Non-GAAP earnings in disclosures : the effect on CEO compensation

(1)

Non-GAAP Earnings in Disclosures

The Effect on CEO Compensation

Name: Boudewijn Weel Student number: 11421673 Thesis supervisor: Dr. P. Kroos Date: 25-06-2018

Word count: 11,424

MSc Accountancy & Control, specialization Control

(2)

Statement of Originality

This document is written by student Boudewijn Weel 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)

Acknowledgments

I would like to thank my thesis supervisor dr. P. Kroos for all his valuable input to support me during my thesis. His knowledge, enthusiasm, and faith helped me tremendously in completing the final challenge of my Master’s degree. Additionally, I really appreciated that I could always visit his office for questions and a small chat.

I would also like to thank EY for providing me with all the necessities for writing my thesis. Great colleagues, an amazing office, and good supervision.

(4)

Abstract

The use of non-GAAP earnings in disclosures has increased over the past years. However, there are concerns about the actual motives for disclosing these earnings. Non-GAAP earnings are GAAP earnings but adjusted for the items which management considers unusual. This leads to opportunities for management to better portray performance and which does not reflect economic reality. Prior research provided evidence of powerful CEOs influencing their performance measures to increase their compensation and CEOs could also influence

performance measures by making non-GAAP adjustments. This study investigates whether CEO compensation drives the motives for non-GAAP disclosures. First, this study demonstrates that there is an association between non-GAAP adjustment levels and CEO compensation. In subsequent analysis, this study finds evidence of a moderating effect of CEO power on this association. The results indicate that this moderating effect is positive when the firm makes large adjustments and this association is negative when the non-GAAP adjustments are smaller. The explanation for the first observation is that compensation committees use large non-GAAP adjustments to justify high compensation which would not have been warranted by profit or stock performance. However, this study cannot provide an explanation for the negative

moderating effect when the firm makes small adjustments. This could be an interesting direction for future research. Altogether, this study provides evidence that CEOs are influencing their compensation through non-GAAP disclosures.

(5)

1 Introduction

Background

Performance measures are important in a market based-economy (Beyer, Cohen, Lys, & Walther, 2010) as it is used for valuation of firms and to assess the stewardship of the CEO. First, the valuation purpose is aimed to provide information to capital market participants to reduce the information asymmetry between them and the CEOs. The disclosure of performance measures in the shape of accounting information should provide capital market participants information to estimate future cash flows and assess risks to support investment decisions (FASB, 1978).

Earnings releases are the primary method to inform investors about firm performance according to Dechow, Sloan, & Zha (2014). This paper shows that trade volumes and abnormal returns almost double around earnings announcements. The second reason for the importance of performance measures is to motivate executives to take actions in the best interests of

shareholders. The separation of ownership between the CEO and the shareholders results in an agency problem because the CEO acts in his self-interest. This is called moral hazard and in order the reduce the effects, shareholders design performance measures to align the interests of the CEO and the shareholders (Darrough & Stoughton, 1986).

Although earnings metrics are used to inform investors and measure performance there are concerns about the reliability of this information. First, earnings can be manipulated to present better performance of the firm. Secondly, earnings information may not truly reflect actual performance because of binding accounting methods. To address the second issue, management can choose to disclose alternative earnings metrics that better reflect their performance. These metrics are called non-GAAP or pro forma earnings which are GAAP earnings but adjusted for items that management believes to be unusual or non-recurring (Bhattacharya, Black, Christensen, & Larson, 2003). An example of nonrecurring items are restructuring costs but Black & Christensen (2009) provide evidence that recurring items like depreciation, research and development and stock-based compensation are also excluded from non-GAAP earnings. The fact that management has an incentive to portray better performance of the firm raises the question about the real motives of non-GAAP disclosure. Are non-GAAP adjustments used to inform investors better or is it used to portray better performance? Prior research provides mixed evidence. On the one hand, Bhattacharya et al. (2003) showed that non-GAAP earnings are more informative to investors than non-GAAP operating earnings. On the other hand, Doyle, Jennings, & Soliman (2013) provide evidence that non-GAAP metrics are a

(8)

which perform better according to Bowen, Davis, & Matsumoto (2005). Most prior research discussed the informativeness and opportunistic motives of use of non-GAAP earnings in external disclosure, whether they are to inform or mislead analysts and investors. However, prior research is relatively silent on the direct relationship between the disclosure of non-GAAP earnings and CEO compensation. This is especially important as research has shown that executives may exercise their power to influence their compensation.

Research question

Based on the aforementioned, my research question is described as follows:

“Do CEOs influence their executive compensation through non-GAAP earnings disclosures?” My first analysis will focus on answering the question of how non-GAAP earnings are associated with compensation levels. Subsequent analysis will focus on the question whether the former relationship varies with proxies for CEO power.

Relevance of the research question

This research is especially important because the disclosure of non-GAAP earnings is rising. D. E. Black, Christensen, Ciesielski, & Whipple (2018) state that in 2014 71 percent of the firms in the S&P 500 disclosed non-GAAP metrics while in 2009 this was only 53 percent. The

magnitude of non-GAAP adjustments is increasing as well. In the period from 2009 till 2014, the expenses that were excluded increased with a rate of 40.6 percent. This increase is attributable to the increase in nonrecurring items since this almost doubled in this period. This is remarkable because the definition of GAAP metrics by Bhattacharya et al. (2003) states that only non-usual or non-recurring items are excluded from GAAP earnings. This study contributes by investigating the role of non-GAAP earnings inflating CEO compensation and whether this relationship varies with proxies for CEO power. This study also contributes to the literature on executive compensation. More specifically on the literature that assumes efficient contracting and potential rent extraction.

Structure of the thesis

The remainder of this paper is as follows. Section 2 outlines the relevant literature regarding the disclosure of non-GAAP earnings and executive compensation and subsequently provides the hypothesis of this research. After that, the paper continues with section 3, which provides an

(9)

overview of the research method. In section 4, the main findings of the analysis are given, and the conclusion is presented in section 5.

(10)

2 Literature review

Agency problems, financial reporting, and executive incentive contracts

2.1.1 Asymmetric information

Prior literature demonstrates that shareholders and the CEO are not on the same page. Shareholders and CEOs have a conflict of interest where they have their own individual goals, characteristics, and knowledge of the firm. This conflict of interest is called an agency problem that arises when two parties cooperate with different personal objectives and division of labor (Eisenhardt, 1989). The division of labor is between the shareholder (principle) and the CEO (agent) where the CEO agrees to work for the shareholder. Agency theory is about solving two problems which arise with this relationship (Eisenhardt, 1989). The first problem is that the CEO has different objectives than the shareholder while it is difficult or expensive for the shareholders to verify the actions of the CEO. Examples could be that the CEO prefers leisure time over maximizing the return of the shareholders. The second problem is that that shareholders and the CEOs have different attitudes towards risk. That is because the CEO has limited opportunities to diversify his income where shareholders can have many investments and therefore can spread their risk.

