents control the physical resources of the enterprise and The impact of supply chain structures on performance Chapter 4 33 share information with planning agents. Table 4.1 summarizes the functional agent framework. Table 4.1. Functional agents framework. Category SCOR model Planning Matrix (Stadtler, 2005) Agents Main Tasks Planning Activities Plan Demand Fulfilment & ATP Demand Fulfilment Agent Demand management Communication with customers Purchasing & Material Requirements Planning MRP Agent Purchase management Communication with providers Demand Planning Demand Forecast Agent Demand forecast Master Planning Master Planning Agent Aggregate production planning Production Planning Production Planning Agent Disaggregate production planning Scheduling Scheduling Agent Jobs sequence Physical Activities Source - Source Agent Reception and storage of raw materials Make - Make Agent Manufacturing process (machines) Deliver - Deliver Agent Storage of finished products and delivery to customers Return - A brief description of the agents is presented below:  Source Agent. It handles the arrival and storage of raw materials, and its delivery to the manufacturing process when needed.  Make Agent. It monitors the manufacturing process by controlling machines and the flow of jobs in the shop floor.  Deliver Agent. It handles the arrival and storage of finished products, and the delivering of orders to customers. Roberto Domínguez Chapter 4 34  Demand Fulfilment Agent. It is in charge of demand management and inventory control. It checks incoming RFQs (Request for Quotations) from customers and quotes the due dates. If the order is accepted, it tries to fulfil them from inventory, if available. If inventory is not enough, it sends a production order (a Job) to the Scheduling Agent if the enterprise is a manufacturer. Otherwise it sends a purchase order to the MRP Agent.  Demand Forecast Agent. It requests the enterprise demand historical data to the Demand Fulfilment Agent in every forecast period. Then it forecasts the demand for each product in the next periods using a forecasting rule (like Simple Moving Average, Exponential Smoothing, etc.).  Master Planning Agent. It uses forecast information from the Demand Forecast Agent and generates an aggregate Master Plan for the products concepts defined by the enterprise by solving a linear programming model (that includes capacity and inventory restrictions, as well as production and inventory holding costs), obtaining the production needs per product concept for each period.  Production Planning Agent. It receives the Master Plan and uses it to generate a detailed Production Plan, obtaining the production needs for the final products to accomplish the Master Plan.  MRP Agent. It creates a detailed material plan to fulfil the Master Plan. If there is no Master Plan, it takes control of the raw material inventory levels by using some inventory policy. This agent is in charge of the purchases in the enterprise.  Scheduling Agent. It schedules production orders (Jobs) coming from the Production Planning/Demand Fulfilment Agent by using first some priority rule to create an initial solution, and then a heuristic to improve the initial solution according to certain objective. It also calculates start and end times for each job (so it can help the Demand Fulfilment Agent for due date calculation), generating a detailed schedule. According to this schedule, at the starting time of each job it sends the job information to the Make Agent to start its production. The impact of supply chain structures on performance Chapter 4 35 The overall configuration of a generic enterprise, with all its functional agents is shown in Figure 4.2. Demand Fulfilment Agent Demand Forecast Agent Master Planning Agent Production Planning Agent MRP Agent Scheduling Agent Source Agent Make Agent Deliver Agent Planning Agents Physical Agents Enterprise Providers Customers C1 C2 Cn P1 P2 Pn Raw Materials Finished Products ... ... Figure 4.2. Multi-Agent Framework: functional layer display of a generic enterprise. 4.3. ENTERPRISE AGENT The Enterprise Agent is able to model any kind of enterprise in the SCN: Suppliers, Manufacturers, Distributors, Retailers, etc. To do so, it can be customized by a combination of different functional agents, which determines the global behavior of the Enterprise Agent and its role in the SCN. Enterprises with similar functions (in terms of functional agents) belong to the same category. The identification of these roles is based on Röder and Tibken (2006): the basic role of an enterprise is determined by its combination of structural agents. These authors identified the following roles: - The SMD-enterprise type (Source-Make-Deliver) represents an enterprise that contains the whole intra-enterprise process chain including source, make and deliver processes. - The MD-enterprise (Make-Deliver) is similar to the SMD-enterprise, have no inventory for incoming goods. The material coming from the suppliers is Roberto Domínguez Chapter 4 36 delivered directly into the production process by “just-in-time” or “just-in- sequence” strategies. - The SD-enterprise (Source-Deliver) does not have production processes. This enterprise is represented by sourcing and delivering processes, having an inventory for incoming goods. According to the above classification, four roles can be adopted by the Enterprise Agent: - Factory, based on the SMD-enterprise for modeling Manufacturers and Assemblers type enterprises. This role needs the entire physical agents for sourcing raw materials, making new products and delivering them to customers, and three planning agents for managing orders and purchases, as well as scheduling production. Optionally, they can include forecasting and production planning abilities. - Intermediary, based on the SD-enterprise for modeling Distributors, Wholesalers and Retailers type enterprises. This role needs one physical agent to store and deliver products, and two planning agents for managing orders and purchases. Optionally, they can include forecasting abilities. - External Provider, which is a simplification of the SD-enterprise for modeling Providers type enterprises that are out of the SCN (Source is not required). This role needs a physical agent to deliver products to members of the SCN and a planning agent to manage the orders received. - External Customer, which is a simplification of the SD-enterprise for modeling Customers that are out of the SCN (Deliver is not required). This role needs a physical agent to source products from members of the SCN and a planning agent to manage the orders placed to them. A summary of these roles is provided in Table 4.2. In order to model a new enterprise with this framework, it is enough to select one of the roles from Table 4.2 for the Enterprise Agent and automatically the required functional agents are assigned to it. The basic agents always belong to the Enterprise Agent for the selected role, while the optional agents are selected depending on the enterprise characteristics. Although the roles included in the framework should be enough to model all kind of enterprises in the SCN, it might be possible to create new The impact of supply chain structures on performance Chapter 4 37 roles in case of necessity by selecting a different combination of functional agents for the Enterprise Agent and/or adding new functions to the functional agents. Table 4.2. Roles of the Enterprise Agent. Roles in Framework Description Examples types Enterprise type (Röder and Tibken, 2006) Basic Agents Optional Agents Factory The enterprise is in the SCN and has manufacturing or assembly capacity Manufacturer, Assembler SMD-type Demand Fulfilment, MRP, Scheduling, Source, Make, Deliver Demand Forecast, Master Planning, Production Planning Intermediary The enterprise is in the SCN and has not manufacturing or assembly capacity Distributor, Wholesaler, Retailer SD-type Demand Fulfilment, MRP, Deliver Demand Forecast External Provider Any enterprise who provide something to the SCN, but it’s out of the SCN Provider - Demand Fulfilment, Deliver No External Customer Any enterprise who demand something to the SCN, but it’s out of the SCN Customer - MRP, Source No 4.4. SUMMARY AND CONCLUSIONS This chapter describes the design of a MAS-based framework for SCN modeling, which takes use of the knowledge obtained from the literature review on MAS applications to SCN modeling described in the previous chapter. Summarizing, the designed framework presents the following highlights:  It presents a simple agent structure, with one generic agent modeling all type of enterprises in the SCN and a collection of functional agents to carry on the main functions of the enterprise. These functions include the three main physical functions (source, make and deliver) and a planning function, according to SCOR (SCC, 2006), and the planning function has been divided into six subfunctions according to the Supply Chain Planning Matrix from Stadtler (2005).  It is intuitive, since each functionality or department is encapsulated in one different agent. Furthermore, this modular design allows to easily improving and Roberto Domínguez Chapter 4 38 customizing each functional agent independently, focusing on the particular function of interest.  The two-layer design allows analyzing inter-enterprise as well as intra-enterprise processes.  The reusability and customizability of the Enterprise Agents allows quickly creating a high number of enterprises with different roles, connecting them and obtaining large SCNs, making possible the analysis of complex SCN structures. 39 Chapter 5: Framework implementation. A SCN simulation tool: SCOPE 5.1. INTRODUCTION In the previous chapter, a MAS-based framework for SCN modeling has been developed. It uses many concepts of previous works found in literature and discussed in Chapter 3 to ensure an accurate modeling of the SCN processes as well as flexibility and modularity. In order to be able to use it to provide results, the framework has to be implemented in a MAS simulation package. Furthermore, it has to be tested and compared with other models to ensure that the results offered by SCOPE are accurate. In this chapter, a description of the simulation package used and the implementation of the framework are provided, as well as a validation test with several models found in literature. 5.2. MAS SIMULATION TOOLS There are many options in choosing a MAS software platform. Most of the commonly used MAS platforms follow the “framework and library” paradigm, providing a framework (a set of standard concepts for designing and describing MASs) along with a library of software implementing the framework and providing simulation tools (Railsback et al., 2006). Some of the most commonly used MAS platforms are summarized next:  Swarm (Minar et al., 1996). The initial version of the library of Swarm was written in Objective-C. Later, a Java version of Swarm was developed, in which a set of simple Java classes allowed use of the Swarm’s Objective-C library from Java. It was designed as a general language and toolbox for MAS, intended for widespread use across scientific domains. A key concept of Swarm is the swarms, which help in organizing models at different levels of detail.  Repast. Its initial design started as a Java implementation of Swarm, but it diverged significantly. Repast did not adopt all of Swarm’s design philosophy and does not implement swarms. It was also clearly intended to support one Roberto Domínguez Chapter 5 40 domain (social science in particular). Furthermore, the schedule executes toplevel actions in randomized order (which is not desirable), while Swarm allows a precise control of the sequence of actions.  MASON (Multi-Agent Simulator of Networks) (Luke et al., 2005). It was designed as a smaller and faster alternative to Repast, with a clear focus on computationally demanding models with many agents executed over many iterations. However, MASON is the least mature of these platforms, with only a few basic capabilities and a complex programming language.  