Earnings Management and the Revelation Principle

When the Revelation Principle (RP) holds, managing earnings confers no advantage over revelation. We construct an explanation for earnings management that is based on limitations on owners’ ability to make commitments (a violation of the RP’s assumptions). Traditionally, earnings management is seen as sneaky managers pulling the wool over the eyes of gullible owners by manipulating accruals; our limited commitment story suggests that the owners, too, can benefit from earnings management. We categorize a variety of extant explanations of earnings management, along with our own, according to which of the assumptions of the RP each explanation violates. Plausibility of multiple simultaneous violations of the assumptions, and strategic use of various accounting and real instruments of earnings management, complicate the task of detecting such management in field data. When managers choose accounting accruals, neutral communication of the firm’s underlying economic reality to the readers of financial reports is not necessarily their only goal. This goal can become enmeshed with managers’ desire to use financial reports, especially earnings, opportunistically to serve their own personal ends. The existence of such mixed motives in managers has given rise to hypotheses about management (or manipulation) of earnings, theoretical analyses of the interaction between the two motives for such management, and an empirical literature that attempts to identify and document this phenomenon. The purpose of the paper is twofold. First, we suggest earnings management is more than just sneaky managers pulling the wool over the eyes of gullible owners. Manipulation can be in the best interests of owners.1 In particular, we study a setting in which the ability of owners to make binding commitments is constrained. Earnings management is useful because it reduces owner intervention. Although such management is not in the best interest of owners ex post (when the earnings report is submitted), it is in their best interest ex ante (when they are trying to induce the manager to join the firm and exert effort to benefit the firm). Economic explanations for earnings management require one or more of the assumptions of the Revelation Principle (RP) to be violated (Dye, 1988). The RP states that any equilibrium outcome of any mechanism, however complex, can be replicated by a truth-telling equilibrium outcome of a mechanism under which the agents are asked to report their private information to the principal (see, for example, Myerson (1979)). Hence, when the RP holds, the performance of any mechanism under which managers manipulate earnings can be replicated by a mechanism under which managers report earnings truthfully.2 As Dye writes, the RP is “a nemesis to the study of earnings management.” 8 ARYA, GLOVER AND SUNDER Nevertheless, the RP is indirectly useful in studying earnings management. We can look to violations of the RP’s assumptions to classify earnings management stories. The second contribution of our paper is to bring out the interrelationships among various earnings management stories. Since multiple simultaneous violations of the assumptions of the RP are plausible, any single explanation of earnings management (including our own) is unlikely to be the explanation of the phenomenon. Two of the better known forms of earnings management are “smoothing” and “big bath.” For example, in estimating their bad debt allowance, companies might be tempted to provide a generous allowance in good years and skimp in lean years in order to smooth the stream of reported earnings.3 In contrast, the big bath hypothesis suggests that managers undertake income decreasing discretionary accruals in lean years. Perhaps managers believe that one very poor performance report is not as harmful as several mediocre performance reports. It has been suggested that big baths often occur under the guise of restructuring charges (see, for example, Elliott and Shaw (1988)) and may coincide with top management transition. The past thirty years have seen an intensive effort to try to document the existence and nature of earnings management in field data and to build formal models in which management of earnings arises endogenously as a consequence of rational choice made by utility-maximizing economic agents. Data gathered from financial reports of corporations have been scrutinized for finger prints of opportunistic managerial manipulation with only mixed results (see, for example, Archibald (1967), Bartov (1993), Copeland (1968), Cushing (1969), DeAngelo (1986), DeAngelo, DeAngelo, and Skinner (1994), Dechow, Sloan, and Sweeney (1995), Gordan, Horwitz, and Meyers (1966), Healy (1985), Liberty and Zimmerman (1986), Lys and Sivaramakrishnan (1988), McNichols and Wilson (1988), Ronen and Sadan (1981)).4 Some of the reasons given for the weak and inconsistent empirical results are: (1) use of unreliable empirical surrogates for managed and unmanaged portions of earnings, (2) the focus of most empirical studies on one accounting instrument of earnings management at a time, (3) a narrow interpretation of earnings management, and (4) managers’ incentives to cover their tracks (Sunder, 1997, pp. 74–78). Our paper suggests two more: (5) owners may have incentives to make it easy for managers to hide information and (6) two or more independent conditions that induce earnings management may exist simultaneously, causing studies that focus on a single condition to yield noisy results. Our limited commitment explanation for earnings management is based on the idea of “at-will” employment contracts. Their implementation depends entirely on the willingness of the parties to continue to subject themselves to their terms. Each employment contract is between an owner and a manager. We assume the owner cannot commit to a policy regarding firing/retention decisions. Also, the manager cannot commit to staying with the firm. This is consistent with observed employment contracts which often specify broad terms and objectives, but rarely specify the exact circumstances under which the employee can quit or be dismissed (Milgrom and Roberts, 1992, p. 330; Sunder, 1997, p. 40). There are benefits and costs associated with dismissing a manager. The benefits to the owner are that (1) the threat of dismissal for bad outcomes provides the manager with incentives to reduce the probability of bad outcomes through his actions, and (2) dismissal allows the owner to replace a manager who is known to have performed poorly with another EARNINGS MANAGEMENT 9 from the pool of candidates whose expected productivity is greater. Because the owner cannot commit to the dismissal decision, she will fire the manager whenever it is in her own ex post best interest. From an ex ante perspective this can result in the manager being fired too often and the cost of the firing option reducing the welfare of the owner.5,6 We compare the owner’s payoff under a system of unmanaged earnings (the owner herself directly and costlessly observes earnings) to her payoff under a system of managed earnings (earnings reports provided by the manager that may or may not be truthful). A proposition establishes conditions under which the owner prefers managed earnings to unmanaged earnings. The manager manipulates earnings to retain his job for as long as possible, and the owner finds the coarsening and delay of information that occurs under earnings manipulation beneficial as a device that effectively commits her to making firing decisions that are better from an ex ante perspective.7 We also present an example in which managed earnings are compared to other benchmarks. The intent is to highlight the particular way in which earnings management coarsens and delays information. Earnings management leads to a time-additive aggregation of performance measures, which prevents the owner from learning the firm’s performance in the short run but allows her to detect persistently poor performance.8 Fudenberg and Tirole (1995) present a story closely related to ours. In their model (and ours) the manager manipulates earnings in order to avoid (delay) dismissal. However, in the Fudenberg and Tirole setting, the owner prefers unmanaged earnings (if available) to managed earnings. This is because, in their setting, the manager does not supply a productive input and the manager’s compensation is not modeled. The benefit to the owner of being able to fire the manager is the possibility of removing an unproductive manager; there is no disciplining benefit to firing and no cost to firing.9 Shades of our story can also be found in business press. First, managers sometimes find earnings management useful as a way of limiting owner intervention. For example, German executives have been described as viewing secret reserves as useful in keeping “gimlet-eyed shareholders [from] calling the shots” (The Wall Street Journal, 1998). The traditional (German) view has been that their reporting standards, which allow for such hidden reserves, encourage managers to focus on the long-run rather than the short-run performance of their companies. Second, placing some bounds on owner intervention is desirable for the company as a whole, and for the owners themselves. An article in Executive Excellence (1997) describes one of the board’s roles as that of providing autonomy: “Boards must give organization members the autonomy to do their jobs.” The presumption here is that, without a certain degree of autonomy, management is not able to perform at its best. Our story links these two views and adds a timing perspective. Earnings management is a substitute for the owners committing ex ante to resist any temptation they might have to intervene ex post. This ex ante commitment is useful in attracting and motivating managers. In our model, the owner intervenes through her decision to retain or replace the manager. Alternatively, the board may respond to the firm’s short-term poor performance by “back seat driving” with respect to decisions normally left to the CEO. By allowing for earnings management, the board gives the CEO more room to work things out. In Section 2.5 of the paper we present a numerical example that illustrates the role earnings management can have in preventing such owner intervention. In the example, owner intervention renders 10 ARYA, GLOVER AND SUNDER earnings less informative about the manager’s actions and, hence, makes it more difficult for the owner to motivate the manager. The remainder of the paper is organized as follows. Section 