Darrough & Stoughton (1986) state that moral hazard (hidden actions) and adverse selection (hidden information) are two ways in which an agency problem can take shape. According to these, moral hazard is the problem that arises “when the action undertaken by the agent is unobservable and has a differential value to the agent as compared to the principal.” Moral hazard in the context of this relationship is that the CEO may have different objectives and therefore his actions are not geared towards maximizing firm value. An example could be that the CEO is golfing instead of creating value for the shareholders. Biggerstaff, Cicero, & Puckett (2017) investigate the relationship between CEO’s golfing activities and firm performance and find that firm value and operational performance is lower when CEOs spend much time on the green. Adverse selection is “when the agent has more information than the principal” (Darrough & Stoughton, 1986). Again in this context, this means that the manager has more information about firm characteristics compared to the shareholders. Adverse selection could be knowledgeable about the future profitability, past performance, pending lawsuits and all other information which is relevant for the valuation of the firm. This information asymmetry combined with the conflict of interest between CEOs and investors is called the “lemon problem” according to Healy &

(11)

Palepu (2001). As a result, investors cannot distinguish between good and bad investments and value all firms at an average level.

2.1.2 Financial reporting to alleviate problems with asymmetric information

A possible solution for firms and investors to alleviate information asymmetry is by financial reporting to capital market participants. Leuz & Verrecchia (2000) demonstrate that firms can decrease information asymmetry by increased corporate disclosure about the performance of the firm. Listed organizations must follow mandatory disclosure requirements which are set by the US Securities and Exchange Commission (SEC). Mandatory disclosure is necessary because of the following two main reasons summarized in the paper of Beyer et al. (2010). First, the different incentives between management and the shareholders can prevent managers to disclose information credibly. Second, because of the public goods characteristics of corporate disclosure, it can lead to situations where the manager’s incentives to voluntarily disclose information are too weak even though this can lead to an improvement of social welfare. The underproduction of disclosures can be explained by the fact that current shareholders indirectly pay for the disclosure costs. When this information is freely available to the public, prospective investors can benefit from this information (Healy & Palepu, 2001). Auditors provide investors assurance that the financial statements are prepared according to the generally accepted accounting principles (GAAP) (Healy & Palepu, 2001). The fact that trade volume and share price react to earning announcements implies that investors value the validated accounting metrics (Dechow et al., 2014). Besides mandatory disclosure, firms can also choose to disclose more information to reduce information asymmetry voluntarily. Healy, Hutton, & Palepu (1999) found the several effects from investors at firms which increase their voluntary disclosures. They observed that investors made upward adjustments to the share price, increased share liquidity, increased interest of institutions and analysts in the shares. Examples of voluntarily corporate disclosure are press releases, management forecasts, conference calls and non-GAAP metrics.

2.1.3 Efficient contracting perspective

Another way to address information asymmetry is by using optimal contracts to align the interests of the investors and the CEO (Healy & Palepu, 2001). An optimal contract or efficient contract is a contract “that maximizes the net expected economic value to shareholders after transaction costs (such as contracting costs) and payments to employees” (Core, Guay, & Larcker, 2003). Or in other words, a contract with the least agency costs. An important implication of this contract is that the CEO is not rewarded for observable luck (Bertrand & Mullainathan, 2001). These contracts are often in the shape of a compensation contract or a debt covenant. This contract

(12)

should alleviate the following two moral hazard issues by aligning the interests of the shareholder and the CEO (Bebchuk, Fried, & Walker, 2002). The first issue is the risk that the CEO will not expand enough efforts to maximize shareholder value. The second issue is the risk that the CEO will take actions in his own self-interest at the expense of the shareholders. The efficient contracting perspective implies that the level of compensation of the CEO is optimal for the shareholder in a way that it minimized agency cost according to Bebchuk et al. (2002).

CEO compensation increased substantially during the last century and especially during the 1990s (Frydman & Saks, 2010). According to some academics, this increase in CEO compensation can be explained by the efficient contracting perspective. Following Gabaix & Landier (2008), the increase in CEO compensation between 1980 and 2003 can be attributed to the increase in market capitalization. The relationship between market capitalization and CEO compensation is as follows: CEOs have different levels of talent and these talents are matched with firms on a competitive basis. As the size of a firm increases, the marginal product of the CEO’s actions increases as well. As a result, talent is more valuable for larger firms because the effect of this talent is bigger. Firms can increase CEO compensation up to the point that the marginal product of the CEO’s actions is equal to the marginal compensation cost to incentivize the CEO. Therefore talented CEOs have a higher equilibrium wage. Murphy & Zabojnik (2006) also state that the increase in CEO compensation is consistent with the efficient contracting perspective. Due to the rise of information technology, firm-specific information like information about product markets, suppliers and clients are widely available to all CEOs, and therefore firm-specific knowledge is a less relevant asset. This leads to increased relevance of CEOs with managerial capabilities like managing assets and people, financial and accounting skills. Due to competition for talent between firms, CEOs can increase their compensation because of their increased transferability. This is effect is most prevalent for the highest-ability CEOs, since competition for these people is more intense.

However, other researchers claim that CEO compensation is not solely determined based on what is optimal for the shareholder. It is also affected by the ability of the CEO to influence the board of directors because they have empathy for the CEO or that the board of directors is not overseeing CEO compensation effectively. Bebchuk & Fried (2005) are proponents of the managerial power approach. The managerial power approach states that powerful CEOs receive more compensation and have a lower pay-for-performance sensitivity than their less-powerful counterparts. The efficient contracting perspective assumes that compensation contracts of CEOs are optimal for the shareholder but Bebchuk & Fried (2005) state that there is no reason to believe that managers or directors seek to maximize shareholder value. Academics found evidence

(13)

consistent with the managerial power perspective that CEOs increase their compensation without creating firm value.

For instance, research by Bergstresser & Philippon (2006) reveals that firms with CEOs whose wealth is more sensitive to stock prices have more discretionary accruals. Furthermore, this research provides direct evidence that these CEOs benefit from the increase in share price as they observe that CEOs exercise substantially more options and sell more shares during these years with high accruals. Also, Garvey & Milbourn (2006) find that rewards and losses because of luck (systematic risk) are asymmetric. They provide evidence that executives suffer 25 percent-40 percent less from bad luck as they would benefit from luck. There are situations when marginal performance does not lead to a higher bonus. Annual cash bonuses often have a minimum threshold and a certain cap. As a result, CEOs will not receive extra compensation when they are outside the incentive zone even though their performance measure increased. Conversely, CEOs have incentives to better influence earnings when they are in the incentive zone. Prior research found several relationships between the relative position of the performance measure in the incentive zone and accruals which implies that managers are influencing their bonus in a direct manner as well (Gaver, Gaver, & Austin, 1995; Guidry, Leone, & Rock, 1999; Healy, 1985; Holthausen, Larcker, & Sloan, 1995). Even though the managerial power perspective contradicts the efficient contracting perspective, it cannot explain the rise in CEO compensation that started in 1970 according to Frydman & Jenter (2010). They note that due to regulation, the corporate governance quality increased since 1970. But since CEO compensation increased as well, this positive association is inconsistent with the managerial power approach because this approach predicts that CEO compensation and corporate governance quality are negatively related.

