Motivation. Coalition Formation in Buyer-Seller Networks. Agenda. Internet: opportunity to form spontaneous, location-independent communities

Transcription

1 Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Coalition Formation in Buyer-Seller Networks Frank E. Walter, Stefano Battiston Frank Schweitzer fschweitzer@ethz.ch Internet: opportunity to form spontaneous, location-independent communities emergence of services based on social networng Application: electronic markets Buying clubs for e-commerce have been around for several years, but the concept itself has not really become popular. advantage: economies of scale increase in quantity decrease in cost-per-item Tsvetovat & Sycara (): formation of groups of buyers to obtain volume discounts from sellers Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Agenda Drawbacks and Incentives Related Work Outline Agent s Utility Extensions Conclusions Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Coalition Formation: alternative to achieve trade-off between economies of scale and matching of preferences:

2 Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Examples buying clubs for food: few self-organised groups co-buying web sites such as Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Related Work Related Work Tsvetovat and Sycara (): Incentive analysis for the formation of buying clubs. Yamamoto and Sycara (): Coalition formation scheme; stability and efficiency analysis. He and Ioerger (): Coalitions as a means of minimising the cost of bundle search. Sarne and Kraus (): Coalitions as a means of sharing the cost of searching specific sellers. Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Drawbacks and Incentives Drawbacks and Incentives more buyers, lower cost principle based on limited selection of products buyers have to compromise buying clubs : waiting time and risk of not concluding a deal (additional overhead) Buyers volume discount customised items: match of preferences sharing of search cost bundle search Sellers better predictability of sales volumes customized items: increase of sales reduction of transaction costs Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Outline agent-based model buyers B and sellers S (represented by their products) focus: dynamics of creation, evolution, and breakup of coalitions of buyers emphasis on: effect of heterogeneity of agents preferences size, number and lifetime of coalitions existence of stationary and non-stationary regimes (stable and unstable coalitions), transition from one regime to the other

3 Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Outline Heterogeneity: products j: vector of features [w j,,..., w j,k ] buyers i: preferences for product features [v i,,..., v i,k ] Example: buyer i v i, seller/product j w j, distributions F(v), G(w) each agent buys only one product different buying modes: individually, in coalitions, new demand v w v v v v w w agents preferences products features Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Agent s actions and decisions purchase product j individually advantage: get product immediately disadvantage: pay higher price pi = P ind (t) [ ] ij P join existing coalition j with a set of other buyers N j advantage: pay lower price pi = P/ N j disadvantage: (i) waiting time until coalition has reached critical size N j N thr, (ii) risk of coalition failure coal (t) Nj P N j [ ] ij N thr Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Agent s Utility Utility agents: rational and self-interested maximise their private utility over time benefit of agent i from purchase of product j depends on: distance between features wj and preferences v i : ij = w j v i price of product j, which depends on quantity sold: pj = P/N β j (price elasticity: β =.) agent s utility: compromise between cheap price and match of preferences U i = p j [ w j v i ] indirect cost for joining a coalition commitment unsuccessful coalition: U i = (risk of failure) Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / initiate new coalition k and wait for other buyers to join advantage: get product k according to preferences: ik = disadvantage: (i) risk of coalition failure: Nk (t ) = N thr, (ii) waiting time until coalition has reached critical size init (t) P N thr postpone decision advantage: no commitment, open for future possibilities disadvantage: (i) wait for product, (ii) uncertainty of future wait (t) exp ( αt)

4 Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Stochastic decision dynamics Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Individual purchasing (N thr = ) each possible action has a certain weight k i decision: stochastic draw among the weighted possibilities path dependence: symmetry break positive feedback: decision affects weights kj of other agents consequences for utility at t tend affects strategy in repeated games t= t= t= t= k wait random number [, Σk] k ind k ind k coal k coal j k j k init Σk i (t) k Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Scenarios of coalition formation individual purchasing behaviour, i.e., no buyer-seller network exists among the agents, formation of several heterogenous coalitions, i.e., a number of buyer-seller networks which are not connected, condensation to a single giant coalition, i.e., a buyer-seller network involving all agents. The transition between these scenarios is governed by heterogeneity of agents preferences, η threshold for successful coalitions, N thr t= t= t= t= Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Several heterogeneous coalitions (ε =., N thr = ) t= t= t= t= t= t= t= t=

5 Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Single giant coalition (ε =., N thr = ) t= t= t= t= Influence of coalition threshold N thr N threshold as a fraction of N agents Total Utility (Normalised) N threshold as a fraction of N agents t= t= t= t= Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Fraction of agents in coalitions vs. time t t t Utilities indiv.p. several c. giant c. Avg... StDev... Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Extensions Extensions buyer s dynamics different preferences (multidimensional case) incomplete, bounded in time information about products buy different products with limited budget competition seller s dynamics products with different features (multidimensional case) offer more than one product limited production ressources competition repeated games buyers: memory about the failure/success of coalitions sellers: memory about agents loyalty reward stationary/non-stationary coalitions

6 Coalitions in Buyer-Seller Networks... Frank Schweitzer Complexity Aix-en-Provence - May / Conclusions Conclusions coalition: social network of agents to reach certain goal get customized products at a lower price compromise between preferences and price risk of failure modeling framework: formation of coalitions vs. individual buying focus: heterogeneity of agents/producs, threshold for success three different scenarios utility maximization at several heterogeneous coalitions extensions towards multiple products/preferences, learning effects, competition scenarios consumer driven economies of scale match of preferences, predictability of sales, reduced costs