Does the fraction of female directors on the board impact CEO compensation?

(1)

“Does the Fraction of

Female Directors on the Board Impact

CEO Compensation?”

M.Sc. Thesis

Executive Programme in Management Studies

Strategy Track

Author Supervisor

Dr. Benjamin Bruhn Dr. Daniel Wäger

10908137 Universiteit van Amsterdam

Faculteit Economie en Bedrijfskunde Plantage Muidergracht 12

Submission 1018 TV Amsterdam

(2)

Statement of Originality

This document was written by Dr. Benjamin Bruhn, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are 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)

List of Tables

Table 1: Correlation matrix for the data set concerning the dependent variable CEO salary. ... 38 Table 2: Regression results for salary as dependent variable. The three models are different steps in a hierarchical regression. Each NAICS group contains at least 20 firms, with a total of 844 firms for all 17 groups. ... 40 Table 3: Correlation matrix for the data set concerning the dependent variable CEO bonus. ... 41 Table 4: Regression results for bonus as dependent variable. The three models are different steps in a hierarchical regression. Only one NAICS group (522) had more than twenty (29) members. ... 42 Table 5: Correlation matrix for the data set concerning the dependent variable CEO stock awards. . 43 Table 6: Regression results for stock awards as dependent variable. The three models are different steps in a hierarchical regression. Each NAICS group contains at least 20 firms, with a total of 712 firms for all 15 groups. ... 44 Table 7: Correlation matrix for the data set concerning the dependent variable CEO option awards. 45 Table 8: Regression results for option awards as dependent variable. The three models are different steps in a hierarchical regression. Each NAICS group contains at least 20 firms, with a total of 235 firms for all 4 groups. ... 46 Table 9: Correlation matrix for the data set concerning the dependent variable CEO pension change. ... 47 Table 10: Regression results for pension change as dependent variable. The three models are

different steps in a hierarchical regression. Each NAICS group contains at least 20 firms, with a total of 184 firms for all 5 groups. ... 48

(5)

List of Figures

Figure 1: Simple model representation of the two hypotheses... 23 Figure 2: Total annual CEO compensation vs. ratio of CEO compensation and average worker

compensation. The data covers a large sample of SEC-registered US companies for the year 2013. The red line indicates a linear relationship between the dependent and independent variables. ... 25 Figure 3: Development of five different components of the total annual CEO compensation in recent years. Data is shown for US companies listed in the Russell 3000® index in year 2015, if available in the Capital IQ database. Grey triangles represent the top 1% and bottom 1% limit, black circles the mean of a log-normal fit. The distributions suffer from noise and a log-normal fit does not always seem suitable, so this graph is only to be understood as a measure of a trend. The analysis here disregards the composition of compensation packages, showing an analysis for the distribution of each individual component (disregarding that e.g. the salary might be smaller in a company where stock awards are very high). ... 27 Figure 4: (left): Nestled histograms of the four components salary (each vertical square represents a 10% increase), pension change (each horizontal square represents a 10% increase), stock awards (each scale bar in vertical direction inside a square represents a 10% increase) and option awards (each scale bar in horizontal direction inside a square represents a 10% increase). The color

represents the occurrence of the given combination, while the dashed lines indicate possible values for stock and options for a given combination of salary and pension change. (right): Complementary table displaying the top 50% of combinations. The top three rows represent 25% of all occurrences. ... 28 Figure 5: EBIT vs. revenue with three NAICS groups indicated by different colors. The dashed black line indicates a scaling factor of unity, lines parallel to it represent linear relationships with a different scaling factor. ... 29 Figure 6: Schematic representation of the model used in the statistical analysis. Relationships

between the control and dependent variables have been combined into one single arrow for clarity. ... 34 Figure 7: Schematic drawing of relations between the independent variables. Average correlation coefficients for all analyses presented in the results section are shown in blue, if potentially

significant. ... 49 Figure 8: Heat map histogram of number of female directors on the board versus total number of directors on the board, with the number of companies exhibiting a specific combination being

encoded by color. ... 50 Figure 9: Histogram of the fraction of female directors on corporate boards of several hundred firms. A reduction at zero goes along with an increase of higher fractions, even if the changes are not major. The inset shows the mean of a log-normal fit to the histogram data for the years 2007 through 2014, with an increase from 14% to 16%. ... 52

(6)

Abstract

Since the 1980s, executive compensation has risen dramatically, which has both caused a great number of public outrages and spawned a whole field of research exploring the development and structure of CEO compensation. Numerous aspects of the underlying ideologies, the reasons for the continuous increase and possible routes for more reasonable pay setting processes have since then been investigated. Due to its monitoring role and its direct influence on the remuneration of the chief executive officer, the board of directors has received a lot of attention, its characteristics and its power relative to that of the CEO being focal points of executive compensation research. One of the factors impacting this power is diversity, as it is believed to improve corporate governance. The rise of women into positions of power increases gender-diversity, and while women are still underrepresented in high tier corporate positions, the situation is slowly but noticeably changing. Due to gender-specific trait differences, the presence of female directors on corporate boards might lead to changes in the structure and level of CEO compensation. This thesis aims at shedding light on the so far unexplored effect the fraction of female directors on corporate boards have on different components of the CEO compensation package, uncovering a rather small, but nevertheless significant relationship between the fraction of female directors on the board and annual bonus payments. Results from a correlation analysis of the independent variables also indicate that large firms tend to deploy larger boards and exhibit higher CEO turnover than small firms and that the fraction of female directors increases superlinearly with board size.

(7)

Introduction

Not long after Donald Trump was elected as the 45th president of the United States of America, Rex Wayne Tillerson, CEO of Exxon Mobile since 2006, was nominated as secretary of state. In order to comply with conflict-of-interest requirements, he stepped back as CEO of Exxon, forfeiting a position earning him $ 30 Mio. per year, however, not without agreeing to a severance package worth $ 180 Mio. News coverage was extensive and not everybody was particularly enthusiastic about these figures.

In a recent study, Kiatpongsan and Norton (2014) asked individuals in different countries and with different socioeconomic backgrounds to estimate the actual ratio between CEO and average worker compensation, as well as state a ratio they would consider fair. It turned out that throughout the whole sample, irrespective of which peer group they belonged to, most people shared a common notion of an ideal ratio (7:1 in the USA). The vast majority guessed that the ratio was considerably higher than that (30:1), but still underestimated the actual ratio (354:1) by far.