1 uses violations of the assumptions of the RP as an organizing principle to explore the relationships among the extant explanations, as well as our own at-will story. Section 2 presents our at-will contracts explanation for earnings management. Section 3 presents concluding remarks and some implications for analysis and interpretation of data. 1. Organizing the Earnings Management Stories Using RP Violations The accounting literature includes many formal and informal explanations for income management, each applicable to a limited range of circumstances. Given the variety of environments in which businesses operate, it is unlikely that any single explanation is adequate for all, even most, earnings management. Perhaps it is better to think of a portfolio of earnings management stories and to identify the criteria that determine which one or more of the members of the portfolio are applicable to individual circumstances. The principal-agent model framework used in this paper is a special case of a mechanism design problem. By a mechanism design problem, we mean a central planner (for example, a principal) constructs a message space (reporting system) and an outcome function (contract) for the purpose of implementing particular actions and/or resource allocations, where the prescribed actions and allocations can depend on the agents’ private information. As mentioned earlier, the RP states that any possible equilibrium outcome of any possible mechanism, however complex, can be replicated by a truth-telling equilibrium outcome of a mechanism under which the agents are asked to report their private information to the central planner. Although the RP is a useful benchmark in establishing the set of actions and allocations that can be implemented, it is difficult to imagine real-world settings (even marriage, let alone employment relationships) that satisfy all of the RP’s assumptions. Weakening each assumption of the RP is a convenient way of organizing a portfolio of earnings management stories because at least one of the assumptions must be violated for earnings management to occur. The RP’s assumptions are related to communication, the contract form, and commitment. The RP assumes: (1) communication is not blocked (it is costless to establish communication channels that allow the agents to report fully their private information),10 (2) the form of the contract is not restricted, and (3) the principal can commit to use the reports submitted by the agents in any prespecified manner.11 We present a summary of the existing stories for earnings management first in Table 1, and illustrate and compare them using a series of numerical examples. We found the examples helpful since variations of the existing stories are relatively easy to incorporate within the risk-neutral framework we use subsequently for our at-will contracts story. In the following examples, we assume that any overstatement or understatement of earnings in the first period is reversed in the second period. Earnings can be managed only intertemporally. EARNINGS MANAGEMENT 11 Table 1. Violations of assumptions of the RP and earnings management stories. Revelation Principle’s Assumptions Revelation Principle’s assumptions do not hold because of: hold Limited Communication Limited Contract Limited Commitment No earnings Earnings management is used Earnings management arises Using earnings management to management to convey information as a response to: conceal information enables: regarding: 1. manager’s expertise/ 1. bonus floors and ceilings 1. risky firms to smooth productive input supplied earnings to pool with safe 2. incomplete debt covenants firms and obtain better credit 2. the constraints placed on terms reporting by partial verifiability 2. firms to manage earnings downward to reduce demands 3. the permanence of earnings from employees, shareholders, tax authorities, and other regulators 3. manager to overstate earnings to benefit one generation of shareholders at the cost of another generation 1.1. Relaxing the Communication Assumption 1.1.1. Conveying Expertise Earnings management can be beneficial to owners because it enables the manager to communicate his acquired expertise. A smooth airline flight is not only comfortable but also reassuring to the passengers about the pilot’s expertise. The expertise explanation for earnings management is developed in Demski (1998). We present an example to highlight the idea, although it does not capture some of the important features of Demski’s model. Two risk-neutral parties, an owner and a manager, contract with each other over two periods. They commit not to fire and not to quit after the first period, respectively. During the first period the manager privately chooses one of two possible productive acts labeled aL (for low) and aH (for high). These acts have a disutility of 0 (for low) and 1 (for high) to the manager. In period t , t = 1, 2, the firm’s earnings, xt , can be either 0 or 200.12 If the manager chooses the low act, the firm earns 0 with probability 0.4 and 200 with probability 0.6 in each period. Under the high act these probabilities change to 0.3 and 0.7, respectively. Given the manager’s act, the distributions of earnings in the two periods are mutually independent; the realization of period-one earnings does not affect the distribution of period-two earnings. The probability distributions and payoffs are common knowledge. 