Essays on repeated principal-agent interaction

In a repeated interaction, one tries to maximize his/her current payoffs and consider the decision's future implications. Reputation concerns arise when there is incomplete information regarding the references of the counterparts. I theoretically and experimentally analyze the finitely repeated interaction between an agent and a principal. The principal can be thought of as a state official who holds either a one-period or a long-term project. The agent is a private firm that holds the necessary information for the project. The principal holds a project which yields a payoff depending on the realized state of the world. He is not informed about the state of the world and hires an informed agent to delegate the authority. The agent may be a good type who shares the same preferences with the principal. Moreover, there is a positive probability that the agent may be a "bad type" who has misaligned preferences with the principal. The principal does not know the agent's type but knows the prior probability that the agent is a bad type. The agent is a long-term player, and the principal is either a short-term or a long-term player. In the short-term-principal setup, the long-term agent meets a new principal in each period of the interaction. If the principal is also a long-term player, then the agent and principal meet once and interact as long as the principal continues to hire the agent. In both setups, the principal observes the past choices of the agent when making the hiring decision. Hence, both types of agents have a long-term incentive to keep a high reputation level in addition to the short-term incentive to play the stage-game optimal action. The good agent wants to separate from the bad agent, whereas the bad agent wants to mimic the good type. This thesis includes three chapters. The first chapter defines a benchmark model. The short-term principal does not hire the agent in the initial periods where there are reputation incentives. The more periods are left to be played, the more the agent's reputation incentives. Hence, the agent tends to play the action that will increase her reputation level even if that action is not optimal in the stage game. The principal expects a negative payoff in such periods. The reputation incentives lead to surplus loss as the principal does not hire the agent in such periods. Hence, the "bad reputation" behavior is observed [Ely and Valimaki, 2003]. When the principal is also a long-term player, the harmful effect of reputation incentives can be avoided. The long-term principal hires the agent in periods with reputation concerns. He can compensate for the losses of reputation-building behavior in the future periods. Moreover, the long-term principal hires the agent for lower reputation levels. Hence, it is seen that a long-term interaction between the agent and the principal decreases the harmful effects of reputation-building behavior and improves equilibrium payoffs . In the second chapter, I introduce endogenous stakes. Either the agent or the principal sets the allocation of the relative stakes. The agent prefers to start the career path with larger stakes, decreasing the stakes gradually. By starting large, the (good) agent creates an environment where future interaction becomes less valuable. Hence the incentive to take the right action in the current period prevails over reputation concerns. On the other hand, the principal allocates the stakes so that the interaction starts small. By doing so, the principal benefits from the reputation concerns of the agent. The principal updates his belief on the agent through interaction - stakes increase as the agent's reputation rises. Likewise, the social planner - whose objective is to maximize social welfare - prefers to start the interaction small. The findings highlight the optimal design of a career path under different conditions. The third chapter theoretically and experimentally analyzes a slightly different repeated principal-agent game with varying relative stakes. I focus on the setup where there is a high probability that the principal and the agent have misaligned preferences (the agent is the bad type with sufficiently high probability). Repeated play becomes valuable in such a setting by improving the equilibrium payoff of the principal. The agent has reputation incentives that motivate her to take action matching the true state in the initial periods rather than maximize her period payoff. As Morris [2001] calls, the "discipline effect" of the reputation incentives benefits the principal. We show it is optimal for the principal to start the interaction small and increase the stakes gradually. The agent's reputation incentives are managed so that the reputation evolves slowly. I test these predictions in four treatments via online experiments. Each period receives an equal stake in the first treatment. The interaction starts small in the other three treatments, and the stakes increase at different speeds. We show that the smaller the interaction starts, the higher the reputation incentives of the agent. More importantly, I show that the principal earns a higher payoff in starting-small (gradualism) treatments than the equal-stakes treatment. I contribute to the literature on gradualism by showing that it is a valuable tool for improving equilibrium payoffs in the principal-agent framework with asymmetric information.