Perhaps this can explain why news items like the one about Tillerson more often than not go along with a public outcry condemning the excessive compensation packages of managers. And this example is just one of many. There is a possibility that if the distribution of wealth is not kept in check to some degree, the stability of society might be endangered, as history has taught mankind on numerous occasions. However, requests of the public to reduce or cap executive compensation have not had any noticeable effect. Recently, Norton (2014) noted that income inequality has steadily risen since the 1980s to levels last seen in the first third of the 20th century (after the Second World War and the subsequent years of general economic growth had decreased it considerably).

This steady rise, probably in conjunction with attention by the general public and media, has lead an increasing number of researchers to investigate the underlying reasons of the structure and development of executive compensation.

(8)

On the one hand, economic reasoning is used to explain increasing remuneration levels, for example a shift in the composition of compensation packages, growing markets, increasing firm sizes, and globalization. On the other hand, there are numerous theories exploring the trend from a corporate governance, cultural aspects and social interaction point of view. More details on these and other considerations will be provided in the Literature Review section of this thesis.

Some of the results could possibly be used to derive implementable measures that – among other effects – might serve to restore a balance of market movements and pay adjustments, as well as couple executive achievements and rewards more tightly. Besides high top income tax levels, solutions aiming at social and psychological mechanisms seem to be a promising route. In addition, high quality corporate governance is recognized by many researchers as a valuable instrument for both increasing the firm performance and avoiding excessive compensation practices.

Diversity is nowadays recognized as one of the factors strengthening the board of directors and rendering it more active. Many big organizations deploy programs for increasing diversity both in their workforce and in their executive ranks. Since this trend is rather recent, the history of research on diversity effects on firms is not very long and many aspects are still unexplored.

This thesis aims to shed light on one of these unknowns. After providing an overview of the most recognized theories dealing with executive compensation, an argument shall be made for investigating the role of female directors in CEO pay practices. After formulating testable hypotheses, the most relevant control variables will be identified using previous literature on CEO compensation. Finally, a statistical analysis will be performed and the results discussed.

(9)

Literature Review

Before delving deeper into considerations about executive compensation, a short detour shall provide an understanding of the importance of those. Fairness regarding pay is not simply an academic exercise, but can have real and profound effects on employees. Akerlof and Yellen (1990) argue that individual effort is reduced if the wage is regarded as unfair, which is supported by Norton (2014) mentioning that while higher salaries might well attract greater talent and increase the willingness to perform better, too large differences can have negative repercussions. Wade, O’Reilly and Pollock (2006) note that social comparison can have an impact on the perception of fairness with regards to compensation and that comparative judgments shape an individual's reactions to rewards and punishments. Citing numerous other works, they identify lower productivity at both the organizational and individual levels, loss of group cohesion, theft, deterioration of delivered quality, and increased turnover as possible negative consequences of perceived inequity. In a Harvard Business Review article (2009), John Mackey, CEO of Whole Foods, notes that in the face of increasing inequality “employee morale is suffering, talented performers’ loyalty is evaporating and strategy and execution is suffering at American companies”. There are indeed numerous studies on team production settings that provide evidence for a deterioration of effort and therefore output when some team members are better paid than others. Cornelißen, Himmler and Koenig (2010) find in a study of German firms that even perceived unfairness – regardless of its actual existence – of CEO pay leads to a deterioration of work morale and what they coin ‘hidden protest’, potentially bearing high economic costs. Cohn, Fehr and Goette (2015) conducted an experiment where workers were paired and one of them experienced a pay cut, upon which that worker reduced effort, while the other worker was not affected at all. A similar finding is presented by Card, Mas, Moretti and Saez (2012), who state that lower paid workers report less job satisfaction when pay inequality is made public, but higher paid workers do not experience any benefit. Just to name a few of the many other contributions to this topic, Pfeffer and Langton (1993),

(10)

Finkelstein and Hambrick (1996) and Bloom and Michel (2002) associate increased dispersion with lower productivity, less cooperation, and increased turnover.

A vast body of literature exists on the topic of fairness considerations, however, not all of them agree with respect to their observations. In order to explain the discrepancies in the conclusions of different researchers, an important distinction might have to be made between pay inequality and pay inequity (see, e.g., Leventhal (1976) and Steers and Porter (1983)). While the former term describes unequal pay as an absolute measure, the latter term is more specific, as it describes unequal ratios of reward to input. It is highly likely that only pay inequity prompts negative reactions in a team production setting (see, e.g. Ambrose and Kulik (1999), Deutsch (1985), Heneman, Judge and Kammeyer-Mueller (2014) and Leventhal (1976)). Trevor, Reilly and Gerhart (2012) build upon this idea, splitting pay dispersion into a fraction that is tied to productive input and a residual that remains unexplained by productivity-related inputs like, for example, political tactics, randomness or discrimination. Citing numerous other works, they describe how in interdependent contexts pay dispersion is particularly detrimental to performance. When individual contributions can be assessed, then less pay dispersion can cause stronger inequity perceptions, given similar inputs (cf. Wade, O’Reilly and Pollock (2006)). A complementary view from a different angle is provided by Ariely, Gneezy, Loewenstein and Mazar (2009), who look at the effects of large incentives on the ones receiving them. Apparently, the opportunity to receive huge bonuses, of course coming along with high expectations of success, makes the recipients perform worse than when trying to obtain a more reasonable bonus.

In summary, it seems that large pay differences are related to a number of negative consequences regarding motivation, performance, loyalty and morale. Therefore, it seems sensible to keep an eye on compensation and prevent excessive deviations. And as Mackey puts it in his previously mentioned HBR article (2009): Above a certain pay level, according to Maslow’s Hierarchy of Needs, other factors than salary are key to retaining skilled employees at the top. Despite this wisdom, however, CEO compensation is staggeringly high

(11)

and has increased steadily since decades. The following paragraphs will attempt to explain at least part of this phenomenon.

