Compensations in the Shapley value and the compensation solutions for graph games

Description

Compensations in the Shapley value and the compensation solutions for graph games? Sylvain Béal†, Eric Rémila‡, Philippe Solal February 24, 2010 Abstract We consider an alternative expression of the Shapley value that reveals a system of compensations: each player receives an equal share of the worth of each coalition he belongs to, and has to compensate an equal share of the worth of any coalition he does not belong to. We give an interpretation in terms of formation of the grand coalition according to an ordering of the players and define the corresponding compensation vector. Then, we generalize this idea to cooperative games with a communication graph. Firstly, we consider cooperative games with a forest (cycle-free graph). We extend the compensation vector by considering all rooted spanning trees of the forest (see Demange [3]) instead of orderings of the players. The associated allocation rule, called the compensation solution, is characterized by component efficiency and relative fairness. The latter axiom takes into account the relative position of a player with respect to his component. Secondly, we consider cooperative games with arbitrary graphs and construct rooted spanning trees by using the classical algorithms DFS and BFS. If the graph is complete, we show that the compensation solutions associated with DFS and BFS coincide with the Shapley value and the equal surplus division respectively. Keywords: Shapley value, compensations, relative fairness, compensation solution, DFS, BFS, equal surplus division.

• communication graph


• over all


• all rooted


• games can


• rooted spanning


• such algorithms


• graph games


• coalition


• component should


• compensation solution

Sujets

• University of Caen Lower Normandy
• Lardon
• Université de Lyon
• Remila

Informations

• Publié par: pefav
• Nombre de lectures: 22
• Langue: English

Extrait

CompensationsintheShapleyvalueandthecompensation solutionsforgraphgames ∗ SylvainBéal † ,EricRémila ‡ ,PhilippeSolal § February24,2010 Abstract WeconsideranalternativeexpressionoftheShapleyvaluethatrevealsasystemof compensations:eachplayerreceivesanequalshareoftheworthofeachcoalitionhe belongsto,andhastocompensateanequalshareoftheworthofanycoalitionhedoes notbelongto.Wegiveaninterpretationintermsofformationofthegrandcoalition accordingtoanorderingoftheplayersanddefinethecorrespondingcompensation vector.Then,wegeneralizethisideatocooperativegameswithacommunication graph.Firstly,weconsidercooperativegameswithaforest(cycle-freegraph).We extendthecompensationvectorbyconsideringallrootedspanningtreesoftheforest (seeDemange[3])insteadoforderingsoftheplayers.Theassociatedallocationrule, calledthecompensationsolution,ischaracterizedbycomponentefficiencyandrelative fairness.Thelatteraxiomtakesintoaccounttherelativepositionofaplayerwith respecttohiscomponent.Secondly,weconsidercooperativegameswitharbitrary graphsandconstructrootedspanningtreesbyusingtheclassicalalgorithmsDFSand BFS.Ifthegraphiscomplete,weshowthatthecompensationsolutionsassociated withDFSandBFScoincidewiththeShapleyvalueandtheequalsurplusdivision respectively. Keywords:Shapleyvalue,compensations,relativefairness,compensationsolution, DFS,BFS,equalsurplusdivision. JELClassificationnumber:C71. 1Introduction TheShapleyvalue(Shapley[17])isthemoststudiedallocationruleforcooperativegames withtransferableutility(TU-gameshenceforth).OnewaytointerprettheShapleyvalue consistsinconsideringequallylikelyorderingsofplayers.Foreachsuchordering,theplayers enterabargainingroomonebyone,anduponenteringeachplayerispaidhismarginal contribution.Thisprocedureyieldstheso-calledmarginalvector,andtheShapleyvalueis theaverageoverallorderingsoftheplayersofthemarginalvectors. ∗ TheauthorsaregratefultoTheoDriessen,AmandineGhintran,AnnaKhmelnitskayaandAymeric Lardonforstimulatingdiscussions.WehavealsobenefitedfromcommentsofseminarparticipantsatCaen UniversityandatthefirstMINTworkshop.FinancialsupportbytheComplexSystemsInstituteofLyon (IXXI)andtheNationalAgencyforResearch(ANR)–researchprogram“ModelsofInfluenceandNetwork Theory”ANR.09.BLANC-0321.03–isgratefullyacknowledged. † Correspondingauthor.UniversitédeSaint-Etienne,CNRSUMR5824GATELyonSaint-Etienne, France.E-mail:[e-mail]:(+33)(0)4.77.42.19.68.Fax:[phone]. ‡ UniversitédeLyon,LIP,UMR5668CNRS-ENSLyon-UniversitéLyon1,France.E-mail: [e-mail].Tél:[phone].Fax:[phone] § UniversitédeSaint-Etienne,CNRSUMR5824GATELyonSaint-Etienne,France.E-mail: [e-mail]:(+33)(0)4.77.42.19.61.Fax:[phone]