14. Solutions to Principal-Agent Problems in Firms - Semantic Scholar

11 downloads 2201 Views 123KB Size Report
and agents) call for different “solutions” to the principal-agent problems. ..... Solutions based on monitoring agent behavior are necessarily more intrusive,.
14. Solutions to Principal-Agent Problems in Firms GARY J. MILLER

There are many settings in which one economic actor (the principal) delegates authority to an agent to act on her behalf. The primary reason for doing so is that the agent has an advantage in terms of expertise or information. This informational advantage, or information asymmetry, poses a problem for the principal—how can the principal be sure that the agent has in fact acted in her best interests? Can a contract be written defining incentives in such a way that the principal can be assured that the agent is taking just the action that she would take, had she the information available to the agent? Solving this problem is a matter of some concern for patients dealing with their doctors, clients dealing with their lawyers, or celebrities dealing with their publicists. It is also a crucial concern for business firms dealing with their employees. Especially in the twenty-first century, employees are often hired precisely because they have information available that is unavailable to the managers of a firm. Making sure that employee expertise is put to work in the interest of the firm can make the difference between success and bankruptcy–as illustrated by the relative performance of Southwest Airlines compared to much of the rest of the airlines industry. This paper examines the large principal-agency literature as it relates to management patterns in the firm. A powerful conclusion emerges, not from any one segment of the literature as much as from a bird’s-eye view of the literature as a whole, that there is no unique “solution” to the principal-agent problems in a firm. Instead, a Coasean “contingency” theory can be constructed in which different conditions inside the firm (characterized by production technology, severity of information asymmetry, and relative risk-preferences of principals and agents) call for different “solutions” to the principal-agent problems. While the first significant papers in principal-agency theory were developed independently of Coasian theory, this chapter of the Handbook will try to establish that there is a natural connection between the two. Coase (1937) hypothesized that transactions may be structured in different ways—in particular, some can be better managed via hierarchy within a firm rather than by the market between firms. This insight has led, in recent years, to a large and successful literature on the “boundaries of the firm”—examining when transactions are best

349

C. M´enard and M. M. Shirley (eds.), Handbook of New Institutional Economics, 349–370.  C 2005 Springer. Printed in the Netherlands.

350

Gary J. Miller

organized within the firm, and when they should be organized between firms. (See the articles by Joskow and Klein in this Handbook.) But the same Coasean logic can be applied to those transactions that occur strictly within the firm—notably, those hierarchical transactions between employer and employee. In particular, incentives, monitoring, and cooperation can and do play different roles in the infinite variety of contractual forms that can govern transactions within the firms. Which kinds of within-firm transactions are best governed by powerful incentives? In which transactions should the firm invest in high levels of monitoring capacity? In which should a long-term, cooperative relationship be encouraged among employees, or between firm and employees? In a Coasian manner, I argue that different intra-organizational transactions can best be structured by different kinds of employment contracts. I also suggest that the literature on principal-agency theory can help to explain why particular contracts are best applied to particular transactions. Fundamentally, principal-agency theory is about trade-offs; it is not surprising that the nature of the tradeoffs shifts subtly as conditions change, resulting in different kinds of solutions in different setting. Furthermore, different solutions create very different types of firms. Some firms that rely heavily on incentives, like marketing powerhouse Pepsico, are known for a free-wheeling, risk-taking, entrepreneurial style of decisionmaking. Others that use monitoring more, like manufacturing giant General Motors, are often characterized as “bureaucratic”—implying that employees tend to avoid risk-taking by looking to hierarchical superiors and standard operating procedures for justification of their actions. In still other firms, like Southwest Airlines, the observer sees high levels of cooperation and teamwork within the firm. These behavioral characteristics may be thought of as derivative of the different kinds of contracts and transactions that emerge in response to different types of principal-agent relationships. While there are multiple solutions, no one solution is perfect. Or rather, there are multiple solutions because no one solution is perfect. Except in “ideal” conditions (with zero risk-aversion and no information asymmetry), agency costs persist with each style of attempted solution. However, there are certain indicators that suggest those situations in which one type of solution may be systematically better than others. One of the tricks of good management is therefore to be sensitive to trade-offs between different kinds of costs associated with different transactional arrangements—within as well as between firms. The purpose of this paper is to show how principal-agency theory has evolved to help explicate differences in within-firm managerial styles. Section I discusses those firms that tend to rely on high levels of risky financial incentives, thereby mimicking the market. Section II discusses why some firms cannot efficiently use outcome-based financial incentives, and turn to bureaucratic oversight, thereby mimicking the state. And Sections III and IV demonstrate that, for some firms, both incentives and monitoring can be improved on by long-term cooperation, between supervisors and subordinates, and among teams of subordinates. The

Solutions to Principal-Agent Problems in Firms

351

paper concludes with a re-affirmation of the Coasean claim that efficiency in transactions calls for variety of contracts, even within the general structure we call “the firm”.

1. SOLUTIONS BASED ON INCENTIVES LINKED TO AGENT OUTCOMES Principal-agency theory is grounded in the study of information asymmetries. The agent takes actions that determine (in combination with a random component) an outcome of interest to the principal. Probably the first influential paper to develop these themes was Spence and Zeckhauser (1971) on insurance. A risk-averse homeowner, facing a ten percent chance of a $100,000 loss, would be willing to pay more than $10,000 to be insured against the loss. A risk-neutral insurance company would be willing to accept any premium larger than $10,000 to cover the loss. Both sides could be made better off by shifting the risk to the insurance company for the appropriate premium; failure to make the trade would be a manifestation of inefficient risk-bearing. However, as Spence and Zeckhauser demonstrate, there are potential obstacles to this efficient trade. When the individual is insured, the individual has no reason to avoid actions that may actually increase the probability of the loss. This notion of moral hazard became crucial for both the insurance literature and the subsequent principal-agency literature. The insurance company would like to be able to monitor whether or not the homeowner engages in moral hazard; when this is impossible or too costly, then the insurance company may have to refrain from fully insuring the homeowner, thereby forcing the homeowner to face the loss as an incentive to discourage moral hazard. That is, the insurance company will have to turn down a full insurance contract that would make both it and the homeowner better off! This is an example of market failure in the market for risk. “Given [the insurer’s] limited information-monitoring capability, his selection of the optimal insurance payoff function is a second-best exercise. Neither complete risk spreading nor appropriate incentives for individual action will be achieved. To find the optimal mixture of these two competing objectives is a difficult problem, here as in the real world.” (1973:387)

The problem of balancing incentives and efficient risk-bearing transfers neatly and quickly to employment relationships. The sales agent (for example) is more risk averse than the employer, just as the homeowner is more risk-averse than the insurance company. The variation in auto sales is in part due to economic conditions beyond the control of either the sales agent or her employer. The agent would prefer a smaller, fixed wage over a risky commission with a higher average payoff—and the employer would prefer the smaller fixed wage as well. Efficient risk-bearing would require that the employer insure the employee against the risk by paying the flat wage. The match to the insurance problem is nearly perfect.

