Shapley Value Analysis of Cooperative Formation of Procurement Networks Y. Narahari Computer Science and Automation, Indian Institute of Science, Bangalore Joint Work with T.S. Chandrashekar, GM ISL, Bangalore September 2007 1 E-Commerce Lab, CSA, IISc OUTLINE 1. Supply Chain Formation Problem: Motivation and Approaches 2. Cooperative Games in Characteristic Form The Shapley Value 3. The Multi-Unit Procurement Network Formation (MPNF) Game 4. Results on Shapley Value Analysis 5. Conclusions and Future Work 2 E-Commerce Lab, CSA, IISc Supply Network for Automotive Stampings Master Coil Cold Rolling Pickling Slitting Stamping 1 2 3 4 2 3 4 5 6 7 6 7 Suppliers 3 E-Commerce Lab, CSA, IISc Procurement Network Formation: Why Economic activity often involves inter relationships at multiple levels of production. Supply chains are deep. Individual entities in the supply chain are rational economic agents. Implicit involvement in deciding the supplier’s supplier with a focus on quality and inventory. Going further one can expect explicit involvement in price setting, capacity planning, etc. Global Supply Chains Vendor with Management specialized suppliers Programs Network Formation is critical to Supply Chain Planning 4 E-Commerce Lab, CSA, IISc Procurement Network Formation: A Cooperative Approach Anecdotal evidence suggests that negotiation and bargaining are key mechanisms to settle procurement contracts. Industrial Electronics - IMEC to Texas Instruments. Automotive Industry - Delphi and Lear to GM, Ford and DaimlerChrysler. Automation Equipment - Symbol Technologies + Paxar to Home Dept and Walmart. Construction Industry. Nagarajan. M and Sosic. G. Game Theoretic Analysis of Cooperation among Supply Chain Agents: Review and Extensions. Technical Report. Sauder School of Business, Univ. of British Columbia, Vancouver, Canada, August 2005. Bajari, P.L and McMillan, R S and Tadelis, S. Auctions versus Negotiations in Procurement: An Empirical Analysis. NBER Working Paper Series, Department of Economics, Stanford University, June 2003. Negotiation and bargaining are at the heart of cooperative game theory. Natural to apply “Negotiation and Bargaining” based mechanisms or cooperative 5 game theoretic techniques to “Procurement Network Formation” E-Commerce Lab, CSA, IISc Cooperative Game with Transferable Utilities (TU Games) T ( N , v) N {0,1,..., n} set of players v : 2 characteristic function ; v( ) 0 N C N is called a coalition. There are 2 | N | 1 possible coalitions 6 E-Commerce Lab, CSA, IISc Given a TU game, two central questions are: Which coalition(s) should form ? How should a coalition that forms divide the surplus among its members ? The second question has implications for answering the first question ! 7 Cooperative game theory offers several solution concepts: The Core Shapley Value Kernel Nucleolus E-Commerce Lab, CSA, IISc Divide the Dollar Game There are three players who have to share 300 dollars. Each one proposes a particular allocation of dollars to players. N {0,1,2} S 0 S1 S 2 {( x0 , x1 , x2 ) 3 : x1 0; x2 0; x3 0; x1 x2 x3 300} 8 E-Commerce Lab, CSA, IISc Divide the Dollar : Version 1 The allocation is decided by what is proposed by player 0 ui ( s0 , s1 , s2 ) xi 0 if s0 ( x0 , x1 , x2 ) otherwise Apex Game or Monopoly Game Characteristic Function v({0}) 300 v({1}) v({2}) v({1,2}) 0 v({0,1}) v({0,2}) v({0,1,2}) 300 9 E-Commerce Lab, CSA, IISc Divide the Dollar : Version 2 It is enough 0 and 1 propose the same allocation ui ( s0 , s1 , s2 ) xi 0 if s0 s1 ( x0 , x1 , x2 ) otherwise Players 0 1nd 1 are equally powerful; Characteristic Function is: v({0}) v({1}) v({2}) 0 v({0,1}) 300 v({0,2}) v({1,2}) 0 v({0,1,2}) 300 10 E-Commerce Lab, CSA, IISc Divide the Dollar : Version 3 Either 0 and 1 should propose the same allocation or 0 and 2 should propose the same allocation ui (s0 , s1 , s2 ) xi 0 if s0 s1 ( x0 , x1 , x2 ) or s0 s2 ( x0 , x1 , x2 ) otherwise Characteristic Function v({0}) v({1}) v({2}) v({1,2}) 0 v({0,1}) v({0,2}) v({0,1,2}) 300 11 E-Commerce Lab, CSA, IISc Divide the Dollar : Version 4 It is enough any pair