A well-known and vastly cited explanation CEO compensation being so much higher than the remuneration of lower ranks is provided by Lazear and Rosen (1981), who argue that ranked wages based on relative productivity increase the equilibrium effort employees spend. In such a system, compensation is regarded as a prize in a contest. All employees assume mutually exclusive positions on a ranking ladder, with a higher rank being associated with a considerably higher salary. By performing better than a higher ranking peer, an employee can frog-leap steps on the ranking ladder, therefore increasing his or her reward considerably. A positive relationship between wage dispersion and productivity is expected, offering an explanation for the high ratio of executive and average worker compensation. Alchian and Demsetz (1972) also noted that when monitoring costs are so high that an input-based compensation scheme yields high risks for moral hazard (costly shirking), then output-based pay, where workers receive compensation output-based on their rank order, may be the best solution, as both cost of measurement and bearing of risk are being reduced. Another concept bearing great similarities to sports, music and fine arts is presented by Rosen (1981) in order to make sense of excessively high CEO compensation. He develops the idea of a superstar phenomenon. Where many talented high-performers exist, even the slightest superiority in talent leads to huge reward differences. Rosen explains the phenomenon by superlinearly increasing demand for talent, where people are willing to pay a much greater amount for being able to enjoy one single outstanding singing or surgical performance than for attending several cheaper, but mediocre performances. One firm typically only employs one CEO, thus large firms with considerable financial resources and high monetary stakes may just be willing to invest heavily in the highest management talent in an increasingly open and competitive market. This reasoning regarding increasing CEO compensation is supported by Gabaix and Landier (2008), who find a simultaneous sixfold increase of CEO

(12)

pay and firm size in the US between 1980 and 2003 and provide a calibrated model that extends the Rosen’s superstar economics.

Other CEOs in the lower part of the peer group income distribution, however, are not contempt with their situation for long, as Gabaix and Landier (2008) note. They describe a phenomenon referred to as contagion, where a salary increase of a fraction of the larger CEO peer group inevitably leads to a raise in compensation for the whole peer group, given a certain time lag. The contagion model was able to explain a major fraction of the roughly 30% compensation increase observed by the authors, thus providing a valuable addition to the hypothesis by Bertrand and Mullainathan (1999) that CEOs of firms without strong principals respond to increased takeover protection by skimming additional compensation, as Bereskin and Cicero (2013) point out. Contagion is without doubt intimately related to social comparison, a topic explored much earlier in a different branch of research by Festinger (1954). He discusses the individuals’ drive for self-evaluation, which is best achieved by comparison to others, preferably those in the near vicinity with respect to social rank. This benchmarking with peers can, according to Aspinwall and Taylor (1993), increase motivation and certainly increases the level of compensation within the CEO peer group. As might be expected, the increase is largest for those ranking lowest, as they feel the most urgent need to improve. O’Reilly, Main and Crystal (1988) support this argument by presenting strong associations between compensation of CEOs and outside members of the board of directors, who are in turn more often than not themselves CEOs of other firms. Faulkender and Yang (2010) also observe that highly paid peers are used as comparison in matters of pay, even after controlling for industry, firm size, CEO responsibility and other factors. They find that the effect is stronger for small peer groups, for CEOs who are simultaneously serving as chairman of the board, for CEOs with higher tenure and in firms where a significant fraction of the directors are also serving on other boards. Ang, Nagel and Yang (2014) even go one step further by postulating that social pressure is not solely applied by job-related peers, but also by other wealthy people who executives are in contact with. In their study, they

(13)

investigate social premiums – pay premiums related to social pressure – and, after controlling for other determinants such as firm characteristics, performance and governance, confirm the existence of such. They also note that these premiums scale with the frequency of relevant social interactions and the physical distance to relevant peers.

A rather material explanation for the sharply rising CEO compensation since the 1980s is provided by Hall and Liebman (1998), among several others, who attribute it to increasing stock option grants in executive compensation packages. Indeed, nowadays stock grants represent the largest portion of total compensation on average, as will be shown with data later (cf. Figure 4). This trend might be a measure aiming at performance increases and monitoring cost reductions advocated by agency theory – probably the most frequently used theoretical framework until recently in executive compensation research. Succeeding Berle and Means (1932), who discussed the separation of ownership and control in modern corporations, Jensen and Meckling (1976) formalized the theory, which postulates a natural misalignment of the interests of owners (i.e., stockholders) and managers, creating the need for monitoring in order to avoid agency costs in the form of shirking. Monitoring has long been regarded as the instrument of choice, if not the only viable tool, for achieving better alignment between principal and agent and thus reduce agency costs. However, as Tosi and Gomez-Mejia (1994) note: “Given the difficulty of directly observing an executive's effort or behavior, monitoring primarily occurs through the use of pay practices that align the interests of agents and principals”. In other words, if the agent (the CEO) is rewarded generously when acting in the best interest of the principal (the shareholder), then a natural incentive to do exactly that exists, mitigating the need for monitoring. Thus, high performance-related pay for the CEO can be understood as a substitute for high monitoring costs. This view opens up the possibility for a second role of the board promoted by, among others, Pfeffer and Salancik (1978) and Westphal (1999): That of a counselor and provider of advice and strategic help to the CEO. Lack of trust and collaboration, as O’Reilly and Main (2010) suggest, may lead to information asymmetry and political struggles, undermining joint efforts

(14)

to tackle strategic challenges facing the firm. They note that the cost of increased compensation might in fact mitigate these negative effects of overemphasized monitoring activities. Westphal (1999) argues that collaboration between the CEO and a board of directors that has enough time and experience at hand can possibly compensate the additional pay expenses that may originate from close social ties between these two parties. Despite all of these arguments hinting at a correlation between CEO pay and firm performance, the results found in the literature are rather mixed (cf. meta analyses of, for example, Daily, Johnson, Ellstrand and Dalton (1998) and Deutsch (2005)). It might still be worth to use firm performance as a control in statistical analyses, as can often be seen in the literature in the form of return on equity, return on assets or market to book value.

