Daan Boekel
The Impact of Chief Executive Officer
Gender on Earnings Quality
Student number: 10003076 Master Thesis, final version Date: 19th June 2014
MSc Accountancy & Control, variant Accountancy Amsterdam Business School
Faculty of Economics and Business, University of Amsterdam Supervisor: Dr. A.K. Sikalidis
Abstract
In this study I examine the relationship between Chief Executive Officer (CEO) gender and the quality of earnings. Earnings quality can be impaired by accrual-based earnings management and real activities manipulation. Based on prior literature, earnings management is viewed as an ethical issue, and women tend to judge more ethically than men. I therefore hypothesize that firms with female CEOs report higher quality earnings than firms with male CEOs, and that the positive effect of a female CEO on earnings quality would be smaller for real activities manipulation compared to accrual-based earnings management. The first hypothesis is not supported by the empirical findings, based on 2114 observations in the United States from 2004 and 2005. There is no significant relationship between CEO gender and earnings quality. I was not able to objectively measure the second hypothesis, as the coefficients for the female CEO dummy variable was insignificant in all regressions. Additional analyses also did not discover significant relationships between CEO gender and earnings quality.
Dutch summary
Tijdens de laatste decennia is de economische status van vrouwen sterk verbeterd. Als gevolg hiervan zijn er steeds meer vrouwen te vinden op topposities binnen het bedrijfsleven. Na de neergang van Enron in 2001, als gevolg van verschillende schandalen, is er steeds meer onderzoek gedaan naar diversiteit in de top van het bedrijfsleven. Het bestuur van Enron bestond voornamelijk uit mannen, wat volgens onderzoekers een rol kan hebben gespeeld in het roekeloos handelen van het management.
Mijn onderzoek was gericht op het geslacht van de algemeen directeur (CEO) van het bedrijf en wat het effect hiervan is op de kwaliteit van de gerapporteerde winst. De kwaliteit van de gerapporteerde winst kan afnemen wanneer deze winst wordt gemanipuleerd (earnings management). Dit kan enerzijds door middel van boekhoudkundige trucs (accrual-based earnings management) en anderzijds door middel van echte acties (real activities manipulation). Bij accrual-based earnings management kan worden gedacht aan het vroegtijdig nemen van winst of het te laat nemen van kosten. Hierdoor neemt de boekhoudkundige winst in de huidige periode toe, ten koste van toekomstige winst. Bij real activities manipulation kan bijvoorbeeld worden gedacht aan bezuinigingen op onderzoeks- en ontwikkelkosten. Hierdoor neemt wederom de winst in de huidige periode toe. Dit zal echter ten koste gaan van toekomstige winst, omdat het bedrijf zich niet ontwikkelt als gevolg van de bezuinigingen.
Beide vormen van earnings management zijn niet tegen de regels. Ze worden meer gezien als ethische kwesties. Men is het er echter over eens dat het ethisch onjuist is om earnings management toe te passen. Eerdere onderzoeken hebben aangetoond dat vrouwen ethischer handelen dan mannen wanneer het aankomt op het nemen van beslissingen. Mijn verwachting was daarom dat vrouwen minder earnings management toepassen dan mannen, en dat zij daardoor winsten rapporteren die van een hogere kwaliteit zijn. Tevens verwachtte ik dat het positieve effect dat een vrouwelijke CEO heeft op winstkwaliteit kleiner is voor real activities manipulation dan voor accrual-based earnings management. Dit verwacht ik omdat men real activities manipulation als minder onethisch bestempelt dan accrual-based earnings management.
Deze hypotheses heb ik onderzocht door middel van een empirisch onderzoek. Mijn sample bestond uit 2114 observaties uit 2004 en 2005. Deze jaren zijn bewust gekozen, zodat
ik mijn resultaten kon vergelijken met de resultaten van een eerder onderzoek naar het effect van het geslacht van de CFO op de winstkwaliteit. Bij 52 van deze observaties was de CEO een vrouw. Om earnings management te kunnen berekenen, maakte ik gebruik van verschillende maatstaven, gebaseerd op eerder academisch onderzoek.
Noch de hoofdanalyses noch de extra analyses leverden significante resultaten op. Op basis van deze steekproef kan ik dus concluderen dat het geslacht van de CEO geen effect heeft op de kwaliteit van de winst. Uit eerder onderzoek bleek dat het geslacht van de CFO wel een significant positief effect op de kwaliteit van de winst heeft. Ik kan dus ook concluderen dat het effect van het geslacht van de CEO op de winstkwaliteit kleiner is dan het effect van het geslacht van de CFO op de winstkwaliteit.
Table of contents
Abstract ... 2
Dutch summary ... 3
1 Introduction ... 6
2 Literature review and hypothesis development ... 8
2.1 Earnings quality ... 8
2.1.1 Earnings persistence and accrual quality ... 9
2.1.2 Real activities manipulation ... 11
2.1.3 Influence of the Chief Executive Officer on earnings quality ... 12
2.1.4 Effect of ethics on earnings quality ... 15
2.2 Differences between men and women ... 17
2.2.1 Ethical differences ... 17
2.2.2 Other differences ... 19
2.3 Hypotheses ... 19
3 Data and method ... 21
3.1 Sample selection ... 21
3.2 Research design ... 22
3.2.1 Proxies for earnings quality ... 23
3.2.1.1 Accrual quality ... 23
3.2.1.2 Real activities manipulation ... 24
3.2.2 Regression models ... 26
4 Results ... 30
4.1 Descriptive statistics ... 30
4.2 Effect of CEO gender on accrual-based earnings management ... 34
4.3 Effect of CEO gender on real activities manipulation ... 35
4.4 Difference between accrual-based earnings management and real activities manipulation ... 37
4.5 Additional analyses ... 37
4.5.1 Earnings persistence ... 37
4.5.2 Positive and negative discretionary accruals ... 39
5 Conclusion and Discussion ... 39
References ... 43
Appendix A: earnings persistence ... 46
Appendix B: positive and negative discretionary accruals ... 47
1 Introduction
During the last decades, the social and economic status improved significantly for women. This caused a significant increase in women at the top executive levels of companies (McKinsey&Company, 2007). The percentage of board seats held by women increased from 5.6% in 1990 to 12.26% in 1999 (Farrell and Hersch, 2005, p. 86). After the Enron scandal in 2001, many researchers started studies about gender diversity at top positions, whereas there was a lack of heterogeneity in the board of directors of Enron (Farrell and Hersch, 2005). Ethical studies pointed out that women behave more ethical than men, especially when it comes to judgment (Nguyen et al., 2008), and managers and accountants often view earnings management as an ethical issue (Bruns and Merchant, 1990; Kim, Park and Wier, 2012). Combining the above arguments could lead us to argue that women potentially manage earnings less than men. Dechow and Dichev found that earnings management decreases the quality of earnings (2002). Earnings can be managed by using discretionary accruals or by manipulating real activities (Kim, Park and Wier, 2012).
