RESEARCH DOSSIER

Principal Agent Contracts and Moral Hazard

Abstract. This research dossier studies principal-agent contracts as arrangements of delegated action, information, risk allocation, and enforcement. It develops the hidden-action problem through participation and global incentive compatibility, binary implementation, likelihood ratios, linear incentives, monitoring, limited liability, multidimensional tasks, team production, and relational enforcement. Conditional mathematical examples are derived explicitly, while empirical results from several economies are compared without pooling incompatible outcomes. The source review distinguishes original theoretical texts, published abstracts, full-text contemporary evidence, and provisional working-paper findings. The synthesis connects compensation design to information production, organizational structure, selection, and welfare incidence. It establishes the limits of inferences from observed outcomes and identifies conditions under which more monitoring or stronger financial rewards may improve or impair economic performance. The work presents an original comparative interpretation of established research, without claiming a new economic theory or independent econometric replication.

Keywords: Principal-agent problem; moral hazard; hidden action; incentive compatibility; participation constraint; contract theory; risk sharing; performance pay; monitoring; agency costs; empirical identification; JEL D82, D86

The economic object of delegation

Agency is an arrangement in which one party assigns decision authority or productive responsibility to another and receives some of the consequences. The principal supplies resources, exercises a contractual right, or purchases a service. The agent selects an action that influences the principal's payoff. The relationship becomes an economic incentive problem when the action is difficult to specify, verify, or enforce and when the parties' preferred actions differ. The same person or firm can be principal in one relationship and agent in another. A shareholder delegates to a board, the board to an executive, and the executive to operational managers. Institutional identity is therefore relational.

Moral hazard is the incentive-sensitive choice of an action under an allocation of consequences in which the decision maker does not personally bear all marginal consequences of that action. Hidden action is a central case in which the decision is privately chosen and unavailable for directly enforceable contracting. The decision may be effort, care, maintenance, project selection, product quality, reporting, loan repayment, medical consumption, or risk exposure. Each application requires a specification of the feasible alternatives, what the parties learn, and which signals enter the agreement. Akerlof-style type selection is a neighboring problem; this dossier centers on the mechanism that changes choices after the contract becomes operative.

The economic analysis measures the difference between feasible arrangements. An observed accident or loss proves neither shirking nor an inefficient contract. Bad luck can produce failure even under high effort. Conversely, success can result from an advantageous shock despite neglected maintenance. Thus attribution of action from outcomes is an inference problem and contract design is a mechanism problem. The relevant benchmark is the best feasible allocation of risk and effort under the actual information and enforcement environment. The word moral in the historical term does not establish a judgment of character.

Agency exists in market transactions and organizational hierarchies, but the canonical bilateral model is a deliberate abstraction. The shareholder may be diversified, the employee may value security, the borrower may lack liquidity, and the provider may possess specialized information that the consumer cannot evaluate. These institutional characteristics determine which simplifying assumptions can be accepted. A scientific account must articulate the mechanism before making a comparative claim about incentives.

Intellectual development and the problem of risk sharing

The modern literature transformed delegation from an intuitive managerial problem into a constrained optimization problem. In corporate finance, Jensen and Meckling (1976) connected agency costs to ownership and financing. In the 1979 literature, Shavell's risk-sharing study and Holmstrom's observability analysis located the central trade-off in compensation contingent on an imperfect record of the agent's actions. The result is an intellectual bridge between insurance and the internal organization of production.

The analytical obstacle is that an insurance arrangement can efficiently protect an individual against randomness while simultaneously changing the individual's private benefit from costly protective action. Similarly, a salaried employee can receive a stable income under uncertainty but have little monetary reason to raise an unobservable margin of productive effort. Introducing a contingent transfer restores an incentive at the price of exposing the agent to variability. The strength of that incentive is itself a choice variable.

The methodological advance of Grossman and Hart (1983) separates the choice of action from the problem of finding a least-cost contract that implements it. This is valuable because differentiating an agent's objective function at a proposed action establishes neither feasibility against every other action nor an optimal decision for the principal. Mirrlees (1999) identifies the restrictions necessary for reasoning based on local optimization and analyzes special environments where near-first-best performance is possible. The chronological publication date of 1999 is used for the latter journal record rather than assigning the journal article an earlier manuscript date.

A further step is the treatment of actions as a vector rather than an intensity. Holmstrom and Milgrom (1991) show why an apparently powerful single measurement can be unsuitable for a job containing competing activities. The contemporary evidence therefore includes monitoring interventions, labor compensation, administrative audits, and the allocation of responsibility within firms. The object survives these historical changes: how to induce valuable actions within real informational and institutional constraints.

Timing, information and participation

Consider a risk-neutral principal and an agent with increasing concave utility u(w) from money w and action cost c(a) measured in utility units. The outside option is U0. At the contracting date the principal proposes a schedule w(s) based on a public, enforceable signal s. The agent accepts or rejects; conditional on acceptance, chooses a from a feasible set A. Nature draws s with conditional probability p(s|a) and produces an economic return y, possibly different from the signal. The transfer is made under the stated schedule. This timing excludes interim renegotiation and unmodeled private types. Both may be added explicitly in other environments.

For a finite signal space, expected utility is U(a,w) = sum_s p(s|a)u(w(s)) - c(a). The participation condition requires U(a,w) >= U0. The global incentive compatibility condition requires U(a,w) >= U(b,w) for every b in A. A risk-neutral principal maximizes E[y|a] - E[w(s)|a] minus monitoring or administrative costs. The action and transfer together form the solution; maximizing output independently of wage cost generally gives the wrong result.

Participation determines whether the relationship forms. Incentive compatibility determines which action is privately preferred if it forms. An agent may take a contract exactly because the expected payoff compensates for risk or effort, without taking the action the principal anticipates. A contract can pass one condition without satisfying the other. A complete analysis also specifies feasibility restrictions such as statutory minimum pay, bankruptcy, collateral, termination rules, or an upper bound on transfers. These conditions modify the feasible set and can determine whether implementation exists.

Expected utility is used here as the baseline, not as a universal behavioral description. Action cost may include time, fatigue, foregone opportunities, responsibility, or other sources of disutility. Some decisions confer intrinsic satisfaction; others are carried out through cooperative norms. Such considerations can be included in the utility specification if they affect choices, but measured monetary cost and subjective utility remain distinct objects.

Observable action and the first-best benchmark

If action is perfectly observable, verifiable, and enforceable without cost, consider a contract enforcing a and providing a constant payment w. The agent obtains u(w)-c(a). A binding participation requirement yields w = u^{-1}(U0+c(a)). Because the principal is risk neutral and the agent risk averse, a constant transfer minimizes expected payment for a fixed utility promise: Jensen's inequality gives E[u(w)] <= u(E[w]). The principal chooses the enforceable action that maximizes E[y|a]-u^{-1}(U0+c(a)). This is a first-best risk-sharing benchmark within the stated model.

