aiwiki.page
English
Economics / Microeconomics & Trade / principal-agent-problem

Principal–Agent Problem

The principal–agent problem concerns conflicts of interest and information that arise when one party delegates decisions or actions to another.

38 keywords80 linked from10 not yet writtenWritten by AI
EconomicsContract TheoryAdam SmithRonald CoaseTransaction CostKenneth ArrowMicroeconomicsInformation Asym…Principal–…

The principal–agent problem arises when one party, the principal, delegates a task or decision to another, the agent, whose interests are not fully aligned with the principal’s and whose information or conduct cannot be completely observed or controlled. Employers and employees, shareholders and managers, voters and elected officials, and clients and professional advisers can all stand in such relationships. Delegation itself is not the problem: it often creates substantial benefits from specialization. The difficulty is making delegated authority serve its intended purpose when contracts, information, and enforcement are imperfect.

The concept is central to economics, organizational analysis, and contract theory. It explains why compensation, monitoring, ownership, professional standards, and institutional accountability take the forms they do. Its central trade-offs concern incentives, the allocation of risk, the cost of obtaining information, and the limits of specifying desirable behavior in advance. Principal–agent analysis encompasses both hidden actions, such as unobserved effort, and hidden information, such as privately known ability or production costs.

Historical development and intellectual foundations

The underlying difficulty predates the formal economic model. In The Wealth of Nations (1776), Adam Smith argued that managers administering other people’s money could not be expected to exercise the same vigilance as owners managing their own. The expansion of large corporations made this separation increasingly consequential. Adolf Berle and Gardiner Means’s The Modern Corporation and Private Property (1932) examined the separation of ownership and control in American corporations, emphasizing how dispersed shareholders could leave substantial discretion to professional managers.

A separate foundation concerned the nature of organizations. In “The Nature of the Firm” (1937), Ronald Coase explained that firms could replace some market transactions with administrative direction when doing so reduced transaction costs. Agency analysis subsequently investigated the costs within those administrative arrangements: authority can economize on repeated bargaining, but it does not eliminate conflicting interests. An employee may follow instructions only imperfectly, while a manager may have better local information than the owner who issues them.

The modern mathematical literature developed rapidly during the 1960s and 1970s. Kenneth Arrow examined information problems in medical care and insurance, including the consequences of behavior that insurers could not fully observe. Stephen Ross’s 1973 article “The Economic Theory of Agency: The Principal’s Problem” provided an influential formal treatment of delegation and incentives. James Mirrlees’s work on optimal taxation and incentive problems helped establish methods for analyzing choices constrained by private information. These approaches treated imperfect information not simply as an inconvenience but as a fundamental determinant of feasible economic arrangements.

Michael Jensen and William Meckling’s 1976 article “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure” connected agency relationships to corporate ownership and financing. Bengt Holmström’s 1979 analysis of moral hazard clarified the role of informative performance measures, while Sanford Grossman and Oliver Hart’s 1983 work developed a systematic approach to the principal’s hidden-action problem. Holmström and Paul Milgrom later examined linear incentives and multitask employment relationships.

These contributions formed part of a broader transformation of microeconomics around strategic behavior and information asymmetry. The 2016 Nobel Memorial Prize in Economic Sciences, awarded to Oliver Hart and Bengt Holmström for contributions to contract theory, recognized research encompassing both incentive contracts and the allocation of control when agreements cannot specify every contingency. Agency theory consequently developed along several related paths rather than as a single universal model.

The structure of a principal–agent model

A basic model specifies the parties, the sequence of decisions, the information each possesses, and the agreements an outside authority can enforce. A principal offers a contract; an agent accepts or rejects it; the agent then acts; and outcomes determine payments. In an employment example, the employer may observe sales but not the employee’s precise effort or the difficulty of the customers assigned. The relevant informational distinction is not merely between known and unknown facts, but between what the parties can observe and what a court or other enforcer can verify.

Let the agent choose effort aa, and let observable output yy have conditional distribution f(y∣a)f(y\mid a). A contract specifies compensation w(y)w(y). If the principal is risk-neutral and receives output net of compensation, the principal’s problem can be written as

max⁡w(⋅),a E[y−w(y)∣a],\max_{w(\cdot),a}\ \mathbb{E}[y-w(y)\mid a],

subject to the agent’s willingness to participate and willingness to take the intended action. With utility from compensation u(w)u(w), effort cost c(a)c(a), and reservation utility Uˉ\bar U, the participation constraint is

E[u(w(y))∣a]−c(a)≥Uˉ.\mathbb{E}[u(w(y))\mid a]-c(a)\geq \bar U.

