Influence of non-GAAP earnings on the cost of capital

(1)

Influence of non-GAAP earnings on the cost of capital

Name: Anne Koppelaar Studentnumber: 10750606

Date: 24-6-2018 Word count: 12,679

Thesis supervisor: Dr. Réka Felleg

MSc Accountancy & Control, specialization Accountancy Amsterdam Business School

(2)

2

Statement of originality

This document is written by student Anne Koppelaar who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

I examine the relation between the quality of non-GAAP disclosures after the Compliance and Disclosure (CDI) update in 2010, and its influence on the cost of capital. The update

introduced more flexibility and a more accessible use of non-GAAP earnings. The earlier regulation, Regulation G resulted in a decreased use of non-GAAP earnings, whereas this update resulted in an increased use of non-GAAP earnings by firms in all sectors. Regulators’ debate from the early 2000s on whether non-GAAP earnings are used to inform or mislead, and this new update is important. I use a recent dataset obtained from U.S. companies from 2003 to 2015, to test my hypotheses. I provide new evidence about the influence of non-GAAP earnings on the cost of capital, by examining the effect of the CDI update. My results show that the quality of non-GAAP earnings decreased after the CDI update, and the use of non-GAAP earnings results in a lower cost of capital.

Keywords: Non-GAAP earnings, Cost of capital, Information asymmetry, Compliance and Disclosure interpretations, Regulation G

(4)

4

1. Introduction

Non-GAAP reporting is widespread. Whereas in 2002 only a few companies provided non-GAAP earnings, recently this type of reporting increased from 53% in 2009 to 71% in 2014 of firms in the S&P 500 (Black, Christensen, Ciesielski, & Whipple, 2017a). As with every form of voluntary disclosure, there are concerns about the quality of non-GAAP earnings. Regulators repeatedly express concerns about whether non-GAAP earnings are used to inform or mislead, while managers argue that non-GAAP numbers better reflect ‘core’ operations than GAAP numbers do (Leung & Veenman, 2016).

Non-GAAP numbers are a subject of interest for regulators, because the numbers are unaudited, and the performance metrics are nonstandard. Non-GAAP numbers, therefore, allow managers discretion in the measurement of these numbers. The Sarbanes-Oxley Act of 2002, appealed for more regulation of non-GAAP reporting (Heflin & Hsu, 2008). The SEC responded to that request and introduced Regulation G, to provide requirements regarding the measurement of non-GAAP numbers and to make non-GAAP numbers less misleading than before.

Cormier, Demaria and Magnan (2017) confirm the effectiveness of Regulation G and state that disclosing EBITDA, a form of non-GAAP reporting, leads to less information asymmetry between managers and investors, when the market for information is not

transparent. Complementing this, Curtis, McVay and Whipple (2013) show that non-GAAP earnings are informative for investors, especially when GAAP earnings are less informative. This means that non-GAAP earnings can be used to clarify or supplement information of GAAP earnings, if GAAP earnings fall short.

However, Bhattacharya, Black, Christensen and Larsson (2003) find that non-GAAP exclusions are sometimes used to meet earnings benchmarks. This is possibly due to the discretion afforded in the measurement of non-GAAP earnings. The discretion in the measurement might lead to an earnings metric that lacks important characteristics that are required under GAAP earnings (Black et al., 2017a). As a result, managers adjust their

numbers based on what they think presents better performance, which can result in misleading information.

In addition, Black, Christensen, Taylor Joo and Schmardebeck (2017c) state that the managerial discretion in non-GAAP reporting is sometimes used to meet or beat strategic earnings targets, which is why they indicate that non-GAAP earnings can be a substitute for earnings management. Black et al. (2017c) state that managers first use real-, or accrual

(6)

6

earnings management or classification shifting, when GAAP earnings do not meet expectations. If these types of earnings management are still not sufficient to meet the expectation, managers are most likely to turn to non-GAAP reporting (Black et al., 2017c).

Merton (1987) argues that investors demand an information risk premium when managers have more information than investors. This results in the fact that firms can reduce their required return, by reducing information risk through increased voluntary disclosure. Graham, Harvey and Rajgopal (2005) complement this and state that disclosing reliable and precise information can in theory reduce ‘information risk’ about a company’s stock, which leads to a lower cost of capital. Graham et al. (2005) determine whether the cost of capital of reduction of information risk is a motivation for providing voluntary disclosure, by

interviewing executives. More than four-in-five respondents agree with this motivation. However, Gietzmann and Ireland (2005) do not find the relation between voluntary disclosure and information asymmetry, possibly due to the quality of the underlying disclosures. This implies that prior research does not always find a relationship between voluntary disclosure and information asymmetry.

The reaction of investors and financial analysts are important, as they play a giant role in determining the cost of capital. Cormier et al. (2017) state that with GAAP, the emphasis lies on producing a single number, which is likely to be useful for the firm, investors and creditors. However, in addition to net income, sophisticated investors need information about earnings components and they want to distinguish operating items from non-operating items. These investors also want to know which items are real and which only reflect the application of accounting rules. Sophisticated investors believe that providing some sort of non-GAAP number results in less information asymmetry (Cormier et al., 2017).

As stated above, the first regulation towards non-GAAP reporting was Regulation G in 2002. Regulation G provides requirements about non-GAAP earnings, and it states that disclosures with a non-GAAP earnings number must contain a directly comparable GAAP number (Heflin & Hsu, 2008). After this regulation, the use of non-GAAP earnings

substantially decreased. However, 11 January 2010, the SEC issued Compliance and

Disclosure Interpretations (CDIs) on the topic non-GAAP earnings, and again in May 2016. The CDI update provided more flexibility towards non-GAAP earnings, which resulted in an increased disclosure of non-GAAP earnings. The quality of the underlying disclosure needs to be re-examined, because after the CDI update disclosing non-GAAP earnings are more

(7)

7

Prior literature examined the practice of non-GAAP earnings and found that non-GAAP earnings influence judgments about firm performance for investors and financial analysts (e.g. Black et al., 2017a; Andersson & Hellman, 2007). Cuijpers and Buijink (2005) examined the influence of non-GAAP on cost of capital in Europe. However, in Europe there is no

regulation and European companies are not subject to the CDI update.

My thesis examines if the quality of non-GAAP earnings influences the cost of capital after the Compliance and Disclosure Interpretations (CDI) update in 2010, because of the ongoing debate about whether non-GAAP earnings are used to inform or mislead, and the fact that there is a differential effect of disclosure on the cost of capital.

The hypotheses of my paper are being investigated in an archival study. I expect that the quality of non-GAAP earnings decreased after the CDI update. Furthermore, I expect that non-GAAP quality leads to a reduction in the cost of capital. The sample consists of 1,192 firms and 3,170 yearly observations, during a time period from 2003 to 2015.The first hypothesis is tested with a logistic regression and the second one with an ordinary least squares (OLS) regression. The findings of my study show that non-GAAP earnings are less informative after the CDI update of 2010 and that there is a negative influence of non-GAAP earnings on the cost of capital, possibly due to the substantially increased use of non-GAAP earnings by firms in all sectors, which is a consequence from from the increased flexibility in non-GAAP reporting. Therefore, both of my hypotheses are supported.

