Implementing Efficient Graphs in Connection Networks∗

We consider the problem of sharing the cost of a network that meets the connection demands of a set of agents. The agents simultaneously choose paths in the network connecting their demand nodes. A mechanism splits the total cost of the network formed among the participants. We introduce two new properties of implementation. The first prop-erty, Pareto Nash Implementation (PNI), requires that the efficient out-come is always implemented in a Nash equilibrium, and that the efficient outcome Pareto dominates any other Nash equilibrium. The average cost mechanism (AC) and other asymmetric variations, are the only rules that meet PNI. These mechanisms are also characterized under Strong Nash Implementation. The second property, Weakly Pareto Nash Implementation (WPNI), requires that the least inefficient equilibrium Pareto dominate any other equilibrium. The egalitarian mechanism (EG), a variation of AC that meets individual rationality, and other asymmetric mechanisms are the only rules that meet WPNI and Individual Rationality. PNI and WPNI provide the first economic justification of the Price of Stability (PoS), a seemingly natural measure in the computer science literature but not easily embraced in economics. EG minimizes the PoS across all individually rational mechanisms. JEL classification: C70, C72, D71, D85