Netlogo. Its primary design objective is clearly ease of use. With a programming language that includes many high-level structures and primitives, it greatly reduces programming effort. However, the language contains many but not all the control and structuring capabilities of a standard programming language. Further, NetLogo was clearly designed with a specific type of model in mind: mobile agents acting concurrently on a grid space with behavior dominated by local interactions over short times, and that are not extremely complex. There are many other tools for MAS design, like JADE, Zeus, JACK, ADE, Anylogic, etc. SCOPE has been implemented in Swarm (Swarm Development Group Wiki, 2003) Java version, using NetBeans IDE 6.7 as implementation framework. Swarm has been chosen due to its high maturity, its model organization (nested swarms), which helps in modeling SCNs, its generic and low-level programming language (Java), and its suitability for modeling CAS (Minar et al., 1996) and SCNs (Lin et al., 1998). It provides object oriented libraries of reusable components for building models and analyzing, displaying, and controlling experiments on those models. A comparison of similarities between SCN’s features and Swarm can be seen in Table 5.1. For solving the planning models included in the Master Planning and Production Planning agents (see Chapter 4), SCOPE can be connected with Gurobi solver through a special library for Java. Gurobi is a commercial software package for solving largescale mixed-integer linear optimization problems. In the next section, a description of the Swarm modeling paradigm is provided. The impact of supply chain structures on performance Chapter 5 41 Table 5.1. A comparison between SCNs and Swarm (adapted from Lin et al., 2002). SCNs Swarm Composition of autonomous and semi-autonomous business entities A swarm of agents with individual based modeling Business entities act different organizationally Agents are constructed with internal state variables and action functions Multiple layer abstraction Nested inherent hierarchy Information flows between business entities Message passes between agents Material flow during procurement, manufacturing, and distribution activities Discrete event simulation and time-stepped scheduling to trigger agent actions Global performance contributed by the processes of individual entities Collective behaviour contributed by the combination of individual behaviours Visibility determined by the information boundary Visibility determined but the boundaries of message passing 5.3. SWARM MODELING TOOL The modeling formalism that Swarm adopts is a collection of independent agents interacting via discrete events triggered by schedules. Within that framework, Swarm makes no assumptions about the particular sort of model being implemented. The basic unit of a Swarm simulation is the agent: any actor in the system, any entity that can generate events that affect itself and other agents. A schedule is a data structure that combines actions in the specific order in which they should execute. The passage of time is modeled by the execution of the events in some sequence. Simulations consist of groups of many interacting agents. The fundamental component that organizes the agents of a Swarm model is a “swarm”, a collection of agents executing a schedule of actions. Swarm supports hierarchical modeling approaches whereby agents can be composed of swarms of other agents in nested structures. The ability to build models at various levels of detail can be very powerful. In addition to being containers for agents, swarms can themselves be agents. In this case, the agent's behavior is defined by the emergent phenomena of the agents inside its swarm. The swarm represents an entire model: it contains the agents as well as the representation of time (schedules). Measurement happens by the actions of observer agents, special objects whose purpose it is to observe other objects. The observer agents themselves are a swarm, a group of agents and a schedule of activity. Roberto Domínguez Chapter 5 48 Since the resultant simulation software has been built-up from its initial design phase around a generic SCN modeling tool concept, it can be used to research on a wide variety of topics within the SCM, as it can be seen from Table 5.2 where the main modeling characteristics of the current version of SCOPE are summarized. Table 5.2. Modeling characteristics of SCOPE. Enterprises  Unlimited number of fully customizable enterprises  Four roles: Manufacturer, Intermediary, External Provider, External Customer Products  Unlimited number of different products  Individual manufacturing characteristics (processing times, raw material requirements, shop floor machine routes)  Individual inventory control policies SCN structure  Total flexibility in modeling different SCN structures (serial, convergent, divergent, conjoined, etc.) Demand control policies  Make-to-Order (MTO): Online orders and batching  Make-to-Stock (MTS): (r,S), (s,S) and (r,Q) policies  Assemble-to-Order (ATO) BWE avoidance  Information sharing, smoothing OUT Shop floor  Individual shop floor configuration for each manufacturer (Job-shop & Flow-shop configurations)  Unlimited number of different processing machines Production scheduling  Scheduling priority rules (FCFS, SPT, LPT)  Optimizing makespan and flowtime by heuristics (Greedy) Production planning  Aggregate master planning and detailed production planning  Solving planning models by an external connection with Gurobi solver Uncertainties  Machine processing time, Lead time, External demand, External supply Demand forecast  Simple Moving Averages (SMA), N-Periods Moving Averages (NPMA), Weighted Moving Averages (WMA), Simple exponential smoothing (SES) Random distributions  Uniform, Normal, Poisson, Exponential, Gamma Other  Random external demand according to distribution  Step demand (step in demand mean)  Reverse Logistics (allowing to return products)  Inventory record data errors  Due dates The impact of supply chain structures on performance Chapter 5 49 Figure 5.3. Source code analysis. 5.5. VALIDATION OF SCOPE In order to validate SCOPE a literature review has been done, looking for SCNs that have been already modeled and simulated by other authors and comparing their results with those provided by SCOPE. The selected work must be published in an important journal and it must provide enough information to reproduce the experiments. Chatfield et al. (2004) validated their software SISCO by comparing their results with those by Chen et al. (2000) and Dejonckheere et al. (2003a). Since they provide enough information to reproduce the validation experiments and use a double comparison with two well-recognized works, SCOPE has been validated using the same procedure described in Chatfield et al. (2004). Furthermore, a third validation scenario has been performed, reproducing some of the experiments conducted in Chatfield et al. (2004) after the validation of SISCO. The SCN’s structure is the same for the first two scenarios (Figure 5.4). There is only one product in the SCN, and it is structured by a serial formation of customer, retailer, wholesaler, distributor, and factory levels. The lower node places orders with the next upper node and this node fills these orders. The customer does not fill orders and the factory places orders with an outside supplier. Roberto Domínguez Chapter 5 50 Figure 5.4. Serial SCN modeled by Chen et al. (2000) and Dejonckheere et al. (2003a). There are other characteristics that remain identical for both scenarios:  Lead Time: there is a fixed lead time between the time an order is placed at a stage i and when it is received at that stage, such that an order placed at the end of period t is received at the start of period t+L.  Customer Demand: these are normally distributed. Negative demands are allowed.  Demand Forecast: enterprises use p-periods moving average (all enterprises use the same “p” parameter).  Inventory Policy: Enterprises use a (r,S) policy with a review period 𝑟=1 and a dynamic OUT level: 𝑆=𝑋+𝑧𝑠𝑥 (5.1) Where 𝑋 is the estimated demand over the protection time 𝐿+𝑟. For simplicity 𝑧=0, so if 𝐷 is the demand estimation, the OUT level is given by the following equation: 𝑆=𝑋=(𝐿+𝑟)𝐷 (5.2) If the inventory level is lower than the 𝑆 level, enterprises are allowed to return goods. The impact of supply chain structures on performance Chapter 5 51 5.5.1. Scenario 1: Chen et al. (2000) Chen et al.’s (2000) coining of the Order Rate Variance Ratio (ORVR) as the quantification metric for demand amplification along a SCN could reasonably be considered the starting point of the current research period in this domain (Cannella and Ciancimino, 2010). They calculated a statistical lower bound for the order variance amplification. For the SCN described above, the proposed lower bound is given by equation (5.3). 𝑉𝑎𝑟(𝑞𝑘) 𝑉𝑎𝑟(𝐷)≥∏(1+2𝐿𝑖 𝑝+2𝐿𝑖2 𝑝2) 𝑘 𝑖=1 ,∀𝑘 (5.3) Results obtained are very close to those offered by SISCO, although greater amplifications than Chen et al. (2000) results have been found at upper SCN stages. Chatfield et al. justify these results arguing that the bounds provided by Chen et al. do not account for interactions and interdependencies present in a multi-stage system. To test this, they perform a “sequential pairs execution” simulation, in which they broke the SCN into four two-node sub-chains (customer-retailer, retailer-wholesaler, wholesaler - distributor, distributor - factory). Then, they simulate each sub-chain using the ordering mean and standard deviation obtained from the simulation of the previous sub-chain. This new scenario has been simulated by SCOPE, obtaining the results in Table 5.3, which are extremely close to those predicted by Chen et al., and SISCO. Table 5.3. Amplification Ratio for Chen et al. (2000). Parameters: demand rate = 𝑵(𝟓𝟎,𝟐𝟎𝟐); protection time = L+R = 4+1 = 5; NPMA(15) forecasting; simulation time = 5200 time periods (200 for warm-up). Enterprise Chen et al. SISCO SCOPE Retailer 1.89 1.90 1.90 Wholesaler 3.57 3.59 3.53 Distributor 6.74 6.70 6.66 Factory 12.73 12.84 12.58 Roberto Domínguez Chapter 5 52 5.5.2. Scenario 2: Dejonckheere et al. (2003a) In Dejonckheere et al. (2003a), authors proposed another measure for the order variance amplification using a Control Engineering methodology. Considering a SCN with the same characteristics that the one described before, they obtained the result shown in equation (5.4). The results obtained for this scenario are summarized in Table 5.4, where it can be noted that SCOPE performs very similar to SISCO and to Dejonckheere et al. (2003a). 𝑇𝐹𝑛=[−2−𝑇𝑝+2𝑧𝑇𝑚+𝑇𝑚𝑧𝑇𝑚+𝑇𝑝𝑧𝑇𝑚 𝑇𝑚𝑧𝑇𝑚 ]𝑛 (5.4) Table 5.4. Amplification Ratio for Dejonckheere et al. (2003a). Parameters: demand rate = 𝑵(𝟏𝟎𝟎,𝟏𝟎𝟐); protection time = L+R = 3+2 = 5; NPMA(19) forecasting; simulation time = 5200 time periods (200 for warm-up). Enterprise Dejonckheere et al. SISCO SCOPE Retailer 1.67 1.67 1.71 Wholesaler 2.99 2.99 3.10 Distributor 5.72 5.72 5.96 Factory 11.43 11.43 11.93 5.5.3. Scenario 3: Chatfield et al. (2004) After the validation of SISCO, in Chatfield et al. (2004) authors analyzed the impact of information sharing and different degrees of information quality on the BWE when the lead time between enterprises is stochastic. The SCN used is similar to the one used in the previous scenarios, with only some differences:  Lead Time: is stochastic and gamma distributed, and the mean lead times at the customer, retailer, wholesaler, distributor, and factory are 0, 4, 4, 4, and 4 time units, respectively.  Customer Demand: negative demand is not allowed. The impact of supply chain structures on performance Chapter 5 53  Inventory Policy: the equation (5.1) is now used with a safety factor of z = 2.0 and the experiments are reproduced for three degrees of information quality, named IQL0, IQL1 and IQL2 (IQL stands for Information Quality Level). IQL0 means that enterprises don’t update the S level. For IQL1 and IQL2 enterprises use demand rate (D) and lead time (L) data available to generate forecasts of lead time demand (𝑋) and variance (𝑠𝑥2). In the case of IQL1: 𝑋=(𝐿+𝑟)𝐷 (5.5) 𝑠𝑥2=(𝐿+𝑟)𝑠𝐷2 (5.6) IQL2 uses lead time variation in the calculation of 𝑠𝑥2, as it appears in equation (5.7): 𝑠𝑥2=(𝐿+𝑟)𝑠𝐷2+𝐷2∙𝑠𝐿2 (5.7)  Forecast: demand estimation (𝐷, 𝑠𝐷2) is doing at each node using a p-period moving averages, NPMA(p), and “moving variances,” NPMV(p), with p = 15. Lead time estimation (𝐿, 𝑠𝐿2) is doing at each node using running averages and variances (“all data” approach). SCOPE is used to reproduce two of the experiments carried out by Chatfield et al. for this SCN. The first experiment analyzes the influence of the different information quality levels on the BWE for a lead time c.v. = 0.50. The results obtained by SCOPE (see Figure 5.5, above) are very close to those obtained by SISCO in Chatfield et al. (2004). The second experiment analyzes the impact of the variance of the lead time on the standard deviation of orders for a given information quality level (IQL2). Again, the results obtained by SCOPE (see Figure 5.5, below) are very close to those obtained by SISCO in Chatfield et al. (2004). Roberto Domínguez Chapter 5 54 Figure 5.5. A comparison between SCOPE and SISCO. 0 50 100 150 200 250 300 Customer Retailer Wholesaler Distributor Factory Std Dev. of Orders Lead time c.v. = 0.50 No information sharing IQL0 IQL1 IQL2 SCOPE 0 50 100 150 200 250 300 Customer Retailer Wholesaler Distributor Factory Std Dev. of Orders Information Quality Level 2 No information sharing C.V.=0 C.V.=0.25 C.V.=0.5 C.V.=0.75 SCOPE The impact of supply chain structures on performance Chapter 5 55 5.6. SUMMARY AND CONCLUSIONS A well-structured framework with a modular design has been developed and implemented in a MAS platform, obtaining a software for SCN simulation that allows modeling a great variety of problems in real-scale SCNs. It is built around the frameworks previously developed by other authors. This tool is useful either for SCN managers or researchers. Managers often have to take local decisions without knowing in advance the consequences of these decisions on the global SCN, mainly due to the complexity of real SCNs and the many interactions existing between the member enterprises. SCOPE is a powerful tool that allows constructing the global system easily, defining the individual enterprises and their interactions. It might be useful for managers in several ways:  Improving the understanding of the current SCN configuration, allowing to calculate the global performance of the system, as well as individual performances of target enterprises, considering real disturbances, like stochastic demand, lead times, and more.  Managers interested in re-engineering the present SCN configuration, implying either structural or operative changes, are able to test several “what-if” scenarios on the SCN model and after the simulations decide which changes are more beneficial for the SCN. If the results obtained do not fulfil the expectations, managers can fine-tune their initial design and improve it by doing several iterations.  Analysis of the performance in a wide variety of topics within the OFP, like the inventory systems, the planning models, the scheduling rules, the configuration of shop-floors, capacity needs, etc.  