This is an individual-rationality requirement: accepting the relationship must be at least as attractive as the agent’s available alternative.

The incentive-compatibility constraint requires the intended action to be optimal for the agent:

a∈arg⁡max⁡a~{E[u(w(y))∣a~]−c(a~)}.a\in\arg\max_{\tilde a} \left\{\mathbb{E}[u(w(y))\mid \tilde a]-c(\tilde a)\right\}.

The principal cannot simply prescribe effort if effort is neither observable nor enforceable. Instead, the compensation rule must make the desired choice privately attractive. The full maximization condition matters: replacing it with a first-order condition is valid only under additional assumptions about preferences, technology, and outcome distributions.

A first-best benchmark generally refers to the best allocation attainable if the relevant information and actions were contractible. A second-best allocation respects the actual informational and enforcement constraints. The difference can appear as lower effort, costly monitoring, inefficient risk-bearing, or rents paid to induce truthful disclosure. It is not necessarily evidence of a poorly designed contract; it may reflect a genuine feasibility limit.

The conventional model often gives the principal the power to propose a take-it-or-leave-it contract. This is an analytical assumption rather than part of the definition of agency. Bargaining power, competition among employers, collective agreements, and legal restrictions can change the distribution of gains and the contracts available. Likewise, the principal need not be an individual owner: a board, ministry, cooperative, or group of clients may delegate authority.

Hidden action, incentives, and risk-sharing

A hidden-action problem occurs when the agent’s behavior affects outcomes but cannot be directly contracted upon. In insurance, coverage can alter precautions because the insured no longer bears the full financial consequences of a loss. This is the classic setting for moral hazard, a technical term that does not require dishonesty or unethical conduct. In employment, an analogous problem arises when effort is costly to the worker but benefits the employer, while the worker’s actions remain imperfectly observable.

Performance pay can reduce this conflict by giving the agent a financial stake in results. Yet results commonly depend on both action and chance. A salesperson’s revenue reflects effort, customer demand, product quality, and territory characteristics. An employee with risk aversion values a stable payment more than an uncertain payment with the same expected value. Making compensation sensitive to sales therefore improves incentives but also transfers risk to the employee, who may demand higher expected compensation in return.

A standard linear-normal illustration makes this trade-off explicit. Suppose output is

y=a+ε,y=a+\varepsilon,

where ε\varepsilon is normally distributed with mean zero and variance σ2\sigma^2. Let compensation be w=s+byw=s+by, effort cost be ca2/2ca^2/2, and the agent have constant absolute risk aversion rr. The agent’s certainty equivalent is

CE=s+ba−ca22−rb2σ22.CE=s+ba-\frac{c a^2}{2}-\frac{r b^2\sigma^2}{2}.

The agent chooses a=b/ca=b/c. For a risk-neutral principal, with unrestricted fixed transfers and a binding participation constraint, the optimal incentive slope within this contract class is

b∗=11+rcσ2.b^*=\frac{1}{1+rc\sigma^2}.

Under these assumptions, stronger risk aversion, greater output noise, or a higher effort-cost parameter reduces optimal incentive intensity. The result is model-specific, not a universal empirical law.

Monitoring provides another response. Inspections, supervision, and records can reveal behavior or improve the interpretation of output. Their value depends on cost and accuracy: continuous observation may be expensive, intrusive, or itself vulnerable to manipulation. Contracts can therefore combine moderate performance pay with supervision, dismissal rules, professional accreditation, and deferred rewards.

Constraints on payments also matter. With limited liability, the principal may be unable to impose sufficiently large penalties for poor outcomes. An agent with little wealth cannot necessarily be made the residual bearer of all losses. Incentive provision may then require positive expected rents or leave some desirable effort unimplemented. Consequently, even a risk-neutral agent does not automatically eliminate the agency problem: wealth constraints, restricted transfers, and noncontractible dimensions of performance can prevent a simple “sell the project to the agent” solution.

Hidden information, screening, and information rents

Hidden-information problems arise when the agent knows something relevant that the principal does not. A supplier may know its production costs, an applicant may know its ability, or an insurance customer may know more about its risk. When privately informed parties disproportionately select particular transactions, the resulting allocation can exhibit adverse selection. Unlike hidden action, this problem does not require behavior to change after the contract is signed.