This paper contributes to the existing literature by examining the effect of the CDI update on the quality of non-GAAP earnings and the influence it has on the cost of capital. This research is distinct from prior studies, because the effect of the CDI update is included, which resulted in a substantial increase in non-GAAP earnings. Regulation G resulted in a decrease in the use of non-GAAP earnings. Prior research mostly examined the effect of Regulation G, and not the CDI update. As stated above, the quality of non-GAAP earnings needs to be re-examined.

There are several limitations to my research. Because I require a dataset with four quarter observations for every year, this leads to a relatively small sample. This can have an impact on the results. When a firm only provides a non-GAAP number at the end of fiscal quarter four, it can be more misleading than a firm that provides a non-GAAP number each quarter. At the end of the fourth quarter, the books need to be closed and managers can have

incentives to meet or beat analyst expectations. Results based on a sample that includes these quarters could show a larger reduction of non-GAAP quality after the CDI update.

(8)

8

Another limitation is that I only used firms that produced a non-GAAP number. This can possibly have an impact on the results. The quality of GAAP earnings might be higher if firms who only provide GAAP numbers are included. This may lead to a lower quality of non-GAAP earnings than in my study.

The last limitation is the GAAP dataset itself, because it only provides the non-GAAP number, and it was not divided into non-recurring and recurring exclusions. If

recurring exclusions and non-recurring were divided, this can possibly have an impact on the quality of non-GAAP earnings. It can lead to a decrease in the quality of non-GAAP earnings, because recurring exclusions are interpreted as low quality exclusions, and therefore it can possibly effect my results.

This thesis proceeds as follows. In the next section, there is a literature review about non-GAAP disclosures, the cost of capital and the possible effects these two have on each other. After that, the hypothesis development is explained. This is followed by the research design, sample selection, and important variables. Section four provides the results of this research. The thesis ends with a conclusion and possible limitations of the study and suggestions for future research. The tables can be found in the appendices.

(9)

9

2. Literature review

2.1 Non-GAAP earnings

2.1.1 Use of non-GAAP earnings

Companies are required to disclose annual reports in line with Generally Accepted Accounting Principles (GAAP), because they provide financial statement users with a standardized performance metric that is comparable between firms. Firms need to measure their earnings number based on this standard. But, in addition to GAAP earnings, companies are allowed to include voluntary disclosures of other earnings figures in their financial statements. Firms usually add this as a supplement to their GAAP numbers, where they exclude some earnings components. Firms can provide this because GAAP earnings may include items that do not relate to core operations or are recurring, and therefore non-GAAP earnings give a clearer understanding of earnings (Whipple, 2015). However, firms can also use it as an earnings management tool, therefore they might use it to appear to meet expectations on a non-GAAP basis when managed GAAP earnings fall short (Black et al., 2017c). That is why companies can use non-GAAP earnings either to inform or mislead stakeholders (Marques, 2017).

Non-GAAP earnings1 disclosures are generated by excluding certain items which managers believe are not relevant for determining the financial performance of the firm (Entwistle, Feltham, & Mbagwu, 2006). The literature partitions the difference between GAAP and non-GAAP earnings into two categories. The first category is labeled as special item exclusions. Special item exclusions represent one-time items that are excluded in the calculation of non-GAAP earnings by managers and analysts (Whipple, 2015). The most common non-recurring exclusions relate to restructuring, tax resolution and acquisition exclusions. Black, Christensen, Ciesielski and Whipple (2017b) state that nonrecurring exclusions have nearly doubled in size over their sample period, from $0.73 in 2009 to $1.03 in 2014, which means that in 2014, nonrecurring exclusions occur frequently. These expenses are predominant in multiple sectors (Black et al., 2017c).

The second category is the remaining part of non-GAAP adjustments, called other exclusions. These exclusions are generally recurring transactions, for example stock-based compensations. Other exclusions are the most common non-GAAP adjustments, as 78% of

1_{Non-GAAP earnings are mostly referred to as non-GAAP earnings, pro forma earnings, or adjusted earnings.}

(10)

10

non-GAAP reporting firms exclude other items (Whipple, 2015). The reason that managers exclude this type of exclusions is that they claim these exclusions are operating’ or ‘non-cash’. Whipple (2015) find that 72% of other exclusions relate to three types of transactions: one-time items (29%), stock-based compensation (22%), and amortization (21%). Other exclusions are comprised of both transitory and recurring items (Whipple, 2015). Marques (2017) reports that especially recurring items are seen as an indication of opportunistic use of non-GAAP earnings. Leung and Veenman (2016) state that these recurring expenses are generally viewed to be more difficult to justify than transitory items, as the motivation for excluding these items is less clear. That is why recurring expenses are interpreted as ‘low quality’ exclusions and indicate aggressive non-GAAP reporting. Barth, Gow and Taylor (2012) suggest that the exclusion of stock based compensation, another widely used exclusion, is not informative because it does not predict future GAAP earnings.

Black et al. (2017a) find that non-GAAP reporting frequency has increased across all industries. This means that non-GAAP reporting is used by all industries. They find that with an average of 3.6 items in 2014 compared to 3.1 in 2009, firms increase the number of

excluded items (Black et al., 2017a). It furthermore appears that excluding recurring items has become more common in 2014, compared to 2009 (Black et al., 2017b).

Leung and Veenman (2016) find that non-GAAP earnings are informative, because they better reflect what companies see as ‘core’ earnings. According to Black et al. (2017b), where some disclosure models focus on a one-period setting, non-GAAP reporting is part of a multi-period disclosure policy. This means that users of financial statements learn from prior actions, indicating that managers who mislead stakeholders with aggressive non-GAAP earnings disclosures are likely harming the firm’s reputation in the long run, which is why they state that non-GAAP disclosure is motivated by a desire to inform investors (Black et al., 2017b).

In addition, Kolev, Marquardt and McVay (2008) state that GAAP earnings have become a noisier measure of true economic income. That is why investors, analysts and managers have adjusted their focus to alternative earnings-performance measures, such as

non-GAAP earnings. However, non-GAAP earnings are unaudited and calculated by what

managers think is the appropriate manner. This managerial discretion potentially causes the concerns that non-GAAP earnings are disclosed opportunistically (Black et al., 2017a).

According to Black et al. (2017c), compared to accruals earnings management and real earnings management, non-GAAP is a relatively costless form of perception

(11)

11

management. Unlike accruals, which reverse in the next period, or real earnings management, which can be extremely costly. Non-GAAP do not attract the attention of auditors, which is why Black et al. (2017c) state that non-GAAP can be seen as a substitute for earnings management, considering the managerial discretion provided. However, it appears that managers generally prefer to meet expectations based on real or accruals management, before employing non-GAAP exclusions (Black et al., 2017c).

2.1.2 Regulation G and S-K

Standard setters and regulators have been worried since the early 2000’s about whether non-GAAP earnings are used to inform or to mislead. These concerns have led to regulation towards non-GAAP reporting from the Securities and Exchange Commission (SEC). The US was the first country that regulated the voluntary disclosure of non-GAAP earnings. In the 1990s, non-GAAP earnings measures became prevalent (Kolev et al., 2008). There is

evidence that managers use these disclosures opportunistically (Black et al., 2017c). This has led to concerns by regulators and standard setters, arguing that these numbers might mislead users of financial statements.