The benefits of collaboration techniques have been proved by several authors (see e.g. Cannella and Ciancimino, 2010; Cannella et al., 2011; Ciancimino et al., 2012). The distributed nature of autonomous MAS and the structure of SCOPE in two layers (enterprise and functional) allow to implement and test these collaboration techniques on several SCN structure with different levels of details (Domínguez et al., 2013), helping managers to decide if its implementation is beneficial or not for the SCN. Roberto Domínguez Chapter 5 56 Researches can benefit from the open-source code of SCOPE (in Java) and its modular design with the main functions of the enterprises encapsulated in different functional agents to make easier the process of adding new functions, allowing to improve and customize the platform in the desired way. PART III Supply chain’ structure and bullwhip effect Roberto Domínguez Chapter 6 64 Chatfield et al. (2004), each experiment consists in 30 replications of 700 periods, with the first 200 periods of each replication removed as a warm-up used to set up the system. The results obtained from the replications are averaged for each experiment. To be able to compare the experiments under both lenses, metrics are calculated in the same simulation period, after the impulse time (t=450). The first set of experiments is summarized in Table 6.1. Table 6.1. Stationary response set of experiments. BWE Lens Demand Pattern Structure of the SCN Metrics Variance Lens 𝑁(50,202) 𝑡∈ [0-700] Serial SCN Φ BwSl 𝑡∈ [450-700] Divergent SCN Shock Lens 𝑁(50,202) 𝑡∈ [0-449] 𝑁(100,202) 𝑡∈ [450-700] Serial SCN Divergent SCN In the second set of experiments, in order to obtain the temporal response, each SCN is evaluated in different simulation periods. In the first observation, named T0, SCNs are simulated until the simulation time is just before the demand impulse occurs, obtaining the initial Φ. Then, Φ is measured in a sequence of experiments where the simulation time starts at the demand impulse instant and the simulation time is increasing in intervals of 25 or 50 periods until the end of the original simulation time is reached (t=700), resulting in the experiments T1-T6. As for the first set, each experiment consists in 30 replicates, and the results obtained are averaged. This set of experiments is summarized in Table 6.2. Time Market demand Time Market demand The impact of supply chain structures on performance Chapter 6 65 Table 6.2. Dynamic response set of experiments. BWE Lens Demand Pattern Simulation Periods Structure of the SCN Metri cs Shock Lens 𝑁(50,202) 𝑡∈ [0-449] 𝑁(100,202) 𝑡∈ [450-700] T0: [200-449] Serial/Divergent Φ T1: [450-475] Serial/Divergent T2: [450-500] Serial/Divergent T3: [450-550] Serial/Divergent T4: [450-600] Serial/Divergent T5: [450-650] Serial/Divergent T6: [450-700] Serial/Divergent 6.4. NUMERICAL RESULTS 6.4.1 Stationary response set Under the variance lens, results obtained for Φ are very similar for both SCNs (see Figure 6.2), being slightly higher for the divergent SCN at the upper stages. However, under the shock lens there is an important difference between both SCNs, as Φ is considerably higher for the divergent SCN (see Figure 6.3). The average results for Φ and BwSl, as well as the differences between both SCNs (∆(%)=𝛷 𝐷𝑖𝑣𝑒𝑟𝑔𝑒𝑛𝑡−𝛷 𝑆𝑒𝑟𝑖𝑎𝑙 𝛷 𝑆𝑒𝑟𝑖𝑎𝑙 ∗ 100%) are shown in Table 6.3, together with the corresponding 99%-confidence intervals. Figure 6.2. Φ under the Variance Lens. Time Market demand 0 10 20 30 40 50 60 70 Customers Retailers Wholesalers Distributors Factory ΦSerial SCN Divergent SCN Roberto Domínguez Chapter 6 66 Under the variance lens, it can be seen that the values of the measures are not statistically different, which indicates a rather similar performance for both SCNS. At the lower stages, the increase of Φ is below 1%, while at the upper stages the differences are slightly higher (Φ is 5.39% higher for the divergent SCN at the distributor stage and 6.08% at the factory stage). BwSl helps to easily compare both SCNs. The propagation of the BWE is very similar for both SCNs, being slightly higher (6.20 %) for the divergent SCN. Table 6.3. Numeric results for Φ and BwSl. Lens SCN structure Φ BwSl Retailer Wholesaler Distributor Factory Variance Lens Serial 2.253±0.031 6.329±0.177 19.153±0.737 57.766±2.667 13.04±0.60 Divergen t 2.258±0.029 6.331±0.169 20.186±0.763 61.276±2.652 13.85±0.60 ∆(%) 0.222 0.032 5.393 6.076 6.20 Shock Lens Serial 2.655±0.013 7.732±0.120 23.453±0.496 69.539±1.739 15.79±0.39 Divergen t 2.690±0.012 8.923±0.119 39.595±0.821 136.196±2.893 30.73±0.65 ∆(%) 1.318 15.404 68.827 95.856 94.62 Figure 6.3. Φ under the Shock Lens. Time Market demand Time Market demand 0 20 40 60 80 100 120 140 160 Customers Retailers Wholesalers Distributors Factory ΦSerial SCN Divergent SCN The impact of supply chain structures on performance Chapter 6 67 Under the shock lens, the Φ curve is clearly stepped for the divergent SCN, with the minimum increase at the retailer stage (1.32% over the serial SCN) and the maximum increase at the factory stage (95.86% over the serial SCN). The bad performance of the divergent SCN in this case is well summarized by the value of BwSl, being 94.62% higher than in the serial SCN. Note that the differences in the indicators for both SCNs are statistically different, thus confirming that the divergent SCN performs worse than the serial SCN in this scenario. 6.4.2 Dynamic response set Figure 6.4 shows the evolution of Φ over time for each stage after the demand impulse (rhomboids dots for the serial SCN and square dots for the divergent SCN). Order Rate Variance Ratio Figure 6.4. Evolution of Φ over time under the Shock Lens. 0 200 400 600 800 1000 T0 T1 T2 T3 T4 T5 T6 Factory 0 200 400 600 800 1000 T0 T1 T2 T3 T4 T5 T6 Distributors 0 5 10 15 20 25 30 35 T0 T1 T2 T3 T4 T5 T6 Wholesalers 0 5 10 15 20 25 30 35 T0 T1 T2 T3 T4 T5 T6 Retailers Roberto Domínguez Chapter 6 68 The differences for Φ observed between both SCNs in Figure 6.4 are plotted in Figure 6.5, showing the temporal evolution of ∆(%) for each stage. Figure 6.5. Divergent SCN Φ increments over the serial SCN. From the above results it can be noted that, under an unexpected impulse in demand average: - Both SCNs react by: 1) immediately incrementing their order variances in all stages, and 2) decreasing their order variances over time. - The highest increase in Φ takes place just after the demand impulse (T1). The difference between both SCNs is maximal at this point, being higher as we move upstream (see Figure 6.5). - The shock recovery is similar for both SCNs at the lower stages (retailers and wholesalers), whereas ∆(%) is near to zero after T3 (see Figure 6.5). However, at the upper stages (distributors and factory), shock recovery is slower for the divergent SCN, obtaining high values of ∆(%) until the end of the simulation time (T6). In Figure 6.6, the order pattern at the factory stage is plotted against the customer order pattern for both SCNs under the shock lens. It is easy to see the high overreaction of the divergent SCN when the demand impulse occurs. -100 0 100 200 300 400 500 600 700 T1 T2 T3 T4 T5 T6 (%) Factory Distributors Wholesalers Retailers The impact of supply chain structures on performance Chapter 6 69 Figure 6.6. Factory vs end customer demand order patterns under the shock lens. Finally, a sensitivity analysis has been performed by systematically increasing the level of end customer standard deviation in the shock lens part of the simulation. The results show that as the impulse in customer demand variability increases, standard deviation of the orders placed in the lower echelons does not increase at the same rate. For example, 47.49% increase in customer demand standard deviation in the shock lens, resulted a 25.96% increase in the standard deviation of the factory orders. In other words, the increase in the shock was transmitted in lower proportions towards the upstream levels of the SCN. 6.5. FINDINGS AND MANAGERIAL IMPLICATIONS The results obtained in the previous section give new insights on the BWE research topic, considering two different lenses for the comparison of two different SCN structures. Under the variance lens, the following comments can be done: - The BWE found in the serial and the divergent SCNs are very similar. When the demand is predictable and the nodes can adequately adjust their inventory levels -600 -100 400 1 21 41 61 81 101 121 141 161 181 201 221 241 261 281 301 321 341 361 381 401 421 441 461 481 Serial SCN Factory Customer -1500 500 2500 4500 6500 1 21 41 61 81 101 121 141 161 181 201 221 241 261 281 301 321 341 361 381 401 421 441 461 481 Divergent SCN Factory Customers Roberto Domínguez Chapter 6 70 to fulfill the demand with a high customer service level, both SCNs are quasiequivalents. A node at the stage i of the divergent SCN causes the same amplification of orders that a node in the same stage i of the serial SCN, because they have the same OUT and forecast policies. The orders received by each node are proportional to the end customer demand, and hence, to the amplification of orders caused by them. As the variance of orders in each stage is rated to the end customer demand variance, each stage produces similar values of Φ for both SCNs. - The small increase observed in Φ for the divergent SCN in Figure 6.2 is caused by eventual excess of stock or by eventual stock-outs. Due to the uncertainties in the end customer demand and lead times, sometimes either the demand received may be different than the demand forecasted in the previous period, or the orders arrive earlier or later than expected, causing this phenomenon. In these cases, where the inventory level is far from the desired OUT level, a node reacts by ordering a big quantity of products (a positive order in case of stock-out and a negative order in case of excess of inventory). These exceptionally high orders are amplified upstream, increasing the variance ratio mainly in the upper stages. In view of the fact that for each node there is a certain probability that this phenomenon occurs, and that the divergent SCN has a higher number of nodes in each stage (higher vertical complexity), it happens more frequently in the divergent SCN, causing the little increment in the values of Φ at the upper stages (distributor and factory). As a summary, it can be concluded that the divergent SCN has almost the same performance in terms of BWE than the serial SCN when the end customer demand does not suffer important changes. Using the shock lens, the following comments can be done: - Under the shock lens both SCNs are stress tested. The end customer demand impulse causes a massive stock-out situation at the retailer stage, which is then propagated and amplified along the SCN, causing stock-outs in all the stages of the SCNs. While the factory in the serial SCN has to manage the instability caused by the stock-out of one retailer, the same factory in the divergent SCN has to manage it with the stock-outs of eight retailers. The disproportional orders of the factory and distributors in the divergent SCN can be observed in Figure The impact of supply chain structures on performance Chapter 6 71 6.6, and are the cause of: the excess of variance observed in Figure 6.3, the high peaks of variances, and the slow recovery observed in Figure 6.4. - The divergent SCN has a bad performance as compared to the serial SCN under important unpredicted changes in demand tendencies. Thus, it can be concluded that divergent SCNs are less robust than serial SCNs. It is worth mentioning the relevance of the framework for the analysis of the BWE proposed by Towill et al. (2007). The authors stated that “the detection of BWE depends on which lens is used”, and they proposed three different lenses for BWE analysis (variance, shock and filter lens). The experiments have shown different behaviors depending on the lens used: while for the classical variance lens analysis (stationary stochastic demand input) the BWE is similar for both SCNs, the shock lens analysis (step demand input) reveals that the divergent SCN performs worse than the serial SCN. With respect to the managerial implications of the study, to face up with the less robustness of divergent SCNs, managers may find useful to consider the following: - Under a shock in end customer demand, the BWE increases when the SCN structure becomes more complex as the number of echelons increases, or as the number of successors at each echelon increases. Thus, to mitigate this incremental BWE, a firm could consider simplifying the SCN structure by reducing the number of echelons or by reducing the number of successors (Sodhi and Tang, 2011). This is particularly important for SCNs characterized by high variations in the end customer demand. On the contrary, traditional arborescent SCNs operaing in markets characterized by a stable consumer demand are less prone to the detrimental consequences of the demand amplification phenomenon. - An adequate forecast method adjusted to the end customer demand would prevent the firm from eventual excess of stock or from stock-outs. Therefore, it is important to make an effort to implement techniques in order to better understand the end customer demand tendencies (i.e. surveys) and to anticipate important changes. - The implementation of well-known techniques for reducing the BWE (i.e. information sharing) is highly desirable. These techniques may help managers to Roberto Domínguez Chapter 6 72 have a better control of the BWE in case of important changes in the end customer demand that cannot be anticipated by the above techniques. However, it has yet to be proved how these techniques (usually tested in serial SCNs) perform in non-serial SCNs. 6.6 SUMMARY AND CONCLUSIONS The literature review on Chapter 2 has revealed a lack of research on the BWE topic when the structure of the SCN is different than a serial SCN. However, real SCNs rarely adopt a traditional serial structure, often following a more complex topology. The work presented in this chapter is an attempt to cover this research gap by analyzing the BWE in a divergent SCN and by comparing its performance with those of a serial SCN already analyzed in the literature by several authors. The BWE has been observed both from a static and a dynamic perspective, being measured at the node level by the Order Rate Variance Ratio, and at the network level by the Bullwhip Slope. The main result obtained show that divergent SCNs are more sensitive to unexpected violent changes in demand signal than serial SCNs. Two situations have been considered: - Variance lens, i.e. stationary demand signal. In this case the performance of both SCNs is very similar, being just a little worse for divergent SCNs. - Shock lens, i.e. demand signal suffers an unexpected violent change. In this case the performance of the divergent SCN is much worse than that of the serial SCN, showing higher variance of orders and taking more time for recovery, incurring in higher costs. As it was pointed out in the previous section 6.5, the less robust structure of the divergent SCN might be