* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Toward the fair sharing of profit in a supply network formation
Survey
Document related concepts
Transcript
Toward the fair sharing of profit in a supply network formation Jean-Claude Henneta,∗, Sonia Mahjouba,b a LSIS, CNRS-UMR 6168, Université Paul Cézanne, Faculté Saint Jérôme, Avenue Escadrille Normandie Niémen, 13397 Marseille Cedex 20, France b FIESTA , ISG Tunis, 41 rue de la liberté, 2000 Le Bardo, Tunisia Abstract The design of a supply chain network can be interpreted as a coalition formation problem in cooperative game theory and formulated as a linear production game (LPG). The companies which are members of the optimal coalition share their manufacturing assets and resources to produce a set of end-products and globally maximize their profits by selling them in a market. This paper investigates the possibility of combining the requirement of coalition stability with a fair allocation of profits to the participants. It is shown that, in general, the purely competitive allocation mechanism does not exhibit the property of fairness. A technique is proposed to construct a stable and fair allocation system when the core of the game does not exclusively contain a set of competitive allocations. Keywords: supply chain, cooperative game theory, coalitions, linear programming, duality * Corresponding author. Tel. : +33 491 05 6016 ; fax: +33 491 05 6033, E-mail address: [email protected] 1 1. Introduction In the context of modern information and communication networks, many firms are open to finding new partnerships and penetrating new markets. In the manufacturing sector, several classical paradigms and tools must be revisited in the light of more adaptability, flexibility and network reconfigurability. This paper describes a supply network formation process as a network of enterprises attempting to collaborate and capture a market share for a particular set of products. The result of such a collaboration may be called an emergent supply network. The basic and necessary ingredients to enable collaboration to prevail are the existence of a targeted market, and resources for manufacturing, communication and logistics. This study focuses on the latter success factor: existing resources could bring more profit if they were used jointly rather than separately. As a consequence, many enterprises are ready to use their own resources and those of other firms to produce services and products in the most advantageous quantity and at the lowest cost. Supply networks are thus characterized by a basic common goal and a highly integrated operational system, combined with a complex decentralized decisional organization. Such characteristics are of major concern in game theory, under the classical decomposition into cooperative (or coalitional) games and non-cooperative (or strategic) games. Models referred to as ‘non-cooperative’ are composed of players with different preference relations or utility functions. Their actions obey a strategy that takes into account information (generally imperfect) on the other players’ actions and preferences. As noted in Cachon and Netessine (2004), most supply chain models based on game theory use a non-cooperative approach. However, cooperative game theory seems more appropriate to analyze a supply network in its design stage, as it is characterized by many possibilities for enterprise coalitions and allocation patterns for tasks and rewards. In such a strategic stage, all the actors are ready to cooperate to create competitive advantages in the market. 2 Partner enterprises share a common goal that is clearly identified: penetrate a market segment to obtain the maximal global expected profit. A general characteristic of cooperative games is the understanding of the players that they can obtain a larger global benefit from pooling their resources than by acting separately. This characteristic is particularly important when resources are scarce, as in the case of spare parts for aircrafts studied by Kilpi et al. (2009), and when important economies of scale in the availability service can be achieved by cooperation. Several authors have used cooperative game theory to represent alliances between retailers in their relation to the market and/or to suppliers. Nagarajan and Sošić (2008) have studied the benefits expected by retailers from price setting, pooling their market share and sharing risks. Profit can also be increased by centralizing inventories. The works of Granot and Sošić (2003), Cachon and Netessine (2004) and Reinhardt and Dada (2005) provide convincing interpretations of supply chain design problems as cooperative games by focusing on manufacturing capacity and inventory pooling. In the light of cooperative game theory, a supply network can be modelled as a coalition of partners pooling their resources and sharing the same utility function (profit). The partnership building problem can then be modelled as a cooperative game with transferable utilities (TU-game). Such a game is firstly characterized by global optimization of the supply chain value (total expected profit). A TU-game can thus be seen as a target model on which the partners can agree to estimate the maximal value of the chain and the shares of the global profit acceptable to all of them. Another practical advantage of this model is that it evaluates and compares different possible coalitions. The results can thus be used as arguments in the supply chain design stage, to convince the partners to be part of the best possible coalition and set up the joint venture. Once the comparative advantage of cooperation has been demonstrated, the key remaining issue is the choice of the scheme for sharing benefits between the members. Expected profit is interpreted as the transferable utility that should be allocated to players in a manner that is 3 acceptable to all of them and guarantees coalitional stability. In cooperative game theory, several concepts have been introduced for approaching the stability issue. A necessary condition for the stability of a coalition is that no set of players is able to increase its members’ profits by forming a different coalition. The set of payoff profiles that verifies this property is known as the core of the TU-game (Gillies, 1959). It is the set of non-dominated feasible payoff profiles (also called imputations) covering all the possible coalitions. Non-emptiness of the core has been shown if the problem is convex (Shapley and Shubik, 1969). By introducing some restrictions on the possible deviations from a coalition, several sets have been defined to characterize stable and/or balanced imputations among the partners of the optimal