Contracting theory and accounting☆
Introduction
This paper reviews agency theory and its applications to accounting.1 Agency theory has been one of the most important theoretical paradigms in accounting during the last 20 years. The primary feature of agency theory that has made it attractive to accounting researchers is that it allows us to explicitly incorporate conflicts of interest, incentive problems, and mechanisms for controlling incentive problems into our models. This is important because much of the motivation for accounting and auditing has to do with the control of incentive problems. For example, the reason we insist on having an “independent” auditor is that we do not believe we can trust managers to issue truthful reports on their own. Similarly, much of the motivation for focusing on objective and verifiable information and for conservatism in financial reporting lies with incentive problems. At the most fundamental level, agency theory is used in accounting research to address two questions: (i) how do features of information, accounting, and compensation systems affect (reduce or make worse) incentive problems and (ii) how does the existence of incentive problems affect the design and structure of information, accounting, and compensation systems?
While agency theory has generated insights into financial accounting and auditing issues, by far its largest contributions have been to managerial accounting. Accounting systems produce numerous measures of financial performance, including costs, revenues, and profits. Each of these financial measures of performance can be calculated at the “local” level or at higher levels, including the firm-wide level. The question of how to best measure performance is an important one because accounting and budgeting systems, performance measurement systems, transfer pricing systems, and decision support systems affect how people and organizations interact. Criticism continues to grow that traditional performance measures motivate dysfunctional behavior by causing managers to pay attention to the “wrong” things.
For example, many firms are beginning to place greater emphasis on nonfinancial measures such as quality, customer satisfaction, on time delivery, innovation measures, and on the attainment of strategic objectives.2 Kaplan and Norton, 1992, Kaplan and Norton, 1993 have developed the notion of a “balanced scorecard” to attempt to reflect the multi-dimensional nature of managerial performance and to capture value drivers in a more timely fashion than conventional accounting numbers. Consulting firms are developing and marketing alternative financial measures of performance such as economic value added, cash flow return on investment, shareholder value, etc. and claiming they provide “superior” measures of performance and better incentives in motivating managers to take the right actions. At the corporate level, the relative merits of stock price versus accounting numbers as measures of performance continue to be debated, and we have witnessed a tremendous increase in the use of stock-based compensation during the 1990s. Agency theory provides a framework for addressing these issues and rigorously examining the link between information systems, incentives, and behavior.
Agency theory has its roots in the information economics literature. As such, accounting and other information is placed into an explicit decision-making setting. The value of information is derived from the better decisions (and higher profits) that result from its use. Another important carryover from information economics is the idea that the most meaningful way to compare accounting/performance measurement systems is by comparing each system when it is used optimally. For example, in order for there to be a role for additional accounting information, it must be the case that the incentive problems being studied cannot be completely resolved via other means. This typically places restrictions on the type of “other” information that is assumed to be available in the model. It also forces the researcher to explicitly build uncertainty and measurement problems into the model.
The primary way agency theory distinguishes itself from “traditional” information economics is its belief that multi-person, incentive, asymmetric information, and/or coordination issues are important in understanding how organizations operate. To have an interesting multi-person model, agency researchers are careful to ensure that conflicts of interests are explicitly built into the analysis. That is, agency theory models are constructed based on the philosophy that it is important to examine incentive problems and their “resolution” in an economic setting in which the potential incentive problem actually exists.3 Typical reasons for conflicts of interest include (i) effort aversion by the agent, (ii) the agent can divert resources for his private consumption or use, (iii) differential time horizons, e.g., the agent is less concerned about the future period effects of his current period actions because he does not expect to be with the firm or the agent is concerned about how his actions will affect others’ assessments of his skill, which will affect compensation in the future, or (iv) differential risk aversion on the part of the agent.
In the simplest agency models, the organization is reduced to two people: the principal and the agent. The principal's roles are to supply capital, to bear risk, and to construct incentives, while the roles of the agent are to make decisions on the principal's behalf and to also bear risk (this is frequently of secondary concern). The principal can be thought of as a “representative shareholder” or the board of directors.4 In more complicated agency models, there can be multiple principals and/or multiple agents. Some agents can even be both a principal and an agent, e.g., in a hierarchical firm a middle level manager might be the agent of managers above him and the principal to employees below him.
In order to more easily keep track of who knows what and when, it is often useful to construct a time line outlining the sequence of events in the model. In the “plain vanilla” principal–agent model, the sequence of events is as follows:
The principal selects a performance evaluation system which specifies the performance measures (or information signals) upon which the agent's compensation will be based and the form of the function that links the performance measures to the agent's compensation. Let s denote the compensation function, and the vector of performance measures to be used in the contract. Based on this contract, the agent selects a vector of actions, , which could include operating decisions, financing decisions, or investment decisions. These decisions, along with other exogenous factors (generally modeled as random variables) influence the realizations of the performance measures, as well as the “outcome” of the firm, which we denote as x.