compensated by a good information system in order to share end customer demand (information sharing) or applying smoothing replenishment orders, allowing a faster and proportional response to violent changes in the end customer demand. Such information system and smoothing replenishment are adapted to a divergent SCN in Chapter 7, testing its efficiency in this SCN configuration. Bhattacharya and Bandyopadhyay (2011) indicated that there are operational and behavioral causes of the BWE, and that the root of all the causes is the lack of The impact of supply chain structures on performance Chapter 6 73 coordination among the SCN members. In this chapter it has been shown that, in addition to the number of stages, there are also other structural factors that amplify the BWE caused by those operational and behavioral factors. The identification of these factors and a quantification of their effects are addressed in Chapter 8. Finally, the different performance observed between the traditional serial SCN model and a divergent SCN model with higher number of companies and higher interconnection confirms the need of increasing the complexity of the SCN models in order to obtain more accurate insights on the dynamics of modern SCNs (see Chapter 2). Roberto Domínguez Chapter 7 80 7.4. RESULTS AND DISCUSSION The numerical output of the experiments is presented. Data are collected and metrics are herein used to assess performance of the SCNs. In order to contrast the scenarios, the Order Rate Variance Ratio measures (Φ) are plotted using the echelon position as independent variable, according to Dejonckheere et al.’s notation (2003a) (Figure 7.1). Discrepancies between the serial SCN and the divergent SCN are plotted in Figure 7.2. Finally, Table 7.2 reports the values of Φ by echelon (columns) and by SCN configuration (rows). Furthermore, in order to concisely compare the different scenarios, the values of the bullwhip slope are also reported in Table 7.2 for every SCN configuration and the discrepancies between the serial SCN and the divergent SCN as well. To test the statistical significance of the scenarios, the 99%-confidence interval is calculated for each one. The confidence intervals are presented next to the Φ and BwSl values in Table 7.2. The values obtained show that all the scenarios simulated are statistically different. Order Rate Variance Ratio Figure 7.1. Order Rate Variance Ratio. 0 50 100 150 Cust Ret Whole Dist Fact Traditional 0 50 100 150 Cust Ret Whole Dist Fact Smoothing 0 2 4 6 8 10 Cust Ret Whole Dist Fact Information Sharing 0 2 4 6 8 10 Cust Ret Whole Dist Fact Smoothing & Information Sharing The impact of supply chain structures on performance Chapter 7 81 Figure 7.2. Order Rate Variance Ratio discrepancies between serial SCN and divergent SCN. Table 7.2. Numeric results (99% confidence intervals). Order Policy SCN structure Φ BwSl Retailer Wholesaler Distributor Factory Traditional OUT Serial 2.655±0.0126 7.732±0.1203 23.453±0.4962 69.539±1.7386 15.790±0.3942 Divergent 2.690±0.0119 8.923±0.1188 39.595±0.8211 136.196±2.8934 30.730±0.6540 ∆ Φ 0.035 1.191 16.142 66.657 14.94 OUT + Smoothing Serial 0.360±0.0015 0.957±0.0058 2.665±0.0221 6.803±0.0655 1.391±0.0150 Divergent 0.530±0.0021 2.190±0.0246 12.127±0.1899 32.168±0.5706 7.393±0.1324 ∆ Φ 0.17 1.233 9.462 25.365 6.002 OUT + Information Sharing Serial 2.120±0.0185 2.657±0.0234 3.093±0.0299 3.508±0.0317 0.599±0.0080 Divergent 2.219±0.0216 4.488±0.0399 7.970±0.0927 9.793±0.1178 2.334±0.0311 ∆ Φ 0.099 1.831 4.877 6.285 1.735 OUT + Smoothing + Information Sharing Serial 0.354±0.0017 0.474±0.0025 0.560±0.0026 0.599±0.0025 -0.060±0.0007 Divergent 0.528±0.0019 1.116±0.0069 2.756±0.0177 3.236±0.0192 0.670±0.0054 ∆ Φ 0.174 0.642 2.196 2.637 0.73 0 10 20 30 40 50 60 70 Ret Whole Dist Fact Traditional Smooth Info Smooth&Info Roberto Domínguez Chapter 7 82 7.4.1. Traditional SCN with classical OUT policy The traditional scenario shows the classical exponential trend of the BWE for the serial SCN, obtaining high values of both 𝛷 and BwSl. The result is in line with several studies dealing with the magnitude of BWE in a traditional SCN under the classical OUT policy (Disney and Lambrecth, 2008). Analogously, the divergent SCN shows the same exponential trend, but with higher values of 𝛷 and BwSl. By analyzing the discrepancies in order variance ratio between the serial SCN and the divergent SCN, an important differentiation between both SCNs is observed, being ∆𝛷=16.142 at the distributor stage and ∆𝛷=66.657 at the factory stage. Finally, it can be appreciated how the discrepancy in the BWE propagation is equal to ∆𝐵𝑤𝑆𝑙=14.94. 7.4.2. Traditional SCN with smoothing replenishment rule The smoothing scenario considerably reduces 𝛷 and BwSl for the serial SCN. In the first stages (retailers and wholesalers) there is no BWE (𝛷≤1) and then, it start to smoothly increases (𝐵𝑤𝑆𝑙=1.391). As for the previous scenario, the benefits provided by the smoothing replenishment rule in term of BWE reduction are confirmed. Likewise, the divergent SCN also experiments a considerable reduction of 𝛷 and BwSl, but still presents a high value of the bullwhip slope (𝐵𝑤𝑆𝑙=7.393), and hence, it still shows high values of 𝛷 at the last stages. Notice that the high discrepancies between both SCNs observed in the previous scenario have been reduced by the use of this technique. 7.4.3. Information sharing SCN with classical OUT policy The reduction of 𝛷 and BwSl in the information sharing scenario is higher than in the smoothing scenario for both SCNs. As this technique uses customer demand in the calculation of orders, the first stage (retailers) shows similar values of 𝛷 to those of the traditional scenario for both SCNs. After this stage, 𝛷 starts to increase in a linear trend (not showing the exponential trend of the above scenarios), with a higher slope in the divergent SCN. The discrepancies between both SCNs have been considerably reduced in this scenario, being ∆𝛷=6.285 at the factory stage and ∆𝐵𝑤𝑆𝑙=1.735. The impact of supply chain structures on performance Chapter 7 83 7.4.4. Information sharing SCN with smoothing order rule Finally, the combination of the above techniques obtains the highest reduction of the BWE for both SCNs. At the retailer stage, similar values to those obtained in the smoothing scenario for the serial SCN are observed (information sharing does not work in this stage). After this stage, 𝛷 starts to increase approximately in a linear trend (like the information sharing scenario), but is deterred by the smoothing factor, obtaining very low values (𝛷≤1) in all stages. The divergent SCN presents the same behavior described for the serial SCN, but with higher bullwhip slope and hence, higher values of 𝛷. However, the discrepancy between both SCNs is very low, with ∆𝛷=2.637 at the factory stage and ∆𝐵𝑤𝑆𝑙=0.73. 7.5. FINDINGS AND MANAGERIAL IMPLICATIONS The results reveal several important features of the divergent SCN and of the BWE avoidance techniques addressed in this study. First of all, the output of the simulation confirms the efficacy of the information sharing and of the smoothing replenishment rule in terms of BWE reduction in the divergent SCN. Until now this efficacy had merely been demonstrated for serial SCN models. However, the most significant results provided by this study concern the differences in term of BWE magnitude between the serial SCN and the divergent SCN. In fact, the divergent SCN configuration always performs worse than the serial SCN. However, a reduction of this discrepancy can be noted for the scenarios characterized by the implementation of one or both of the BWE avoidance techniques. Furthermore, these techniques are not only able to reduce the BWE in both SCN structures, but are even able to increase the resilience and the robustness of the divergent SCN. However, there are some differences in the impact of the information sharing and of the smoothing replenishment rule. More specifically, by adopting only the smoothing replenishment rule a significant reduction of the BWE on both SCNs can be noted, but it is still high in the last stages of the divergent SCN. With this technique, the orders placed by each node are just reduced by the smoothing factor, but are still affected by the demand pattern of the downstream nodes. When the shock in demand occurs, leading to a multiple stock-out situation occurs, the high order amplification is reduced (smoothed), but not eliminated. Furthermore, the BwSl is high, so a divergent SCN with high number of stages would present high values of 𝛷. Thus, Roberto Domínguez Chapter 7 84 the smoothing technique does not work properly for long divergent SCNs under a shock demand. On the contrary, information sharing performs better than the smoothing, obtaining good values of the BWE for both SCNs. The benefit of this technique is twofold: 1) nodes can adapt faster to the violent changes in market demand, and 2) the high amplification of orders due the multiple stock-out problem commented above is stopped, because nodes use the customer demand order patterns to update the base stock level instead of the order pattern of their downstream partners. Combining the benefits of the information sharing and the smoothing together, the BWE in the divergent SCN almost disappears and its propagation is very low (near zero). From a managerial view point, a significant implication for the designing and management of SCNs has been precisely captured. In fact, till now, the unique proposed solution in scientific literature to reduce poor dynamics in divergent SCN has been the elimination of channel intermediaries (direct channel, “the Dell model”) (Disney and Lambrecth, 2008). The work of Sodhi and Tang (2011), one of the few papers that have reported some insights on the differences between a serial SCN and a no-serial SCN in terms of their dynamic behavior, reveals that a firm should consider simplifying the SCN structure by reducing the number of levels or by reducing the number of successors (i.e. transforming the current SCN structure into a serial structure) to mitigate the incremental BWE. In this work, it is shown how the discrepancies between the divergent SCN and the serial SCN can be considerably reduced by an appropriate implementation of the smoothing replenishment rule and/or the information sharing (see e.g. Figure 7.2). Thus, it can be argued that information sharing and smoothing replenishment rule not only limit the BWE, even SCN characterized by more than one node in the same layer, but also are able to increase the resilience and robustness of SCNs. By reducing this incremental BWE, the differences in operation performance between the traditional structure and the divergent structure are reduced (merging their dynamic behavior) and hence, increasing the robustness of the divergent SCN without modifying its structure (suppressing nodes). The above-mentioned result bring us to further concern about the efficient management of the SCNs. Nowadays we are not facing a temporary shock that will quickly pass, but in fact are on the verge of an “era of turbulence”, that will feature higher variance in key business parameters (Christopher and Holweg, 2011). Obviously, this context exposes SCNs to tremendous shocks and impetuous alterations of the The impact of supply chain structures on performance Chapter 7 85 market. Thus, the SCN crash test adopted in this work do not merely emulate the potential response of the real-world SCNs for an extreme and rare condition of the business environment. On the contrary, this response realistically represents the dynamic behavior of the real-world SCNs under the current and the advocated future business environment. In the light of the results, companies should pay more attention with respect to the past decades, when decide to reengineer and even design new SCNs. Consider the case of a company that operates with traditional control strategies and is yet able to perform well in the current market. If this company is willing to enhance their market by covering further geographical positions, probably should increase their distribution, wholesaler and retailer centers. Obviously this would amplify the complexity of the chain structure. As direct consequence, this company would risk to experiment a decrement of the whole operational performance. Thus, the potential benefit provided by the acquisition of new market share can be leveraged by a structurally decaying of the dynamic behavior. On the contrary companies adopting these BWE avoidance strategies, such as the external collaboration by information sharing strategies, pursuing the “new supply chain agenda” (see e.g. Stank et al., 2011), would reduce these risks and in any case would be more protected against the effect of the “era of turbulence” than the traditional SCN. 7.6. SUMMARY AND CONCLUSIONS The work presented in this chapter explores the impact of some well-known BWE avoidance strategies (i.e. the smoothing replenishment rule and the information sharing) when applying on different configurations of the SCN (i.e. a serial SCN and a divergent SCN). The analysis has been carried out using the shock lens proposed by Towill et al. (2007), which is a stress-test related to the robustness of the system. The results confirm that the BWE avoidance features of the strategies are also significant for the divergent SCN. Nevertheless, under these conditions, the divergent SCN performs worse than the serial SCN in all the scenarios. This bad behavior is caused by the higher complexity of the divergent SCN, which leads to a loose in robustness in relation to the serial SCN. However, the discrepancies in performance between both SCNs can be considerably reduced by the adoption of the two BWE avoidance strategies analyzed. Furthermore, it is shown how these strategies not only Roberto Domínguez Chapter 7 86 reduce the BWE in SCNs, but also increase the robustness of complex SCNs, such as the divergent SCN. The best results are offered by the combination of the smoothing replenishment rule with the information sharing. However, the discrepancies between both SCNs still persist, not being completely removed. This observation opens a new research line in developing new techniques which implicitly consider the inherent complexity of the divergent topology and attempt to totally erase the discrepancies with the serial SCN. These techniques would allow managing a divergent SCN with the same robustness than the classical serial SCN. Finally, it is worth mentioning that the results obtained in this chapter and the results obtained in Chapter 6 confirm the existence of differences between the dynamic behavior of the serial SCN and other SCN configurations (i.e. the divergent SCN). 