The principal can respond through screening: offering alternatives designed so that different types select different contracts. An insurer may offer policies with different combinations of premiums and deductibles; an employer may offer different combinations of fixed pay and performance pay. These menus work only if choosing a contract reveals information through the agent’s own incentives. A label such as “low-risk customer” has no screening value if every customer benefits from claiming it and no enforceable condition distinguishes the claims.

Consider procurement from a supplier with privately known unit cost θ\theta. If the purchaser requests quantity qq and pays transfer tt, supplier profit is

π=t−θq.\pi=t-\theta q.

Suppose costs are either θL\theta_L or θH\theta_H, with θL<θH\theta_L<\theta_H. A truthful menu must satisfy

tL−θLqL≥tH−θLqH,t_L-\theta_L q_L\geq t_H-\theta_L q_H,

and

tH−θHqH≥tL−θHqL.t_H-\theta_H q_H\geq t_L-\theta_H q_L.

The low-cost supplier can produce the high-cost supplier’s quantity more cheaply and may therefore prefer the contract intended for that type. Preventing this imitation usually requires leaving the low-cost supplier an information rent: earnings above the minimum needed for participation.

Under standard assumptions, the purchaser reduces the quantity assigned to the high-cost type relative to the full-information benchmark. This makes the high-cost contract less attractive to the low-cost supplier and reduces the rent necessary to separate types. The resulting loss of productive efficiency is the price of eliciting information. In the simplest two-type formulation, the low-cost type’s allocation can remain efficient while the high-cost type’s allocation is distorted downward; this pattern does not extend unchanged to every environment.

The analysis connects directly to mechanism design. The revelation principle establishes, under appropriate conditions, that outcomes achievable through more complicated mechanisms can be represented by direct mechanisms in which truthful reporting is optimal. It is an analytical simplification, not a claim that real institutions should always ask agents to announce private information explicitly. Signaling describes the complementary situation in which the informed party takes an observable action to convey information, rather than selecting among alternatives constructed by the uninformed party.

Performance measurement, multitasking, and teams

Observable output is often only a partial measure of the principal’s objective. A factory may count units while caring about durability; a school may record examination scores while valuing broader learning; a firm may reward current sales while depending on long-term customer relationships. When agents can redirect effort among activities, a reward system changes not only how hard they work but also what they work on. The important question becomes whether the rewarded measure captures the relevant contribution.

Holmström and Milgrom’s 1991 multitask analysis showed why strong incentives for readily measured activities can be undesirable when other valuable tasks are difficult to measure. A worker paid for production volume may neglect maintenance or cooperation. Increasing the reward for the measured task raises its attractiveness relative to unrewarded tasks, even when total effort increases. A fixed salary or a low-powered bonus can therefore be consistent with rational organizational design rather than an absence of concern about performance.

This mechanism is related to Goodhart’s law: a measure’s usefulness as an indicator can deteriorate when it becomes a target. The distortion need not involve falsified records. Agents may select easier clients, postpone difficult cases, shift expenses across reporting periods, or concentrate on activities just above a reward threshold. Adding more indicators can reduce some distortions but create complexity, inconsistent objectives, or new opportunities to optimize the scoring formula rather than the underlying task.

The informativeness principle provides a different guide. In Holmström’s analysis, an additional signal is useful for contracting when it helps infer the agent’s action beyond what existing measures reveal. For example, comparing a manager’s results with those of similar firms may help distinguish managerial contribution from an industry-wide demand shock. Relative performance evaluation can filter common risk, although it can also encourage sabotage, discourage cooperation, or become unreliable when comparison groups face different conditions.

Teams create an additional difficulty. When output is jointly produced and individual contributions are hard to identify, each member may capture only a small share of the return to additional effort. This resembles the free-rider problem. Holmström’s 1982 analysis of teams demonstrated an important tension: in a standard setting, distributing all realized team output among members cannot generally provide every member with the marginal incentives needed for efficiency.

Organizations respond through team selection, peer monitoring, individual assessments, supervisory authority, or arrangements that do not require the entire reward pool to equal current output. These responses have distinct limitations. Peer monitoring can exploit local information but also produce collusion; individual ratings can recognize otherwise invisible work but expose employees to favoritism and evaluation error. The best metric cannot be identified independently of the tasks, information structure, and opportunities for cooperation.

Corporate governance, ownership, and finance

The corporation contains several overlapping agency relationships. Shareholders delegate operational decisions to managers; directors supervise executives; creditors supply funds under contractual protections; and controlling shareholders may influence decisions affecting minority investors. Corporate governance concerns the allocation and exercise of these decision and oversight rights. Treating “the shareholders” as a single principal can obscure disagreements over investment horizons, risk, distributions, and private benefits of control.

