A Game Theoretic Framework for Analysing Coopetition
J van Rooij ANR: 656090
Thesis written in conjunction with an internship at TNO, and submitted in partial fulfillment of the requirements for the degree of Master of Science, and of the Master’s program in Econometrics & Mathematical Economics at the Tilburg School of Economics and Management
Thesis committee: prof. dr. P.E.M. Borm (Supervisor) dr. J.H. Reijnierse (Chairman) dr. M. Groote Schaarsberg (Company supervisor)
TNO The Hague, The Netherlands
Tilburg University Tilburg School of Economics and Management Tilburg, The Netherlands
August 2015 [Page intentionally left blank] Management Summary
In this thesis we develop a game theoretic framework to analyse coopetition. Coopetition stems from a contraction of the words cooperation and competition. When firms are engaged in coopetition this means that the firms are competitors of each other that are working together to achieve a common interest. In real life there are many examples of such situations. One could think of the joint production facility that has been set up by PSA Peugeot Citroen and Toyota Motor Corporation to produce cars, or examples of joint delivery networks. Another example of coopetition is network sharing in the mobile telecommunications industry. In many countries there are several operators that own and operate a mobile network. Upgrading and maintaining such a network is costly, therefore operators are more and more looking for alternatives to reduce costs. One of these alternatives is network sharing: operators jointly invest in network upgrades. Examples of network sharing are already seen in for example Sweden. The theoretical framework that we develop can be used to analyse these scenarios. Our main interest is to find coalitions or coalition structures that are stable. To make predictions about the stability of coalitions we make use of two different stability concepts: individual and core stability. These concepts have to do with the incentives and possibilities for firms to deviate from a certain coalition structure. The framework combines concepts from both cooperative and noncooperative game theory. One of the main advantages of the framework is that it can be used to analyse scenarios with a large (up to 7 or 8) number of players. This is an advantage over many non-cooperative models that can only analyse scenarios with 2 or at most 3 players. Furthermore the framework can be used in a pragmatic way and can be easily adopted to fit different scenarios. Using this framework we analyse theoretical examples of coopetition in a Cournot oligopoly where firms can reduce their marginal costs by working together. We find several results that are of interest. First of all we show that the stability of coalition structures greatly depends on how the firms divide the surplus that arises from cooperation. We also find that coalition structures where all the firms but one work cooperate with each other are often individual stable. The coalition of size n − 1 does not want the single firm to join their coalition. Core stable coalition structures most of the time have a similar structure. Furthermore we will show that getting all the firms to work together is often hard. Due to the competitive nature the gain of one firm often comes at the cost of the loss of another. Finally we show how the framework can be used in a more pragmatic way to analyse network sharing in the mobile telecommunications industry. We discuss an example regard- ing the Dutch market with three operators. We find that the most likely coalition structure to form is where the two smallest operators invest together and the largest operator invests on its own. Acknowledgements
This thesis marks the end of a period of 9 years as a student. A long ride of which in the end I enjoyed every single bit. During those 9 years, and over the course of my internship at TNO I have got to learn many people that I am grateful for. First of all I am grateful for my supervisor Peter Borm for his excellent guidanc and his efforts in keeping me focussed on the subject instead of diverting and wandering through the beautiful realm of game theory and coalition formation theory. Thank you also for making sure that I actually finished this thesis. I also owe many thanks to my company supervisors Mirjam Groote Schaarsberg and Frank Berkers. Mirjam, thank you for giving me the opportunity to do an internship at TNO and providing me with all the space to pursue my own ideas and to explore all that TNO has to offer. Frank, thank you for always making time to discuss my ideas. I enjoyed every single conversation and discussion. You have always challenged me to look beyond the theory. To all the wonderfull people at the Strategic Business Analysis department of TNO: thank you for making my internship such a pleasant stay with all the outdoor lunches, coffee breaks and improv workshops. Special thanks need to go to my fellow interns Marieke and Tim for supplying me with the occasional motivational coffee break or peptalk. To my friends in Tilburg: I cannot name you all, nonetheless, thanks for everything. Arnoud, Gijs, thank you for being part of my econometrics dream team. You guys kept me going! Marleen, you know all too well how many thanks I owe you. Mom, dad, thank you for always being there. During the past 9 years as a student you have always been there to support me in whatever way possible, I cannot explain in words how grateful I am. I know from time to time you have had your doubt whether or not there would ever come an end to this expedition. Well, to take away the last of them: here it is. Contents