87 Chapter 8: A systematic analysis of the structure of divergent SCNs and bullwhip 8.1. INTRODUCTION In previous chapters, the divergent SCN configuration has been benchmarked against the classical serial SCN. The results reveal different dynamics behaviors in terms of BWE. In particular, under a shock in the end customer demand, the divergent SCN performs worse than the serial SCN. Furthermore, even though the effectiveness of the information sharing and the smoothing replenishment rule on reducing the BWE in the divergent SCN has been proved, this configuration still perform worse than the serial SCN. These results lead to think that the structure of the SCN may impact on the BWE. In order to confirm this hypothesis, this chapter analyzes the potential relation between the structure of the SCN and the BWE. More specifically, this chapter presents: (1) an analysis of the divergent SCN configuration in search of its structural factors, (2) a structured full factorial design of experiments in which the configuration of the SCN is systematically varied through its different structural factors, remaining the rest of the parameters fixed, (3) a statistical analysis (ANOVA) and a discussion of the results obtained for two different demand perspectives (i.e. the variance lens and the shock lens, see Chapter 2). 8.2. THE DIVERGENT SCN CONFIGURATION In this section, the structural elements of a generic SCN are described, and then, the inherent characteristics/constrains of the divergent configuration are formalized. The SCN structure arises from the connected facilities that work together in order to supply products or services. In a SCN, each link represents the flow of materials and information that makes possible the functions of procurement, processing (or manufacturing), storage and distribution. For any given SCN, each functional level comprises an echelon, and there may be numerous facilities within each echelon (Beamon and Chen, 2001). This definition of the SCN structure is in accordance with the growing literature on complex networks, in which the SCN is modeled as a network by a set of “nodes” that represent autonomous business units (firms or facilities), and a Roberto Domínguez Chapter 8 88 set of “connections” (links) that link these firms together in demand-supply relationships for the purposes of creating products or services (Hearnshaw and Wilson, 2013; Gerschberger et al., 2012; Wen et al., 2012; Kim et al., 2011; Li et al., 2010a; Li et al., 2010b; Li et al., 2009; Choi et al., 2001). Hence, in line with this literature, the structure elements of a SCN are formalized as follows:  Echelons: the number of echelons is denoted by 𝑖∈(1,𝐸), with E the total number of echelons in the SCN. Echelons are numbered downstream from the suppliers, which are in echelon i = 1.  Nodes: a generic node j in echelon i is denoted by 𝑛𝑖𝑗. The number of nodes in a specific echelon i is 𝑁𝑖. The total number of nodes in the SCN is: ∑𝑁𝑖 𝐸𝑖=1 =𝑁.  Links: a link between nodes 𝑛𝑖𝑗 and 𝑛𝑖′𝑗′ is denoted by 𝑙(𝑛𝑖𝑗,𝑛𝑖′𝑗′) and the total number of links is L. There are two commonly used indicators to measure the degree of linkage in a SCN, namely the connection degree and the cluster coefficient (see e.g. Wen et al., 2012; Kim et al., 2011; Xuan et al., 2011; Li et al., 2010a; Barabási et al., 2002) . The connection degree 𝐷𝑖𝑗 is defined as the sum of a node’s links (Li et al., 2010a). The number of suppliers linked with node 𝑛𝑖𝑗 is the in-degree (𝑑𝑖𝑖𝑗), and the number of customers linked with a node 𝑛𝑖𝑗 is the out-degree (𝑑𝑜𝑖𝑗) (Kim et al., 2011; Xuan et al., 2011). The sum of the in-degree and the out-degree is the connection degree: 𝐷𝑖𝑗=𝑑𝑖𝑖𝑗+𝑑𝑜𝑖𝑗. The clustering coefficient C is the probability that two nearest neighbors of a node are also nearest neighbors of one another (Li et al., 2010a). Given node 𝑛𝑖𝑗 linked to 𝑘𝑖 other nodes in the system, if these 𝑘𝑖 nodes form a fully connected clique, there are 𝑘𝑖(𝑘𝑖−1)/2 links between them. Let us denote by 𝜆𝑖 the number of links that connect the selected 𝑘𝑖 nodes to each other. The clustering coefficient for node 𝑛𝑖𝑗 is then 2𝜆𝑖/𝑘𝑖(𝑘𝑖−1) (Barabási et al., 2002). The number of nodes, the number of echelons, and the structure of the material and information flows (links) has given rise to a structural classification scheme of SCNs based on the material relationship between nodes (Beamon and Chen, 2001). Up to now, most of the existent literature on the BWE topic has analyzed the classical serial SCN. In this SCN, the number of nodes in each echelon is limited to one (𝑁𝑖=1), and hence, the number of nodes and echelons in the SCN is the same (𝑁=𝐸). The connection degree is also limited: each node supplies to one node in the successor The impact of supply chain structures on performance Chapter 8 89 echelon (𝑑𝑜𝑖𝑗=1) and it is supplied by one node in the predecessor echelon (𝑑𝑖𝑖𝑗=1), thus limiting the total number of links to 𝐿=𝑁−1. Summing up, the structure of the serial SCN configuration is very restrictive: by selecting the quantity of one of the structural elements above mentioned (echelons, nodes or links), the SCN structure is defined, thus limiting the analysis of the influence of the SCN structure on the BWE to the number of echelons. The divergent SCN configuration is less restrictive than the serial configuration. The inherent structural restrictions of divergent SCNs are described and formalized next: 1. The number of nodes in each echelon is equal or greater than the number of nodes in its predecessor, i.e.: 𝑁𝑖≥𝑁𝑖−1. Furthermore, in order to exclude the serial SCN, the total number of nodes is constrained to 𝑁≥𝐸+1. 2. A node 𝑛𝑖𝑗 can supply to any number of nodes in the successor echelon (𝑑𝑜𝑖𝑗≥ 1), but can be supplied only by one node from the predecessor echelon (𝑑𝑖𝑖𝑗= 1) (Beamon and Chen, 2001). 3. Nodes in the same echelon are not linked. Hence, the network clustering coefficient C is zero. This is consistent with most cases in real-world SCNs (e.g. divergent SCN), that is, entities in the same echelons normally have no demandsupplier relations (Li et al., 2010a). This constraint, together with the previous restriction, limits the total number of links to the total number of nodes minus one: 𝐿=𝑁−1. By observing the above constrains, it can be noted that N is greater than E in divergent SCNs and thus, echelons are allowed to contain more than one node. Furthermore, for a given E, there is no upper bound for N. Thereby, any distribution of nodes across the SCN satisfying restriction (1) is allowed. In addition, nodes can supply to any number of nodes downstream, as indicated by restriction (2). Hence, there might be nodes with a high connection degree while others with low connection degree, resulting in SCNs with different degree distributions. In a first attempt to measure the impact of the SCN structure on the BWE, the connection degree is not considered as a factor, and for this reason, the divergent SCNs under analysis have homogeneous degree distributions: all nodes in the same echelon have similar connection degrees. Instead, this work focuses on the number of echelons, the number of nodes, and the distribution of links (or nodes) along the SCN. Roberto Domínguez Chapter 8 96 average (N/E)). In this situation, demand is also uniformly distributed among the different nodes, thus limiting the amplification effect. However, when the divergence of the SCN increases (∑𝑛𝑖𝑗𝑗 is far from N/E), there are one or more critical echelons in which the number of nodes abruptly increases and therefore, there a few nodes supplying a high number of nodes downstream in these echelons, as it can be seen in Figure 8.3-below. This situation increases the variability of orders received by these nodes and hence, increases the BWE. Figure 8.3. Increasing N (above) and divF (below) in a divergent SCN. Finally, there is one significant interaction between the number of echelons and the number of nodes, although it has a low impact on the overall BWE. A simple but powerful “interaction plot” is used to determine the severity of the interaction between The impact of supply chain structures on performance Chapter 8 97 these factors. Interaction plots are obtained by graphing the combined effects of the pairs of the factors studied. Due to the exponential nature of the obtained interaction curves, logarithms have been used to transform them into linear curves, in order to clarify its interpretation (see Figure 8.4). In fact, there is a small interaction between the number of echelons and the number of nodes, since the interaction curves are not parallel at all. More specifically, the NH curve has lower slope than the other curves. Hence, increasing the number of echelons in a SCN with a high number of nodes has a slightly lower impact than in SCNs with low or medium number of nodes. Another interpretation is that increasing the number of nodes results in a lower impact since the SCN has a higher number of echelons. This interpretation has been statistically contrasted by running a single-variable test (see Table 8.4). In this test, the effect of factor N is contrasted for each level of factor E. In fact, increasing the number of nodes has a significant impact on BWE no matters the number of echelons of the SCN (p<0.001). However, its impact decreases as the number of echelons increases (check partial 𝑅2 in Table 8.4). Table 8.4. Single-variable test for the interaction between E and N in Variance Lens scenario. Echelons F-ratio p-value 𝑹𝟐 (percent) Low 63.364 <0.001 4.5 Medium 31.747 <0.001 2.3 High 22.765 <0.001 1.7 Figure 8.4. Interaction between E and N in Variance Lens scenario. 7 8 9 10 11 12 13 EL EM EH Ln(PeakO1) NL NM NH Roberto Domínguez Chapter 8 98 8.4.2. Shock Lens ANOVA results are summarized in Table 8.5. When all factors are considered together, the model is statistically significant at the 95% confidence level with an overall 𝑅2 of 0.892, indicating that 89.2% of the variation in 𝑃𝑒𝑎𝑘𝑂1 can be explained by the structural factors considered. Furthermore, all factors are found to be statistically significant, as well as two of the interactions. Figure 8.5 show the main effects of the structural factors (E, N, D) by plotting the 𝑃𝑒𝑎𝑘𝑂1 averages for each level of the factors (Low, Medium, High). As in the previous analysis, all 𝑃𝑒𝑎𝑘𝑂1 values appear divided by 104 (1E4). Table 8.5. ANOVA results in Shock Lens scenario. Factors DOF F-ratio p-value 𝑹𝟐 (percent) Model 17 1305.427 <0.001 89.2 Echelons 2 10231.880 <0.001 88.4 Nodes 2 439.303 <0.001 24.7 Divergence 1 693.143 <0.001 20.5 Echelons * Nodes 4 1.877 0.112 0.3 Echelons * Divergence 2 65.821 <0.001 4.7 Nodes * Divergence 2 7.909 <0.001 0.6 In view of the main effects in Figure 8.5 and the data from Table 8.5, it is noticeable that the most significant factor on the BWE is the number of echelons: SCNs with higher number of echelons show higher BWE, following an exponential trend. The average shock in demand causes an unexpected multi stock-out at the retailer level. Nodes at this level react by placing higher orders than usual to the upstream nodes, which fall in a stock-out situation too. This effect is amplified from one echelon to another, increasing the fluctuation of orders through the SCN and causes the high 𝑃𝑒𝑎𝑘𝑂1 values observed in Figure 8.5. The number of nodes and the divergence of the SCN are both significant and have a relative high impact on the BWE, but such impact is lower than that of the number of echelons, as it can be deducted from partial 𝑅2 in Table 8.5. More specifically, the number of nodes is slightly more significant than the divergence of the SCN. SCNs with higher number of nodes show higher BWE, and SCNs with higher divergence show higher BWE, both factors following a linear trend. The impact of supply chain structures on performance Chapter 8 99 Figure 8.5. Main effects in Shock Lens scenario. There are two significant interactions in this scenario. The most important is the number of echelons and the divergence of the SCN. As in the variance lens scenario, interaction plots as well as single-variable tests are used to determine the interaction between factors. Once again, the natural logarithm is used to transform the exponential interaction curves into linear curves in order to clarify its interpretation. In Figure 8.6 it can be seen that the linearized interaction curves are not parallel, which means that an interaction occurs between both factors. The DivFH curve shows a higher slope than the DivFL curve. Therefore BWE is more sensitive to the number of echelons in SCNs with high divergence than in SCNs with low divergence (see partial 𝑅2 in Table 8.6). In addition, BWE is more sensitive to the divergence of SCNs with high number of echelons than SCNs with low number of echelons (see partial 𝑅2 in Table 8.6). These results are confirmed by the single-variable test in Table 8.6, where all contrast were statistically significant (p<0.001). Figure 8.6. Interaction between E and the DivF in Shock Lens scenario. 