352

Gary J. Miller

But as in the insurance industry, the problem is moral hazard: the flat wage leaves the agent with no incentive to avoid behaviors (shirking) that increase the risk of a “bad month” for auto sales. Nor can anyone work backward from the outcome to deduce the agent’s effort level. When a month with low car sales occurs, the sales agent can blame those external conditions that are known to impact sales. Efficiency in incentives must be traded off against efficiency in risk-bearing. The flat wage contract has insufficient incentives for effort, but contracts that provide incentives for agent effort are more costly to the employer. This problem of balancing efficiency in risk-bearing against incentives, when effort levels are costly to discover, became the defining problem for principalagent theory. Harris and Raviv (1978) assume that an agent’s costly effort, combined with an exogenous random variable (e.g., the state of the economy), determine some outcome of interest to the principal. Assuming that the outcome (but not the agent’s effort) is public information, Harris and Raviv are interested in discovering if and when information about the agent’s effort level is valuable. That is, should the employer spend money to monitor whether the agent is shirking? They conclude: “There are no gains to be derived from monitoring the agent’s action when the agent is risk neutral.” (1979: 233) This is the case because an outcome-based incentive contract can be written that will induce optimal effort, consistent with efficient risk-sharing. In particular, the optimal contract is one in which the risk-neutral agent bears all the risk, making a fixed payment to the principal. However, there are potential gains from monitoring when the agent is both effort-averse and risk-averse (Harris and Raviv 1978). In that case, the absence of information about effort means that the principal must motivate effort by outcome-contingent incentives. But these incentives imply that the risk-averse agent must bear some of the risk, since the outcome is in part dependent on an exogenous random variable. Therefore, efficient risk-sharing is sacrificed for the second-best incentive contract. Shavell (1979) and Holmstrom (1979) elaborate on this basic result—that efficiency losses result from a situation in which the principal is ignorant of the action taken by a risk-averse and effort-averse agent. Shavell reiterates that, just as in the insurance problem, suboptimal risk-sharing is necessary to induce effort; equivalently, optimal risk-sharing undermines the incentives for efficient levels of effort. Shavell also shows that if the agent is risk averse, the most efficient possible contract is one that nevertheless forces the agent to bear some of the risk—i.e., his fee always depends to some extent on the risky outcome. Furthermore, “the achievable level of welfare approaches the first-best level” as the agent’s effectiveness (i.e., the impact of the agent’s efforts compared to Nature’s impact) approaches either zero or one (1979:56). If the agent’s impact is zero, the outcome is not affected by the agent’s effort, and optimal risk-sharing (a flat wage) is first-best . If the agent’s effort swamps the effect of the random variable with nearly zero cost, then a relatively small (infinitesimal) imposition

Solutions to Principal-Agent Problems in Firms

353

of risk on the risk-averse agent will produce a nearly-best solution. It is in the intermediate levels, when the agent has a significant but not overwhelming impact, that efficiency losses are greatest. Holmstrom (1979) provides a significant paper that rounds out our understanding of the principal-agent problem. He agrees with Shavell that when the agent is risk-averse and effort-averse, “Pareto-optimal risk sharing is generally precluded, because it will not induce proper incentives” (1987:74). Given this constraint, Holmstrom demonstrates that the second-best compensation scheme will be one in which the agent’s share of the output is always increasing in output. The shape of that relationship may be steep or flat, convex or concave, depending on the agent’s risk preferences. The more risk-averse agents will, in general, have flatter payoff functions in output (more like a flat wage). For firm settings in which agent action is costly to monitor, the best solution is one in which agents are paid based on an outcome-based commission, and allowed to go their own way without much hierarchical supervision. The employee-employer relation resembles a market transaction. The difference between such an employee and the outside contractor may, therefore, be a fuzzy one. Incentives for Sales Agents

The literature on principal-agency theory found a ready application to the subject of sales force compensation. In an early paper, Basu et al. (1985) note that, within the firm, sales is a particularly appropriate setting for applying principal-agency theory. Normally, they point out, “it is very difficult to monitor the actual efforts of each salesperson” (1985: 268); and a low or high level of sales in a given month may be due either to the agent’s efforts or to sales conditions. A firm has available a variety of different compensation contracts for sales personnel, from straight commission, to variable commissions (based on sales), to straight salary, to various combinations of salary and commissions. Basu et al. (1985:277–79) show how this variation in compensation schemes can be understood in terms of different parameters in the solution to the principalagency problem. In particular, they apply and extend Holmstrom’s second-best sharing (compensation) scheme. The best compensation will be monotonic in sales, but the compensation function can be many shapes. The more risk averse the sales agent, the flatter the compensation plan in sales: a decrease or increase in sales should result in less impact on the sales agent’s compensation. The more risk tolerant, the more the firm should rely on intense incentives in the form of higher commissions and lower flat wages. If the agent becomes more risk acceptant with income, then the commission rate should increase with sales. If the agent’s risk tolerance is constant with income, then the agent’s compensation function will be a straight line against sales revenue. The more uncertainty in the environment, the less control the sales agent has over sales; this implies that the sales agent will receive a higher fixed salary parameter and a lower commission rate—and a lower expected total compensation (Basu et al. 1985:282).

354

Gary J. Miller

Empirical research on sales has supported many of these results. Eisenhart (1988) found that routine sales jobs like operating a cash register were paid with fixed salaries, while sales positions that involved establishing a close relationship with the customer were paid with a commission. This is consistent with principal-agency theory in that employers bore the risk for those positions in which employees could be cheaply and easily monitored, but risk was shifted to those employees where monitoring was more expensive. Presumably, agents with the most risk-aversion were more likely to stick with lower-paid but safer fixed wages, operating the cash register, while agents who had the least risk-aversion were those who accepted the commission pay for the less easily monitored personal sales positions. Experiments provide further support for the empirical applicability of principal-agency theory. McLean Parks and Conlon (1995) provide a controlled laboratory study with owner/employee dyads that produced solutions to a mathematical problem. The employee’s contribution was to spend money for computer-generated solutions that stochastically increased in accuracy with the agent’s expenditures. The pairs negotiated compensation contracts composed in part of a flat wage and in part of an outcome-based bonus. Each dyad participated in twenty trials, negotiating a contract and completing the exercise in each trial. In half of the dyads, the agent’s expenditure could be monitored by the owner in the next period, and in half, the agent’s expenditure could never be made public. Another, independent treatment was the profitability to the owner. In profitable environments, dyads agreed to an increased use of outcomecontingent bonuses when monitoring was not possible. This result is entirely analogous to Eisenhardt’s observation that commissions are used more frequently for difficult-to-monitor sales assignments. However, in unprofitable environments, dyads used contingent bonuses less frequently. Most interestingly, employees tended to over-produce in all settings. In the profitable environments, employees spent on average 2.6 times the rational amount on information, regardless of monitoring. In the unprofitable treatments, the ratio was 1.7 or 5.6, depending on whether monitoring was or was not possible (McLean Parks and Conlon 1995: 826). The striking over-investment in information indicates that some other motivation (e.g., competition with other subjects or an inner desire to solve the math problem) was driving the experimental subjects to provide the overly large expenditures that they did. Like sales agents, CEOs have outcome measures which are impacted by individual effort and exogenous random variables. Jensen and Murphy (1990: 227) use regression (of CEO wealth on corporate performance) to find that a typical CEO’s wealth increases by only $3.25 for every $1000 in the corporation’s value. They believe that this regression coefficient is so low that it must be “inconsistent with the implications of formal agency models of optimal contracting. Garen (1994) believes that Jensen and Murphy exaggerate the importance of their finding. While Jensen and Murphy find an average pay-performance sensitivity coefficient b of CEO compensation on organizational performance,