We will assume the outcome is measured in monetary terms, although in some contexts such as health care choices or government policy choices, the outcome might be better thought of as being nonmonetary. In a single-period model, the monetary outcome is well defined; it represents the end-of-period cash flow or the liquidating dividend of the firm gross of the compensation paid to the agent. For now, we will assume that the outcome x is observable and can be contracted on. This assumption will be relaxed later. After the performance measures are jointly observed, the agent is paid according to the terms of the contract. Note that this formulation implicitly assumes that the property rights to the outcome belong with the principal. A few papers consider the opposite situation in which the agent has the property rights to the outcome by allowing him to keep any “unreported income”.
The “plain vanilla” version of the agency model has been extended in a number of ways. For example, as mentioned above, the outcome might not be observable. In this case there is potentially a role for information that helps estimate the outcome. Considerable effort in accounting research has also been directed at modeling different mechanisms by which the information signals, , are produced. The simplest case is that they are simply “generated” by the actions and “automatically” observed by the parties. Other papers have modeled the situation where the principal observes some information at the end of the period and then decides whether to conduct an investigation to obtain more information (e.g., a variance investigation).5 Another possibility is that the information is generated by a report made by the agent. In this case, there may be moral hazard problems on the agent reporting truthfully. The information signal might also be generated via a third party such as an auditor. In this case, incentives problems with the auditor (e.g., independence or how intensively does he audit and does he report his findings honestly) can be modeled and analyzed.6 Finally, the performance measure may come from the security market's process of aggregating information into stock prices. Again, issues regarding what information is available to investors, and how this information is affected by operating and reporting decisions by the agent can be modeled and analyzed.
Agency papers have also extended the basic model by allowing the agent and/or the principal to obtain information prior to the agent selecting his action. This information could relate to the productivity of different operating actions, the general “favorableness” of the environment, or information about the employee's type (e.g., his skill or his risk aversion). The pre-decision information could be received before the contract is signed or between the time the contract is signed and the time the agent selects his actions. In these papers, communication of the agent's information via participative budgeting can be studied. Agency papers have also extended the basic model to include multi-periods (either where a single-period model is repeated over time or where there are explicit interdependencies between the periods).
Finally, papers have modeled issues that arise when there are multiple agents in the firm. This enables us to examine the role of encouraging/discouraging competition among agents, and the use of relative performance to compare the performance of agents. With multi-agent models we can also study the interaction between management accounting and organizational structure, including hierarchies, job design, and task allocation. Multi-agent models are also necessary to studying the role of incentive problems in allocating resources (and costs) among agents, and analyzing transfer pricing between subunits.
In the next section, I discuss single-period, single-action agency models in which the incentive problem arises because the agent's actions are unobservable to the principal. These types of incentive problems are referred to as moral hazard or hidden action problems. I describe the features the models must possess in order for a genuine incentive problem to exist that cannot be costlessly resolved. I then discuss the role of performance measures in reducing the magnitude of the agency problem. The key characteristic here is the informativeness of the performance measure about the agent's action. The informativeness of a performance measure is a function of its sensitivity to the agent's actions and its noisiness. I discuss the implications of these models for the shape of the optimal contract, the conditions where performance measures are combined in a linear fashion (which is how accounting systems aggregate line items), and the ideas of responsibility accounting and the controllability principle commonly discussed in managerial accounting textbooks.
In Section 3, I continue to analyze hidden action models, but in models where the agent is responsible for multiple actions. In this section, I discuss the Linear contracting, Exponential utility function, Normal distribution (LEN) framework for formulating agency problems. In a multi-action model, the emphasis shifts from that of motivating the intensity of the agent's effort to the allocation of his effort. Accordingly, the congruity of a performance measure (or how it contributes to constructing an overall performance measure that is congruent) becomes important. I discuss the application of the results to accounting “window dressing” and earnings management, to incomplete or myopic measures of performance, to the role of nonfinancial measures of performance, divisional versus firm-wide performance, the valuation versus stewardship uses of information, and stock price versus accounting numbers in compensation contracts.
In Section 4, I focus on agency problems caused by the agent possessing superior information about a parameter that affects the outcome-generating process or perhaps about the outcome itself. In these models, accounting systems are used to communicate information within the organization, to coordinate actions across parties, as well as to evaluate the actions that have been taken and the outcome that has occurred. A new role for accounting systems is to reduce the “information rent” that the agent is able to extract based on his information advantage. I discuss the application of these results to issues of participative budgeting and target setting (including the creation of organization slack), to the “confirmatory” role of accounting numbers, hurdle rates for allocating capital, transfer pricing, and cost allocation.