0 50 100 150 200 250 EL EM EH NL NM NH DivFL DivFH PeakO1/1E4 7 8 9 10 11 12 13 14 15 16 EL EM EH Ln(PeakO1) DivFL DivFH Roberto Domínguez Chapter 8 100 Table 8.6. Single-variable test for the interaction between E and DivF in Shock Lens scenario. A contrast of the effect of DivF for each level of E Echelons F-ratio p-value 𝑹𝟐 (percent) Low 41.010 <0.001 1.5 Medium 282.493 <0.001 9.5 High 501.283 <0.001 15.7 A contrast of the effect of E for each level of DivF Divergence F-ratio p-value 𝑹𝟐 (percent) Low 4343.363 <0.001 76.4 High 5954.338 <0.001 81.6 The other significant interaction occurs between the number of nodes and the divergence of the SCN, but it has a lower R2 than the previous interaction. BWE is more sensitive to the number of nodes in SCNs with a high divergence factor than in SCNs with a low divergence factor. Moreover, BWE is more sensitive to the divergence of SCNs with medium or high number of nodes than SCNs with low number of nodes. These results are supported by the single-variable test in Table 8.7, where all contrast were found to be significant (p<0.001). Table 8.7. Single-variable test for the interaction between N and DivF in Shock Lens scenario. A contrast of the effect of DivF for each level of N Nodes F-ratio p-value 𝑹𝟐 (percent) Low 142.922 <0.001 5.1 Medium 286.426 <0.001 9.6 High 279.612 <0.001 9.4 A contrast of the effect of N for each level of DivF Divergence F-ratio p-value 𝑹𝟐 (percent) Low 171.984 <0.001 11.4 High 275.228 <0.001 17.0 8.4.3. A comparison between the variance lens scenario and the shock lens scenario There are three important differences between the variance and the shock lens scenarios. First of all, the number of nodes and the divergence of the SCN have a higher impact on the BWE in the shock lens scenario than in the variance lens scenario (see 𝑅2 in Tables 8.3 and 8.5). This fact is confirmed by comparing the main effects of the The impact of supply chain structures on performance Chapter 8 101 number of nodes and the divergence of the SCN in both scenarios (see Figure 8.7), obtaining that 𝑃𝑒𝑎𝑘𝑂1 curves show higher slopes in the shock lens scenario than in the variance lens scenario and hence, BWE is more sensitive to these factors in the former scenario than in the latter. The higher number of nodes per echelon and/or the presence of critical echelons (SCNs with high divergence) make the SCN more vulnerable to an unexpected shock in demand and the consequent multi stock-out situation. Figure 8.7. A comparison of the main effects of E, N and DivF between Variance Lens and Shock Lens scenarios. A second important difference between both scenarios is that the BWE is higher in the shock lens scenario in all cases with a 95% confidence level (see Table 8.8 and Figure 8.7). Furthermore, since the shock lens scenario presents higher values of 𝑃𝑒𝑎𝑘𝑂1 and higher slopes than the variance lens scenario, the discrepancies in terms of BWE between both scenarios increase as the levels of the three structural factors become higher. In order to quantify these discrepancies a measure of the relative increment of the average BWE is employed in the shock lens scenario over the average BWE in the variance lens scenario (see equation 8.3). -40 10 60 110 160 210 260 EL EM EH PeakO1/1E4 Shock Lens Variance Lens 0 5 10 15 20 NL NM NH DivFL DivFH PeakO1/1E4 Shock Lens Variance Lens Roberto Domínguez Chapter 8 102 ∆=(𝑃𝑒𝑎𝑘𝑂1𝑠ℎ𝑜𝑐𝑘𝐿𝑒𝑛𝑠−𝑃𝑒𝑎𝑘𝑂1𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒𝐿𝑒𝑛𝑠) 𝑃𝑒𝑎𝑘𝑂1𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒𝐿𝑒𝑛𝑠 ∗100 (8.3) By plotting ∆ for each level of the structural factors in Figure 8.8, in fact, it can be observed how the discrepancies between both scenarios show an increasing linear trend for each factor. Focusing on this figure, it can be noted that the curve for the number of echelons show the highest slope. Thereby, the number of echelons has the highest impact on ∆. Meanwhile, the number of nodes and the divergence of the SCN have similar slopes (the slope of the former slightly higher than the slope of the latter), thus having similar impacts on ∆. Table 8.8. Average 𝑷𝒆𝒂𝒌𝑶𝟏 and 95% confidence intervals from ANOVA. Factor Level Lens Average 𝑷𝒆𝒂𝒌𝑶𝟏/𝟏𝑬𝟒 95% confidence Lower Bound 95% confidence Upper Bound Echelons Low Variance 0.155 0.147 0.163 Shock 0.313 0.293 0.334 Medium Variance 1.929 1.833 2.030 Shock 8.009 7.504 8.548 High Variance 33.591 31.925 35.344 Shock 259.515 243.143 276.990 Nodes Low Variance 1.612 1.532 1.696 Shock 4.224 3.958 4.509 Medium Variance 2.237 2.126 2.353 Shock 9.058 8.487 9.668 High Variance 2.778 2.640 2.922 Shock 16.976 15.905 18.119 Divergence Factor Low Variance 1.804 1.730 1.880 Shock 5.226 4.955 5.511 High Variance 2.576 2.471 2.685 Shock 14.352 13.609 15.137 The impact of supply chain structures on performance Chapter 8 103 Figure 8.8. BWE discrepancies between Variance Lens and Shock Lens scenarios. Finally, the third important difference between both scenarios refers to the interactions between the structural factors: while there is a significant interaction between the number of echelons and the number of nodes in the variance lens scenario, in the shock lens scenario the significant interactions take place between the divergence with the number of nodes and the number of echelons. A summary of findings is shown in Table 8.9. 8.5. SUMMARY AND CONCLUSIONS The structural design of the SCN, defined by the number of echelons, the number of nodes and the divergence of the SCN, has been analyzed in terms of BWE. A collection of divergent SCNs with random structures according to different levels of their structural factors have been modeled and simulated, and output data has been statistically analyzed. Two independent scenarios with different demand patterns have been considered: the former is characterized by a stationary and normally distributed demand input, while the latter is characterized by a normally distributed demand input which suffers, at a given time, a violent increment in mean. It has been shown that, in fact, the structural design of a SCN is statistically significant and influences the BWE in both scenarios: increasing the number of echelons, nodes, or the divergence of the SCN, will increase BWE. Furthermore, BWE is always higher in the case of an impulse in the end-customer demand. Additionally, BWE is more sensitive to the structural design of the SCN in this scenario than in the scenario with stationary demand. Hence, as SCNs size increases in terms of number of echelons or number of nodes, or increases its divergence, they also become more vulnerable to unexpected violent changes in demand 0 100 200 300 400 500 600 700 Low Medium High Δ(%) E N DivF Roberto Domínguez Chapter 8 104 mean. In other words, as SCNs become more complex, they fall in a more vulnerable situation under uncertainties in market demand. There are some managerial implications that can be derived from this work. The effect of the number of echelons (horizontal complexity) on the BWE has been widely analyzed in literature, mostly in serial SCNs. This study, through a statistical analysis of different structural designs, shows that the amplification of the variability of orders caused by the number of echelons persists in divergent SCNs. Hence, solutions proposed by other authors, like the elimination of channel intermediaries (Disney and Lambrecth, 2008), are also applicable. Factor N is related to the average number of nodes (or entities) within each echelon, also known as vertical complexity. Since the analysis focuses on divergent SCNs, the level of N is directly related to the number of retailers (e.g. a high value of N also means a high number of retailers). The present study shows that this factor has a direct impact on BWE. Therefore, managers and designers should pay special attention in optimizing the geographical distribution of entities in each echelon to avoid unnecessary stock points and retailers, thus limiting the factor N and reducing the BWE. The last structural factor analyzed, DivF, describes the distribution of nodes along the echelons of the SCN. The present study shows that this factor has a direct impact on BWE. Hence, managers and designers should try to smoothly increase the number of entities downstream to avoid the presence of critical echelons, in which a few entities have to deal with the supply of a high number of other entities in the subsequent echelon which, in fact, increases the BWE. All the above implications are more critical for SCNs in a shock demand situation, i.e., SCNs facing unpredictable violent changes in demand mean. In this case, BWE is much more sensitive to the structure of the SCN than in the case of a stationary demand. Due to the important economic lost caused by the BWE, in this situation it is necessary to reorganize the structure of SCN. Other options are a good analysis of the market tendencies to anticipate these violent changes and the use of certain techniques to smooth the BWE, like information sharing or smoothing replenishment rules. 105 Table 8.9. A summary of findings. Variance Lens Shock Lens Variance Lens Vs Shock Lens Factors All the studied factors impact on performance with different magnitude. In particular E reveals the highest impact. As the levels of the factors increase, the BWE always increase but with a different magnitude. More specifically, BWE exponentially increases as the structure shifts from an EL configuration to an EH one, and linearly increases as the structure shifts from a NL to NH and from DivFL to DivFH, respectively. All the studied factors impact on performance with different magnitude. In particular E reveals the highest impact. As the levels of the factors increase, the BWE always increase but with a different magnitude. More specifically, BWE exponentially increases as the structures shifts from an EL configuration to an EH one, and linearly increases as the structure shifts from a NL to NH and from DivFL to DivFH, respectively. In both scenarios all the factors are statistically significant. Even though, the impact of factors noticeable differs: N and DivF have a higher impact in the shock lens scenario. BWE is always higher in the shock lens scenario and main effect curves show higher slopes, thus showing a higher sensitivity of BWE to variations of the structural factors. As the levels of the factors increase, the discrepancies between both scenarios increase following a linear trend. These discrepancies are especially sensitive to E. Interactions Only the interaction E*N reveals a significant impact. As E increases, BWE is characterised by a lower sensitivity to N. Furthermore, as N increases, BWE is characterised by a lower sensitivity to E. The interactions E*DivF and N*DivF are statistically significant. As E increases, BWE is characterised by a higher sensitivity to DivF. Furthermore, as DivF increases, BWE is characterised by a higher sensitivity to E. As DivF increases, BWE is characterised by a higher sensitivity to N. However, as N increases, BWE is characterised by an almost unaltered sensitivity to DivF. Statistically significant interactions completely differs between both scenarios: Variance Lens: E*N Shock Lens: E*DivF and N*DivF Roberto Domínguez Chapter 9 112  Domínguez, R., Framinan, J.M. 2013. A decision management tool: Modelling the order fulfilment process by multi-agent systems. International Journal of Management and Decision Making, 12 (3), 240-258. (in Scopus). 9.3.2. Conferences  Domínguez R., Cannella S., Bruccoleri M., Framinan J.M. Arborescent Supply Chain Dynamics: the impact of OUT policies. Proceedings of the 20th European Operations Management Association Conference, Dublin, June 10-12, 2013.  Domínguez R., Framinan, J.M., Cannella, S., Póvoa, A.P. Using multi-agent systems to explore information sharing in arborescent supply chain networks. Proceedings of the 5th International Conference on Industrial Engineering and Systems Management, Rabat, October 28-30, 2013.  Cañizares, R.D., Framiñán, J.M. 2012. Development of a multi-agent platform for supply chain-wide order fulfillment. Proceedings of the 4th International Conference on Agents and Artificial Intelligence, 2, 203-208.  Domínguez, R., Framinan, J.M., Dios, M.A., León, J.M. Revisión del estado del arte en modelado del order fulfillment process mediante sistemas multi-agente. XIV Congreso de Ingeniería de Organización, September 8-10, 2010. 9.4. FURTHER RESEARCH LINES The limitations of the present work also represent opportunities for further research. First, in order to simplify the analysis, this work has been focused on divergent SCNs. Once the structure of the divergent SCN has been analyzed and its impact on the BWE has been determined, this study can be extended to other SCN configurations, like convergent or conjoined SCNs. Secondly, since the focus of the analysis was the structure of the SCN, other operational factors (forecast method, inventory policy, lead time, order batching, etc.) have been maintained fixed. Hence, a further research in which common operational factors are variables is needed in order to determine their relative importance in such SCNs as well as some possible interactions with their structural factors. The impact of supply chain structures on performance Chapter 9 There is also some research implications derived from this work. As it was reported in Chapter 2, SCNs are complex systems and thus, the analysis performed on this field should turn on considering complex SCNs in further studies, in order to get closer to the dynamics of real SCNs. The present work is a step through modeling SCNs as complex systems. According to Reiß (1993), four dimensions of complexity exist: 1. Multiplicity, which leads to the variety of a system. 2. Variance, resulting in the heterogeneity of the system. 3. Changeability, determining the dynamic behavior of the system. 4. Ambiguity, leading to uncertainty. In the current work, one of these drivers of complexity is addressed: the multiplicity. As compared with the traditional serial SCN, the divergent SCNs analyzed here have a higher number of elements and interdependence. In order to better understand complex SCNs, future research should consider exploring all the dimensions of complexity identified by Reiß (1993). In other words, future research should focus on modeling SCNs with high number of elements and interdependence (multiplicity), including diversity of elements, i.e., elements are different between them (variance, heterogeneity), and the ability of elements to change their status over time (changeability, chaos), as well as ambiguity and uncertainty. 