Relative performanceeEvaluation and earnings management : An empirical study after the effect of relative performance evaluation on earnings management

(1)

Relative Performance Evaluation and

Earnings Management

An empirical study after the effect of Relative Performance Evaluation

on Earnings Management

June 2nd, 2014 – Final version Michelle Maas

10001858

MSc Accountancy & Control, Variant Control Amsterdam Business School

Faculty of Economics and Business, University of Amsterdam Supervisor: M. Schabus MSc

(2)

2

Abstract

From an agency perspective, relative performance evaluation (RPE) as a performance measure, seems to be a positive development, as it filters out common risks between companies. However, this performance measure may also have some adverse effects. One of them may be its effect on earnings management. To this end, Bagnoli and Watts (2000) theoretically find that, under certain circumstances, RPE leads to higher earnings management. However, this relationship has not yet been empirically investigated. That is where this study contributes. As a result, the research question was: Does the use of relative performance evaluation lead to higher earnings management? By investigating this relationship, a distinction was made between accrual-based and real earnings management. Unfortunately, not enough evidence was found to be able to conclude that the proposed relationship between RPE and earnings management exists. As the signs of some coefficients were as predicted, it was suggested that future studies take larger samples, as the coefficients will then probably become significant. In addition, it was proposed to take a sample that has income-increasing earnings management on average, as the theory of Bagnoli and Watts (2000) said that RPE would lead to higher income-increasing earnings management. The sample in this study, however, had income-decreasing earnings management on average, which may be the reason that some of the coefficients were not of the predicted sign.

(3)

3

1. Introduction

Nowadays, it is normal for a firm to have a separation of management and ownership. Agency theory describes aspects of their relationship, in which management is the agent and the shareholders are the principals. One of the problems that is described in agency theory, which is caused by the separation of management and ownership, is moral hazard. It refers to the principal being unable to observe everything that the agent does. This gives the agent incentives to act in his own interests. It has been widely known that the use of performance measures in the contract of the agent helps to mitigate this problem. However, most of the time these are individual performance measures (e.g. accounting earnings) that are influenced by noise (Gibbons & Murphy, 1990). This means agents are rewarded (or punished) for something they have almost no influence on. This is not the intention of using performance measures. Relative performance evaluation (RPE) offers a solution to this problem. When a firm uses RPE, the performance of an employee is evaluated partly by looking at the performance of peers. So the performance of a CEO may be compared to the performance of a CEO of another company. In this way, if the peer company has the same characteristics (size, industry, etc.), RPE could be a way to filter out common risks (e.g. noise) from executives’ compensation contracts (Antle & Smith, 1986). As agents are assumed to be risk averse (Jensen & Meckling, 1976), this can be considered to be a benefit of RPE.

However, also RPE has some adverse effects. One of them could be an increase in earnings management. There are two forms of earnings management, these are real earnings management and accrual-based earnings management. Real earnings management is changing the earnings by means of real economic decisions. For example, decreasing discretionary expenses, like R&D and advertising expenses (Cohen et al., 2008). Accrual-based earnings management is changing the earnings by using discretionary accruals. This means that a manager has some discretion in recording a certain component of accounting and can use this discretion to manage the earnings. RPE could give incentives to executives to use these methods, because it makes them look better than their peers and consequently they can get higher awards.

Bagnoli and Watts (2000) theoretically examine the effect of RPE on earnings management. They conclude that executives that are evaluated by means of RPE will be more inclined to manage their earnings, because they expect that executives from other firms will do so. However, this can only be concluded if there is information asymmetry so that it is

(5)

5

difficult to recognize earnings management, if investors and creditors make comparisons between firms before deciding where to invest their money, and if firms care about their own fundamental value as well as about how the market estimates their value. This would mean that the use of earnings management can be reduced if RPE is no longer used in the evaluation of an employee. However, the relationship they claim to exist, has never been tested for in an empirical setting. Therefore, the research question of this thesis is: Does the use of relative

performance evaluation lead to higher earnings management?

Now, Bagnoli and Watts (2000), as discussed above, did already investigate the relationship between these two topics. However, the relationship they find has never been tested for empirically. This thesis will contribute to the literature by being the first to empirically test this relationship and, by doing so, also distinguishing between accrual-based and real earnings management. In addition, it is important to answer this question from a societal point of view, because earnings management can be value destroying and needs to be mitigated. Mitigation strategies become more effective when the effect of RPE on earnings management is taken into account.

The remainder of this thesis proceeds as follows. Section 2 first gives an overview of some relevant literature and then proceeds to discuss the hypothesis. Section 3 discusses the empirical methodology, which also includes the description of the sample and of the variables that will be used in the regressions. Section 4 first outlines the descriptive statistics, then discusses the results and finally discusses some additional analyses that were performed. Finally, section 5 concludes and gives recommendations for future research.

(6)

6

2. Literature Review

In this chapter the most important literature is reviewed. First, agency theory and theory on performance measures are discussed. Second, research on RPE is reviewed. And finally, research on earnings management is reviewed. This is divided into a general description of earnings management and its determinants and an elaboration on performance measures as determinants of earnings management, which leads to the development of the hypothesis.

2.1 Agency theory & Performance Measures

Agency theory is about a principal (shareholders) and an agent (manager), and within this relationship problems arise because they have different interests and both are trying to maximize their own utility (Jensen & Meckling, 1976). One of the assumptions of agency theory is information asymmetry. It implies that the agent has more information than the principal, which he can use to his own advantage. One way he can do this, is by not producing the effort that is required by his contract. This problem can arise because the principal is unable to check and oversee everything and is also known as moral hazard (Eisenhardt, 1989). Making compensation sensitive to performance helps mitigate this problem, because it aligns the interests of the agent with those of the principal (Indjejikian, 1999).

