Dynamic incentives and Markov perfection : putting the 'conditional' in conditional cooperation

This paper experimentally examines the selection of equilibria in dynamic games. Our baseline treatment is a two-state extension of an indefinitely repeated prisoner's dilemma, which we modify in series of treatments to study the focality of efficiency and symmetry, the effect dynamic and static strategic externalities, and the size of the state-space. Subjects in our experiments show an affinity for conditional cooperation, readily conditioning their behavior on both the state of the world, and recent history of play. With strong dynamic and static externalities present we see most subjects coordinate on efficiency by conditioning on past play. However, when we remove either type of strategic externality, conditioning on just the state becomes more common, and behavior is consistent with the Markov-perfect prediction. Changes to the environment's state-space are more nuanced: perturbations of the game with small-sized noise does not lead to more state-conditioned behavior; however, a richer set of endogenous states does lead to more Markov-perfect behavior.