What learning models tell us to expect in Three by Three Bimatrix games.

This paper extends several existing learning models to investigate their xed points (their long run predictions of play). The xed points of the model are not necessarily at the Nash equilibria of the payo matrices but are a function of both the Nash equilibria and the pa-rameters of the model. The stability of these xed points also depends on both the characteristics of payo matrix used and the parameters of the model. These new ndings indicate that behaviour previously thought to be inconsistent with theory may not necessarily be so