Abstract:
Network formation is frequently modeled using link-formation games and typically present a multiplicity of Nash equilibria. Cooperative refinements - such as strong or coalitional proof Nash equilibria - have been the standard tool used for equilibrium selection in these games. Non-cooperative refinements derived from the theory of global games have shown also that, for a class of payo¤ functions, multiplicity of equilibria disappears when the game is perturbed by introducing small amounts of incomplete information. We conducted a laboratory study evaluating the predictive power of each of these refinements in an illustrative link-formation game. Compared with cooperative game solutions, the global game approach did significantly better at predicting the strategies played by individuals in the experiment.