In 2001, the SEC issued a cautionary warning about non-GAAP earnings, in which they noted that disclosure of non-GAAP earnings may mislead investors, because non-GAAP earnings carry no defined meaning and no uniform characteristics (SEC, 2001). Entwistle et al. (2006) state that the U.S. Congress enacted a new regulation in 2002, to restore confidence in the stock markets and improve the reliability of corporate disclosures. This regulation is known as the Sarbanes-Oxley Act, and Section 401(b) is devoted to the use of non-GAAP earnings. The SOX act directed the SEC to issue rules about non-GAAP earnings, because there was a concern that non-GAAP earnings could mislead investors. In January 2003, the SEC responded to this by issuing regulation G and amending Item 10(e) of Regulation S-K (Webber, Nichols, Street, & Cereola, 2013). Regulation G requires that disclosures with a non-GAAP earnings number must contain the most directly comparable GAAP number and may not present non-GAAP earnings in a way that misleads investors. Management also needs to explain why they believe the non-GAAP measure provides investors useful

information (Heflin & Hsu, 2008). The new rule also includes amendments to Item 10(e) of Regulation S-K requires additional disclosures, therefore it has stricter rules than Regulation G. Unlike Regulation G, Regulation S-K directly prohibits non-GAAP financial measures that exclude ‘non-recurring’ items, if the firm reports or is likely to report the same or similar items in the previous or following two years (Heflin & Hsu, 2008).

(12)

12

The new regulation suggests that the SEC believes that more consistency and

transparency in non-GAAP disclosure provides investors with more understanding about the non-GAAP earnings information (Marques, 2006). Kolev et al. (2008) confirm this and find that the SEC intervention had the desired effect of mitigating the opportunistic use of non-GAAP earnings, and that firms who disclosed lower quality exclusions before Regulation G stopped releasing non-GAAP earnings afterwards. Kolev et al. (2008) find that the quality of non-GAAP disclosures increased significantly after Regulation G. Most of the firms with the lowest quality exclusions in the pre-Regulation period did not report non-GAAP earnings anymore after the regulation. While exclusions are still not perfectly transitory in the post-regulation period, SEC intervention seems to have the effect that they desired, in mitigating opportunistic behavior of non-GAAP earnings.

Furthermore, the research of Kolev et al. (2008) suggests that the increased costs of non-GAAP disclosure, as a result of the SEC intervention, discouraged at least some

opportunistically motivated firms from continuing this practice. But when total exclusions are decomposed into special items and other exclusions, it appears that the quality of special items has decreased after the SEC regulation (Kolev et al., 2008). This suggests that managers shifted recurring expenses into special items.

However, prior literature disagrees about the effectiveness of Regulation G. Black et al. (2017a) state that there is still evidence of aggressive non-GAAP reporting after the Regulation G period. They also find evidence suggesting a substitute relation between aggressive non-GAAP reporting and earnings management in the Post-Regulation G period. Curtis et al. (2014) also state that managers are more willing to report non-GAAP earnings that exclude one-time losses relative to one-time gains. Despite above research, it seems that the quality of non-GAAP earnings increased due to Regulation G.

2.1.3 Compliance and Disclosure interpretations

After the introduction of Regulation G in 2003, the regulation of non-GAAP earnings has remained fairly the same, until 2010. In January 2010, the SEC updated its Compliance and Disclosure Interpretations (CDI) related to non-GAAP earnings. The CDI update is a response to the ‘two-tiered reporting system’ of Regulation G and Item 10(e) of Regulation S-K, where a company discloses certain non-GAAP earnings in its press release and then discloses

different financial measures in its SEC filings, because Item 10(e) of Regulation S-K had stricter restrictions than Regulation G (Webber et al., 2013).

(13)

13

Webber et al. (2013) state that the revised CDI signaled substantially increased flexibility regarding the inclusion of non-GAAP financial measures, in relation to the earlier regulation. The SEC’s update suggests that management should be able to provide investors with ‘appropriate’ non-GAAP financial performance measures. Measures are appropriate as long as the measures are explained by adequate disclosures. If the firm cannot indicate a certain gain as being non-recurring, infrequent or unusual, it does not mean that the gain cannot be adjusted. In Item 10(e) of Regulation S-K, this kind of exclusions were prohibited (Webber et al., 2013).

After the CDI update, more companies started disclosing non-GAAP earnings, and the frequency increased as well. Webber et al. (2013) note that the increase can also be due to several other factors, such as providing better information to their investors, and more accessibility towards presenting non-GAAP measures.

In 2011, Howard Scheck, SEC’s former chief accountant, labeled non-GAAP

performance measures as a ‘fraud risk factor’. This resulted in a taskforce formed by the SEC to closely examine potentially misleading non-GAAP earnings (Black et al., 2017a). In 2016, this led to an update in the Compliance and Disclosure Interpretations yet again, where the SEC encouraged companies to ‘self-correct’ their reporting practices that are not in

compliance with the regulation, because they classify firms’ non-GAAP reporting as

misleading, if they are inconsistently measured across quarters (Black et al., 2017a). Overall, the update in the CDI of 2010 has initiated more flexibility towards non-GAAP reporting.

Since the update of the CDIs in 2010, non-GAAP reporting has increased and is now at an all-time high point. Non-GAAP reporting has increased flexibility since the CDI update (Webber et al., 2013), and the effects have not been examined yet.

2.2 Cost of capital

Cost of capital is the minimum rate of return that investors require in return for the provision of capital to organizations, being the sum of the risk-free return and a risk premium (Botosan, 2006). Whether more information in financial reporting affects the cost of capital, is an important question. The quality of underlying information also needs to be taken into consideration. Cost of capital affects corporate decisions, like investment and acquisition decisions. It is favorable for companies to have a low cost of capital, because it results in cheaper financing. Information asymmetry and non-diversifiable estimation risk both

(14)

14

future returns, or lower the information asymmetries between managers and outside investors, resulting in increased liquidity and lower required rates of return (Daske, 2006).

2.2.1 Information asymmetry

According to Healy and Palepu (2001), the agency problem plays a big part in information asymmetry. The agency problem arises because the shareholders, better known as the

principal, who invest in a business, do not play a part in daily management. The management is therefore a different party, which is better known as the agent. This can result in the fact that principal and agent have different incentives. For example, a self-interested agent might make decisions that expropriate funds of the principal.

Daske (2006) states that not every company is transparent in disclosing information, because making all information public is costly in the capital market. If the benefits from publishing information outweigh the costs, a firm is likely to choose to disclose this

information. Information asymmetry among different groups of investors introduce adverse selection into stock transactions and reduce market liquidity. Firms must issue their shares at a discount and face a higher cost of capital (Daske, 2006), because of this. If a firm does not want that to occur, it needs to increase disclosure. Companies are turning private information into public information, which in theory leads to lower information asymmetry, increased liquidity and a lower required return, and therefore a lower cost of capital. Graham et al. (2005) confirm this by stating that additional disclosures can reduce ‘information risk’ and this reduces the required return. Lambert, Leuz and Verrechia (2007, p.385) complement this by quoting Neel Foster, a former member of the FASB, who claims that “More information always equates in less uncertainty, and people pay more for certainty. Better disclosure results in a lower cost of capital”.

