Fictitious Play: A Statistical Study of Multiple Economic Experiments

We propose a methodology for combining a number of experiments to make an overall assessment of the empirical justifiability of certain economic hypotheses. We illustrate our methodology by an application to models of learning in repeated games. The application of our proposed (purely Bayesian) procedure allows us to combine two sets of experiments (for a total of nine experiments) to update our beliefs on the relative justifiability of the Cournot and the fictitious play learning hypotheses. The experiments we analyzed indicate that the fictitious play hypothesis is impressively more likely than the Cournot hypothesis.