This paper considers the problem of resource allocation for a large-scale wireless network consisting of multiple amplify-and-forward cooperative systems. The traditional competitive design leads to a network collapse where every system obtains almost zero throughput. We resolve this issue by resorting to the novel economic model of coalition formation with externalities by first letting the systems (modeled as players) self-organize into mutually beneficial coalitions, followed by a sequential non-cooperative game with improved player profiles. The results on the two-player game has been generalized to the case of an arbitrary number of players, incorporating additional peak power constraints as well. The uniqueness of the Nash equilibrium is proved analytically. Simulation results show that the proposed method improves the sum rate over 200% relative to the non-cooperative case due to improved frequency reuse.