Then, performance measures are needed in order to determine whether effort was sufficient for a bonus (for example) to be earned. Bushman & Smith (2001) argue that the quality, and therefore the weight put on a measure in the compensation contract, depends on sensitivity of the measure to the agent’s actions, on the one hand, and the precision of the measure, on the other hand. Together, they evaluate the agent’s controllability of the measure and the amount of noise that influences the measure. The amount of noise also has an effect on the controllability, because noise is not controllable for the agent. So when noise increases, controllability decreases. When a performance measure is of low quality, and therefore has low controllability and is influenced by a considerable amount of noise, the risk of the compensation contract of the agent increases. In this case, because the agent is risk averse, which is a maintained and important assumption of agency theory (Eisenhardt, 1989), a risk premium has to be paid to the agent (Bushman & Smith, 2001). Examples of these kind of performance measures are the more traditional financial measures, such as accounting earnings and the stock price of the firm. These are measures at the firm level and, therefore, the agent’s effort represents only a small part of the forces (e.g. all employees) that influence the measure (low controllability). In addition, these measures tend to be influenced by forces

(7)

7

outside the company, like the state of the economy (high noise). These measures are being used a lot, though, because the information needed for them is often readily available, making it possible to have a ‘low cost measurement regime’ (Tangen, 2003). However, a consequence of using these measures, is that the compensation contract is more costly, as a result of having to pay a risk premium to the agent. This reduces the efficiency of the contract. Another kind of performance measures are non-financial performance measures. These evaluate the performance of the company by looking at for example customer satisfaction and product quality and are being used increasingly. One reason for this is that non-financial performance measures are associated with the future financial performance of the company (Banker et al., 2000). These measures include less noise than financial measures, because there are less external factors influencing for example product quality than accounting earnings. However, these measures do also have drawbacks. For example, the sensitivity to the agent’s effort is again low (low controllability), because customer satisfaction, for example, is influenced for a large part by the customer service department and a manager does not have direct control over how these employees treat customers.

RPE, on the other hand, filters out common risks (noise) from compensation contracts as a result of comparing performance between companies with similar characteristics in addition to increasing the conditional controllability, as it is not literally controllable by the agent but it is informative and therefore useful (Indjejikian, 1999). Therefore, the quality of RPE as a performance measure is higher, as the (conditional) controllability increases, in addition to being a less noisy measure of performance. As a result, when using RPE as a performance measure instead of traditional financial and non-financial performance measures, the risk of the compensation contract of the agent will be lower, and therefore the risk premium that has to be paid to agents will be lower, making the contract less costly and more efficient. Before jumping into conclusions, though, elaborating more on both the benefits and the costs associated with RPE seems appropriate, which will be done in the next section.

2.2 Relative Performance Evaluation

I start by describing the circumstances in which it may be beneficial to use RPE. The results of the study of Hannan et al. (2008) show that performance is higher when they are evaluated using RPE compared to using an individual performance measure. But this relationship only holds when there is no relative performance feedback, because when there is, their performance is higher when they are evaluated using an individual performance measure compared to using RPE. Furthermore, the paper of Holmstrom (1982) shows that RPE can

(8)

8

help in decreasing moral hazard costs, because the agents will be exposed to less risks. Finally, Nalebuff and Stiglitz (1983) find that competitive systems (like RPE) are more flexible and adapt easier to a changing environment than other ways of evaluation. However, this is only the case when there is a lack of information about the difficulties that come with different tasks, when it is costly to observe inputs in a direct manner and when it is difficult to measure the outputs in a precise manner.

There are also circumstances in which it may not be beneficial to use RPE. Gibbons and Murphy (1990) note that relative performance contracts can give incentives to sabotage the performance of the other (peer) groups, to plot with co-workers and to choose unsuitable peer groups. According to them, relative performance contracts are also less desirable when the output of peer groups is difficult to measure. In addition, Knoeber and Thurman (1994) find that when agents have different abilities, the less able agents will choose riskier activities than the more able agents. So RPE may influence the agents’ risk choices. This would be the case when CEOs of two different companies are compared and one of the CEOs knows that he has less abilities than the other CEO. As a consequence he might choose riskier activities with the idea of increasing the chance of outperforming his peer. However, taking more risk may turn out to be disadvantageous to the company. Furthermore, Garvey and Milbourn (2003) find that when executives have the ability to provide insurance from market risks for themselves, most of the time the firm will not make use of RPE. The firm will only use RPE if it has a comparative advantage in providing insurance from market risks. Further, Rajgopal et al. (2006) find that the use of RPE is related to the CEO’s talent. Here, talent is proxied by the CEO’s visibility in the financial press and the CEO firm’s past industry-adjusted ROA. More specifically, they find that when a CEO’s talent is scarce, it is optimal to reward CEOs on market-wide risk. As RPE does not reward for market-wide risk (it actually filters this risk out), doing so implies that RPE is not being used. However, this relationship only holds if market-wide risk increases the market value of the firm and increases the CEO’s outside employment opportunities. The last two discussed papers show that certain attributes of the CEO (in these cases self hedging ability and talent) can mitigate the benefits of using RPE in compensation contracts. Lastly, Bagnoli and Watts (2001) find that RPE may be costly to use from an earnings management point of view. However, this effect of RPE is only present if there is enough information asymmetry so that investors may not recognize earnings management, if investors and creditors make comparisons between firms before deciding where to invest their money, and if firms care about their own fundamental value important as well as about how the market estimates their firm value.

(9)

9

Finally, it is remarkable to note that two studies, cited by Bushman and Smith (2001) have complete opposite results. First, Aggarawal and Samwick’s (1999b) model implies that strategic considerations decrease the use of RPE. This is because their model implies that the optimal contract also looks at the rival-firm performance (instead of just the own-firm performance), which creates a less aggressive pricing environment, and thus increases return to shareholders. In addition, the model implies that the weight put on the rival-firm performance in the contract, increases with the competition. As strategic considerations are more present in more competitive industries, more competitive industries make less use of RPE. Second, DeFond and Park (1999) look at CEO turnover and find that turnover is higher in industries that are classified as being more competitive, indicating that the board of directors is better able to detect CEOs that perform poorly in more competitive industries, as a result of RPE. This is consistent with their predictions of RPE being more valuable in more competitive industries, because of it, inter alia, being better able to capture common noise. These two studies are difficult to compare, as a result of totally different methods. However, they are a good representation of the RPE literature, as I can conclude from this section that there is not really a consensus on what the potential benefits and costs of RPE are, and under which circumstances these materialize. One of the costs resulting from the use of RPE was found to be an increase in earnings management, which will be elaborated on more in the next section.

2.3 Earnings Management

2.3.1 Earnings Management and its determinants

There are two ways in which managers can conduct earnings management. Real earnings management is changing the earnings by means of real economic decisions. Such decisions could be 1) to increase discounts so that sales will increase, 2) lowering the cost of goods sold by increasing production so that fixed overhead costs can be spread over a larger number of units (however, this can only be the case if the decrease in fixed costs is not offset by increases in marginal costs per unit, otherwise total costs still increase) or 3) decreasing discretionary expenses, like R&D and advertising expenses (Cohen et al., 2008). Accrual-based earnings management is changing the earnings by using discretionary accruals. This will exist, for example, if a company uses the discretionary component of its provision for bad debts to manage its earnings, as McNichols and Wilson (1988) found in their study after the

(10)

10

existence of earnings management in such provisions. So, with discretionary accruals, managers use their discretion to manage earnings.