Solutions to Principal-Agent Problems in Firms

355

Garen estimates a different such b term for different industries, and examines determinants of this coefficient. In his view, principal-agency theory implies that “As the output the agent produces becomes inherently riskier, the insurance component of pay is increased and the incentive component is reduced.” This hypothesis is consistent with his finding that the CEO pay-performance sensitivity coefficient is lower in firms with a higher variability in return. In industries where exogenous variables have more impact on outcomes than does the CEO’s own effort, the compensation is less fixed to those outcomes. An outcome-based contract, in which monitoring is minimal, will almost certainly attract certain kinds of employees—in particular, those who are least risk-averse, and most appreciative of the absence of close scrutiny by hierarchical supervisors. The more risk-neutral are the employees who are attracted to such an employment opportunity, the more intensely incentivized is the optimal contract. Some production jobs are compensated by piece-rates, which rely on outcomebased compensation rather than supervision to motivate effort. For example, Lazear (1996) examined the effect of a change from fixed wages to piece rates in a firm that installed auto windshields. Lazear argued that productivity rose by approximately a third in this case, and wages by 12 percent. He also found evidence for a selection effect—those who were less willing to have their compensation linked to outcome (due to either risk- or effortaversion) left the firm and were replaced by employees who preferred the new compensation package. One can readily imagine that the recruitment process and compensation practice reinforce each other, and result in an entrepreneurial, market-like environment within the firm. Tournaments

In general, information about an individual’s effort is economically valuable— especially if that information can be gathered with little expense. One form of information that can often be cheaply collected is that regarding relative performance: how well do agents do compared to each other? Can that information be helpful to firms trying to solve principal-agent problems? The assumption is that, if all agents are facing similar risks, then relative performance provides information about relative effort. Even a hard-working agent may not produce much in the way of sales when some event, such as a recession, reduces overall demand. Considering that the best possible outcomebased compensation is upward sloping (and steeply upward sloping for some agents), then a recession such as this would produce low compensation for all— and presumably poor motivation and morale as well, considering that everyone would be aware that the sales force was not responsible for the recession. A tournament among sales agents may be implemented by paying the topselling agent a bonus or prize, paying the second-ranked agent a smaller bonus, and so on. The motivation to do as well as possible would persist—even in the face of a recession that would discourage an agent based on a commission only.

356

Gary J. Miller

The prize may be a promotion—with opportunities to continue promotion up the line. For any such prize, competition may lead to increased effort by all agents (Lazear and Rosen 1981). The effort required to win a tournament depends on how hard other agents work—so a tournament implies the existence of a game between agents. Nalebuff and Stiglitz (1983) show that, as in a non-competitive output-based compensation scheme, the first-best solution is available with tournaments among risk-neutral agents. With risk-averse agents, a tournament can induce a “better” second-best compensation scheme than can an outcome-based noncompetitive scheme. It can induce risk-averse agents to supply more effort, and to take “riskier” and more profitable actions (1983:23) The competitive tournament is more desirable when agents face common risks, such as the risk of a recession. It is less useful when agents faces different risks, because a comparatively low or high outcome may be due to the agent’s unique risk factors, not the agent’s effort. Each agent’s chances of winning the tournament depend on the particular risk factors that affect each other agent’s performance. In the limit, when all agents face no common risk, the best possible compensation scheme is a non-competitive outcome-based contract. (1983:25); with enough agents facing a common risk, a competitive tournament can approach the first-best solution. Eriksson (1999) finds consistent results in executive compensation. He notes that tournament theory predicts a convex relationship between compensation and rank, as each promotional “prize” needs to be greater than the last. This result—convexity—is affirmed in his data on Danish executives. As Nalebuff and Stiglitz point out, tournaments do have a downside. A big disadvantage seems to stem from differential abilities: “the presence of a sure loser destroys everyone’s incentive to work hard.” (1983:40). So does the presence of a sure winner. Tournaments work best among agents of equal abilities. A system of “handicapping” by abilities requires shared, public knowledge of each person’s abilities. Considering that most people regard themselves in the upper half of any ability measure, there is likely to be little possibility of mutual agreement on abilities. Another limitation is that a tournament can eliminate any cooperation between agents. In some settings, such as sales, cooperation is not desirable; but in others, such as designing a new computer, it may be essential to the production process. As Lazear (1989) notes, “It is not sensible to create rivalry by setting up implicit promotion contests. Up to this point, the focus has been on contracts in which the agent’s compensation is pegged to observed (individual or relative) outcomes. The presence of “high-powered” incentives creates a non-hierarchical, market-like atmosphere that encourages entrepreneurialism, competition, and risk-taking. But Holmstrom showed (1979) that contracts based on direct observation of the agent’s actions (e.g. shirking) are better as long as the cost of monitoring is sufficiently low. The following section suggests that most firms find monitoring preferable most of the time—and the result is a very different kind of firm.

Solutions to Principal-Agent Problems in Firms

357

2. SOLUTIONS BASED ON DIRECT MONITORING OF AGENT ACTIONS The assumption for the previous section was that information about agent actions is costly to obtain, forcing compensation based on risky outcomes. Holmstrom also observes that since the source of the problem is information asymmetry, a “natural remedy is to invest resources into monitoring of actions” (1978:74). Of course, monitoring may be prohibitively expensive, or it may be inaccurate, and therefore just as risky as outcome-based incentive contracts. But Holmstrom (1979) shows that even “noisy” monitoring—where the information about agent performance is subject to random error—can lead to a Pareto improvement. A partial reliance on even a faulty, subjective measure of agent effort can allow the use of a smaller, less risky compensation scheme, making both the principal and agent better off. As long as the noise in the monitoring signal is not perfectly correlated with the noisy outcome of the production process, then the combination serves (like an investor’s portfolio) to decrease risk. As a result, we would expect to see performance-based contracting solely in those principalagent relationships where monitoring is very noisy or very costly, or agents are relatively risk-acceptant. Otherwise, we would expect to see principals invest in monitoring. In the sales context, Joseph and Thevaranjan (1998) show that monitoring permits a less intense incentive system. This allows the firm to pay risk-averse sales staff a smaller and flatter wage, making both happier; it also allows the firm to hire more risk-averse sales staff. If the monitor cannot observe all aspects of the agent’s productive behavior equally well, then the agent has an incentive to try to please the monitor by emphasizing the more observable behaviors (and spending less effort on the less observable aspects of productive behavior). But monitoring changes the nature of the employment relationship drastically. Monitoring, I will argue, is filled with hazards, and results inevitably in a much more hierarchical and bureaucratic organizational culture. Hierarchical Monitoring