Another economic effect no individual can exert any meaningful influence on, but that nevertheless impacts CEO compensation strongly, is mentioned in Beasley’s Research Handbook on Executive Pay (2012). Macroeconomic factors represent the state of the broader market, and when the market is thriving, shares prices rise, which both leads to generous raises and an increasing value of stocks granted as part of the executive compensation package. The authors estimate that without the influence of macroeconomic factors CEO pay levels would have risen by 27% instead of the actual 58% between the years 2001 and 2007. Compared to pay level increases in Bull markets, pay cuts in Bear markets, are far less pronounced, explaining the monotonous increase of CEO compensation despite the occurrence of several economic downturns within the last two decades. Especially firms without a large institutional investor base do not tend to change pay practices in the face of inferior corporate performance, according to Bell and van Reenen (2013). One macroeconomic factor that can be steered actively by politics happens to be the one with the probably most profound effect on executive pay levels: The top marginal tax rate. Alvaredo, Atkinson, Piketty and Saez (2013), four of the five executive committee members of the World Wealth and Income Database, note that the top 1% fiscal income share in the United States of America increased from below 10% in the late 1970s to over

(15)

20% to date. They observe a similar trend for a group of other countries (e.g. UK, Australia), which seem to have one important factor in common that is different in technologically similar countries (e.g. Germany, France and Sweden) not having experienced such a sharp rise: The slashing of top income tax rates. They argue that with high top marginal tax rates the net increase of compensation is not worth the effort of extensive bargaining, but that the executives’ resources are rather spent on improving the firm’s efficiency, cutting unnecessary expenses and concentrate on growth. When these taxes are cut, on the other hand, more of the executives’ energy might be spent on trying to increase their own compensation, as the effort comes with a high financial reward.

Numerous ways for CEOs to do that have been identified in the literature, many of which focus on social influence and the accumulation of power. Perhaps the most effective and enabling measure is the simultaneous occupation of two essential positions in the management – chief executive officer and chair of the board – most often referred to as duality or interlock. Many authors (e.g. Sauerwald, Lin and Peng (2014), Elhagrasey, Harrison and Buchholz (1999) and David, Kochhar and Levitas (1998)) report elevated CEO pay levels in firms exhibiting duality. In such a constellation, the board is likely to experience difficulties in fulfilling the primary task assigned to it by agency theory, namely thorough monitoring and consequently the efficient reduction of agency costs. The CEO, on the other hand, can exert direct influence on, for example, compensation decisions and the appointments of new directors. Those appointed by the CEO might be subject to reciprocity – a felt obligation and executed process of doing each other favors – which, according to O’Reilly and Main (2010), yields unreasonable pay raises. More examples are provided by Lorsch and MacIver (1989) and Westphal (1998). Westphal (1999) adds that the action of the CEO appointing board members may render the board more passive, caused by closely related factors such as friendship, financial rewards and prestige. A similar effect might occur when directors are paid generously for their services, see, for example Brick, Palmon and Wald (2006). As Crystal (1991) notes: “Whenever you find highly paid CEOs, you will find

(16)

highly paid directors. It’s no accident.” Demographic similarity seems to add to these effects, as Westphal and Zajac (1995) show. In their study, board members were found to favor demographically similar new appointments and demographic similarity also increased the executives’ cash compensation. Tsui and O’Reilly (1989) argue that demographic similarity increases social influence, which is in turn generally understood as having an effect on the compensation level. Social ties and the influence a CEO gains on the environment in general should further strengthen with time served in that position. It is therefore not surprising that a number of authors (see, for example, Sauerwald, Lin and Peng (2014), Elhagrasey, Harrison and Buchholz (1999) and Hill and Phan (1991)) report elevated CEO compensation levels for higher tenure. Internal networks can be nurtured and a greater sense of dependence created. This argument goes along well with the preferential treatment in form of lower pay-performance sensitivity towards the CEO by insider members of the compensation committee Newman and Mozes (1999) report. Outside directors are generally viewed as more independent from the CEO, and thus more capable of enforcing high quality governance. Since the US stock exchanges issued requirements for board members to be independent in the years following the economic downturn in the year 2001, it has become more difficult for CEOs to grow their power by appointing subordinated internal employees. Chhaochharia and Grinstein (2009) conducted a study entailing firms that were strongly and weakly affected by these requirements. They report a significant decrease of roughly 17% in CEO pay for firms that were subject to changes following the introduction of those requirements. This effect was stronger for firms without outside block holders and with few institutional investors, as these have the general tendency to control the management’s actions more tightly, anyway. It should be noted at this point that Guthrie, Sokolowsky and Wan (2012) commented on this staggering decrease, attributing a major fraction of it to outliers that were not removed before the regression analysis. Part of the effect remains, though, and is reinforced by other authors like Beatty and Zajac (1994).

(17)

Not only internal networks prove to be beneficial for CEOs, but also connections to external nodes. Geletkanycz, Boyd and Finkelstein (2001) find a positive correlation between external network connections and CEO compensation. Apparently, organizations – especially highly diversified ones being exposed to greater external constraints – seem to value connections they see fit to potentially provide access to resources, knowledge and other advantages. Additionally, being a recognized member of the corporate elite brings about advantages similar to those of demographic similarity and connected to hegemony theory. Members of the same group, in this case CEOs and directors who are or have themselves been managers and CEOs, tend to treat each other favorably and all protect the group’s interests. Bilimoria (1997) uses this argument to show how CEO pay and the use of non-stock market performance criteria increase when an organization assumes a key position in a network of interlocked firms, where the same individuals or at least members of the same elite occupy positions within several organizations. Similarly, Fich and White (2003) find higher levels of executive compensation and lower turnover rates in firms where one or more pairs of board members are also active in the boards of other firms.

CEO power is not an absolute measure, but only meaningful when compared to the power of the board of directors. Well working, powerful boards can enforce high quality corporate governance and monitor the CEO effectively, but also provide valuable strategic support. Four classes of boards are identified by Pearce and Zahra (1991), with board power and CEO power as differentiators, but only two of them seem to add value in the economic sense. Proactive boards, representing true corporate governance, are strong, independent boards wielding more power than the CEO. They are primarily comprised of outside directors. Participative boards share the same level of power as the CEO and resolve differences by debate and vote and typically consist of a majority of outside directors. Reforms of corporate boards are often aiming at increasing the relative power of the board with respect to the CEO, as powerful boards are generally associated with organizational effectiveness. There are, according to Pearce and Zahra (1991), four underlying reasons

(18)

justifying this assessment. Firstly, they provide useful business contacts, strengthening the link between the firm and its environment. Secondly, they actively contribute to the development to the organization’s mission and goals. Thirdly, they assume a monitoring role, taking action in case the company performance does not fulfill expectations. And last, but not least, they play a crucial role in creating corporate identity.