With an action-dependent cost in utility units, the marginal monetary compensation for a higher continuous action is c'(a)/u'(w) when differentiable. A planner with transferable utility or a money-metric cost can use a different surplus representation. Mixing a disutility in utils with a cost in dollars is a mathematical error. The separation between units is important when comparing experiments that measure time, monetary incentives and subjective preference.

Perfect observability alone does not yield this benchmark when the observed action cannot be verified by a court or enforced through a reliable agreement. The parties might observe each other's contribution while a third-party adjudicator cannot assess its quality. A forcing contract must be feasible, and enforcement must not impose additional welfare costs. Without those requirements, constant pay plus instructions does not automatically implement the desired behavior.

The benchmark is also not a prescription that every agent ought to be insured completely. It is conditional on the risk-bearing and technology assumptions. A principal with limited financial capacity may be unable to absorb output risk efficiently, and an agent who obtains productive information may need discretion. The first best helps separate losses from hidden action from losses arising for independent reasons, including financing constraints, production uncertainty, and restricted contracting rights.

Binary effort and the cost of incentives

Let high effort H and low effort L generate high output with probabilities pH and pL, where 0 < pL < pH < 1. Let cH > cL. Contract payments are wG following success and wB following failure. High effort is incentive compatible exactly when

(pH−pL)[u(wG)−u(wB)]≥cH−cL.(p_H-p_L)[u(w_G)-u(w_B)]\ge c_H-c_L.

High effort participation requires pH u(wG)+(1-pH)u(wB)-cH >= U0. When both constraints bind, and the required utility values are attainable by admissible wages, write delta = (cH-cL)/(pH-pL). Then u(wG) = U0+cH+(1-pH)delta and u(wB) = U0+cH-pH delta. This solution equalizes marginal implementation cost subject to the two binding constraints in the unrestricted two-action case with strictly concave utility and positive probabilities. It can fail if an imposed wage floor makes the lower payment infeasible or if other actions have been omitted.

The calculation has three immediate implications. A more discriminating output distribution requires a smaller utility difference to implement a given costly action, holding the action costs fixed. Greater action cost requires a larger utility difference. Greater risk aversion can increase the monetary compensation needed for given incentive and participation utility levels, but comparisons must normalize preferences and wealth rather than treating the curvature coefficient as a complete sufficient statistic.

For a deterministic illustrative computation, set u(w)=sqrt(w) for nonnegative w, U0=4, cL=0, cH=1, pL=0.4, and pH=0.8. The required difference in utility is 2.5; u(wG)=5.5 and u(wB)=3. Consequently wG=30.25 and wB=9, with high-effort expected wages equal to 26. Under verifiable effort, a flat wage of 25 achieves participation. The extra expected wage cost of 1 is the hidden-action risk premium in this example. Both candidate utilities equal 4 under the contract, so high-effort implementation is weak: a tie-breaking rule in favor of H or an arbitrarily small additional incentive must be specified.

With success worth 100 monetary units and failure worth zero, high effort yields expected principal profit 54, while a constant low-effort wage of 16 yields profit 24. Under the first-best high-effort benchmark, profit is 55. These values are constructed, not estimated. They identify a transparent separation of productive value, expected wages and the incremental incentive cost. Changing pH or imposing liability constraints can reverse the comparison; the efficient policy is selected by comparing all feasible actions.

Global implementation and likelihood ratios

For several actions, the principal must include every global deviation. The conditional cost program for action a is to minimize sum_s p(s|a)u^{-1}(v_s), with v_s denoting state utility, subject to participation and sum_s [p(s|a)-p(s|b)]v_s >= c(a)-c(b) for every feasible b. Under concave increasing u, its inverse is convex, and the constraints are affine. For finite actions and signals, this is a convex minimization problem whenever the inverse is defined on a convex utility domain. Actions with no feasible solution are excluded from the principal's comparison.

The approach reproduces the structure of the implementation-cost method within an additive utility specification. It distinguishes a feasible transfer determining behavior from the principal's ranking of productive outcomes. In a continuous-action problem the agent's first-order condition alone can be inadequate. For an elementary independent counterexample, let the induced objective over a in [0,2] be (a-1)^2. The derivative vanishes at a=1, which is its minimum. A principal choosing a=1 through a local stationarity constraint has chosen an action the agent would reject in favor of either endpoint. Additional curvature or global conditions are essential.

When the target action has full support and the appropriate constraints bind, the first-order conditions for state utilities give a useful characterization. Let lambda be the multiplier on participation and mu_b the nonnegative multiplier on the target action's comparison with b. At an interior wage,

1u′(ws)=λ+∑bμb(1−p(s∣b)p(s∣a)).\frac{1}{u'(w_s)}=\lambda+\sum_b\mu_b\left(1-\frac{p(s|b)}{p(s|a)}\right).

An outcome strongly diagnostic of the target relative to binding deviations may receive a higher reward, even if that outcome is not directly the most valuable to the principal. This formula requires regularity for the multipliers, strictly positive target probabilities and interior transfers; boundary solutions have inequalities rather than this equality. When target probability is zero in a particular state, the ratio is undefined and the original optimization should be used.

The economic unit of evidence is a comparison of probabilities conditional on actions. Ranking signals by their unconditional frequency or their dollar value conflates informativeness with productivity. A monotone wage schedule needs additional conditions; it is not a free theorem of incentive compatibility.

Informativeness, relative evaluation and risk

An additional signal is useful for designing a contract when it changes the information available about relevant deviations. If z has the same conditional distribution for all actions after conditioning on the original signal s, then z contributes no new information about action at a fixed s. In the risk-neutral-principal and risk-averse-agent setting with enforceable transfers, any wage w(s,z) can be replaced with a deterministic amount at each s that preserves conditional expected utility. Jensen's inequality weakly reduces expected wage cost. Thus such a signal cannot improve the hidden-action contract through incentive provision alone.

Conversely, additional discrimination of action can make contingent compensation less costly. The conditional likelihood ratios determine the usefulness of the new signal, not its correlation with revenue by itself. The baseline argument is motivated by Moral Hazard and Observability. A strict improvement requires suitable regularity, an active incentive constraint, and a signal that changes the relevant information. The new variable may still be useful for other purposes, such as learning about production technology, if it has no incremental value for the specified action problem.

Relative performance evaluation illustrates this logic. If two sales regions experience the same demand shock, another region's sales may help distinguish individual effort from common economic conditions. A contract based on adjusted sales can avoid rewarding or penalizing the individual for common movements. But when agents can manipulate the comparison group, coordinate reported output, or affect one another's performance through cooperation, relative evaluation introduces strategic interactions. Its usefulness must be assessed relative to the resulting game.

For two outcomes with common shock theta and independent agent-specific noise, suppose y=a+theta+epsilon and peer output z=theta+eta. All shocks have zero means and are mutually independent. A signal y-kz has noise variance Var(theta)(1-k)^2+Var(epsilon)+k^2 Var(eta). Minimization gives k=Var(theta)/(Var(theta)+Var(eta)), assuming a positive denominator. Perfectly measured common shocks give k=1, while very noisy peer signals produce a smaller correction. When the peer measure responds to a, this calculation no longer preserves the action loading; both its risk and its incentive effect must be recalculated.