The literature mentioned above could indicate that women at executive positions will improve the quality of earnings, however little research in this specific field has been done. But there are some studies that suggest the existence of a positive relationship between female employees and earnings quality. Krishnan and Parsons for instance found that earnings quality increases when the senior management is more gender diverse (2008). There are also indications that female accountants are less likely to engage in earnings management (Shawver, Bancroft and Senneti, 2006).
There is one study that addresses the specific subject that will be covered in this study. That is the paper of Ye, Zhang and Rezaee (2010). They examined whether the gender of top executives affects the quality of earnings of companies in China. They classify this economy as an emerging economy. Their findings suggest that there is no significant difference between earnings quality of firms with a male chief executive officer (CEO) or chief financial officer (CFO) and earnings quality of firms with a female CEO or CFO (Ye, Zhang and Rezaee, 2010). They declare this result by the belief that males and females do not have different ethical values in China since the founding of communist China. But no empirical archival research has been done in more developed countries without a communist past. However Barua, Davidson, Rama and Thiruvadi (2010) examined the effect of CFO gender
on earnings quality in the United States. They found that earnings quality increases when the CFO is a woman. This indicates that the situation in more developed western countries differs from the situation in China. But the relation between CEO gender and earnings quality remains unclear.
Above paragraphs show that there are studies that suggest that women behave more ethical and that earnings management is an ethical issue. This indicates that the quality of earnings should be higher when the CEO is female compared to a male CEO. However prior studies showed inconsistent results. On the one hand, Barua, Davidson, Rama and Thiruvadi (2010) found in the United States that earnings quality increases when the CFO is a female. But on the other hand, studies in China suggest that there is no significant difference in earnings quality of companies with a female CEO or CFO compared to a male CEO or CFO (Ye, Zhang and Rezaee, 2010). The effect of CEO gender on earnings quality in, more developed, western countries remains unclear.
This gap in literature leads to the following research question that will be addressed in this study:
“What is the impact of the Chief Executive Officer’s gender on the quality of earnings?”
Besides the theoretical contribution, answering this question is also important for society, as higher quality earnings increases the decision usefulness of earnings information (Dechow, Ge and Schrand, 2010; Scott, 2012). This could be helpful for investors. When earnings quality seems to increase in case of a female CEO, investors could keep this in mind when considering investing in certain companies. It could also play a role in the hiring process of a new CEO by companies. If it turns out that female CEO causes higher quality earnings, companies could keep this in mind when hiring a new CEO.
The rest of this thesis will be structured as follows. In chapter 2 I will discuss prior literature to provide background information on the main topics. I will also formulate my hypotheses. The purpose of chapter 3 is to discuss the research methodology. I will explain the sample selection process, the proxies for earnings quality and the regression models. The results of the main analysis and the additional analyses will be discussed in chapter 4. Finally
chapter 5 contains the conclusion and discussion, in which I will provide a brief summary, the answer on the research question, the limitations of this study and some directions for future research.
2 Literature review and hypothesis development
In this chapter the main theoretical constructs used in this study will be explained. In the first paragraph I will cover the term earnings quality. I will point out what factors affect earnings quality and I will give in-depth explanations of those factors. I will also point out how CEOs can influence earnings quality and what the effect of ethics will be on earnings quality. The second paragraph contains information about differences between men and women. First I will discuss ethical differences between men and women. After that I will discuss some other differences. Based on the effect of ethics pointed out in the first paragraph and ethical differences between men and women explained in the second paragraph, I will develop my hypotheses about the effect of CEO gender on the quality of earnings. The other differences also play a role in this process. Examples of those differences are that women tend to be more risk-averse and are more cautious than men. These hypotheses are stated in the third paragraph.
2.1 Earnings quality
The term earnings quality can best be described by the definition provided by Dechow, Ge and Schrand in their article: “Higher quality earnings provide more information about the features of a firm’s financial performance that are relevant to a specific decision made by a specific decision-maker” (2010, p. 344). Thus, earnings quality is determined by the extent to what earnings provide relevant information about the firm’s financial performance to the users, on which the users of financial information can base their decisions. So, earnings
quality is determined by the decision usefulness of earnings information (Dechow, Ge and Schrand, 2010; Scott, 2012).
Earnings quality is determined by two important factors: earnings persistence and accrual quality (Dechow and Dichev, 2002; Dechow, Ge and Schrand, 2010). Those two factors will be addressed in the first sub-paragraph. Earnings quality can also be decreased by real activities manipulation (Kim, Park and Wier, 2012). This will be discussed in the second sub-paragraph. In the third sub-paragraph I will point out how earnings quality can be influenced by a firm’s CEO. The last sub-paragraph contains literature that states what the effect of ethics will be on earnings quality.
2.1.1 Earnings persistence and accrual quality
The section above indicated that two important determinants of earnings quality are earnings persistence and accrual quality. Those two dimensions will be discussed in this section. I will also discuss how those two dimensions are related.
Earnings persistence captures how a change in current earnings will affect an entire stream of future earnings (Ye, Zhang and Rezaee, 2010). So to what extent good or bad news in earnings will persist in the future (Scott, 2012). Earnings persistence comes in three forms: permanent, transitory and price-irrelevant (Scott, 2012). Permanent persistence means that the change in current earnings will fully persist to future earnings. For example: a production firm improved its production process, which leads to more efficient production. As a result they use less man-hours and raw materials. This results in lower production costs. The improvements made are expected to last forever. This leads to an increase in current earnings, but as the improvements are permanent, also to an increase in future earnings. The second form is transitory earnings persistence. This means that a change in current earnings might persist in future earnings, but it could also not persist. An example of this is a one-time tax reduction. As a result of the lower tax payable, the firm’s current earnings will increase. But at the time of the reduction, the firm is not sure whether future year’s tax will be reduced as well. Price-irrelevant earnings changes are changes in earnings that will never persist in the future. An example of this is the sale of a piece of land. The gain on sale will increase current
earnings, but will not persist in future earnings. The land is sold this year, so it could not be sold again in another year. Therefore the sale will not lead to future earnings.
Earnings persistence is an important determinant of earnings quality. When earnings are more persistent, current earnings are more likely to persist into future earnings. This makes it easier for the users of financial information to predict future earnings and to value the firm. Therefore, more persistent earnings are more decision useful; thus more persistent earnings lead to higher earnings quality (Dechow, Ge and Schrand, 2010).
The second dimension of earnings quality is accrual quality. Dechow and Dichev proposed this in their article (2002). They stated that net income is a composition of cash flows and accruals. So: Earnings = Cash Flow from Operations + Accruals. Those accruals can be either positive or negative. Accruals consist of changes in non-cash working capital accounts. Examples of this are: accounts receivable, accounts payable, etc. Accruals are used to shift the recognition of cash flows over time to match them with the real situation (Dechow and Dichev, 2002). For example when a customer buys a car in January 2014, and he is allowed to pay in January 2015, the car dealer can already recognize the revenue in January 2014. The dealer also recognizes a receivable for the same amount. When the customer pays the amount in January 2015, the dealer derecognizes the receivable and recognizes the received cash.
