COOPERATION DYNAMICS IN REPEATED GAMES OF ADVERSE SELECTION: TO FORGIVE IS NOT TO FORGET∗

ABSTRACT. We study repeated games with Markovian private information and characterize op-timal equilibria as players become arbitrarily patient. We show that seemingly non-cooperative actions may occur in equilibrium and serve as signals of changes in private information. Play-ers forgive such actions, and use the information they convey to adjust their continuation play. However, to forgive is not to forget: players keep track of the number of aggressions and enter into a punishment phase if that number becomes suspiciously high. Our model explains features of long-run relationships that are only barely understood, such as equilibrium defaults, unilateral price cuts, collusive price leadership, graduated sanctions, and restitutions. We also explore a model in which interactions are frequent and show how increasing the persistence of the process of types reduces informational frictions