Non-GAAP metrics

Firms can choose to disclose an unaudited alternative earnings measure next to GAAP earnings to inform investors better. Non-GAAP earnings are “GAAP earnings adjusted for items that management deems to be ‘‘unusual’’ or ‘‘non-recurring’’ (Bhattacharya et al., 2003). The following example provides a better understanding of how non-GAAP earnings can be used in practice:

Assume that you are the CEO of a small store. The store is fully owned by an external shareholder who does not have an active position within the firm. Every month, your profit is $1,000, and you report this to your shareholder. This month unfortunately, you got a parking fine of $100, and therefore your profit is $900 in this particular month. The shareholder is mainly interested in your performance of selling products and does not consider the fine relevant. You

(14)

can choose to disclose a non-GAAP profit of $1,000 in addition to your GAAP profit of $900 to better inform the shareholder about your core performance.

The non-GAAP earnings metric in the example above provides a better reflection of the core performance of the store compared to the GAAP earnings metric. This is consistent with the stated objective of non-GAAP metrics, which is to provide a more relevant earnings metric that is more informative to investors.

There are doubts regarding the actual motives of non-GAAP disclosure because management can use their own judgment when they make non-GAAP adjustments. Prior research focused on the informativeness versus opportunism aspect of non-GAAP (D. E. Black, Christensen, et al., 2017). Managers have incentives to overstate earnings to meet or beat the analysts’ earnings expectation since this lead to an abnormal return increase of 3.2 percent (Bartov, Givoly, & Hayn, 2002). While the definition of non-GAAP earnings states that management only adjusts items that are “unusual” or “non-recurring,” prior research showed that recurring items are also adjusted. For instance, Black & Christensen (2009) find that recurring items like research and development, depreciation and amortization and stock-based compensation are excluded from GAAP earnings. Non-GAAP metrics are intended to provide a better picture of the firm performance but it also offers management possibilities to overstate firm performance.

Prior research states that before 2003, investors were misled by non-GAAP earnings because firms were not required to clarify how they derived to this non-GAAP metric (Jennings & Marques 2011). In 2003, the SEC adopted regulation G which forced firms to report a reconciliation table to the most comparable GAAP equivalent. Jennings & Marques (2011) find that after the adoption, investors were no longer misled due to the increased transparency. Zhang & Zheng (2011) find consistent evidence that securities were no longer mispriced after regulation G. Bhattacharya et al. (2003) and Bradshaw, Christensen, Gee, & Whipple (2018) found that non-GAAP earnings are more informative to investors than non-GAAP earnings. Additionally, the research of Entwistle, Feltham, & Mbagwu (2010) had consistent evidence. They showed that share prices react more to non-GAAP earnings than GAAP or I/B/E/S earnings. One could argue that the comparability decreases when every firm has the discretion to adjust items to derive at their own non-GAAP earnings. However, D. E. Black, Christensen, Ciesielski, & Wipple (2017) find evidence inconsistent with this statement. They discovered that non-GAAP earnings are more comparable across industry peers than their GAAP equivalent. This is because managers can exclude items which are firm-specific and do not affect other firms. Additionally, they showed that there is a positive relationship between an inconsistency of adjustments and high non-GAAP

(15)

reporting quality. This indicates that firms are selecting adjustments to improve the informativeness of non-GAAP earnings rather than for opportunistic reasons. Finally, Leung & Veenman (2016) conducted research about loss converters. Loss converters are firms who made a loss according to GAAP but reported a non-GAAP profit. They found that the motive behind this loss conversion is to increase the informativeness and to help investors better assess future performance. Additionally, investors conclude that these firms are more likely to become profitable compared to companies who report losses for both GAAP and non-GAAP earnings.

However, there are signals that the use of non-GAAP earnings is also motivated by incentive contracts. First, there are indicators that executives base their disclosure decisions on the compensation of the CEO and the CFO. For example, Isidro & Marques (2013) provide evidence that there is a higher probability of non-GAAP disclosure when the compensation contract is based on market performance measures. They also find a relationship between market performance measures and non-GAAP metrics in the headline, more adjustments of recurring items and not providing a reconciliation between the non-GAAP and GAAP metric.

CEO power and rent extraction

Financial economists that studied executive compensation assumed that the compensation contract of the CEO was determined according to the efficient contracting perspective according to Bebchuk & Fried (2005). Because of the conflict of interest between CEOs and shareholders, these authors claim that there is no indication to think that CEOs seek to maximize shareholders’ wealth. They propose that managerial power determines CEO pay. The managerial power approach states that powerful CEOs receive a higher compensation or that their compensation is less sensitive to performance compared to less powerful CEOs. CEO power reflects the amount of influence that the CEO has over the board (Nanda, Silveri, & Han, 2016). The more power the CEO has, the more he is able to extract rents.

Bebchuk & Fried (2005) discuss incentives and circumstances that result in directors behaving differently than in the best shareholders’ interests. The first reason is that directors have incentives to be re-elected. Holding a board seat is beneficial, both financial and nonfinancial. Directors have incentives to agree with the proposed compensation of the CEO because he has a major influence on the nomination process. Still, the proposed compensation contract must still be justifiable towards the shareholders. In addition, actively blocking compensation proposals of the CEO can decrease the chances of being invited to other boards. Another reason is that the CEO has influence on the compensation of the directors by supporting or opposing director compensation proposals. When directors support the compensation of the CEO, they might

(16)

expect that he will support their increase in compensation in turn. Research has shown that firms whose CEO receives a high compensation, the directors receive higher compensation as well. However, this high compensation of both the CEO and the directors is not associated with superior performance but with underperformance. This implies mutual back-scratching or cronyism according to Brick, Palmon, & Wald (2006). Another indication could be that directors may feel an obligation to return a favor when he is appointed during the current CEO’s tenure. Directors may also treat CEOs with certain deference because the CEO is the leader of the firm. In both cases, it is harder to disagree with the CEO about his compensation. Because the directors typically have small ownership in the firm, the costs are small to agree with the compensation contract of the CEO.

Hypotheses development

2.4.1 Hypothesis 1

The stated objective for disclosure of non-GAAP metrics is to inform investors better. Besides the informativeness objective, it can also be exploited opportunistically by CEOs by making large adjustments which do not reflect economic reality. Dichev, Graham, Harvey, & Rajgopal (2016) find that CFOs already use their discretion within GAAP standards to influence executive compensation. These authors conducted a survey among 375 CFOs about their motives to report earnings that misrepresent economic performance. Approximately 90 percent of all CFOs stated that executive compensation was a relevant factor.