In the case illustrated above, where the cash flow occurs after the revenues and expenses are recognized, managers have to estimate the amount that will be received in the future. They also have to estimate the amount that will not be received and recognize this as an allowance for doubtful accounts. This requires judgment of the corresponding manager. The difference between the cash flow received and the expected receivable is called the accrual estimation error (Dechow and Dichev, 2002).
The higher this accrual estimation error, the lower the quality of accruals. So high quality accruals are accruals that will lead to future cash flows. Low accrual quality also has a negative effect on earnings quality, as earnings will be less aligned with real firm performance. Therefore the decision usefulness of earnings information decreases when accrual quality decreases. Those accruals can be either managed upward or downward (Dechow and Dichev, 2002). So lower accrual quality leads to lower earnings quality.
Another interesting finding of the study by Dechow and Dichev is that accrual quality and earnings persistence are related to each other. They found a negative relationship between
accrual estimation errors and earnings persistence, which is the same as a positive relationship between accrual quality and earnings persistence. This indicates that higher quality of accruals also leads to more persistent earnings. This is the case because when earnings are made up from low quality accruals, the chance that they will lead to future cash flows is low. Therefore it is not likely that the earnings will persist in the future, what leads to lower earnings persistence. A practical benefit of this finding is that accrual quality can also be used to measure earnings persistence (2002).
2.1.2 Real activities manipulation
Real activities manipulation, or real earnings management, can best be described by using the definition of Roychowdhury: “departures from normal operational practices, motivated by manager’s desire to mislead at least some stakeholders into believing certain financial reporting goals have been met in the normal course of operations” (2006, p. 336). So instead of manipulating accounting choices, real operational activities are manipulated to, for example, reach targets (Walker, 2013). Roychowdhury (2006) introduced three methods to engage in real earnings management. Those are: sales manipulation, reduced discretionary expenditures and overproduction.
Sales manipulation is a managers’ attempt to increase sales temporarily during the year to reach sales targets. This can be done in two ways. The first one is offering price discounts. As a result of those temporary discounts, sales volumes will increase, but are likely to decrease when the discounts reverse. Total sales will increase, leading to higher cash flows, but as the products are sold at a discount, the margin per product will decline, what has a negative effect on the profit margin. Production costs margin relative to sales increases due to the discount. A second way to manipulate sales is by providing more lenient credit terms to customers. A company can for example extend the credit term from 30 to 60 days. As a result, more customers are willing to buy products, as they have to pay later. So sales increase, but the cash flows related to those sales will be received in later periods than usual. So the relative cash flow from operations will decline compared to the sales level (Roychowdhury, 2006).
Discretionary expenditures are expenditures like advertising, R&D and maintenance. Usually, those costs are expensed in the period they incurred. Those expenditures often do not lead to direct profits. R&D expenditures for example might lead to new products and extra profits, however those profits will incur in later periods. Therefore it is likely that firms reduce those expenses to report higher profits in the current periods, however profits in later periods might decrease as a result of less advertising and R&D. So the reduction of discretionary expenditures has a positive effect on current cash flow from operation, however there is a risk that future cash flow will decline (Roychowdhury, 2006).
Overproduction is the third way to manage earnings upward. Manufacturing firms’ managers can decide to produce more than necessary to meet demands. When production levels increase and total fixed cost remain the same, fixed costs per produced unit decrease. This results in lower costs of goods sold and a higher profit margin. But this also results in higher production and storage costs. Those costs are often not incurred in the current period, but in later periods. So, as a result, current earnings will be higher and future earnings will decrease (Roychowdhury, 2006).
In section 2.1.1 it was described that accrual-based earnings management has no direct effect on cash flows. This differs for real earnings management, because in this case real economic decisions are changed to reach financial targets (Walker, 2013). Firms can take actions in the current period that increase current earnings. But as a result future cash flows are influenced negatively. This destroys future firm value (Roychowdhury, 2006). Gunny examined the effects of real earnings management and found that future return on assets declines for firms that engage in real earnings management. Future scaled cash flows will also decrease (2005).
2.1.3 Influence of the Chief Executive Officer on earnings quality
In this section I will discuss how earnings quality can be influenced by CEOs. First I will discuss how earnings management affects earnings quality. After that I will point out to what extent the CEO has power to manage earnings. When earnings management affects earnings quality and the CEO is able to manage earnings, the CEO is able to influence earnings quality.
Earnings quality is affected by earnings management (Dechow and Dichev, 2002; Dechow, Ge and Schrand, 2010). Walker provided a broad definition of earnings management in his article: “The use of managerial discretion over (within GAAP) accounting choices, earnings reporting choices, and real economic decisions to influence how underlying economic events are reflected in one or more measures of earnings” (2013, p. 446). In section 2.1.1 I already pointed out that the recognition of accruals requires judgment of the corresponding manager. Within General Accepted Accounting Principles (GAAP) managers got a great deal of discretion over recognition of accruals (Bergstresser and Philippon, 2006). Some examples of discretionary accounting choices are: depreciation rates, assumptions about deferred tax, valuation of stocks, provisions for bad debts and recognition of revenues (Walker, 2013). Earnings are often managed to move reported earnings in the direction of a certain earnings target (Merchant and Rockness, 1994).
Discretionary use of accruals can be either good or bad. Good earnings management can be used to transfer inside information to outsiders. When management has more timely information about free cash flows in the future, they may transport this information to the users via the recognition of accruals. However, accruals can also be used in a bad way. When managers want to increase revenues to for instance reach a certain target, they can recognize revenues more aggressively by making use of accruals to increase income for the current period. They can also manage earnings downward, when they already got a loss in a particular year, they can manage earnings more downward to be able to report higher profits next year (Walker 2013). However, many researchers believe that the quality of accruals is decreased by the amount of discretionary accruals (Dechow and Dichev, 2002; Walker, 2013). So earnings management has a negative effect on earnings quality. A key aspect of accrual-based earnings management is that there are no cash flow effects (Walker, 2013). Accruals only shift income over time, without affecting underlying cash flows. At the end of section 2.1.1 I pointed out that Dechow and Dichev also found that lower accrual quality leads to lower earnings persistence (2002). Accrual-based earnings management has therefore also a negative impact on earnings persistence.
Real activities manipulation also got a negative impact on the quality of earnings. Gunny (2005) pointed out that real earnings management destroys future firm performance, as a manager boosts current income at the cost of future cash flows. When we take a look at the definition provided by Dechow, Ge and Schrand again: “Higher quality earnings provide more information about the features of a firm’s financial performance that are relevant to a
specific decision made by a specific decision-maker” (2010, p. 344), I can conclude that real earnings management reduces the quality of earnings as the reported earnings do not provide information about the real financial performance of the firm, because they are manipulated by changing real activities. Those reported earnings are influenced by the judgment of the manager in charge (Roychowdhury, 2006).