coalition: stable set, bargaining set, nucleolus, kernel and the ‘Shapley value’ (see e.g. Osborne and Rubinstein, 1994). In recent years, several imputation mechanisms have been proposed in the literature on supply chain analysis by cooperative game theory. Several authors have proposed the use of the ‘Shapley value’ approach in various contexts related to supply chain design (Nagarajan and Sošić, 2008). In particular, Reinhardt and Dada (2005) have studied the case of n firms cooperating by pooling their critical resources. Due to the difficulty of computing the Shapley value for a large number of players, these authors have proposed a pseudo-polynomial algorithm to compute the Shapley value allocation of benefits for particular games, called ‘coalition symmetric’. In addition to its computational complexity, the Shapley value also has the drawback of not necessarily belonging to the core, as pointed out by Granot and Sošić (2003). Bartholdi and Kemahlıoglu-Ziya (2005) have also used the Shapley value to allocate profit in a network composed of two retailers and one supplier. They have shown, in particular, that such an allocation coordinates the supply chain but may appear unfair to the supplier. Another example of an alliance between retailers can be found in Guardiola et al. (2007). The authors analyze the case of a supply chain with one supplier and n retailers selling the same product in separate markets. Through cooperation, the retailers increase their profit by obtaining a lower 4 wholesale price from the supplier. In counterpart, the supplier receives equal side payments from all the retailers. The maximal value of these payments is computed to guarantee that the allocation mechanism belongs to the core of the retailers–supplier game. In an almost symmetrical manner, Jin and Wu (2006) have shown that in online reverse auctions there is one unique strongly stable suppliers’ coalition that allows for maximizing profit. Several studies have concentrated on the costing aspects rather than on profit sharing. In particular, Charles and Hansen (2008) have applied cooperative game theory to global cost minimization and cost allocation in an enterprise network. They have shown that under classical concave cost functions for all members, the cost allocation obtained by the activity based costing (ABC) technique is rational and belongs to the core of the game. In many practical situations, a supply network can be viewed as a multistage production system in which the different production stages are performed by different enterprises. Material requirement planning (MRP) theory (Grübbström, 1999; Grubbström and Huynh, 2006) can then be used to define and distribute responsibilities and manufacturing orders among the partners. In this view, the product structure supports the enterprise network organization, especially under an extended view of the bill of materials (BOM) such as the generic BOM (Lamothe et al., 2005). Additionally, multistage production by several producers highly differs from multistage production by a single producer because of the need for coordinating requirements and contracts negotiation. It also carries new possibilities for increasing the effectiveness of ordering and inventory policies (Arda and Hennet, 2006). In the design stage, the largest possible set of firms should be investigated for selecting partners and sharing resource costs and/or rewards in the most efficient manner. This paper analyzes the problem of optimally choosing the partners to constitute a multistage production system. By assumption, the firms who own manufacturing resources are ready to make them available to the supply chain if they obtain sufficient rewards. In such a scheme, some enterprises may be complementary (if they own complementary resources) and some may compete 5 (if they own resources of the same type). In addition, they all compete with each other to obtain the largest possible share of the global profit. Using MRP theory, this paper shows that the supply network formation problem can be formulated as a linear production game (LPG). Since the pioneer works of Shapley and Shubik (1972) and Owen (1975), many results have been obtained for this particular class of cooperative game. LPGs are characterized by a linear value function to be optimized by linear programming (LP) under linear constraints relating production to resources under capacity constraints. Since the work of Owen (1975), it has been well-known that the core of an LPG is generally not empty and that a purely competitive allocation scheme, called the Owen set, is contained in the core. This set is constructed from the optimal solution of the dual linear program, which defines the shadow prices of the resources. In the Owen set, each player receives the reward corresponding to the shadow prices of the resources that he owns. Then, if the dual problem has a unique solution, the Owen set consists of a single point, called the purely competitive allocation (Van Gellekom et al., 2000). These results have been extended to semi-infinite LP situations, under the assumption of a finite bound on the maximal profit, in two basic cases. Fragnelli et al. (1999) have considered the case of a potentially infinite number of end products, and Tijs et al. (2001) have studied the possibility of an infinite number of transformation techniques. Based on the previous works of Owen (1975) and Van Gellekom et al. (2000), this paper explicitly constructs the Owen set of the multistage LPG under study. Then, the quality of the solution is discussed in terms of stability and fairness. In particular, the existence of partners obtaining a null profit is considered as an unfair property that decreases the robustness of the coalition’s stability. The main issue of this paper is to study the existence of fair allocations in the core of LPGs. Theorem 1 in section 4 gives a sufficient condition for existence and a technique to construct such allocations. To solve this problem, some side payments are introduced from the partners with a positive payoff to their associates with a null payoff in the Owen allocation. The 6 maximal value of these side payments is computed under the condition of maintaining the allocation in the core of the game. If this value is strictly positive, as may occur if the competition is not too severe, then a new set of fair imputations lying in the core is constructed. Section 2 presents some preliminaries on cooperative game theory. Then, section 3 represents the multistage supply chain design problem as a cooperative linear production game. The problem is solved in terms of maximizing profit and forming a coalition with the smallest number of partners. The Owen set of this game is constructed and analyzed. It is shown via an example that, under the Owen allocation, some participants of the optimal coalition may receive no payoff. To correct this drawback as much as possible, the side-payments mechanism is proposed and studied in section 4. The main result of the paper is stated in Theorem 1. It provides a test of existence and a technique to construct an efficient, rational and fair allocation. Some conclusions on the applicability of the method are finally exposed. 