Graham et al. (2005) organized a survey around CFOs about their choices related to accounting numbers and voluntary disclosures and found that, four-in-five respondents agree or strongly agree with the motivation to voluntarily disclose information because it reduces information risk. But this is not evidence that the relationship is truly there, between voluntary disclosures and lower information risk. For example, Gietzmann and Ireland (2005) find an effect of disclosure on cost of capital for those firms that make aggressive accounting choices, but they do not find a relationship between disclosure and cost of capital for conservative firms. Concluding this, the relationship between voluntary disclosures and information asymmetry is not clear.

(15)

15

2.2.2 Non-diversifiable estimation risk

Daske (2006) states that higher information quality reduces a firm’s cost of capital, because it reduces non-diversifiable estimation risk. This results from the fact that investors have to estimate the parameters of a security’s return from available information, and thus they face estimation risk. Therefore, investors demand an uncertainty premium in their rate of return. Additional information can reduce estimation risk and this is rewarded by risk-averse investors, by demanding a lower risk premium. This is relevant, because when more

disclosure is related to a lower cost of capital, firms are more keen to increase disclosure. An example of such a disclosure may be non-GAAP measures.

2.2.3 Quality versus quantity of disclosure

Suggesting that more disclosure leads to less information asymmetry is not self-evident, because the underlying quality of the disclosures may be even more important. As stated above, Neel Foster states that more information always leads to less uncertainty (Lambert et al., 2007). However, as can be seen from the agency theory, this is not always right. Managers have gained discretion in voluntary disclosure, because they decide what information is published. And if the agent uses this discretion for his own benefit, the disclosure may lead to more information asymmetry (Francis, Nanda, & Olsson, 2007), not less, which is what is normally assumed with more disclosure. In theory, more disclosure always leads to less information asymmetry (Lambert et al., 2007). However, investors demand a risk premium when they see that disclosures are not informative.

However, companies who choose to disclose voluntarily face additional costs and risks, because they make private information public (Black et al., 2017a). Black et al. (2017c) suggest that non-GAAP is a relatively costless form of earnings management, which means that the quality of the underlying disclosures is doubtful. Huang and Skantz (2016) however, find a greater reduction in information asymmetry after earnings announcements, when firms disclose non-GAAP earnings in addition to GAAP earnings.

2.3 Investors influenced by non-GAAP earnings

It is important who is influenced by non-GAAP earnings, as there needs to be a motive why so many companies are using non-GAAP earnings. Marques (2017) states that investors and financial analysts respond to non-GAAP information. In an earlier paper, Marques (2010) states that when the focus in press releases lays more on non-GAAP values than GAAP values, non-professional investors’ reliance on these values increases, and the focus tends to

(16)

16

be on non-GAAP values instead of GAAP values. This creates incentives in the way how managers use non-GAAP reporting, when managers want to influence investors about the performance and valuation of the firm. However, Marques (2010) furthermore states that this influence is mitigated by the disclosure of a side-by-side reconciliation.

In these reconciliations of non-GAAP figures, the value of the adjustments made by managers and the GAAP figures are compared. Black et al. (2017a) also find that the existence of non-GAAP earnings affects the judgment of less sophisticated investors. But experimental research indicates that, although less susceptible, more sophisticated parties can be influenced by non-GAAP reporting as well. Analysts’ who revise forecast surrounding non-GAAP disclosures are generally viewed as sophisticated financial statement users (Black et al., 2017a). Andersson and Hellman (2007) find that non-GAAP earnings can influence analysts’ earnings per share forecasts. However, Johnson and Schwartz (2005) find that investors are not misled by the disclosure of non-GAAP earnings, because the possibility of misleading is mitigated by the required reconciliation of the adjustment to a direct GAAP number. Most researchers find that investors are influenced by non-GAAP earnings.

2.4 Hypotheses development

Most studies show that more voluntary disclosure leads to a decrease in information asymmetry, and therefore reduces the cost of capital (Francis et al., 2007). However, the empirical relation is not clear, as not every researcher finds that relation (Gietzmann & Ireland, 2005). The outcome of my research is important, as Marques (2017) states that investors and financial analysts respond to non-GAAP information. Black et al. (2017a) complement this by stating that less sophisticated investors are affected by non-GAAP, and, although less susceptible, sophisticated parties as well. Johnson and Swartz (2005) find that investors are not misled, because of the reconciliation to a direct GAAP number. The influence of non-GAAP on investors is thus not clear. The literature mostly states that investors do react on non-GAAP information, which is why I expect that non-GAAP influences the cost of capital.

According to Marques (2006), non-GAAP disclosures are less misleading after

Regulation G. She furthermore finds that non-GAAP disclosures provide extra information in addition to GAAP earnings. The quality of non-GAAP earnings after Regulation G is high, and therefore information asymmetry decreases, following Graham et al. (2005). Non-GAAP earnings, in the period prior to the CDI update, provide additional information, and thus reduce information asymmetry. However, the CDI update from the SEC in 2010 signaled

(17)

17

increased flexibility regarding non-GAAP financial measures (Webber et al., 2013). Following the CDI update, management needs to provide investors with ‘appropriate’ non-GAAP measures, but as long as non-non-GAAP measures are accompanied by an adequate disclosure explaining the adjustment, it is appropriate (Webber et al., 2013). This made the use of non-GAAP earnings more accessible for companies. The CDI update resulted in a substantial increase in the use of non-GAAP earnings by firms in all sectors. This increased use and flexibility can have an impact on the quality of non-GAAP earnings, taking into account managers’ incentives to provide non-GAAP numbers. Based on the above, I argue that non-GAAP reporting has gained more flexibility from the CDI update and resulted in more managerial discretion in providing non-GAAP numbers. Therefore, I expect that the CDI update results in a decrease in the quality of non-GAAP earnings. Therefore, my first hypothesis is stated as follows:

H1: Non-GAAP earnings are less informative after the CDI update of 2010.

Informative non-GAAP earnings reduce information asymmetry, when a market is not transparent in providing information. If non-GAAP earnings are informative, I expect that non-GAAP earnings are negatively related to the cost of capital. Non-GAAP earnings can reduce information asymmetry, which has an impact on the cost of capital, and therefore lower the required return. That is why I expect that informative disclosure of non-GAAP earnings results in a lower cost of capital. However, if non-GAAP earnings are misleading, and investors are tricked by the disclosures, the cost of capital also reduces. If investors are not tricked by misleading non-GAAP earnings, non-GAAP earnings either have no influence on the cost of capital, or investors punish firms by demanding a higher required return, and a higher cost of capital. Therefore, I expect the cost of capital to increase after the CDI update, when the quality of the underlying disclosures decreases. To sum up, I expect that the relation between non-GAAP earnings and the cost of capital is negative after the CDI update.