Linear incentives and a transparent continuous benchmark

The familiar linear-normal-quadratic example is useful because the risk-sharing cost and the action response can be shown explicitly. Assume output y=a+epsilon, where epsilon is normally distributed with zero mean and variance sigma squared. The principal receives value y monetarily, the agent has constant absolute risk aversion r>0, the monetary cost of action is k a^2/2 with k>0, and compensation is w=alpha+beta y. Effort is nonnegative and the feasible beta is nonnegative. Normal income and exponential utility imply a certainty equivalent

CE=α+βa−ka22−rβ2σ22.CE=\alpha+\beta a-\frac{k a^2}{2}-\frac{r\beta^2\sigma^2}{2}.

For beta>=0 the agent selects a=beta/k. Let the monetary outside certainty equivalent be U0. Binding participation sets alpha=U0-beta a+k a^2/2+r beta^2 sigma^2/2. Expected principal profit after substituting the participation constraint is a-k a^2/2-r beta^2 sigma^2/2-U0. Inserting the effort response yields beta/k-beta^2/(2k)-r beta^2 sigma^2/2-U0. The strictly concave quadratic is maximized at

β∗=11+krσ2.\beta^*=\frac{1}{1+kr\sigma^2}.

This is optimal within the specified one-period linear class, not a proof that linear compensation is globally optimal for arbitrary environments. If variance tends to zero, the optimal share tends to one. If risk aversion, residual noise or the cost curvature increases, beta falls in this benchmark. The effect of cost curvature on the share should be separated from its direct effect on effort.

For k=2, r=0.5, and variance 1, beta equals 0.5 and effort equals 0.25. Holding k and r fixed, reducing variance to 0.25 raises beta to 0.8 and effort to 0.4. The increase follows from a specific assumption: the monitoring improvement lowers residual noise without changing action technology, preferences, enforcement, or signal manipulation. A field study must determine which of these conditions is credible. The stylized numerical inputs carry no empirical units beyond their internally consistent normalization.

The example highlights the meaning of incentive intensity. A large bonus paid independently of an individual's actions may have no marginal monetary incentive despite raising total compensation. A small but steep change around a target can generate a large conditional incentive while leaving total pay modest. Contract evaluation therefore requires the schedule, its slope, its relevant thresholds, and the distribution of performance rather than its advertised bonus pool.

Monitoring as an endogenous economic choice

Monitoring is an input into the production of contractible information. It can also be a direct enforcement instrument, a source of managerial learning, or a signal affecting the agent's interpretation of the relationship. A useful analysis specifies which of these mechanisms is operating and which costs are borne by each party. Monitoring costs include equipment, data collection, oversight, dispute adjudication, lost discretion and errors. Some are monetary expenditures and others are welfare costs, and a social evaluation must place them in consistent units.

Let m denote a chosen monitoring intensity and V(m) the optimized gross contractual surplus after solving the action and compensation problem at that intensity, excluding monitoring cost C(m). The principal chooses m to maximize V(m)-C(m) over the feasible set. For an interior differentiable solution V'(m)=C'(m), but boundaries and discontinuities are possible. A low-cost informative additional signal weakly expands the feasible contract set if it can be ignored. This statement holds for a fixed environment; introducing a monitoring system may itself change preferences, available actions, or the cost of cooperation.

In the linear-normal example, let residual variance S(m) decline with monitoring, with all other primitives fixed. The optimized profit before monitoring cost is 1/[2k(1+krS(m))]-U0. Its derivative is -r S'(m)/[2(1+krS(m))^2]. The sign is nonnegative for S'(m)<=0. An interior choice equates this quantity to C'(m). This calculation provides a measurable condition, rather than the universal recommendation that firms collect more data.

The physical existence of a monitoring device does not imply the production of usable evidence. A device may record movement but not workmanship, presence but not comprehension, speed but not customer safety. An optimal record system must address both sensitivity to the action of interest and the false-positive rate for compliant behavior. False positives impose expected sanctions on productive agents, weakening risk sharing and, potentially, participation. The monitoring problem is consequently joint with contract design and procedure for challenging incorrect observations.

A useful distinction concerns the value of measurement before and after a contract can be adapted. If observations arrive only after the contract expires, they may influence a subsequent relationship but cannot enforce the original schedule. If principals can change compensation after learning favorable information, initial promises lose credibility unless the contract or repeated relationship constrains their discretion.

Limited liability, wealth and enforceability

A contract that requires a large penalty in an unfavorable state may be formally optimal without being collectible. Limited liability means that the agent's payment or loss is bounded by available wealth, law or contractual protection. An incentive comparison requires rewards feasible in each signal state; participation must be evaluated separately. In binary effort with a risk-neutral agent, threshold spread is (cH-cL)/(pH-pL), with costs in money units. If a permitted maximum wage spread is smaller, the target action cannot be implemented by that signal and transfer alone.

As a constructed example, suppose pH=0.8, pL=0.4 and cH-cL=12 monetary units. High effort requires a success-minus-failure wage difference of at least 30. If wage regulations and budget restrictions require wB>=10 and wG<=35, the maximum spread is 25, so the high-effort target is infeasible. Simply solving a differentiable unconstrained problem would misstate what the principal can achieve. Additional signals, different technology, training, job reassignment or a legally enforceable future bonus may expand the feasible set. Each changes a specified primitive or contractual instrument.

Limited liability also creates a potential divergence between privately attractive and socially valuable risks. An equity holder receives the residual after obligations are satisfied; a sufficiently adverse outcome can exhaust equity while further losses fall on creditors or other claimants. Debt covenants and collateral alter the private payoff distribution. A full welfare calculation includes the distribution of losses, risk premiums demanded by other claimants, expected resolution costs, and transfer rules rather than interpreting any increase in risk as evidence of opportunism.

A risk-neutral and wealthy agent who can pay a franchise fee and receive the entire output may internalize the marginal productive return in simple models. That limiting result helps explain why leasing and residual claimancy matter. It is conditional on wealth, enforceability, no additional external effects, and the absence of other uncontractible actions such as asset maintenance. When those conditions fail, residual claimancy may align one margin and distort another.

Multiple tasks and distorted performance measures

Jobs ordinarily contain activities whose outputs vary in measurability. A worker may produce units, preserve equipment, assist colleagues, and detect defects. Some margins are immediately recorded and others manifest only after a delay. The multitask analysis of 1991 supplies the intellectual foundation for recognizing that contract design changes the direction of effort as well as its total level.

