Gender differences in conservative financial reporting in the pre- and post-sarbanes-oxley period.

(1)

1

Master thesis

Fenna Boon - 10003986

Msc Accountancy & Control

Track – accountancy

1

st

Supervisor: dr. B. Qin

2

nd

Supervisor: dr. A. Sikalidis

June 20

th

2014

Gender Differences in Conservative Financial Reporting in the Pre-

and Post-Sarbanes-Oxley Period

(2)

2 Abstract

In this paper, I study the effect of the introduction of SOX on the relationship between gender and conservatism. Prior studies have shown that companies have reported more conservatively after SOX. Furthermore, from the field of psychology and sociology it stems that women are less prone to risk than men, and since conservative reporting is associated with less risk-taking I expect that women will make more conservative reporting choices than men. This paper builds on the emerging literature on differences in gender in the financial reporting area. Because there is no universal measure for conservatism, I use three different measures in my research model: skewness of earnings, discretionary accruals and the market-to-book ratio. Using data of S&P1500 firms in the two years before and two years after SOX, I find no evidence that there is a significant relationship between gender and conservatism. I attribute the insignificance to two factors, first there is only a small percentage of firms that hire female CFO in my sample. Second, it might be that firms who hire a female CFO have different incentives influencing the accounting conservatism than firms who hire a male CFO, and that I was not able to control for those factors in my research model. Given that I was not able to establish a significant relationship between gender and conservatism, I can also not conclude that SOX influenced the relationship between conservatism and gender.

(3)

3

Over the past ten years there has been an increase in the number of female executives. In psychology, sociology and economics there have been a lot of studies examining the differences in behavior between male and female individuals, however in accounting literature this issue is relatively under-researched. The general finding in the other research areas such as psychology is that women are more risk averse then men (Eckel and Grossman ,2008; Croson and Gneezy, 2009). Drawing on this risk aversion and the recent increase in the number of female executives, financial and economic researchers have an increasing interest in the effect of gender on economic decision-making. Studies have shown that women make significantly different choices with respect to corporate decisions (Huang and Kisgen, 2008; Levi, Li and Zhang, 2008; Mohan and Chen, 2004). In the accounting literature little research has been done with respect to gender differences in financial reporting. Francis, Hasan, Park and Wu (2014) were one of the first to study the relationship between CFO gender and financial reporting decision-making, especially with respect to conservatism in financial reporting. Francis et al. conducted a quasi-natural experiment by examining companies who changed their male CFO for a female CFO. They assessed the difference in conservatism in financial reporting before and after the transition period for a sample of S&P 500 companies and found considerable evidence that female CFOs report more

conservatively than male CFOs.

Following the ‘upper echelons theory’ which was introduced by Hambrick and Mason (1984), the individual characteristics of managers have the potential to affect their decisions. In the early years of the twenty-first century these managerial characteristics became of interest to researchers, who linked specific characteristics to economic outcomes (Geiger and North, 2006; Bamber, Djiang and Wang, 2010; Dyreng, Hanlon and Maydew, 2010). Studies on how individual characteristics influence economic results showed that the specific managerial characteristics had significant explanatory power for various accounting

outcomes. For what concerns the gender-effect on specific accounting outcomes, there are generally no significant effects. This might however be due to a lack of female executives in the examined samples (Ge, Matsumoto and Zhang, 2011).

While there is only little evidence that gender influences conservatism in financial reporting, there are a lot of studies on the effect of the Sarbanes-Oxley act (hereafter

(5)

5

referred to as SOX) on conservative reporting. In 2002, SOX was introduced to enhance corporate governance and restore investor confidence. The introduction of the act was accelerated by several accounting scandals which were discovered in late 2001. The act imposed several new responsibilities on the management reporting as well as on the auditor side (Zhang, 2005). The act also increases the litigation risk and imposes penalties on

executives who do not comply with SOX. As a result, previous studies mention that

managers are likely to engage in less risky business transactions, which might not be in the best interest of investors (Zhang, 2005; Ribstein, 2002).

In the period preceding SOX, 39 percent of the 919 financial statements announced between 1997 and 2002 involved an aggressive form of earnings management. Management recognized earnings from future periods before they were even realized, which gave them opportunity to smooth out income over time. The accounting scandals discovered in late 2001 originated partly from management using the rules to their advantage in a very aggressive manner (Zhang, 2005).

SOX is aimed at protecting investors by ensuring the accuracy and completeness of financial statements. Executives who knowingly certify financial statements that do not meet the requirements of SOX are exposed to penalties of as much as 5,000,000 dollar or 20 years in prison. This potential threat should prevent managers from engaging in aggressive

earnings management.

The increased penalties decrease the incentives for managers to use their discretion in managing earnings upwards, and makes them pursue more conservative behavior when faced with uncertainty (Ribstein, 2002). Lobo and Zhou (2006) found that firms are indeed more conservative in their financial reporting in the two years after SOX (post-SOX period) than in the two years preceding SOX (pre-SOX period).

2. Motivation of the study 2.1. Research question

The research question in this paper is ‘Is there a gender difference in conservative financial reporting and is this difference moderated by SOX?’ I will first test whether there is a gender difference in conservative financial reporting, and consequently I will use a

(6)

difference-in-6

difference regression model to conclude whether SOX moderated the effect of gender on conservatism.

2.2. Motivation of the study

Up until now there are little studies examining the effect of gender on financial reporting decision-making. Francis et al. (2014) were one of the first to conduct a study on CFO gender and the difference in conservatism in financial reporting. They provide initial evidence that female CFOs make more conservative financial reporting choices than male CFOs. A factor that could have modified this assumed relationship between gender and conservatism, is the introduction of SOX. Many studies have focused on the influence of SOX on reporting

behavior and found that companies reported more conservative following the introduction of SOX (Lobo and Zhou, 2006; Lobo and Zhou, 2010 and Zhou, 2008).

Given the increased number of female executives, it is interesting to see whether female executives behave differently than male executives. Charness and Gneezy (2012) provided strong evidence that women are less prone to risk than men, which might have important implications for the profitability of the firm. Johnson and Powell (1994) argue that decision-making is one of the most important tasks of an executive. Decision-making mostly involves an outweigh of possible costs and benefits and therefore involves risk. When women are not capable to make business decisions involving a certain amount of risk, this might destroy firm value on the long-term. On the other hand, taking too much risk can destroy firm value on the long-term as well. Thus, when there are differences in conservative financial reporting between male and female CFO’s this might have important consequences for firm value.

Barua, Davidson, Rama and Thiruvadi (2010) found that there are differences between male and female executives with respect to accrual quality. In their study, they encourage other researchers to study gender differences in other settings. They argue that the implementation of new rules and frameworks, such as SOX, might have a different impact on male than on female executives. Several studies provided evidence that

companies reported more conservatively following the introduction of SOX (Lobo and Zhou, 2006; Lobo and Zhou, 2010 and Zhou (2008)). However, for this increase in conservatism, no distinction is made between firms with a male or a female CFO. Given the consequences that gender might have for accounting conservatism, it would be interesting to see if SOX had a

(7)

7

moderating effect on the possible relationship between gender and conservative financial reporting.

