| Cournot's theory of best reply has been widely accepted as the most likely method a rational player will employ in a game where one's payoffs are dependent on their own strategy as well as their opponent's. This paper tests this theory, among others, in the context of an extremely limited information environment that contains eighteen two-player experimental environments. All participants face a sufficiently complicated and symmetric profit function. Derived best response functions contain both linear and non-linear components, yielding three Nash equilibria: two stable, one unstable. I find that less than half of all subjects play the Pareto optimal Nash equilibria, and about five percent play the unstable Nash equilibrium. Two tests of convergence are proposed: one focusing on the level of the individual, and the other at the level of the group; both proving to be weak. By performing ordinary least squares regression analysis, I find that aggregate path of play is best described by a combination of imitating one's opponent, and inertia---playing the same strategy in the current period as the prior. Furthermore by expanding analysis to the level of the individual, I find that there is significant heterogeneity in learning rules utilised among individuals, with most individuals employing less rational theories of learning. |