An independent illustrative example makes the trade-off transparent. Let an agent allocate exactly one unit of activity between measured production e and unmeasured maintenance 1-e, with e in [0,1]. The principal's gross return is 4e+8(1-e)=8-4e. The agent receives alpha+beta e and bears a personal adjustment cost 2(e-0.25)^2. The agent's unconstrained optimum is e=0.25+beta/4, clipped to [0,1]. With beta=0, gross return is 7; with beta=2, e=0.75 and gross return is 5. The bonus increases the recorded production metric while lowering gross return. This is a hypothetical allocation example, not a claim that maintenance is always more valuable than production.

A stronger incentive on an observed action can be appropriate when the action has a large return and the complementary margins are protected by other instruments. The relevant variables include substitution between tasks, task assignment, minimum quality rules, asset ownership, training, monitoring of inputs, and implicit rewards. An institution may split duties among specialists, adopt a balanced set of indicators, or use a lower-powered financial reward combined with experienced supervision. Each arrangement carries a cost and changes the production relationship.

A performance measure should therefore be judged by alignment with the economic objective conditional on the agent's feasible responses. The score can be statistically correlated with value before the incentive is introduced and lose that relationship when the agent learns how to optimize the score. In that sense, validation must be policy-dependent. A principal who evaluates a bonus program by the same metric used to calculate the bonus risks treating desired performance and strategic response as a single outcome.

The central implication is not that proxies are fundamentally unusable. Rather, a proxy provides economic value when the behavioral changes it induces increase the objective sufficiently to justify the associated risk, reallocation and administration. An independent quality measure, preferably collected without sharing the reward threshold, helps determine whether the contract has improved what its designer actually values.

Teams, hierarchies and shared performance

If many agents contribute to a common output, individual effort can become harder to identify. A production team often receives a shared payment, yet each member controls a different input. Under an equally divided fixed team reward, the agent captures only part of the marginal return to individual effort while bearing its full private cost. This can lead to free riding, despite the mutual advantages of higher effort.

For n risk-neutral agents, suppose total output equals the sum of individual efforts and each receives one n-th of output. Each agent bears cost k a_i^2/2. Individual optimization gives a_i=1/(nk). The social optimum over efforts, ignoring fixed rewards and assuming unit value of output, gives a_i=1/k. With n=5 and k=1, total equilibrium effort under equal sharing is 1, compared to efficient total effort 5. A principal offering each agent a full marginal unit of team output could restore effort in this stylized environment, but total transfers would then no longer equal output state by state. This example explains the budget constraint rather than recommending unlimited bonuses.

Holmstrom's team analysis establishes the relevance of free riding and the role of contract arrangements beyond simple balanced sharing in the formal multiagent setting. More informative individual records, peer observation, contribution-based allocation, and repeated cooperation can mitigate the problem. Such institutions introduce further incentive constraints: peers may exchange favorable evaluations, managers may allocate desirable assignments opportunistically, and common outcomes may encourage workers to conceal one another's failures.

Hierarchy changes the identity of the agent. A team supervisor who is paid for aggregate output may direct training or resources to the highest-productivity subordinates, alter which workers remain employed, or discourage valuable tasks whose returns are hard to measure. Those actions affect the distribution of pay and access to productive opportunities. Agency analysis must therefore extend to the measurement of managerial actions and their downstream consequences.

Team performance measures and relative performance measures solve distinct problems. Sharing a total can strengthen cooperation among agents but dilute individual marginal returns. Relative evaluation can filter common shocks but can discourage helping colleagues. A feasible combination may use a common reward for shared quality and a separate individual signal for a margin that can be attributed reliably. Its merit turns on the production technology and the possibility of strategic behavior.

Repeated relationships, career incentives and discretion

Many contracts operate in repeated interactions where both sides expect future gains. A manager can reward performance that is observable internally yet difficult to verify externally. Such a reward is credible only if the principal expects the cost of breaking its promise to exceed the immediate benefit of doing so. Reputation, future cooperation and termination may support the implicit arrangement.

In a simple stationary repeated situation, suppose deviation gives an immediate private gain G relative to compliance, and loss of cooperation reduces each future period's payoff by L>0. With discount factor d in (0,1), a sufficient no-deviation inequality under permanent credible punishment is G <= dL/(1-d). For G=12 and L=4, the required discount factor is at least 0.75. The calculation is hypothetical. It presupposes a detectable deviation, a feasible punishment, and credible responses on the other side; an actual relationship may have stochastic survival and private monitoring.

Baker, Gibbons and Murphy (1994) identify conditions under which subjective and objective measures reinforce each other and conditions in which explicit contracts undermine informal incentives. A compulsory objective payout may lower the surplus that supports discretionary rewards. Alternatively, a reliable objective measure may make a continuing relationship more valuable. The scope of the repeated game must therefore include principal behavior as well as agent behavior.

Career concerns introduce another dynamic incentive. If a worker's current performance informs future employers about ability, today's effort may influence future opportunities even without a current explicit bonus. Early-career incentives can be substantial, but signals also mix ability and effort. A worker may prioritize visible achievements over investments whose returns appear later. Compensation, evaluation horizons and promotion systems must be evaluated jointly.

The principal may also have private information about the availability of future work. If the agent expects promised rewards to be revised once effort is sunk, the action can be inefficient even when the published schedule looks generous. Credible commitment, transparent criteria and independent review can reduce the discrepancy between formal and expected rewards. Contract design is therefore an institutional arrangement of rights and future expectations, not merely a formula mapping output into pay.

Incentives in financial intermediation and ownership

Financial contracts allocate future cash flows and thereby distribute marginal consequences of project choice. Managers may select investments, lenders choose screening and collection effort, borrowers allocate resources to repayment, and intermediaries generate or verify information. Each relationship has a distinct principal, agent and hidden margin. The theory is strongest when those roles are explicitly mapped.

The ownership problem examined in the 1976 theory of the firm links the financing structure to the incentives of decision makers. Equity-based pay can expose management to owners' residual return, but exposure to upside and downside is not necessarily symmetric across all claimants. A highly leveraged firm may prefer a project with larger upside and worse expected total value if the adverse outcomes fall mainly on creditors. Bankruptcy law, covenants, liquidity and capital requirements shape the feasible incentive regime.

For example, suppose a project returns 130 with certainty and requires repayment of 100, so equity receives 30. An alternative project returns 220 with probability 0.4 and zero otherwise. Expected gross value is 88; expected equity is 48, assuming nonrecourse debt and no other costs. Thus the equity claimant favors the second project while expected gross value falls by 42. The creditor's expected receipts fall from 100 to 40. Expected total distributed value is 88, equal to 48+40. The calculation describes a hypothetical already-financed decision, not a market estimate or a prediction of an equilibrium financing contract. At issue, lenders may price, prohibit or monitor the decision.

Executive compensation is similarly multidimensional. Share price incorporates aggregate information, expectations and common economic shocks; it is neither a pure effort measure nor a complete welfare measure. Deferral, vesting, clawbacks and holding requirements can alter horizons and downside exposure. A principal must also account for the agent's undiversified wealth and the possibility of manipulating information disclosure. The optimal schedule depends on which choices can be verified, not solely on the nominal value of outstanding shares.