To the best of my knowledge, the moderating effect of SOX on gender differences in financial reporting has not been studied. This paper contributes to prior literature in several ways. First, by providing insights in the difference between male and female financial reporting decision-making. Different studies on behavioral differences between men and women do not yield unambiguous results, therefore it is important to re-examine the relationship between conservative financial reporting and gender. By controlling for other factors influencing the relationship between conservatism and gender, I attempt to find an unbiased relationship. Second, by contributing to the literature on the effects of SOX on financial reporting practices. Third, by providing insights in the possible different impact of SOX on male and female CFOs. The moderation that SOX might have on the association between gender and conservatism might be of particular importance to the users of financial statements. In making their investment decisions, it is important to fully understand the incentives of the firm that shape the financial statements. Thus the study contributes to the interpretation of the differential effects that regulations like SOX might have on financial statements of firms with a female versus a male CFO. The results of the paper might also be of interest to corporate management in the process of hiring a new CFO. To conclude, this paper can serve as a reference point for future studies in the emerging field of behavioral accounting research.

3. Literature review

3.1. The influence of specific managerial characteristics on economic and financial reporting outcomes

In 1984, Mason and Hambrick introduced the ‘upper echelons theory’. Several studies before that time had discussed different organizational ‘moves’ (Hambrick, MacMillan and Day, 1982; Harrigan, 1980; Porter, 1980). However, these studies mostly focused on rather technical aspects, separating the information streams within the company from the people involved with them. Mason and Hambrick were the first to build a theory on different organizational moves, linking them to the people involved in top-level decision-making. In organizational theory, one of the most important questions is why companies act the way they do. Since decisions are mainly made in the top of the organization, executives (referred

(8)

8

to as the ‘upper echelon’) are of particular interest in this respect.

Several studies have examined the relation between specific managerial

characteristics and economic outcomes. Geiger and North (2006) examined the changes in discretionary accruals surrounding the hiring of a new CFO. They found that discretionary accruals declined significantly more for firms hiring a new CFO than firms that did not hire a new CFO. This means that the amount of discretionary accruals is probably dependent on the position of the CFO, meaning that it could take some time for a new CFO before he or she gets the incentive to use more discretionary accruals. Bamber et al. (2010) found evidence that individual managers had significant influence over voluntary financial disclosures in a company. Managers’ individual disclosure preferences were shaped by demographical background characteristics such as their education, when they were born and their military experience.

Ge et al. (2011) conducted a very comprehensive study of the influence of CFO characteristics on accounting choices. They examined 359 CFOs across different times and different time periods to investigate whether they have a personal ‘style’ that influences the accounting choices. The CFOs in their sample held positions in several firms, thereby the researches wanted to control for the firm-specific characteristics that might influence the accounting choices. Controlling for other factors that might also influence the accounting choices, they found that the individual CFO style explained a significant portion of the variances in accounting practices.

3.2. The influence of gender on the general decision-making process

One of the most important tasks of a manager is decision-making. Since women play an increasingly important role in business management, researchers have an increasing interest in differences in the decision-making process between male and female executives. Studies on the decision-making differences between men and women focus on both the quality and nature of decision-making. Evidence from the social and behavioral sciences mostly supports the general assumption that women are more socially oriented (selfless) and men are more individually oriented (selfish) (Eckel and Grossman (1998)).

Until the 1980’s, most studies suggested that decision-making of women was of inferior quality than that of men, women were seen as less capable of problem solving (Maier and Hoffman, 1961; Milton, 1957; Sweeney, 1953). However, a closer examination of

(9)

9

the pre-1980’s research conducted by Johnson and Powell (1994) showed that the results of most of these papers were biased by creating either a very feminine or a very masculine problem-setting. The quality of decision-making of women improved when the setting was more feminine and likewise for men when there was a more masculine setting.

From the 1980’s onwards, researchers have become more skeptical towards the general assumption that there is a difference in gender with respect to decision-making. Several studies show that there are no significant differences in self-esteem, risk attitude, locus of control, flexibility and thinking and the ability to persuade others, which are factors that influence the decision-making process. A factor that differed between the male and female sample was self-confidence (Ashburner, 1991; Chaganti, 1986; Welsch and Young, 1984). An interesting finding by Eckel and Grossman (2004) is that there are no systematic differences in decision-making between male and female participants in an experiment when they are exposed to risk. In this study, several experiments were conducted in which female and male participants were sometimes exposed to risk. In the experiments where the participants were not exposed to risk, systematic differences were revealed. The choices that women make are more socially oriented than those of men, conditioned by the level of risk involved (Eckel and Grossman, 2004).

When analyzing 150 studies on gender differences in risk-taking, Byrnes et al. (1999) found considerable evidence that male participants took greater risk than female

participants. By distinguishing between several forms of risk-taking, 14 out of 16 types of risk-taking indicated significantly greater risk-taking by male participants as compared to female participants. More recently, Charness and Gneezy (2012) found strong evidence for gender differences in risk-taking. By using the results of a simple investment game,

conducted in different countries, with different instructions and different durations, Charness and Gneezy find evidence that women are more financially risk averse than men. The authors argue that their evidence is stronger than that of most other studies, since they used the results of one simple investment game, conducted in different settings. Other studies either use settings which are specifically designed to find gender differences or report the results of just one experiment, while they may have conducted several

experiments in which they did not find gender differences (Charness and Gneezy, 2012). An important conclusion that can be derived from previous literature is that there are a lot of factors influencing decision-making. Researchers have not reached a consensus

(10)

10

about the differences in decision-making between men and women. The different outcomes of the studies on gender differences in decision-making are mainly due to the different research settings. Studies conducted before 1980 mostly found significant differences in decision-making between men and women, as opposed to studies after 1980 which mostly found no significant differences. The recently conducted study on risk-taking by Charness and Gneezy on the other hand does provide strong evidence for a significant difference in decision-making between men and women. Also, Eckel and Grossman (2004) provide evidence that women are more socially oriented in their decision-making than men. This relationship is moderated by the amount of risk involved.

3.3. The influence of gender on financial reporting decision-making

Decision-taking mostly involves risk, because the outcomes of a decision are mostly not fully predictable. Because of this uncertainty the manager has to make a trade-off between the possible costs and benefits. Decision-making is therefore arguably one of the most critical factors to a firm’s success. If there is a significant difference in the decision-making process between men and women, this might have important implications for a firm’s success and the hiring of a new executive (Johnson and Powell, 1994).

There is a widespread belief that women are less prone to risk than men. This belief has caused a difference in the approach to female executives on the one hand and male executives on the other hand. An example of the different approach to female executives is that they are often offered less risky investment opportunities. This alternative approach to female executives might not be surprising, given that research has demonstrated that women make less risky investment decisions than men. Barber and Odean (2001) find that women on average hold their securities longer than men do. This results in higher returns. Barber and Odean attribute the differences to men’s overconfidence in their ability to trade. Another example of the different approach is that women are often discriminated when applying for an executive position. The belief that women are less risk-prone than men makes them less desirable for certain executive positions, in which risk-taking is seen as necessary to a firm’s success (Johnson and Powell, 1994).

