Revisiting collusion in routing games: A load balancing problem

Is it profitable for players to unite and merge to a single player? Obviously, the sum of utilities at an equilibrium cannot exceed the sum obtained if all players join together. But what happens if only a subset of players join together? Previous work on collusion have already shown that the society may either gain or loose from collusion of a subset of players. In this paper we show for a simple load balancing example that not only the society may loose, but also the subset of players that collude may end up with a worse performance than without collusion. In doing so, we introduce new concepts that measure the price of collusion.