Non-GAAP adjustments can impact CEO compensation in two ways. First, it can directly influence the earnings metric which they are evaluated upon. Prior research provided evidence that CEOs are influencing their GAAP performance measures already with accrual management when this leads to an increased bonus (e.g., Gaver et al., 1995; Guidry et al., 1999; Healy, 1985; Holthausen et al., 1995). An indication that a CEO can yield more compensation when they are evaluated based on non-GAAP earnings compared to GAAP earnings is provided by Pozen & Kothari (2017). They state that in 2015, most of the firms that had the largest non-GAAP adjustments also used this metric in CEO compensation contracts. The non-GAAP metrics in these contracts, determined for at least 40 percent the annual bonus, long-term grant or both. The second way that non-GAAP adjustments can impact CEO compensation is through better stock price performance by reporting opportunistically. Prior research already showed that CEOs increase their wealth by earnings and accrual management (e.g., Bergstresser & Philippon, 2006). Just like earnings management, non-GAAP adjustments can also be used to portray a better

(17)

performance. Consistent with this statement, E. L. Black, Christensen, Taylor Joo, & Schmardebeck (2017) find that non-GAAP reporting and earnings management have a substitute relationship. Firms are more likely to disclose non-GAAP metrics when they cannot use accrual and earnings management. Doyle et al. (2013) and Elshafie, Yen, & Yu (2010) provide consistent evidence of non-GAAP adjustments and earnings management being substitutes. Shiah-Hou & Teng (2016) also state that the personal wealth of the CEO motivates non-GAAP disclosure. They find that firms are more likely to disclose non-GAAP earnings when the CEO and CFO plan to sell shares within two weeks after the earnings announcement.

While this evidence points to the opportunistic use of non-GAAP adjustments, prior research also provided evidence against the concerns discussed above. For instance, Curtis, Li, & Patrick (2017) find that the use of non-GAAP performance measures is consistent with the efficient contracting perspective. They do not find evidence that CEOs are more likely to meet or beat bonus targets and receive a higher bonus when they are evaluated on non-GAAP performance measures. Consistent with this finding, D. E. Black, Black, Christensen, & Gee (2017) find that non-GAAP earnings are of higher quality when these metrics are used in disclosures and for compensation purposes. These findings argue against that CEOs use non-GAAP metrics to increase their pay. As earlier discussed, another motive to disclose non-GAAP earnings could be to increase stock performance. However, Zhang & Zheng (2011) and Jennings & Marques (2011) find that after the reconciliation requirement of regulation G, investors are no longer misinformed and securities are not misprized. Finally, Guest, Kothari, & Pozen (2017) propose an alternative explanation. They state that firms do not disclose non-GAAP earnings to inform or to portray a better performance. They suggest and find evidence that non-GAAP reporting is used by compensation committees to justify high CEO compensation which is not warranted by profit or stock price performance. The first hypothesis follows the same rationale as the paper of Guest et al. (2017) and is as follows:

H1: The level of CEO compensation is positively associated with the use of non-GAAP adjustments in disclosures.

According to the managerial power approach, powerful CEOs receive a higher compensation or have a less pay-for-performance sensitivity compared to their less powerful peers (Bebchuk & Fried, 2005). First, Morse, Nanda, & Seru (2011) and later Morse, Nanda, & Seru (2014) provide evidence that powerful CEOs can shift performance measures to better-performing ones to increase their compensation. They call this rigging, and this is similar to adjusting non-GAAP

(18)

earnings to increase compensation. Abernethy, Kuang, & Qin (2015) also investigated the ability of powerful CEOs to influence their compensation contract design. They observed that powerful CEOs were able to shift option vesting conditions from market to EPS based on which targets were easier to meet.

Next to personal characteristics, CEOs can also be powerful as a result of low corporate governance quality. Prior research found a negative relationship between corporate governance quality and the probability of non-GAAP reporting and the magnitude of non-GAAP adjustments according to Isidro & Marques (2013). They also find that market-based incentives lead to emphasizing non-GAAP earnings in press releases, more adjustments of recurring items and avoidance of reconciliation of the non-GAAP metric to their GAAP equivalent. These authors demonstrate that high-quality corporate governance can alleviate these opportunistic disclosure practices. Consistent with this finding, Frankel, McVay, & Soliman (2011) and (Jennings & Marques (2006) also find that low corporate governance is associated with more frequent and more opportunistically non-GAAP disclosure. The evidence of manipulating performance measures and opportunistic use of non-GAAP reporting by powerful CEOs results in the following hypothesis: H2: The power of the CEO moderates the relation between the use of non-GAAP adjustments in disclosures and the level of executive compensation.

(19)

3 Data and research design

Sample

The sample consists of 3,210 US-listed firms that disclose non-GAAP EPS in their earnings announcements. The time frame of this research is the period between the fiscal years from 2003 to 2015. The reason that the start date of the sample is 2003 is that of the adoption of regulation G at the beginning of that year. This regulation forced companies to report a reconciliation table between the non-GAAP metric and their GAAP equivalent. In order to ensure consistency, the years prior to 2003 are not used in this research. The composition of the sample is based on the database of Bentley, Christensen, Gee, & Whipple (2017). This database consists of 49,919 observations of non-GAAP EPS disclosed in quarterly earnings announcements. The non-GAAP per year is the aggregated value of four quarters of the same fiscal year if those are available. If not, the quarterly observations are dropped. After dropping all the incomplete fiscal years and aggregating the four quarters, the total observations of firms with an annual non-GAAP EPS is 4,872. Subsequently, 1,560 and 102 observations are dropped due to the unavailability of compensation and control variable data respectively. This leads to the final sample size of 3,210. In addition, the sample size to investigate the moderating role of CEO power is 1,677 because of the lack of CEO power data for 1,533 observations.

Empirical model

The main independent variable in this research is Non-GAAP adjustment, and it is constructed based on research by Guest et al. (2017). This variable is operationalized by calculating the difference between GAAP EPS and non-GAAP EPS. Subsequently, the observations are categorized into six groups based on their magnitude. Group 0 contains firms with income-decreasing adjustments and the other observations are categorized in quintiles based on the adjustment levels where the firms with the lowest adjustments are in Category 1 and the firms that have the biggest adjustments are in Category 5. As earlier stated, the non-GAAP data is extracted from the database of Bentley et al. (2017). The GAAP EPS data is the “Earnings Per Share (Diluted) - Including Extraordinary Items” variable from Compustat. The dependent variable is CEO compensation. CEO compensation is extracted from Execucomp and Incentive Lab. Consistent with the study of Faulkender & Yang (2010), CEO compensation data is winsorized at the 1st_{and 99}th_{percentile to control for outliers.}

(20)

supports the following two analyses. First, this analysis can determine whether there is a linear relationship between non-GAAP adjustment and CEO compensation. Secondly, this can provide evidence whether this association differs across categories and is nonlinear. The control variables which are used in the regression are discussed in section 3.2.3.