List of Figures
List of Tables
1 Introduction 1
2 Coopetition Framework 3 2.1 Introduction ...... 3 2.2 Overview ...... 4 2.3 Preleminaries and notation ...... 4 2.4 Coopetition Game ...... 5 2.5 Cooperation ...... 6 2.6 Competition ...... 10 2.7 Stability ...... 12 2.8 Computational Complexity ...... 14
3 Cournot Oligopoly: A Theoretical Example 16 3.1 Introduction ...... 16 3.2 Cournot Coopetition Game ...... 16 3.3 Examples ...... 17 3.3.1 Example 1: One Large Firm, Three Small Firms ...... 17 3.3.2 Example 2: Economies of Scale ...... 22 3.4 Beyond a Division Rule ...... 24
4 Case Study: Network Sharing In The Mobile Telecommunications Indus- try 26 4.1 Introduction ...... 26 4.2 The Competition Model ...... 27 4.2.1 Investment in LTE ...... 27 4.2.2 Multinomial Logit Demand Model ...... 29 4.2.3 Linking the LTE Investments and the Logit Demand Model Together 31 4.3 Scenarios ...... 32 4.3.1 Scenario 1: The Dutch Market with 3 Operators without an Outside Good...... 32 4.3.2 Scenario 2: The Dutch Market with 3 Operators and an Outside Good 35
5 Conclusions 38
6 Further Research 39
A Bibliography 40
B Mathematical Proofs 42 B.1 Derivation of the Cournot-Nash equilibrium ...... 42 B.2 Proof of Proposition 1 ...... 43 B.3 Proof of Proposition 2 ...... 45
C Matlab Code 47 C.1 ChiValue.m ...... 47 C.2 StabilityChecker.m ...... 48 C.3 CoreStabilityChecker.m ...... 51 C.4 CournotProfit.m ...... 53 C.5 CournotExamples.m ...... 54 C.6 MarketShareSolver.m ...... 55 C.7 BULRIC.m ...... 56 C.8 NewSMARTModel.m ...... 56 List of Figures
1 Schematic overview of the general coopetition model ...... 7 2 Schematic overview of the coopetition model combined with a CS-value . .8 3 Schematic overview of the coopetition model combined with a CS-value and stability concepts ...... 14 4 4 Player Cournot coopetition game: initial configuration ...... 18 5 4 Player Cournot coopetition game: {1234} ...... 19 6 Dutch market with three operators without an outside good ...... 34 7 Dutch market with three operators and an outside good ...... 37
List of Tables
1 Four player Cournot example ...... 19 2 Three player Cournot example: Small and large economies of scale . . . . . 24 3 Coverage and capacity sites ...... 28 4 Dutch market with three operators without an outside good ...... 34 5 Market share of the outside good ...... 35 6 Dutch market with three operators with an outside good ...... 37 [Page intentionally left blank] 1 Introduction
The research for this thesis was conducted as part of an internship at TNO in the Strategic Business Department. TNO is the largest research institution in the Netherlands with over 3,000 employees. Originally this thesis would be connected to the SMART 24 project that was conducted by TNO and commissioned by the European Commission. The SMART 24 project was aimed at analysing the effects of network sharing scenarios in the mobile telecommunications industry. Unfortunately the SMART 24 project was aborted prema- turely due to circumstances. This resulted in this thesis more deviating from modelling the telecommunications market to a theoretical and conceptual treatment of coopetition in general. However, we did apply the framework to network sharing scenarios. Our analysis of analysis of these network sharing scenarios uses many ideas and concepts from Offergelt (2011) who also wrote a thesis In a competitive environment firms are often faced with large investments or high costs to keep up with trends in the market. One could think of examples such as investing in a R&D project, building production facilities or investing in a delivery network. A solution to overcome these high costs and save money would be to jointly initiate such projects together with one or more competitors. There indeed exist real life examples where several competitors jointly established a R&D project, build a joint production facility or jointly set up a delivery network. In this thesis we will study such cooperative relations between competitors using game theory. The general idea is that cooperation takes place at a lower level than competition. At this lower level firms can act collectively to create a win-win relationship, but in the market they remain competitors. A situation where competitors are engaged in a cooperative relation is called coopetition. The term coopetition was first introduced by Brandenburger and Nalebuff (1996). In particular we will concentrate on oligopolistic situations where each firm is likely to be aware of the actions of the others. The decisions of one firm therefore influence and are influenced by the decisions of other firms. Given this oligopolistic nature of the market we believe that game theory is the right tool to analyse the coopetitive situation. We will be combining concepts from bot cooperative and non-cooperative game theory since a coopetitive situation entails both cooperation and competition. Eventually we want to be able to predict which coalitions will form. To do this we make use of different stability concepts to assess whether or not certain coalitions or coalition structures are stable. Bogomolnaia and Jackson (2002) and Banerjee et al. (2001) both define stability concepts for hedonic games which we will generalise to fit within our coopetition framework. For TNO it would also be