of players has the same proposal ui ( s0 , s1 , s2 ) xi if s0 s1 ( x0 , x1 , x2 ) or s0 s2 ( x0 , x1 , x2 ) or s1 s2 ( x0 , x1 , x2 ) 0 otherwise Also called the Majority Voting Game Characteristic Function v({0}) v({1}) v({2}) 0 v({0,1}) v({0,2}) v({1,2}) v({0,1,2}) 300 12 E-Commerce Lab, CSA, IISc Shapley Value Shapley (1950) Gives a unique payoff allocation for a TU game (N, v) that describes a fair way of dividing the gains from cooperation. Axiomatically developed 13 Symmetry Carrier Linearity E-Commerce Lab, CSA, IISc Shapley Value : Expression Given : A Coalitional Game - (N , v) Shapley value (v) ( 0 (v),..., n (v)) where i (v ) 14 | C |!(| N | | C | 1)! {v(C {i}) v(C )} | N |! C N i E-Commerce Lab, CSA, IISc Shapley Value: Examples Version of Divide-the-Dollar Shapley Value Version 1 (300,0,0) Version 2 (150,150,0) Version 3 (200,50,50) Version 4 (100,100,100) 15 E-Commerce Lab, CSA, IISc The MPNF Game (Multi-Unit Procurement Network Formation Game) 16 E-Commerce Lab, CSA, IISc Supply Network for Automotive Stampings Master Coil Cold Rolling Pickling Slitting Stamping 1 2 3 4 2 3 4 5 6 7 6 7 Suppliers 17 E-Commerce Lab, CSA, IISc Procurement Feasibility Graph Cold – Rolled Stage Post-Picking Stage Post Slitting Stage Stamped Stage Finished Stage S T Master Coil Stage • Each vertex represents an intermediate state of the stamping • Each edge represents a value adding operation 18 E-Commerce Lab, CSA, IISc Master Coil Stage Supplier-ID, Per unit-cost, Capacity ColdRolled Stage • Capacity is specified as an upper-bound on the flow along the edge. 19 E-Commerce Lab, CSA, IISc MPNF Situation (G, N , , b, d ) G Procuremen t feasibility Graph (V,E ) N Set of Suppliers : E N edge ownership mapping b Valuation of the buyer for a single unit of the item d number of units demanded by the buyer An MPNF situation leads to an MPNF game in characteristic form, (N,v) 20 E-Commerce Lab, CSA, IISc MPNF Problem G (V , E ) ; N Set of Suppliers c(e) cost per unit flow of edge e l (e) minimum flow along edge e u (e) maximum flow along edge e f (e) flow on edge e (e) Supplier owning edge e I ( j ) Set of incoming edges at vertex j O ( j ) Set of outgoing edges at vertex j 21 E-Commerce Lab, CSA, IISc b valuation of the buyer for a single unit C of the item Coalition of the suppliers EC Set of edges owned by Suppliers in coalition C FC Flow in the netwok from S to T using only the edges in EC ( FC ) Suppliers in C who facilitate the flow FC 22 d quantity demanded by the buyer x decision variable, actual flow from S to T E-Commerce Lab, CSA, IISc Objective Function Maximize the surplus v(C) for C = N where v(C ) bx f ( e) c ( e) eEC bx Total value to buyer from a flow x f (e)c(e) Total cost to the suppliers in C to achieve the eEC required flow 23 E-Commerce Lab, CSA, IISc Constraints (1) Flow conservati on constraint s f(e) f(e) eI ( j ) Ec j V \ {S , T } eO ( j ) Ec (2) Flow constraint s f(e) x f(e) eI (T ) Ec eO ( S ) Ec (3) Demand constraint s 0 xd (4) Lower and upper bounds on flow l (e) f (e) u (e) e EC 24 E-Commerce Lab, CSA, IISc MPNF Cooperative Game • (N, v) • N = Set of all suppliers = {1,2,…,n} • v(C) = Maximum flow that can be derived using suppliers in coalition C CN Immediate Questions • What is the core of (N, v) ? • What is the Shapley value of (N, v)? 25 E-Commerce Lab, CSA, IISc Shapley Value of MPNF Games • Makes a positive allocation of surplus to agents who own edges in any flow that generates positive surplus • Need to understand conditions under which the Shapley value makes positive allocation to only agents involved in surplus maximizing flows • Need to understand conditions under which the Shapley value allocation is stable – Convex Games 26 E-Commerce Lab, CSA, IISc An example to show implications of using Shapley value Sample procurement networks (4, [0,2]) T s t (3, [0,2]) b = 5, d=2 a (2, [0,2]) s L (3, [0,2]) a (2, [0,2]) t (2, [0,2]) s (2, [0,2]) R (3, [0,2]) b (0, [0,2]) b = 5, d=2 t b = 5, d=2 Agent 1 Agent 2 27 3 Agent (4, [0,2]) E-Commerce Lab, CSA, IISc Characteristic function and Shapley value allocations Example