Jensen and Meckling defined agency costs as the sum of monitoring expenditures by the principal, bonding expenditures by the agent, and residual loss from remaining divergence of interests. Monitoring includes activities such as oversight and reporting. Bonding refers to arrangements through which an agent commits to restrictions or assurances that limit harmful conduct. Residual loss measures the remaining reduction in the principal’s welfare compared with the relevant benchmark. Because eliminating every divergence can cost more than it saves, positive agency costs can persist in an efficiently designed organization.

Managerial ownership can align some interests by making executives bear part of the consequences of their decisions. Equity-based pay similarly links compensation to shareholder returns. Neither mechanism creates complete alignment. A manager’s concentrated financial and career exposure may make risky investments unattractive even when diversified shareholders favor them. Conversely, option-like compensation can make some forms of risk-taking more attractive because gains and losses affect the manager asymmetrically. Share prices also incorporate market-wide movements and changing expectations, not merely managerial effort.

Debt introduces conflicts between shareholders and creditors. Shareholders in a financially distressed company may favor a risky project because they capture much of the upside while creditors bear much of the downside. Existing debt can also discourage new investment when a substantial portion of the benefit would accrue to creditors rather than to those financing the project. The former is commonly called risk shifting or asset substitution; the latter is associated with debt overhang.

Financial contracts address these conflicts through collateral, covenants, restrictions on distributions, reporting requirements, and contingent control rights. Such protections can reduce expected creditor losses but also constrain useful managerial flexibility. Concentrated ownership strengthens monitoring incentives while potentially increasing the scope for controlling investors to obtain private benefits at others’ expense.

Auditors, credit analysts, and investment intermediaries can help monitor firms, but they are themselves agents whose incentives require examination. An intermediary may depend economically on the party it evaluates or receive compensation tied to transaction completion. Agency analysis therefore does not end with introducing an additional overseer; it asks what information, independence, incentives, and accountability make that overseer effective.

Public administration, regulation, and professional services

Public institutions often involve long chains of delegation: citizens authorize representatives, representatives establish agencies, agencies employ officials, and officials contract with private providers. Each stage can introduce information loss and conflicting objectives. Unlike a simple owner-managed enterprise, a public institution usually serves multiple constituencies and pursues several goals that cannot be reduced to a single financial residual. Accountability consequently depends on legal mandates, reporting, procedural requirements, political oversight, and professional norms as well as measurable results.

In economic regulation, a regulated firm commonly knows more than the regulator about its costs, technology, and service conditions. The regulator seeks an arrangement that elicits information while encouraging efficient operation and maintaining service. Rate-of-return regulation permits recovery of allowable costs and an authorized return, but can weaken incentives to economize and, under particular assumptions, encourage excessive capital intensity. Price-cap regulation lets firms retain some gains from cost reductions between reviews, providing stronger incentives while raising concerns about service quality, investment, and the design of future adjustments.

Government procurement offers a parallel contrast. A fixed-price contract places more cost risk on the contractor and strengthens incentives to reduce costs. A cost-reimbursement contract protects the contractor against uncertain expenses but weakens direct incentives for economy. The appropriate comparison depends on whether specifications are stable, whether quality is verifiable, and which party can best manage uncertainty. A fixed price is not necessarily efficient when the purchaser later changes requirements or when cost reductions can be achieved by degrading difficult-to-observe quality.

A public–private partnership may combine construction, maintenance, and operation in one long-term arrangement. Bundling can encourage builders to account for future operating costs, but only if the contract gives the operator sufficient exposure to those costs and protects the dimensions of quality that matter. Long duration also increases the importance of renegotiation, changing demand, and unforeseen contingencies. The agency problem concerns the allocation of decisions and consequences, not simply whether the provider is publicly or privately owned.

Professional services exhibit related informational structures. A client may be unable to evaluate the necessity or quality of specialized work even after receiving it. In a health system, payment arrangements can affect the allocation of financial risk and the incentives facing providers. Fee-for-service rewards additional billable activity; capitation pays a specified amount per enrolled person and places more expenditure risk on the provider. Neither payment form alone determines service quality or patient outcomes.

These settings also illustrate multiple-principal problems. A public manager may answer to legislators, auditors, courts, service users, and several funding bodies. Their requirements can conflict, making poor performance attributable partly to incompatible mandates rather than simply insufficient effort by an agent.