interesting to use this model for large multi-stakeholder innovation projects. In these projects the interests of many stakeholders need to be aligned in order to get them to cooperate. The coopetition framework could provide insight in the considerations of these stakeholders. The buildup of this thesis is as follows. In Chapter 2 we formally will introduce our game theoretical framework for analysing coopetition. To our knowledge this is a completely new model. The definition of a coopetition game will be introduced in this chapter and we will present a method to solve such games using different stability concepts. Chapter 3 takes the framework developed in chapter 2 and applies this approach to the theoretical example of a Cournot oligopoly. In this oligopoly the firms can reduce their marginal costs of production
1 by working together. In chapter 4 we will use our framework in a more pragmatical way to take a look at network sharing in the mobile telecommunications industry. Finally the conclusions are summarised in chapter 5 and chapter 6 points out directions for further research.
2 2 Coopetition Framework
2.1 Introduction A coopetitive situation is different from a cooperative relationship between actors who are not rivals of each other. When a firm saves money because it jointly invested in a production facility these savings can be invested in other activities such as marketing or product development. By doing so the firm strengthens its strategic position. This also means that when a firm works together in a coalition the position of other members of the coalition will be strengthened. When rivals decide to cooperate there are a lot of effects that play a role: Not only direct effects such as the savings itself, but also strategic effects because a firm is able to produce at a lower price. These effects depend on how well firms are able to convert the benefits of a cooperative relationship into strategic advantage. It could happen that a cooperative deal seems to be fair and justifiable at the lower level but has devastating effects for one of the involved firms because all other partners are far better at converting the gains from the collaboration into strategic advantage. The contrary could also be true, a firm might agree on a deal that seems unreasonable at the lower level because otherwise it would be even worse off. Bengtsson and Kock (2000) argue that the most complex, but also the most advantageous relationship between competitors is coopetition. The complexity arises due to the fundamentally different and contradictory logics of interaction that competition and cooperation are built on. We are analysing situations in which firms are engaged in competition with each other. Formally we define competitors as actors that produce and market the same product. Firms try to use the resources available to them in the best way possible to maximise their own interests. The different self-interests are in conflict with each other, which in consequence means that firms compete against each other to best fulfil their own self-interests. A firm’s payoff will be determined by its access to resources, how well it can convert these resources in value and the characteristics of the other firms it is competing with in the market. The concept of cooperation is based on a diametrically opposite assumption; individuals participate in collective actions to achieve common goals. We will think of cooperation as a way of working together - on a lower level than competition - to gain access to more resources or to be able to use already available resources in a more efficient way. We will be more specific regarding what we mean with ‘working together at a lower level than competition’. Let us start by stating what we do not mean by it. We do not consider the formation of cartels where firms maximise their joint profit and redistribute this among the members of the cartel. Regardless of the coalition a firm belongs to, it acts in such a way only to maximise its own interests and not those of the coalition. Firms are not allowed to transfer their individual profits. A good example of such a coopetitive situation is the consortium of PSA Peugeot Citroen and Toyota Motor Corporation. This consortium owns a joint production facility that produces three cars that are, in essence, the same; the Citroen C1, Peugeot 107, and Toyota Aygo. While the cars are produced in the exact same factory and only differ in cosmetics, Toyota and PSA remain competitors in the market. The firms need to negotiate how to split the costs and allocate the capacity of this facility. However, in the market they will compete as fiercely with each other as they would do with other competitors. Since the firms are not allowed to transfer profits, the only compensations they can make are at the lower level by paying more for the joint
3 facility or allocating more capacity to a competitor. Summing up the above discussion we can distinguish two key features of the coopetitive situations we will study: 1. Cooperation takes place at a lower level than competition. Individual profits are nontransferable. Side payments can only be made at the lower level. 2. Cooperation has strategic effects. Market conditions change depending on the coali- tion structure that is formed and the allocation at the lower level. The next sections will discuss our approach in more detail. We explain the theoretical foundations and introduce definitions and notation.
