Noisy Equilibrium Selection in Coordination Games

We analyse symmetric coordination games à la Bryant (1983) where a number of players simultaneously choose efforts from a compact interval and the lowest effort determines the output of a public good. Assuming that payoffs are concave in the public good and linear in effort, this game has a continuum of Pareto-ranked equilibria. In a noicy variant of the model an error term is added to each player's choice before his effort is determined. An equilibrium of the original model is noise-proof if it can be approximated by equilibria of noisy games with vanishing noise. There is a unique noise-proof equilibrium and, as the noisy games are supermodular, this solution can be derived by an iterated dominance argument. Our results agree with the experimental findings in Van Huyck, Battalio and Beil (1990). We also show that the unperturbed game is a potential game and that the noise-proof equilibrium maximizes the potential.