12 ARYA, GLOVER AND SUNDER Regardless of his action, the manager privately observes period-one earnings at the end of that period. By choosing the high act, the manager acquires expertise to forecast perfectly period-two earnings at the end of period one. The low act does not furnish him with the foresight and he must await the end of period two to learn the earnings of that period.13 At the end of each period, the manager submits a report to the owner on that period’s earnings. We restrict communication by not permitting him to submit a forecast of period-two earnings at the end of period one. Suppose unmanaged earnings (xt ) are available for contracting. The owner’s objective is to maximize expected earnings less compensation subject to the constraints that the contract: (1) is individually rational (provides the manager with at least his reservation utility), (2) is incentive compatible (it is in the manager’s own interest to choose the act the owner intends), and (3) avoids bankruptcy (payments in each period are nonnegative).14 Assume the manager’s two-period reservation utility is 2. Denote by s(x1, x2) the compensation paid to the manager as a function of the firstand second-period earnings. The owner’s program under unmanaged earnings is as follows. (It can be verified that motivating aH is optimal.) Max s (.3)(.3)[0− s(0, 0)]+ (.3)(.7)[200− s(0, 200)]+ (.7)(.3)[200− s(200, 0)]+ (.7)(.7)[400− s(200, 200)] subject to: (.3)(.3)s(0, 0)+ (.3)(.7)s(0, 200)+ (.7)(.3)s(200, 0)+ (.7)(.7)s(200, 200) −1 ≥ 2 (.3)(.3)s(0, 0)+ (.3)(.7)s(0, 200)+ (.7)(.3)s(200, 0)+ (.7)(.7)s(200, 200) −1 ≥ (.4)(.4)s(0, 0)+ (.4)(.6)s(0, 200)+ (.6)(.4)s(200, 0)+ (.6)(.6)s(200, 200) −0 s(0, 0), s(0, 200), s(200, 0), s(200, 200) ≥ 0. An optimal solution to the above program is: s(200, 200) = 7.69 and s(·, ·) = 0 otherwise. Under unmanaged earnings the owner’s payoff is 276.23. The owner can do better with managed earnings—that is, allowing the manager the freedom to choose what to report at the end of period 1. An optimal contract is to pay the manager 3 if his firstand second-period earnings reports are equal and 0 otherwise. This contract motivates the manager to choose the high act. If the manager chooses the high act, he is able to smooth earnings (report 0.5[period-one earnings + period-two earnings] at the end of each period) and earn his reservation utility of 2. If the manager chooses the low act, he will not learn period-two earnings at the end of period one; the probability with which he will be able to produce identical firstand second-period earnings reports is only 0.6. His best guess (maximum likelihood estimate) of the second-period earnings level is 200, so he reports .5[period-one earnings +200] at the end of the first period. With probability 0.4 the second-period earnings report will be different than the period-one earnings report. The manager earns only 0.6(3) − 0 = 1.8 if he chooses the low act. Hence, he prefers the high act. Under managed earnings the owner’s payoff is 277, which is higher than the 276.23 she obtains under unmanaged earnings. EARNINGS MANAGEMENT 13 Both the level and the smoothing of earnings are informative about the manager’s action. While the level alone is not enough for the owner to obtain the first-best solution (the solution under the assumption that the owner observes the manager’s action), the smoothing of earnings is. If the manager’s ability to predict earnings were imperfect, the optimal contract would depend on both the level as well as the pattern of reported earnings (e.g., earnings smoothing). In this example earnings management is important, smoothing is not. Any number of earnings management conventions would effectively tell the owner if the manager is an expert (for example, the first-period report is 75 percent of the second-period report). However, smoothing has the advantage of being a simple and, therefore, an easy convention on which agents can coordinate. The RP does not apply in this example because communication is restricted. If the manager were allowed to submit a period-one earnings report and a forecast of period-two earnings at the end of period one and a period-two earnings report at the end of period two, a revelation mechanism could be used to identify the expert.15 1.1.2. Partial Verifiability Restricted communication is also a key part of the earnings management explanation presented in Evans and Sridhar (1996). They consider an internal control system that sometimes prevents the manager from misreporting the outcome. In the single-period version of Evans and Sridhar’s story, because high outcomes indicate the manager worked harder, reports of higher outcomes are associated with larger compensation. As a result, the manager overreports the outcome whenever the control system allows him to do so. The owner prefers to induce the manager to lie than to bear the cost of motivating truth-telling. However, it is even better for her to observe the actual outcome herself. That is, lying is tolerated but does not benefit the owner as much as an ability to observe the outcome. The impact of partial verifiability is also studied in Green and Laffont (1986) and Lipman and Seppi (1995). The following numerical example applies the basic idea to earnings management. Earnings, x , can be 0, 1, or 2, with equal chance. The manager privately observes x and reports x̂ . The contract between the owner and the manager specifies a dividend amount, d, which is contingent on the manager’s earnings report. The owner consumes the dividend and the manager consumes the remainder of earnings. The dividend is constrained to being less than or equal to actual earnings, d ≤ x . The extent of misreporting by the manager is constrained by a partial verification process. Suppose when earnings are 0 the manager can report only 0; when earnings are 1 the manager can report 0 or 1; and when earnings are 2 the manager can report 1 or 2. The owner can motivate the manager to report earnings truthfully, but only by paying a high price and setting her own dividends to zero irrespective of the report: d(x̂ = 0) = d(x̂ = 1) = d(x̂ = 2) = 0. The owner is better off letting (encouraging) the manager to misstate earnings. The owner optimally specifies d(x̂ = 0) = 0 and d(x̂ = 1) = d(x̂ = 2) = 2. This motivates the manager to report that earnings are 0 when they are 0 or 1 and to report earnings of 2 when they are 2. (Note that the dividend paid is always less than or equal to earnings.) 14 ARYA, GLOVER AND SUNDER If the manager could report the set of outcomes the control system will allow him to report instead of reporting a single outcome, there would be an optimal mechanism under which the manager truthfully reveals the set of possible reports. As in Evans and Sridhar (1996) the owner is even better off under unmanaged earnings. 1.1.3. Conveying the Permanence of Earnings Although conflicting interests (common to the preceding stories) are useful in understanding earnings management, a simple explanation can be given without appealing to such considerations. If managers are not otherwise able to communicate whether earnings changes are permanent or transitory in nature, earnings management can be a way of conveying this information (Fukui, 1996).16 When a manager believes the increase in earnings of a period to be transient, he hides some to create an earnings reserve. If a drop in earnings is judged to be transient, he reports more by drawing down the reserve. In contrast, when a change in earnings is believed to be permanent, the manager allows his report to reflect the change. Under the assumed restriction on communication, such a policy helps the shareholders arrive at a more accurate valuation of their shares. If we reintroduce the possibility of a divergence in incentives, managers can have shortrun incentives to pretend temporary earnings increases are permanent and/or permanent earnings declines are temporary. In some cases long-run considerations (for example, maintaining one’s reputation) dominate short-run considerations. In other cases short-run considerations win out—this is likely to be the case when the manager is near retirement or earnings are so poor that he is likely to be dismissed if the truth is revealed. 1.2. Relaxing the Contract Assumption 1.2.1. The Form of Bonus Schemes Another explanation for earnings management, due to Healy (1985), is based on the form of linkage between earnings and bonus compensation. Bonus schemes often specify lower and upper bounds on earnings; no bonus is paid if the lower bound is breached, and a fixed bonus is paid if the earnings exceed the upper bound. Between the lower and upper bounds bonuses increase with earnings. Opportunistic managers can increase the present value of their compensation by managing earnings down (up) when earnings fall outside (inside) the range defined by these bounds. However, there is at least anecdotal evidence that managers sometimes manage earnings down even when they are inside the bonus range. As an extension of Healy’s work one could try to explain this phenomenon. Suppose the share of earnings paid to the manager increases over his tenure. This would tend to provide a manager who is currently in the bonus range and thinks it is likely he will be within the bonus range in future periods with incentives to manage current earnings down to save up for future periods. On the other hand, a manager who thinks it is not likely he will be within the bonus range in future periods will manage current earnings up. In this argument the form of the compensation EARNINGS MANAGEMENT 15 contract is exogenous. One could take this a step further and derive conditions under which such bonus schemes arise endogenously (for example, because they induce the manager to reveal information about his assessment of the firm’s future prospects through earnings management). 1.2.2. The Form of Debt Covenants Debt covenants can be viewed as incomplete contracts in that they are not conditioned on all accounting methods a firm can choose. The standard story is that debt covenants motivate a firm to adopt income increasing accounting methods when a firm is in danger of violating its covenants (see, for example, Sweeney, 1994). For an analysis of income management (that can be interpreted as a method of avoiding default) in a dynamic agency setting, see Boylan and Villadsen (1997). A signaling story involving debt covenants is presented in Levine (1996). An incomplete debt contract—a single and fixed debt covenant—induces firms with favorable future prospects to use a conservative accounting method to account for stock-based compensation. By doing so, they can distinguish themselves from firms with unfavorable future prospects in the eyes of their creditors, and thus obtain better credit terms.17 If a menu of debt covenants could be offered, the choice of a tight debt covenant (instead of accounting method choice) could itself be used to separate firms. 