Therefore, my second hypothesis is stated as follows:

H2: The use of non-GAAP earnings is negatively related to the cost of capital after the CDI update of 2010.

(18)

18

If both H1&H2 hold, this means that shareholders are misled by non-GAAP disclosures, because although the quality is not good, the cost of capital decreases. If H1 does not hold and H2 does, this means that non-GAAP earnings are informative and that investors value the information provided in non-GAAP earnings. However, if H1 holds and H2 does not, investors see through, and do not reward the information provided in non-GAAP earnings.

(19)

19

3. Methodology

3.1 Sample selection

The dataset that I use in this thesis contains U.S. companies. In the U.S. non-GAAP earnings are regulated since the early 2000’s by Regulation G, and since 2010 the regulation is updated in the CDIs. Black et al. (2017a) state that the use of non-GAAP earnings has increased from 53% in 2009 to 71% in 2014, and the amount of exclusions also increased. The change in regulation in 2010 from the CDI update can be a cause of this, which is why I am examining what effect the CDI update has on GAAP earnings, and what the relation is with non-GAAP earnings and the cost of capital.

The non-GAAP dataset that I use is from Bentley, Christensen, Gee and Whipple (2016) and covers the period from 2003 to 2016. In 2003, Regulation G was initiated, and therefore 2003 is my starting point. Following Leung and Veenman (2016), SIC codes 6000-6999 are excluded for financial firms, because they have different regulations than those of nonfinancial firms (Marques, 2006). I gather the data for GAAP earnings from

CRSP/Compustat, through the Wharton Research Data Services (WRDS), using the variable earnings per diluted share (“EPSFX”). I only keep the observations that have both non-GAAP earnings and GAAP earnings. This results in a large decrease in the sample size.

The cost of capital is determined yearly, which is why I remove observations where non-GAAP information is not available for four quarters per year, to match my non-GAAP dataset with the cost of capital. The year 2016 is dropped, as no observations were available for this year. The data for the cost of capital calculation is gathered through the Thomson Reuters Institutional Brokers’ Estimate System, often abbreviated as I/B/E/S, where I use EPS forecasts for 2 years ahead. For the share price, which is part of the cost of capital calculation, I use the CRSP/Compustat database, with the variable share price (Prcc). I add up all four quarter observations per year, resulting in a final sample of 1,192 firms, and 3,170 yearly observations, during a time period of 2003 to 2015. This sample is relatively small, comparing to Whipple (2015), who had a sample of 48,101 firm quarter observations. As stated above, this is due to the fact that I lost many observations, where there were no four quarter observations per year available. Table 1 summarizes the sample selection.

(20)

20

3.2 Research design

The sample period is 2003-2015, as fiscal year 2015 is the last available period in the dataset, in which there are four quarterly observations per year available. The period 2003-2009, and 2010-2015 are compared in my hypotheses. All continuous variables are winsorized at 1st and 99th percentiles, therefore large outliers are eliminated.

3.2.1 H1: Quality of non-GAAP disclosures

In my first hypothesis, I examine the quality of non-GAAP earnings. To determine what the quality of non-GAAP earnings is, I follow the research of Leung and Veenman (2016) who classify non-GAAP earnings as informative if they are more positively related to future earnings than GAAP earnings only. I follow Leung and Veenman (2016) in predicting future GAAP earnings. However, as they used quarterly observations, I use yearly observations. This does not result in a change in the model. I estimate ordinary least squares (OLS) models: Future GAAP earningst+1 = β0 + β1 GAAP earningst + Control variables+ ԑ (1)

Non-GAAP earnings are split in the second equation into non-GAAP earnings and exclusions, as follows:

Future GAAP earningst+1 = β0 + β1 Non-GAAP earningst + β2 Exclusions + Control

variables + ԑ (2)

Where:

Future GAAP earningst+1 The outcome variable. This variable is used to predict

future GAAP earningst+1, for both equations.

GAAP earnings The GAAP earnings variable using EPSFX in the CRSP/Compustat database.

Non-GAAP earnings The yearly non-GAAP earnings, added up all four quarter observations through the database of Bentley et al. (2016) Exclusions The remainder of the non-GAAP earnings minus

GAAP earnings, to measure the quantity of the exclusions in the non-GAAP number.

I use the regressions to estimate future GAAP earningst+1. The predicted variables are defined

as GAAP predictions and non-GAAP predictions. The one which has the best predictive ability, is the most informative one. I measure the predictive ability by first determining the

(21)

21

difference between GAAP earnings per share one year ahead and GAAP predictions, and the same for GAAP earnings per share one year ahead and non-GAAP predictions. I generate the variable final differences such that the value is negative when non-GAAP predicts better and positive when GAAP predicts better. Finally, I create an indicator variable, which has a value of 0 if GAAP has better predictive ability and a value of 1 if non-GAAP has better predictive ability. If I use a continuous variable, a large difference between GAAP prediction and non-GAAP prediction has more effect than a smaller difference. Therefore, I use an indicator variable for non-GAAP quality, to appreciate every observation equally. I use a logistic regression, because the dependent variable non-GAAP quality, is a dummy variable. To determine whether the quality of non-GAAP disclosures changed after the CDI update, I estimate a logistic regression using the following model:

NonGAAPquality= β0 + β1 CDI + Control variables + ԑ (3) Where:

Non-GAAP quality The outcome variable. It is an indicator variable, equal to one if the quality of non-GAAP earnings is more informative than GAAP, and zero otherwise.

CDI An indicator variable, that has a value of one for the observations after the CDI update, which took place in January 2010, and a value of zero for the observations prior to the CDI update.

H1 is supported, if my coefficient β1 is negative. I expect the quality of non-GAAP earnings to reduce after the CDI update.

3.2.2 H2: Cost of capital

In the second hypothesis, the effect of the cost of capital is tested. To test the second hypothesis, I first calculate the cost of capital and after that I estimate an OLS regression to determine what influence non-GAAP quality has on the cost of capital. The traditional method of determining the cost of capital is the Capital Asset Pricing Model. However, for example Daske (2006) states that this method has produced some disappointing results and the results are questionable, as this method uses average realized returns instead of measures of expected returns. The residual income valuation (RIV) model is becoming more widely used, but analysts’ reports have an earnings focus, rather than a book value focus. In this study the price earnings growth (PEG) ratio is used to determine the cost of capital, based on

(22)

22

Easton (2004), because this model is popular in use and has appropriate measures for firm-specific estimates of the cost of equity (Sheng, Barron, & Thevenot, 2012). Even though there is still significant debate about what the better measures available are, Botosan and Plumlee (2005) argue that this model suffers from less measurement errors and is recommended as one of the superior cost of equity capital estimators. This model takes account of differences in short-run earnings growth. The key elements of the model are mostly similar to the key elements of the residual income valuation model, only this model also focuses on earnings, instead of book values only. The model that I use in this thesis is consistent with the studies of Easton (2004) and Sheng et al. (2012), because it is based on firm-specific estimates of the expected rate of return, so the cost of capital is based on an ante basis, instead of the ex-post return. The estimation model is stated as follows:

re =

EPS_t+2-EPS_t+1

p_t (4)

r_" = The implied cost of capital

EPS_&'( = Estimated earnings per share in two years

EPS_&') = Estimated earnings per share in one year

p_& = Price per share

Re The outcome variable of the formula. Measured as the cost of capital of a firm in a certain year.