Clearly, monitoring is not a trivial task, and there is no special reason to believe that the principal will be the best person to supply the monitoring services. Someone else, a specialist whom we may call a “supervisor”, may be able to provide the monitoring services more cheaply and/or more expertly than the principal herself. The result will be a three-level (at least) hierarchy, in which the principal attempts to induce a supervisor to act in the principal’s own best interests, as the supervisor is monitoring the agent. Tirole (1986) shows that, ideally, the principal will pay the supervisor a flat wage for learning about the agent’s efforts; this will result in a decrease in agency costs as the risk-averse agent is in turn paid a (smaller) flat wage that makes both principal and agent better off than would be possible without monitoring. However, this ideal solution is not immune to a problem that Tirole identifies: collusion between the supervisor and the agent. The supervisor and agent can

358

Gary J. Miller

make themselves better off (with appropriate side-payments) when the supervisor acts to protect the agent from the sanctions that should result from inadequate performance. In Tirole’s analysis, this makes hierarchies less efficient, and, frequently, more rigid, as a result of bureaucratic rules that the principal may impose to protect herself from this collusion. Tirole also allows that, outside of his model, there are situations in which collusion (more properly termed “cooperation”) may be helpful. This possibility will be discussed later in this essay. Even if the supervisor has no special technological advantage, there will be good reasons for the principal to hire a supervisor to supply the principal with monitoring services for the agent. One important reason is that the principal’s profit-maximizing offer will be a higher compensation if the principal observes a high effort; however, the principal may not be able to credibly commit to this “carrot”, leading the agent to doubt that the principal will ever hand it out (Strausz 1997: 341).1 Delegating control of the carrot to the supervisor may make it credible as an inducement, creating appropriate incentives for the agent to supply high effort, and making everyone (including the principal) better off. Strausz is able to prove (1997: 351) that the principal can achieve a strictly higher payoff by hiring a monitor than by monitoring herself. Creating Authority: Efficiency Wage

Solutions based on monitoring agent behavior are necessarily more intrusive, and potentially resented, than contracting for outcomes. Agents may dislike having their actions observed and sanctioned by supervisors. While outcomebased incentives resembles a normal market relationship—with all that that implies in the way of agent autonomy—contracts requiring monitoring may create an atmosphere of subservience, antagonism, and covert resistance. The sanction for non-performance may be getting fired—but as long as employment markets clear, there is no excess supply of employees in the marketplace, and consequently little wait before finding an equivalent job. The anonymity of the large labor market guarantees that the cost of getting fired is small, and provides little disincentive for shirking as long as unemployment rates are low. The result will be that the scope of the supervisor’s authority to monitor and direct employee behavior is sharply constrained. In order to give greater scope to supervisorial monitoring and direction, one answer is to contract at a wage level higher than the market-clearing wage. At this wage, called an “efficiency wage”, the quantity of labor supplied increases, meaning that the market no longer clears (Akerlof and Yellen 1986). Agents queue up seeking relatively scarce jobs, and the prospect of quitting and finding a replacement job looks more costly. The net result is that the agent’s willingness to accept a broad scope of direction and monitoring increases substantially. This 1 In Strausz’s model, the agent is risk-neutral, but neither the effort nor outcome are directly observable. The agent provides a service (like car maintenance or termite inspection) that the principal cannot tell for sure she received.

Solutions to Principal-Agent Problems in Firms

359

is precisely what Henry Ford was seeking when he began paying a $5 day at Ford Motor Company. (Raff 1988). Solutions Relying on Programmed Behavior: Bureaucracy

Each “solution” to principal-agent problems seems to initiate another set of problems. A “monitoring” solution creates the necessity of additional authority. But once the problem of creating authority in a firm is created, then comes yet another problem: agent attempts to influence authority. Milgrom (1988) points out that one of the last things the principal wants an agent to do is to spend his time and energy in attempts to lobby the supervisor in order to influence task assignments and distribution of rewards. These “influence activities”, at a minimum, detract from the effort being put into the job. And at the worst, influence activities can generate the kind of collusive coalition between agent and supervisor that Tirole (1986) warns us against. Consequently, principals will try to restrict the degrees of freedom of supervisors by binding their actions with rules. The supervisor will be judged by his willingness to abide by those rules. The more rules, and the more inflexibly they are enforced by supervisors and monitored by principals, the more the firm begins to look like a Weberian bureaucracy. A firm that designs a rule-bound hierarchy for itself will inevitably lose the capacity to respond quickly to a change in its operating environment. The fluidity and spontaneity of the market may be approximated in firms based on performance-incentive contracts, but will be scarce indeed in effort-monitoring organizations. Contingency Theory

The net result is that organizations which rely on monitoring solutions to principal-agent problems tend to look a great deal different than those based on outcome incentives. They are more hierarchical, more rule-bound, and more intrusive of agent autonomy. In a very interesting analysis, Holmstrom and Milgrom (1994) show that worker freedom and worker ownership of assets should theoretically co-vary with intensity of performance incentives: the three are complementary organizational characteristics, characterizing a more marketlike organizational setting. Furthermore, Holmstrom and Milgrom (1994) argue that each bundle of characteristics is more appropriate in a particular organizational setting. In some sales settings, for example, the agent will be asked to engage in multiple tasks, some of which will be hard to measure. In such a setting, incentivizing one easy-to-measure task may result in the employee spending too much effort on that task at the expense of other, equally important but less measurable, tasks: when the cost of measuring sales performance is high (e.g., because it involves team selling) or when hard-to-measure nonselling activities are important, it is more likely that the agent’s optimal incentives will conform with the attributes of

360

Gary J. Miller

employment: modest commissions, firm ownership of customers, and no right for the agent to sell the products of other manufacturers. (Holmstrom and Milgrom 1994:974)

In other cases, multi-tasking and joint production may not be a problem; then the firm should loosen the rules, allow the agent to “own” access to customers, and intensify incentives. In this agents (whether formally considered inside employees or outside contractors) may more closely resemble autonomous market agents than employees of a hierarchy.