The first regression will test whether there is a linear relationship between the Category and CEO compensation levels. In this regression, Category is treated as a continuous variable. On the basis of the hypothesis, I expect β1 > 0. The following equation is constructed and tested by using OLS regression:

Equation (1):

The subsequent analysis is going to allow for a nonlinear relationship between the individual non-GAAP adjustments categories and CEO Compensation. In contrast to equation (1), the different categories are treated as dummy variables instead of treating Category as a continuous variable. Note that the group 0 (the income-decreasing adjustments) acts as the reference group. On the basis of the hypothesis, I expect β1 to β5> 0. This relationship is tested by using OLS

regression with the following equation: Equation (2):

The Libby Boxes provide a visualization of the first hypothesis and can be found in figure A1 of appendix A.

CEO Compensationt = β0 + β1Categoryt + β2ROAt + β3ROAt-1

+ β4RETt+ β5RETt-1 + β6Revenuet-1

+ β7BTMt-1 + β8Leveraget-1 + + β9IndustryVolatilityt + ε

CEO Compensationt = β0 + β1Category1t + β2Category2t + β3Category3t

+ β4Category4t+ β5Category5t + β6ROAt

+β7ROAt-1+ β8RETt + β9RETt-1

+ β10Revenuet-1 + β11BTMt-1

(21)

The second hypothesis addresses the moderating effect of CEO power on hypothesis 1. CEO power reflects the amount of influence that the CEO has over the board (Nanda et al., 2016). CEO power is operationalized based on the research of Morse et al. (2011). They operationalize CEO power with a power index variable which addresses the concentration of titles. In this research, I extend this power index by using more proxies for CEO power than used in the work of Nanda et al. (2016). The Power Index Score in this research consists of CEO Pay Slice, Duality,

Triality, Tenure, Inside Directors, and Directors Appointed by Current CEO. When the value of the variable

is higher than the industry median, the indicator variable equals to one for all proxies except for

Duality and Triality. These indicator variables equal to one when the definition corresponds with

titles of the CEO. The maximum of the Power Index Score equals the value of 7.

CEO Pay Slice is described by Bebchuk, Cremers, & Peyer (2011) as the fraction of CEO

compensation out of the aggregated sum of compensation of the top 5 executives. They state a higher pay slice does not reflect the relative importance of the CEO but is a signal of power, and it indicates that the CEO is better able to extract rents1_{. The source of this data is the “Executive}

Rank by Salary + Bonus” variable from Compustat.

Prior research used concentration of titles as a proxy for CEO power according to Nanda et al. (2016). Duality is when the CEO is also the chairman of the board and when this is the case the index variable increases by one. Triality describes a CEO who is both the chairman of the board and the president. When this is the case, the index variable increases by one again. The function of the CEO is derived from the “Annual Title” variable of Execucomp. The search function examined whether “president” and “chairman” were part of the annual title to determine the concentration of titles of the CEO.

Tenure describes the period that the individual of interest is the CEO. Research by Ryan &

Wiggins (2004) showed that the bargaining power of the CEO is positively related to CEO tenure.

Tenure is calculated by subtracting the end of the relevant fiscal year to the “Date Became CEO”

variable from Execucomp.

Stock Ownership is the percentage of stock owned by the CEO. When the CEO has more

shares, he has more power over the board (Finkelstein 1992). This data is extracted from the “Percentage of Total Shares Owned – Options Excluded” variable from Execucomp.

11_{In contrast to the research by Bebchuk et al. (2011), the pay slice of the CEO is in this research the fraction of the}

(22)

Prior research used the Inside directors as a proxy for CEO power (Morse et al., 2011; Ryan & Wiggins, 2004). When the board consists of less independent directors, the CEO experiences less resistance. To determine whether a director was an inside director is based on the “Board Affiliation” variable from the Institutional Shareholder Services (ISS) database where directors with the “E – employee” and “L – Linked” classification are considered inside directors.

Directors Appointed by Current CEO is used as a proxy for CEO power by Morse et al. (2011).

The director may feel an obligation to return the favor if he is appointed during the current CEO’s tenure. The “Year Service Began” variable from ISS is used to determine whether the director is appointed during the current CEO’s tenure. Because this database only contains the year that the director started his service, it is assumed that the starting date is the first of July of that particular year.

The moderating effect of CEO power is investigated by using two methods. First, the complete sample is divided into two groups to test the difference between the high and low power group. The second analysis considers the interaction effect between CEO power and non-GAAP adjustments levels on CEO compensation. CEOs have low power when their Power Index Score is in the range of 0-3. He is a high power CEO when this score is in the range of 4-7. The difference between the coefficients is tested to determine whether CEO power has a moderating effect. The subsampling analysis focuses on a linear and non-linear association by using equation (1) and equation (2) respectively.

Secondly, I am going to exploit all available information in the Power Index Score and interact this with non-GAAP adjustment levels. The following equation is constructed and tested by OLS regression:

Equation (3):

The relationship between the use of non-GAAP adjustments and CEO compensation is given by the sum of coefficients (β1 + β3*Power Index Score). This implies that β3 gives the effect

CEO Compensationt = β0 + β1Categoryt + β2Power Index Scoret

+ β3(Categoryt * Power Index Scoret)

+ β4ROAt + β5ROAt-1 + β6RETt

+ β7RETt-1 + β8Revenuet-1 + β9BTMt-1

(23)

of increases in CEO power on the relationship between non-GAAP adjustments and CEO

compensation.

Finally, I’m also going to interact the Power Index Score with all the individual non-GAAP adjustment categories with the following equation:

Equation (4):

The Libby boxes of the second hypothesis can be found in figure A2 of appendix A.

3.2.3 Control variables

To control for firm characteristics that affect CEO compensation, I included the following control variables which are also used by Curtis, Li, & Patrick (2017). ROA and ROAt-1 are used to control for earnings performance of the current and prior period. The next control variables address stock price performance of the current and the prior period, RET and RETt-1 respectively. Revenuet-1 is

included to control for size. As earlier stated, Gabaix & Landier (2008) find that the CEO equilibrium wage for CEOs of bigger firms is higher because firms can benefit more from the talent of the CEO. Smith & Watts (1992) provide evidence that CEO compensation is higher when the firm has more investment opportunities. This is because the marginal product of investment decisions is higher compared to good decision makers and supervisors. To control for this, the book-to-market ratio (BMTt-1) is included. Peters & Wagner (2014) state that CEOs of firms in volatile industries are more likely to be fired. As CEOs are risk averse, they are compensated for being exposed to this risk. They find a positive relation between CEO turnover and CEO compensation. To control for this, I use the industry average of the standard deviation

CEO Compensationt = β0 + β1Category1t + β2Category2t + β3Category3t

+ β4Category4t + β5Category5t

+ β6Power Index Scoret

+ β7(Category1t * Power Index Scoret)

+ β12ROAt + β13ROAt-1 + β14RETt

+ β15RETt-1 + β16Revenuet-1 + β17BTMt-1

(24)

of the stock price returns of the last five years, SDRET. Finally, the last control variable is Leverage t-1.

