On this page
A CEO is hired by shareholders to maximize firm value. The CEO knows far more about the firm's operations, strategic options, and her own effort level than shareholders do. She may rationally prefer perquisites, empire-building acquisitions, or reduced work effort — activities that benefit her but reduce shareholder value. Shareholders can observe results (quarterly earnings, stock price) but cannot monitor exactly how hard the CEO worked, what alternatives she considered, or whether her decisions truly maximized value or primarily protected her own position. This is the principal-agent problem at the apex of the corporate hierarchy — and some version of it runs through every delegated decision-making relationship.
The setup
The principal-agent problem arises whenever:
- A principal (shareholder, patient, employer, voter) delegates decision-making to an agent (manager, doctor, employee, politician)
- The agent has private information about their own actions, effort, and capabilities that the principal cannot directly observe
- The agent's interests are not perfectly aligned with the principal's — the agent may benefit from actions that cost the principal
Two distinct information problems create this:
Hidden action (moral hazard): the agent's effort or behavior after contracting cannot be fully observed. An employee paid a flat salary may shirk; a fee-for-service physician may over-treat; a commissioned salesperson may recommend products that earn higher commissions.
Hidden information (adverse selection): the agent knows more about their own type or the decision environment than the principal. A financial advisor knows more about the return-commission tradeoff of recommended products than the client.
What happens — and why
The principal's problem is to design a contract that aligns the agent's incentives with her own interests as closely as possible, given that the agent's actions cannot be fully observed.
Outcome-based pay: linking agent compensation to observable outcomes (profit sharing, equity, sales commissions, performance bonuses) gives agents a stake in principal-aligned outcomes. The SEC's executive compensation disclosure rules require detailed reporting of CEO pay structures — making visible how boards attempt to solve the principal-agent problem through compensation design.
Monitoring: direct observation of agent behavior (managerial oversight, audit requirements, performance reviews). Costly but reduces hidden action.
Reputation: agents who depend on long-term relationships have incentives to behave well — reputation destruction is a cost that aligns interests over time.
The DOL's fiduciary rule — requiring financial advisors to act in clients' best interests — is a regulatory response to the principal-agent problem in financial advice: advisors (agents) historically had incentives to recommend higher-commission products regardless of whether they best served clients (principals).
Where you see it in the wild
Hospital-physician relationships create principal-agent problems in healthcare. Hospitals hire physicians who know far more about appropriate treatment intensity than administrators monitoring costs. Fee-for-service payment creates incentives for over-treatment (more procedures → more revenue); capitated payment creates incentives for under-treatment (fewer services → higher margin). The CMS's value-based payment programs attempt to align physician incentives with patient outcomes rather than service volume — a direct principal-agent problem solution.
The fix (or why it's hard to fix)
Complete alignment of principal-agent interests would require the agent to bear all risk — a pure profit share contract. But agents are typically risk-averse and less diversified than principals; bearing full risk is inefficient and often impossible. The practical solution is always second-best: balancing incentive alignment against risk-sharing. This tradeoff explains why corporate pay always mixes salary (risk insurance for the agent) with equity (incentive alignment for the principal) — the optimal contract depends on how well performance is measurable and how risk-averse the agent is.