3. SOLUTIONS BASED ON COOPERATION BETWEEN PRINCIPAL AND AGENT All of the contracting solutions discussed so far fall short of a first-best solution, as long as agents are risk-averse, effort-averse, and monitoring is costly. It is possible to minimize agency costs by selecting output-based incentive contracts when agents are efficacious or not very risk-averse, or monitoring contracts when monitoring is relatively inexpensive. But in each of these games, the Coasean prospect of negotiating to a first-best solution is an attractive solution. Managers who can somehow negotiate cooperative solutions on the Pareto possibility frontier have a skill that is worth a great deal on the market for managerial talent. Solutions Based on Gift Exchange

In the case of either output-based or monitoring solutions, it is clear what cooperation would look like. In output-based contracts, cooperation would consist of a trade of optimal risk sharing (flat wages) for effort; in monitoring contracts, cooperation would consist of a trade of increased agent autonomy (decreased monitoring) for extra effort. Akerlof (1982), in his article on “Gift Exchange”, proposes that such outcomes may well be negotiated in organizational settings. Akerlof used Homans’ (1954) classic study of “cash posters” to motivate his model. The “cash posters” registered payments to a utility company against customer ledgers. The firm set a standard of performance, and employees were paid a flat wage for satisfactory performance of this repetitive, even monotonous work. Since there was no compensation for work beyond the standard, Akerlof notes that “the standard economic model of contract would predict that workers set their work habits to meet the company’s minimum standards of performance” (1982:547). Surprisingly, however, cash posters exceeded the minimum by an average 17.7%, with a range from 2% to 46%. Akerlof interprets this extra effort as a “gift” to the firm. The firm reciprocated the employees’ extra effort with what Akerlof calls “leniency”. Leniency may be valued by employees both instrumentally and for its own right. Instrumentally, it is valued because it means a reduction in the possibility of

Solutions to Principal-Agent Problems in Firms

361

punishment. Leniency is also directly valued as “freedom” or “autonomy”. Employees experience a utility cost from “having someone breath down their neck”. A “good” (i.e., lenient) employer is one that will allow the employees the flexibility to arrive late occasionally when the children are ill, will allow changes in work rates to correspond to varying energy levels during the day, and will turn a blind eye to minor rule violations, like conversing on the job, that relieve tedium. The fact is that cash posting was a hard and dull job. A number of girls who were offered it had turned it down. The supervisors realized that “a group of young girls like this one” would have resented a managerial style of “bearing down” on them, meaning that the supervisors would have a still harder time getting recruits and maintaining production (Homans1954, 726). As a result, a variety of work “rules” were simply not enforced—notably the work rule forbidding conversation among the employees. They were convinced they could do their work without concentrating on it—they could work and talk at the same time. In theory, talking was discouraged. In practice, the supervisors made little effort to stop it . . . (Homans 1952: 727)

Homans notes that there was a time in the not-too-distant past in which relations between employees and supervisor were much more hostile. Supervisors “cracked down” on a variety of work rules that were intended to boost performance but had little real impact, and employees gave only the minimum effort, grudgingly. Greater effort by employees, and greater leniency from supervisors, emerged as a negotiated solution over time. The negotiated “gift exchange” was one in which supervisors got what they really wanted most (high output) and the employees got what they wanted most (a relaxed and lenient work environment). Repeated Game Solutions to Principal-Agency Problems

The problematic principal-agency relationships we have looked at resemble a prisoners’ dilemma game in that, in each, the Nash equilibrium is Pareto dominated by some outcome that is not a Nash equilibrium. For example, when monitoring is impossible and the agent is risk-averse, the Pareto preferred outcome is one in which the principal pays a low flat wage and the agent provides the optimal effort level. (Miller and Whitford 2002) This is not a Nash equilibrium because the flat wage induces moral hazard in the agent. Radner (1985) points out that, like a repeated prisoners’ dilemma game, the repeated principal-agent game could recapture these agency losses (and approach the first-best solution). The obvious way to re-capture some of the losses associated with agency problems is to play a repeated game. Radner shows that, with a sufficiently high discount factor, cooperative outcomes may be supported as Nash equilibria in a (non-cooperative) supergame. This result, a manifestation of the Folk Theorem to principal-agent problems, occurs with no incentive other than self-interest on the part of principal and

362

Gary J. Miller

agent; both are induced to play cooperatively by contingent strategies in which non-cooperative play by either side is met with punishment by the other. In the case of Homans’ cash posters, envisioning the outcome as a repeated game means that the employees’ “gift” of extra effort was in fact in equilibrium because it was enforced by the threat of reduced leniency; similarly, the supervisors’ leniency was enforced by the threat of reduced effort. In these cooperative solutions, optimal risk-sharing and effective incentives are reconciled in an interesting an important way. In the efficient equilibrium, the principal pays a flat wage, contingent on a long-term pattern of outcomes that is consistent with high effort by the agent. While the agent still has a shortterm incentive to shirk, the agent knows that shirking in the short-run will result in an imposition of unwanted risk in the long run. As long as the agent and the principal both value the long run enough, the threat of a risky compensation scheme will serve to avert moral hazard as effectively as the reality—and without the efficiency loss.

4. SOLUTIONS BASED ON COOPERATION WITHIN TEAMS Holmstrom (1982) discuss a particular aspect of principal-agent problems: team production. In many technologies, each member’s productivity depends on the effort levels of other agents in the team. As an example, consider a computer programming team that is trying to develop a new computer game. The output of the team may be readily observable, even though the efforts of each individual member of the team may be completely obscure, at least to someone outside the team. As Holmstrom argues, this creates a special form of moral hazard. There is no risk in the formal sense, because the output may be completely determined by the actions of the members of the team (with no random exogenous variable entering the production function). But even without risk, there is a similar problem of tracing backward from the observed output to the effort levels of individual members. Each agent would prefer to cut back effort while hoping that the other team members pick up the slack. The efficient outcome is not a Nash equilibrium because each would prefer to cut back when the other members of the team are working at the efficient level. In contrast to the single-agent case, moral hazard problems may occur even when there is no uncertainty in output. The reason is that agents who cheat cannot be identified if joint output is the only observable indicator of inputs.” (Holmstrom 1982:325)

Monitoring individual team members is always a possibility, but this implies an efficiency loss due to supervision, as long as supervision does not have zero cost. It is tempting, then, to ask if there isn’t an incentive solution–some way of allocating the team’s joint product so that moral hazard is eliminated. Holmstrom looks for a sharing rule that induces a Pareto optimal Nash equilibrium. He would

Solutions to Principal-Agent Problems in Firms

363

also like for such a solution to budget-balancing, requiring that the team’s output exactly equal the sum of all the payoffs to the productive team members. An allocation is budget-balanced if there is no “deficit” financing agent shares and no surplus left over. Holmstrom shows that these requirements are logically inconsistent. There is no budget-balanced allocation of the team’s revenue product that eliminates moral hazard in teams. As long as the sharing rule is budget-balanced, some team member will have an incentive to shirk: budget-balancing implies moral hazard. Dealing with team-induced moral hazard can be manifest within the team members, or at the level of the principal herself. Each of these possibilities, and some of the solutions attempted by the firm, are addressed in what follows. Moral Hazard in a Team of Agents