The analysts’ forecasts are included for a fiscal year ending, that is for example, December 31, 2003 for the fiscal year ending December 31, 2004 (which is EPS₎). For the eps forecasts I use median forecasts. I measure the price per share as the share price at the moment the forecast is made, which is in line with Easton (2004). I estimate equation 5 using the ordinary least squares (OLS) regression with robust standard errors clustered by firm, and the model is stated as follows:

Cost of capital = β0 + β1 Non-GAAP quality + β2 CDI + β3 Non-GAAP quality * CDI

+ Control variables+ ԑ (5)

Cost of capital The outcome variable, which is a result of the calculation of equation four.

(23)

23

The variables non-GAAP earnings and CDI are interacted, to test the hypothesis.

I expect the interaction coefficient β3 to be negative, because when non-GAAP quality is high, this reduces information asymmetry and I expect that it therefore reduces the cost of capital. This result implies that H2 is supported.

3.2.3 Control variables

The control variables I include are, following Cuijpers and Buijink (2005) and Daske (2006), as Daske (2006) examined an effect of a standard on the cost of capital, and Cuijpers and Buijink (2005) examined the influence of non-GAAP on the cost of capital, in Europe. The control variables are used in both hypotheses. They are stated as follows:

Firm size Measured as the natural logarithm of the book value of total assets (AT). It is included to control for a firms’ information environment. Firm performance Determined by the roa, which is measured as net income divided by

total assets. Firm performance is included because the return on assets is an important measure for investors to assess a company’s

performance, and the return they are demanding (Easton, 2004). Loss An indicator variable, equal to one if a company has a GAAP loss,

and zero otherwise. It is included, because when a firm has a GAAP loss, they have more incentives to provide misleading non-GAAP measures.

Leverage Measured by the ratio of long-term debt divided by the market value of equity at fiscal year-end. Leverage is included, because following Daske (2006), A higher leverage is perceived riskier by investors, which results in a rise in the required return.

Time Measured as an indicator variable equal to one if the observations fall into a certain fiscal year, and zero otherwise.

Industry Measured based on 2-digit SIC codes. It is created as an indicator variable, equal to one if a company falls into that industry and zero otherwise. This resulted in 58 industries. Firms in a specific industry share similar business risks and often similar accounting choices (Daske, 2006), which is why industry is included as a control variable. Time and industry controls are included to control for fixed effects.

(24)

24

4. Results

4.1 Descriptive statistics

The descriptive statistics stated in Table 2 provide the following insights. The dependent variable for hypothesis one, non-GAAP quality, has a mean of 0.577 and a standard deviation of 0.494. This means that overall, the predictive ability of non-GAAP is slightly higher than the predictive ability of GAAP. This is in line with Leung and Veenman (2016), who found that about 58% of non-GAAP earnings have a better predictive ability. The mean of Non-GAAP earnings per share is $1.78. For Non-GAAP, the mean earnings per share is $1.01. This is not in line with Black et al. (2017a), who found a mean for both non-GAAP earnings per share and GAAP earnings per share of approximately $3.00. The mean of exclusions is approximately $0.77, which is in line with prior research (e.g. Black et al., 2017b; Whipple, 2015).

The indicator variable CDI has a mean of 0.64, which means that in my thesis, 64% of the observations are from the period after the CDI update, and only 36% from the period prior to the CDI update. The mean of the cost of capital is 0.12, indicating that the mean annual cost of capital in this sample contains a value of 12%. The standard deviation is 0.08, which is not considered high. This is also in line with what Sheng et al. (2012) found. Size has a mean of 7.5, and a standard deviation of 1.7. This means that the firms’ sizes are divergent, which is in line with prior research (e.g. Bentley et al., 2016; Black et al., 2017c). The control variable firm performance has a mean value of 0.009, and the standard deviation is 0.12. Loss has a mean of 0.27 and a standard deviation of 0.45. This means that in my sample, 27% of the observations have a GAAP loss, which is in line with Bentley et al. (2016).

Leverage has a mean of 0.66, which is in line with Black et al. (2017a), who had a leverage with a mean value of 0.61. There is a wide spread in leverage, as the smallest observation contains -5 and the biggest observation contains 11.5. Some firms are all equity based, and firms with a high debt ratio. The dependent variable and control variables are checked for normality in hypothesis two. The dependent variable in hypothesis one and the independent variable in both hypotheses are not tested for normality, as they are all dummy variables.

(25)

25

Table 3 presents the test with the descriptive statistics of the differences of the mean of the variables before and after the CDI. Non-GAAP earnings per share is significantly higher after the CDI, with a mean of 1.242 before the CDI and a mean of 2.081 after the CDI (p < 0.001). GAAP earnings per share is also significantly higher after the new CDI, with a mean of 0.678 before the CDI and a mean of 1.201 after the CDI (p < 0.001). These observations are

included before the CDI and can be a result of a lower mean before the CDI. Total exclusions significantly increased, with a mean of 0.576 before the CDI and a mean of 0.888 after the CDI (p < 0.001). This is in line with Black et al. (2017b), who stated that total exclusions increased significantly. The mean of non-GAAP quality significantly increased, with a mean of 0.558 before the CDI and 0.592 after the CDI (p = 0.06).

[Insert Table 3]

Table 4 provides an overview of the Pearsons’ correlation test, among the dependent, and independent variables that I used. This coefficient measures the strength of a linear correlation between variables. The correlation coefficient value is spread with a minimum of -1 and a maximum of 1. A negative value indicates a negative relationship between the variables. When the value is positive, this indicates a positive relationship between the variables. Non-GAAP earnings per share, Non-GAAP earnings per share and exclusions are highly correlated. This makes sense, because:

Non-GAAP earnings + exclusions = GAAP earnings.

Future GAAP earnings is highly correlated with non-GAAP and GAAP earnings, which implies that the GAAP earnings and non-GAAP earnings both have the ability to, at least partly, predict future earnings. Cost of capital is negatively correlated with both non-GAAP and GAAP earnings. This means that higher non-GAAP/GAAP earnings are correlated with a reduction in the cost of capital. This is not surprising, as firms with higher earnings are less risky, and therefore the required return is lower. Size is positively correlated with non-GAAP earnings and GAAP earnings as well. This is logical, because it means that larger companies have higher earnings. Size is negatively correlated with the cost of capital, which means that larger companies experience a lower cost of capital. Large firms carry less risk, as there is more to gain from them then from smaller firms. Firm performance is positively correlated with non-GAAP earnings and GAAP earnings, and negatively correlated with cost of capital.

