Amsterdam Business School
The influence of CEO power on the use of
compensation consultants
A study of US firms over the period 2007-2013
Name Roy Steinvoort
Student number 10088369
Program MSc Business Economics
Specialization Finance
Document Master’s Thesis
Number of ECTS 15
Thesis supervisor dhr. dr. I.J. Naaborg
Date July, 2016
Abstract
This research captures the relation between CEO power and the use of compensation consultants. Furthermore, this study examines the association between economic and corporate governance determinants and the level of CEO compensation. Using a longitudinal dataset consisting of compensation consultant data and data from ExecuComp, Compustat and ISS on 1,580 US firms, the results show that CEO compensation is higher in larger firms and in firms with weaker governance which is consistent with prior research. Furthermore, consistent with previous studies, this paper finds that firms in which the CEO exhibits higher relative power to the other executives and firms with weaker governance are more likely to hire compensation consultants. The findings of this study imply that claims of prior studies that CEOs use compensation consultants as a justification device for extracting rents from the company, are not unfounded. The use of compensation consultants seems not to mitigate agency problems.
Keywords: Executive compensation, CEO power, Compensation consultants, Corporate Governance, Agency problems
- 1 -
Statement of Originality
This document is written by Roy Steinvoort 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.
Acknowledgements
Through this way I would like to thank Ilko Naaborg for his guidance throughout the process of writing a Master’s Thesis, with the purpose of graduation. In addition, my gratitude also go out to Torsten Jochem for giving a nudge in the right direction with respect to the topic of the thesis as well as making available his personal data regarding the use of compensation consultants.
- 2 -
Table of Contents
1. Introduction ... 3
2. Related Literature ... 5
3. Methodology and Data ... 9
3.1. Methodology ... 9
3.1.1. Model specifications ... 9
3.1.2. Endogeneity issues ... 13
3.1.3. Reverse causality/Simultaneity ... 13
3.2. Data and descriptive statistics... 14
3.2.1. Sample ... 14
3.2.2. Variables ... 15
3.2.3. Descriptive statistics ... 17
4. Analysis ... 19
4.1. Empirical Results ... 19
4.1.1. Economic and governance determinants of CEO compensation .... 19
4.1.2. Economic and governance determinants of CEO power... 22
4.1.3. CEO power and the use of compensation consultants ... 25
4.2 Robustness checks ... 26
4.2.1. Consultant use ... 26
4.2.2. Additional disclosure rules ... 27
4.2.3. Pooled Probit model ... 28
5. Conclusion and discussion ... 29
References ... 31
-1. Introduction
Bebchuk, Cremers and Peyer (2011) investigate the relation between CEO power, which is the total annual compensation of the CEO relative to the total annual compensation of the other executives (CEO pay slice) and the value, performance, and behavior of public firms. For a sample of 2,015 US firms over the period 1993-2004, they find evidence that the CEO pay slice is correlated with lower accounting profitability, negatively related to firm value measured by Tobin’s Q and that it is associated with agency problems (CEOs act in their own self-interest and are assumed to extract rents from the firm by allocating a larger pay slice to himself or herself). Hence, high CEO power could lead to neglecting shareholder interests. Bebchuk and Fried (2003) explain in their paper that beyond the excess pay executives receive, there might be substantial costs imposed on shareholders because managers influence their own pay by diluting and distorting managers’ incentives and thereby damage corporate performance.
According to Adams, Almeida and Ferreira (2005) both executive variables (i.e. ‘power of the CEO over other executives and his role on the board) and firm variables (i.e. firm size, leverage and the capital expenditure to sales ratio) could influence firm performance. They hypothesize that firms with less powerful CEOs will have less volatile performances. These less extremely high or low performances are, in principle, not necessarily better or worse for a company, but just more moderate when compared to firm performance under powerful CEOs.1 Since with less power, the CEO has to compromise with other executives when they disagree. Therefore, more moderate operating decisions will be taken by taking into account shareholder value and pursuing the interest of the company. They provide statistically and economically significant proof by using a sample of US fortune 500 firms for the period 1992-1999, that when decision-making power becomes more centralized in the hands of the CEO, firm performance measured by stock returns, ROA or Tobin’s Q will be more variable.
One way how firms are trying to align the interests of the shareholders with the interests of the management is by hiring compensation consultants for advice on the compensation of executives. Chu, Faasse and Rau (2015) examine the rent extraction hypothesis through a sample of 1,051 US publicly listed firms for the period 2006-2012.