While there are a lot of papers on behavioral differences between men and women, there is little research on the relationship between gender and financial reporting choices.

(11)

11

However with the increasing number of female executives, more researchers have become interested in this specific area of accounting research. Most studies focus on differences in reported accruals (Barua et al., 2010; Peni and Vähämaa, 2010; Francis et al., 2014). Barua et al. (2010) conducted a study on the differences in accrual quality between male and female CFOs. Based on prior literature they assume that female CFOs report less aggressively and therefore are likely to report lower absolute normal accruals. They find that firms with a female CFO indeed have lower absolute normal accruals and lower estimation errors, which is associated with lower risk taking and therefore a more conservative means of reporting. Another difference in reported accruals is found by Peni and Vähämaa (2010). In their study on the relationship between executive gender and earnings management, they find that firms with a female CFO or CEO are associated with income-decreasing discretionary accruals.

Francis et al. (2014) more specifically focused on the influence of gender on conservative financial reporting. In their research they conducted a quasi-natural

experiment, by investigating the financial reporting changes following the change from a male CFO to a female CFO. Using three models to measure accounting conservatism, Francis et al. find a significant increase in accounting conservatism following the hiring of a female CFO. Additionally Francis et al. emphasize the relationship between accounting conservatism and risk-aversion of female CFOs. They find that female CFOs are less likely to choose equity-based compensation, which is seen as a risky form of compensation. Furthermore

discretionary accruals are lower after a transition from a male to a female CFO, and earnings volatility is lower. Another indication of risk-averseness is that firms with a female CFO are likely to influence the investment pattern of the firm from more intangible assets to more tangible assets. Finally they find that female CFOs are more likely to reduce dividend pay-outs.

3.4. Explanations for conservatism in financial reporting

Conservative accounting has been the standard in financial reporting for a long time. There are different explanations for conservative reporting, however all of them suggest that conservatism should be beneficial to the users of the financial statements. The first

(12)

12

more information on the actual performance of the company, which they might use to their own advantage. Conservative reporting can contain this type of behavior by managers. By demanding a higher degree of verification to recognize good news as gains, managers have less opportunity to bias the financial reports. Contracting parties, such as banks, have more certainty that they will get their funds repaid, therefore conservatism leads to more efficient contracting (Watts, 2003a).

Another explanation for conservative reporting is the litigation hypothesis. When companies overstate their earnings, they can face high litigation costs. Especially with the introduction of new regulations, such as the Sarbanes-Oxley act. In most cases, the litigation cost for overstating earnings is far above the cost of understating earnings. Therefore, to protect themselves from high litigation costs, companies might choose to report

conservatively (Watts, 2003a).

Conservatism can also arise because of taxation incentives. Managers could

purposely delay high profits to later years, in order to create a tax liability in one year which will reduce the tax payments made in later years, when there is more profit. Thus, by reporting conservatively in the first (few) year(s) a company could lower the overall tax payments (Scott, 2011).

Finally, political costs can also lead to more conservative reporting. Financial reporting standard setters and regulation bodies, such as the IASB and the FASB, have different incentives with respect to the reporting style. Just as companies, regulators also face costs when companies reporting under their regimes overstate their assets. As with the fines imposed on companies, the litigation costs for overstating net assets will be higher than for understating net assets, which might lead the standard setters to create more conservative accounting standards. Under the old conceptual framework, conservatism (or prudence) was a desired qualitative characteristic of financial reporting information. Under the new conceptual framework, neutrality is promoted (Scott, 2011).

In summary, there are four explanations for conservative reporting, which have some important features in common. The contracting, litigation and political cost explanations of conservatism are all consistent with reducing opportunistic payments to either their own managers or outside parties. Also, the four explanations have in common that they require a high degree of verification.

(13)

13

3.5. Conservative reporting in the pre- and post-Sarbanes-Oxley period

The introduction of the Sarbanes-Oxley act has led to significant changes in the accounting and auditing environment. Barua et al. (2010) suggest that the introduction of SOX might have a differential impact on female CFOs and auditors than on male CFOs and auditors, given the behavioral differences they find in their study.

The enormous accounting scandals at the beginning of the 21st century including Enron, WorldCom and Arthur Andersen, have functioned as a catalyst for the introduction of SOX. Most of the accounting scandals resulted from the aggressive use of generally accepted accounting principles (GAAP) by company executives. Examples include very premature revenue recognition and income smoothing. This aggressive behavior was the cause of many corporate failures. Because of the enormous accounting scandals, investors lost their

confidence in the integrity of financial reporting. One of the most important aims of the introduction of the SOX law was to restore investor confidence. This goal needs to be achieved by improving the accuracy and reliability of financial statements (Lobo and Zhou (2006)).

Since the passage of the act in 2002, all exchange-listed companies are obliged to disclose their internal control weaknesses and furthermore executives are required to certify the material accuracy and completeness of the financial statements. Executives who

knowingly certify statements that do not meet the requirements face high penalties and might even end up in prison. Given the high financial risks that executives are exposed to, researchers have become interested in financial reporting changes following the

introduction of SOX.

Since penalties for overstating earnings are significantly higher than penalties for understating earnings, it seems a logical consequence that firms will report more

conservatively following the introduction of SOX. On this issue, Browning (2002) notes that: “requiring chief executives at the nation’s largest companies to personally endorse the company’s financial filings and face punishments if the filings are false, could make some companies file unexpectedly conservative numbers”. Consistent with this fear for

punishments, Heflin and Hsu (2008) found that companies use less non-GAAP disclosures to improve their perceived performance. On the one hand, this reduction in non-GAAP

disclosures was due to increased regulation. However even when companies were allowed to report certain non-GAAP earnings figures, they were less willing to do so.

(14)

14

In 2006, Lobo and Zhou (2006) provided some initial proof on Browning’s prediction that companies would report more conservatively after SOX. With the use of the Basu (1997) conservatism measure, they proved that companies required relatively more verification to recognize good news as gains than to recognize bad news as losses. Furthermore they show that there is a significant reduction in discretionary accruals in the post-SOX period relative to the pre-SOX period, which also indicates a more conservative form of reporting. However the study of Jain and Razaee (2004) was not able to establish a positive relationship between conservatism and the post-SOX period. Using a market-value-based, an accrual-based and an earnings/return-based measure of conservatism, they were unable to establish a

significant relation. Given that Jain and Razaee (2004) use different conservatism measures than Lobo and Zhou (2006), this might indicate that the way in which conservatism is measured determines for a large part the outcomes of the study.

Seemingly inconsistent with the findings of Lobo and Zhou (2006), Cohen, Dey and Lys (2005) found that earnings management declined after SOX. Since conservatism is associated with income-decreasing earnings management, one could expect earnings management to increase after SOX (Kim, Chung and Firth, 2003). In 2008, Zhou re-examined the relationship between SOX and conservatism and provided evidence that firms reported lower discretionary accruals, both the number and absolute value of discretionary accruals decreased (Zhou (2008)). This also explains the results from the Cohen et al. (2005) study in which it is evidenced that companies engage in less earnings management after SOX.