(25)

4 Empirical results

The findings of this study are presented in this section. First, the descriptive statistics are analyzed and discussed and subsequently the two hypotheses are tested. The process is similar for both hypotheses. The first analysis treats the adjustment category as a continuous variable to determine whether there is evidence for a linear relationship. The Second analysis focuses on this association in more detail. By testing the associations between the individual adjustment categories and CEO compensation I identify whether the association differs among adjustment magnitudes. The regression outputs in this section consist of five models. The first model in the regression output is without control variables. In model 2, the control variables are included to incorporate other determinants of CEO compensation. Model 3 controls for year effects and model 4 also for industry effects. Additionally, model 5 is a robust regression (r-reg) which controls for outliers as well as year and industry effects.

Descriptive statistics

4.1.1 Full sample

Table 1: Descriptive statistics (full sample)

Variable N Mean Std. Dev. 10% 25% 50% 75% 90%

Non-GAAP Adjustment 3,210 0.909 2.474 -0.285 0.140 0.450 1.070 2.395

Non-GAAP EPS 3,210 2.519 3.240 0.190 0.900 1.880 3.370 5.340

GAAP EPS 3,210 1.610 4.031 -1.055 0.250 1.320 2.790 4.940

CEO Compensation 3,210 8.305 1.011 6.921 7.673 8.417 9.029 9.506

Power Index Score 1,677 3.189 1.693 1.000 2.000 3.000 4.000 5.000

Play Slice of CEO 1,677 0.272 0.148 0.131 1.173 0.222 0.360 0.484

Duality 1,677 0.562 0.496 0.000 0.000 1.000 1.000 1.000

Triality 1,677 0.274 0.446 0.000 0.000 0.000 1.000 1.000

Tenure 1,677 8.684 7.283 1.667 3.417 6.750 11.500 18.417

Stock Ownership 1,677 0.015 0.041 0.000 0.001 0.003 0.009 0.031

Inside Directors 1,677 0.102 0.118 0.000 0.000 0.091 0.167 0.273

Directors Appointed by Current CEO 1,677 0.504 0.350 0.000 0.200 0.455 0.833 1.000

ROA 3,210 0.023 0.131 -0.054 0.008 0.037 0.070 0.114 ROAt-1 3,210 0.026 0.134 -0.046 0.010 0.040 0.073 0.119 RET 3,210 0.201 0.696 -0.346 -0.119 0.098 0.373 0.768 RETt-1 3,210 0.289 4.532 -0.380 -0.129 0.101 0.389 0.822 Revenuet-1 3,210 14.402 1.581 12.374 13.303 14.366 15.420 16.509 BTMt-1 3,210 0.576 0.628 0.157 0.273 0.465 0.747 1.090 Leveraget-1 3,210 0.876 8.854 0.000 0.085 0.386 0.875 1.777 Industry Volatility 3,210 1.922 5.092 0.402 0.487 0.858 1.764 4.762

(26)

The descriptive statistics are presented in table 1. The sample consists of 3,210 observations to determine whether the magnitude of non-GAAP adjustments affect CEO compensation. The mean of non-GAAP EPS is $2.519 and of GAAP EPS is $1.610. This means that the firms in our sample made on average an income increasing non-GAAP adjustment of $0.909 which equals to an increase of 36.1 percent. This increase is consistent with the opportunistic view of non-GAAP disclosure. The mean of total compensation earned by CEOs is 8.305 which corresponds to $6,264,000.

The sample of CEO power consists of 1,677 observations and is used to determine whether CEO power has a moderating effect on the relationship between the magnitude of non-GAAP adjustments and CEO compensation. The power index score in our sample has a mean of 3.189. The pay slice of the CEO is a 27.2 percent fraction of the total compensation earned by the top 4 executives in the firm which is interestingly a little above 25 percent. The average CEO’s tenure was 8.684 years, and the mean stock ownership was 1.5 percent. Additionally, the CEO appointed half of the people on the board of directors and 10.2 percent of all board members were previously affiliated with the firm. Lastly, the firms in the sample generated a mean revenue of 14.402 which is the natural log of $6,326,834,000. The main takeaway from the descriptive statistics is that the average of non-GAAP adjustments is positive. This means that companies made on average income-increasing non-GAAP adjustments, therefore non-GAAP EPS reflected a stronger performance than GAAP EPS.

4.1.2 Summary statistics per adjustment category

The observations are categorized into six groups to determine whether there is an association between the magnitude of non-GAAP adjustments and CEO compensation. Group 0 consists of firms which made income-decreasing adjustments and group 1 till 5 are sorted by magnitude of income-increasing adjustments where 1 has the lowest adjustments and 5 the biggest. Table B1 in the appendix presents the summary statistics per adjustment category. This preliminary analysis results in the following interesting observations. First, there are only 526 observations which made income-decreasing adjustments compared to 2,684 income-increasing observations. Guest et al. (2017) find a similar distribution and they interpret this as an argument against the informativeness objective of non-GAAP reporting. Second, CEO compensation is higher in category 5 than category 0. To be more specific, the mean in category 0 is higher than category 1 and 2, but the mean CEO compensation increases gradually from the categories 3 to 5. This could indicate that there is an association between the magnitude of non-GAAP adjustments and CEO compensation. Another remarkable observation is that both the means of non-GAAP EPS as

(27)

GAAP EPS are higher in category 0 than of category 1 till 4. Although category 5 has the highest mean non-GAAP EPS, the firms in that subsample made an average loss per share of $0.85. On average, the CEOs in this category converted a GAAP loss into a non-GAAP profit.

Hypothesis 1

This section estimates whether there is an association between non-GAAP adjustments in disclosures and CEO compensation. First, with equation (1) I investigate whether there is a linear relationship and subsequently, this relationship is explored in more detail to see whether this relationship differs across adjustment categories with equation (2).

4.2.1 Linear association between non-GAAP adjustments and CEO compensation

Table 2 displays the results of equation (1) which estimates whether there is a linear association between the adjustment categories and CEO compensation. The F-statistic is significant for all models. In addition, the adjusted R2 _{for all these models are sufficient}2_{. This leads to the conclusion}

that these models have explanatory power except for model 1.