2.2 Overview In this chapter we will introduce our approach to model coopetition and formally define of a coopetition game. Our model links concepts from both cooperative and non-cooperative game theory. In a coopetition game the firms are basically playing two games at once. At the market level they are competing with each other for the same group of customers and try to maximise their individual profits. However, at a lower level the firms are cooperating with each other in order to exploit synergies and economies of scale. The competition of the firms in the market will be modelled using a non-cooperative, strategic game whereas the cooperation at the lower level will be modelled using a cooperative, coalitional game. The approach assumes that competition, i.e. the strategic game, takes place in the short run, while cooperation, i.e. the coalitional game, occurs more slowly or only occasionally. Therefore we can treat cooperation and the resulting outcome as prior to the determination of each firm’s strategy in the competitive stage. The two different stages are linked in the sense that the outcome of the coalitional (cooperative) game defines the strategic (competitive) game that the firms will play. In a way firms are choosing which strategic game they play in the market by means of cooperation at a lower level. Our main interest is to find out which coalition structures are stable and which are not (with respect to different notions of stability). Unstable coalition structures are less likely to form since there will be incentives and possibilities for firms to deviate.
2.3 Preleminaries and notation Let N = {1, . . . , n} denote the set of firms.. A coalition is a nonempty subset S ⊆ N and the set 2N is the collection of all coalitions. A coalition structure is a collection 1 m k l m k P = {S , ..., S } of coalitions such that S ∩ S = ∅ for all k 6= l and ∪k=1S = N. The set P shall be the collection of all coalition structures of N, and for a coalition S we denote the set of all coalition structures containing S by PS. For i ∈ N and a coalition structure P we denote by P (i) the (unique) coalition S ∈ P with i ∈ S. To condense notation, we shall drop the braces around coalitions in coalition structures, for example, denote {{1}, {2, 3}} by {1, 23}. In case we have variables that depend on a coalition structure P we use the superscript 1 to refer to the coalition structure P = {1, . . . , n} (using the 1 condensed notation), for example πi = πi({1, . . . , n}) refers to the profit of firm i in the absence of any cooperation. Usually we will call this the initial profit of firm i or the profit of firm i in the initial configuration.
4 2.4 Coopetition Game At the start of the game there is no cooperation. Each firm starts with a bundle of resources which might be assets such as intellectual property, production technology, government permits, machinery or simply cash. In a competitive market firms try to use their resources as efficiently as possible in order to maximise their individual profits. We will assume that there is only one type of resources and call this the capital stock of a firm. The vector 1 k = (k1, . . . , kn) denotes the capital stock of all firms i ∈ N, and k denotes the initial level of capital stock. Based on the initial level of capital stock, k1, the firms play a strategic 1 1 1 1 1 game G which is given by G = {(Xi , πi )}i∈N where for each firm i ∈ N, Xi denotes the 1 Q 1 strategy space and πi : j∈N Xj → R the payoff function. We will assume that the game G1 has a unique Nash equilibrium. In mathematical terms this means that there exists a 0 Q 1 xˆ ∈ j∈N Xj such that 1 1 1 1 1 πi (ˆx ) ≥ πi (xi , xˆN\{i}) 1 1 for all xi ∈ Xi , and all i ∈ N. In a Nash equilibrium there are no unilateral deviations that are profitable for the firms. Therefore the result of competition between the firms will be this equilibrium. This is nothing new, we follow a classic approach to modelling competition between firms. One could for example think of a Cournot oligopoly in which the cost functions or the maximum production capacity of the firms depend on the level of capital stock. So both the payoff functions (cost functions) and the strategy spaces (maximum production capacity) of the firms can be functions of the level of capital stock. At some point in time it becomes possible for the firms to cooperate with each other. By cooperating the firms can exploit synergies or economies of scale. In this way they can gain access to more capital stock then they would on their own. The firms need to decide two things: who to cooperate with and how to split the surplus of capital stock arising from their cooperation. We will model this cooperative stage using a coalitional game (N, v) where N, as before, is the set of firms and v : 2N → RN a characteristic function describing the amount of capital stock each coalition can acquire. It is assumed that v ≥ 0 and v(∅) = 0. The outcome of this coalitional game will be an allocation configuration which is defined as follows:
Definition 1. Let (N, v) be a coalitional game. The outcomes of this game can be rep- resented by an allocation configuration. An allocation configuration is a pair α = (P, k) where P ∈ P is a coalition structure of N and k ∈ RN is an allocation vector. Given an allocation configuration α = (P, k) the coalition structure P describes which coalitions have formed and the allocation vector k denotes how the firms split the surplus of capital stock arising from their cooperation. Capital stock is transferable within, and non- transferable between coalitions in a coalition structure. Firms use capital stock efficiently, which means that they use all the capital stock that is available to them. Therefore we re- quire an allocation configuration to be component efficient. An allocation vector k is called P N component efficient with respect to P if i∈S ki = v(S) for all S ∈ P . Let KP (N, v) ⊂ R denote the set of all component efficient allocation vectors with respect to P for (N, v). An allocation configuration α = (P, k) is called component efficient if k is component efficient with respect to P . Let A(N, v) ⊂ RN×P denote the set of all component efficient allocation configurations for (N, v).
5 When firms cooperate at a lower level this changes the competition in the market. By acquiring more capital stock firms might be able to produce at a lower cost, sell new products or improve their marketing strategy. Somehow the outcome of the coalitional game must be linked to a strategic game. In a coopetition game we propose to link the two different stages by a mapping Γ which assigns to each component efficient outcome α = (P, k) of the coalitional game (N, v) a strategic game Gα = Γ(α). As mentioned before both the strategy spaces and the payoff functions can depend on the level of capital stock. Therefore they also both depend on the allocation configuration α, so Gα = Γ(α) = α α {(Xi , πi )}i∈N . Now we have discussed the different ingredients of a coopetition situation we are ready to give a formal definition.
Definition 2. A coopetition game is a tuple (N, v, Γ) where
(a) N = {1, . . . , n} is a set of firms;
(b)( N, v) is a coalitional game, v : 2N → RN is a characteristic function, and the out- comes of this game are represented by component efficient allocation configurations α = (P, k) ∈ A(N, v);
(c) Γ is a mapping from the space A(N, v) of component efficient allocation configurations of (N, v), to the space of strategic games. Each outcome α = (P, k) ∈ A(N, v) of the α α α coalitional game (N, v) maps to a strategic game G = {(Xi , πi )}i∈N = Γ(α); and (d) for each α = (P, k) ∈ A(N, v) the strategic game Gα has a unique Nash equilibrium α Q α α α α α α α α xˆ ∈ j∈N Xj such that πi (ˆx ) ≥ πi (xi , xˆN\i) for all xi ∈ Xi and all i ∈ N. Note that a coopetition game is a generalisation of both the strategic-form non-cooperative and TU cooperative game models. In the absence of competition the model reduces to a coalitional game and in the absence of cooperation the model reduces to a strategic game. Figure 1 shows a schematic overview of our model. In a coopetition game firms both cooperate and compete with each other. Cooperation at a lower level results in different allocation configurations that describe which firms work together and how they divide the proceeds of their cooperation. In general here are many possible allocation configurations. Each allocation configuration α = (P, k) in turn maps to a strategic game Gα = Γ(α). This strategic game describes how the firms compete with each other in the market. Different forms of cooperation lead to different forms of competition: in the coalitional game the firms choose which strategic game they will play in the market. Each strategic game Gα has a unique Nash equilibrium. The result of competition will be this equilibrium and the firms will receive their equilibrium payoffs. In the next two sections that follow we will discuss both the cooperative and competitive stage and their complexities in more detail. We will show how the number of possible allocation configurations can be reduced and how we can use the competitive stage to rank the different outcomes of the cooperative stage.
