Learning purified mixed equilibria

We examine the local stability of mixed equilibria in a smoothed fictitious play model. Our model is easy to analyze and yields the same conclusions a s other models in 2 x 2 games. We focus on 3 x 3 games. Contrary to some pr evious suggestions, learning can sometimes provide a justification for comp licated mixed equilibria. Whether an equilibrium is stable often depends on the distribution of payoff perturbations. The totally mixed equilibria of zero sum games are generically stable, and the totally mixed equilibria of symmetric games with symmetric perturbations are generically unstable. Clas sification Number: C72. (C) 2000 Academic Press.