This leads to the discussion of when to use which method of earnings management, that is, the trade-off between real and accrual-based earnings management. Cohen and Zarowin (2010) show that the tendency for firms to trade-off between the two methods of earnings management varies cross-sectionally. They find that the firms’ choice depends on the firms’ ability to use accrual-based earnings management and the costs of using it. Furthermore, Zang (2012) finds that managers treat accrual-based and real earnings management activities as substitutes (there is a negative relation between the two). Specifically, she finds that managers trade them off based on their relative costs and that they determine the level of accrual-based earnings management based on the level of real earnings management realized. The costs of each method are influenced by certain factors, which can be seen as determinants of accrual-based and real earnings management. Zang (2012) finds that the costs of accrual-based earnings management depend on the level of scrutiny of accounting practice, the amount of accounting flexibility and the length of the operating cycle. Furthermore, the costs of real earnings management depend on the competitive status in the industry, the financial condition, the level of monitoring from institutional investors and the amount of tax expenses incurred during the period.

In addition to these general determinants, other firm characteristics have been examined to find if there existed a relationship with earnings management in more specific cases. Klein (2002) examines whether audit committee and board characteristics are related to earnings management. She finds that there is a negative relation between audit committee independence and abnormal accruals. The same holds between board independence and abnormal accruals. This effect is most present when there is just a minority of outside directors in the audit committee or board. In addition, Becker et al. (1998) find that clients of Big Six auditors report lower discretionary accruals than clients of non-Big Six auditors. Their result indicates that lower audit quality is associated with more flexible accounting, and thus earnings management.

But what are the reasons to conduct earnings management in the first place, especially when considering the costs that are associated with it? A considerable amount of authors explain this practice. Burgstahler and Dichev (1997) present a theory that could explain this. The theory is based on the theory of transaction costs and tells us that in general the terms of transaction are more advantageous for firms with higher earnings, on the one hand, and that information costs are high enough for at least some stakeholders to simply determine the

(11)

11

terms of transaction based on cutoffs at zero earnings levels or zero changes in earnings, on the other hand. Both of these provide incentives to manage earnings upwards. Related to the terms of transaction, Lev (2003) states that managers manage earnings to ensure the continued support of investors and suppliers, otherwise there may be no term of transaction at all. He also acknowledges that earnings may be manipulated in order to satisfy contractual arrangements, like the restrictions that are included with loans and bonds (e.g. debt covenants). Furthermore, Bagnoli and Watts (2001) state that the market value of the company will increase because of reporting higher earnings, which also provides an incentive to manage earnings upwards. Arya et al. (1998), similarly (although reasoning the other way around), state that managers just meet or beat earnings benchmarks in order to prevent significant punishments by the market. In addition, Gunny (2010) finds that firms using real earnings management to just meet earnings benchmarks perform better in the future than firms missing the earnings benchmark or just meeting it without engaging in real earnings management. Further, Bartov et al. (2002) find that earnings management is a way for managers to signal private information they have about the firm’s future performance. Finally, and most importantly, Guidry et al. (1998) and Healy (1985) find that managers manage earnings in order to maximize the value of their bonuses (known as the bonus maximization hypothesis). More specifically, Healy (1985) presents the theory that income decreasing earnings management is more likely when the boundaries (upper and lower) of the bonus plan are binding and income increasing earnings management when these are not binding. He finds additional support for this by showing that accruals are lower at companies that have bonus plans with binding upper boundaries than at companies that do not have an upper boundary. Guidry et al. (1998) more specifically find that business unit managers use earnings management to maximize their short-term bonus. Because in both studies the bonus was based on earnings, it can be argued that financial performance measures are also determinants of earnings management. This relationship between performance measures and earnings management is discussed in the next section.

2.3.2 Performance Measures and Earnings Management

When looking at performance measures as determinants of earnings management, the studies of Guidry et al. (1998) and Healy (1985) can be used to notice that managers manipulate earnings because their bonus depends on it. This shows that financial performance measures are determinants of earnings management. That is, when compensation is tied to for example accounting earnings (meaning accounting earnings is the performance measure), this gives

(12)

12

managers incentives for manipulating earnings and thereby increasing compensation. This too will be the case with other financial performance measures, like ROE. On the other hand, Ibrahim and Lloyd (2011) find that firms that include non-financial performance measures in their compensation contracts, next to financial performance measures, have lower discretionary accruals. However, they do not get this result when looking at the incidence of meeting or just beating analyst earnings benchmarks. Here, the incentive in both cases is a cash bonus. When the incentive is equity offerings, they again find that firms with non-financial performance measures have lower discretionary accruals. In conclusion, they find (little) evidence that non-financial performance measures decrease earnings management. However, it is the relationship between RPE as a performance measure and earnings management, that was found in section 2.2, that needs to be explored. That is what the rest of this section is devoted to.

When looking at the relationship between peer performance and earnings management, the existing literature provides some indirect evidence. For example, DeChow et al. (2011) find that differences exist between firms that misstate and those that do not. Specifically, they differ in their cross-sectionally estimated and performance-matched discretionary accruals. Moreover, Fernandes and Guedes (2010) find that the expected level of economic activity (the incentive effect) and the realized level of economic activity (the need effect) jointly interact to determine whether financial fraud will take place. Both these articles thus implicitly take peer performance as a point of reference. Fung (2014) does explicitly take peer performance as a point of reference, making his findings being classified as more direct evidence. He finds that fraudulent financial reporting is more probable when low-probability reference gains and high-probability reference losses take place. On the other hand, fraudulent financial reporting is less probable when high-probability reference gains and low-probability reference losses take place. Here, reference gains and losses represent the amount that the own performance is higher or lower than the mean rival performance, this representing the fact that peer performance is taken as a reference point. The findings of this study indicate that managers are more inclined to take risks when there is a high-probability of performing lower than its rivals and a low-probability of performing higher than its rivals. When it is the other way around, managers are more inclined to be risk averse. Although these studies provide a nice starting point, none of these studies take the effect of peer performance into account when it is being explicitly used as a performance measure.