In regulated financial institutions, creditors, customers, deposit insurers, regulators and taxpayers may share exposure. An individually incentive-compatible contract between owners and managers can still generate external costs. The private design problem should be separated from the social evaluation and supplemented by an assessment of the consequences borne by third parties.

Insurance, healthcare and the welfare meaning of utilization

Insurance alters the allocation of financial loss conditional on an event. Ex ante moral hazard concerns preventive action before a loss; ex post moral hazard in health economics commonly refers to a change in demand after exposure to an insurance contract or illness. The latter usage captures a price response, even where the relevant utilization is observed. It extends beyond the strict unobservable-effort case. Einav and Finkelstein (2018) review this empirical distinction and the complementary roles of experimental and structural methods.

Consider a risk-neutral person choosing prevention e with private monetary cost c(e), loss probability p(e), and potential loss L. Under insurance that reimburses fraction q of the loss, and assuming a fixed premium independent of the current choice, the private objective includes c(e)+(1-q)p(e)L. At an interior minimum c'(e)=-(1-q)p'(e)L. When p'(e)<0, additional coverage lowers the private reward to prevention. The direction of the resulting effort change requires the appropriate second-order conditions and a stable action technology. An experience-rated premium or contractual prevention requirement can change the derivative.

Medical utilization has a different structure because a service often confers direct value to the patient. Suppose marginal benefit declines with quantity and marginal social resource cost is constant C. A patient paying share q of C chooses quantity at marginal benefit qC in an unconstrained interior model. The gap between private and social marginal cost may lead to additional consumption. But illness shocks, liquidity constraints, treatment effectiveness, public-health spillovers and access barriers complicate the welfare comparison. Insurance may enable a treatment whose value exceeds its real resource cost. Additional utilization alone is an economic response, not a sufficient statistic for waste.

The first-year Oregon study by Finkelstein and colleagues (2012) uses a lottery opportunity rather than randomly assigning universal insurance acquisition. The published abstract reports that selection raised the probability of coverage by about 25 percentage points, increased use of care and reduced some measures of financial burden. Those outcomes represent multiple welfare margins of the intervention. A defensible inference preserves the distinction between assignment effects and effects of actual coverage. The estimates concern the low-income population and institutional program studied.

A scientific insurance evaluation consequently combines risk protection, marginal behavioral responses, premium revenue, administrative cost and distributional consequences. It distinguishes a change in service use from a change in physical health and from a change in out-of-pocket spending. Deductibles, coinsurance, provider incentives and prior authorization involve different parties and require separate enforcement and welfare assessments.

Empirical identification: from outcome changes to hidden actions

An empirical researcher observes a realized outcome, a measurement system and an institutional agreement. The action is latent by construction if the target is a strict hidden-action problem. The research challenge is to establish a counterfactual outcome or an independent action proxy under a different contract. A difference between groups with different compensation can reflect existing differences in ability, risk, task composition and selection into the arrangement. Merely controlling for measured characteristics does not remove private information.

Let Z represent randomized assignment to a contract offer or monitoring intervention, D actual receipt, and Y a designated outcome. With complete randomization, a difference E[Y|Z=1]-E[Y|Z=0] identifies the intention-to-treat effect under consistency and appropriate treatment of interference and missingness. When take-up is imperfect, a Wald ratio of that difference to E[D|Z=1]-E[D|Z=0] identifies a local average treatment effect for compliers if relevance, exclusion, independence and monotonicity hold. The excluded direct effect is particularly consequential when an offer itself conveys trust, information or managerial intention.

Panel designs can help separate within-worker changes from composition. If a firm adopts piece rates, comparing output of continuing workers before and after the change removes time-invariant individual heterogeneity, but concurrent changes in demand, technology or task opportunities remain. Entry and exit records should be analyzed in their own right because the compensation regime may change which agents participate. A within-worker effect and a workforce-composition effect answer different questions; neither necessarily equals the welfare effect.

Structural estimation takes an explicit action and preference model to the data. It may infer costly effort or response parameters from choices across incentive schedules, then simulate unobserved contracts. Such simulations rely on maintained assumptions, identification of parameters, error structure and external validity. A flexible model that reproduces the observed output may still be wrong about another schedule if latent action and productivity are confounded. Validation using holdout variation or independent effort measures strengthens the exercise.

A useful empirical design separately varies available information, the compensation rule and enforcement. A factorial experiment can test whether monitoring without contingent pay alters actions, whether pay responds to information, and whether the combination is complementary. Where randomization is infeasible, institutional discontinuities or phased adoption may provide useful variation, conditional on defensible assumptions about sorting, anticipation and simultaneous policies.

Labor-market evidence: effort and composition

In the United States, the Safelite change studied by Lazear (2000) provides evidence that a change in compensation can alter both individual incentives and workforce composition. The original published abstract reports a 44 percent increase in output per worker after the shift to piece rates. It also identifies sorting as a relevant prediction of the personnel-economics framework. The percentage is a proportional output statistic, not a probability difference or a portable estimate of all workers' response to contingent pay.

This case is useful precisely because the choice of compensation changes more than one margin. Suppose initially there are five workers generating 10 units each. After a change, the same five generate 12 each and five entrants generate 16 each. Average output changes from 10 to 14; total output changes from 50 to 140. The 40 percent average gain mixes a 20 percent continuing-worker change with a composition change. The 180 percent total increase also reflects employment. These numbers are illustrative; they are not the Safelite data.

The implication for personnel records is that the denominator must be specified. Output per worker, output per hour, value added per unit of labor cost and quality-adjusted output are different measures. Performance pay can raise the count of completed tasks while leaving hourly productivity unchanged if workers supply additional time. It can raise output per hour while increasing rework or generating later maintenance costs. Identification of net value requires measurement of input use and downstream quality.

Agency theory therefore supplies a research program rather than a direct conversion from a productivity estimate to an optimal pay rate. The researcher must estimate how the wage schedule changes actions and participation, and how the resulting output is valued. Welfare analysis also considers exposure to income variability and whether workers choose the new contract under a meaningful outside option.

Managerial incentives and the distribution of opportunity

The firm-level study of Bandiera, Barankay and Rasul (2007) extends the question from individual effort to managerial choices. Changing supervisors' pay to depend on subordinate output altered both mean and dispersion of worker productivity. The published abstract identifies changed assignment of managerial attention and the exclusion of some less-able workers as mechanisms. These findings are specific to the experimental firm and its organization.

A manager's reward may induce a choice of who receives assistance, where scarce resources are allocated, and which workers are retained. If managerial time is productive in combination with worker ability, allocating it to high-ability workers can raise aggregate output. The same decision may widen gaps in opportunity and earnings. An economic objective specifying only aggregate output treats that distribution as irrelevant; a different objective can value minimum service or equal opportunities directly.