cont’d: Network T S Network L Value v(S) S Value v(S) {i} 0 {i} 0 {1,2} 0 {1,2} 0 {1,B} 4 {1,B} 4 {2,B} 2 {2,B} 0 {1,2,B} 4 {1,2,B} 4 1(N,v) = 4/3 1(N,v) = 2 2(N,v) = 2/6 2(N,v) = 0 B(N,v) = 14/6 B(N,v) = 2 Network R S Value v(S) {i}, {1,2}, {1,3}, {2,3}, {2,B}, {3,B}, {1,2,3}, {2,3,B} 0 {1,B} 2 {1,2,B} 2 {1,3,B} 4 {1,2,3,B} 4 1(N,v) = 9/6; 2(N,v) = 1/6 3(N,v) = 5/6; B(N,v) = 9/6 Makes allocations to agents who are not critical to network formation 28 E-Commerce Lab, CSA, IISc Characterization of Shapley value allocations and ownership structure Definition 1: The set of all agents, denoted SM(F) who own edges in a surplus maximizing flow F of the procurement graph are called the SM-agents, i.e., SM = {i = ψ(e), i N: e ψ(F)} associated with the surplus maximizing flow F. Proposition: The characteristic function of the MPNF game with the buyer included as an agent is zero-monotonic, i.e., vb ( N ) vb ( N \ {i}) vb ({i}), i N 29 E-Commerce Lab, CSA, IISc Characterization of Shapley value allocations and ownership structure (cont’d) Theorem: If the Shapley value rule allocates all the surplus value in the MPNF game only to agents i SM then for every flow FS provided by a coalition S that includes an agent i SM, either: 1. FS is not profitable, i.e., vb(S) = 0 or 2. If FS is profitable, then we have (FS) SM ≠ 0, and there is a set SSM (FS) SM such that v(SSM) = v((FS) 30 E-Commerce Lab, CSA, IISc Implementation of the Shapley value of the MPNF game A non-cooperative way to the cooperative solution • The Shapley value is an exogenous viewpoint of a cooperative scenario. • We need a succinct game form that allows us to implement the Shapley value. • What do we mean by “implement the Shapley value”? • The Shapley value allocation vector must correspond to the Nash equilibrium of a non-cooperative (extensive form) game. • Implementation Theory has investigated this topic rigorously. • See: J Moore, Implementation, contracts, and renegotiation in environments with complete information. Chapter in Advances in Economic Theory, VI World Congress of the Econometric Society. 31 E-Commerce Lab, CSA, IISc Implementing the Shapley value of the MPNF game A high level description of the extensive form game • Actions in Stage 1: • Agents make their choices simultaneously. • Every agent chooses an action from the action set given by • A {(bi | bi R ; bij R} 1 |N| Net bid of agent i • Bi jN \{ i} bij b jN \{ i} Value that agent i is willing to transfer to agent j in order to become the proposer |N| dimensional bid vector of agent i ji • Agent with highest net bid chosen as the proposer 32 E-Commerce Lab, CSA, IISc Implementing the Shapley value of the MPNF game A high level description of the extensive form game • Actions in Stage 2: • Agent makes an offer (N, ) to form the coalition N and implement the outcome • Outcome consists of a reassignment of edge capacities and division of surplus among the agents in the coalition. • Actions in Stage 3: All agents other than the proposer either accept or reject the offer made by agent . • If offer is accepted than edge capacities and surplus is transferred as agreed otherwise the game is replayed without agent . • Result: This extensive form game implements the Shapley value of the MPNG game in sub-game perfect Nash equilibrium. 33 E-Commerce Lab, CSA, IISc Conclusion and Future work Investigated the interrelation between using the Shapley value as a solution concept for the MPNF game and the ownership structure of the procurement network. Also, developed a protocol to implement the Shapley value of the MPNF game. Future work: 34 Examine other solution concepts such as the Kernel and Nucleolus for the MPNF game. Investigate the incomplete information versions of this game. E-Commerce Lab, CSA, IISc Key References • T.S. Chandrashekar. Procurement Network Formation: A Cooperative Game Theoretic Approach. PhD Thesis, February 2007, CSA, IISc. • Roger B. Myerson. Game Theory: Analysis of Conflict. Harvard University Press. 1998 35 E-Commerce Lab, CSA, IISc Questions and Answers … Thank You … 36 E-Commerce Lab, CSA, IISc