Essays in dynamic games: Bargaining, elections, and self -control

Chapter 1 investigates the impact of optimism on multilateral bargaining. I show that extreme optimism results in costly delays in arbitrarily long finite games but with moderate optimism, all sufficiently long games end in immediate agreement. Bounding the efficiency loss from delay, I demonstrate that inefficiencies disappear in the continuous-time limit of the game; however, the cost of delay can be substantial in the presence of bargaining frictions. The dynamics engendered by extreme optimism involve cyclic delays: certain periods emerge endogenously as stalemates while others become ripe for agreement. I also consider other voting rules, and show that the majority-rule may be more efficient than the unanimity rule when agents are optimistic.;Chapter 2 (co-authored with Navin Kartik) develops a rational theory of behavior in sequential elections that accounts for momentum. We analyze an election with two candidates in which some voters are uncertain about which candidate is more desirable. Voters obtain private signals and vote in a sequence, observing the history of votes at each point. We show that, regardless of the voting rule, voters can herd on a candidate with positive probability, and such a "bandwagon" can occur with probability approaching one in large electorates. Our theory is distinct from the standard information cascades literature because voting is a collective decision problem, and consequently voters have forward-looking incentives to consider the actions of those after them.;Chapter 3 examines how a decisionmaker who is only partially aware of his temptation problem learns about it over time. In an environment in which the decisionmaker could eventually learn the exact nature of his temptation problem, rational learning may be perpetually incomplete with positive probability. In particular, partial awareness can generate overcommitment whereby decisionmakers with no or mild self-control problems choose to commit forever. My results cast doubt on the learning-theoretic justification for assuming that individuals perfectly understand their self-control problems, and demonstrate that partial awareness can have long-lasting behavioral implications.