Gompers, Ishii and Metrick (2003) found that firms scoring a low G-index, representing high shareholder power, are more valuable and profitable and have higher sales growth and lower capital expenditures than firms with a high G-index. These results were in part confirmed by Gabaix and Landier (2008), who report a positive relationship between G-index and CEO pay. Numerous other authors (e.g. Tosi and Gomez-Mejia (1994)) similarly observed that higher board control was associated with lower CEO compensation packages. Arguing in the same direction as Gompers et al., Pearce and Zahra (1991) note that dispersion of ownership causes a ‘widespread lack of accountability of directors and executives alike, resulting in a near-universal decline of the role of board’. The presence of institutional investors owning significant portions of the firm’s stock, on the other hand, tends to lower the level of CEO compensation and increase the portion of long-term incentives within the compensation package, according to David, Kochhar and Levitas (1998), except when the investors are dependent on the firm. Similar findings were presented by Ozkan (2007), Hartzell and Starks (2002) and Sauerwald, Lin and Peng (2014). Other factors than block holders seem to impact the accountability and effectiveness of the board, though. Yermack (1996) and other works cited therein state that small boards work more effectively than large ones, improving financial performance and operating efficiency measures, enhancing the firm value and increasing the pay-performance sensitivity of CEO compensation. It seems reasonable that smaller boards are more engaged, feel more responsible and are able to act faster and more cohesively. An explicit positive relation between board size and CEO pay level can be found in a study on UK companies by Ozkan (2007).

(19)

A few studies investigate the impact of diversity on firm performance. O’Reilly and Main (2010), Dobbin and Jung (2011) and Herring (2009), for example, claim to observe increased firm performance as a result of greater gender diversity. Erhardt, Werbel and Shrader (2003) find a positive relationship between diversity and effectiveness in the oversight function of boards of directors, as diversity generates more conflicts, which spark more discussions, eventually leading to better performance. Since gender equality in the workplace is a rather recent trend, though, research focusing on the effects of diversity on performance, pay and other key figures has just begun. Probably, due to the current scarcity of well distributed data, this new field of research will only grow and improve significantly throughout the next decades. Questions opening up with more and more females overcoming the obstacle known as the glass ceiling are obviously: Do women, considering gender-specific traits, have a different management style? Will they change the way things are done at the top? And will they impact those around them? The next section will explore this topic in some more detail.

(20)

Hypotheses Development

Recently, the so-called glass ceiling has become more brittle and slowly but surely the number of females in corporate top positions is increasing. Large international corporations are promoting inclusiveness and diversity and take measures to put these ideals into practice. In many countries, those aspirations have climaxed into strict quotas, headed by Norway having passed a law in the year 2003 that requires 40% of all board members to be women. Partly this development is due to social and political pressures, as described by Farrell and Hersch (2005) and O’Reilly and Main (2010) and references therein. However, there are also numerous other reasons, most of which are subject to heated debates. The opposition of this trend is emphasizing negative impacts on economic figures, like increased labor costs and reduced short-term profits reported by Matsa and Miller (2013) or the average negative effect of gender diversity on firm performance claimed by Bøhren and Strøm (2010). It should be noted, though, that these studies were performed using Norwegian firms affected by the new quota requirement, so the described issues are more likely to be caused by side-effects of the quota than the presence of women, namely the promotion of younger and less experienced individuals as board members, as described by Ahern and Dittmar (2012). In addition, changing the focus from shareholders to all stakeholders might be able to mitigate some of the criticism. McGuiness, Vieito and Wang (2017), as well as numerous works cited within their paper, observe a positive correlation between gender balance in top management and CSR-performance, which is a long-term strategic advantage that should not be overlooked. Some studies even directly contradict the above-mentioned negative results of women on the board. O’Reilly and Main (2010), Dobbin and Jung (2011) and Herring (2009), for example, report increased firm performance as a result of greater gender diversity. Other optimistic views present advantages like access to larger talent pools, infusion of new ideas and first hand accessibility to the thinking of female consumers (cf. Useem (1993)).

While these arguments for and against increasing the number of females on corporate boards are essential, as becomes clear through diverse coverage in media from all sides of

(21)

the spectrum, the focus shall now be directed towards certain characteristics that distinguish women from men. Those gender-specific traits might indeed have an impact on corporate culture, if Upper Echelons Theory formulated by Hambrick and Mason (1984) applies. Their theory is built on the premise of bounded rationality (cf. Cyert and March (1963) and March and Simon (1993)), i.e. the idea that absolute knowledge of complex, uncertain situations cannot exist in reality, so that only interpretation allows drawing conclusions from the limited amount of data inputs (Mischel 1977). Taking into account numerous extensions provided by other authors in the meantime, Hambrick updated it in 2007. The theory suggests that executives' past experiences, acquired values, and personalities greatly influence their field of vision (i.e. information intake), selective perception and interpretation and, in turn, affect their choices. It further stipulates that the characteristics of the top management team as a whole should be taken into account when attempting predictions of organizational outcomes. Hambrick (2007) finds that executives' functional backgrounds, industry and firm tenures, educational credentials and affiliations can be used reliably to predict strategic actions, and that demographic profiles of executives are highly related to strategy and performance outcomes, indeed. Of course gender is merely one aspect of the diverse set of characteristics constituting a demographic profile, but it might nevertheless have a noticeable impact.

Plenty of studies have investigated how women and men behave when playing games in which the players need to take decisions that increase either their personal gain or the gain of others or the group as a whole. The dictator game and the ultimatum game are prominent examples. In the dictator game, the (first) player assumes the role of a dictator who must split an endowment between him- or herself and the other players, upon which the other players simply have to accept the decision. There can be a price of giving, meaning that giving one unit to a recipient might cost the dictator more than that one unit. Andreoni and Vesterlund (2001) found that women’s willingness to give depends less on the price of giving than it does for men, meaning that women are more generous when tradeoffs are expensive,

(22)

whereas men tend to be more generous than women when the price of giving is low. In other words, women were found to be more likely to insist on equality and thus share evenly, whereas men tended to be either selfish or to maximize total payoffs by taking into account the price of giving. This equalitarian behavior hints at a possible effect that female directors might have on the CEO salary, as a limited amount of resources needs to be distributed among many recipients, of which one is the CEO. In the ultimatum game, the first player receives an endowment and needs to propose how to split it between him- or herself and a second player. Subsequently, the other player can either accept or decline the offer; if he or she decides to opt for the latter, then none of the players receives anything. Eckel, Oliveira and Grossman (2008) report women to be more egalitarian and more sensitive to fairness considerations than men. They tend towards more equal distributions of the split parts of the endowment. An important point made is also that women are actually expected to be fair. Due to those reasons, they are believed to be more successful in negotiations aiming at long-term relationships rather than short-term gains, and would probably put larger emphasis on not-so-excessive compensation decisions in a setting where the board of directors has to decide on executive pay. Additionally, Dufwenberg and Muren (2006) find that both men and women gave more money to female than to male second movers, with the notion that males are less in need or deserving large amounts of money. These results are complemented, for example, by Vinacke (1959), Vinacke (1964) and Brenner and Vinacke (1979), who all confirm the stereotype-based expectation that females are more accommodative, aiming at fairness for everyone by focusing on strategies oriented towards social and ethical considerations, and males more exploitative, i.e. playing to win, with the end justifying the means.