1.3. Relaxing the Commitment Assumption 1.3.1. Improved Credit In discussing conservatism, Sanders, Hatfield, and Moore (1938, p. 16) argue that some procedures are “undertaken for the purpose of averaging profits over the years, so as to make a better showing in the lean years than the facts warrant. This, it is asserted, enhances the company’s credit and prestige.” A similar story is presented in Trueman and Titman (1988). A numerical example highlights the idea. There are two types of risk-neutral firms, safe and risky. Each type is equally likely. Firms have a life of three periods. A risky firm’s periodic earnings are 0 with probability 0.1 and 300 with probability 0.9. Earnings across periods are independently distributed. A safe firm’s periodic earnings are 150 with probability 1. In the first two periods the firm’s financing is provided by its owners. Period-one and period-two earnings (and paid-in capital) are distributed to the owners by the end of period two. At the beginning of the third period, the firm can contract with a risk-neutral lender to borrow 100 for the third period. The firm repays the principal plus interest to the lender at the end of period three. The payment to the bank at the end of the third period is bounded by the firm’s period-three earnings. Because the lender operates in a competitive market, it charges the firms an interest rate such that its expected return is equal to a market rate of return, rM , say 10 percent. 16 ARYA, GLOVER AND SUNDER At the end of each period the firm observes actual earnings while the bank observes only the firm’s reported earnings. There are infinitely many partially separating equilibria, all of which involve earnings management and are equivalent in terms of the payoffs to all parties. One such equilibrium is for a safe firm to report its earnings truthfully and for a risky firm to report 150 in the first period and x1 + x2 − 150 in the second period. A risky firm is able to mimic a safe firm if x1 = 0 and x2 = 300 or x1 = 300 and x2 = 0. For the other two possible earnings combinations, a risky firm is separated since it cannot report 150 in each period; these firms are charged an interest rate rR such that: .9(1+rR) = 1 + rM . Hence, rR = 22.22 percent. The lender cannot differentiate the remaining risky firms (18 percent of the risky firms which constitute 9 percent of the total population) from the safe firms (which constitute 50 percent of the total population) and charges a pooled interest rate, rP , such that: (.5/.59)(1 + rP) + (.09/.59)(.9)(1 + rP) = 1 + rM . Hence, rP = 11.71 percent. By smoothing earnings, a risky firm sometimes is able to obtain a lower interest rate (of 11.71 percent) than the rate (of 22.22 percent) it would be charged if the lender knew its type. Of course a safe firm is worse off since it is charged a higher interest rate (of 11.71 percent) than it would have been charged if the lender knew its type (of 10 percent). The RP’s assumptions do not hold in this setting since the lender cannot commit to using only the sum of period-one and period-two earnings in determining the interest rate. If this commitment were possible (by all lenders), the same outcomes could be implemented under a truth-telling equilibrium. 1.3.2. Avoiding Demands by Various Constituents In the above story, one group (risky firms) benefits at the expense of another (safe firms). There are other related stories. Some firms choose to smooth earnings to avoid increased tax levies and demands by employees in good years (Hepworth, 1953). Earnings also may be managed down in good years to avoid regulatory scrutiny and demands from shareholders for increased dividends distributions.18 In discussing “secret reserves” maintained by some railroads, Cole (1908, p. 217) explains: “The reason given for such practices is that such roads prefer to maintain large margins of safety for poor years, rather than to distribute extra earnings to stockholders.” Dye (1988) shows that earnings management can be used by a manager to make the current shareholders better off at the expense of the future shareholders. In his paper the presence of third-party users (potential new shareholders) of financial statements makes it beneficial for two parties (existing shareholders and managers) to write contracts that induce earnings management. We next present a variation of this story: earnings management is beneficial not because of the presence of third-party users but because of limits on the ability of the current owners to commit to future actions. In such a world of limited commitment, owners find it beneficial to install reporting systems that enable managers to credibly communicate some information while withholding other information. EARNINGS MANAGEMENT 17 2. At-will Contracts