The first part of this section pointed out that both accrual-based and real earnings management have a negative effect on earnings quality. In the rest of this section I will discuss whether and how CEOs are able to manage earnings and affect earnings quality.
Bergstresser and Philippon (2006) showed that CEOs had opportunities to manage earnings. Those opportunities arise because of the fact that earnings are, besides cash flows, also made up from accruals (Dechow and Dichev, 2002). The cash flow part of earnings is relatively easy to measure, while the accrual part involves a lot of discretion. CEOs do have great possibilities to use their discretion to manipulate earnings. This was mostly driven by incentives. The more the total compensation of the CEO depends on the share price of the firm, the more earnings are managed within the firm (Bergstresser and Philippon, 2006).
Jiang, Petroni and Wang (2010) built further on the study done by Bergstresser and Philippon by examining the relation between CFO incentives and earnings management and comparing those results with the results of Bergstresser and Philippon. They argue that the main responsibility of a CFO is financial reporting. Therefore they should have more possibilities to manage earnings compared to CEOs. Jiang, Petroni and Wang (2010) found that the relationship between accrual-based earnings management and CFO compensation was stronger than the relationship between CEOs compensation and earnings management found by Bergstresser and Philippon. From this finding they were able to conclude that CFOs have more opportunities than CEOs to manage earnings. So, CEOs do have opportunities to manage earnings, but those opportunities are not as strong as the opportunities of CFOs.
Besides accrual-based earnings management, CEOs also got opportunities to engage in real earnings management (Gunny, 2005). Real earnings management is often more costly than accrual-based earnings management, but there are still some reasons why managers would engage in real earnings management. First, aggressive accrual choices are more susceptible to SEC investigations. The litigation risk of accrual-based earnings management is therefore higher. Another reason is that accrual-based earnings management requires a certain extent of accounting flexibility for the manager, which in some cases does not exist. The last
reason is that the auditor has to accept accounting policies; this means that managing accruals is uncertain for managers, as they are not sure what manipulations will be allowed by the auditor. Managers can decide about operating activities themselves, which makes real earnings management more attractive in some cases (Gunny, 2005). Graham, Harvey and Rajgopal (2005) examined the willingness of executives to engage in real earnings management. They found that 78% of the interviewed executives are willing to sacrifice future economic firm value to improve current earnings. They also found that many of them really engaged in real earnings management. So, managers commonly use real earnings management.
From this section one can conclude that both accrual-based earnings management and real activities manipulation have a negative impact on earnings quality. The second part of the section described that CEOs do have opportunities to engage in both accrual-based earnings management and real activities manipulation; both forms of earnings management are based on judgment of the manager. So CEOs do have possibilities to influence earnings quality by managing earnings.
2.1.4 Effect of ethics on earnings quality
Earlier sections pointed out what earnings quality is and that earnings management has a negative impact on earnings quality. It also became clear that CEOs do have opportunities to manage earnings. What remains unclear is what the impact of ethics on earnings management, and thus earnings quality is. That is what will be discussed in this section.
There are some studies that examined the effect of ethics on earnings management. The studies do agree on one point: the role of ethics. They all state that earnings management is an ethical issue (Bruns and Merchant, 1990; Merchant and Rockness, 1994). Merchant and Rockness (1994) pointed out that most of the earnings management practices that have been studied are not illegal. They are in line with regulations like GAAP or IFRS. But what is questioned, is whether those earnings management actions are the right or wrong thing to do. That’s the point where ethics starts playing a role, as ethics has to do with moral reasoning
about what is considered to be right or wrong within society. So based on one’s ethical norms and values, earnings management actions can be considered to be either right or wrong.
Kim, Park and Wier (2012) argue that engaging in earnings management activities is unethical. They studied whether companies that exhibit corporate social responsibility engage less in earnings management. Their argument was that firms that exhibit corporate social responsibility would behave more ethically than firms that do not exhibit it. They will therefore engage less in both accrual-based earnings management and real earnings management. They found support for their hypothesis, I can therefore conclude that both accrual-based earnings management and real earnings management are perceived to be unethical actions (Kim, Park and Wier, 2012).
However, some researchers examined the difference between both forms of earnings management. Bruns and Merchant (1990) designed a questionnaire in which thirteen earnings management actions were described. Managers had to undertake this questionnaire and decide whether the earnings management actions are ethical or unethical. Merchant and Rockness (1994) extended and improved this method for their own study. They used the same thirteen earnings management actions but introduced a 1-5 scale to rate the actions, where a score of 1 means that the action is ethical and a score of 5 means totally unethical. The managers had to rate all the actions by using this scale.
Merchant and Rockness (1994) found some interesting results, which are really useful for this study. Their first finding was that the type of earnings management matters. They distinguished real operations earnings management and accounting based earnings management (discretionary accruals). Managers rated real earnings management as less unethically, where they rated accounting based earnings management as more unethically. Another important finding is that the direction of earnings management does not matter. In section 2.1.1 is already stated that earnings can be either managed upwards or downwards. This study pointed out that the managers rate both upward and downward earnings management the same. As a reaction on manager’s ethical perceptions about earnings management, Kaplan (2001) examined the ethical perceptions of outside parties about earnings management. His results suggest that outsiders also consider accrual-based earnings management as more unethically than real earnings management.
According to this section I can conclude that earnings management is seen as an ethical issue. Both accrual-based earnings management and real earnings management are
seen as unethical actions. However, Merchant and Rockness (1994) found that both managers and external parties rate accrual-based earnings management and real earnings management differently. Accrual-based earnings management is seen as more unethically than real earnings management.
2.2 Differences between men and women
In this paragraph I will address differences that exist between men and women, which play a role in this study. First I will focus on ethical differences that could lead to an effect in earnings quality. After that I will address some other differences that could also affect earnings quality.
2.2.1 Ethical differences
A lot of research has been done about ethical differences between men and women. However, results of those studies are mixed (McGabe, Ingram and Dato-on, 2006; Nguyen et al., 2008). Peterson, Rhoads and Vaught (2001) found out that many studies pointed out that females behave more ethical than men, but other studies show that there is no difference between men and women. There are no studies suggesting that men behave more ethical than women. In this section I will discuss the different results and try to come up with a general statement about ethical differences between men and women.
Betz, O’Connell and Shepard (2009) describe the two approaches that exist to explain ethical behaviour of men and women; the structural approach and the gender socialization approach. The structural approach assumes ethical differences between men and women, which occur in an early socialization process (childhood, upbringing, religion) will be eliminated by rewards and costs of their jobs. Followers of the structuration approach believe that behaviour of men and women is shaped by their work and the structure of the rewards related to their work. Initial differences between men and women will become smaller and
smaller when they work for a longer period for the same employer. Supporters of the structuration theory therefore predict that there are no ethical differences between men and women. On the other hand, the gender socialization approach assumes that men and women bring in different norms and values to their work. As a result, their work-related decisions, interests and practices differ. This results in different responses to rewards and costs by men and women. According to this approach, men are more driven by success, money and promotion, where women are primarily driven by performing well at their tasks and build up relationships. Supporters of the gender socialization approach expect that there are ethical differences between men and women (Betz, O’Connell and Shepard, 2009).