1 Adams et al. (2005) consider CEOs as powerful if they are able to consistently influence key
- 4 -
This hypothesis argues that CEOs have great influence on the pay-setting decision and use their power to extract rents from the board in the form of excessive pay. Since the executives own interest is deemed more important than the interests of the shareholder an agency problem arises. They want to determine whether compensation consultants intentionally help the executives undermine the interests of the shareholders by deliberately advising higher compensation packages. On the other hand they examine if CEO pay is efficiently set according to the optimal contracting hypothesis, which is to attract and motivate candidates for the CEO executive positions and provide the optimal level of motivational incentives to increase shareholder value. Armstrong, Ittner and Larcker (2012) state that proponents of managerial power theories argue that consulting practices as mentioned by Chu et al. (2015) provide a way for CEOs with significant influence over the board of directors to justify excessive pay packages. A primary assumption in the literature of managerial power theories is that ‘weak’ corporate governance and ‘excess’ CEO pay are not in shareholders’ best interests (Armstrong et al., 2012).
However, these studies do not examine whether relative CEO power influences the decision on whether to hire compensation consultants or not. It could be the case that when a CEO is very powerful relative to the other executives, he or she does not want to hire compensation consultants because the CEO does not want its compensation package to go down or the CEO does not want to lose its relative power advantage over the other executives. On the other hand, the CEO might intend to increase the compensation packages of all executives by hiring compensation consultants who assist in extracting rents. The latter obviously is not in line with the interests of the shareholders. Therefore, this study is relevant to society – represented by e.g. shareholders, tax payers and a company’s clients – since there is additional evidence that touches upon the fact that the use of compensation consultants may strengthen the selfish behavior of CEOs. In addition, consultant use might not reflect a very strong form of corporate governance and firms might want to take other measures to reduce weak governance and agency problems within firms even more.
In short, this study contributes and adds to the scarce quantity of prior literature on the use of compensation consultants and CEO power since the above mentioned literature serves as guidance regarding the use of control variables, methodological parts and theory in order to conduct research on compensation consultants with another angle of incidence. This other angle of incidence is an examination of what
characteristics influence CEO pay levels and the association between relative CEO power and the use of compensation consultants. This is complementary to the influence of (weak) corporate governance and compensation consultants on CEO pay as examined by prior studies.
Following from this, the main aim of this research is to provide valuable insights on the relationship between hiring compensation consultants and the relative power of CEOs with respect to the other executives within firms. The research question: “How
does the level of CEO power affect the use of compensation consultants in the US for the period 2007-2013?” will be investigated by means of a longitudinal dataset of 1,580
publicly listed US firms that is compiled from data that comes from a variety of databases such as ExecuComp, Compustat, CRSP/Compustat Merged, ISS and a personal database on compensation consultants. The panel dataset spans a period from 2007 to 2013. In this study, a nonlinear Correlated Random Effects model is used to capture the effect of CEO power on the probability that firms hires a compensation consultant.
The remainder of this study is organized as follows. Section 2 reviews prior and relevant literature on the use of compensation consultants, CEO/executive compensation and CEO power. Section 3 provides the methodology and describes the data sample and variables. Section 4 provides empirical results. Section 5 concludes and discusses.
2. Related Literature
Bebchuk et al. (2011) serves as a key example regarding CEO power. By means of a pooled panel with firm and year fixed effects they investigate the relationship between CEO power (measured by the CEO pay slice) and the value, performance and behavior of public firms based on US data for the period 1993-2004. They argue that the CEO pay slice could reflect the relative performance of the CEO as well as the extent to which the CEO is able to extract rents which is tested following the agency hypotheses. The authors find that Return on Assets (ROA), the Industry Median CEO pay slice and the CEO also being the chairman of the board are positively related to CEO power. They also find that higher CPS (the power of the CEO is relatively stronger) is associated with lower firm value. This research is related to Bebchuk et al. (2011) because the same measure for CEO power within the top executive team (i.e. CEO pay slice) is used to see how relative CEO power is affected by economic and governance
- 6 -
characteristics. It differs from their study in the examination whether CEO power can be associated with hiring compensation consultants in a longitudinal dataset.
Focusing on the structural power of CEOs relative to the board and other executives that arises from his formal position, status as a founder and as the board’s sole insider, Adams et al. (2005) find that stock returns are more volatile for firms led by powerful CEOs. They state that the higher the number of important decision-makers, the less powerful the CEO is likely to be. Furthermore, they find that diluting CEO power may lead to less variable firm performance while simultaneously the likelihood of extreme high or low performance will be lower. The authors use data on US fortune 500 firms for the period 1992-1999. To test their main hypothesis that firms with CEOs that have less power will have less extreme performances they apply Glejser’s heteroscedasticity test for which they specify performance models.