Moral hazard in the team should be especially salient whenever the compensation for the team members is dependent on team, rather than individual, production levels. With a team piece-rate, each individual will be experiencing the full cost of his own efforts, but will experience only l/N of the benefits—what Prendergast (1999: 39) calls “the 1/N problem.” In medical partnerships (Gaynor and Pauly 1990) and law partnerships (Leibowitz and Tollison 1980), cost containment and productivity figures decrease as larger proportions of revenue are shared among partners. However, the empirical evidence suggests that when teams are compensated as teams, social processes can be developed to address the moral hazard problem. This does not mean the moral hazard problem disappears. Rather, group-based compensation makes moral hazard such a salient issue for the group that solving the problem becomes an essential first order of business (Miller 1992: 188–195). Firms delegate the problem of monitoring effort within the group to the group itself, and implicitly at least encourage the group social norms that include mutual assistance and sanctions for free-riding. This strategy for addressing moral hazard has been called “high commitment” or “high involvement” management (Lawler 1986). Explicit tests of team-based incentives are relatively few, but a recent careful examination shows striking results. Hamilton et al. (2003) analyze data from a garment manufacturing facility that undertook a well-documented transition from individual piece-rates to team-based compensation as the demand for just-in-time put a premium on more flexibility in production methods. Team-based production increased productivity by about 21 percent (Hamilton et al. 2003: 481). More heterogeneous teams, in terms of individual ability, were more productive than teams of uniform quality. Most workers received more hourly compensation after moving to teams, with the exception of high ability workers who actually received an average 8 percent reduction in wage; they were nevertheless no more likely to leave than low quality workers. The authors note that this result suggest that “nonpecuniary

364

Gary J. Miller

benefits of team membership, such as more control over the work environment and less repetition, appear to be important for many workers.” (2000, 487). The Principal’s Moral Hazard

Holmstrom (1982), together with related studies (Eswaran and Kotwal 1984), is able to show that moral hazard can be eliminated among team members by the appropriate incentives as long as budget-balancing is sacrificed. That is, residual profits must be allocated to a budget-breaker who is removed from the team production process. However, then moral hazard is necessarily relocated in the owner of the budget residual. This poses its own principal-agent problems for firms. Holmstrom shows that group forcing contracts eliminate moral hazard within the team, while necessarily violating budget-balancing. The group forcingcontract rewards everyone in the team if the efficient outcome is observed— without trying to monitor any individual’s effort. No team member is paid if the efficient outcome is not observed. Among the team, this creates a Nash equilibrium at the optimal outcome, because any one shirker would deprive herself (and everyone else) of the payment. The budget-breaker absorbs the surplus revenues after the efficient outcome is observed and individual team members are paid. Holmstrom points out that a standard solution in firms to this particular principal-agent problem, the separation of ownership and control, is a common feature of firms. The shareholders of the modern firm do not, and should not, provide any contribution to the production process. Their role in the firm should be limited simply to absorbing the surpluses and deficits generated by the productive members of the firm. As Eswaran and Kotwal (1984) demonstrate, the budget-breaking principal is endowed with perverse incentives; moral hazard has been removed from the team but deposited with her. Consider the joint-forcing contract. The productive team members, together with the residual owner, constitute a budget-balanced whole that must include at least one actor with moral hazard. But by construction, moral hazard has been eliminated among the productive team members. Therefore, the residual owner must be endowed with moral hazard. As shown by Eswaran and Kotwal, the residual owner would be better off by bribing one team member to shirk, thereby relieving her of any contractual obligation to pay any of the team members. This thought experiment demonstrates that the ownership of the residual does not in general guarantee incentives that are aligned with efficiency. Because of this, the principal must not only be passive, she must be credibly constrained from altering the incentive schemes, because the incentive systems that maximize her residuals are inconsistent with the incentive schemes that encourage Pareto optimal levels of team effort (Falaschetti and Miller 2001). Consequently, one fundamental aspect of any solution to principal-agent problems is a delicate constitutional balancing act, in which the principal is

Solutions to Principal-Agent Problems in Firms

365

credibly constrained from acting on her moral hazard. Without this constraint, an efficient incentive scheme could not be effective in inducing the efficient Nash equilibrium, because the team members would expect that the principal would act on her moral hazard and ultimately deprive them of their compensation. As an example, Prendergast (1999: 29) notes that principal moral hazard is especially threatening with subjective performance evaluations; “even though an agent exerts effort and performs well, the supervisor may claim otherwise” because the bonus comes from the pocket of the residual claimant. As evidence of this, she notes the tendency of film companies to manipulate book profits in order to keep down contractual obligations to film stars and others. James Garner sued Universal Studies when his promised 37.5 percent of net profits proved to be worthless in the wake of zero reported profits. Kahn and Huberman (1988) point out that firm-specific human capital provides another setting for moral hazard by the residual claimant. A firm may have reason to offer a bonus to an employee for investing in firm-specific human capital—then wish to renege after the employee has made the investment. One solution that Kahn and Huberman point to is a kind of credible commitment device—an enforceable “up or out” contract that would require the firm to either give the bonus or fire the employee—and the latter would be prohibitively costly to the firm. The up-or-out contracts at law partnerships, for example, are thus understandable as a way of constraining moral hazard by firms and thereby inducing the immense commitment in human capital that is required by its employees. Another potential object of moral hazard is the deferred compensation plans studied by Lazear (1981). In many firms and industries, employees are often paid less than their marginal products early in their careers, and then paid more than their marginal product later in their careers. One interpretation is that employees are in effect bonding themselves as diligent workers: they accept the low wage early for a long enough time that their true effort can be discerned and rewarded by the firm. This pattern also has the benefit of rewarding investments in firmspecific human capital. As a reward scheme, it therefore has efficiency benefits with regard to both adverse selection and moral hazard by agents. However, it is vulnerable to moral hazard by owners. Owners have every reason to renege on the higher compensation schemes the employees come to expect late in their careers. Shleifer and Summers (1988) argue that these deferred compensation schemes offer a temptation to owners: they can let employees work at low wages early in their careers, invest in firm-specific capital, and then either fire them or refuse to pay the career-ending bonuses. The downsizing accompanying a takeover has the same profit-maximizing effect. A downsizing takeover therefore represents a one-time appropriation of employees’ investments—which have the effect of destroying future credible commitment. Later employees and managers are unlikely to negotiated deferred compensation plans, and employees are less likely to make firm-specific human capital investments.