Kidwell, Stevens and Bethke (1987) examined if there exists a difference between the ethical decisions of female and male managers. They found no significant differences between ethical perceptions of men and women. But in the study of Betz, O’Connell and Shepard (2009) they did find a significant difference between men and women. Men were twice as likely as women to engage in unethical actions.
The study that is the most important for this research is the study done by Nguyen et al. (2008). They looked specifically at differences in ethical judgment between men and women. This is important, as earlier sections of my research pointed out that the recognition of accruals and the decisions about real activities require judgment from the corresponding manager. Nguyen et al. (2008) used the model introduced by Reidenbach and Robin (1990) to measure ethical judgment: the Multidimensional Ethics Scale (MES). It is a 1-6 scale where 1 is most unethical and 6 is least unethical. This scale measures ethical judgment by using three dimensions: moral equity, relativism and contractualism. Moral equity focuses on the individual’s perceptions of fairness, judgment and right or wrong. Relativism on the other hand, is defined as the perceptions of a social/cultural system, instead of the individual. Contractualism focuses on perceptions of right or wrong that are based on a contractual agreement between the firm and society. The respondents had to rate moral issues, which were concerned to be unethical, by using the 1-6 scale. Nguyen et al. found that female respondents rated the issues lower than men, thus they saw the issues as more unethical. Based on this finding they concluded that women judge more ethically than men. They argued that age could influence the ethical judgment of an individual, but after they controlled for the variable ‘age’, the results remained the same. Those results support the gender socialization approach of Betz, O’Connell and Shepard (2009).
So different studies came up with different results about ethical differences between men and women. But at the end Nguyen et al. (2008) found that women judge more ethically than men, which supports the gender socialization approach of Betz, O’Connell and Shepard (2009).
2.2.2 Other differences
Besides the ethical differences there are more differences between men and women that could have impact on earnings quality. Those differences are discussed in this section.
The most important difference between males and females is that females are more risk-averse and cautious than their male compounds. It is also found that this cautiousness and averseness is prevalent in decision and business judgment contexts (Byrnes, Miller and Schafer, 1999). As a result females are less likely to recognize revenues aggressively, they require more evidence before recognizing revenues. This could result in less aggressive judgment-related revenue recognition and the recognition of accruals (Barua, Davidson, Rama and Thiruvadi, 2010).
Another difference between men and women is that financial reporting of women is more in compliance with accounting regulation (Barua, Davidson, Rama and Thiruvadi, 2010). However the result of this difference would probably be small, as most of the earnings management actions are allowed according to GAAP, thus are not illegal (Merchant and Rockness, 1994).
2.3 Hypotheses
In summary, earnings quality is determined by accrual quality and earnings persistence. Both accrual-based earnings management and real activities manipulation decreases earnings quality and CEOs do have opportunities to engage in both forms of earnings management. Higher accrual quality also leads to higher earnings persistence. Prior research pointed out
that earnings management is seen as an ethical issue, where the recognition of accruals and the decisions about operating activities require judgment from the manager. It is also found that women judge more ethically than men. Women also tend to be more cautious and risk-averse, and their financial reporting is more in compliance with accounting regulations compared to men. Women would therefore probably recognize revenues less aggressive as men. I therefore expect that women will be less engaged in both accrual-based earnings management and real activities manipulation. As earnings persistence and accrual quality are related, I also expect that earnings persistence of firms with female CEOs is higher compared to firms with male CEOs. Formally stated, my first hypothesis is:
H1: Firms with female Chief Executive Officers will have higher earnings quality
than firms with male Chief Executive Officers.
However, here I take into account the findings of Merchant and Rockness (1994) about managers’ and external parties’ perceptions of earnings management activities. They found that real activities manipulation is perceived to be less unethical than accrual-based earnings management. It is therefore likely that the effect of a female CEO on real earnings management is smaller than the effect on accrual-based earnings management, because the effect of ethics plays a smaller role in real earnings management decisions. This leads to my second hypothesis:
H2: The positive effect of a female Chief Executive Officer on earnings quality will be smaller for real activities manipulation compared to accrual-based earnings management.
3 Data and method
I will perform a quantitative archival study (database research) to answer my research question. I use data from the Compustat Fundamentals Annual database. In this chapter I will point out what sample will be used for this research. After that I will discuss the design of my study. I will point out how earnings quality is measured and I will discuss the regression model.
3.1 Sample selection
For this study I will use data from the United States. This has two reasons. The first one is that Barua, Davidson, Rama and Thiruvadi (2010) examined the effect of CFO gender in the US, but the effect of CEO gender remains unclear. This makes it possible for me to compare my results about female CEOs with the results of Barua, Davidson, Rama and Thiruvadi (2010) about female CFOs. I can use their results to evaluate my own results about the effect of a female CEO on earnings quality. The second reason is a more practical reason. The United States is the only country with available executive information in the Execucomp database. The sample selection process is showed in table 1. To make it possible to compare my results with the results of Barua, Davidson, Rama & Thiruvadi (2010) I use data from the same fiscal years as they did, 2004 and 2005. I started with 1.753 firms with available CEO data in 2005. Pourciau (1993) found that executive turnover and the appointment of new executives are associated with lower earnings quality. Therefore I removed firms with a new CEO in 2005 from my sample, to prevent bias. For this reason, 329 firms were removed from the 2005 data. For the remaining 1.424 firms I downloaded data for the fiscal year 2004 and removed 229 firms with a new CEO in 2004.
After the removing of observations, there were 2.619 firms in my sample. I then removed 322 (143 in 2004 and 179 in 2005) observations that had not all data available to calculate the earnings quality proxies. Kim, Park and Wier (2012) document that financial firms have unique characteristics. I therefore removed 145 observations (68 in 2004 and 77 in 2005) in SIC codes 60 till 67. Lastly, I removed 38 (18 in 2004 and 20 in 2005) observations
with missing control variable data. My final sample consisted of 2.114 observations: 966 in 2004 and 1.148 in 2005.
3.2 Research design
To answer the research question, I designed a regression model, which is a combination of the models by Barua, Davidson, Rama and Thiruvadi (2010) and Kim, Park and Wier (2012). I use different proxies to measure earnings quality. Those are discussed in section 3.2.1. In section 3.2.2 I will explain the regression model I used in this study.
3.2.1 Proxies for earnings quality
In this study, accrual quality and real activities manipulation are the proxies for earnings quality. Accrual quality is measured by the amount of discretionary accruals and real activities manipulation is measured by four different proxies. All those proxies are calculated by separate regressions per industry-year. I require at least 15 observations per industry year (Barua, Davidson, Rama and Thiruvadi, 2010). I used a larger sample to calculate the proxies, as I otherwise did not have enough observations per industry-year to run the regressions. This sample contained 14.309 firms: 7.245 in 2004 and 7.064 in 2005. In the next sections I describe how the proxies were calculated.