(26)

26

This means that better firm performance leads to higher earnings, and with that a lower cost of capital. It makes sense that a firm with better performance has higher earnings, and is less risky, which is rewarded by investors, resulting in a lower cost of capital. Loss is negatively correlated with non-GAAP and GAAP earnings. This is logical, considering that loss is an indicator variable, which is equal to one, when there is a GAAP loss. Loss is positively correlated with both exclusions and the cost of capital. This means that exclusions are higher for firms who experience losses, and cost of capital is higher for firms with losses. This makes sense, because a loss firm has incentives to exclude more in their non-GAAP earnings, as they have more to gain with misleading information, than a profitable firm. Considering that a firm with a loss contains more risk, and investors demand a higher return for riskier firms. Loss is also negatively correlated with size and firm performance, which implies that firms with losses are smaller in size and have a lower firm performance. When a firm has a loss, this is almost always due to a lower firm performance.

[Insert Table 4]

Table 5 delivers more insight in the frequency of observations per year. The frequency is in the first years low, and later on the frequency increases. It is noticeable that the frequency of non-GAAP earnings only decreased in 2010, which may indicate that this was due to the CDI update. However, after 2010 it can be seen that the frequency of non-GAAP observations constantly increases.

[Insert Table 5]

A variance inflation factor (VIF) test conducts a multivariate analysis and checks whether high correlation coefficients lead to multicollinearity in the OLS regression. The VIF test estimates if an independent variable can be considered a linear combination of the other independent variables. A VIF value of 10 or higher indicates that this linear combination is likely in place. Table 6 and Table 7 provide the VIF tests for the regressions based on non-GAAP quality (Table 6) and cost of capital (Table 7). Neither tables show a value of 10 or above, therefore multicollinearity is not an issue in both regressions.

(27)

27

4.2 Results of hypothesis tests

H1: Non-GAAP quality after the CDI update

Hypothesis 1 tests whether non-GAAP quality decreased after the CDI update. The results of the logistic regression are provided in Table 8. An odds ratio between zero and one indicates a decrease, because you cannot have negative odds ratios. When the value is higher than one, there is a positive effect of the CDI. The column CDI is between zero and one and significant, as expected (β1 = 0.722, p = 0.047). In this context this means that the odds of having better quality of non-GAAP earnings after the CDI update is 0.722 times that of having better quality before the CDI update, holding all other independent variables constant. This implies that each time there is an observation after the CDI update (CDI=1), the odds of having a decrease in non-GAAP quality is 27,8% (1-0.722). In logistic output, the coefficients are in log-odd units. The estimates describe the amount of increase in the predicted log odd units of non-GAAP quality = 1 that is predicted by a one-unit increase in CDI. The value of the log odd from the variable CDI is -0.326 (p = 0.047). Therefore, H1 is supported. The quality of non-GAAP earnings is negatively related to the CDI update. The logistic regression outcomes are only available in probabilities. Therefore, it is not possible to discuss the economic

significance of my results.

With respect to the control variables, firm performance has an odds ratio of 2.629 (p = 0.02), which implies that non-GAAP quality is positively affected by firm performance, as the value is more than one. For each additional firm performance, the odds of having a better quality of non-GAAP earnings if 2.629. Loss has an odds ratio of 1.71 (p < 0.001). This means that non-GAAP quality is positively affected by loss. For each additional loss

observation (loss=1), the odds of having a better quality of non-GAAP earnings is 1.71. This is in line with prior research done by Leung and Veenman (2016), in which they state that non-GAAP is especially informative in loss firms. Size and leverage do not have a significant influence on non-GAAP quality. This implies that there is no significant relationship between non-GAAP quality and size or leverage. The model that is being used has not been used in previous literature, which makes it impossible to compare the results completely with other papers.

(28)

28

H2: Cost of capital after the CDI update

Hypothesis 2 tests whether the use of non-GAAP earnings results in a lower cost of capital after the CDI update. The results are provided in Table 9. The interaction effect between non-GAAP quality and CDI is the main variable of interest. This interaction effect provides results when non-GAAP quality is higher than GAAP, after the CDI update. The result is significant. The coefficient, β3 is negative and significant, as expected (β3 = -0.008, p = 0.083). This implies that when non-GAAP information is more informative than GAAP earnings, after the CDI update, results in a lower cost of capital. This means that after the CDI update, disclosing an additional non-GAAP number which is more informative than GAAP, results in a decrease in the cost of capital of 0.8%. I find this effect economically significant, because this leads to cheaper financing. The results are in line with Botosan (2006) & Gietzmann and Ireland (2005), who also found that greater disclosure results in a decrease in the cost of capital. The column non-GAAP quality states what effect non-GAAP quality has on the cost of capital, when CDI is zero (β1 = 0.009, p = 0.041). This means that non-GAAP quality has a

significant positive relationship with the cost of capital, before the CDI update. The column CDI states what effect CDI has on the cost of capital, when non-GAAP quality is zero. β2 is negative and significant (β2 = -0.02, p < 0.001). This implies that CDI has a negative effect on the cost of capital, even when the quality of the underlying GAAP is higher than non-GAAP. For each additional observation after the CDI this results in a decrease in the cost of capital of 2%. The results from the regressions imply that H2 is supported. Providing non-GAAP earnings has a negative effect on the cost of capital, unconditional of the quality of the underlying disclosure.

With respect to the control variables, firm performance has a value of -0.179 (p < 0.001). This indicates that when firm performance increases, cost of capital decreases. This is in line with Francis et al. (2007). Size has a value of -0.008 (p <0.001). This has no real economic consequences, thus it is not economically significant. Loss has a value of 0.037 (p < 0.001). This implies that when a firm experiences a GAAP loss, this results in a higher cost of capital. This is logical, as firms who have a GAAP loss, have a higher risk than profitable firms. When size increases, this results in a slightly lower cost of capital. Leverage is not significantly related to the cost of capital, which implies that there is no significant relationship between leverage and the cost of capital. The adjusted R-squared is 0.345, indicating that almost 35% of the regression analysis is explained by the model. The adjusted R-squared is lower than Cuijpers and Buijink (2005) (R2 = 0.529), the most comparable study.

(29)

29

However, this can be explained by the fact that they used more control variables in their research.

[Insert Table 9]

4.3 Robustness tests

As discussed in the regression analysis, both hypothesis 1 and 2 are supported. In addition to these regressions, a robustness test is performed about the operationalization of the dependent variable. In the main analysis, I suggested that a dummy variable is a better way to measure the quality of non-GAAP quality. In the robustness test, non-GAAP quality is measured as a continuous variable. The difference between a dummy variable and a continuous variable is that a large difference in quality is appreciated higher with a continuous variable than with a dummy variable. With a dummy variable, every observation is coded, as it has a value of one when non-GAAP earnings have better predictive ability, and zero if GAAP earnings have better predictive ability. With the continuous variable, a positive value implies that non-GAAP earnings have better predictive ability, and a negative value means that non-GAAP earnings have better predictive ability.

H1: Non-GAAP quality after the CDI update

To test alternative explanations for the change in quality in non-GAAP earnings, a robustness test is performed. The regression model is almost the same, with the difference that this an OLS regression, whereas the main analysis tests the hypothesis based on a logistic regression. Because the interpretation might differ, this robustness test is performed. The following regression model is used:

NonGAAPquality= β0 + β1 CDI + Controls + ԑ (6)

Where:

NonGAAPquality is the outcome variable. It is measured as the difference between the predictive ability of GAAP and non-GAAP, compared to the GAAP earnings per share one year ahead.