3.6. Hypothesis

The majority of the studies on the relationship between gender and financial reporting decision-making provides evidence for behavioral differences between male and female CFO’s, especially with respect to their risk averseness (Johnson and Powell, 1994; Francis et al., 2009; Barua et al., 2010). Barua et al. (2010) provide suggestions for further research on the relationship between CFO gender and conservatism. Given the significant influence of SOX on the accounting and auditing environment in general, they suggest that the

introduction of SOX might have a different impact on firms with a female CFO than on firms with a male CFO. Following the literature on the risk averseness of women, and the findings of Barua et al. (2010) and Francis et al. (2014) I expect that female CFOs will report more conservatively than firms with a male CFO. Formally stated my first hypothesis is:

(15)

15

H1: Firms with a female CFO will report more conservatively than firms with a male CFO.

The introduction of SOX has led to significant changes in the environment of auditing and accounting (Barua et al., 2010). By requiring the CEO and CFO of a company to personally guarantee that the financial statements fairly represent the underlying economics, CEOs and CFOs are exposed to more risk. Several studies have shown that male executives are more prone to risk than female executives (Charness and Gneezy, 2012; Johnson and Powell, 1994; Welsch and Young, 1984). There is also empirical evidence that females are more likely to comply with tax and accounting rules, implying that companies with a female CFO will be more compliant with SOX than companies with a male CFO (Barua et al., 2010). So given these gender differences in compliance with regulations and given the increased exposure to risk following the introduction of the SOX act, it might be that female executives will report even more conservative than they did before SOX. This assertion is supported by the paper of Francis et al. (2014), who found that firms with a transfer from a male to a female CFO reported more conservatively after SOX than before SOX, indicating that SOX led to an increase in conservatism for female CFOs. This would imply that SOX strengthens the relationship between conservatism and gender.

However, the evidence on the strengthened relationship between gender and conservatism following SOX is very thin. Given that female CFOs were already reporting more conservatively than male CFOs, it is fairly possible that SOX has weakened the relationship between gender and conservatism. Several studies have found a significant increase in conservatism following SOX (Lobo and Zhou, 2006; Lobo and Zhou, 2010; Zhou, 2008). Since female CFOs were already reporting more conservatively before SOX, it could be that the level of conservatism for companies with a male CFO increased more strongly than the level of conservatism for companies with a female CFO.

Summarizing, it could either be that firms with a female CFO will experience a greater increase in conservatism than firms with a male CFO, given that females are more compliant with regulations and are less prone to risk. However given that companies with a female CFO were already reporting more conservatively than firms with a male CFO, it could also be that SOX weakens the relationship between gender and conservatism, implying that the increase in conservatism will be more significant for firms with a male CFO. Giving the competing

(16)

16

theories, my second hypothesis is formulated non-directionally. Formally stated my second hypothesis is:

H2: The increase in conservatism following SOX will be different for firms with a

male versus a female CFO.

4. Research method

4.1. Introduction to the research method

To examine the change in conservatism in financial reporting following the introduction of SOX, an archival study will be undertaken. In order to be able to test H1, I created a model that tests the relationship between conservatism and gender. For H2, I use a difference-in-difference model, which allows me to both examine the relationship between gender and conservatism and the impact of SOX on the relationship between gender and conservatism. The model will be discussed in more detail in section 4.2.

The construct for H1 that needs to be measured is ‘conservatism.’ There is no universally accepted measure for conservatism. In this paper I will use the same three conservatism measures as used in the Francis et al. (2014) paper: CON_MTB, CON_ACCRUAL and CON_SKEW. By using three different measures of conservatism, my results will be less sensitive.

The first measure, CON_MTB is the market-to-book ratio of a firm (as used in Beaver and Ryan, 2000; Ahmed, Billings, Morton and Stanford-Hellis, 2002; Ahmed and Duellman, 2007). When a company reports more conservatively, it should have a lower book-to-market ratio than when it is reporting less conservatively. The idea behind this is that firms who report conservatively have an understatement of net assets relative to the market value, thus the book-to-market ratio of a firm is inversely related to the level of conservatism (Beaver and Ryan, 2000).

The second measure, CON_ACCRUAL is the cumulative of discretionary accruals and is deflated by cumulative total assets. CON_ACCRUAL is negatively related to conservatism, meaning that the lower the CON_ACCRUAL measure, the more conservative a firm reports. I multiply CON_ACCRUAL by -1, for ease of interpretation, making it positively related to accounting conservatism. Discretionary accruals are measured by the non-operating accruals

(17)

17

of a firm. The composition of this variable is described in detail in appendix A. Discretionary accruals are often used as a measure of conservatism in the accounting literature (Ahmed and Duellman, 2007; Zhang et al., 2008; Zhou, 2008). Discretionary accruals give the

manager some freedom in whether or not to recognize future expected expenses or income. When a company reports more conservatively, it would recognize future expected expenses sooner than future expected income. Most likely, future expected income will only be recognized when the company is certain that it will flow to the entity (Ahmed and Duellman, 2007).

The third measure, CON_SKEW is a measure that is focused on the skewness of earnings over time. Given that under the conservative approach companies will recognize (expected) losses immediately and (expected) income gradually, the earnings distribution over time will be negatively skewed (Givoly and Hayn, 2000). Therefore, CON_SKEW is negatively related to conservatism. I multiply CON_SKEW by -1, for ease of interpretation, making it positively related to conservatism. Furthermore, the skewness of earnings could also be due to variations in firm performance. Therefore the skewness of earnings is deflated by the skewness of cash flows (Francis et al. (2014).

4.2. Research model

4.2.1. Research model for hypothesis 1

For hypothesis 1, I test the relationship between conservatism and gender. The model is a simplified modification of the difference-in-difference model as introduced by Ge, Tanlu and Zhang (2014), and is represented by the following equation:

Model 1: CON = β0 + β1*dummyGENDER+ β2*dummySOX+ βx*CONTROL_VAR

CON = conservatism as measured by CON_MTB, CON_ACCRUAL and CON_SKEW

dummyGENDER = dummy variable, 1 if CFO is female and 0 otherwise dummySOX = dummy variable, 1 if year = post-SOX, 0 if otherwise CONTROL_VAR = control variables:

ROA – measured by: earnings before interest, taxed and

(18)

18

SIZE - the natural log of total assets (-)

LEVERAGE - total long term liabilities divided by total assets (+)

CASHHOLDING - cash and short-term investments divided by total

assets (+)

dummyLITRISK - dummy variable which is 1 if the industry belongs to a high-

litigation risk industry (SIC code 2833-2836, 3570-3577, 7370-7374, 3600- 3674, 5200-5961) and 0 otherwise (+)

Except for SIZE, all of the control variables used in the model are positively associated with accounting conservatism. This means that there is positive association with CON_ACCRUAL and a negative association with CON_MTB and CON_SKEW. The control variables used in model 1 are the same as those used in model 2. An explanation on all the control variables is given in section 4.2.2.