2. Some preliminaries on TU-games Classically, a game involves a finite set of N players, denoted N = {1,..., N } , with N = Card( N ). A coalition S is a subset of N : S ⊆ N . The set P (N ) is the set of all the subsets of N . In a TU cooperative (or coalitional) game in the sense of Von Neumann and Morgenstern (1944), each coalition S ∈ P (N ) is characterized by a value function v( S ) ≥ 0 . The value v(S ) is the maximal utility (or payoff) that can be obtained by coalition S. Each player i ∈ N seeks to maximize his utility function, which is the payoff that he can obtain from belonging to a coalition. By convention, all the utilities are nonnegative. The minimal utility that a player may obtain is zero. In this model, the players’ actions within a coalition are implicit. In order to maximize his utility function, the only possible decision a player can take is to belong to a coalition or not. If, as the result of the game, coalition S , S ⊆ N prevails, then each player i ∈ S obtains a share vi ( S ) ≥ 0 such 7 that: ∑ vi (S ) = v(S ) , and each player j of N not belonging to S ( j ∈ N \ S ) has a null payoff: i∈S v j ( S ) = 0 . Notation N \ S represents the set of players that belong to N but not to S . All the utilities are transferable (TU-game) in the sense that they are all shares of the global payoff. A TU-game is thus simply noted (N, v) . It raises two basic problems: • global utility maximization: determination of the maximal value function and the coalitions S for which this value is obtained, • allocation problem: determination of the endowments of the agents by distributing the global payoff among them. As in Osborne and Rubinstein (1994), an S-feasible payoff profile is defined as a vector (u i ) i∈S such that ∑ ui = v(S ) , and a feasible payoff profile as a vector (u i ) i∈N i∈S such that ∑ ui = v(N ) . i∈N Let v * be the maximal global payoff of the TU-game (N, v) : v* = max v( S ) . S ∈P (N ) (1) Consider a feasible payoff profile (u i ) i∈N . With every coalition S we associate a payoff u (S ) defined by expression (2): u(S ) = ∑ ui . (2) i∈S Several properties will be now defined. These properties appear necessary for a payoff profile to solve a coalitional game. Property 1: Efficiency (Pareto optimality) The feasible payoff profile (ui )i∈N is said to be efficient (or Pareto optimal) if and only if 8 N u (N ) = ∑ ui = v * . (3) i =1 Another condition for a feasible payoff profile (ui )i∈N to be accepted by all the players is (coalitional) rationality, as defined below. Property 2: Rationality A feasible payoff profile (ui )i∈N is said to be rational if the payoff of every coalition S is larger than its value v(S ) : u ( S ) ≥ v( S ) ∀S; S ∈ P (N ) . (4) The core of the game (N, v) is a game-theoretic concept introduced in Gillies (1959). It is defined as follows. Definition 1: Core The core of a TU-game is the set of feasible payoff profiles (u i ) i∈N that satisfy conditions (3) and (4). Namely, it is the set of feasible payoff profiles that are both efficient (Pareto optimal) and rational. A major issue in the analysis of a TU cooperative game consists in characterizing its core. It can be noted that equations (2) and (3) and linear inequalities (4) define a closed and convex set. However, due to the combinatorial growth of the number of coalitions (cardinal of P (N ) ) with the number of players, N, the core is generally difficult to construct and fully explore. The approach that will be followed in this study for the LPG will rather be to construct a core solution by a known algorithm and then to locally modify this solution without escaping the core. An interesting index to characterize the value of a subset S' ⊆ S ⊆ N is its marginal contribution in S , denoted ∆ S (S ' ) and defined as follows. 9 Definition 2: Relative marginal contribution The marginal contribution of a subset S ' of a coalition S is: ∆ S ( S ' ) = v( S ) − v( S \ S ' ) (5) where notation S \ S ' represents the set of players that belong to S but not to S ' . In a classical manner, marginal contributions relative to the so-called ‘grand coalition’ N are simply called ‘marginal contributions’. Property 3: Marginal contribution principle An allocation that lies in the core satisfies the marginal contribution principle: u(S ) ≤ ∆ N (S ) ∀S; S ∈ P (N ) . (6) The proof of this property is straightforward: a feasible payoff profile (u i ) i∈N the core satisfies u (N \ S ) ≥ v(N \ S ) for every coalition S. that lies in Then, from u ( S ) = u (N ) − u (N \ S ) = v(N ) − u (N \ S ) , one derives u ( S ) ≤ v(N ) − v(N \ S ) . Hence, a core allocation is such that any coalition cannot obtain a payoff greater than its marginal contribution. With the objective of studying the core of a TU-game, inequalities (6) are particularly useful since they provide upper bounds to coalition payoffs, while rationality conditions (4) provide lower bounds. The maximal global payoff of the TU-game (N, v) , v * , has been defined by equation (1). As this value can be obtained for several coalitions, it is possible to introduce a secondary criterion to define the optimal coalitions. Definition 3: Optimal coalition 10 The optimal cardinality the TU-game (N, v) is: s* = min{card ( S ) v( S ) = v *}. An optimal coalition of the TU-game (N, v) is a coalition S * ⊆ N that satisfies v( S * ) = v * and card ( S ) = s * . The most critical situation for a player in a TU-game is when his utility remains at its minimal value, zero. In other words, his share of the global payoff, v * , is null. For a given feasible payoff profile (ui )i∈N , a player who does not receive any payoff is called a ‘null payoff player’ (NPP), according to the following definition. Definition 4: Null payoff player (NPP) An NPP with respect to a feasible payoff profile (u i ) i∈N is a player i ∈ N such that ui = 0 . For every optimal coalition S * , two types of NPPs may be distinguished: players outside and players inside the optimal coalition S * . The following properties derive from the definitions above. Property 4 Under a core allocation profile, any player who does not belong to every optimal coalition is an NPP. Proof. Suppose j ∈ N \ S * with S * as an optimal