(30)

30

The dependent variable is positive and significant (β1 = 0.0698, p = 0.079, one-tailed). This result differs from my main analysis, in where there was a negative effect between the CDI update and non-GAAP quality. In this robustness test, the results imply that CDI has a positive effect on non-GAAP quality, and therefore, H1 is not supported.

Firm performance has a value of 0.398 (p = 0.012), implying that a higher firm performance has a significant positive effect on non-GAAP quality, which is the same as in my main analysis. Loss has a value of 0.0216 (p = 0.03), which is in line with the main analysis, in where it said that non-GAAP quality is positively affected by loss. Size has a value of 0.164 (p < 0.001), indicating that larger firms result in a better non-GAAP quality. In the main analysis, this effect was insignificant. The value of leverage is insignificant, which is confirming the main analysis. Overall, the robustness test does not support the main analysis, which is why I did not chose this method in the first place, because the dummy variable codes every observation equally, and a continuous variable does not do that.

[Insert Table 10]

H2: Cost of capital after the CDI update

To test an alternative explanation for the relationship between non-GAAP and the cost of capital, a robustness test is performed. The OLS regression stays the same, only non-GAAP quality is measured as a continuous variable. The following regression model is used:

Cost of capital = β0 + β1 Non-GAAP quality + β2 CDI + β3 Non-GAAP quality * CDI

+ Control variables+ ԑ (7)

I expect the coefficient of β3 to be negative, as before. The results are provided in Table 11. The dependent variable is negative and significant (β3 = -0.015, p = 0.070). My main analysis is supported, similar to the main analysis. When Non-GAAP quality is high, before the CDI, it has a value which is not significant, indicating that there does not exist a significant

relationship between non-GAAP quality and the cost of capital, prior to the CDI update. This is not in line with my main analysis, in where I found that non-GAAP quality was related to a higher cost of capital, before the CDI update. When CDI is one and non-GAAP quality is zero, the outcome variable has a value of -0.019 (p < 0.001), which means that CDI is

(31)

31

negatively related to the cost of capital, which confirms my main analysis. Firm performance has a value of -0.179 (p < 0.001), which indicates a negative relationship between firm performance and the cost of capital. This value is almost completely the same as the main analysis. Loss has a value of 0.037 (p < 0.001). This implies that loss is positively related to the cost of capital, which is the same outcome as my main analysis. Size has a value of -0.008 (p < 0.001), which implies a negative relationship between size and the cost of capital. This is the same value as my main analysis. And finally, leverage has a value of 0.002 (p = 0.052), which implies a positive relationship between leverage and the cost of capital. All of the control variables are in line with the main analysis, except for leverage. In the main analysis, leverage was not significant, and in here it is.

(32)

32

5. Conclusion

There is much debate about non-GAAP earnings disclosures, whether they are used to inform or mislead. This can be due to the quality of the underlying disclosures. With every kind of voluntary disclosure, there is debate about whether they are used to inform or mislead. However, non-GAAP disclosures are a distinct category, as managers argue that non-GAAP numbers better reflect ‘core’ operations, than GAAP numbers do (Leung & Veenman, 2016). As a result, regulators initiated a change in regulation, firstly by Regulation G in 2003 and again in 2010, with new Disclosure and Compliance Interpretations. The rules were initiated to reduce the misleading use of non-GAAP earnings, and provided additional requirements when disclosing non-GAAP numbers.

Most studies find that there has been a decline in non-GAAP earnings after Regulation G, and an increase in non-GAAP earnings after the CDI update. The CDI update increased flexibility in non-GAAP reporting, to the extent that it states that non-GAAP numbers are appropriate, as long as they are explained by adequate disclosures. Therefore, this study examines the influence of non-GAAP earnings on the cost of capital, after the CDI update of 2010. More specifically, I investigate whether the use of non-GAAP earnings results in lower earnings informativeness, after the CDI update, and if a lower quality results in a higher cost of capital.

First, I examined whether non-GAAP quality decreased after the CDI update, by predicting future GAAP earnings for both non-GAAP and GAAP. The one with better predictive ability, is the more informative one. This results in an indicator variable, after which I examined what influence the CDI had on the quality of non-GAAP earnings. A logistic regression was performed to test my analysis. My findings show that there is a

negative effect between the CDI update and the quality of non-GAAP earnings. However, my findings differ when non-GAAP quality is a continuous variable, and an OLS regression is performed.

Next, I examined what influence non-GAAP earnings have on the cost of capital after the CDI update. My findings show that there is a negative relationship between the CDI update and the cost of capital, indicating that the cost of capital decreases through non-GAAP earnings, after the CDI update. This is confirmed in the additional analysis, measuring non-GAAP quality as a continuous variable.

Taken together, my results show that while the quality of non-GAAP earnings is lower than that of GAAP earnings, there is still a negative influence on the cost of capital after the

(33)

33

CDI update. This implies that investors are tricked into believing possibly misleading non-GAAP numbers, and reward it with a lower cost of capital.

I contribute to the non-GAAP literature by analyzing whether the CDI update has an effect on the quality of GAAP earnings, and if the cost of capital is influenced by non-GAAP earnings after the CDI update. Prior literature has examined whether non-non-GAAP earnings were used to inform or mislead (e.g. Black et al., 2017c; Leung & Veenman, 2016), and examined what effect voluntary disclosure has on the cost of capital (Gietzmann & Ireland, 2005). My thesis is distinct from other research, as I examined the effect of the CDI update on the quality of non-GAAP earnings, and the influence on the cost of capital. This is a unique setting, which has not been examined before. I also contribute to the literature of voluntary disclosure, by examining if non-GAAP disclosures influence the cost of capital. My findings showed that non-GAAP disclosures lead to a reduction in the cost of capital, where prior research not always found that relation.

My results have the following implications for managers, investors and regulators. Managers now know that providing a non-GAAP measure, informative or not, reduces the cost of capital. Managers can decide to provide a non-GAAP measure, because it leads to a reduction in the cost of capital of the firm. The findings might also be useful to investors, in determining carefully, whether non-GAAP earnings are informative or misleading. My research implies that investors reward both informative and misleading non-GAAP numbers with a lower cost of capital, after the CDI update. My findings suggest that investors should check the quality of the underlying non-GAAP disclosures carefully, or they might misprice a firm. The regulatory implications of my findings are that regulators are right in questioning whether non-GAAP earnings are used to inform or mislead. My results imply that regulators should not stop questioning whether non-GAAP earnings are used to inform or mislead. In my main analysis, I found that the quality of non-GAAP earnings decreased after the CDI update. This was not confirmed with my robustness test.

There are several limitations to my research. The first one is the relatively small sample, which started with about 146,000 observations. Due to the fact that not every year contained four quarter-observations, a lot of observations needed to be removed. This can have an impact on the results. When a firm only provides a non-GAAP number at the end of fiscal quarter four, it can be more misleading than a firm that provides a non-GAAP number each quarter. At the end of the fourth quarter, non-GAAP can be used as a perception management tool, to make it seem that non-GAAP earnings expectations are met, where

Influence of non-GAAP earnings on the cost of capital