3.2.1.1 Accrual quality
The absolute value of discretionary accruals (ABS_DA) is my proxy for accrual quality. Accrual quality is decreased by accrual-based earnings management, so by the amount of discretionary accruals (Walker, 2013). Thus, a higher amount of discretionary accruals leads to lower accrual quality and lower earnings quality. To measure discretionary accruals, I will make use of a modified version of Jones (1991) model. Dechow, Sloan and Sweeney (1995) found out that this model is the most effective in recognizing discretionary accruals. The model is displayed below as equation 1. In this model, a firm’s normal accruals are estimated by its change in revenues, change in receivables and property, plant and equipment. All variables are scaled by beginning of the year total assets. After running the company-year regression, this model estimates a firm’s normal accruals. The difference between the normal accruals and the actual total accruals of the firm are the discretionary accruals. This is the residual of the model. I will use the absolute value of this residual, as earnings can be managed either upward or downward (Dechow and Dichev, 2002).
Where:
TAit = Total accruals of firm i for year t. Calculated by the difference between income before
extraordinary items and cash flow from operations. Ait-1 = Total assets of firm i at the beginning of year t.
ΔREVit = Change in revenues of firm i for the year t-1 to year t.
ΔRECit = Change in receivables of firm i for year t-1 to year t.
PPEit = Property, plant and equipment of firm i for year t.
εit = Residual, representing discretionary accruals of firm i in year t.
3.2.1.2 Real activities manipulation
Following the approach of Kim, Park and Wier (2012) and Roychowdhury (2006) I will use four proxies to detect real earnings management. The first measure is abnormal operating cash flows (AB_CFO), the next one is abnormal production costs (AB_PROD), the third one is abnormal discretionary expenses (AB_EXP) and the last one is a combined measure of the first three proxies (COMBINED_RAM). To calculate the first three proxies, I use the equations used by Roychowdhury (2006). The first three proxies are calculated as follows:
CFOit/Ait-1 = β0 + β1(1/Ait-1) + β2(Sit/Ait-1) + β3(ΔSit/Ait-1) + εit (2)
PRODit/Ait-1 = β0 + β1(1/Ait-1) + β2(Sit/Ait-1) + β3(ΔSit/Ait-1) + β4(ΔSit-1/Ait-1) + εit (3)
Where:
CFOit = Cash flow from operations of firm i for year t.
PRODit = Production costs of firm i for year t. Calculated by: COGSit + ΔINVit. Thus, cost of
goods sold + change in inventory of firm i for year t.
DISEXPit = Discretionary expenses of firm i for year t. Discretionary expenses are the sum of
advertising expenses, R&D expenses and Sales, General and Administration (SG&A) expenses.
Ait-1 = Assets of firm i at the beginning of year t.
St = Sales of firm i for year t.
Sit-1 = Sales of firm i for year t-1.
ΔSit = Change in sales of firm i for year t.
ΔSit-1 = Change in sales of firm I for year t-1.
εit = Error term, representing the abnormal portions of the variables.
Equation 2, 3 and 4 calculate the normal values of cash flow from operations, production costs and discretionary expenses. Normal cash flows from operations are estimated by current year’s sales and the change in sales. Normal production costs are estimated by current year’s sales and change in sales and by the past year’s change in sales. Normal discretionary expenses are estimated by past year’s sales. Like in the modified Jones model, all variables are scaled by beginning of the year’s assets.
After the company-year regression, the models can be used to calculate the normal values of cash flow from operations, production costs and discretionary expenses for each firm. The difference between the normal value and the actual value is the abnormal cash flow (AB_CFO), abnormal production costs (AB_PROD) or abnormal discretionary expense (AB_EXP). Those are the residuals of the models.
The calculation of the combined proxy (COMBINED_RAM) is determined based on the expected directions of the first three proxies. Abnormal operating cash flows (AB_CFO) are likely to decrease, abnormal production costs (AB_PROD) are likely to increase and
abnormal discretionary expenses (AB_EXP) are likely to decrease when real activities are manipulated (Roychowdhury, 2006). Therefore, the combined proxy is calculated as follows:
COMBINED_RAM = AB_CFO – AB_PROD + AB_EXP (5)
This combined proxy is likely to decrease when real activities are manipulated.
3.2.2 Regression models
As mentioned earlier, I conducted a regression analysis to test my hypothesis that earnings quality of firms with female CEOs will be higher than earnings quality of firms with male CEOs. I designed my model with variables from the models of Barua, Davidson, Rama and Thiruvadi (2010) and Kim, Park and Wier (2012). The variable of interest that will be introduced is the dummy variable ‘FEMALECEO’. I will make use of two slightly different models. Equation 6 will be used for the accrual quality measure and equation 7 for the real activities manipulation measures. The models are stated below. Under the model, the variables will be discussed individually.
Regression models:
EQit = βo + β1FEMALECEOit + β2SIZEit + β3ATit + β4BMit + β5SGROWTHit +
β6ROAit + β7CFOit + β8BIG4it + β9LEVit +β10DEit + β11OPCYCLEit +
β12TOTALCOMPit + β13RD_INTit + β14AD_INTit + β15AB_CFOit +
β15AB_PRODit + β16AB_EXPit + εit (6)
EQit = βo + β1FEMALECEOit + β2SIZEit + β3ATit + β4BMit + β5SGROWTHit +
β6ROAit + β7CFOit + β8BIG4it + β8LEVit + β10DEit + β11OPCYCLEit +
Where:
EQit = Earnings quality proxies: ABS_DA, AB_CFO, AB_PROD, AB_EXP and
COMBINED_RAM for firm i in year t.
FEMALECEOit = 1 when the CEO of firm i is a female in year t and 0 otherwise. This is the
most important variable, which actually measures the main effect of this study.
SIZEit = Size of the firm; natural logarithm of the market value of equity for firm i in year t.
ATit = Second proxy for firm size; natural logarithm of the total assets for firm i in year t.
BMit = Book value to market value of equity ratio for firm i in year t.
SGROWTHit = Change in sales, measured as (Sit – Sit-1)/Sit-1.
ROAit = Return on assets of firm i in year t. Computed by income before extraordinary items
divided by average total assets.
LEVit = Leverage ratio of firm i in year t. Calculated by: long-term debt divided by total
assets.
CFOit = Cash flow from operations of firm i in year t, divided by assets for the year t-1.
BIG4it = 1 if firm i is audited by a Big 4 audit firm in year t and 0 otherwise.
DEit = Debt/Equity ratio of firm i in year t.
OPCYCLEit = Natural log of the operating cycle length of firm i in year t.
TOTALCOMPit = Total compensation the CEO of company i received in year t scaled by
beginning of the year assets.
RD_INTit = Research and Development intensity for firm i in year t. Calculated by: R&D
expense divided by sales.
AD_INTit = Advertising Intensity for firm i in year t. Calculated by: Advertising expense
divided by sales.
ABS_DAit = Absolute value of discretionary accruals for firm i in year t.