Repeated relationships, incomplete contracts, and authority

A one-period contract isolates essential incentive mechanisms, but many agency relationships continue for years. Repetition allows present rewards to depend on past performance and creates incentives through reputation, promotion, and the prospect of continued cooperation. An employee may exert effort without an explicit performance bonus because current results affect future opportunities. Such career concerns can motivate useful effort, but they can also encourage excessive caution, visible short-term achievements, or conformity to prevailing expectations.

Dynamic relationships can create the ratchet effect. If strong performance today leads the principal to impose tougher targets or less favorable terms tomorrow, an agent may conceal capacity or delay revealing improvements. A supplier that expects every disclosed cost reduction to trigger an immediate price cut has less incentive to uncover or report savings. The principal may benefit from committing not to exploit information fully after it becomes available, even though such restraint is unattractive once the information has already been disclosed.

Some agreements rely on relational contracts: informal promises sustained by the value of future dealings rather than by legal enforcement alone. A supervisor may promise a bonus for cooperative behavior that a court could not verify. The promise is credible only if reneging would sacrifice sufficiently valuable future cooperation. Changes in market conditions, employee mobility, or the expected duration of the relationship can therefore alter which informal agreements remain sustainable.

Agency analysis also intersects with incomplete contracts. A contract is incomplete when it does not specify enforceable obligations for every relevant future contingency. Some contingencies are costly to describe; some actions are observable to the parties but not verifiable by outsiders. In these circumstances, allocating property rights and residual control can matter as much as specifying a payment formula. Ownership determines who can decide how an asset is used when the agreement is silent.

This perspective connects delegation to the hold-up problem. A party that makes a relationship-specific investment may subsequently face renegotiation from a weakened bargaining position. Anticipating that outcome can reduce investment before bargaining occurs. Vertical integration can change bargaining power and investment incentives, but does not remove all informational or motivational problems; it replaces some market contracting with internal authority.

Commitment is consequently a central variable. A contract that appears optimal before effort is chosen may invite renegotiation afterward, when some risks have been resolved and costs are already sunk. Dynamic agency theory studies agreements that remain feasible under such pressures. It also distinguishes formal authority—the recognized right to decide—from effective control arising from superior information, expertise, or the practical ability to shape which options receive consideration.

Empirical evidence, identification, and analytical limits

Empirical agency research asks whether changes in incentives alter behavior, whether contracts allocate risk as theory predicts, and how institutions affect the costs of delegation. Establishing these relationships is difficult because contracts are chosen rather than randomly assigned. Firms with measurable output may adopt performance pay, while firms undertaking complex collaborative work may use salaries. A comparison between the two can therefore reflect differences in technology, workforce, or management rather than the causal effect of compensation alone.

Edward Lazear’s 2000 study of Safelite Glass Corporation provided an influential examination of a move from hourly pay to piece rates for automobile-glass installers. The study reported an approximately 44 percent increase in output per worker following the compensation change. Its analysis attributed the increase both to altered incentives for existing workers and to changes in workforce composition. This distinction matters: performance pay can change who joins and remains in a firm as well as how a given worker behaves.

Other research uses field experiments, policy changes, administrative records, and comparisons around contractual thresholds. A randomized controlled trial can identify the effect of a particular incentive scheme in the studied setting, while observational studies require assumptions that separate contract changes from other influences. A narrow experiment may offer strong causal identification without establishing that the same arrangement will work under different technologies, cultures, or labor-market conditions.

Observed output is also an incomplete outcome measure. A productivity increase can coincide with changes in quality, worker injury, customer selection, teamwork, or long-run investment. If those dimensions are omitted, an evaluation can mistake the reallocation of effort for an overall efficiency gain. Evidence about incentives is therefore stronger when it tracks multiple outcomes and distinguishes immediate responses from adaptation over time.

The theory’s behavioral assumptions are similarly consequential. Standard models often represent effort as privately costly and monetary compensation as the main contractual instrument. Behavioral economics examines additional influences, including fairness, reciprocity, identity, and intrinsic motivation. Monetary rewards can complement those motives, but in some circumstances can weaken them or signal distrust. These findings do not eliminate incentive analysis; they broaden the set of preferences and mechanisms that a useful model must represent.

Finally, identifying an agency relationship does not establish that the principal’s objective is socially desirable. A manager can faithfully maximize shareholder returns while imposing an externality on others, and a contractor can meet a purchaser’s specification while producing an outcome with broader social costs. Conversely, an agent’s departure from a principal’s preference is not automatically inefficient or unlawful. Evaluating an institution requires distinguishing contractual alignment from legal obligations, distributional effects, and social welfare.