115 References Abid, C., D'Amours, S., Montreuil, B. 2004. Collaborative order management in distributed manufacturing. International Journal of Production Research, 42 (2), 283-302. Adenso-Díaz, B., Moreno, P., Gutiérrez, E., Lozano, S. 2012. An analysis of the main factors affecting bullwhip in reverse supply chains. International Journal of Production Economics, 135 (2), 917-928. Aggelogiannaki, E., Doganis, P., Sarimveis, H. 2008. An adaptive model predictive control configuration for production-inventory systems. International Journal of Production Economics, 114 (1), 165-178. Agrawal, S., Sengupta, R.N., Shanker, K. 2009. Impact of information sharing and lead time on bullwhip effect and on-hand inventory. European Journal of Operational Research, 192 (2), 576-593. Ahn, H., Park, S. 2003. Modeling of a multi-agent system for coordination of supply chains with complexity and uncertainty. Lecture Notes in Artificial Intelligence (Subseries of Lecture Notes in Computer Science), 2891, 13-24. Akkermans, H., Voss, C. 2013. The service bullwhip effect. International Journal of Operations and Production Management 33 (6), 765-788. Ali, M.M., Boylan, J.E. 2011. On the effect of non-optimal forecasting methods on supply chain downstream demand. IMA Journal Management Mathematics, 23 (1), 81-98. Ali, M.M., Boylan, J.E., Syntetos, A.A. 2012. Forecast errors and inventory performance under forecast information sharing. International Journal of Forecasting, 28 (4), 830-841. Alony, I., Munoz, A. 2007. The Bullwhip effect in complex supply chains. International Symposium on Communications and Information Technologies Proceedings, 4392227, 1355-1360. Alpay, S. 2007. Agent Based Dynamic Job Shop Simulation System. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 4570, 364-373. Alpay, S., Sançiçek, I. 2008. Agent-based shop-floor control system in a dynamic environment. Proceedings - International Conference on Information Technology: New Generations, 4492515, 414- 419. Álvarez, E., De la Calle, A. 2009. Information Exchange and Synchronized Scheduling in the Supply Chain. 3rd International Conference on Industrial Engineering and Industrial Management, 43-49. Audy, J.-F., Lehoux, N., D'Amours, S., Rönnqvist, M. 2012. A framework for an efficient implementation of logistics collaborations. International Transactions in Operational Research, 19 (5), 633-657. Roberto Domínguez References 116 Azevedo, A., Toscano, C., Sousa, J., Soares, A. 2004. An advanced agent-based order planning system for dynamic networked enterprises. Production Planning and Control, 15(2), 133-144. Babai, M.Z., Ali, M.M., Boylan, J.E., Syntetos, A.A. 2013. Forecasting and inventory performance in a two-stage supply chain with ARIMA (0,1,1) demand: Theory and empirical analysis. International Journal of Production Economics, 143 (2), 463-471. Balan, S., Vratb, P., Kumarc, P. 2009. Information Distortion in a Supply Chain and Its Mitigation Using Soft Computing Approach. The International Journal of Management Science, 37 (2), 282–299. Barabási, A.L., Jeong, H., Néda, Z., Ravasz, E., Schubert, A., Vicsek, T. 2002. Evolution of the social network of scientific collaborations. Physica A: Statistical Mechanics and its Applications, 311 (3-4), 590-614. Barlas, Y., Gunduz, B. 2011. Demand forecasting and sharing strategies to reduce fluctuations and the bullwhip effect in supply chains. Journal of the Operational Research Society, 62 (3), 458-473. Bayraktar, E., Koh, L. S. C., Gunasekaran, A., Sari, K., Tatoglu, E. 2008. The Role of Forecasting on Bullwhip Effect for E-SCM Applications. International Journal of Production Economics, 113 (1), 193–204. Beamon, B.M., Chen, V.C.P. 2001. Performance analysis of conjoined supply chains. International Journal of Production Research, 39 (14), 3195-3218. Ben-Tal, A., Golany, B., Shtern, S. 2009. Robust Multi-Echelon Multi-Period Inventory Control. European Journal of Operational Research, 199 (3), 922–935. Bhattacharya, R., Bandyopadhyay, S. 2011. A review of the causes of bullwhip effect in a supply chain. International Journal of Advanced Manufacturing Technology, 54 (9-12), 1245-1261. Bo, X., Zhiming, W. 2003. Modeling of supply chain: A multi-agent approach. Proceedings of the American Control Conference, 2, 1056-1061. Bonabeau, E. 2002. Predicting the unpredictable. Harvard Business Review, 80(3), 109-116. Bottani, E., Montanari, R., Volpi, A. 2010. The impact of RFID and EPC network on the bullwhip effect in the Italian FMCG supply chain. International Journal of Production Economics, 124 (2), 426-432. Boute, R.N., Disney, S.M., Lambrecht, M.R., Van Houdt, B. 2007. An Integrated production and inventory model to dampen upstream demand variability in the supply chain. European Journal of Operational Research, 178 (1), 121-142. Boute, R., Disney, S.M., Lambrecht, M. and Van Houdt, B. 2009. Designing replenishment rules in a two-echelon supply chain with a flexible or an inflexible capacity strategy. International Journal of Production Economics, 119 (1), 187 -198. Bray, R.L., Mendelson, H. 2012. Information transmission and the bullwhip effect: An empirical investigation. Management Science, 58 (5), 860-875. The impact of supply chain structures on performance References Byrne, P.J., Heavey, C. 2006. The impact of information sharing and forecasting in capacitated industrial supply chains: A case study. International Journal of Production Economics, 103 (1), 420-437. Bozarth, C.C., Warsing, D.P., Flynn, B.B., Flynn, E.J. 2009. The impact of supply chain complexity on manufacturing plant performance. Journal of Operations Management, 27 (1), 78-93. Bray, R.L., Mendelson, H. 2012. Information transmission and the bullwhip effect: An empirical investigation. Management Science, 58 (5), 860-875. Butner, K. (2010). The smarter supply chain of the future. Strategy and Leadership, 38 (1), 22-31. Byrne, P.J., Heavey, C. 2006. The impact of information sharing and forecasting in capacitated industrial supply chains: A case study. International Journal of Production Economics, 103 (1), 420-437. Cachon, G., Fisher, M. 2000. Supply chain inventory management and the value of shared information. Management Science, 46 (8), 1032-1048. Cachon, G. P., Randall, T., Schmidt, G. M. 2007. In Search of the Bullwhip Effect. Manufacturing & Service Operations Management, 9 (4), 457–479. Caloiero, G., Strozzi, F., Zaldívar Comenges, J.M. 2008. A supply chain as a series of filters or amplifiers of the bullwhip effect. International Journal of Production Economics, 114 (2), 631-645. Campuzano, F., Guillamon, A., Ros, L. 2009. The Influence of Lead Time Variability on Supply Chain Costs: Analysis of Its Impact on the Bullwhip Effect. The IUP Journal of Supply Chain Management, 6 (3), 15-26. Cannella, S., Ciancimino, E., Marquez, A.C. 2008. Capacity constrained supply chains: A simulation study. International Journal of Simulation and Process Modelling, 4 (2), 139-147. Cannella, S., Ciancimino, E. 2010. On the bullwhip avoidance phase: Supply chain collaboration and order smoothing. International Journal of Production Research, 48 (22), 6739-6776. Cannella, S., Ciancimino, E., Framinan, J.M. 2011. Inventory policies and information sharing in multiechelon supply chains. Production Planning and Control, 22 (7), 649-659. Cannella, S., Barbosa-Povoa, A.P., Framinan J.M., Relvas S. 2013. Metrics for bullwhip effect analysis. Journal of the Operational Research Society, 64 (1), 1-16. Cantor, D.E., Katok, E. 2012. Production smoothing in a serial supply chain: A laboratory investigation. Transportation Research Part E: Logistics and Transportation Review, 48 (4), 781-794. Caloiero, G., Strozzi, F., Zaldívar Comenges, J.M. 2008. A supply chain as a series of filters or amplifiers of the bullwhip effect. International Journal of Production Economics, 114 (2), 631-645. Caridi, M., Cigolini, R., De Marco, D. 2005. Improving supply-chain collaboration by linking intelligent agents to CPFR. International Journal of Production Research, 43 (20), 4191-4218. Caridi, M., Cigolini, R., De Marco, D. 2006. Linking autonomous agents to CPFR to improve SCM. Journal of Enterprise Information Management, 19 (5), 465-482. Roberto Domínguez References 118 Cattani, K.D., Jacobs, F.R., Schoenfelder, J. 2011. Common inventory modeling assumptions that fall short: Arborescent networks, Poisson demand, and single-echelon approximations. Journal of Operations Management, 29 (5), 488-499. Chaharsooghi, S.K., Heydari, J. 2010. LT variance or LT mean reduction in supply chain management: Which one has a higher impact on SC performance? International Journal of Production Economics, 124 (2), 475-481. Chan, F.T.S, Prakash, A. 2012. Inventory management in a lateral collaborative manufacturing supply chain: a simulation study. International Journal of Production Research, 50 (16), 4670-4685. Chandra, C., Grabis, J. 2005. Application of multi-steps forecasting for restraining the bullwhip effect and improving inventory performance under autoregressive demand. European Journal of Operational Research, 166 (2), 337-350. Chatfield, D. Harrison, T. Hayya, J. 2001. SISCO: A supply chain simulation tool utilizing Silk™ and XML. Winter Simulation Conference Proceedings, 1, 614-622. Chatfield, D., Kim, J., Harrison, T., Hayya, J. 2004. The Bullwhip Effect—Impact of Stochastic Lead Time, Information Quality, and Information Sharing: A Simulation Study. Production and Operations Management, 13 (4), 340-353. Chatfield, D., Harrison, T., Hayya, J. 2006. SISCO: An object-oriented supply chain simulation system. Decision Support Systems, 42 (1), 422-434. Chatfield, D., Hayya, J., Harrison, T. 2007. A multi-formalism architecture for agent-based, order-centric supply chain simulation. Simulation Modelling Practice and Theory, 15 (2), 153-174. Chatfield, D., Harrison, T., Hayya, J. 2009. SCML: An information framework to support supply chain modeling. European Journal of Operational Research, 196 (2), 651-660. Chatfield, D.C., Hayya, J.C., Cook, D.P. 2013. Stockout propagation and amplification in supply chain inventory systems. International Journal of Production Research, 51 (5), 1491-1507. Chatfield, D.C. 2013. Underestimating the bullwhip effect: a simulation study of the decomposability assumption. International Journal of Production Research, 51 (1), 230-254. Chatfield D.C., Pritchard A.M. 2013. Returns and the bullwhip effect. Transportation Research Part E: Logistics and Transportation Review, 49 (1), 159–175. Cheeseman, M., Swann, P., Hesketh, G., Barnes, S. 2005. Adaptive manufacturing scheduling: A flexible and configurable agent-based prototype. Production Planning and Control, 16 (5), 479-487. Chen, F., Drezner, Z., Ryan, J., Simchi-Levi, D. 2000. Quantifying the bullwhip effect in a simple supply chain: the impact of forecasting, lead times, and information. Management Science, 46 (3), 436-443. Chen, Y.F., Disney, S.M. 2007. The myopic Order-Up-To policy with a proportional feedback controller. International Journal of Production Research, 45 (2), 351-368. The impact of supply chain structures on performance References Chen, L., Lee, H.L. 2009. Information Sharing and Order Variability Control under a Generalized Demand Model. Management Science, 55 (5), 781-798. Chen, X., Hao, G., Li, X., Yiu, K.F.C., 2012. The impact of demand variability and transshipment on vendor's distribution policies under vendor managed inventory strategy. International Journal of Production Economics, 139 (1), 42-48. Chen, H.-H. 2012. Simulating analysis of complex supply chain networks invulnerability. Lecture Notes in Electrical Engineering, 154, 1229-1235. Cheng, L.C. 2009. Impact of Inventory Policy Consistency on the Three Stage Supply Chain Performance. Journal of Academy of Business and Economics, 9 (4), 33–53. Childerhouse, P., Disney, S. M., Towill, D. R. 2008. On the Impact of Order Volatility in the European Automotive Sector. International Journal of Production Economics, 114 (1), 2–13. Cho, D.W., Lee, Y.H. 2013. The value of information sharing in a supply chain with a seasonal demand process. Computers and Industrial Engineering, 65 (1), 97-108. Choi, T.Y., Dooley, K.J., Rungtusanatham, M. 2001. Supply networks and complex adaptive systems: Control versus emergence. Journal of Operations Management, 19 (3), 351-366. Choi, T.Y., Hong, Y. 2002. Unveiling the structure of supply networks: case studies in Honda, Acura, and Dailmer Chrysler, Journal of Operations Management, 20 (5), 469-94. Christopher, M., Holweg, M. 2011. "Supply Chain 2.0": Managing supply chains in the era of turbulence. International Journal of Physical Distribution and Logistics Management, 41 (1), 63-82. Ciancimino, E., Cannella, S., Bruccoleri, M., Framinan, J.M. 2012. On the Bullwhip Avoidance Phase: The Synchronised Supply Chain. European Journal of Operational Research, 221 (1), 49-63. Coyle, R.G. 1977. Management System Dynamics. New York: John Wiley & Sons. Cox, A., Sanderson, J., Watson, G. 2006. Supply chains and power regimes: toward an analytic framework for managing extended networks of buyer and supplier relationships. Journal of Supply Chain Management, 37 (2), 28-35. Crespo Márquez, A. 2010. Dynamic Modelling for Supply Chain Management: Dealing with Front-end, Back-end and Integration Issues. London: Springer. Croson, R., Donohue, K. 2005. Upstream versus Downstream Information and Its Impact on the Bullwhip Effect. System Dynamics Review, 21 (3), 249–260. Croson, R., Donohue, K. 2006. Behavioral causes of the bullwhip effect and the observed value of inventory information. Management Science, 52 (3), 323-336. Dass, M., Fox, G.L. 2011. A holistic network model for supply chain analysis. International Journal of Production Economics, 131 (2), 587-594. Roberto Domínguez References 120 Dejonckheere, J., Disney, S.M., Lambrecht, M.R., Towill, D.R. 2002. Transfer function analysis of forecasting induced bullwhip in supply chains. International Journal of Production Economics, 78 (2), 133-144. Dejonckheere, J., Disney, S., Lambrecht, M, Towill, D. 2003a. The impact of information enrichment on the Bullwhip effect in supply chains: A control engineering perspective. European Journal of Operational Research, 153, 727-750. Dejonckheere, J., Disney, S.M., Lambrecht, M.R., Towill, D.R. 2003b. Measuring and avoiding the bullwhip effect: A control theoretic approach. European Journal of Operational Research, 147 (3), 567-590. Disney, S.M., Towill, D.R. 2003a. On the bullwhip and inventory variance produced by an ordering policy. Omega, the International Journal of Management Science, 31 (3), 157-167. Disney, S.M., Towill, D.R. 2003b. The effect of vendor managed inventory (VMI) dynamics on the Bullwhip Effect in supply chains. International Journal of Production Economics, 85 (2), 199-215. Disney, S.M., Naim, M.M., Potter, A. 2004a. Assessing the impact of e-business on supply chain dynamics. International Journal of Production Economics, 89 (2), 109-118. Disney, S.M., Towill, D.R., Van De Velde, W. 2004b. Variance amplification and the golden ratio in production and inventory control. International Journal of Production Economics, 90 (3), 295-309. Disney, S.M., Farasyn, I., Lambrecht, M., Towill, D.R., de Velde, W.V. 2006. Taming the bullwhip effect whilst watching customer service in a single supply chain echelon. European Journal of Operational Research, 173 (1), 151-172. Disney, S.M, and Towill, D.R. 2006. A methodology for benchmarking replenishment-induced bullwhip. Supply Chain Management, 11 (2), 160-168. Disney, S.M., Farasyn, I., Lambrecht, M.R., Towill, D.R. and Van De Velde, W. 2007. Controlling bullwhip and inventory variability with the golden smoothing rule. European Journal of Industrial Engineering, 1 (3), 241-265. Disney, S.M., Lambrecht, M., Towill, D.R., Van de Velde, W. 2008. The value of coordination in a twoechelon supply chain. IIE Transaction, 40 (3), 341-355. Disney, S.M., Lambrecht, M., 2008. On Replenishment Rules, Forecasting, and the Bullwhip Effect in Supply Chains. Foundations and Trends in Technology, Information and Operations Management, 2 (1), 1-80. Domínguez, R., Cannella, S., Bruccoleri, M., Framinan, J.M. 2013. Arborescent supply chain dynamics: the impact of OUT policies. Proceedings of the 20th EurOMA Conference. Dong, S.