In addition to the stronger focus on equality, fairness and broader social and societal considerations, women were observed to exhibit higher discipline with respect to their duties as a director than men. Adams and Ferreira (2009) find that females have better attendance records than male directors and that males’ attendance statistics are improved in more

(23)

gender-diverse environments. Apparently, gender-diverse boards engage more strongly in governance processes, aiding both the monitoring and the counseling role of the board. This higher quality governance is probably the reason for the same authors also reporting CEO turnover becoming more sensitive to stock performance. A positive relationship between diversity and effectiveness in the oversight function of boards of directors was also observed by Erhardt, Werbel and Shrader (2003). They argue that diversity generates more conflicts, which spark more discussions, eventually leading to better performance, as was discussed previously with regard to the participative type boards (cf. Pearce and Zahra (1991)) and gets confirmed by Selby (2000) noting that through the presence of women the board of directors can acquire an improved ‘questioning culture’.

In summary, women seem to put a strong emphasis on equality, even if it does not necessarily maximize the overall utility. They are apparently taking into account a broader picture than their male counterparts, substituting the interests of shareholders by the interests of the larger group of stakeholders. And their presence on the board of directors improves attendance statistics, proactivity and monitoring quality. Taking into account these points, following hypotheses emerge with respect to the research question:

Hypothesis 1a: The fraction of female directors on the board impacts CEO compensation. Hypothesis 1b: More specifically, CEO pay levels are expected to be lower when the fraction of female directors on the board is higher.

The effect might differ for different components of the compensation package, which is why it will be investigated not for total compensation, but for different components that will be described in the next section.

In the literature review it became clear that tenure allows CEOs to build and strengthen networks and ties that allow them to exert stronger influence on their environment and therefore affect their compensation. Similarly, female directors should be able to use their

(24)

time served as director to gain power and influence, partially countering the increased CEO power. Therefore, a second set of hypotheses shall be formulated:

Hypothesis 2a: The tenure of female directors moderates the relationship between the fraction of female directors on the board and CEO compensation.

Hypothesis 2b: More specifically, the higher their tenure, the stronger the negative effect of female director fraction on CEO compensation is expected to be.

Since the first hypothesis will be tested for different components of the executive compensation package, so will the second.

(25)

Research Method

This section first explores the structure of CEO compensation, in order to arrive at a suitable selection of dependent variables for the statistical analysis. Then, the other variables chosen for the regression are presented, with a complete schematic model summarizing the first subsections. Sources to acquire the necessary data from, as well as calculation methods and variable transformations will be described, before finishing with a brief description of the analysis approach.

CEO Compensation and Dependent Variables

As the introduction already suggested, many, especially the public and public media, like to use the concept of pay ratio between CEO and average worker. This figure makes it easy to put executive pay into context, rather than providing absolute values that are difficult to interpret due to a lack of meaningful comparison for the layman. A plot of the total annual CEO compensation versus the ratio of total CEO compensation to average worker compensation, shown in Figure 2, reveals a rather intuitive fact: The data scatter around a linear relationship over the whole observed range, spanning several orders of magnitude. This means that the ratio of CEO compensation to average worker combination is strongly dominated by total annual CEO compensation. Due to the high correlation the two constructs can almost be used interchangeably and cross-concept conclusions can be drawn to some extent without major distortions.

(26)

Figure 2: Total annual CEO compensation vs. ratio of CEO compensation and average worker compensation. The data covers a large sample of SEC-registered US companies for the year 2013. The red line indicates a linear relationship between the dependent and independent variables.

Which concept shall be used here then? While the pay ratio of CEO and workers might be best suitable for public media, where provision of context for the general public is most important, absolute CEO compensation might be more useful for thorough statistical analysis and modeling, as differences in worker pay in different organizations add noise to the more basic absolute data. This is a good argument to use absolute total CEO compensation data for a statistical analysis. There are other good arguments to even go further, though. Different components of the compensation package can be determined by different underlying strategies and ideologies. Murphy (1999) provides a detailed description of different compensation components in his review on executive compensation. Owing to the exhaustive coverage in the literature, only a brief overview shall be provided in this thesis.

Salary is a fixed monthly payment that provides a secure income required to cover continually occurring costs. Traditionally, it is the component that is traded for time spent on the job rather than for leisure. Most ordinary workers only receive a salary, as their contribution to the overall success of a big firm is considered minor and their talent easily substitutable.

Bonuses typically depend on short-term performance targets, which are often tied to measurable inputs affecting the success of the firm, as well as its financial performance. For

(27)

CEOs, the bonus might well be tied to measures such as return on assets; however, different rules and requirements in different companies – even within the same industry – are likely to obscure this relationship. Additionally, bonuses are often granted only when a performance threshold is surpassed, and are often capped by an upper limit.

Stock shares and options are more strategic incentives, intended to provide rewards for a positive long-term development of the organization. They are not necessarily granted annually and are more difficult to evaluate with respect to their value in the corresponding fiscal year. Vested or restricted stock and options are rather common as long-term incentives; they cannot be realized for a certain amount of time called the vesting period. Due to the potentially significant time lag, their realized value upon selling and exercise, respectively, is not necessarily a good measure of CEO compensation, since that depends both on the CEO’s skills of choosing the right moment in time, as well as on general economic conditions, industry development and luck. Consequently, it makes sense to use the fair value – an accounting value that assumes the market price equivalent at the date of measurement – as an estimate for this type of compensation. Shares are the most common long-term compensation used to pay CEOs and in conjunction with firms growing ever bigger and bull markets presiding, one of the major reasons for vastly increasing total CEO compensation.