The paper of Bagnoli and Watts (2000) does consider RPE when it is being used as a performance measure and provides some (theoretically) direct evidence on the relationship

(13)

13

between the use of RPE as a performance measure and earnings management. The findings of this paper are used to develop the hypothesis. Bagnoli and Watts (2000) provide a theoretical model in which they present a few propositions. Two of the propositions rely on the fact that there are two types of benefits from stating higher earnings. First, the market value of the company will increase because of reporting higher earnings (they refer to this as the direct effect). Second, the market value of the company will increase because of reporting higher earnings relative to its competitors (they refer to this as the indirect effect). Based on these benefits, their first proposition is the following: “For those parameterizations for which no earnings management would be done absent relative performance evaluation, there exists a level of reliance on relative performance evaluation that induces earnings management in each period in equilibrium” (p. 387-388). This means that if RPE is being introduced, firms, that would otherwise have not found it beneficial, start managing their earnings. Their second proposition, following the above mentioned benefits, is as follows: “The amount of earnings management is greater in the presence of relative performance evaluation than in the absence of it” (p. 388). This means that if a firm already manages its earnings, it will increase the amount of earnings management even more if RPE is introduced. Following these propositions, I expect that the use of RPE will increase (income-increasing) earnings management. Therefore, the hypothesis is:

Firms that use RPE have a higher amount of earnings management than firms that do not use RPE.

(14)

14

3. Methodology

3.1 Sample

The year of analysis was 2010, which was randomly chosen. The sample consisted of firms from North America (the United States (U.S.)) with an Industry Classification Code between 2000 and 3999, which is the division of manufacturing. Compustat’s global fundamentals annual file (S&P 1500) was used for gathering data for the fiscal years 2008, 2009 and 2010. These three years of data were used in determining earnings management, as can be seen later on. However, for detecting the use of RPE, the compensation discussion and analysis (CD&A) in the proxy statements of just 2010 sufficed. In addition to Compustat, Execucomp was used to obtain information about the compensation of the CEOs. Bagnoli and Watts (2000) do not expect much earnings management by any firm that operates in an industry which has a dominant firm. These kind of industries would not be useful to examine in this thesis. So I verified that there is no dominant firm in the manufacturing division before I started the examination. I did this by looking at the assets, sales and net income levels. Specifically, I looked at how close the second largest company was to the largest, and I concluded from this that this space was not large enough to say that there is a dominant firm in this industry.

There are a few disadvantages of using just the manufacturing division from the U.S. The first one is that it may be difficult to generalize the findings of this thesis to other industries. Furthermore, because of only using companies from the U.S., the findings may not be generalizable to other countries or continents. Finally, the final sample may be too small to be able to detect any relationship between RPE and earnings management at all.

The data collection resulted in an initial sample of 748 companies. However, some companies had to be deleted as a result of missing values in Compustat or Execucomp. Most of the time the problem was that these firms did not provide any information about their advertising and research and development expenses. After this, there were 207 companies left. One of these had to be deleted because there was no proxy statement in electronic form publicly available for 2010. So the final sample consisted of 206 companies.

3.2 Detecting the use of RPE

To determine which companies in the final sample use RPE in their executive compensation contracts in 2010 and which do not, I used the method of Gong et al. (2011), which means that proxy statements (especially the CD&A part) from the year 2010 were read to detect the

(15)

15

use of RPE. So the explicit use of RPE was examined, and the reason for this was that “a lack of knowledge of both actual peer-group composition and the link between RPE-based performance targets and future peer performance significantly hinder the traditional implicit test from detecting RPE use” (Gong et al., 2011, p. 1007). Moreover, implicit tests rely on simplified assumptions about RPE contract details (like the composition of the peer groups) which will cause measurement errors in these tests (Murphy, 1999). Indeed, a lot of studies using the implicit test were not able to detect the use of RPE (e.g., Antle and Smith, 1986, Gibbons and Murphy, 1990, Janakiraman et al., 1992). So using the explicit test to examine the use of RPE seemed more appropriate.

A firm was classified as using RPE when at least one of the components of executive compensation was dependent on the performance of the company compared to its peers or an index (for example the S&P 500). Some companies have a general paragraph where they explain terms they use in the proxy statement. Most of the time this includes the term “performance criteria”, of which it is then explained that these can be absolute criteria or relative criteria. However, this is too general and certainly not evidence that the company uses RPE. So when this was the only thing that could be found about relative performance, the company was classified as a non-RPE firm. Furthermore, a considerable amount of companies use a peer group to benchmark their compensation to make sure that their compensation is competitive. Otherwise it will become difficult to attract the right professionals. When there was no further information given about using relative performance or other uses of the peer group, these firms were also classified as non-RPE firms. A few examples on how it was decided under which category a company belongs can be found in the appendix.

3.3 Estimating accrual-based and real earnings management

The amount of accrual-based earnings management was calculated using the modified cross-sectional Jones model (Jones, 1991) as developed in the paper of Dechow et al. (1995). It was being used in the exact same manner as it is being used by Cohen et al. (2008). For the estimation of total accruals, the model looks as follows:

with firm i and year t. Here TA stands for total accruals (calculated as the difference between earnings before extraordinary items and discontinued operations and operating cash flows),

(16)

16

Assets is total assets from the previous year, REV is the change in revenues from the preceding year and PPE is the gross value of property, plant and equipment. After that, normal accruals were estimated by using the estimated coefficients of equation (1) in the second equation. The model looks as follows:

Here NA stands for normal accruals and AR stands for the change in accounts receivable from the preceding year, all other variables are similar to equation (1). Now, it was possible to determine the amount of discretionary accruals as follows: .

Cohen et al. (2008) also provide a model for estimating the amount of real earnings management. This includes regressions of operating cash flows (CFO), discretionary expenses (DiscExp) and production costs (Prod), where production costs are defined as the sum of cost of goods sold and the change in inventory from the preceding year. Operating cash flows represent the method of giving more price discounts (or milder credit terms) in order to accelerate sales, as both discounts and milder credit terms will lead to less cash in this period relative to the level of sales. Discretionary expenses represent the method of decreasing discretionary expenses, like advertising and R&D expenses, relative to the level of sales. And production costs represent the method of lowering cost of goods sold by increasing production. However, because of the firm incurring other extra production costs, this will lead to higher annual production costs relative to the level of sales, as explained in paragraph 2.3.1. So lower levels of CFO and DiscExp and higher levels of Prod imply higher real earnings management. The model for each looks as follows:

(17)

17

Here, abnormal levels were found by subtracting normal levels from actual levels. Normal levels were calculated as follows. First a regression was run on each model in order to estimate the coefficients. These coefficients were then used to multiply with the firm year observations in order to determine the normal levels of CFO/Assetst-1, Prod/Assetst-1 and DiscExp/Assetst-1 for each firm. Then subtracting these normal levels from the actual firm year observation of CFO/Assetst-1, Prod/Assetst-1 and DiscExp/Assetst-1 (represented by CFO, PROD and DISCEXP) gave an estimation of the abnormal level of each variable (represented by RM_CFO, RM_PROD and RM_DISCEXP). Real earnings management (represented by REAL) was then defined as the sum of the three abnormal levels. As was done by Badertscher (2011), before summing them up, the proxies for RM_CFO and RM_DISCEXP were first multiplied by -1 in order to make sure that higher levels of those variables proxy for higher levels of real earnings management. This is the case because lower levels of cash flows from operations and lower levels of discretionary expenses would imply higher levels of real earnings management. So to make higher levels of these variables proxy for higher levels of real earnings management, they have to be multiplied with -1. This is not necessary for RM_PROD because higher levels of abnormal production costs proxy for higher levels of real earnings management. By looking at the three components of real earnings management separately, it was possible to conclude which method is the least costly and the least easy to see through, because that method may be used the most.