coalition. Then, by the efficiency property of the payoff profile, u (N ) = u (N \ S *) + u ( S *) = u ( S *) = v * , therefore ∑u i∈N \ S * i = 0 and, by the nonnegativity of utilities, u j = 0 ∀j ∈ N \ S * . Property 5 Under a core allocation profile, any NPP which belongs to an optimal coalition, i ∈ S * satisfies the two following conditions: v({} i ) >0 . i ) = 0 and ∆ S* ( {} 11 Proof. Condition ui ≥ v({} i ) ∀i ∈ N is a well-known property, called ‘individual rationality’. For a core allocation, this property is verified as a particular instance of rationality property (4), for sets S ⊂ N with cardinality 1. For an NPP, condition ui = 0 implies v({} i ) = 0 . To show ∆ S * ( {} i ) > 0 , it suffices to apply definition 5 to note that the reverse condition, ∆ S *( {} i ) = 0 , would contradict the minimality of coalition S * . With this background, it is now possible to propose a definition of fairness for payoff profiles. Definition 5: Fairness A feasible payoff profile (u i ) i∈N is fair if all the members of an optimal coalition S * obtain a strictly positive payoff : ui > 0 ∀i ∈ S * . This definition raises the problem of the existence of fair solutions belonging to the core of a TU-game. To show that some TU-games cannot admit any fair solution, it suffices to construct games that possess several different optimal coalitions, optimality meaning that the coalition creates the maximal utility value, v* , and has minimal cardinality in the set of coalitions S verifying v( S ) = v* . It will now be shown that games with this property can be constructed, as inspired by Aumann (1964), by replicating one or several players of an optimal coalition. Consider a production game (N, v) in which the players own resources that they put into common use to manufacture products that create value. Consider a player i who belongs to an optimal coalition S * and suppose that he owns resources that are useful but not critical, in the sense that he possesses all of them in excess of what is required for the optimal production mix. Other resources that he does not own are critical. Let us now define a new game, (N ∪ {i' }, v) , by introducing a new player, denoted i’, exactly identical to player i. 12 Because of the non-criticality of the resources owned by players i and i’, the act of increasing their quantity does not change the optimal production mix. Hence, the optimal value of the new ( ) game is the same as the optimal value, v*, of the original game. Coalitions S * and S * \ {} i ∪ {i '} are both optimal for the new game. Therefore, Property 4 implies ui = ui ' = 0 for any core allocation u of the game (N ∪ {i' }, v) . And yet player i belongs to the optimal coalition S * , and player i’ to the ( ) optimal coalition S * \ {} i ∪ {i '} . Therefore, player i is an NPP in S * and player i’ an NPP in (S * \ {}i )∪ {i'}. As a consequence, a necessary condition for the existence of a fair solution in the core of a TUgame is the uniqueness of the optimal coalition S * . Note that this is an intrinsic property of the game, independent from the core allocation considered. 3. Coalition formation for profit maximization 3.1. A multistage model Consider a network of N firms represented by numbers in the set N = {1,..., N } . These firms are willing to cooperate to produce commodities and sell them in a market. These commodities belong to a set of g manufactured products (or families of products) i=1,…,g. Typically, manufacturing recipes are fixed and well defined. The gozinto graph describes the product structure and has no cycle. It can then be decomposed into levels: level 0 products are the g final products. Then, intermediate and primary products are numbered in the increasing order of their level. The level of product i, for i=g+1,...,n is the maximal number of stages to transform product i into a final product. Each production stage is supposed to have several input products but only one output product. The BOM technical matrix, G, is defined as follows: according to the given manufacturing recipe, production of one unit of product i requires the combination of components j=1,...,n in 13 quantities G ji . It can be noted that under a level-consistent ordering of products, matrix G then has a lower triangular structure (Hennet, 2003). In the context of MRP theory (Grübbström, 1999; Grubbström and Huynh, 2006), matrix G is the input matrix. When combined with the lead-time (output) matrix, it can be used to define the generalized Leontieff inverse of the system. However, the scope here will be restricted to stationary balance equations under a simple production structure. In this case, the output matrix reduces to the identity matrix. Let x = (( xij ) be the matrix of the quantities of product i produced (or purchased) at firm j and y = ( y1 yn )T be the output vector of products during a reference period. The components of this matrix and vector are the variables of the design problem. For simplicity, quantities per period (or throughputs) are supposed to be continuous: x ∈ ℜ n+× N , y ∈ ℜ n+ . Due to the sharing of resources, the problem can be formulated in terms of the global throughput vector, denoted ω and related to matrix x through the elementary summation relation (7): ω = x1 with 1 being the unit vector of dimension N. (7) The output vector can be computed from the global throughput vector by the following relation: y = ( I − G )ω (8) with I the identity matrix of dimension n × n . Matrix G being componentwise nonnegative and lower triangular (with 0s on the diagonal), matrix ( I − G ) is a regular –M-matrix and ( I − G ) −1 is lower-triangular (with 1s on the diagonal) and nonnegative (Berman and Plemmons, 1979). Then the BOM can be expressed as follows: 14 ω = ( I − G )−1 y . (9) There are N enterprises which are candidates to be part of the supply chain to be created. Each candidate enterprise is characterized by its production resources: manufacturing plants, machines, work teams, robots, pallets, storage areas, etc. As in Van Gellekom et al. (2000), a coalition S is defined as a subset of the set N of N enterprises with characteristic vector e S ∈ {0,1}N such that: (e S ) j = 1 (e S ) j = 0 if j ∈ S . if j ∉ S (10) For the R types of resources considered (r=1,…,R), let crj be the amount of resource r available at enterprise j, C = ((crj )) ∈ ℜ R× N , and mri the amount of resource r necessary to produce one unit of product i, M = ((mri )) ∈ ℜ R×n . Capacity constraints for coalition S are written: n ∑ mriωi ≤ ∑ crj or equivalently i =1 j ∈S n ∑ mriωi ≤ ∑ crj (e S ) j . i =1 (11) j∈N It can be noted that the left-hand side quantities in inequalities (11) fully characterize coalition S. One originality of the proposed model is that it combines the coalition decision problem through the choice of vector eS , with the multistage manufacturing model represented by constraints (9) and (11). The second originality is related to the choice of the objective function that aggregates the benefits expected from manufacturing activities. The value chain concept introduces transfer prices for intermediate goods so that each manufacturing stage should be profitable. 