Retirement plans are mostly put into place by big organizations and usually only make a relatively small contribution to the total compensation package. Their value cannot be assessed exactly at the time of issue, but statistically estimated.

Murphy (1999) stated that pay levels and structures were converging from the point of view of the mid-nineties, reflecting an increasingly global market for managerial talent, in which large international corporations compete for the same resources. Frydman and Jenter (2010) complement Murphy’s work with another review, covering the development in subsequent years. Figure 3 shows data for more recent developments, still following the same trend. On

(28)

average, all components of CEO compensation increase, but except for bonus payments, a narrowing of the distribution can indeed be observed.

Figure 3: Development of five different components of the total annual CEO compensation in recent years. Data is shown for US companies listed in the Russell 3000® index in year 2015, if available in the Capital IQ database. Grey triangles represent the top 1% and bottom 1% limit, black circles the mean of a log-normal fit. The distributions suffer from noise and a log-normal fit does not always seem suitable, so this graph is only to be understood as a measure of a trend. The analysis here disregards the composition of compensation packages, showing an analysis for the distribution of each individual component (disregarding that e.g. the salary might be smaller in a company where stock awards are very high).

It is also apparent, however, that different components have been developing differently, which makes sense when taking into account that they are influenced differently by independent variables. It could, for example, be that industries tend to reward their c-suit staff with higher bonuses at the expense of salary than other industries. Likewise, female directors may favor an increase in performance-related compensation over a salary raise, creating a perception of higher fairness. These effects might be obscured if only the total compensation package was to be analyzed.

By far the most frequently used compensation components are salary and stock awards, as can be seen in Figure 4.

(29)

Figure 4: (left): Nestled histograms of the four components salary (each vertical square represents a 10% increase), pension change (each horizontal square represents a 10% increase), stock awards (each scale bar in vertical direction inside a square represents a 10% increase) and option awards (each scale bar in horizontal direction inside a square represents a 10% increase). The color represents the occurrence of the given combination, while the dashed lines indicate possible values for stock and options for a given combination of salary and pension change. (right): Complementary table displaying the top 50% of combinations. The top three rows represent 25% of all occurrences.

The outcomes of the analysis matter most for salary and stock options for two reasons. On the one hand, these are the most prominent components of executive pay. On the other hand, any influence that can be traced back to female traits on the board of directors are most likely – at least following the discussion in the introduction – to affect performance-based long-term incentives and performance-unrelated fixed remuneration.

Since all abovementioned components can be accessed in the same database, though, all of them shall be used in the analysis. This yields five dependent variables:

 Annual salary

 Annual bonus

 Stock awards (fair value, FAS 123R)

 Option awards (fair value, FAS 123R)

(30)

Control Variables

Over the years a great number of factors potentially having an impact on CEO compensation were discussed and used in analyses in the literature. Many of those are difficult or even impossible to assess within the given time constraints. Others are heavily debated, with very mixed results for their predictive power, ranging from relatively strong positive associations with executive pay through no detectable influence to a negative impact on CEO pay levels. In this thesis, factors shall be chosen that are relatively basic, meaning that they can be used without computing complex constructs first, readily accessible in existing databases and reported as significant predictors by many scientific articles.

There are two predictors for CEO compensation that all scholars seem to agree on: Industry membership and firm size. Accordingly, these variables should be included in every analysis exploring the effect of any independent variable on CEO compensation.

Different classification systems allow placing organizations in industry groups, which significantly differ in some properties. Figure 5 shows an example for the relation of earnings with sales, where three distinct industry groups are marked by different colors. The data follow linear relationships for all three groups, but with different coefficients, so a regression on the joint data would leave a significant fraction of the variance unexplained.

Figure 5: EBIT vs. revenue with three NAICS groups indicated by different colors. The dashed black line indicates a scaling factor of unity, lines parallel to it represent linear relationships with a different scaling factor.

(31)

In the US, the Standard Industrial Classification (SIC) code has been used for a long time, but has mostly been replaced by the North American Industry Classification System (NAICS). NAICS deploys a two-through-six digit code, where the first two digits designate the economic sector, the third represents the subsector, the fourth stands for the industry group, the fifth designates the NAICS industry and the sixth the national industry (since Canada, Mexico and the USA are using the code). A greater number of digits signifies greater classification details, but also reduces the number of data points in specific groups. Typically, membership in an industry group is recoded into a dummy variable that assumes unity when a firm is a member and zero when it is not. A wide enough definition has to be chosen to retain enough data points within one group to allow a meaningful regression analysis, while simultaneously differentiating sufficiently in order to preserve intergroup differences.

Firm size is the most powerful predictor of CEO compensation, expected to explain up to 30% of the variance. It can be measured in many ways, be it as number of employees, total assets, market capitalization, revenue, or other figures. All of them exhibit large correlations with each other, rendering them almost interchangeable in statistical analyses. Sanders and Carpenter (1998) found that the number of employees and a firm’s sales were equivalent and mutually exclusive in explanatory power and Gabaix and Landier (2008) and Gabaix, Landier and Sauvagnat (2013) tested three firm size proxies and reported that market capitalization, earnings before interest and taxes, as well as sales, are roughly equally good predictors of CEO compensation with roughly 30% of CEO compensation variance explained. Adding any of the remaining two variables did not have any additional explanatory effect. Baker, Jensen and Murphy (1988) state that the relationship between firm size and CEO pay is very reproducible over many studies, industry groups and time, with a 10% increase in size yielding a 3% increase in executive compensation. A possible explanation, according to this paper and Murphy (1985), is that boards use firm size as an indicator for compensation decisions. Murphy shows that firms whose value does not change, but whose sales increase by 10%, subsequently raise the CEO’s remuneration by 2-3%. Note that regardless of the

(32)

actual measure used, due to the highly asymmetric (log-normal) distribution of firm sizes, this control variable is typically transformed by computing its decadic logarithm. As revenue is probably the most widely used measure of firm size in the literature, its logarithm shall also be used in this thesis.