4.2.2. Research model for hypothesis 2

The model I will use to test the moderation by SOX on the relationship between gender and conservatism is an extension of the model I use to test hypothesis 1. I add an interaction-term (GENDER*SOX) to be able to see whether SOX moderates the relationship between gender and conservatism. The model is represented by the following equation:

Model 2: CON = β0 + β1*GENDER+ β2*SOX+ β3*GENDER*SOX+ βx*CONTROL_VAR

CON = conservatism as measured by CON_MTB, CON_ACCRUAL and CON_SKEW

dummyGENDER = dummy variable, 1 if CFO is female and 0 otherwise dummySOX = dummy variable, 1 if year = post-SOX, 0 if otherwise GENDER*SOX = interaction term

CONTROL_VAR = control variables:

ROA – measured by: earnings before interest, taxed and

depreciation divided by total assets (+*) SIZE - the natural log of total assets (-)

(19)

19

CASHHOLDING - cash and short-term investments divided by total

assets (+)

dummyLITRISK - dummy variable which is 1 if the industry belongs to a high-

litigation risk industry (SIC code 2833-2836, 3570-3577, 7370-7374, 3600- 3674, 5200-5961) and 0 otherwise (+)

The relationship between conservatism and gender, and the relationship between conservatism and SOX is influenced by a lot of factors related to the firm-specific characteristics. These factors are controlled for in the regression. Profitability (ROA) is included in the regression since highly profitable firms can more easily bear the costs of conservative accounting (Ahmed et al., 2002). Thus I would expect that high profit firms are positively associated with conservatism.

Cash holdings (CASHHOLDING) are included for the same reason as profitability.

Firms with a high amount of cash holdings can bear the costs of conservative accounting better than firms who have low cash holdings. Consequently, firms with high profitability are more likely to report conservatively.

Another firm specific factor that I control for is Size (SIZE). Lafond and Watts (2008) and Givoly et al. (2007) found that large firms face less information asymmetry. This reduced information asymmetry for large firms reduces their demand for conservative accounting. Thus I expect a negative relationship between Size and conservatism.

Leverage (LEVERAGE) is included because highly leveraged firms are more likely to

have more conflicts with their share- and bondholders. Because of the risk involved with the company, share- and bondholders likely ask for more conservative accounting practices. Research has shown that those conservative accounting practices have mitigated share- and bondholder conflicts and reduced the cost of debt (Ahmed et al., 2002; Zhang, 2006). I thus expect that highly leveraged firms are positively associated with conservatism.

Litigation risk (dummyLITRISK) is included because firms with high litigation risks are

more likely to report conservatively in order to avoid a litigation. Watts (2003a) argued that litigation risks is one of the main drivers of conservatism. Following prior studies on

conservatism, litigation is controlled for by a dummy variable which is one if the firm is in a high-litigation risk industry and zero otherwise (Ahmed and Duellman, 2007; Lafond and Roychowdhury, 2008; Francis et al. (2014). I expect the relationship between litigation risk

(20)

20

and conservatism to be positive, indicating that firms in high litigious industries will report more conservatively.

4.3. Expected outcomes of the data analysis

In accordance with hypothesis 1 I expect that firms with a female CFO will report more conservatively, thus I expect coefficient β1 to be significantly positive for all the three

measures of conservatism. Furthermore prior studies (Cohen et al., 2005; Lobo and Zhou, 2006; Barua et al., 2010) showed that firms report more conservatively following the introduction of SOX, so I expect coefficient β2 to be significantly positive as well. Coefficient

β3 represents the interaction between gender, SOX and conservatism. Following hypothesis

2 which states that the increase in accounting conservatism is different for firms with a male versus a female CFO, I have no upfront expectation on coefficient β3. Assuming a positive

relationship between dummyGENDER and conservatism, a positive β3 would indicate that

SOX strengthens the association between conservatism and (female) CFO gender, whereas a negative β3 would indicate that SOX weakens the relationship between conservatism and

(female) CFO gender. The expectations on all the different coefficients are summarized in table 3.

4.4. Sample selection

For the sample selection, I retrieved data from the Compustat database. I used firms from the S&P 1500 database, since these firms are all obliged to comply with SOX after its enactment. Data on the gender of the CFO will mainly be retrieved from the Execucomp database. The time-sample I want to use is from 2000 to 2001 and from 2003 to 2004, in order to cover the two years before and the two years after the enactment of SOX. Since 2002 is the year that SOX was introduced and corporate management probably had a lot of difficulties implementing the necessary controls, I exclude this year from my sample.

In the final sample of 516 observations I had 156 observations that lacked data on the CFO gender. Since this is a crucial data item in my analysis, I manually searched for the gender of the CFO during the sample years. I found the information on the company websites, Forbes.com, via Google.com and Yahoo.com. In table 1, the composition of the final sample is presented.

(21)

21 Table 1: data collection

Observations including missing data and data from the years 1999 and 2002

10990

Observations that are not in the sample for the full test period

(9940)

Observations 1999 (156)

Observations 2002 (327)

Observations that lack data on SIZE (23) Observations that lack data on LEVERAGE (27) Observations that lack data on

CASHHOLDING

(1)

Deleting outlier observations (77)

Final sample 439

After the initial data retrieval, I had 10990 observations. Since I require that every company is in the sample for the years 2000, 2001, 2003 and 2004, I deleted 9940 observations that did not meet this requirement. The 10990 observations also included data on the years 1999 and 2002, which I needed to calculate certain variables that measure the difference in a particular balance between two years(for example Δaccounts payable, see appendix A for a complete overview). The years 1999 and 2002 are deleted from the dataset, because these years do not belong to the final sample (2000-2001 and 2003-2004). For 1999 and 2002, I deleted respectively 156 and 327 observations. Some observations had missing data on the control variables SIZE, LEVERAGE and CASHHOLDING. I also deleted these observations, which totaled 51 observations. Furthermore, I deleted outlier observations which were in the upper or under 1% of the observations. It left me with a final sample of 439

(22)

22 4.5. Descriptive statistics

In table 2.1, the descriptive statistics for the variables used in the regressions are presented.

Table 2.1: descriptive statistics

Variable Obs Mean Std. Dev Min Max

CON_SKEW 439 0.17587 2.08728 -1.030575 5.89743 CON_ACCRUAL 132 -0.01879 0.06437 -1.56879 1.32451 CON_MTB 425 1.22433 1.64681 0.062444 20.6486 dummyGENDER 439 0.054670 0.22759 0 1 dummySOX 439 0.72210 0.44848 0 1 ROA 439 0.02083 0.10333 -0.89161 0.28457 SIZE 439 7.56969 1.66846 3.94600 12.07255 LEVERAGE 439 0.18940 0.14904 0 0.667231 CASHHOLDING 439 0.12234 0.15965 0 0.81323 dummyLITRISK 439 0.25474 0.43770 0 1

CON_SKEW and CON_MTB both have substantially more observations than CON_ACCRUAL, which will yield more reliable results. The reason that CON_ACCRUAL has less observations than CON_SKEW and CON_MTB is that CON_ACCRUAL is composed of a detailed set of variables, on which I missed a lot of data. The means for CON_SKEW, CON_ACCRUAL and CON_MTB are -0.17587, -0.01879 and 1.84710. These variables are similar to the descriptive statistics in other studies on conservatism (Ahmed et al., 2002; Zhang, 2008; Francis et al., 2014). Furthermore we see that the mean for the dummyGENDER is 0.054670, which means that 5.4622% of the sample consists of female CFOs. The dummy statistic on litigation risk (dummyLITRISK) is 0.25474, indicating that 26.6807% of the sample belong to high risk industries. The mean for dummySOX is 0.72210, which means that 72% of the observations is post-SOX, so the results might be driven by the post-SOX period.