15 3.2. The Value chain Let χ = ( χ 1 , , χ R ) T be the vector of unit costs for resources. The unit cost for product i, γ i , is assumed to be the same for all the firms possessing the required resources. It is determined by the cost resource requirement for producing one unit of product i: R γ i = ∑ χ r mri . (12) r =1 Let γ = (γ 1 ,, γ n )T be the vector of unit production costs for the products. The set of relations (12) for all the products can be written in vector form: γ = MTχ . (13) Prices may be partly or totally exogenous. Let p = ( p1 pn )T be the vector of market prices for the final products. A necessary and sufficient condition for the global profitability of the supply chain is: pT y − γ T ω ≥ 0 , (14) or, using relation (9), [ pT − γ T ( I − G ) −1 ] y ≥ 0 . (15) The prices of final products can be supposed fixed and exogenous. The global profitability condition (15) can be considered sufficient for the supply chain to be viable. More restrictive conditions will now be established to allow for a total decomposability of the multistage manufacturing process. It is now assumed that the prices of intermediate products are 16 negotiated in the network so that each manufacturing stage is profitable. Under this more restrictive requirement, the following inequality should be verified for any product i: n pi ≥ γ i + ∑ Gij p j j =1 (16) . Let π = (π1,, π n )T be the vector of unit profits for products. The unit profit π i associated with product i is: n π i = pi − γ i − ∑ Gij p j j =1 (17) . Condition (16) corresponds to the profitability conditions π i ≥ 0 ∀i ∈ {1,, n} that characterize the value chain. These conditions can be gathered into the following condition in vector form: π = (I − G) p − γ ≥ 0 . (18) Condition (18) may appear unnecessarily restrictive. However, it can be used to fix the transfer prices between products in the network so that each manufacturing stage is intrinsically profitable. 3.3. Global profit maximization The maximal total payoff, v * , is obtained from the solution of the mixed variables’ Linear Programming problem (P): T N Maximize v =< ( I − G ) p − M χ , y >= ∑ π i yi i =1 subject to M ( I − G ) −1 y ≤ Qe (P) y ∈ ℜn+ , e ∈ {0,1}N with the cardinality of the coalition as the complementary objective to be minimized: 17 N s = ∑ (e) j . (19) j =1 The secondary objective can be solved together with the global profit maximization problem by adding a ‘small’ term − ε ( N ∑ (e j =1 S ) j ) with 0 <ε <<1 to v in the objective function of (P). The modified problem, (P'), is formulated as follows: n N Maximize ϕ = ∑ π i y i − ε (∑ e j ) i =1 j =1 −1 subject to M ( I − G ) y ≤ QeS y ∈ ℜ +n , e ∈ {0,1} N The term − ε ( N ∑ (e) j =1 j (P') . ) is said to be ‘small enough’ if the optimal solutions of (P) and (P’) are identical with respect to the optimal vector y * . This property is achieved for any value of ε smaller than a threshold value ε 0 > 0. The optimal coalition of lowest cardinality S* is then obtained by resolution of problem (P'): S * = {j ; j ∈ {1, , N }, e j * = 1}. This coalition is supposed not empty. Accordingly, the maximal value function of the TU-game is supposed strictly positive. It is exactly computed from the solution of (P’) by: v* = ϕ * +ε ( N ∑ e *) . j =1 j In problem (P), vector e is a vector of binary variables. By definition, problem ( PS ) relates to a particular coalition S ⊆ N . It is defined by the same constraints as (P), but with the vector of variables e replaced by the known vector eS given by (10): (e S ) j = 1 (e S ) j = 0 if j ∈ S . if j ∉ S 18 N Maximize vS = ∑ π i yi i =1 subject to M ( I − G ) −1 y ≤ QeS (PS) y ∈ ℜ +n . The TU-game (N, v) defined in this section is an LPG as defined in Owen (1975). It is an N-persons game in which the value v(S ) of a coalition S is obtained as the solution of an LP problem, (PS), defined by the nonnegative production matrix M ( I − G ) −1 , the nonnegative resource matrix Q and the nonnegative unit profit vector π . Several properties can be derived from this definition. In particular, it is super-additive: v( S ) + v(T ) ≥ v( S ∪ T ) for all disjoint coalitions S and T , and it has a non-empty core (Owen, 1975). Superadditivity implies, in particular, that if S * is an optimal coalition, then v( S ) = v* for any coalition S such that S * ⊆ S . In particular, the maximal total payoff, v* can simply be obtained for the grand coalition, N , by solving problem PN . 3.4. The core of the linear production game The core of a TU-game is often difficult to determine. Some subsets are of particular importance. It is a classical result that in exchange economies (markets with transferable payoffs), every competitive allocation is in the core (Shubik, 1959). For LPGs, the set of competitive allocations has been characterized by Owen (1975) and called the Owen set by Van Gellekom et al. (2000). Conversely, for coalitions with a limited number of players, as in the supply chains example, the core may be substantially larger. On the other hand, the set of competitive allocations often reduces to a single point. It is thus interesting, in particular regarding improving negotiation convergence and robustness, to consider payoff allocations that lie in the core but possibly outside the set of competitive allocations. 