Firm performance is one of the perhaps not most powerful, but certainly most frequently discussed, predictors of CEO compensation. Three measures of firm performance are widely used. Firstly, return on equity (RoE), defined as a firm’s net income divided by its shareholders’ total equity. Secondly, return on assets (RoA) as a measure of accounting performance, defined as a firm’s net income divided by its total assets. And finally, Tobin’s Q, often computed as a firm’s market capitalization divided by its book value. Controlling for firm size and industry membership, the authors of Beasley’s Research Handbook on Executive Pay (2012) tested numerous different measures for profitability in a regression, of which return on assets (RoA), return on equity (RoE) and Tobin’s Q as an indicator of existing opportunities, measured as market value relative to book value, are the most powerful predictors of executive compensation. Any of these three measures – but not in combination – is frequently found in analyses in the literature. The book further states that total CEO compensation changes 2.3% for a 10% change in size, affecting mostly the salary component, and 1% for a 10% change in Tobin’s Q, affecting mostly the bonus component. These results strongly support the 2.14% and 2.11% changes for 10% change in size and Q, respectively, that Bebchuk and Grinstein (2005) report, the deviation in the effect of Q being explained by controlling for macroeconomic factors separately by Beasley (2012). Even though the value seems unusually high in comparison to most other results in the literature, Belliveau, O’Reilly and Wade (1996) report a 57% explained variance in executive compensation by using sales and return on equity as control variables. In a meta-analysis of 137 articles on CEO pay, Tosi, Werner, Katz and Gomez-Mejia (2000) concluded that size – measured as total market value, total assets or sales, or the logarithm of any of those, depending on the study and combined into one aggregated size in their paper – accounts for

(33)

more than 40% of the variance in total CEO pay, while firm performance – mostly represented by RoA or RoE – accounts for less than 5% of the variance. Newman and Mozes (1999), use not only RoE, but also stock returns as a predictor with positively association with CEO compensation. Whether returns or one of the ratios is preferable they do not elaborate on. Return on equity is one of the most frequently used measures in the literature and is probably the best representative of shareholder interests. Therefore, RoE shall be used as a control variable in this thesis.

The size of the board of directors has been found by some researchers to be positively associated with CEO compensation, due to inferior governance. On large boards, directors are believed to feel less responsible and get less involved in decision making, which weakens the power of the board relative to that of the CEO. Some researchers use board size as one component of a more complex construct often coined board power, while others use the unmodified figure. For the sake of simplicity and to avoid introducing errors by modifying variables, board size shall be used as is in this thesis.

Tenure, as the number of years a CEO has served in the current position, grants specific experience, allows for subsequent annual raises and enables the CEO to build a larger network, as well as reinforce his or her entrenchment. It is used by many researchers as one dimension of CEO power and often found to have a small, yet significant impact on CEO compensation. Again, many researchers build CEO tenure into a complex construct referred to as CEO power, but for simplicity’s sake tenure in its raw form shall be used in this thesis.

Independent Variable

Female presence on the board of directors was reported to increase the attendance to and frequency of board meetings, which in turn are seen as a strengthening factor of board power. Since the number of females naturally correlates with the board size, the fraction of female directors, defined as the number of female directors on the board divided by the

(34)

board size shall be used as the main independent variable. It is yet to be determined whether there is any significant impact on CEO compensation.

Moderator variable

Similarly to the tenure of a CEO increasing his or her experience, network ties and confidence, the tenure of female directors should increase their expertise, network ties and assertiveness. Therefore, it might act as a moderator on the relationship between female fraction on the board and CEO pay. Since there can be several female directors present, tenure in this case is a distribution rather than a scalar value. Using the average would cause a strong downward bias regarding the influence of more seasoned female directors in the presence of new female directors. Although the sum is likely to lead to a positive bias on the expected increase of influence, the error is deemed to be small due to the overall low representation of women on corporate boards. If in the future this situation changes, a different construct might be required, for example ranking the female directors by tenure and then summing the distinct tenure values divided by their rank.

(35)

Full Regression Model

Figure 6: Schematic representation of the model used in the statistical analysis. Relationships between the control and dependent variables have been combined into one single arrow for clarity.

Data Sourcing and Preparation

Since a vast pool of data both on firms, their CEOs and their directors is readily available for the US market, the choice of firms fell onto an American stock market index covering roughly 98% of the American public equity market: Russell 3000. The ticker symbols were extracted from that list and used to source all data for the year 2015 required for the statistical analysis in this thesis from the following data bases:

 Compustat Capital IQ Segments (firm specifics)

 Compustat Capital IQ Execucomp (CEO specifics)

 Institutional Shareholder Services (director specifics)

Out of the 3003 organizations contained in the index list, data was available for 2823 firms concerning financial information about the organization, for 1388 firms regarding CEO specifics excluding compensation, for 1470 firms regarding director specifics, for 1538 firms

(36)

regarding non-zero CEO salary figures, for 254 firms regarding non-zero CEO bonus figures, for 1374 regarding non-zero CEO stock award figures, for 708 regarding non-zero CEO option award figures and for 444 regarding non-zero CEO pension change figures.

All firms that were missing any of the control variables were sorted out, leaving 1349 data sets for which all control variables had valid values.

Then, in five separate steps, all firms were sorted out that did not contain any valid values for the five different compensation components. The resulting portfolio was grouped with respect to the firms’ industry membership, using the first three digits of the NAICS code. All groups containing less than 20 firms were excluded, leaving 17 groups with a total of 844 firms for a regression on salary, 1 group with a total of 29 firms for a regression on bonus, 15 groups with a total of 712 firms for a regression on stock awards, 5 groups with a total of 235 firms for a regression on option grants and 6 groups with a total of 184 firms for a regression on pension change.

This methodology substitutes listwise exclusion of variables in the regression, while simultaneously providing higher transparency on what is happening during analysis.

For all relevant NAICS three-digit codes a Boolean dummy variable is created, which assumes zero if a firm is not member of that NAICS subcategory and unity if it is. That would be 17 dummy variables for the regression on salary, 15 for the one on stock awards, and so on. Revenue as a measure of firm size exhibits a highly asymmetrical distribution, so the decadic logarithm was computed, as is common practice in the vast majority of all articles using firm size as an independent regression variable. Return on equity was computed by dividing pretax income by shareholder’s total equity. Board size was determined by summing all directors listed for the same firm. Female directors were likewise summed, in order to obtain the fraction of female directors as the number of female directors divided by the total number of directors. Tenures were derived from the installment date; the accumulated

Does the fraction of female directors on the board impact CEO compensation?