According to the first model, CEO compensation and non-GAAP adjustments are positively associated with a significant level of 1 percent. Since there are more determinants of CEO compensation, this does not necessarily mean that a higher CEO compensation is attributable to non-GAAP adjustments alone. After including firm characteristics that influence CEO compensation in model 2, the category coefficient is 0.073 and significant at a 1 percent level. This means that the total compensation of the CEO increases by 7.3 percent per category compared to the reference group. For instance, if the CEO is in adjustment category 4, he receives compensation which is 29.2 percent (= 4 * 7.3%) higher compared to CEOs of firms with income-decreasing non-GAAP adjustments. After controlling for year effects in model 3, the category variable is still significant at the 1 percent level, but the coefficient decreased to 0.044 (4.4 percent pay increase per category advancement). Model 4 which also controls for industry effects finds that CEO compensation increases by 2.6 percent per category at a 1 percent significance level. Also, all control variables in this model are significant except for Leveraget-1 and RETt-1. Consistent

with our prediction ROA, ROAt-1, RET, and Revenuet-1 are positively associated with CEO

compensation. Finally, the category variable is also significant in the robust regression model 5 which controls for outliers.

(28)

To conclude, model 4 provides evidence of a 2.6 percent compensation increase per category advancement at a 1 percent significance level. This provides strong evidence that there is a linear association between the use of non-GAAP adjustments and CEO compensation. Additionally, all the other models have results consistent with model 4. These findings lead to supporting hypothesis 1.

Table 2: The linear association between non-GAAP adjustments and CEO Compensation

Y = Ln(CEO Compensation) r-reg Variable Prediction (1) (2) (3) (4) (5) Category (+) 0.067*** 0.073*** 0.044*** 0.026*** 0.032*** (0.000) (0.000) (0.000) (0.004) (0.000) ROA 0.620*** 0.472*** 0.369*** 0.240** (0.000) (0.000) (0.004) (0.034) ROAt-1 0.495*** 0.272** 0.220* 0.160 (0.000) (0.021) (0.055) (0.117) RET 0.018 0.065*** 0.079*** 0.164*** (0.433) (0.003) (0.000) (0.000) RETt-1 0.001 0.000 0.000 0.092*** (0.726) (0.982) (0.938) (0.000) Revenuet-1 0.333*** 0.313*** 0.342*** 0.370*** (0.000) (0.000) (0.000) (0.000) BTMt-1 -0.054** -0.064*** -0.084*** -0.186*** (0.029) (0.007) (0.001) (0.000) Leveraget-1 -0.003* -0.002 -0.001 -0.001 (0.100) (0.126) (0.347) (0.388) Industry Volatility -0.002 -0.002 -2.303** -2.284** (0.576) (0.427) (0.039) (0.021) Constant 8.136*** 3.332*** 2.714*** 5.920*** 5.542*** (0.000) (0.000) (0.000) (0.004) (0.002)

Controlled for year effects No No Yes Yes Yes

Controlled for industry effects No No No Yes Yes

Observations 3,210 3,210 3,210 3,210 3,207

R-squared 0.013 0.325 0.409 0.461 0.549

Adj.R2 0.013 0.323 0.405 0.447 0.538

F-test 41.96 171 105.1 33.47 48.78

***, **, * correspond to 1 percent, 5 percent, and 10 percent significance levels respectively

4.2.2 Nonlinear association between non-GAAP adjustments and CEO compensation

The subsequent analysis focuses on the associations between the individual adjustment categories and CEO compensation. Table 3 displays the results of this regression analysis. Both the adjusted

(29)

R2 _{and the F-statistic provides evidence that this model has explanatory power}3_{. After}

incorporating firm characteristics that influence CEO compensation in model 2, category 1, 3, 4, and 5 are significantly associated. To be more specific, CEOs of firms in the first category earn 18.5 percent less than the reference group of CEOs of firms which make income-decreasing adjustments. Moreover, CEOs of firms in category 3, 4, and 5 receive higher compensation than the reference group, 18.2, 19.7, and 23.7 percent respectively. CEO compensation decreases when the firm makes little non-GAAP adjustments, i.e., category 1 compared to the reference group. Conversely, CEO compensation increases at firms that make larger adjustments like the firms in category 3,4, and 5. The findings of model 2 suggest a nonlinear association between the use of non-GAAP adjustments and CEO compensation. After controlling for year effects in model 3, the betas coefficients are lower but still significant. Subsequently, when the model also controls for industry effects, CEO compensation is only affected by non-GAAP adjustments which magnitude lies in the range of either category 1 or 3. This means that CEOs of firms that made the smallest income-increasing adjustments in our sample earn 12.4 percent less than the reference group and that CEOs of category 3 receive 8.8 percent more compensation. The difference between the coefficients of category 1 and 3 is significant at the 1 percent level. Finally, when the model also controls for outliers, CEO compensation is only negatively associated with category 1 and positively associated with category 5.

To sum up, the results provide evidence of an association between CEO compensation and category 1 and 3. To be more specific, the compensation of category 1 CEOs are negatively associated with non-GAAP adjustments while this association is positive for category 3 CEOs. This provides evidence that it depends on the magnitude of the non-GAAP adjustments whether CEO compensation is positively or negatively associated with non-GAAP adjustments in disclosures. The results demonstrate that CEO compensation and non-GAAP adjustments are not always positively associated and therefore do not support hypothesis 1.

(30)

Table 3: The association between non-GAAP adjustment categories and CEO compensation Y = Ln(CEO Compensation) Y = Ln(Total Compensation) Y = Ln(Total Compensation) r-reg Variable Prediction (1) (2) (3) (4) (5) Category 1 (+) -0.353*** -0.185*** -0.148*** -0.124** -0.132*** (0.000) (0.000) (0.002) (0.013) (0.003) Category 2 (+) -0.167*** 0.010 -0.024 -0.010 -0.006 (0.007) (0.843) (0.619) (0.842) (0.900) Category 3 (+) 0.026 0.182*** 0.102** 0.088* 0.071 (0.671) (0.000) (0.036) (0.080) (0.113) Category 4 (+) 0.110* 0.197*** 0.117** 0.079 0.053 (0.073) (0.000) (0.016) (0.119) (0.239) Category 5 (+) 0.149** 0.237*** 0.122** 0.024 0.087* (0.015) (0.000) (0.019) (0.650) (0.068) ROA 0.460** 0.323 0.184 0.221* (0.027) (0.113) (0.370) (0.054) ROAt-1 0.463** 0.231 0.209 0.180* (0.013) (0.207) (0.252) (0.078) RET 0.018 0.071** 0.086** 0.173*** (0.581) (0.039) (0.014) (0.000) RETt-1 0.002 0.001 0.001 0.092*** (0.706) (0.885) (0.842) (0.000) Revenuet-1 0.320*** 0.302*** 0.323*** 0.369*** (0.000) (0.000) (0.000) (0.000) BTMt-1 -0.030 -0.033 -0.066* -0.196*** (0.409) (0.358) (0.088) (0.000) Leveraget-1 -0.002 -0.002 -0.001 -0.001 (0.379) (0.445) (0.712) (0.337) Industry Volatility -0.002 -0.002 -2.389 -2.172** (0.713) (0.662) (0.178) (0.028) Constant 8.324*** 12.724*** 11.956*** 15.547*** 5.485*** (0.000) (0.000) (0.000) (0.000) (0.002)

Controlled for year-effects No No Yes Yes Yes

Controlled for industry effects No No No Yes Yes

Observations 3,210 3,210 3,210 3,210 3,209

R-squared 0.029 0.331 0.413 0.464 0.552

Adj.R2 0.028 0.329 0.408 0.450 0.540

F-test 19.38 121.8 89.55 32.22 45.87

(31)

4.2.3 Conclusion hypothesis 1

In this section, two analyses are performed to determine whether non-GAAP adjustments are positively associated with CEO compensation. First, there is strong evidence of a linear relationship of 2.6 percent per category advancement at the 1 percent significance level. Subsequently, a more thorough analysis is conducted to see whether this relationship could be attributed to individual categories. The results found evidence that only category 1 and 3 are associated with CEO compensation. To be specific, a positive association for category 3 and a negative association with category 1. This latter finding is not consistent with the positive association hypothesis, and therefore it provides no support for hypothesis 1.

