逆向选择下信息系统的选择标准

逆向选择情形下,代理人拥有的私人信息对委托人与代理人博弈的均衡结果有着至关重要的影响.实际上,逆向选择过程中委托人面临的问题可分为两个阶段:一是对信息系统的选择,二是在给定信息系统下的机制设计问题.现有文献往往关注后一阶段信息系统既定情况下的机制设计问题,本文研究的是第一阶段信息系统的选择标准.在代理人效用函数特定的基础上,当代理人能力的分布函数G(y)一阶随机占优于分布函数F(y),委托人相对于信息系统G(y)更偏好于信息系统F(y),即一阶随机占优(FSD)是逆向选择情形下信息系统占优的充分条件.

The Ranking Criteria for the Information System under Adverse Selection

Under adverse selection, the agent's private information has a significant affect on the outcome of the game between the agent and the principal. In fact, the question the principal faces in the process can be divided into two stages: first, choosing the right information system; second, designing mechanism under given information system. Present articles usually deal with the second problem, while this paper studies the ranking criteria for the information system in the first stage. On the basis of specific utility function of the agent, if the distribution function G(y) of the agent's ability first-order stochastically dominates another distribution function F(y), then the principal prefers information system F(y) to information system G(y), that is, First-order Stochastic Dominance(FSD) is a sufficient condition in ranking information system under adverse selection. This conclusion has a very intuitive explanation in economics, Usually, First-order Stochastic Dominance implicates higher mean, and that is, when confronted with agents with higher abilities, principal has relatively higher expected revenue in average. Similar studies show that the precision of prior distribution does not reveal agents' private information, so prior distribution is useless in mechanism design. Our conclusion indicates prior distribution has other meanings, it affects the information system the principal choose and in the end affects' principal's welfare. To some extent, our study is close to signaling or screening models, which admit the difference in agents' ability and design different contracts to distinguish different agents. However, unlike these models, we study only separating equilibrium based on revelation theorem, while they permit both separating and pooling equilibrium, also, they consider the costs of sending and screening signals which is not included in our model, and adding these costs means adding one layer of gaming, which may make our model too complicated to get an explicit solution.