(23)

23

The firms in the sample on average have an ROA of 2.1%, which means that net income is on average 2.5 percent of total assets. The average size of the assets is 8,188,320 US dollar. The mean debt-to-equity ratio is 0.1867, indicating that for every 1 dollar of equity there is 0.1867 dollar of debt.

In table 2.2, the correlation-matrix is reported. There are a few variables that are strongly correlated with each other. First, dummyGENDER and dummyGENDER*dummySOX are strongly correlated (0.8587). This strong correlation exists because dummyGENDER is part of the variable dummyGENDER*dummySOX. Furthermore, ROA is strongly related with CON_SKEW (-0.6879), because the skewness of earnings is a modification of the ROA (see appendix A). There is also a strong negative correlation between ROA and CON_MTB (-0.7019) indicating that firms with a high ROA have a low market-to-book value.

In table 2.3, the distribution of the female CFOs over the sample is presented. As can be seen, there are more female CFOs in the post-SOX period than in the pre-SOX period. This is in line with the claimed increase in the number of female executives. From 2000 to 2001, the percentage of female executives has declined from 4.762% to 2.970%. In the two years after SOX, the percentage of female CFOs has increased from 7.752% to 5.600%. Thus on average, the amount of female CFOs is significantly larger in the post-SOX period than in the pre-SOX period. The overall mean of female CFOs in the entire sample is 5.4622%.

Table 2.3: Distribution of female CFOs over the sample years

Year/statistics Obs Mean Std. dev.

2000 84 0.04762 0.21424

2001 101 0.02970 0.17061

2003 129 0.07752 0.26846

(24)

24 Table 2.2: Correlation matrix

CON_SKEW CON_ACCRUAL CON_MTB Dummy GENDER

dummySOX DummyGENDERx dummySOX

ROA SIZE LEVERAGE CASH-HOLDING Dummy LITRISK CON_SKEW 1.0000 CON_ACCRUAL .0045 1.0000 CON_MTB -.6445 -.1115 1.0000 dummyGENDER .0382 -.0253 -.0760 1.0000 dummySOX -.0520 .1691 -.0501 .0434 1.0000 dummyGENDERx dummySOX .0274 .1243 -.1195 .8587 .1569 1.0000 ROA -.6879 .0273 -.7019 -.0462 -.1143 0.0535 1.0000 SIZE -.0444 .0533 -.1989 -.0515 -.2011 -.0166 -.1542 1.0000 LEVERAGE .1849 -.1021 -.3267 -.0352 .1103 -.0481 -.2332 .01357 1.0000 CASHHOLDING -.1497 -.0942 -.3484 -.0948 .1472 -.0807 .1918 -.1666 0.5274 1.0000 dummyLITRISK -.2364 .0057 .2466 .0827 -.0064 .0491 .3705 -.0535 -.1613 .2705 1.0000

(25)

25 5. Results

In this section, the results will be described for the variables CON_SKEW, CON_ACCRUAL and CON_MTB. Before presenting the main results, I will first test for differences in the mean and median of the most important firm-level characteristics for the firms with a female CFO and firms with a male CFO. The purpose of these tests is to show whether there are structural differences in firms with a female or a male CFO, which might explain their reporting

behavior. In table 3, the statistics for the difference in the mean are presented, using a t-test for independent samples.

Table 4: t-test for difference in means

Mean value male Mean value female t-statistic p-value Total assets 8829.3600 6721.4810 -0.5440 0.5867 Market value 6944.4990 4833.6440 -0.6457 0.5188 Net income 293.7545 294.3629 0.0033 0.9974 ROA 0.0192 0.0376 1.0814 0.2801 SIZE 7.5863 7.3942 -0.6780 0.4981 LEVERAGE 0.1928 0.1532 -1.5681 0.1176 CASHHOLDING 0.1203 0.1438 0.8659 0.3870 dummyLITRISK 0.2668 0.1579 -1.4683 0.1428

As can be seen in table 4, all firm-level characteristics have no significantly different means at the p=0.05 level. The only two firm-level characteristics that come close to having a significantly different mean are LEVERAGE (with an uncertainty level of 11.76% and dummyLITRISK (with an uncertainty level of 14.28%). However I must acknowledge that I have substantially more observations for firms with a male CFO than for firms with a female

(26)

26

CFO (respectively 401 and 38 observations), which might be of influence of the results.

Table 5: Wilcoxon rank-sum test for difference in the median

ranksum value male ranksum value female Z-statistic P-value Total assets 88827 7753 -0.812 0.4168 Market value 83815 6710 -1.070 0.2845 Net income 89013 7567 -1.061 0.2887 ROA 87805 8775 0.555 0.5788 SIZE 88827 7753 -0.812 0.4168 LEVERAGE 89480 7090 -1.7000 0.0892 CASHHOLDING 88113 8467 0.143 0.8862 dummyLITRISK 89050 7530 -1.466 0.1462

As can be seen in table 5, all firm-level characteristics do not have significantly different medians at the p=0.05 level. The only firm-level characteristic that is significant at the p=0.1 level, is LEVERAGE. Given that the difference in the mean value of leverage between firms with a male and female CFO was also almost significant, and given the significantly different median that I found, it might be that the amount of leverage influences the reporting behavior of a firm. However, I should note again that I only have 38 observations for firms with a female CFO and 401 for firms with a male CFO.

In table 6, the outcomes of the regressions for H1 are shown for each of the three variables measuring conservatism.

(27)

27 Table 6: results for hypothesis 1

Predicted Sign

CON_SKEW CON_ACCRUAL CON_MTB R-squared:0.3792 R-squared:0.0884 R-squared:0.2667

Constant ? 0.5356 (p=0.244) -0.3153 (p=0.436) 1.4389*** (p=0.000) dummyGENDER + 0.0345 (p=0.921) -0.1510 (p=0.499) 0.0093 (p=0.976) dummySOX + -0.1495 (p=0.405) 0.0278** (p=0.024) -0.2127 (p=0.174) ROA + -12.9115*** (p=0.000) 0.6006 (p=0.681) 2.6480*** (p=0.000) SIZE - 0.0300 (p=0.559) 0.0494 (p=0.224) -0.0682 (p=0.133) LEVERAGE + -1.2878** (p=0.035) -0.9889** (p=0.039) -0.9616* (p=0.073) CASHHOLDING + -0.2736 (p=0.638) -1.0746** (p=0.028) 3.7763*** (p=0.000) dummyLITRISK + 0.2535 (p=0.191) 0.0412 (p=0.780) -0.4345*** (p=0.010)

*** = significant at the p=0.01 level ** = significant at the p=0.05 level * = significant at the p= 0.10 level

In table 6 the results for hypothesis 1 are described. Hypothesis 1 tests the relationship between conservative accounting on the one hand, and CFO gender and SOX on the other hand. The three different measures for conservatism, CON_SKEW, CON_ACCRUAL and CON_MTB have an explanatory power of respectively 37.92%, 8.84% and 26.67%. This means that the CON_SKEW measure is the most useful measure for accounting conservatism in this setting, whereas CON_ACCRUAL is the least useful.