AB_CFOit = Abnormal cash flow from operations for firm i in year t.
AB_EXPit = Abnormal discretionary expenses for firm i in year t.
The first variable (FEMALECEOit) is the most important variable for my study. This variable
captures my main effect, namely the effect a female CEO has on earnings quality. In section 2.3 I explained my hypotheses. I expect that earnings quality increases when the CEO is a female.
Results about the relation between firm size and earnings quality are mixed. Prior literature pointed out that management of larger firms has more pressure to show more constant/predictable earnings (Pincus and Rajgopal, 2012). This may cause managers to engage in either income decreasing or income increasing earnings management. However, Dechow and Dichev (2002) found that smaller firms face lower accrual quality and lower earnings quality. And more recent studies found that fixed cost related to maintaining good internal control procedures are relatively smaller for larger firms (Ball and Foster, 1982). It is therefore likely to predict that firm size is positively related to earnings quality. Roychowdhury also documented firm size can explain variations in earnings management. To control for this effect I included two variables for firm size: SIZEit and TAit.
Menon and Williams (2004) found that the amount of abnormal accruals is associated with growth of the firm. Furthermore, Nissim and Penman (2001) found that earnings of high growth firms are less persistent. Richardson, Sloan, Soliman and Tuna (2005) document that overall earnings management possibilities increase when a firm grows. I therefore included two variables to control for growth. The first one is the book value to market value ratio (BMit), which is related to expectations of growth prospects by the market. The second one is
growth in sales in the last year (SGROWTHit) (Barua, Davidson, Rama and Thiruvadi, 2010).
The absolute value of abnormal accruals is positively related to the growth in sales and negatively to the book to market value ratio (Menon and Williams, 2004).
Dechow, Sloan and Sweeney (1995) found that earnings quality is associated with firm performance. I control for firm performance by including the variable return on assets (ROAit). Dechow and Dichev (2002) examined this effect and found that earnings quality is
negatively influenced when a firm performs poor. A second variable I included to control for firm performance is cash flow from operations (CFOit). Subramanyam (1996) found a
There are several studies that examined the relationship between earnings quality and whether a firm is audited by a Big N auditor. Becker, DeFond, Jiambalvo and Subramanyam (1998) found that earnings quality increases when a firm is audited by a Big N auditor. Chi, Lisic and Pevzner (2011) also studied the effect Big N auditors on earnings management. They found that accrual-based earnings management is constrained by Big N auditors. However as a result, firms that are audited by Big N auditors switch to real earnings management. Thus it is likely that accrual-based earnings management decreases when a firm is audited by a Big 4 auditor and that real earnings management increases.
The firm’s leverage ratio (LEVit) and debt to equity ratio (DEit) are included to control
for leverage-related incentives to manage earnings. When debt covenants are about to be violated, this could lead to earnings management to prevent this violation (Sweeney, 1994; Teoh, Welch and Wong, 1998).
The length of the operating cycle (OPCYCLEit) also has an effect on earnings quality.
Dechow and Dichev (2002) documented that accrual quality, thus earnings quality is decreased when the operating cycle is longer.
The next control variable is the total compensation received by the CEO (TOTALCOMPit). This variable is included to control for two possible effects. The first one is
corporate governance. Core, Holthausen and Larcker (1999) found that CEO compensation is associated with a firm’s corporate governance structure. A weaker corporate governance structure leads to higher CEO total compensation. At the same time a weak corporate governance structure results in more opportunities to manage earnings (Xie, Davidson and DaDalt, 2003). So higher CEO total compensation is an indicator for a weak corporate governance structure, which results in more opportunities for earnings management. Thus CEO total compensation will have a negative impact on earnings quality.
Besides this indirect effect, there is also a direct effect of CEO total compensation on earnings quality. Bergstresser and Philippon (2006) found that CEO compensation is positively associated with earnings management. So higher CEO compensation will lead to more earnings management and lower earnings quality.
Research and development intensity (RD_INTit) and advertising intensity (AD_INTit)
are also included in the model. McWilliams and Siegel (2000) found that the R&D intensity and advertising intensity can influence earnings management behaviour and thus the quality of earnings.
Firms do often use a mix of real activities manipulation and accrual-based earnings management to manipulate earnings. The decision of which method to use depends on the costs of the different methods (Zang, 2012). So real activities manipulation and accrual-based earnings management are substitutes (Cohen, Dey and Lys, 2008). When a firm manipulates more real activities, it is likely that its accrual-based earnings management practice is lower and vice versa. Therefore, I included the variables abnormal cash flows from operations (AB_CFOit), abnormal production costs (AB_PRODit) and abnormal discretionary expenses
(AB_EXPit) in equation 6 and the absolute value of discretionary accruals (ABS_DAit) in
equation 7.
4 Results
In this section I will discuss the results of my study. First I will evaluate the descriptive statistics of the sample. Then I will discuss the results of my main analysis, the effect of CEO gender on earnings quality. Lastly I will discuss some additional tests that I performed.
4.1 Descriptive statistics
The descriptive statistics of the used sample are displayed in two tables. Table 2 contains descriptive statistics about the full sample. I winsorized all continuous variables at 2,5% to control for outliers. The mean absolute value of discretionary accruals (ABS_DA) is 0,53. This is higher than in the studies of Kim, Park and Wier (2012) and Barua, Davidson, Rama and Thiruvadi (2010), who found means of 0,20 and 0,053. This gap can be explained by a difference in the calculation of the value. I did not perform the process to match discretionary accruals to performance groups per industry based on ROA. In the other two studies, they both used a different way of performance matching. My sample was relatively small, with around 1.000 observations per year. As a result most industry groups only contained 2-10 observations, thus forming performance groups did not make sense. The means of the
abnormal cash flows (AB_CFO) and abnormal production costs (AB_PROD) are 0,22 and -0,15. So, averagely real activities are not manipulated through cash flows and production costs. However, the mean values of abnormal discretionary expenses (AB_EXP) and the combined real earnings management proxy (COMBINED_RAM) are -0,63 and -0,23. This indicates that the firms in my sample tend to cut discretionary expenses to manage earnings. Most of those values are in line with the means of Kim, Park and Wier (2012), only the mean value of abnormal discretionary expenses (AB_EXP) is more negative in my sample.
In table 3 I divided my sample into two groups: one with female CEOs and one with male CEOs. This is a preliminary univariate analysis, which answers to my hypotheses before I will continue with my multivariate analysis. It enables me to spot some early differences in the earnings management variables and control variables between the two groups. There are differences in the mean and median of several earnings management variables, however those differences are far from significant. The difference closest to significant is the difference in mean abnormal cash flows (AB_CFO). This difference is 0,09 (0,31 for female CEOs and 0,22 for male CEOs), the p-value is 0,139. So, based on the descriptive statistics I cannot say that earnings quality of firms with a female CEO is higher than that of firms with a male CEO.
