发表在 Management Science, 2005. DOI: https://doi.org/10.1287/mnsc.1040.0217
Keywords: salesforce; incentives; asymmetric information; screening; production planning
A firm sells a single product through a sales agent. The total sales or demand $$ X = a + \theta + \epsilon $$ where
- $a$ is the agent’s selling effort
- $\theta$ is the market condition, $P(\theta = \theta_H) = \rho = 1-P(\theta=\theta_L); \; \theta _H > \theta_L$
- $\epsilon \sim N(0, \sigma^2)$ is a random noise
The principal designs the agent’s wage contract and makes production decisions; the agent endowed with private information about the market condition, decides whether or not to accept a contract and, if so, how much selling effort to exert.
The following sequence of events:
- The firm (or principal) offers a menu of wage contracts
- The agent privately observes the value of $\theta$
- The agent decides whether or not to participate (work for the firm) and if so, which contract to sign
- Under a signed contract, the firm determines the production quantity, and the agent makes the effort decision
- Both parties observe the total sales
The firm cannot directly observe the agent’s effort level, and thus must compensate the agent based on the realized value of $X$ .
Assume the agent’s untility for net income $z$ is $U(z) = - e^{-rz}$ with $r > 0$ .
Suppose $s(\cdot)$ is the contract being considered, and cost of effort is assumed to be $V(a)=a^2/2$, the agent solves the following optimization problem $$ \max _a E\left[-e^{-r(s(X)-V(a))}\right] $$
Let the unit selling price be $1 + c$ (the profit margin is thus normalized to $1$), the principal faces a newsvendor-like problem when choosing a proper production quantity.
Let $Y$ be the normally distributed net income. The agent's expected utility is thus $E\left[-e^{-r Y}\right]$. The certainty equivalent of $Y$, denoted by $C E[Y]$, is the fixed net income that provides the agent with a utility level equal to $E\left[-e^{-r Y}\right]$; i.e., $-e^{-r C E[Y]}=E\left[-e^{-r Y}\right]$. It can be easily verified that $$ C E[Y]=\mu_Y-\frac{1}{2} r \sigma_Y^2 $$
Let $a(s, t)$ be the optimal effort decision given contract $s(\cdot)$, and agent’s type $t (= H \text{ or } L)$.
Let $u(s, t)$ be the corresponding expected utility for the agent, i.e., the maximum achievable expected utility under $s$ and $t$.
The agent’s reservation utility is represented by $-U_0$ .
逆向选择与道德风险
说到委托代理理论,那么一定会涉及到「逆向选择」和「道德风险」这两个概念。
用一个例子来说明这二者分别是什么
为了缓解同学们挂科的忧虑心情,你开发了一款产品叫“挂科险”,给挂科的同学一点金钱上的安慰。
但是,这款产品的运营容易出现的一个很大的问题是,买挂科险的往往都是那些不好好学习的学渣,学霸是不会买挂科险的,这样按照学校平均的挂科率来设计保险产品肯定是有问题的。
这里出现的一个现象就是逆向选择(adverse selection),作为担保方,因为你不知道同学们真实的学习水平(不对称信息,asymmetric information),导致买挂科险的大都是不好好学习的。一群不读书的人来买你的挂科险,你分分钟经营不下去。
总的来说,逆向选择指的是因为信息不对称导致的资源配置扭曲的情况。因为信息不对称,所以劣币驱逐良币,好人没法生存下去。
这时候可以通过设计 screening contract 来解决不对称信息的问题。
比如说,提供两种保单
- 当考试成绩比60略低时(55-59),能获得较高赔付
- 考试分数越低,赔付额越高
这样,那些徘徊在及格线边缘努力挣扎的同学就会选择第一种保单,而那些躺平摆烂的同学就会选择第二种保单。
于是,我们通过合同的设计就能消除掉不确定信息,使我们知道每个同学的类型。
至于道德风险(moral hazard),当同学买了挂科险之后,就很容易懈怠,不努力学习;买了挂科险,反而挂科的概率更高了。就像人买了车险之后就不好好开车了。
道德风险也是可以通过适当的合同设计来规避的,比方说,规定同学在图书馆学习至少多少个小时的时间,保单才会生效,这就避免了同学摆烂骗保的发生。再比如现在雇佣销售,都是底薪+绩效的工资设计,使用绩效防止员工摆烂。
逆向选择和道德风险这两个概念具有相似性,关于它们的区分也有几种办法。
比方说,事前的信息不对称是逆向选择(你不知道来买保险的人是学霸还是学渣);事后的信息不对称是道德风险(你不知道一个人买了挂科险之后还会不会认真学习)。这是一种比较简单的区别办法。
另一种区别办法是隐蔽信息和隐蔽活动。「不知道来买保险的人是学霸还是学渣」是隐蔽信息,「不知道一个人买了挂科险之后还会不会认真学习」是隐蔽活动。
还有一种是内生 vs 外生。一个人是学霸亦或学渣,在某一时刻是外生的信息;而买了挂科险之后有没有动机摆烂不学习,这是由模型内生的。