Hypothesis 2

This section estimates the moderating effect of CEO power on the relationship between the use of non-GAAP adjustments and CEO compensation to test hypothesis 2. This is done by both subsampling and with interacting the non-GAAP adjustment category with the power index score. Just as hypothesis 1, first the linear relationship is examined. Subsequent analysis will focus on the moderating effect of CEO power on the different adjustment categories. The only models included in this section are controlled for year and industry effects.

4.3.1 Moderating effect of CEO power on the linear association

In the first analysis, the observations are divided into two groups based on their power index score, which incorporates seven proxies for CEO power. Table 4 compares the results of the regression analysis between the two subsamples after controlling for year and industry effects. The subsample of CEOs with high power has a 1 percent significant beta coefficient of 0.055, which means that each advancement in a non-GAAP adjustment category results in a 5.5 percent increase in compensation. The category variable in the low power subsample is 0.027, which equals to a 2.7 percent increase in category advancement. Although there is a difference between the high and low power group of 1.8 percent, the “suest” function of Stata didn’t provide evidence that this difference is significant.

To check for robustness of these results, I performed an additional test by dividing the sample into a high, medium and low power group. The result is consistent with table 4 and also find a higher beta coefficient for the high power group compared to the low, group but again, this difference is not significant. The results of the subsampling analysis provide no evidence that CEO power has a moderating effect.

(32)

Table 4: The moderating effect of CEO power on the linear association

Y = Ln(CEO Compensation)

CEO Power Interaction

Variable High Low Complete sample

Category 0.055*** 0.027* -0.019

(0.006) (0.060) (0.409)

Power Index Score -0.060***

(0.001)

Category * Power Index Score 0.017***

(0.006)

Controls Yes Yes Yes

Controlled for year and industry effects Yes Yes Yes

Observations 729 948 1,677

R-squared 0.556 0.618 0.517

Adj.R2 0.515 0.586 0.494

F-test 13.47 19.37 22.49

The subsequent analysis focuses on whether there is an interaction effect between category and the power index score to exploit all available information. As displayed in table 4, there is a significant interaction term between category and the power index score. The interaction coefficient is 0.017 at a 1 percent significance level. This means that with every increase of the product of category and power index score, CEO compensation increases by 1.7 percent. Important to note is that including the interaction term in the model leads to an insignificant category coefficient which was significant in the model without interaction. The result of the model shows that non-GAAP adjustment magnitudes only affect CEO compensation when these are combined with CEO power. This provides evidence that CEO power influences the effect which the category variable has on CEO compensation which supports hypothesis 2.

4.3.2 Moderating effect of CEO power on the nonlinear association

Consistent with our analysis of hypothesis one, the second regression will focus on the association between CEO compensation and the individual adjustment categories, but this time including the moderating effect of CEO power. Table 5 compares the adjustment categories of the high and low power group after controlling for year and industry effects. The results of the subsample analysis lead to two observations, one in each subsample. First, in the high CEO power group, there is only an association between the fifth category CEO compensation. This significant beta coefficient provides evidence that high power CEOs of firms that make the largest income-increasing adjustments receive 20 percent more compensation than the reference group. The compensation

(33)

of low power CEOs is only associated with non-GAAP adjustments when these adjustments fall into the smallest category. When this is the case, CEO compensation is 16.8 percent lower than the reference group. This table leads to two conclusions. First, the use of non-GAAP adjustments is positively associated with the compensation of high power CEOs but only when these adjustments have high magnitudes. Second, the use of non-GAAP adjustments is negatively associated with the compensation of low power CEOs, but only when these adjustments are small and fall in the first category. These results provide evidence that powerful CEOs can benefit from non-GAAP adjustments while low power CEOs can only “suffer” from non-GAAP adjustments depending on the magnitude. These findings are consistent with the moderating effect of CEO power and therefore support hypothesis 2.

The interaction analysis demonstrates that there are no direct associations between the individual adjustment categories and CEO compensation. However, there are two significant interaction effects in this model. Category 2 has a negative interaction coefficient of 0.059 at the 10 percent level, and category 5 has an interaction coefficient of 0.094 at 5 percent level. These findings provide evidence that the moderating effect of CEO power is positive when firms make the largest adjustments and have a negative moderating effect on firms that make smaller adjustments that lie within the range of category 2. Overall, these results also provide evidence of the moderating effect of CEO power and therefore supports hypothesis 2.

(34)

Table 5: The moderating effect of CEO power on the individual adjustment categories

Y = Ln(CEO Compensation)

CEO Power Interaction

Variable High Low Complete sample

Category 1 -0.056 -0.168** -0.121 (0.607) (0.034) (0.372) Category 2 -0.116 -0.039 0.152 (0.276) (0.618) (0.255) Category 3 0.157 0.083 0.070 (0.139) (0.277) (0.582) Category 4 0.127 0.058 0.019 (0.239) (0.457) (0.883) Category 5 0.200* 0.011 -0.222 (0.095) (0.898) (0.105)

Power Index Score -0.025

(0.346)

Category 1 * Power Index Score 0.002

(0.959)

Category 2 * Power Index Score -0.059*

(0.094)

(0.709)

(0.511)

Category 5 * Power Index Score 0.094**

(0.013)

Controls Yes Yes Yes

Controlled for year and industry effects Yes Yes Yes

Observations 729 948 1,677

R-squared 0.506 0.585 0.523

Adj.R2 0.457 0.548 0.498

F-test 10.27 15.91 20.82

4.3.3 Conclusion hypothesis 2

In this section, the moderating effect of CEO power is examined. These results provide strong evidence for a moderating effect of CEO power. The findings suggest that CEO power is a positive moderator when a firm makes large non-GAAP adjustments and that this is negative when the firm makes small adjustments. The subsampling analysis provides evidence of two associations, one in each group. The compensation of powerful CEOs is positively associated with category 5 adjustments, while the compensation of low power CEOs is only negatively associated

Non-GAAP earnings in disclosures : the effect on CEO compensation