3.4 Empirical model

3.4.1 Regression formulas

After collecting the necessary data from the databases, the RPE variable and earnings management data, I ran a regression on earnings management and RPE to see if the suggested effect of RPE on earnings management is really the result of using RPE. Here, a distinction is made between accrual-based and real earnings management. Testing for this is necessary because it could be that the companies that use RPE have different characteristics than the other companies, which could be the reason that they choose RPE while the other group did not. I use control variables in order to exclude the possibility of an omitted variable bias as much as possible. The model for the accrual-based earnings management looks as follows:

(18)

18

Where DA is the outcome of the modified Jones model (the dependent variable) and RPE is an indicator variable equal to one if the firm uses RPE and zero otherwise (the independent variable). The other variables are control variables, which will be discussed in more detail below. This is the same model as Baker et al. (2003) used, which is extended by including RPE, BIG4 and REAL. BIG4 is added because Becker et al. (1998) find that clients of Big Six auditors (which is now reduced to big four) report lower discretionary accruals than clients of non-Big Six auditors, so the model controls for this.

In addition, I also ran a regression on real earnings management as it might be the case that RPE not only has an effect on accrual-based earnings management but also (or only) on real earnings management. The model for real earnings management looks as follows:

Where REAL is the outcome of the real earnings management model of Cohen et al. (2008) (the dependent variable) and RPE is an indicator variable equal to one if the firm uses RPE and zero otherwise (the independent variable). The other variables are control variables. This is the same model as Cohen et al. (2008) used, which I simplified by excluding the executive compensation controls they use and is extended by including RPE. This was also done by Kim et al. (2010) in their paper on the relationship between debt covenant slack and real earnings management. However, I excluded their GDP variable because the variable is the same for every company and as I just examine one year of data, it does not make sense to include that variable.

3.4.2 Control Variables

In the first regression formula (equation (6)), LEVER is a measure of debt, which is included because high debt gives incentives to manage earnings in order to prevent having to pay extra as a result of violating debt covenants. So a positive relationship between DA and LEVER is expected (suggesting that income-increasing accruals are used). ASSET controls for firm size effects, and is included because it is suggested that larger firms defer earnings into the future. Consequently, a negative relationship is expected here. STOCK refers to the amount of CEO equity ownership, and is included because when there is a higher amount of such ownership, the agency problems are lower, giving the agent less incentives to manage earnings. The sign of this coefficient will depend on whether there are incentives to increase (negative) or

(19)

19

decrease (positive) earnings. FIRSTYR is an indicator variable for first-year CEOs, which is included because those CEOs are associated with “big-bath related income-decreasing accruals” (Baker et al., 2003, p. 564). Therefore, a negative relationship is expected. OPTRATIO is stock option compensation scaled by salary, bonus and stock option exercises, and is included because option compensation gives incentives to manage earnings downward prior to the grant while the denominator represents the disincentive to downward earnings management. So if OPERATIO is high (more stock option compensation), then there will be more incentives to manage earnings downward. This explains the expected negative sign of the coefficient. BIG4 is an indicator variable equal to one if the company is being audited by one of the big four accountancy firms and zero otherwise and a negative sign is expected. REAL is the outcome of the real earnings management model of Cohen et al. (2008) and controls for the amount of real earnings management.

Lastly, SMOOTH refers to the pressure on management to smooth earnings. The calculation of the SMOOTH control variable requires more explanation. SMOOTH is calculated by subtracting premanaged earnings from target earnings scaled by assets. Premanaged earnings is the earnings of the current year less an amount to eliminate the effect of earnings management. The effect of earnings management is estimated by the change in the ratio of accounts receivable to revenue, the change in the ratio of current liabilities less current maturities of long-term debt to operating expenses and the change in the ratio of inventory to operating expenses. All changes in these ratios are assumed to be related to earnings management, because Baker et al. (2003, p. 563) assume that the ratio of accruals to revenue and accruals to operating expenses are constant over time. Target earnings is prior year earnings plus a four-year-average historical growth factor. So actually SMOOTH measures the gap between premanaged earnings and target earnings, and it is suggested that if the higher this gap, the more incentives to manage earnings. So, a positive sign of the coefficient is expected.

In the second regression formula (equation (7)), LMVE refers to the natural log of market value of equity and MTB is a firm’s market to book ratio, which both are included to eliminate the systematic variation in CFO, Prod and DiscExp as a result of growth opportunities (measured by MTB) and size (measured by LMVE). For both, negative coefficients are expected. LEV is a firm’s leverage defined as the ratio of total liabilities to assets, which is included because, as with the first regression formula, higher debt gives incentives to manage earnings in order not to violate debt covenants. A positive coefficient is expected. ROA is a firm’s return on assets which is defined as earnings before extraordinary

(20)

20

items divided by prior period assets, and is included because Dechow et al. (1995) argue that abnormal accruals have measurement errors with firm performance. A negative coefficient is expected. DA is the absolute value of the modified Jones model of discretionary accruals (to control for accrual-based earnings management). For the sake of completeness I showed all of the variables that are used in determining the earnings management variables and in performing the regressions, together with their definitions, the source that is used to get the needed data and the way of calculating them, in Table 1.

(21)

21

TABLE 1 Variable Definitions

Variable

name Variable definition [source]

Calculation of the variables (showing which items in the database are used)

Income Net income (loss) [Compustat]. Net income (loss)

TA

Earnings before extraordinary items and discontinued operations minus operating cash flows [Compustat].