-H., Xi, B., Tian, L.-N., Huang, Q.-G., Chen, H.-X. 2006. An agent-based architecture for supply chain management. Proceedings of the 2006 International Conference on Machine Learning and Cybernetics, 4028046, 137-141. The impact of supply chain structures on performance References Dooley, K.J., Yan T., Mohan S., Gopalakrishnan, M. 2010. Inventory management and the bullwhip effect during the 2007-2009 recession: Evidence from the manufacturing sector. Journal of Supply Chain Management, 46 (1), 12-18. Dvorak, P. 2009. Clarity Is Missing Link in Supply Chain. The Wall Street Journal. Forget, P., D'Amours, S., Frayret, J.-M. 2008. Multi-behavior agent model for planning in supply chains: An application to the lumber industry. Robotics and Computer-Integrated Manufacturing, 24 (5), 664- 679. Forget, P., D’Amours, S., Frayret, J.-M., Gaudreault, J. 2009. Study of the performance of multibehaviour agents for supply chain planning. Computers in Industry, 60, 698-708. Framinan, J.M. 2009. Managing resources for order promising in Available-To-Promise (ATP) systems: A simulation study. International Conference on Industrial Engineering and Systems Management. Framinan, J.M., Leisten, R. 2009. Available-to-promise (ATP) systems: a classification and framework for analysis. International Journal of Production Research, 48 (11), 3079-3103. Fung, R., Chen, T. 2005. A multiagent supply chain planning and coordination architecture. International Journal of Advanced Manufacturing Technology, 25 (7-8), 811-819. Ganeshan, Ram. 1999. Managing supply chain inventories: A multiple retailer, one warehouse, multiple supplier model. International Journal of Production Economics, 59 (1), pp. 341-354. Garcia Salcedo, C.A., Ibeas Hernandez, A., Vilanova, R., Herrera Cuartas, J. 2013. Inventory control of supply chains: Mitigating the bullwhip effect by centralized and decentralized Internal Model Control approaches. European Journal of Operational Research, 224 (2), 262-272. Geary, S., Disney, S. M., and Towill, D. R. 2006. On Bullwhip in Supply Chains—Historical Review, Present Practice and Expected Future Impact. International Journal of Production Economics, 101, 2- 18. Gerschberger, M., Engelhardt-Nowitzki, C., Kummer, S., Staberhofer, F. 2012. A model to determine complexity in supply networks. Journal of Manufacturing Technology Management, 23 (8), 1015- 1037. Giard, V., Sali, M. 2013. The bullwhip effect in supply chains: A study of contingent and incomplete literature. International Journal of Production Research, 51 (13), 3880-3893. Goh, W., Gan, J. 2005. A dynamic multi-agent based framework for global supply chain. Proceedings of International Conference on Services Systems and Services Management, 2, 981-984. Gonçalves, P., Hines, J., Sterman, J. 2005. The impact of endogenous demand on push-pull production system. System Dynamics Review, 21 (3), 187-216. Govindu, R., Chinnam, R.B. 2007. MASCF: A generic process-centered methodological framework for analysis and design of multi-agent supply chain systems. Computers and Industrial Engineering, 53 (4), 584-609. Roberto Domínguez References 128 Sterman, J. 1989. Modeling managerial behavior: misperceptions of feedback in a dynamic decision making experiment. Management Science, 35 (3), 321-339. Strader, T., Lin, F., Shaw, M. 1998. Simulation of Order Fulfilment in Divergent Assembly Supply Chains. Journal of Artificial Societies and Social Simulation, 1 (2). Strozzi, F., Noè, C., Zaldívar, J.-M., 2012. Divergence control of a one-level supply chain replenishment rule. International Journal of Production Research, 50 (7), 2046-2057. Su, Y., Geunes, J. 2012. Price promotions, operations cost, and profit in a two-stage supply chain. Omega, the International Journal of Management Science, 40 (6), 891-905. Sucky, E. 2009. The Bullwhip Effect in Supply Chains – An Overestimated Problem?. International Journal of Production Economics, 118 (1), 311–322. Sun, H., Wu, J. 2005. Scale-free characteristics of supply chain distribution systems. Modern Physics Letters B, 19 (17), 841–848. Supply Chain Council. 2006. Supply-Chain Operations Reference-model—Overview of SCOR Version 8.0, Washington. Available from: <http://www.supply-chain.org>. Surana, A., Kumara, S., Greaves, M., Raghavan, U.N. 2005. Supply-chain networks: a complex adaptive systems perspective. International Journal of Production Research, 43 (20), 4235–4265. Swaminathan, J.M.; Smith, S.F.; Sadeh, N.M. 1998. Modeling supply chain dynamics: A multiagent approach. Decision Sciences, 29 (3), 607-631. Swarm Development Group Wiki. 2003. Centre for the Study of Complex Systems, University of Michigan. http://www.swarm.org [last access November 2011]. Syntetos, A.A., Boylan, J.E., Disney, S.M. 2009. Forecasting for inventory planning: a 50-year review. Journal of the Operational Research Society, 60 (5), 149-160, Syntetos, A.A., Georgantzas, N.C., Boylan, J.E., Dangerfield, B.C. 2011. Judgement and supply chain dynamics. Journal of the Operational Research Society, 62 (6), 1138-1158. Teunter, R., Sani, B. 2009. Calculating order-up-to levels for products with intermittent demand. International Journal of Production Economics, 118 (1), 82-86. Thonemann, U. W. 2002. Improving Supply-Chain Performance by Sharing Advance Demand Information. European Journal of Operational Research, 142 (1), 81–107. Towill, D.R. 1982. Dynamic analysis of an inventory and order based production control system. International Journal of Production Research, 20, 369-383. Towill, D.R., Zhou, L., Disney, S.M. 2007. Reducing the bullwhip effect: Looking through the appropriate lens. International Journal of Production Economics, 108 (1-2), 444-453. Trapero, J.R., Kourentzes, N., Fildes, R. 2012. Impact of information exchange on supplier forecasting performance. Omega, the International Journal of Management Science, 40 (6), 738–747. The impact of supply chain structures on performance References Turrisi M., Bruccoleri M., Cannella S. 2013 Impact of reverse logistics on supply chain performance. International Journal of Physical Distribution & Logistics Management, 43 (7), 564 - 585. Villegas, F.A., Smith, N.R. 2006. Supply chain dynamics: Analysis of inventory vs. order oscillations trade-off. International Journal of Production Research, 44 (6), 1037-1054. Voigt, G., Inderfurth, K. 2012. Supply Chain Coordination with Information Sharing in the Presence of Trust and Trustworthiness. IIE Transactions, 44 (8), 637-654. Wang, J., Jia, J., Takahashi, K. 2005. A Study on the Impact of Uncertain Factors on Information Distortion in Supply Chains. Production Planning & Control, 16 (1), 2–11. Wang, K., Zeng, Z., Sun, D. 2008. Structure analysis of supply chain networks based on complex network theory. 4th international conference on semantics, knowledge and grid. Wang, S.-J., Wang, W.-L., Huang, C.-T., Chen, S.-C. 2011. Improving inventory effectiveness in RFID- enabled global supply chain with Grey forecasting model. Journal of Strategic Information Systems, 20 (3), 307-322. Wang, X., Disney, S.M., Wang, J. 2012a. Stability analysis of constrained inventory systems with transportation delay. European Journal of Operational Research, 223 (1), 86-95. Wang, X., Disney, S.M., Wang, J. 2012b. Exploring the oscillatory dynamics of a forbidden returns inventory system. International Journal of Production Economics. (Article in press) DOI: http://dx.doi.org/10.1016/j.ijpe.2012.08.013. Wangphanich, P., Kara, S., Kayis, B. 2010. Analysis of the Bullwhip Effect in Multi-Product, Multi- Stage Supply Chain Systems– A Simulation Approach. International Journal of Production Research, 48 (15), 4501–4517. Warburton, R.D.H. 2004. An analytical investigation of the Bullwhip Effect. Production and Operations Management, 13 (2), 150-160. Wei, Y. Wang, H. Qi, C. 2013. On the stability and bullwhip effect of a production and inventory control system. International Journal of Production Research, 51 (1), 154-171. Wen, L., Guo, M., Wng, L. 2012. The directed complex network application in the supply chain. Proceedings - 2012 3rd International Conference on Digital Manufacturing and Automation, 6298664, 911-914. Wikner, J., Naim, M. M., Towill, D.R., 1992. The system simplification approach in understanding the dynamic behaviour of a manufacturing supply chain. Journal of Systems Engineering, 2 (3), 164-178. Wright, D., Yuan, X. 2008. Mitigating the bullwhip effect by ordering policies and forecasting methods. International Journal of Production Economics, 113 (2), 587-597. Wu, D.Y., Katok, E. 2006. Learning, communication, and the bullwhip effect. Journal of Operations Management, 24 (6), 839-850. Roberto Domínguez References 130 Xuan, Q., Du, F., Li, Y., Wu, T.-J. 2011. A framework to model the topological structure of supply networks. IEEE Transactions on Automation Science and Engineering, 8 (2), 442-446. Yang, T., Wen, Y.-F., Wang, F.-F. 2011. Evaluation of robustness of supply chain information-sharing strategies using a hybrid Taguchi and multiple criteria decision-making method. International Journal of Production Economics, 134 (2), 458-466. Yoo, M.-J.; Glardon, R. 2009. Combining JADE and repast for the complex simulation of enterprise value-adding networks. Lecture Notes in Computer Science, 5386, 116-130. Yu, C.-Y., Huang, H.-P. 2001. Development of virtual foundry fab based on distributed multi-agents. Proceedings of the IEEE International Conference on Systems, Man and Cybernetics, 2, 1030-1035. Yuan, X., Shen, L., Ashayeri, J. 2010. Dynamic Simulation Assessment of Collaboration Strategies to Manage Demand Gap in High-tech Product Diffusion. International Journal of Computer Integrated Manufacturing, 25 (6), 647-657. Zhang, X., 2004. The impact of forecasting methods on the bullwhip effect. International Journal of Production Economics, 88 (1), 15-27. Zhang, L.; Lee, C.; Xu, Q. 2010. Towards product customization: An integrated order fulfilment system. Computers in Industry, 61 (3), 213-222. Zhang, C., Wang, H. 2012. A state-space based study of stability, bullwhip effect and total costs in twostage supply chains. International Journal of Innovative Computing, Information and Control, 8 (5A), 3399-3410. Zhou, L., Disney, S., Towill, D.R. 2010 A pragmatic approach to the design of bullwhip controllers. International Journal of Production Economics, 128 (2), 556-568. Zotteri, G. 2013. An empirical investigation on causes and effects of the Bullwhip-effect: Evidence from the personal care sector. International Journal of Production Economics, 143 (2), 489-498. Zhu, S., Xu, Y. 2012. Complexity measurement of supply chain networks. Proceedings of the 24th Chinese Control and Decision Conference, 6244356, 2220-2224. 131 Appendix A: Abbreviations ADE Agent Development Environment ANOVA ANalysis Of Variance APVIOBPCS Automatic Pipeline Variable Inventory and Order Based Production Control System ATP Available To Promise ATO Assemble To Order BWE Bullwhip effect BwSl Bullwhip Slope CAS Complex Adaptive System DOF Degree Of Freedom DSOPP Distributed Simulation of Order Promising Protocols FCFS First Come First Served GUI Graphical User Interface IDE Integrated Development Environment IOBPCS Inventory and Order Based Production Control System Roberto Domínguez Appendix A: Abbreviations 132 IQL Information Quality Level JADE Java Agent Development framework LPT Longest Process Time MAS Multi-Agent System MASCF Multi-Agent Supply Chain Framework MASON Multi-Agent Simulator of Networks MD Make-Deliver MRP Material Resource Planning MTO Make To Order MTS Make To Stock NPMA N-Periods Moving Averages NPMV N-Periods Moving Variances OFP Order Fulfillment Process ORVR Order Rate Variance Ratio OUT Order Up To RFQ Request For Quotation SCC Supply Chain Council The impact of supply chain structures on performance Appendix A: Abbreviations SCM Supply Chain Management SCML Supply Chain Modeling Language SCN Supply Chain Network SCOPE Sistemas COoperativos para la Producción y Ejecución de Pedidos SCOR Supply Chain Operations Reference SD Source-Deliver SES Simple exponential smoothing SISCO Simulator for Integrated Supply Chain Operations SMA Simple Moving Averages SMD Source-Make-Deliver SPT Shortest Process Time Std Dev. Standard Deviation WMA Weighted Moving Averages XML eXtensible Markup Language