19 3.5. The Owen set The numerical resolution of problem (P’) solves the global utility maximization problem presented in section 3.3.: it defines the maximal total payoff, v*, and an optimal output vector y*. Problem (P) can then be replaced by problem ( PS * ), in which all the variables are real numbers. At this stage, however, the allocation problem, which determines the share of the total payoff to be distributed to each coalition partner, still remains open. Consider the dual of ( PS ), denoted ( DS ): Minimize wS = R ∑ qr ( S ) z r r =1 −T subject to ( I − G ) MTz ≥π z = (z1, ,z R )T ∈ ℜ+R ( DS ) . The coefficient of variable zr in the objective function is the quantity of resource r available for production if coalition S is selected: N qr ( S ) = ∑ (eS ) j crj = ∑ crj . j =1 (20) j ∈S It can be noted that the set of constraints of ( DS ) is the same for any coalition S. Further, since the optimal dual variables zr* ( S ) can be interpreted as shadow prices for resources, they determine a vector of payoffs, the so-called ‘Owen set’ for this TU-game, which is optimal in the context of a purely competitive economy (Van Gellekom et al., 2000). Owen (1975) has shown that the Owen set of an LPG is contained in the core of the LPG. 20 Consider the optimal solution (S*, y*) of problem (P). The purely competitive payoff profile u* = u ( S *) is also called the Owen set or the Owen point, since it is generally a single point. It is obtained from the solution (wS*, z*(S*)) of ( DS * ) through the following relation: R * u j * = ∑ crj zr if j ∈ S * . u j * = ∑ (eS * ) j crj zr * or equivalently, r =1 r =1 = ∉ u j S * 0 if * j R (21) Expression (21) shows that the payoff of each player equals the value of his resource bundle under the marginal price. Moreover, this vector of payoffs forms a subset of the core in this production game. However, it will be shown that the Owen set allocation has some drawbacks, in particular a lack of fairness that may induce a corresponding lack of stability robustness against the objections to coalition S*. 3.6. Example Consider the BOM of the example in Hennet (2003) with two final products (1 and 2), three intermediate products (3, 4 and 5), the unit profit vector π = (22 25 0 0 0)T and the technical 0 0 matrix G = 2 0 1 0 0 2 0 2 0 0 0 3 2 0 0 0 0 0 0 0 . Four resources are necessary for the five products at the different 0 0 0 1 0 manufacturing stages, with the following requirement matrix M = 0 2 1 0 0 2 0 1 1 1 0 1 1 1 0 0 . 2 1 Ten enterprises are candidates for partnership in the supply chain. The amounts of the four resources owned by the ten firms are represented in the following matrix: 21 0 0 C= 0 3 0 0 0 0 1 0 0 3 0 0 5 5 9 10 0 0 0 2 . 0 12 10 7 13 0 9 19 0 3 0 5 5 0 27 20 0 2 The optimal total payoff and optimal coalition are obtained from the solution of the LP (P) (with the additional term to obtain a coalition of lowest cardinality). The maximal total payoff is v* = 87.5 , obtained for y* = [0 3.5 0 0 0 ]T . It is obtained by the minimal coalition S * = {3 4 5 6 7 8 9} and also, indeed, by any coalition containing S*. The associated purely competitive payoff profile (Owen set) is obtained by formula (21) for z * = [0 0 1.25 0] : u* = [0 0 15.00 12.50 8.75 16.25 0.00 11.25 23.75 0] . In this solution, the endowment of partner 7 is null. Yet, partner 7 is important to the coalition since for coalition S * −{7} , the total payoff drops down to 46.875! Its individual marginal contribution in S* is: ∆ S * ({7}) = v(S *) − v(S * \{7}) ≅ 40.625 . The reason for a null endowment is that, with partner 7 in the coalition, resource 4 is in excess and its shadow price falls to 0. The individual marginal contribution of partner 7 in the grand coalition N is also strictly positive: ∆ N ({7}) = v(N ) − v(N \ {7}) = 87.5 − 59.375 = 28.125 . From the marginal contribution principle (6), non nullity of ∆ N ({7}) is a necessary condition for the existence of a core payoff profile (ui )i∈N with u7 > 0 . 22 4. Improving payoff functions with side payments 4.1. Some criticisms of the Owen set allocation The property of not allocating any payoff to the players with resources in excess is inherent in the Owen set since this solution rule derives from the duality principle in linear programming. In duality theory, a shadow price indicates the value of one additional unit of the resource associated with the corresponding primal constraint. Thus, if a dual variable is equal to zero (zr* = 0), this means that the addition of one unit of resource r has no effect on the optimal objective function. Hence the allocations based on the values of the optimal dual variables exhibit this property: the players with scarce resources share the total worth of the coalition between them, and the players with excess resources get a null payoff. In light of these shortcomings, the Owen set solution may become critically stable since the players with excess resources are indifferent as to whether they participate. Such a lack of incentive toward NPPs is not desirable in supply chain design because it does not provide stability robustness against objections. Therefore, even if the allocation mechanism may better reward the players with scarce resources, allocating a null imputation to the players with all resources in excess may not be the best solution. The next section explores the possibility of constructing a core solution with a strictly positive payoff for each player of the optimal coalition. 4.2. Trying to combine fairness with rationality Due to the criticisms directed at the Owen set as a competitive solution, other practical allocation schemes will be explored to ensure that the maximal payoff is fairly allocated to the players of the LPG game. So, the objective is to find an allocation that is fair and belongs to the core of the LPG. These requirements can be formulated by the following set of constraints: 23 n ∑x i =1 i = v* ; ∑ x j ≥ v( S ) (22) ∀S ⊂ N ; and (23) j ∈S x j > 0 ∀j ⊂ S * . (24) As shown in the previous example, the optimal payoff may strongly decrease if a player who only owns resources that are globally in excess decides not to participate in the cooperative game. This is why condition (24), denoted ‘fairness property’, has been added to core requirements (22) and (23); it provides a positive gain to every partner of the optimal coalition. However, the existence of solutions to the set of conditions (22), (23) and (24) is not guaranteed in general. The purpose of this section is precisely to characterize the cases when such solutions exist. By construction, the core of an LPG is not empty if problem ( PN ) has a finite optimal solution that is strictly positive, v(N ) > 0 . This is clear from the fact that an Owen solution exists whenever the dual problem ( DN ) , or equivalently ( DS * ), has a solution. Then, in order to explore whether the core of the game contains solutions that also satisfy (24), the following set of imputations can be constructed. The purely competitive payoff profile u* = u ( S *) has been defined by (21). The set N can be decomposed into three disjoint subsets, as follows: N = S 0 * ∪S 1 * ∪S 2 * (25) with S0 * = {i; i ∈ S * and ui* = 0} , S1 * = {i; i ∈ S * and ui* > 0} , S 2 * = N − S * . 