There is no contradiction between a more effective incentive for the manager and less favorable outcomes for some workers. The manager's contract can correctly implement the principal's stated aggregate objective while producing effects omitted from that objective. The empirical problem is to measure the induced distributional changes separately from changes in mean productivity.

A firm assessing such a policy needs records of original assignments, time with supervisors, individual capacity proxies measured before treatment, pay, and retention. Changes in the dispersion of observed output can result either from changed treatment of the same workers or from a change in the selected sample. Reporting both surviving-cohort outcomes and flows in and out of the workforce helps establish the mechanism. A more complete evaluation additionally accounts for the cost of recruitment and for the continuation value of investing in lower-productivity workers.

Education and public administration

In Indian nonformal schools, Duflo, Hanna and Ryan (2012) evaluated a combined attendance-monitoring and salary-incentive intervention. The journal abstract reports a decline in teacher absence of 21 percentage points and an improvement of 0.17 standard deviations in student test scores. A change in absence is an absolute probability difference, whereas a standardized test-score effect uses a distribution-dependent unit. Their combination supports an interpretation extending beyond the attendance metric, but the overall treatment does not isolate a separate pure salary slope effect. The publisher lists an erratum; these quantitative statements are limited to the published abstract and are not a reanalysis of the replication package.

Public administration provides a related setting where the ultimate principal is dispersed. In the randomized study of Indonesian village road projects by Olken (2007), the higher formal audit probability reduced the discrepancy between project-account expenditures and independent engineering cost estimates. The abstract gives the probability change as 4 percent to 100 percent and the missing-expenditure reduction as eight percentage points. The measure is an expenditure discrepancy rather than directly observed private appropriation; it may contain measurement error or legitimate differences between records and engineering estimates.

The studies identify distinct margins: attendance and learning in one, expenditure discrepancies in the other. Both involve institutions capable of applying a credible action-contingent response. The effects should not be pooled into a common percentage because their denominators and mechanisms differ. A school principal's information technology, a public auditor's sampling regime, and a centralized pay schedule have different implementation costs and legal constraints.

Where public-service objectives are multidimensional, a principal must account for access, service quality, safety, and distribution of benefits. A narrow performance target could improve the recorded margin while shifting attention away from valuable activities. Experimental evaluation should collect both primary outcomes and prespecified secondary quality measures. Transparency about complaint and appeal procedures is important because enforcement based on faulty data can impose sanctions on legitimate behavior.

Consumer credit and selection after an offer

Credit markets frequently mix hidden information and hidden action. Borrowers may possess information about existing income risk, repayment capacity and intended use of funds at the contracting date. Once a loan is granted, the contract may alter repayment effort or project choice. Default is therefore an outcome jointly determined by privately known risk, realized shocks, contractual burdens and action.

The South African experiment of Karlan and Zinman (2009) randomized offer interest rates, interest rates subsequently revealed to borrowers who accepted the offers, and prospective benefits attached to maintaining good repayment standing. The design allowed the authors to separate some selection and incentive mechanisms. The institutional repository reproduces the original abstract, which characterizes evidence for hidden action as stronger than evidence for hidden information and gives an approximate, setting-specific moral-hazard share of default of 13-21 percent. This is the authors' approximation under their empirical framework, not a general fraction of loan defaults.

The timing matters. If a rate is announced before take-up, the accepted-borrower sample can change; if a lower rate is assigned after a borrower accepts the initial offer, selection on the new rate is curtailed by design. A future conditional benefit can affect incentives even if the immediate loan terms remain fixed. However, a repayment response may reflect liquidity effects as well as costly repayment effort. Mechanism claims rely on the authors' experimental variation and maintained interpretation rather than on default data alone.

Creditor and debtor welfare must also be evaluated separately. Lower default can reflect improved productive capacity, reduced consumption, assistance from family, asset liquidation or reduced willingness to take future loans. A high recovery rate is neither a complete measure of economic efficiency nor a complete measure of borrower welfare. A policy assessment includes cost of capital, contract participation, selection into borrowing, and dynamic effects on future access to finance.

Monitoring technology in small firms

The Kenyan public-transit experiment described by Kelley, Lane and Schonholzer (2024) introduces a concrete modification to owners' information. Monitoring devices were installed and data access through an app was randomized among owners. The abstract reports changes in contract terms, greater driver effort, reduced risky behavior, lower firm costs and higher profitability. Evidence of expansion is characterized as suggestive rather than definitive.

The setting highlights a point often lost when monitoring is treated as a fixed constraint: access to information can change the entire contract. An owner who previously relied on a daily remittance target may renegotiate that target or alter enforcement upon obtaining trip information. The reduced-form effect of access then includes those modifications. It does not equal the mechanical effect of a particular signal holding all contract terms fixed.

The intervention also illustrates a division between owners' private profit and the welfare of drivers, passengers and other road users. Lower risky behavior is an important potential external benefit, but a complete welfare analysis would require explicit measures of safety, work intensity and the cost of monitoring, among other variables. Profit growth may help small firms finance expansion, yet statistical support for expansion must be distinguished from a proven causal effect on firm scale.

For application beyond Kenya, it is necessary to examine the contract structure, device reliability, enforcement rules, credit access and the ability of workers to switch employers. Identical tracking hardware in two countries can have different effects because the underlying delegation problem differs. International comparisons should identify institutional variables that mediate mechanisms rather than assign a universal elasticity to monitoring.

Control, motivation and contemporary evidence

The improvement of information conditional on fixed preferences and fixed technology should be distinguished from a real intervention that changes the work environment. The online labor experiment by Herz and Zihlmann (2024) provides a counterexample to the unconditional claim that adding control always raises output. The publishers' text describes a restriction on declared unreadable pictures, with another dimension of work remaining difficult to enforce. Average performance fell by 5.5 percent against the counterfactual; losses were concentrated among difficult tasks and workers with greater pretreatment motivation. The control device was intentionally weak at preventing some forms of shirking, which limits generalization.

This does not overturn the formal proposition that an optional informative signal expands a fixed contracting set. The field intervention changes an actual work relationship and can shift motivation or effort allocation. The empirical and theoretical statements concern different operations. Treating them as contradictory would obscure the condition under which the mathematical proposition applies.

An emerging research direction independently varies pay and monitoring. The working-paper abstract of Cai, Luo and Wang (2025) reports randomized interventions involving newly hired workers in a Chinese automobile firm. Compensation and monitoring raised output through different channels: increased hours in one case and higher manager-rated performance in the other. This is provisional working-paper evidence, and manager ratings should not automatically be equated with independently validated productivity. Full-paper methods and current journal status were not independently verified here.

The immediate research implication is to distinguish a worker's time contribution from the quality and intensity of the activity performed during that time. If digital data become available, the analyst should specify whether the intervention changes actual supervision, the expected probability of detection, the agent's perception of trust, the compensation schedule, or all of these. Treatment labels such as monitoring or incentives conceal these distinct operations.