(Net income (loss) minus extraordinary items and discontinued operations) minus Operating activities - Net cash flow

Assets Total assets [Compustat]. Assets - Total

∆REV The change in revenues from the preceding year [Compustat]. Revenue - Total this year minus Revenue - Total previous year PPE The gross value of property, plant and equipment [Compustat]. Property, Plant and Equipment - Total (Gross)

∆AR

The change in accounts receivable from the preceding year

[Compustat]. Receivables – Total this year minus Receivables – Total previous year

CFO Cash flows from operations [Compustat]. Operating Activities - Net cash flow

Sales Sales [Compustat]. Sales/Turnover (Net)

∆Sales The change in sales from the preceding year [Compustat].

Sales/Turnover (Net) this year minus Sales/Turnover (Net) previous

year

DiscExp The sum of Advertising, R&D and SG&A expenses [Compustat].

Advertising expense plus Research & Development expense plus Selling, General & Administrative expense

Prod

The sum of cost of goods sold and the change in inventories [Compustat].

Cost of Goods sold plus (Inventories – Total this year minus Inventories – Total previous year)

DA Outcome of the modified Jones model [own estimation].

SMOOTH

Target earnings - premanaged earnings where premanaged is earnings for the current year - the changes in the ratios of accounts receivable to revenu, current liabilities - current maturities of long- term debt to operating expenses and inventory to operating expenses, and target earnings is prior year earnings plus a four-year-average historical growth factor [Compustat]. The four-year average historical growth factor is a manual calculation.

(Net income (loss) previous year times (one plus the four year average historical growth factor)) minus (Net Income (loss) this year minus (Receivables - Total divided by Revenu - Total), (Long-term debt due in one year divided by Operating expenses - Total) and (Inventories - Total

divided by Operating expenses - Total))

LEVER Current maturities of long-term debt / current assets [Compustat]. Long-term debt due in one year divided by Current Assets - Total

(22)

22

TABLE 1 (Continued)

Variable

name Variable definition [source]

Calculation of the variables (showing which items in the database are used)

STOCK Year-end holdings of stock and options [Execucomp].

Shares owned - options excluded plus unexercised exercisable options

plus unexercised unexercisable options

FIRSTYR

An indicator variable equal to 1 if it is the first year of the CEO and 0

otherwise [Execucomp]. Date became CEO

OPTRATI O

Stock option compensation / (salary, bonus and stock option exercises) [Execucomp].

Value of option awards - FAS 123R ($) divided by (Salary ($) plus Bonus ($) plus Value realized on option exercise ($)

BIG4

Indicator variable equal to 1 if the firm is audited by one of the big four

auditing firms and zero otherwise [Compustat]. Auditor (4, 5, 6, and 7 are Big Four companies) RPE

Indicator variable equal to 1 if the firm uses RPE and 0 otherwise [own investigation].

REAL

The outcome of the real earnings management model estimated in this thesis.

LMVE The natural log of the market value of equity [Compustat]. Log (Stockholders Equity - Total)

MTB A firm's market to book ratio [Compustat].

Market Value - Total - Fiscal divided by (Book Value per share times Common shares used to calculate Earnings per share – Basic)

LEV Ratio of total liabilities to assets [Compustat]. Liabilities - Total divided by Assets - Total ROA Earnings before extraordinary items / prior period assets [Compustat].

(Net income (loss) minus extraordinary items) divided by Assets - Total

(23)

23

4. Results

4.1 Descriptive statistics

Table 2 presents the descriptive statistics of the sample. These numbers represent the sample after having winsorized the data. This means that the observations in the smallest percentile are set equal to the value of the 2nd percentile, while the observations in the highest percentile are set equal to the value of the 99th percentile. This is a way of filtering outliers.

From this it can be seen that total accruals are negative (-0.045). This is similar to the descriptive statistics of Cohen et al. (2008), who report a mean for total accruals of -0.10 in their full sample and, moreover, a mean of -0.07 in their Execucomp sample. Discretionary accruals was found to have a mean of -0.033, which is similar to the mean of -0.01, that

TABLE 2 Descriptive Statistics

Variable Obs Mean Median Std. Dev. Min Max

Total Accruals 206 -0.0447 -0.0435 0.0696 -0.3353 0.1920 ASSET 412 7490.99 1397.554 18760.03 51.905 114799 Sales 618 5521.901 1205.057 11493.96 55.364 67791 PPE 206 2573.284 490.295 5976.408 7.755 33924 Net Income 412 624.6971 69.2655 1960.332 -1037.56 12075 NA 206 -0.0115 -0.0107 0.0132 -0.0706 0.0384 Receivables 412 735.7454 175.5905 1584.278 3.758 9782 DA 206 -0.0332 -0.0306 0.0687 -0.3048 0.2216 CFO 206 0.1319 0.1095 0.1059 -0.2133 0.6023 DISCEXP 206 0.4085 0.3501 0.3337 0.0637 3.3680 PROD 206 0.6147 0.5313 0.4397 0.0328 2.4336 SMOOTH 201 -0.0343 -0.0190 0.4925 -5.2406 2.2584 LEVER 206 0.0386 0.0018 0.0719 0 0.4149 STOCK 206 2054.949 1083.646 3154.945 0 25253.01 FIRSTYR 206 0.0437 0 0.2049 0 1 OPERATIO 206 0.8299 0.4182 1.1282 0 7.9692 BIG4 206 0.9029 1 0.2968 0 1 RM_CFO 206 0.0630 0.0582 0.0932 -0.2197 0.3213 RM_PROD 206 -0.0420 -0.0530 0.2327 -1.4651 0.8663 RM_DISCEXP 206 0.0049 -0.0002 0.2506 -0.8122 1.4395 REAL 206 -0.1099 -0.1148 0.5116 -3.0406 1.7118 RPE 206 0.2767 0 0.4485 0 1 LMVE 200 6.5654 6.5048 1.7273 -0.4910 10.9605 MTB 206 4.2466 2.2830 22.9010 -71.5819 275.0643 LEV 206 0.4785 0.4525 0.2485 0.0329 1.1875 ROA 206 0.0313 0.0397 0.1338 -0.6550 0.4197

(24)

24

Cohen et al. (2008) found in their Execucomp sample. This negative number indicates that accruals are used to decrease earnings. Real earnings management (represented by REAL) has a negative mean (-0.1099) as well. However, this negative mean is obtained by multiplying RM_CFO and RM_DISCEXP with -1. This can be seen from looking at the three components of real earnings management before multiplication, of which both RM_CFO and RM_DISCEXP have a positive value, meaning that actual levels are higher than normal levels. Higher amounts of cash flows from operations and discretionary expenses will decrease earnings, which implies that these two methods are used to manage earnings downwards. On the other hand, the third component of REAL, RM_PROD, has a negative value, which indicates that this method is also used to manage earnings downward. So, all three components of real earnings management measure that earnings are managed downwards, but their signs are not consistent. The result of summing these components is 0.026, which is a number that does not give any insights. This is why multiplying with -1 was necessary, after which summing up RM_CFO, RM_PROD and RM_DISCEXP results in a mean of -0.1099, suggesting that REAL is used to manage earnings downward. So both DA and REAL suggest that firms manage their earnings downward. However, it has to be remembered that earnings management often reverses over time, so this downward earnings management in 2010 may be the result of income increasing earnings management in previous years.