24 The cardinals of these sets are respectively denoted s 0 *, s 1 *, s 2 * with, by assumption, s 0 * ≥ 1, s 1 * ≥ 1 . This corresponds to the case when the total payoff is strictly positive and some of the players in the optimal coalition are NPP: their Owen allocation is null. A new set of imputations, denoted w , is defined as follows: wi = ui * +α = α ∀i ∈ S0 * wi = ui * − β ∀i ∈ S1 * , w = u * = 0 ∀i ∈ S * i i 2 (26) α ≥ 0 , β ≥ 0 , α s0 * = β s1 * . (27) with ∑ u i * −v ( S ) i∈S1 β ≤ min with S1 = S ∩ S1 * and s 1 = card ( S1 ) s1 S ⊂N (28) β < min ui * (29) i∈S * 1 Theorem 1 A set of imputations w that satisfies conditions (26)-(29) belongs to the core of the LPG. Furthermore, there exist strictly positive values of β that satisfy (28)-(29) and define fair imputations if and only if s0 * β = min~ S ⊂ S s 1 s 0 * −s 0 s 1 ∑ u j * −v( S ∪ S 2 *) > 0 * j ∈S 1 with ~ S ⊂ S* and s1 s 0 * − s0 s 1 * > 0 . Proof 25 • Efficiency : From condition (26), n ∑ wi = ∑ ui * + α s0 * − β s1 * . Then, Property (22) for i =1 i∈N imputation w is obtained from relation (27) and the efficiency property of the Owen n n i =1 i =1 imputation ( • ∑ ui * = v * ): ∑ wi = v * . Fairness: ∑ u i * −v ( S ) i∈S1 By construction of S1 * , min u i > 0 . Then, If β = min > 0 , it is possible to s1 i∈S * S ⊂N 1 select β > 0 such that (22) and (23) are satisfied. • Rationality: The Owen imputation belongs to the core and thus satisfies the rationality property: ∑ u j * ≥ v( S ) j ∈S ∀S ⊂ N (30) . Consider now the new imputation w and a coalition S ⊂ N . Then, define S0 = S ∩ S0 * , with S1 = S ∩ S1 * , S2 = S ∩ S2 * and s 0 = card ( S 0 ) , s 1 = card ( S1 ) . ∑ w j = ∑ u j * + α s0 − β s1 j ∈S j ∈S If α s 0 − β s 1 ≥ 0 , then condition (25) implies ∑ w j ≥ v(S ) for any value of β satisfying j ∈S (27), (28) and (29). 26 Consider now the case α s 0 − β s 1 < 0 . Using relation (18), this condition can be equivalently replaced (with s 0 * ≠ 0 ) by s1 s 0 * − s0 s 1 * > 0 . Then, condition ∑ w j ≥ v(S ) is satisfied for any j ∈S value of β that verifies s0 * β≤ s 1 s 0 * −s 0 s 1 ∑ u j * −v( S ) . * j∈S 1 (31) Then, by super-additivity of v , v( S 0 ∪ S1 ∪ S 2 ) ≤ v( S 0 ∪ S1 ∪ S 2 *) and the two sets have the same parameters s 0 , s 1 . Therefore it suffices to test condition (31) for sets such that s1 s 0 * − s0 s 1 * > 0 and s 2 = s 2 * . Conversely, if condition (31) for some set S implies β = 0 , then imputation u * is the only one satisfying conditions (26)-(29). 4.3. Example In the example of section 2.5, the optimal coalition is S * = {3 4 5 6 grand coalition N 7 8 9} and the can be partitioned as follows: N = S0 * ∪S1 * ∪S 2 * with S0 = {7} , S1 = {3,4,5,6,8,9} and S 2 = {1,2,10}. The total number of coalitions in N is 210 − 1 = 1023 . However, using theorem 1, only 2 6 − 1 = 63 coalitions have to be tested to determine the value of β . In this example, s 0 * = 1, s 1 * = 6 and condition s1 s 0 * − s0 s 1 * > 0 requires s0 = 0 . Then, the possible values of s1 ( s1 =1,…,6) generate the 2 6 − 1 sets for which condition (27) has to be tested. The minimal bound, β = 1.25 is obtained for the set S * = {1,2,6,8,10} and since β > min ui , the value β defines the following imputation which belongs to the core and satisfies i∈ S 1 the condition of fairness: 27 w = [0.00 0.00 13.75 11.25 7.50 15.00 7.50 10.00 22.50 0.00] . The proposed technique has actually constructed the set of core allocations w( β ) defined by parameter β in the interval [0 1.25] and such that: w( β ) = [0 0 15.00 - β 12.50 - β 8.75 - β 16.25 - β 6 β 11.25 - β 23.75 - β 0] . For β > 0 , the solution w( β ) belongs to the core and has the property of fairness. The feasible interval for β can be used as a negotiation space by the players of the optimal coalition. 4.4. Discussion The situation described in the numerical example has only 10 players with resource capacities that are not oversized. It has been shown that, in this particular case, the core is not reduced to a single point and it is possible to construct imputations that are both fair and rational. In contrast to this situation, other numerical experiments have been performed with more players and several owners for each resource. These results show that, in this case, the core of an LPG game is generally reduced to a single point, which is precisely the Owen competitive allocation. This result is not surprising if one realizes that in an open market in which many players have similar abilities and equipment, the core of an LPG tends toward a single point that characterizes the perfectly competitive situation. Convergence of the core to the Owen set of an LPG has been shown and characterized by Owen (1975) and Semet and Zamel (1984) for games in which players are replicated when the number of replications tends to infinity. In a broader context, a well-known result is that, for large numbers of players, the core of the game tends to be the set of competitive allocations (Aumann, 1964). Using the LPG framework, this study has provided a rational explanation for the difficulty in conciliating economic efficiency with a fair repartition of profit. A related finding is that, in Linear 28 Production Games, cooperation and competition are complementarily entangled rather than opposed. Many practical examples of cooperative competition, also called “co-opetition” (Brandenburger and Nalebuff, 1997), can be found in today’s industrial world. A well-known example is the short-term cooperation that took place between the competing companies IBM and Oracle to develop ERPs for SMEs. On the one hand, coalition formation between retailers, manufacturers, and suppliers can provide a competitive advantage on the market because of the increased manufacturing possibilities generated by the sharing of competence and resources. On the other hand, competition between owners of similar resources or between manufacturers of similar products tends to decrease the value of these resources or products. As a consequence, it may jeopardize the coalitional stability by decreasing the partners’ motivation to cooperate. If a resource becomes potentially available beyond its necessary level, its marginal value decreases to zero, and so does the profit allotted to its owners for possessing it. In this respect, a good business strategy for a company could be to maintain the uniqueness of its products through technical progress and innovation. The search for cases when fair repartitions of profits exist also indicates that in order to maintain good profit margins, an enterprise should become a partner of its complementors and avoid associations with its competitors. The difficulty is that the same company may be both the enterprise’s complementor and its competitor. In such a case, a possible path to reach a win–win situation is not to enter the game as it is, but rather to change its rules, typically through negotiation with the partners. The current industrial and commercial practices show that many agreements and contracts are currently established to reach mutually profitable situations, with a particular attention paid to the business legal regulations to be respected. In terms of practical significance, this study has thus provided some clues for achieving economic profitability, by identifying some key properties that generate positive profits: ownership or production of assets that are not easily available elsewhere and cooperation with firms producing complements of the enterprise’s own products. 