For what concerns the relationship between conservatism and female CFO gender, none of the three conservatism measures yields a significant relationship. The relationship

(28)

28

between SOX and conservatism is significant and positive only for CON_ACCRUAL. The other two measures do not show a significant relationship. The results on the control variables are mixed. ROA is significantly positive (p=0.000) for CON_MTB where it is significantly negative for CON_SKEW (p=0.000). The control variable SIZE is not significantly related to either of the three conservatism measures. Leverage is, against my predictions, significantly negative related to CON_ACCRUAL (p=0.039) and CON_MTB(p=0.073). CASHHOLDING is negatively related to conservatism for CON_ACCRUAL (p=0.028) and positively related to conservatism for CON_MTB (p=0.000). dummyLITRISK is significantly negative related to conservatism for CON_MTB (p=0.010).

Table 7: Results for hypothesis 2

Predicted Sign

CON_SKEW CON_ACCRUAL CON_MTB R-squared:0.3792 R-squared:0.1433 R-squared:0.2695

Constant ? 0.5318 (p=0.252) -0.1528 (p=0.701) 1.3705*** (p=0.001) dummyGENDER + 0.0658 (p=0.919) -1.2615*** (p=0.006) 0.5889 (p=0.290) dummySOX + -0.1470 (p=0.428) 0.1923 (p=0.120) -0.1640** (p=0.309) dummyGENDER* dummySOX ? -0.0442 (p=0.954) 1.4353*** (p=0.006) -0.8316 (p=0.212) ROA + -12.9116*** (p=0.000) 0.4452 (p=0.755) 2.6461*** (p=0.000) SIZE - 0.0303 (p=0.558) 0.0359 (p=0.366) -0.0644 (p=0.157) LEVERAGE + -1.2873** (p=0.035) -0.9732** (p=0.037) -0.9550* (p=0.075) CASHHOLDING + -0.2729 (p=0.640) -1.0523** (p=0.027) 3.7873*** (p=0.000) dummyLITRISK + 0.2540 (p=0.192) 0.0624 (p=0.664) 0.4432** (p=0.009)

(29)

29

*** = significant at the p=0.01 level ** = significant at the p=0.05 level * = significant at the p=0.10 level

In the table it is shown that the three different variables each yield very different results. Also, the R-squared values differ significantly. It seems that for this research, CON_SKEW is the best measure for conservatism with an explanatory power of 37.30%, followed by CON_MTB with an explanatory power of 26.95%. CON_ACCRUAL has the least explanatory power as an indicator for conservatism, with 4.11%.

Furthermore it can be seen that each control variable is significant in one or two of the conservatism variables, indicating that the control variables explain a significant part of the variation in the model. Especially the return on assets seems to have a big influence on the conservative reporting methods, as the p-value for both CON_SKEW and CON_MTB is 0.000 and the coefficients are large (-12.9116 and 2.6461). Interestingly enough, when measuring ROA with CON_SKEW there is a positive relationship with conservatism and when using the CON_MTB measure there is a negative relationship with conservatism, which is the same result as I reported for H1 in table 6.

The debt-to-equity ratio also determines for a significant part the conservative reporting choices of a company. For both the CON_SKEW and the CON_MTB measure there is a significantly (p<0.05) negative relationship between LEVERAGE and conservatism. This is however not consistent with prior studies evidencing a positive relationship between

leverage and conservatism (Ahmed et al., 2002; Zhang, 2006). For CON_ACCRUAL there is also a positive relationship with LEVERAGE, however this relationship is not statistically significant.

Consistent with the litigation hypothesis, firms that operate in a high-litigation risk industry are expected to be positively associated with conservatism (Scott, 2011). I find that this is the case only for the measure CON_MTB. The other two measures are positive, however not significant.

5.1. The relationship between conservatism and SOX

Prior research has shown that there firms reported more conservatively in the years

following the introduction of SOX (Lobo and Zhou, 2006; Lobo and Zhou, 2010; Zhou, 2008). When I tested for the relationship between conservatism and SOX, I indeed found a positive

(30)

30

relationship for the CON_ACCRUAL measure (p=0.05).The coefficients on CON_SKEW and CON_MTB were both negative, however they were also insignificant. Assuming that CON_MTB and CON_SKEW are adequate measures for conservatism, as they are used in several studies, the insignificant results on those two measures might indicate that I have not included enough control variables for those two models.

5.2. The relationship between gender and conservatism (H1)

Hypothesis 1 states that: Firms with a female CFO will report more conservatively than firms with a male CFO. Following this hypothesis, I would expect a positive relationship between dummyGENDER and each of the three conservatism measures. As opposed to my

expectations, I found no significant evidence for a positive relationship between conservatism and female CFO gender. CON_SKEW and CON_MTB showed a positive

relationship and CON_ACCRUAL a negative relationship, however all of the three measures were insignificant.

Francis et al. (2014) did found a positive significant relationship between female CFO gender and conservatism for the CON_MTB and CON_ACCRUAL measure. Also they found an insignificant but positive relationship between CON_SKEW and female CFO gender. Their research had a different set-up: Francis et al. (2014) examined firms who changed their CFO from female to male and from male to female, thereby eliminating any firm-specific

conditions that may bias the results. An explanation for my insignificant results might be that I was not able to control for all firm-level characteristics influencing the relationship

between conservatism and gender.

As can be seen in table 2, only 5.46 percent of the firm-years in the sample consisted of firms with a female CFO. This might explain why the results were not significant and why the coefficient on dummy GENDER was not consistent with my expectations. When

conducting this experiment in a later time-period, the number of firms in the sample with a female CFO will probably increase. However, the intent of this research was to examine the pre- and post-SOX period. I could have used an extended post-SOX period, however then I would also have to use an extended pre-SOX period which would then lower the number of female executives in the sample. Also, the extension of the sample period could introduce more noise to my results, given the changing economic conditions over time.

(31)

31

CFO have different incentives than firms who hire a male CFO. However, until this moment there has been little research about the incentives of hiring a female or male CFO.