Further, the means of LEV (0.479) and the market-to-book ratio (MTB = 4.247) can be considered to be similar to the value that Cohen et al. (2008) found in their full sample, which were 0.41 and 4.94 respectively. Lastly, the value I found for SMOOTH (-0.034) is similar to the one found by Baker et al. (2003), which was -0.025. All these findings lead to the belief that there are no significant errors present in the sample. Moreover, the findings of this study could be considered to be representative for a larger group and generalizable, because the studies which are used to compare the descriptive statistics with generally have a much larger sample.

Lastly, some noteworthy facts that can be found from Table 2. Only 4.4% of the CEOs were in their first year as a CEO in 2010. In addition, 90.6% of the companies were audited by one of the big four accountancy firms. And, perhaps most importantly, 27.7% of the companies were using some form of RPE in their executive compensation contracts in 2010. This is comparable to 25% which was found by Gong et al. (2011) and almost equal to the 28% which was found by Bannister and Newman (2003). Again this is an indication that the sample is representative.

(25)

25

Table 3 represents the correlations between the different variables used in the regressions. When looking at the most important variable, RPE, it can be seen that RPE has some significant positive correlations with other variables. For example, the correlation between ASSET and RPE is 0.2913, meaning that there is a weak positive linear relationship between the two. This weak positive relationship applies to all significant correlations that the RPE variable has with the other variables, ranging from 0.1216 to 0.2913. Unfortunately, the RPE variable is not significantly correlated to any of the earnings management variables, which could be an indication that the RPE variable will not be significant in the regression analysis. It is also noteworthy that discretionary accruals are significantly negatively correlated with OPERATIO (-0.1418), meaning that these two tend to move in opposite directions. In addition, real earnings management moves in the opposite direction of BIG4. More importantly, Table 3 tells us that accrual-based and real earnings management are positively correlated (0.3285), referring to the fact that the variables move in the same direction. On the other hand, RM_CFO and RM_DISCEXP are negatively correlated to DA, showing that they move in opposite directions. However, this actually means that they move in the same direction as a negative number for DA means income-decreasing earnings management while this is shown by a positive number for RM_CFO and RM_DISCEXP. In addition, RM_PROD is positively related to DA, showing that they move in the same direction. Furthermore, I would like to emphasize that there are only three very high correlations, which is between the real earnings management measures of RM_DISCEXP and RM_PROD 0.8476), REAL and RM_PROD (0.9664) and RM_DISCEXP and REAL (-0.9311). This is favorable for the study, because too high correlations could be indicating that those variables are measuring the same thing. If this happens, one of the variables should be eliminated. In this case, this is not necessary, as it is normal that these variables measure almost the same thing, because RM_DISCEXP and RM_PROD are a component of REAL.

Lastly, it is noteworthy that the values which correspond to the variables of LMVE, MTB, LEV and ROA are missing in the correlation table because these are lagged variables. In other words, these variables are only used for their values in the year 2009, while all other variables, except for ASSET, are used for their values in 2010. Thus, the calculation for correlations is not possible. However, it is still possible that those lagged variables are correlated to the others, since most variables do not change drastically from year to year.

(26)

26 * Numbers in bold are at the 10% significance level.

TABLE 3 Correlation Matrix

ASSET DA SMOOTH LEVER STOCK

FIRST YR

OPERA

TIO BIG4 RM_CFO

RM_ PROD

RM_DISC

EXP REAL RPE LMVE MTB LEV ROA

ASSET 1 DA 0.0026 1 SMOOTH 0.0205 -0.0371 1 LEVER 0.0525 -0.0675 0.0066 1 STOCK 0.1975 0.0576 -0.0435 0.0009 1 FIRSTYR -0.0248 -0.1035 0.0775 -0.067 -0.0874 1 OPERATIO 0.0749 -0.1418 0.0487 0.1377 0.1574 -0.0011 1 BIG4 0.1260 -0.0488 -0.0302 0.1315 -0.0436 -0.0903 0.1326 1 RM_CFO 0.0983 -0.3835 -0.0543 0.069 0.1166 -0.0586 -0.0843 -0.0048 1 RM_PROD -0.0803 0.2858 0.0394 0.0645 -0.0633 -0.0719 0.0239 -0.1721 -0.5287 1 RM_DISC EXP 0.0221 -0.2624 -0.1399 -0.0996 0.0981 -0.0558 0.0452 0.1605 0.3053 -0.8476 1 REAL -0.0653 0.3285 0.0966 0.0655 -0.0981 0.0053 0.0041 -0.1578 -0.5723 0.9664 -0.9311 1 RPE 0.2913 -0.0408 -0.0050 0.1216 0.1225 -0.0791 0.1555 0.2028 -0.0452 0.0374 -0.0434 0.0465 1 LMVE 0.6513 - - - - - - - - - - - - 1 MTB -0.0183 - - - - - - - - -0.2802 1 LEV 0.1212 - - - - - - - - - - - - -0.0009 0.1212 1 ROA 0.1575 - - - - - - - - - - - - 0.2142 0.0449 -0.0244 1

(27)

27

4.2 Multivariate Analysis

Table 4 and 5 present the results of the regressions that were performed. Table 4 presents the results of the first regression on accrual-based earnings management and Table 5 presents the results of the second regression on real earnings management. First, the results for the accrual-based earnings management are discussed.

From Table 4 it can be seen that discretionary accruals are significantly positively related to both STOCK and REAL. Regarding STOCK, this means that if the amount of year-end holdings of stock and options increases, the amount of negative discretionary accruals will decrease (because the variable will move to become more positive or less negative as in most cases here). This is consistent with the expectation, because when equity ownership increases, less agency problems will arise, and therefore there will be less incentives to manage earnings. This causes the amount of income-decreasing accruals to decrease. The same relationship was found with REAL, which implies that if real earnings management increases, the amount of negative discretionary accruals will decrease (or will become positive). This is contrary to the trade-off relationship which was found by Zang (2012). A possible explanation for these different findings could be that if income is decreased using earnings management, the two methods are used to complement each other (as would be the case here, because the mean of DA shows negative accruals), while if income is increased using earnings management, the two methods are traded-off against each other (as was found by Zang (2012)). This possible explanation, however, needs further examination. Furthermore, discretionary accruals are significantly negatively related to OPERATIO. This means that if the percentage of stock option compensation relative to other forms of compensation, like salary, increases, the amount of negative discretionary accruals increases (or will become less positive). This is also as expected, because options give incentives to manage earnings downward prior to the grant. Therefore, income-decreasing accruals will increase when more options are granted.

