Switching Away From Probability One Beliefs

This paper considers a class of repeated signalling games to gain some intuitive insights into the effects and the desirability of modelling players in a dynamic game of incomplete information as being obstinate in the sense that their beliefs satisfy a support restriction. We demonstrate that such a restriction is rather dubious on a-priori grounds and in general imposes "too much" pooling on sequential equilibrium outcomes. Equilibria violating a support restriction should therefore not be dismissed in dynamic models of incomplete information and may actually reflect the possibility of reputation effects present in such a setting.