There are some significant differences in the control variables between the two groups. First of all the mean market value of equity (SIZE) for the male CEO group is significantly higher than for the female CEO group (difference of 0,47, p-value: 0,028). The mean total assets (TA) is for the male CEO group marginal significantly higher than for the female CEO group (difference of 0,37, p-value: 0,0907). The median is significantly higher for the male CEO group (difference of 0,51, p-value: 0,035). This indicates that companies with a female CEO are smaller compared to companies with a male CEO. The mean of BIG 4 is also significantly higher for the male CEO group (difference of 0,07, p-value: 0,040). So, firms with male CEOs are more likely to be audited by a BIG 4 auditor. This difference can be linked to the difference in firm size, as it is logical that larger firms are more likely to be audited by a BIG 4 auditor. Furthermore, I see that firms with male CEOs are more leveraged than firms with female CEOs. The medians of both the leverage ratio (LEV) and debt to equity ratio (DE) are significantly higher for the male CEO group (differences of 0,08 and 0,33, p-values: 0,035 and 0,008). The last difference is in the advertising intensity (AD_INT). The mean of the female CEO group is significantly higher (difference of 0,01, p-value: 0,003). This indicates that firms with female CEOs spend relatively more money on advertising.
4.2 Effect of CEO gender on accrual-based earnings management
The results of the regression analyses with the accrual-based earnings management proxy are displayed in table 4. The coefficients and significance levels are presented. The proxy I used to measure accrual-based earnings management is the absolute value of discretionary accruals (ABS_DA). The variable of interest is FEMALECEO. Higher absolute values of discretionary accruals imply more earnings management and thus lower accrual quality and earnings quality. I therefore expected that the FEMALECEO coefficient would be negative. The FEMALECEO coefficient is -0,035, however the p-value is 0,728. So the coefficient is, in line with my expectations, negative, but it is also far from significant. In total, seven of the control variables in the model are significant. Thus the absolute value of discretionary accruals is influenced by those variables, but the FEMALECEO variable has no significant effect. So, based on this regression there is no indication that female CEOs engage less in accrual-based earnings management than male CEOs. Therefore I reject my first hypothesis.
I can now evaluate these results by comparing them with the results of Barua, Davidson, Rama and Thiruvadi (2010). In their study, they used four different proxies for accrual-based earnings management. Their coefficients for the female CFO dummy variable ranged from -0,011 to -0,007 and all coefficients were significant. In my study, the coefficient for the female CEO dummy variable was also negative, but insignificant. So, based on their and my evidence, I can say that there is an effect for female CFOs on earnings quality but not for female CEOs on earnings quality. This indicates that the effect for female CFOs on earnings quality is larger than the effect for female CEOs on earnings quality, which is in line with the findings of Jiang, Petroni and Wang (2008), who found that CFOs have more opportunities to manage earnings than CEOs. However, an objective comparison is not possible, as I did not test the effect of female CFOs on earnings quality.
The reason for not finding a significant relationship between CEO gender and accrual quality can be declared by my sample size. When I compare my study with that of Barua, Davidson, Rama and Thiruvadi (2010) my sample is not much smaller (2.114 compared to 2.622). But what does differ is the amount of female executives. Where they got a total of 233 female CFOs in their sample, I only got 52 female CEOs in my sample.
4.3 Effect of CEO gender on real activities manipulation
The results of the regression analyses with the real activities manipulation proxies are also displayed in table 4. For each proxy, the coefficients and significance levels are displayed. I used four proxies to measure real activities manipulation. Those were: abnormal cash flows from operations (AB_CFO), abnormal production costs (AB_PROD), abnormal discretionary expenses (AB_EXP) and the combined real activities manipulation proxy
(COMBINED_RAM). When real activities are manipulated, this results in negative AB_CFO, AB_EXP and COMBINED_RAM and positive AB_PROD. I therefore expected that the FEMALECEO coefficient would be positive for the AB_CFO, AB_EXP and COMBINED_RAM regressions and negative for the AB_PROD regressions.
The FEMALECEO coefficient in the AB_CFO regression is 0,087; this is in the expected direction. However the p-value is 0,141, thus the coefficient is not significant. Six of the control variables are significant, so the level of abnormal cash flows is influenced by those variables, but not significantly by the FEMALECEO variable. In the AB_PROD regression, the FEMALECEO coefficient is -0,045. This is also in the expected direction. However the coefficient is again not significant, as the p-value is 0,592. There are eight significant control variables in the model that influence the level of abnormal production costs, but the FEMALECEO variable does not significantly influence it.
The coefficient for the FEMALECEO variable is -0,333 in the AB_EXP regression. This is not in the expected direction and the coefficient is insignificant. The p-value is 0,198. A total of six control variables are significant and influence the level of abnormal expenses, but the FEMALECEO variable has no significant effect on it. The last regression was the COMBINED_RAM regression. Here the FEMALECEO coefficient is -0,167, which is not in the expected direction. Again, the coefficient is insignificant, as the p-value is 0,325. So the FEMALECEO variable has no significant effect on the combined real activities manipulation proxy, but seven control variables do significantly influence the value.
After performing the four different regressions to examine the effect of a female CEO on real activities manipulation, I can conclude that there is no significant relationship between the gender of the CEO and the engagement in real activities manipulation. Therefore the first hypothesis is also rejected for the real earnings management proxies. Like in the discretionary accrual regression, the low number of female CEOs in the sample might play a role in the fact that I could not find a significant effect. However it is also possible that CEOs are not the ones that make the decisions regarding real activities manipulation. It is likely that those decisions are made by the Chief Operating Officer (COO). Bennett and Miles (2006) found that the COO is the person who is responsible for the oversight of daily organizational operations, so that the CEO can focus on the longer term challenges.
4.4 Difference between accrual-based earnings management and real activities manipulation
My second hypothesis was that the positive effect of a female CEO on earnings quality would be smaller for real activities manipulation compared to accrual-based earnings management. I expected this because real activities manipulation is viewed as less unethically than accrual-based earnings management, according to Merchant and Rockness (1994). It is therefore probable that women, who judge more ethically than men, engage as much in real activities manipulation as men do. I would therefore expect smaller FEMALECEO coefficients (in any direction) in the real activities manipulation regressions than in the discretionary accrual regression.
The coefficients in the real activities manipulation regressions are bigger than in the discretionary accrual regression. However the coefficients in all regressions are insignificant. It’s therefore impossible to objectively measure the differences. I can therefore not accept or reject my second hypothesis.
4.5 Additional analyses
In the main analysis, I was not able to find a significant effect of CEO gender on earnings quality. I therefore performed several additional analyses. First I tested whether earnings persistence differs between firms with male and female CEOs. I also tested the effect of CEO gender on positive and negative discretionary accruals.
4.5.1 Earnings persistence
In chapter 2, I explained that earnings persistence is a determinant of earnings quality. I therefore included an additional analysis, to examine whether earnings of firms with female CEOs are more persistent than earnings of firms with male CEOs. Earnings persistence captures how a change in current earnings will affect an entire stream of future earnings (Ye,
