The principal-agent model represents situations in which a supervisor (principal) delegates a task to a supervisee (agent) who has specialized knowledge about the task but also has objectives that are different from those of the principal. This is what occurs, for example, between an employer and an employee, between legislators and bureaucrats, between regulators and regulated firms, between voters and public officials in a representative democracy, and between managers and stockholders. In such situations the principal cannot directly accomplish some specific tasks and needs to delegate them to the agent, who acts on behalf of the principal. The agent has superior private information that can be of two types. If the agent has superior information regarding the characteristics of some goods (e.g., their quality) or one or more of his or her own characteristics that are relevant to the task (e.g., skills, preferences, honesty), then the principal faces a problem of adverse selection. If the agent can take actions (e.g., the effort exercised in pursuing the task or the act of accepting a bribe from a third party to underperform) that are unobserved by the principal, then the problem is moral hazard. Because of the conflicting interests, the agent has to be compensated in order to be induced to act according to the principal's preferences. The principal's problem is therefore to design a contract that selects the right type of agent (to solve adverse selection problems) or that provides the agent with the right incentives, that is, induces the agent to act according to the principal's objective (to solve moral hazard problems). This process goes under the name of mechanism design.
In the absence of asymmetric information, the principal and the agent could stipulate a contract in which the payment for the agent is made contingent on the actions or type of actions of the agent. This would be the first best outcome. With asymmetric information, however, the problem is more complex, and it will usually be impossible to reach the first best. Asymmetric information induces an agency cost because providing the right incentives to the agent is costly. This is a special form of transaction cost that only allows the players to reach the second best, an outcome that is Pareto-dominated by the first best but that can be considered optimal given the added informational constraint faced by the principal.
When there is adverse selection, agents of the “good” type can try to convey information on their quality (or the quality of the products they try to sell) by acquiring a costly signal (signaling). Alternatively, the principal can propose a menu of contracts, trying to infer the quality of the agents from their choice of contract (screening). In any event, agents of inferior quality have an incentive to mimic the behavior of better agents. The problem is illustrated by Michael Spence in 1973 in a signaling model of the labor market in which education is used as a signaling device by workers with (innate) high skills. Education levels work as a signal of skill only if education is more costly to acquire for less skilled individuals. This assumption is also known as the Spence-Mirrlees condition, or single-crossing property, because the indifference curve (in this case mapping in a wage-education diagram) of the low-ability worker (or, in general, of the seller of bad products) crosses the corresponding curve of the high-ability worker once from below. In other words, a variable can be used as a signaling device only if it is manipulable by individuals and if we can expect it to be related in the right way (i.e., it is more easily obtained by good types) with the characteristic of interest. Two kinds of equilibria can emerge. In a separating equilibrium all workers with high productivity choose an education level that is higher than that chosen by less productive workers. Firms pay a higher wage (corresponding to workers' anticipations) to people with a higher education, correctly assuming that they have higher productivity. In a pooling equilibrium, all workers have the same education level and get the same wage. The outcome depends on the distribution of skills among workers, on the costs of education, and on the beliefs of agents.
Another possibility is that the principal tries to distinguish the different type of agents by proposing a menu of different contracts. In the model of Michael Rothschild and Joseph Stiglitz in 1976, an insurance company offers different contracts in order to separate high-risk from low-risk individuals. The role of education in Spence's model is played here by the willingness to afford more risk, because risk is more costly for people with higher probability of having an accident. Hence, the cost of separating (by getting incomplete insurance) is lower for agents with low risk of accidents. This is again the Spence-Mirrlees condition.
To analyze the case of moral hazard, consider the owner of a piece of land (the principal) and someone who is using the land to produce potatoes (agent). The amount of potatoes produced at the end of the year depends on the effort exercised by the agent and on the weather during the year. Assume also that the principal lives on a different continent and has no way to find out about the weather on his plot of land. The problem of the principal is then to design the appropriate reward scheme for the agent in order to maximize his or her own return, that is, the difference between the revenue from selling the potatoes and the payment to the agent. The possible reward schemes range between two extremes. At one extreme, the principal could pay a fixed wage to the agent, then sell the potatoes and keep the revenue. At the other extreme, he or she could rent the land to the agent for a fixed amount of money (that does not depend on the final output) and then let the agent keep all the revenue from the sale of the potatoes. These contracts differ not just in the way the reward is distributed but also in the incentives they provide to the agent and in the way the risk is distributed. In the first case agents have no incentive to provide any effort in cultivating the land and bear no risk: he or she is unaffected by weather conditions. In the second case the agent's incentive to provide effort is maximized, but the agent also bears all the risk. If both principal and agent are risk-neutral then this is the optimal contract. It is, however, often assumed that the agent is more risk-averse than the principal. A risk-averse agent is willing to accept more risk only if appropriately compensated, that is, if the risk is offset by a higher expected reward. The fundamental trade-off faced by the principal is therefore that incentives can only be provided by letting the agent bear some risk: hence, giving incentives to the agent (and therefore generating more output) also means having the agent bear more risk (and therefore having to pay him or her more in expected terms). With a risk-neutral principal and a risk-averse agent the optimal contract involves some risk sharing between the two.
The principal-agent model has been widely used in political science to analyze accountability problems of the sort that occur in the relationship between legislators and bureaucrats, or in that between voters and politicians. In this second example, the principal is a representative voter who tries to get his or her preferred policies enacted, and the agent is a politician who tries to be reelected. The incentive scheme that the voter can use is obviously much coarser that the sort of reward schemes that can be implemented in other situations: it consists essentially of the decision whether or not to retain the agent. An important conclusion of this class of models is that the perspective of being reelected acts as a discipline device for incumbent politicians (the performance effect). There is also a selection effect, in the sense that better types will generally have more chances to be confirmed in office. Empirical work by Tim Besley and Anne Case in 1995 finds that gubernatorial term limits have significant effects on economic policy choices, which confirms that incumbent governors facing the perspective of reelection behave differently from “lame duck” governors.
Adverse Selection, Manipulation, Moral Hazard
Imagine the worst happens, and you are falsely accused of a crime in a city in which you are vacationing and in which you do not know anyone....
Principal–agent games address the situation where one party (the agent) can take an action unobserved by the other party (the principal), who...
Principal-agent theory (PAT) was developed in the field of economics in the 1970s to understand the prevailing problems that appear every time...