366

Gary J. Miller

To the extent that managers earn only a tiny fraction of increased profits, their incentive to renege on such long-term arrangements is much less than that of shareholders. The credible constraint of budget-breakers often relies, then, on effective delegation from shareholders to managers. Falaschetti (2002) finds evidence of the use of “golden parachutes” as a mechanism for insulating managers, and therefore credibly constraining shareholders. He argues that if golden parachutes were simply a manifestation of managerial shirking, then they would be more in evidence in those firms in with lower concentration of ownership. However, he finds that golden parachutes are in fact more common in those firms with high concentration, where the owner’s moral hazard is more likely to be acted on. Thus, in his perspective, diffusion of ownership and golden parachutes are alternative mechanisms of credible constraint of owners’ moral hazard; both mechanisms allow managers to make contractual commitments in such a way that employees, suppliers, and other stakeholders need not worry about shareholder reneging. Credible Commitment and Group Incentives

Credible commitment may also explain some of the variability in the effectiveness of group incentive plans. In a large firm, group incentives by themselves do little to encourage effort by any individual team member; the reason is that each person has such a tiny impact on the probability or size of the bonus that it motivates little extra effort. This is similar to the argument that there is little to no reason for any citizen in large democracy to go to the trouble of voting, no matter how much the citizen may care about the outcome. However, just as has been argued in the literature on voting, one obstacle to free-riding is the mutual monitoring and sanctioning within groups which support voting as a group norm. Similarly, Knez and Simester (2001) analyze “mutual monitoring” among employees of Continental Airline. In 1995, Continental had had the worst record among the top ten airlines in on-time arrival, baggage handling, and customer complaints, and was in danger of bankruptcy (2001: 747). Management created a group incentive scheme paying a monthly bonus, contingent on firm-wide ontime performance. In the succeeding months, Continental moved to the top half of on-time performance, and made sizable profits for the first time in recent years (2001:748). Knez and Simester were able to compare Continental’s performance at airports where Continental employees were subject to the bonus plan with airports where services had been outsourced to firms without such a bonus; their statistical analysis indicated that the outsourced airports did not experience the same boost in performance. Managers and employees both felt that the incentive system had been effective in changing group norms. They reported that employee work groups began initiating their own performance reviews when flights were delayed; they increasingly chastised team members who seemed to be shirking, even calling colleagues from break rooms; gate employees would enter aircraft to identify the source of delays. (2001:767)

Solutions to Principal-Agent Problems in Firms

367

If Continental’s incentives were in fact instrumental, then the case stands in contrast with other firms where such bonus plans were ineffectual and abandoned. One such was at Dupont fibers division, where group work norms seemed to be little affected by the group bonus plan (Hays 1988). One difference seems to be the credibility of the firm’s commitment to the plan (Miller 1992). At Dupont, employees openly suspected management of reneging. Firm managers could avoid paying the bonus by manipulating tax write-offs or other accounting procedures. Said one union representative: “There are so many loopholes for management. . . . How do we know if we’ve reached our goal?” (Hays 1988) On the other hand, the more successful Continental program seemed to have satisfied its employees that the firm could not renege on the bonus plan. They did this by making the bonuses contingent, not on the firm’s own accounting figures, but publicly announced Department of Transportation on-time arrival figures (2001:747). The transparency of the plan made the effort of changing group norms a more rational investment. It is worth emphasizing that the problem of the principal’s moral hazard, as revealed in the analysis of Holmstrom (1982) and in Eswaran and Kotwal (1984), drastically redefines the nature of the principal-agent problem. Rather than simply being an incentive design problem in which the principal constrains the agent, what emerges is a constitutional problem in which the principal must first be constrained not to act in her own self-interest. This credible commitment allows for the negotiation of contracts such as deferred compensation schemes that provide incentives for efficient long-term investment in human capital and cooperative work group norms. Piece-rates are notorious for failing to have the desired productivityenhancing benefits when employers have the power to reduce piece-rates when productivity goes up or to fire employees when inventories grow (Miller 1992). Black and Lynch (2001) note that unionized firms that allow for employee participation in decision-making generate more efficiency benefits from incentivebased compensation plans than nonunionized plants; presumably this is because unionization is one means by which management can credibly commit to the incentive scheme. When management is credibly constrained, combined with the emphasis on autonomy and production flexibility within the work teams, this creates a much less hierarchical and bureaucratic style than the monitoring-intensive organizations discussed in Section 2. Furthermore, the emphasis on cooperation and socialization within work teams results in a culture that is less competitive than the entrepreneurial teams discussed in Section 1. The result is sometimes called a “clan” (Ouchi 1981) or “community” style of culture.

5. CONCLUSION Coase (1937: 388) observed, “The main reason why it is profitable to establish a firm would seem to be that there is a cost of using the price mechanism.”

368

Gary J. Miller

However, the firms that are established for this common reason are not themselves uniform. While the price mechanism enforces certain basic similarities among markets for a variety of different goods and services (the wheat market looks a lot like the bond market), the absence of the price mechanism allows for a great diversity of transactional forms (a wheat farm is fundamentally different from bond-trading firm). Ouchi noted as long ago as 1980, some firms don’t look so very different from markets, while others resemble hierarchical bureaucracies or cooperative clans. The theme of this essay has been that the development of principal-agency theory since 1980 can aid our understanding of this diversity of form and behavior across firms. In some cases, the price mechanism must be modified very little in order to serve the firm’s purposes: the individual agent’s output is easily verified; the agent’s impact on that output is neither too low nor too high (Shavell 1979); the agent’s degree of risk aversion does not require an exorbitant risk premium. In these cases, the firm may use commissions or piece-rates—which create marketlike incentives for the agent. The result is what organizational behavioralists observe as a “market culture” within the firm (Kerr and Slocum 1987), with little socialization needed among employees, limited supervision from relatively distant supervisors, an emphasis on individual initiative, and a sense of autonomy and ownership by the agent in the work setting. The firm’s managers will have more authority than that of a client over a contractor, but only marginally so. In other settings, the price mechanism must be wholly replaced by what Coase calls “direction” (1937: 4) or authority. If monitoring agent activities is both informative and sufficiently inexpensive, contracting on that information makes both risk-averse agent and owner better off. The appropriate form of the contract is a flat wage contingent on satisfactory response to direction, as reported by the monitor (Harris and Raviv 1979). This employment relationship is characterized, in general, by strong hierarchical authority and programmed behavior by agents. The authority of the monitor is supplemented by bureaucratic rules that limit the incentive for the agent and the monitor to collude. The result is a firm culture that is more bureaucratic than market-like. If the production process is sufficiently inter-dependent, then it may be impossible either to contract on the individual agent’s output, or to monitor cheaply the actions of individual agents. As Holmstrom suggested, the best-possible response may be team-based incentives. The cost of implementing team-based incentive systems may nevertheless be significant: group socialization and training sessions in order to coordinate expectations on a high performance equilibrium; and cross-training in technical skills (Lawler 1986). A more subtle requirement is a credibly constrained budget-breaker (Eswaran and Kotwal 1984). The result, when the requirements are met, is a “clan” culture as seen at Hewlett-Packard or Southwest Airlines (Miller 1992). Coase is correct that “the distinguishing mark of the firm is the supersession of the price mechanism” (1937:389); however, the price mechanism may be replaced in a variety of ways. The development of a rigorous theory of principalagent contracts in recent decades has allowed us to be much more precise about

Solutions to Principal-Agent Problems in Firms

369

the essential reasons for the myriad variations in the organization we call the “firm”.