Privacy, data retention and the ability to challenge a performance record are economic design variables because they alter the costs and enforceability of the arrangement. They are also governed by jurisdiction-specific rights. This dossier makes no claim that a particular monitoring or sanction is legally permissible in every country. Scientific comparisons should explicitly account for the institutional feasible set.

Measurement conventions and reproducibility

A research dossier on hidden actions requires an observable-data dictionary and a mapping from those measurements to theoretical quantities. A probability of absence is a frequency over designated observation opportunities; its denominator may be teacher-days or scheduled visits. Labor output is a count or value over a stated period. Loan default follows a defined delinquency threshold and horizon. Audit missing expenditure is a constructed discrepancy. Healthcare utilization comprises events, expenditures or quantities of services; the choice of measure changes interpretation.

A data table for a new study should identify the following minimum fields:

Variable Unit and definition Primary limitation
Contract assignment Treatment indicator and date May differ from take-up
Contract exposure Actual schedule, duration, eligibility Endogenous compliance
Action proxy Hours, attendance, observed care Measures only a component of action
Outcome Defined output, quality, loss, repayment Noise and reporting incentives
Agent type proxy Pre-intervention measures Incomplete information about type
Monitoring Signal access, timing, audit probability Distinguish access from use
Cost Monetary resources and opportunity cost Valuation and incidence
Participation Acceptance, exit, entry Selection and attrition
Context Market, jurisdiction, technology Limits external validity

A replication should record the data vintage, acquisition permissions, sample restrictions, randomization unit, comparison group, attrition, weights and variance estimator. Confidence intervals and uncertainty in estimates should be reproduced in preference to point estimates alone when original data or full tables are accessible. Where only published abstracts are verified, as in several cases in this dossier, reported numerical values remain abstract-level findings, without new confidence intervals or independently reproduced regressions.

Percentage changes require a reference value. Percentage-point effects require risks or proportions on the same scale. Standard-deviation effects depend on the reference population and outcome transformation. Combining these magnitudes into one aggregate index would entail additional valuation and measurement assumptions. This dossier uses comparisons of documented mechanisms rather than an unsupported cross-study pooled effect.

A reproducible analytical calculation differs from an empirical replication. The numerical examples here use declared inputs and equations; another reader can reconstruct them without access to restricted data. They are diagnostic calculations for model implications. The examples carry no claim to newly measured parameters or universal optimal contracts.

Comparative institutional interpretation

The sources cover production workers in the United States, a managerial compensation intervention in a firm's labor organization, teachers in India, roads in Indonesia, borrowers in South Africa, public health insurance in Oregon, transit operators in Kenya, workers in online labor markets, and a recent Chinese manufacturing study. This selection reflects variation in institutional mechanisms and source accessibility. It is a purposive comparison, not a globally representative sample.

The unit of institutional comparison is the concrete modification of the agent's incentives. In the labor cases the payment function or information set changes. In the credit case the timing and terms of offers change. In the insurance case a lottery changes access to coverage. In the public audit case the probability of oversight changes. A common abstract agency problem can illuminate these settings, but it cannot convert their different treatment effects into identical structural parameters.

Setting Primary manipulation Reported outcome margin Principal inference limit
Safelite Compensation transition Output per worker Sorting and concurrent changes
Indian schools Attendance records and pay Absence and tests Bundled treatment
Indonesian roads Audit probability Missing expenditure Proxy for unobserved diversion
South African lending Sequential offer and loan terms Default Local clientele and model dependence
Oregon Medicaid Lottery opportunity Use and financial strain Imperfect insurance take-up
Kenyan minibuses Owner access to records Contract and operating behavior Profit and broader welfare differ
Online tasks Control threshold Correct task output Specific restriction and task setting
Chinese auto firm Bonus and monitoring Hours and rated performance Working paper; ratings as proxy

External validity requires a mapping across task structure, market alternatives, the risk-bearing capacity of agents and technology. For example, an attendance indicator for a single-teacher school is not an automatic substitute for the measurement of educational quality elsewhere. An audit regime with credible penalties cannot be replicated merely by generating more reports in a jurisdiction where findings are unenforceable.

Research comparing countries should preserve this institutional specificity, articulate the mechanisms that travel, and report what remains contingent. The scientifically portable result is a conditional relationship among incentive constraints, information and welfare. The numerical magnitude belongs to the study population and intervention unless further transport evidence is supplied.

Original analytical synthesis and boundary conditions

The formal and empirical record supports a synthesis in which compensation, information and organization are jointly chosen elements of agency arrangements. The principal's problem changes across three margins: the cost of implementing a target, the value of the target relative to alternatives, and the reliability of measuring the consequences. A well-designed contract addresses all three. In this synthesis the terminology, examples and comparisons are authorial analysis of established results; no new named theory or empirical discovery is asserted.

An especially important distinction concerns information that is useful conditional on an arrangement and a monitoring intervention that changes the arrangement itself. The first can expand available contracts when the signal is optional and the environment stays fixed. The second can introduce new costs, strategic actions, selection or motivational effects. The Kenyan example shows a positive configuration with contract adaptation; the online-task experiment shows a negative response to a particular restrictive control. Each is consistent with a correctly specified feasible-set analysis, once the intervention's actual effects are included.

A second distinction separates incentive intensity from the expected value of rewards. The success bonus in the binary example is effective through differences in conditional utility. The linear example connects a marginal share to effort and risk. In the managerial experiment the reward changes allocation of support and employment selection. In dynamic credit arrangements a future benefit can affect present incentives. Thus equal expected compensation need not produce equal behavior, and equal behavior in one recorded margin does not demonstrate equivalent economic value.

A third distinction concerns private objective attainment and social welfare. A firm can raise output per worker while transferring additional risk to workers. An insurer can reduce utilization at the cost of protection or access. A lender can lower default while imposing costly adjustment on households. A financial contract can raise expected equity value by transferring losses to creditors. Comparative welfare analysis requires the gains and costs to all affected parties. These are established microeconomic considerations applied jointly to the evidence, not a separate doctrine.

One practical implication for research design is to evaluate any incentive arrangement using two sets of outcomes: the signal that enters the contract and independently assessed measures of the principal's economic objective. The first identifies whether the schedule functions as intended; the second detects distortions or transfers of effort. A third set of measurements, including participation, earnings risk and external effects, clarifies welfare incidence. This is an analytical recommendation derived from the reviewed mechanisms rather than a claimed universal algorithm.

The scope is bounded by the assumptions in each derivation. The baseline model assumes known type, commitment, and an enforceable signal. The examples add risk aversion, wealth bounds, task allocation or repetition one at a time. Real settings may combine private types, hidden action, bargaining, incomplete contracts, dynamic learning and multiple principals. A rigorous analysis reports which deviations matter instead of treating a simplified example as a structural estimate.

Open research questions and extension requirements

Several questions follow from the present analysis. First, what dimensions of latent action can be reliably measured without changing the behavior being measured? A valid research design should compare a monitoring intervention with a separate, independently collected performance measure and document the agent's knowledge of both. Changes in metric accuracy and changes in perceived control should be independently varied where possible.

