Abstract

Thomas Hobbes famously proposed the solution of an absolute sovereign to foster cooperative behavior. Yet reputation effects can encourage cooperation in repeated games where players' interests conflict, such as prisoner's dilemma, which can model common trading encounters. Institutions such as private judges and credit bureaus can collect and transmit information on a player's history, so that the players may form a reputation. Players may thus cooperate now in the expectation that others will cooperate with them later. This dissertation offers a Hobbesian argument, made in terms of the self-interest of the players and using the tools of modern game theory, to argue that such institutions form a cheaper and more libertarian mechanism to encourage cooperation than Hobbes' Leviathan. And while more expensive than more informal mechanisms such as tit-for-tat and grim-trigger, these institutions can offer adjudication of disputes and are more robust with respect to errors, and thus can especially work better for a large and transient population. In order to show robustness, the effects of different kinds of errors are analyzed, including correlated versus uncorrelated errors, and type I versus type II errors. The models in which these possible errors are considered are based on the Law Merchant model of Milgrom et al. and the Credit Bureau model of Klein . Reputation effects can in general survive in repeated game models accurate enough to include errors, but the errors will reduce the parameter space over which cooperation can be preserved. But where the system is especially sensitive to permanent type I errors of recording, a correction mechanism is proposed. Corruption is also a potential problem for such institutions, so this dissertation analyzes the "Bribe Solicitation Strategy" of model of Milgrom et al. and finds it to be implausible in that it benefits only the judge. A more plausible strategy is presented that benefits the briber as well. Nonetheless, under certain conditions, a small rate of corruption would be indistinguishable from the errors against which the system is already robust