REFERENCES Akerlof, George A. 1982. “Labor Contracts as a Partial Gift Exchange”. Quarterly Journal of Economics 96(4): 543–569. Akerlof, George A. and Janet L. Yellen. 1986. Efficiency Wage Models of the Labor Market. Cambridge: Cambridge University Press. Basu, Amiya, Rajiv Lal, V. Srinivasan, and Richard Staelin. 1985. “Salesforce Compensation Plans: An Agency Theoretic Perspective”. Marketing Science 4: 267–291. Black, Sandra E. and Lisa M. Lynch. “How to Compete: The Impact of Workplace Practices and Information Technology on Productivity”. Review in Economics and Statistics 83: 434–445. Coase, Ronald. 1937. “The Nature of the Firm”. Economica 4: 386–405. Conlon, Edward J. and Judi McLean Parks. 1990. “Effects of Monitoring and Tradition on Compensation Arrangements: An Experiment with Principal-Agent Dyads”. The Acadamy of Management Journal 33: 603–622. Eisenhart, Kathleen. 1988. “Agency and Institutional-Theory Explanations: The Case of Retail Sales Compensation”. Academy of Management Journal 31: 488–511. Eriksson, Tor. 1999. “Executive Compensation and Tournament Theory: Empirical Tests on Danish Data”. Journal of Labor Economics 17: 262–280. Eswaran, Mukesh and Ashok Kotwal. 1984. “The Moral Hazard of Budget Breaking”. Rand Journal of Economics 15(4): 578–581. Falaschetti, Dino. 2002. “Golden Parachutes, Credible Commitments or Evidence of Shirking?” Journal of Corporate Finance 8: 161–180. Falaschetti, Dino and Gary Miller. 2001. “Constraining Leviathan: Moral Hazard and Credible Commitment in Constitutional Design”. Journal of Theoretical Politics 13(4): 389–411. Garen, John. 1994. “Executive Compensation and Principal-Agency Theory”. Journal of Political Economy 102: 1175–1199. Gaynor, Martin and M. Pauly. 1990. “Compensation and Productive Efficiency in Partnerhsips: Evidence from Medical Group Practice”. Journal of Political Economy 98: 544–574. Harris, Milton and Artur Raviv. 1979. “Optimal Incentive Contracts with Imperfect Information”. Journal of Economic Theory 20: 231–259. Hamilton, Barton H., Jack Nickerson, and Hideo Owan. 2003. “Team Incentives and Worker Heterogeneity”. Journal of Political Economy 111: 465–497. Hays, L. 1988. “All Eyes on Du Pont’s Incentive-Pay Plan”. Wall Street Journal 1: p. A1. Holmstrom, Bengt. 1979. “Moral Hazard and Observability”. Bell Journal of Economics 10: 74–91. Holmstrom, Bengt. 1982. “Moral Hazard in Teams”. Bell Journal of Economics 13: 324– 340. Holmstrom, Bengt and Paul Milgrom. 1994. “The Firm as an Incentive System”. American Economic Review 84: 972–991. Homans, G.C. 1954. “The Cash Posters”. American Sociological Review 19: 724–33. Jensen, Michael and Kevin Murphy. 1990. “Performance Pay and Top Management Incentives”. Journal of Political Economy 98(2): 225–264. Joseph, Kissan and Alex Thevaranjan. 1998. “Monitoring and Incentives in Sales Organizations: An Agency-Theoretic Perspective”. Marketing Science 7: 107–123. Kahn, Charles and Gur Huberman. 1988. “Two-Sided Uncertainty and Up-or-Out Contracts”. Journal of Labor Economics 6: 423–444. Kerr, Jeffrey and John W. Slocum. 1987. “Managing Corporate Culture Through Reward Systems”. Academy of Management Executive 1: 99–108.

370

Gary J. Miller

Knez, Marc and Duncan Simester. 2001. “Firm Wide Incentives and Mutual Monitoring at Continental Airlines”. Journal of Labor Economics 19: 743–772. Lawler, Edward E. 1986. High-Involvement Management. New York: McGraw-Hill. Lazear, Edward. 1981. “Agency, Earnings Profiles, Productivity, and Hours Restrictions”. American Economic Review 71(4): 606–620. Lazear, Edward. 1989. “Pay Equality and Industrial Politics”. Journal of Political Economy 97: 561–580. Lazear, Edward. 1996. “Performance Pay and Productivity”. NBER Working Paper 5672. Lazear, Edward and Shewin Rosen. 1981. “Rank-Order Tournaments as Optimal Labor Contracts”. Journal of Political Economy 89: 841–864. Leibowitz, Aleen and Robert Tollison. 1980. “Free Riding, Shirking and Team Production in Legal Partnerships”. Economic Inquiry 18: 380–394. Milgrom, Paul. 1988. “Employment Contracts, Influence Activities, and Efficient Organizational Design”. Journal of Political Economy 96: 42–60. Miller, Gary J. 1992. Managerial Dilemmas: The Political Economy of Hierarchy. Cambridge: Cambridge University Press. Miller, Gary J. and Andrew B. Whitford. 2002. “Trust and Incentives in Principal-Agent Negotiations”. Journal of Theoretical Politics 14: 231–267. Nalebuff, Barry and Joseph Stiglitz. 1983. “Prizes and Incentives: Towards a General Theory of Compensation and Contracts”. Bell Journal of Economics 14: 21–43. Ouchi, W.G. 1980. “Markets, Bureaucracies, and Clans”. Administrative Science Quarterly 25: 124–141. Prendergast. Canice. 1999. “The Provision of Incentives in Firms”. Journal of Economic Literature 37: 7–63. Radner, Roy. 1985. “Repeated Principal-Agent Games with Discounting”. Econometrica 53: 1173–1198. Raff, Daniel. 1988. “Wage Determination and the Five-Dollar Day at Ford”. Journal of Economic History 48: 389–399. Sappington, David E.M. 1991. “Incentives in Principal-Agent Relationships”. Journal of Economic Perspectives 5: 45–66. Shavell, Steven. 1979. “Risk Sharing and Incentives in the Principal and Agent Relationship”. Bell Journal of Economics 10: 55–73. Shleifer, Andre and Lawrence Summers. 1988. “Hostile Takeovers as Breaches of Trust” in Alan Auerbach (ed.). Corporate Takeovers: Causes and Consequences. Chicago, IL: University of Chicago Press. Spence, Michael and Richard Zeckhauser. 1971. “Insurance, Information, and Individual Action”. American Economic Review (Papers and Proceedings) 61: 380–387. Stiglitz, Joseph. 1974. “Risk Sharing and Incentives in Sharecropping”. Review of Economic Studies 61: 219–256. Strausz, Roland. 1997. “Delegation of Monitoring in a Principal-Agent Relationship”. Review of Economic Studies 64: 337–357. Tirole, Jean. 1986. “Hierarchies and Bureaucracies: On the Role of Collusion in Organizations”. Journal of Law, Economics, and Organisation 2: 181–214.