2.5 Cooperation In this section we will focus on the cooperative part of a coopetition game. The complexities of this stage will be discussed and we will propose a way of dealing with those complexities. By cooperating with each other at a lower level than competition, firms can gain access
6 Coopetition N, v, Γ Situation
Allocation α = (P, k) with ··· ··· Configuration k ∈ KP (N, v)
Strategic Game ··· Gα ···
Equilibrium ··· πα(ˆxα) ··· Payoff
Figure 1: Schematic overview of the model. The starting point is a coopetition game (N, v, Γ). In the cooperative stage firms play a coalitional game (N, v) to exploit synergies and acquire additional capital stock. The outcome of this coalitional game is an allocation configuration α = (P, k). For each coalition structure P ∈ P the corresponding allocation vectors k are component efficient, l α k ∈ KP l (N, v). Based on the outcome α the firms play a strategic game G = Γ(α) in the competitive stage. Therefore different outcomes of the coalitional game lead to different stategic games. The strategic game Gα has a unique Nash-equilibrium, xˆα, therefore the result of the competitive stage will be a vector of payoffs πα(ˆxα). to more resources than they would be able to individually. The question that needs to be answered in the cooperative stage is the following: which coalitions will form and how will their members split the capital stock? Each coalition can gain acces to an amount of capital stock which depends on the members of the coalition. This relation is given by a characteristic function v : 2N → R. We can also incorporate the initial level of capital stock, k1, in this characteristic function 1 by simply defining v({i}) = ki . Whether or not to do this depends on the context. As explained before, the firms play a coalitional game (N, v) and the outcomes of this game will be allocation configurations denoted by α = (P, k) where P represents the coalition structure that forms and k is a vector describing how capital stock is allocated. Since we assumed that capital stock is transferable within and nontransferable between coalitions and that firms use the available capital stock efficiently, the outcomes of the coalitional game need to be component efficient. The set P = {P 1,...,P m} denotes all possible coalition structures.1 The coalition structure of only singletons will be denoted by P 1 = {{1},... {n}}. When P 1 forms there is no cooperation and firms will only be able to use their initial level of capital stock. Therefore the allocation configuration will be α = (P 1, k1). For all other coalition structures P l, with l ∈ {2, . . . , m}, the firms are free in negotiating the division of capital stock within their coalition. Therefore in general multiple allocation vectors are possible as
1 m = Bn, and Bn is the Bell number which counts the number of partitions of a set that has exactly n elements.
7 l long as they are component efficient with respect to the coalition structure, k ∈ KP l (N, v). The freedom to choose any allocation vector, as long as it is component efficient, makes the model complicated. In principle there is an infinite amount of allocation configurations that can be the result of the cooperative stage. Given a coalition structure we do not know what the corresponding allocation vector will be. Therefore we would like to simplify the model by fixing the allocation of capital stock given a coalition structure. Fixing the allocation also allows us to overcome possible problems concerning bargaining. Take for example a three player coopetition game and suppose that the coalition structure {1, 23} forms. In this case it might be that firm 1 has a very low payoff and would like to propose a new allocation configuration in which it works together with either firm 2 or firm 3. In this case firm 1 only needs to propose a deal in which it is better off than in the current situation even if the deal seems to be unfair. This will result is the formation of the coalition structure {12, 3} and now firm 3 finds itself in the same position as firm 1. Later in section 2.4 we will see that in some examples we can lift this restriction, but for now we will keep the allocation of capital stock fixed. One way of fixing the allocation of capital stock would be to specify a value for games with a given coalition structure (henceforth CS-values and CS-games). Such games and values were first introduced by Aumann and Dreze (1974). A CS-value is an operator ϕ that assigns allocation vectors to all CS-games, ϕ(N, v, P ) ∈ RN . Basically it is the same as a solution concept for TU-games (such as the Shapley value or the Nucleolus) however this time it is not the grand coalition that forms but the coalition structure P . All allocation configurations will be of the form α = (P, ϕ(N, v, P ) for all P ∈ P. Instead of a manifold of allocation vectors there we use a CS-value to fix just one allocation vector for each coalition structure. Figure 2 shows how this changes the model. Instead of an infinite l amount of branches at each coalition structure (one branch for every k ∈ KP l and each l ∈ {1, . . . , m}) there is just one branch per coalition structure.