Section 5 discusses communication, earnings management, and the revelation principle. In particular, I describe the qualities a model must possess to circumvent the revelation principle so that earnings management issues can be addressed. In Section 6, I briefly discuss multiple period agency models. Multi-period models are essential for earnings and cash flows to be different and for accruals to have a substantive role. I discuss multi-periods models in motivating long-term investment decisions, the use of cash flow versus accrual accounting versus residual income measures of performance, and the role of depreciation policies. The final section outlines some suggestions for future research.
Section snippets
Single-action agency models
In words, we express the principal's problem as a constrained maximization problem in which he chooses the compensation function (its form and the variables it is based on) to7
Multi-action models
While single-action agency models have been useful in generating many insights, they are too simple to allow us to address some important features of performance measures. In particular, in single-action models the sensitivity of a signal is an important feature, but the single-action framework precludes us from asking whether the measure is sensitive to the “right things”. In reality, we know that agents are generally responsible for a rich set of actions. They can vary how much attention they
Private information and communication
In this section, I discuss research that extends the basic agency model by allowing one party to have private information. For the most part, researchers have assumed that the agent is the party who obtains private information.53
Earnings management versus the revelation principle
Earnings management is viewed as an activity that is widely practiced by managers. Even though the agency framework seems to be a natural one to use to study earnings management, the agency literature to date has not made much progress in helping us understand how, why, and when earnings management takes place. The primary obstacle has been the revelation principle. As discussed in the previous section, when the revelation principle applies, any equilibrium that involves nontruthful reporting
Multi-period models and investment problems
While there are a number of interesting issues that arise in multi-periods agency models, the one I believe is of greatest interest to accounting relates to the role of lead–lag effects in performance measures. For example, we need a multi-period model to be able to talk about accrual accounting problems because in single-period models, cash flow and accrual accounting numbers are identical. Despite the obvious importance, not much work has been done on multi-period models in the agency
Directions for future research
I have attempted to indicate unresolved or unexplored issues at various places in the paper; however, in this section I will outline what I feel are the most important areas and issues for future research to explore. The first area relates to the aggregation of performance measures. A fundamental property of accounting systems is that they aggregate “basic” signals. Moreover, the aggregation is done in specific ways; in most cases, the aggregation is linear and all dollar amounts are weighted
References (122)
- et al.
Contracts without memory in multiperiod agency models
Journal of Economic Theory
(1985) - et al.
Introducing convexity into optimal compensation contracts
Journal of Accounting and Economics
(1999) - et al.
Dynamic incentives and responsibility accounting
Journal of Accounting and Economics
(1999) The use of accounting and security price measures of performance in managerial compensation contractsa discussion
Journal of Accounting and Economics
(1993)- et al.
Using delegation and control systems to mitigate the trade-off between the performance-evaluation and belief-revision uses of accounting signals
Journal of Accounting, Economics
(1998) - et al.
Strategic transfer pricing
Management Science
(1998) - Antle, R., 1982. The auditor as an economic agent. Journal of Accounting Research...
- Antle, R., Demski, J., 1988. The controllability principle in responsibility accounting. The Accounting Review...
- Antle, R., Eppen, G., 1985.Capital rationing and organizational slack in capital budgeting. Management Science...
- Antle, R., Fellingham, J., 1995. Information rents and preferences among information systems in a model of resource...
An empirical investigation of the relative performance evaluation of corporate executives
Journal of Accounting Research
Earnings management and the revelation principle
Review of Accounting Studies
Contracts and performance measurement
Journal of Political Economy
Sensitivity, precision, and linear aggregation of signals for performance evaluation
Journal of Accounting Research
Aggregate performance measures in business unit manager compensationThe role of intrafirm interdependencies
Journal of Accounting Research
Cost allocation games
Cited by (573)
Revising the theory behind corporate governance and management control: A reflection and roadmap
2024, Management Accounting ResearchEnvironmental claims and executive compensation plans: Is there a link? An empirical investigation of Italian listed companies
2023, Journal of Cleaner ProductionThe effect of peer-to-peer recognition systems on helping behavior: The influence of rewards and group affiliation
2023, Accounting, Organizations and SocietyDEVELOPMENT OF FINANCIAL RISK HEDGING STRATEGIES
2024, Financial and Credit Activity: Problems of Theory and Practice
- ☆
I would like to thank Stan Baiman, Ronald Dye, Robert Magee, Madhav Rajan, Robert Verrecchia, and Jerold Zimmerman for their useful comments.