5.3. The influence of SOX on the relationship between gender and conservatism (H2)

Hypothesis 2 states that: The increase in conservatism following SOX will be different for firms with a male versus a female CFO. Hypothesis 2 builds further on hypothesis 1. I expected the coefficient on dummyGENDER to be positive, however I got insignificant results. When the interaction term dummyGENDERxdummySOX would be positive, this would indicate that SOX strengthens the supposed relationship between dummyGENDER and conservatism. When the interaction term dummyGENDERxdummySOX would be negative, this would indicate that SOX weakens the relationship between dummyGENDER and conservatism, which would mean that the difference in the level of conservatism between male and female CFOs would shrink as a consequence of SOX.

I found that the interaction-term is positive and significant for the CON_ACCRUAL measure (p=0.006). If I would have found a positive significant relationship between dummyGENDER and conservatism in the test of hypothesis 1, this would imply a strengthened relationship between female CFO gender and conservatism after SOX.

However I was not able to establish a significant relationship between dummyGENDER and conesrvatism. The other two measures, CON_SKEW and CON_MTB, were both negative and insignificant for the interaction-term. For a positive significant relationship between

dummyGENDER and conservatism, a negative interaction-term would indicate a weakened relationship between dummyGENDER and conservatism. This would imply that following the implementation of SOX, the increase in the level of conservatism would be greater for male CFOs than for female CFOs.

5.4. Concluding remarks

The area of gender differences in financial reporting is relatively under researched, which also means there is little evidence to build a research upon. I have created a difference-in-difference regression model to capture the influence of SOX on the relationship between gender and conservatism. Even though I carefully designed the model and retrieved the variables, I was not able to establish significant relationships and therefore I have to reject my hypotheses and accept the null hypothesis that there is no relationship between CFO

(32)

32

gender and conservatism. Given that I did not find significant results for my first hypothesis, I was not able to draw conclusions upon the second hypothesis, even though I found a

significant result for the CON_ACCRUAL measure.

Given that Francis et al. (2014) were able to establish a significant positive

relationship between gender and conservatism, it could be that there is a type II error in my research. It could also be that, in the examined time period, there is no significant

relationship between gender and conservatism. Francis et al. (2014) conducted their research in a larger time period (1988-2007).

I measured conservatism in three ways, which all did not yield significant results. The conservatism indicators CON_SKEW, CON_ACCRUAL and CON_MTB had R-squared indicators of respectively 0.3792, 0.0884 and 0.2667. This means that CON_SKEW was the best

indicator for conservatism in this setting, followed by CON_MTB and CON_ACCRUAL. I found that the control variables were in most cases significant, which means they explain an important part of the relationship. It might well be that firms with a female CFO do not per se report more or less conservatively.

Another explanation why I could not establish a significant relationship could be that firms who hire female CFOs have certain incentives that I was not able to control for.

Furthermore the amount of firm-years with a female CFO was significantly lower than the amount of firm-years with a male CFO. This could have influenced the results, however it most likely represents the real situation at that moment and should therefore be a representative sample.

To the best of my knowledge, my study is one of the first to examine the effect of (a certain) regulation on the amount of accounting conservatism executed within a firm. While I was not able to draw a conclusion upon the effect of SOX on the relationship between CFO gender and accounting conservatism, this does not mean that SOX (or other accounting regulations) have no effect on this relationship.

A suggestion for further research could be to conduct an experiment likewise to Francis et al. (2014) in the years prior to and after SOX, where male-to-female CFO transitions are examined. This research method controls for the bias associated with firm-specific characteristics. One disadvantage to their research method is that they examined firms with a male-to-female CFO or female-to-male CFO transition, and in the time-period 2000-2004 there are are probably little of such transitions. This problem can be overcome by

(33)

33

extending the time-period.

6. Conclusion

The increase in the number of female executives since the beginning of the twenty-first century has opened up new research possibilities. In the field of psychology and sociology, there already is a vast body of research on the differences in decision-making between men and women, however in the accounting literature this phenomena is relatively under-researched. This paper is an attempt to fill in that gap.

Even though there are different findings with respect to differences in the decision-making process between men and women, the general assumption is that men are more prone to risk than women. In financial reporting, this could imply that female executives make more conservative reporting choices, thereby exposing their company to less risk. Francis et al. (2014) indeed found that companies who switched from a male to a female CFO make more conservative reporting choices.

Another factor that more certainly led to an increase in conservatism was the introduction of SOX. By exposing company executives to more rules and regulations – including severe punishments – SOX was intended to enhance the quality of financial reporting and restore investor confidence. Several studies found that the introduction of SOX indeed led to an increase in conservatism (Lobo and Zhou, 2006; Lobo and Zhou, 2010; Zhou, 2008).

In this study I examined whether the introduction of SOX has moderated the expected positive relationship between female CFO gender and conservative financial

reporting. Because of the competing theories, I had the non-directional expectation that SOX moderated the relationship between gender and conservative reporting. On the one hand there was evidence that SOX strengthened the relationship between gender and

conservatism (Francis et al., 2014). However given that female CFOs already reported more conservatively than male CFOs, and given the overall increase in conservatism following SOX, it could also be that the increase in conservatism would be greater for male CFOs, meaning that SOX would weaken the relationship between gender and conservatism.

To test the assumptions I created a regression model including an interaction-term between SOX and GENDER. Because there is no single accepted measure of conservatism, I use three different measures of conservatism: skewness of earnings, discretionary accruals

(34)

34

and the book-to-market ratio. Furthermore I controlled for firm-level characteristics influencing the accounting conservatism.

I found no significant positive relationship between GENDER and conservatism. Thus, I cannot accept hypothesis 1 and therefore have to accept the null hypothesis; there is no relationship between CFO gender and conservatism. Given that Francis et al. (2014) do find a significant positive relationship between female CFO gender and conservatism, this might imply that there is a Type II error in my research. I attribute the insignificance of my results to two different factors. First, only 5.46 percent of the sample consisted of firm-years with a female CFO. Second, it could be that firms who hire female CFOs have certain characteristics that are associated with less conservative financial reporting. Francis et al. (2014) control for those firm specific characteristics by examining male-to-female and female-to-male

transitions within a company, thereby eliminating the firm-specific factors that might bias the relationship between conservatism and gender. It would have been hard to conduct a likewise experiment in my research, because there are so little firms who have a male-to-female transition in the pre-SOX period.

Given that I could not accept hypothesis 1, I can also not conclude that SOX moderates the relationship between CFO gender and conservatism. Thus I cannot accept hypothesis 2. I expected a positive relationship between conservatism and CFO gender and a negative or positive relationship between conservatism and the interaction-term on CFO gender and SOX (dummyGENDERxdummySOX). This would have indicated that SOX

respectively weakened or strengthened the relationship between gender and conservatism. I did found a significant positive relationship for the CON_ACCRUAL measure. However I was not able to conclude that SOX strengthens the relationship between CFO gender and conservatism, given the insignificant results for hypothesis 1.

In this study I extended existing research on the relationship between gender and conservatism. Although I could not establish a significantly positive relationship between gender and conservatism, I did offer some alternative explanations for the insignificance of the relationship. First, there is only a small number of firms with a female executive, this might have influenced the results. Second, it could be that firms who hire a female executive have certain characteristics or incentives for which I was not able to control in my research model.