Coopetition Situation (N, v, Γ),ϕ + CS-value Allocation α1(ϕ) = (P 1, k1) with αm(ϕ) = (P m, km) with ··· Configuration k1(ϕ) = ϕ(N, v, P 1) km(ϕ) = ϕ(N, v, P m)
Strategic Game Gα1(ϕ) ··· Gαm(ϕ)
Equilibrium α1(ϕ) ··· αm(ϕ) Payoff π π
Figure 2: By combining a coopetition situation (N, v, Γ) with a CS-value ϕ ww can simplify the model. Instead of an infite amount of branches there now is a finite amount of branches, one for each coalition structure P ∈ P.
In this thesis we will use two CS-values in particular. Both values are based on the well known Shapley value. The first CS-value is the one introduced by Aumann and Dreze
8 (1974) (henceforth AD-value) which assigns to each player in a coalition structure element the Shapley value of the subgame restricted to that element:
ADi(N, v, P ) = Shi(P (i), v|P (i)), where Sh(N, v) is the Shapley value of (N, v).2 The second CS-value we will investigate is the χ-value of Casajus (2009). This value uses the Shapley value as a yardstick to distribute the allocation within a structural coalition. It compares the sum of the Shapley allocations in a coalition with the worth of that coalition; the difference, positive or negative, is distributed evenly. The χ-value is defined as: P v(P (i)) − Shj(N, v) χ (N, v, P ) = Sh (N, v) − j∈P (i) . i i |P (i)|
Note that in theory the χ-value could lead to a negative allocation of capital stock. This would mean that a firm pays another firm to cooperate. The intuition behind the χ- value is the following. Suppose that we believe that the Shapley value does a good job at distributing the worth of the grand coalition, v(N). If a group of players decides to split off, these players should all win/lose an equal amount since much. Similar to solution concepts for TU-games, CS-values can be characterised using prop- erties and axioms. We list the following properties from Casajus (2009) and refer the reader for more properties and an in-depth review to his paper:
Axiom 1. (Additivity, A). ϕ(N, v + v0,P ) = ϕ(N, v, P ) + ϕ(N, v0,P ) for all characteristic functions v, v0.
The first axiom states that if there are two CS-games with the same coalition structure P that are being played, the sum of the CS-values for both games should equal the CS-value of the game that would consist of putting those two games together. For the second axiom we need to define what symmetric players are. Two players i and j are called symmetric if v(S ∪ {i}) = v(S ∪ {j}) for all S ∈ 2N \{i, j}. That is, we can change one player for the other in any coalition that contains one of the players and not change that coalitions worth. The next axiom captures the idea that symmetric players should be treated equal when they are part of the same coalition in a coalition structure.
Axiom 2. (Component Restricted Symmetry, CS). If i, j ∈ N are symmetric and j ∈ P (i), we have ϕi(N, v, P ) = ϕj(N, v, P ). Earlier we already defined component efficiency for an allocation configuration. We can do the same for a CS-value. When a CS-value is component efficient this relates to the idea that the coalitions are the productive units; the players within a component work together in order to generate that coalition’s surplus. P Axiom 3. (Component Efficiency, CE). For all S ∈ P , we have i∈S ϕi(N, v, P ) = v(S).
2 P |S|!(|N|−|S|−1)! The Shapley value is defined as: Shi(v) = S∈2N :i/∈S |N|! (v(S ∪ {i}) − v(S)).
9 For the fourth and fifth axiom we again need to define a new type of player, a Null player. Formally a Null player i satisfies v(S ∪ {i}) = v(S) for every S ∈ 2N \{i}. A Null player never contributes anything to a coalition. The following axiom states that therefore this player should also receive nothing.
Axiom 4. (Null Player, N). If i ∈ N is a Null player, then ϕi(N, v, P ) = 0. In some situations the previous axiom might be too restrictive. Therefore this axiom can be replaced with the following axiom which only states that a Null player should recieve nothing when the grand coalition forms.
Axiom 5. (Grand Coalition Null Player, GN) If i ∈ N is a Null player, then ϕi(N, v, {N}) = 0.
A coalition structure P 0 ∈ P is finer than P ∈ P if P 0(i) ⊆ P (i) for all i ∈ N. Intuitively this means that P 0 can be obtained by splitting the elements of P . The final axiom now states that if a group of players within a coalition decides to leave that coalition they should all win/lose the same amount.
Axiom 6. (Splitting, SP). If P 0 is finer than P then for all i ∈ N and j ∈ P 0(i), we have
0 0 ϕi(N, v, P ) − ϕ(N, v, P ) = ϕj(N, v, P ) − ϕj(N, v, P ).