29 5. Conclusions In the light of cooperative game theory, supply networks can be modelled as profit-maximizing systems creating value in their socio-economic environment, considered as a market. Under the assumption of a perfect sharing of resources and production capacities in their manufacturing environment, supply network design problems can be represented as problems of optimal coalition formation. Then, coalition stability requires efficiency and rationality in the distribution of profit among the enterprises of the supply network. However, if the manufacturing environment is highly competitive, only the owners of resources that are marginally scarce receive a strictly positive share of the profit. A firm does not receive a share of the profit if all its resources are also possessed by other firms in the manufacturing environment and are globally in excess. Its revenue simply covers its costs. This result is consistent with the general equilibrium analysis in competitive economies. However, such profit allocation rules are not motivating and may appear unfair to the firms that belong to the optimal coalition without owning any scarce resource. This paper has shown that in moderately competitive manufacturing environments, where resources are not very abundant, the core of the game, which is the set of efficient and rational profit allocations, is not always restricted to the purely competitive profit allocation rule. Under these conditions, a set of feasible allocation rules has been constructed to guarantee a positive profit to all the enterprises in the supply network. A possible extension of this work could be to use this feasible set as a negotiation set for the firms of the supply network. References Arda, Y., Hennet, J.C., 2006, Inventory control in a multi-supplier system. International Journal of Production Economics, 104 (2), 249–259. Aumann, R.J., 1964, Markets with a continuum of traders. Econometrica, 32, 39–50. 30 Bartholdi III, J.J., Kemahlıoglu-Ziya, E., 2005, Using Shapley value to allocate savings in a supply chain. In: Geunes, J., Pardalos, P., (Eds.). Supply chain optimization. Springer Science+Business Media Inc., New-York, U.S.A., 169–208. Berman A., Plemmons, R.J., 1979, Nonnegative matrices in the mathematical sciences. Providence, RI: Academic Press. Brandenburger A.M., Nalebuff B., 1997, Co-Opetition: A revolution mindset that combines competition and cooperation, Currency Doubleday, New York. Cachon, G., Netessine, S., 2004, Game theory in supply chain analysis. In: Simchi-Levi, D., Wu, S.D., Zuo-Jun, M.S., (Eds.). Handbook of quantitative supply chain analysis: Modeling in the ebusiness era. Kluwer Academic Publishers, USA. Charles S.L., Hansen, D.R., 2008, An evaluation of activity-based costing and functional-based costing: A game-theoretic approach. International Journal of Production Economics, 113 (1), 282– 296. Fragnelli, V., Patrone, F., Sideri, E., Tijs, S., 1999, Balanced Games arising from infinite linear models. Mathematical Methods of Operations Research, 50, 385–397. Gillies, D.B., 1959, Solutions to general non-zero-sum games. In: Tucker, A.W., Luce, R.D., (Eds.). Contributions to the theory of Games vol. IV. Annals of math studies, vol. 40. Princeton, NJ: Princeton University, 47–85. Granot, D., Sošić, G., 2003, A three stage model for a decentralized distribution system of retailers. Operations Research, 51, 771–784. Grubbström, R.W., 1999, A net present value approach to safety stocks in a multi-level MRP system. International Journal of Production Economics, 59 (1-3), 361–375. 31 Grubbström, R.W., Huynh, T., 2006, Multi-level, multi-stage capacity-constrained production– inventory systems in discrete time with non-zero lead times using MRP theory. International Journal of Production Economics, 101 (1), 53–62. Guardiola, L.A., Meca, A., Timmer, J., 2007, Cooperation and profit allocation in distribution chains. Decision Support Systems, 44, 17–27. Hennet, J.-C., 2003, A bimodal scheme for multi-stage production and inventory control. Automatica, 39, 793–805. Jin, M., Wu, S.D., 2006, Supplier coalitions in on-line reverse auctions: Validity requirements and profit distribution scheme. International Journal of Production Economics, 100 (2), 183–194. Kilpi, J., Töyli, J., Vepsäläinen, A., 2009, Cooperative strategies for the availability service of repairable aircraft components. International Journal of Production Economics, 117 (2), 360–370. Lamothe, J., Hadj-Hamou, K., Aldanondo, M., 2005, Product family and supply chain design. In: Dolgui, A., Soldeck, J., Zaikin, O., (Eds.). Supply chain optimisation – Product/process design, facility location and flow control. Springer, New-York, U.S.A., 175–190. Nagarajan, M., Sošić, G., 2008, Game-theoretic analysis of cooperation among supply chain agents: review and extensions. European Journal of Operational Research, 187, 719–745. Osborne, M.J., Rubinstein, A., 1994, A course in game theory. The MIT Press, Cambridge, Massachussetts, U.S.A, London, England. Owen, G., 1975, On the core of linear production games. Mathematical Programming, 9, 358–370. Reinhardt, G., Dada, M., 2005, Allocating the gains from resource pooling with the Shapley value. Journal of the Operational Research Society, 56, 997–1000. 32 Samet, D., Zemel, E., 1984, On the core and dual set of linear programming games. Mathematics of Operations Research, 9, 309–316. Shapley, L.S., Shubik, M., 1969, On market games. Journal of Economic Theory, 1, 9–25. Shapley, L.S., Shubik, M., 1972, The assignment game 1: The core. International Journal of Game Theory, 1, 111–130. Shubik, M., 1959, Edgeworth market games. In: Tucker, A.W., Luce, R.D., (Eds.). Contributions to the theory of games vol. IV. Annals of Math Studies, vol. 40. Princeton, NJ: Princeton University Press, 267–278. Tijs, S., Timmer, J., Llorca, N., Sánchez-Soriano, J., 2001, The Owen set and the core of semi-infinite linear production situations. In: Goberna, M.A., López, M.A., (Eds.). Semi-infinite programming, recent advances. Dordrecht: Kluwer, 365–386. Van Gellekom, J.R.G., Potters, J.A.M., Reijnierse, J.H., Engel, M.C., Tijs, S.H., 2000, Characterization of the Owen set of linear production processes. Games and Economic Behavior, 32 (1), 139– 156. Von Neumann, J., Morgenstern, O., 1944, Theory of games and economic behavior. Princeton, NJ: Princeton University Press. 33