Lastly, unfortunately no significant relationship was found with the RPE variable. Moreover, the coefficient of the RPE variable was found to be negative, while a positive sign was predicted. This suggests that the use of RPE decreases the use of income-increasing discretionary accruals, or increases income-decreasing discretionary accruals (as was frequently found here). However, as was said, this relationship was not found to be statistically significant.

(28)

28

Table 5 presents all four regressions on the different real earnings management variables. First of all, when looking at the aggregate measure of real earnings management (REAL), it can be seen that it is significantly positively related to LEV. This means that if the ratio of total liabilities to assets increases, the value of the real earnings management variable will also increase. In other words, the variable will become less negative in most cases of this sample, because it involves income-decreasing earnings management, as was the case with discretionary accruals. This is as expected, because if liabilities are very high, this gives incentives to manage earnings upwards (or decrease downwards earnings management) in order to prevent the firm from violating debt covenants. Such a significant positive relationship was also found with the DA variable. This again shows that REAL and DA move in the same direction instead of the opposite one (as was discussed above in the DA model). Furthermore, a significant negative relationship was found with ROA, meaning that if return on assets increases, real earnings management will become more negative or less positive (so more income-decreasing real earnings management).

TABLE 4 Regression Results regarding DA

Number of observations: 201 Adjusted R-squared: 0.1158

Prediction Coefficient Significance level Standard Error

Intercept -0.0283 * .0150 RPE + -0.0103 .0108 SMOOTH -0.0069 .0092 LEVER -0.0513 .0644 ASSET 0.0000 .0000 STOCK 0.0000 * .0000 FIRSTYR -0.0323 .0219 OPERATIO -0.0086 ** .0041 BIG4 0.0082 .0159 REAL 0.0464 *** .0091

* indicates a significance level of 10% ** indicates a significance level of 5% *** indicates a significance level of 1%

(29)

29

Lastly, again no significant relationship was found with the RPE variable. However, in this case the coefficient was of the right sign. Meaning that if RPE is being used, the REAL variable will become more positive (or less negative in most cases), which suggests that the use of RPE increases increasing real earnings management, or decreases income-decreasing real earnings management. This could be indicating that if a bigger sample is used, the relationship would have been found to be significant.

When looking at the results containing the three components of real earnings management, it has to be remembered that RM_CFO and RM_DISCEXP are formulated in the opposite way. A positive number for those means that income is decreased using real earnings management (instead of a negative number that indicates this for REAL and RM_PROD). Keeping this in mind, it can be seen that most relationships found are the same as for the aggregate measure. For example, a significant negative relationship was found between LEV and RM_CFO. This indicates that if the ratio of total liabilities to assets

TABLE 5 Regression Results regarding the Real Earnings Management Variables

REAL RM_CFO RM_PROD RM_DISCEXP

Prediction Coef. (Stand. Err) Sign. Level Coef. (Stand. Err) Sign. Level Coef. (Stand. Err) Sign. Level Coef. (Stand. Err) Sign. Level Intercept -0.0493 0.0346 0.0211 0.0358 (.1566) (.0249) (.0739) (.0806) RPE + 0.0476 -0.0109 0.0240 -0.0127 (.0811) (.0129) (.0383) (.0417) LMVEt-1 -0.0247 0.0058 * -0.0123 0.0066 (.0220) (.0035) (.0104) (.0113) MTBt-1 -0.0023 0.0005 ** -0.0013 * 0.0005 (.0016) (.0003) (.0008) (.0008) LEVt-1 0.4550 ** -0.0851 ** 0.1299 * -0.2400 ** (.1549) (.0246) (.0731) (.0797) ROAt-1 -0.7853 ** 0.2791 *** -0.2919 ** 0.2143 (.2575) (.0409) (.1216) (.1325) DA 2.7293 *** -0.6345 *** 1.0625 *** -1.0322 *** (.5000) (.0794) (.2360) (.2573) Number of observations 200 200 200 200 Adjusted R-squared 0.1674 0.3596 0.1067 0.094

* indicates a significance level of 10% ** indicates a significance level of 5% *** indicates a significance level of 1%

(30)

30

increases, the variable RM_CFO will decrease, meaning that less income-decreasing real earnings management will take place (and some will become more positive). However, there were some differences with the aggregate measure. First, MTB was found to have significant relationships with RM_CFO and RM_PROD, while it did not with the aggregate measure. The relationship found indicates that when a firm’s market-to-book ratio increases, more income-decreasing real earnings management will take place (a more positive outcome for RM_CFO and a more negative outcome for RM_PROD). This is consistent with the expectation, because when firms get more growth opportunities (greater market-to-book), they are also more visible, so as a result of being afraid of getting caught, they will use less income-increasing real earnings management (or more income-decreasing real earnings management, as will be the case in this sample). Second, LMVE was found to have a significant positive relationship with RM_CFO, while it did not with the aggregate measure, which indicates that when the log of the market value of equity increases, more income-decreasing real earnings management will take place. This is consistent with the expectation, again because of increasing visibility. Finally, ROA was not found to have a significant relationship with RM_DISCEXP, while it did with all other measures of real earnings management. This could be the result of having a too small sample and not big enough results on the RM_DISCEXP variable, as the mean of this variable was close to zero (0.0049).

Unfortunately, again, none of the components of real earnings management have a statistically significant relationship to the RPE variable. However, the estimated coefficients do have the sign that was predicted, as was the case with the aggregate measure. Again, it has to be remembered here that RM_CFO and RM_DISCEXP are formulated in the opposite direction, so the fact that they have a negative sign for RPE, is consistent with the prediction.

In conclusion, after interpreting the regression results, it turns out that there is not enough evidence to support the hypothesis. Moreover, the sign of the coefficient of the RPE variable in the regression of the discretionary accruals was not even of the right sign. Fortunately, the sign of the RPE variable was right in all of the real earnings management regressions. So it could be interesting for future research to take a larger sample, and then perform the study again. This could be delivering significant results.

4.3 Additional Analyses

The multivariate analysis did not turn out to have significant results for the RPE variable. Therefore, it was proposed to take a larger sample, but there may be another improvement of this study. Namely, taking a year in which the earnings management variables show