Using the axioms defined above it is possible to (uniquely) characterise both the AD- value and the χ-value. The following two theorems do so. For the proofs of these theorems we refer to the original work.
Theorem 1 (Aumann and Dreze (1974)). The AD-value is the unique CS-value that satisfies A, CS, CE and N.
Theorem 2 (Casajus (2009)). The χ-value is the unique CS-value that satisfies A, CS, CE, GN and SP.
Both the AD-value and the χ-value are based on the Shapley value and whenever the Grand Coalition forms they both distribute the Shapley value amongst the players. Now what is the main difference between both values? Casajus (2009) points out that the AD- value fails to account for outside options. It ignores the fact that a firm could also form a coalition with firms that are not in present in its current coalition structure element since it only uses information from the subgame restricted to that coalition structure element. We are interested in whether or not different CS-values lead to different outcomes and different stable coalition structures. One remark considering the initial level of capital stock k1 and the normalisation of v. Since we are considering two CS-values that both satisfy additivity we can choose whether or not to include k1 in v since ϕ(v + k1) = ϕ(v) + ϕ(k1). So in some cases we will choose to zero-normalise v and explicitly specify k1.
10 2.6 Competition Once alliances have been made and capital stock has been divided firms will try to maximise their individual profits by competing with each other in the market. Based on the outcome α = (P, k) of the coalitional game (N, v) the firms play a non-cooperative strategic game α α α α G = {(Xi , πi )}i∈N = Γ(α). When playing G firms independently and simultaneously choose their strategies. Furthermore they make their choices irrespective of the coalition they belong to. This means that they compete as fiercely with their fellow coalition mem- bers as they do with other firms. We will assume that the game Gα has a unique, strong Nash equilibrium for every allocation configuration α ∈ V (N, v).3 In mathematical terms α Q α this means that for every α there exists ax ˆ ∈ j∈N Xj such that
α α α α α πi (ˆx ) ≥ πi (xi , xˆN\{i})
α α for all xi ∈ Xi , and all i ∈ N. Since for every allocation configuration there exists a unique Nash equilibrium the strategic game reduces to a payoff function. The assumption of a unique Nash equilibrium for every allocation configuration also allows us to use our framework in a more pragmatic way. In many situations competition may be hard to model. Specifying strategy spaces and interactions between firms might be extremely complicated or even impossible. However, there might be analysts or other experts who are able to predict what will happen to the market and the payoffs of the firms in different situations. Most likely their models are not formulated in terms of strategic games and Nash equilibria. We could use the models of these experts to calculate the payoffs for different allocation configurations while keeping all the other steps in our analysis the same. Basically we assume that the experts are able to capture the complex, competitive interactions between the firms in their models without explicitly specifying all the game theoretic specifics. In this way our approach remains useful even if the competitive stage is not formulated as a strategic game with a unique Nash equilibrium. In that case the boxes containing the strategic games in figure 1 and 2 could be viewed as ‘black boxes’ that model the competition in the market. In a later chapter of this thesis we will discuss such an example. Since firms seek to maximise their individual profits the outcome of the second stage can be used to rank the different outcomes of the first stage. The payoff function provides us with a valuation over the different allocation configurations. A firm will prefer one configuration, α = (P, k) over another configuration α0 = (P 0, k0) if and only if it obtains a higher payoff in the competitive stage of the game. This can be formalised in mathematical terms: 0 α α α0 α0 α i α ⇐⇒ πi (ˆx ) ≥ πi (ˆx ), 0 0 where α i α means that player i thinks that the configuration α is at least as good as α . As explained in the previous section we can apply different CS-values, ϕ, to the coalitional game that the firms play in the cooperative stage. For each coalition structure the division of capital stock will then be fixed by ϕ, and outcomes of the cooperative stage will always
3This uniqueness assumption is common in most models in the literature. If the second-stage game has multiple Nash-equilibria the analysis would become problematic. In this case the preferences of the firms (and stability of configurations) would depend on their expectations.
11 be of the form α = P, ϕ(N, v, P ). Instead of ranking allocation configurations we are ϕ 4 now able to derive a profile of preferences (i )i∈N over all coalition structures P ∈ P:
ϕ 0 α α α0 α0 P i P ⇐⇒ πi (ˆx ) ≥ πi (ˆx ),