The strategically ignorant principal

A principal-agent model is considered in which the principal decides how much private information to acquire before making an offer to the agent (e.g., an insurer decides whether to gather information about the risk status of a potential client through medical testing before offering the client a policy). I prove in a general environment that there is a nontrivial set of parameters for which it is strictly suboptimal for the principal to be completely informed, regardless of the continuation equilibrium following any information acquisition choice. This result is robust to the notion that an informed principal could select any desired equilibrium via persuasion over the agent’s beliefs. The intuition is that to convince the agent that she is contracting honestly given her private information, the principal may need to severely distort the allocation. This distortion can be very costly ex ante. Choosing to be partially ignorant frees the principal from these incentive constraints and partially mitigates the damage to her ex ante payoff. I also determine in a quasilinear, three state case the optimal information acquisition choice for the principal as a function of the parameters of the model. In particular, I characterize when it is optimal to be fully ignorant of the state, when partial ignorance is preferred and when the principal wants to know the state precisely. Although a small literature has looked at the principal’s information acquisition problem in specific environments or for restricted mechanisms, this paper is the first to take a mechanism design approach to the problem in a general environment. This generality is important since it allows the principal to make full strategic use of any information she acquires. Outside of this literature, the principal is assumed to have acquired information; this paper demonstrates that this assumption may be undesirable. ∗Department of Economics, University of Western Ontario (email: [email protected]). The author wishes to thank Charles Zheng for his tremendous guidance and support in this project, participants in the Micro Theory Group at the University of Western Ontario for helpful comments especially Maria Goltsman and Al Slivinski, as well as Ben Lester, Maciej Kotowski, Tymofiy Mylovanov and Gabor Virag. The author also gratefully acknowledges financial support of Social Sciences and Humanities Research Council of Canada and the Ontario Graduate Scholarship.