Second, how can a contract system adapt over time while making current promises credible? Learning about productivity can improve future task assignment, yet retrospective revision can make present rewards unreliable. Longitudinal experiments should record agreement versions, dates of disclosure, and decisions to terminate or renew. Repeated-interaction models can then be evaluated against a specified data-generating process rather than fitted only to outcomes from surviving relationships.

Third, how do heterogeneous risk preferences, private types and mobility constraints influence participation in incentive contracts? A controlled experiment with a common assignment can estimate an offer effect; understanding voluntary contract adoption requires behavioral and market information on the available alternatives. Where contracts change selection, policy simulations must model both the supply of effort and participation.

Fourth, when does an improvement in measurement permit a genuinely better welfare outcome? An answer requires more than evidence of greater recorded performance. The relevant data include monitoring costs, quality, worker earnings risk, prices, service accessibility, and the allocation of costs to third parties. An experimental design that randomizes access to signals while tracking these measures can identify different efficiency mechanisms.

Finally, the empirical record reviewed here should be deepened through independent access to original data, full tables, appendices, published corrections and subsequent versions of working papers. The analysis has intentionally bounded claims drawn from abstracts and indexed summaries; it does not reproduce individual empirical estimates. An extension could address the newer literature with systematic inclusion criteria and comparable effect measures. That would be a separate research undertaking, not a claim of completeness for the entire international literature.

Conclusion

The principal-agent contract organizes economic cooperation where control, consequences and information are distributed among parties. Hidden action requires implementation through feasible contingent rewards, monitoring, institutional rights or credible continuation of relationships. The fundamental trade-off is between the productive effects of induced action and the costs of providing incentives, sharing risk, obtaining information and enforcing commitments.

The formal examples establish conditional rather than universal conclusions. A risk-averse agent may require additional expected compensation for bearing incentive risk; informative monitoring can reduce that cost under a fixed environment; limited liability can rule out otherwise desirable schedules; multiple tasks and shared outcomes create distinct incentive distortions; and repeated interactions can support rewards that are difficult to enforce through a third party. Each proposition depends on its explicit assumptions.

The empirical record demonstrates that changes in incentives and monitoring influence both actions and the composition or organization of economic relationships. The cases in employment, education, administration, credit, health insurance and transit differ substantially in the mechanisms observed and in the units of reported effects. Their strongest joint implication is that the design of a contract must be evaluated together with the measurement system and institutional context.

A complete welfare assessment follows the economic consequences beyond the monitored signal and beyond the bilateral private objective. The decisive question is which feasible contract increases value after accounting for risk, administrative costs, distributional effects, strategic responses and relevant external consequences. This is the organizing conclusion of the dossier and the basis for its place within Microeconomics.

References

  • Baker, George, Robert Gibbons, and Kevin J. Murphy. 1994. Subjective Performance Measures in Optimal Incentive Contracts. Quarterly Journal of Economics 109 (4): 1125-1156. DOI: 10.2307/2118358. Research record.

  • Bandiera, Oriana, Iwan Barankay, and Imran Rasul. 2007. Incentives for Managers and Inequality among Workers: Evidence from a Firm-Level Experiment. Quarterly Journal of Economics 122 (2): 729-773. DOI: 10.1162/qjec.122.2.729. Research record.

  • Cai, Jing, Sai Luo, and Shing-Yi Wang. 2025. Money or Monitoring: Evidence on Improving Worker Effort. NBER Working Paper 33977. DOI: 10.3386/w33977. Research record.

  • Duflo, Esther, Rema Hanna, and Stephen P. Ryan. 2012. Incentives Work: Getting Teachers to Come to School. American Economic Review 102 (4): 1241-1278. DOI: 10.1257/aer.102.4.1241. Research record.

  • Einav, Liran, and Amy Finkelstein. 2018. Moral Hazard in Health Insurance: What We Know and How We Know It. Journal of the European Economic Association 16 (4): 957-982. DOI: 10.1093/jeea/jvy017. Research record.

  • Finkelstein, Amy, Sarah Taubman, Bill Wright, Mira Bernstein, Jonathan Gruber, Joseph P. Newhouse, Heidi Allen, Katherine Baicker, and the Oregon Health Study Group. 2012. The Oregon Health Insurance Experiment: Evidence from the First Year. Quarterly Journal of Economics 127 (3): 1057-1106. DOI: 10.1093/qje/qjs020. Research record.

  • Grossman, Sanford J., and Oliver D. Hart. 1983. An Analysis of the Principal-Agent Problem. Econometrica 51 (1): 7-45. DOI: 10.2307/1912246. Research record.

  • Herz, Holger, and Christian Zihlmann. 2024. Adverse Effects of Control? Evidence from a Field Experiment. Experimental Economics 27: 469-488. DOI: 10.1007/s10683-024-09823-3. Research record.

  • Holmstrom, Bengt, and Paul Milgrom. 1991. Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design. Journal of Law, Economics, and Organization 7 (Special Issue): 24-52. DOI: 10.1093/jleo/7.special_issue.24. Research record.

  • Holmstrom, Bengt. 1979. Moral Hazard and Observability. The Bell Journal of Economics 10 (1): 74-91. DOI: 10.2307/3003320. Research record.

  • Holmstrom, Bengt. 1982. Moral Hazard in Teams. The Bell Journal of Economics 13 (2): 324-340. Research record.

  • Jensen, Michael C., and William H. Meckling. 1976. Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure. Journal of Financial Economics 3 (4): 305-360. DOI: 10.1016/0304-405X(76)90026-X. Research record.

  • Karlan, Dean, and Jonathan Zinman. 2009. Observing Unobservables: Identifying Information Asymmetries with a Consumer Credit Field Experiment. Econometrica 77 (6): 1993-2008. DOI: 10.3982/ECTA5781. Research record.

  • Kelley, Erin M., Gregory Lane, and David Schonholzer. 2024. Monitoring in Small Firms: Experimental Evidence from Kenyan Public Transit. American Economic Review 114 (10): 3119-3160. DOI: 10.1257/aer.20210987. Research record.

  • Lazear, Edward P. 2000. Performance Pay and Productivity. American Economic Review 90 (5): 1346-1361. DOI: 10.1257/aer.90.5.1346. Research record.

  • Mirrlees, James A. 1999. The Theory of Moral Hazard and Unobservable Behaviour: Part I. Review of Economic Studies 66 (1): 3-21. DOI: 10.1111/1467-937X.00075. Research record.

  • Olken, Benjamin A. 2007. Monitoring Corruption: Evidence from a Field Experiment in Indonesia. Journal of Political Economy 115 (2): 200-249. DOI: 10.1086/517935. Research record.

  • Shavell, Steven. 1979. Risk Sharing and Incentives in the Principal and Agent Relationship. The Bell Journal of Economics 10 (1): 55-73. Research record.