Both Armstrong et al. (2012), using a cross-sectional sample that consists of 2,110 publicly traded firms with fiscal year ending after December 31st, 2006 (thereby meeting the new disclosure requirements of consultant use) and Core, Holthausen and Larcker (1999) using survey collected data of 205 publicly listed US firms for the years 1982, 1983 and 1984, find that when governance structures are less effective – e.g. with large boards, busy boards,2 old directors on the board,3 the CEO is also chairman of the board –, CEOs receive higher compensation. In addition, Armstrong et al. (2012) find that companies with weaker governance are more likely to hire compensation consultants. They raise the question how CEOs justify excess pay levels to the board, the shareholders, and outside observers (e.g. the media). One way is possibly through the use of compensation consultants that design and justify excessive pay packages in firms with weak governance. They also show that the majority of total executive compensation is explained by economic variables such as market capitalization, return on assets, book-to-market ratio – which serves as an inverse measure of the firm’s investment opportunities - and prior stock return of the firm. This study is related to Armstrong et al. (2012) because part of the methodology and control variables that are used, are presented in their study. This research complements Armstrong et al. (2012) through a more comprehensive database and focusing on different aspects with respect to governance determinants of CEO pay and the use of compensation consultants. In
2 Directors are classified as busy if they serve at least on two boards (Armstrong at al., 2012 and Chu et
al., 2015).
3 Directors are classified as old if they are at least 69 years of age. This is conform the specifications of
order to conduct their research Armstrong et al. (2012) use Most companies are part of the Russell 3000, but there are also somewhat smaller firms that fit the criteria. For determining the explanatory power of economic and governance characteristics of CEO compensation they use OLS model with two digit SIC industry fixed effects included. Furthermore, for testing whether higher pay levels at firms using compensation consultants are due to firms with weaker governance rather than by making use of compensation consultants per se, they employ a propensity score matched pair research design. They estimate some multivariate logistic models.
Chu et al. (2015) also find in their study - based on an extensive longitudinal dataset consisting of US public firms for the period 2006-2012 - that compensation consultants can be used as a way of justifying higher executive pay, whereas Cadman, Carter and Hillegeist (2010) – which analyze 880 S&P 1500 firms with fiscal year ending after December 15th, 2006 (in order to capture effects of the new disclosure requirements) – find evidence suggesting that potential conflicts of interest between the firm and its consultant are a primary driver of excessive CEO pay for their one year data sample. They do this by estimating a multivariate OLS model with industry fixed effects included to capture CEO pay differences across industries. Chu et al. (2015) investigate this matter via the rent extraction hypothesis which argues that CEOs have a great deal of power which they use to extract rents from the board. They also use propensity matching and logistic regressions to estimate their models on the determinants of CEO compensation and compensation consultant use. Consistent with this rent extraction hypothesis from Chu et al. (2015), Bebchuk and Fried (2003) state that the managerial power approach is also part of the agency problem and is not only considered to be a potential instrument for addressing agency issues, because some features of compensation structures seem to reflect managerial rent-seeking behavior rather than the alignment with shareholder incentives. Bebchuk and Fried (2003) also argue that managerial power and rent extraction are likely to have an important impact on the shaping of compensation packages. Chu et al. (2015) argue that Armstrong et al. (2012) pose an important finding against this rent extraction hypothesis. Namely, using compensation consultants is not a justification device for high executive pay by itself, but simply an expression of weaker corporate governance at firms using those consultants. They find that switching to related newly spun-off specialist compensation consultants serves as a screening mechanism for client firms that hired conflicted consultants with other business interests at stake. CEOs were paid more by these firms
- 8 -
than by firms that remained with multi-service consultants. Furthermore, they ascertain that consultants have incentives to help CEOs justify higher pay as they find that larger pay increases are significantly and negatively associated with subsequent consultant turnover. This study is related to Chu et al. (2015) since some specifications that they impose to their dataset are followed and a few tests are used as examples.
Core et al. (1999) find that both board characteristics and ownership structure are substantially associated with CEO pay levels. They estimate the effects of economic determinants and board and ownership structure on CEO pay by means of a fixed effects model accounting for year and industry specific effects. The authors also argue that weaker governance is related to greater agency problems which in turn is related to CEOs receiving higher pay and worse firm performance. Their results show that the level of total CEO compensation is associated with firm size, investment opportunities, prior performance, and firm risk. They also state that bigger firms and firms with more investment opportunities (i.e. market-to-book ratio acts as a proxy) pay higher CEO compensation. They interpret this as reflecting firm’s demand for higher-quality managerial talent.
Using a cross-sectional sample of 1,046 US firms, Murphy and Sandino (2010) examine the relation between conflicting consultants and CEO compensation for 2007, falling under the new SEC disclosure rules effective as of end fiscal year 2006. In a parallel analysis they use 124 Canadian firms for 2007 that meet the Canadian disclosure rules as of the start of fiscal year 2005. Their results are robust to multiple different OLS specifications, but fail to provide significance when estimated using a propensity-score matching analysis. They find that in US and Canadian firms where consultants also provide other services than compensation advice, CEO compensation is higher. They also find that for US firms, CEO pay is higher when the consultants work for the board of directors which is inconsistent with their predictions. Contrary to Murphy and Sandino (2010) and Cadman et al. (2010) – using a sample of the 250 largest UK and S&P 500 US firms for respectively 2003 and 2007, Conyon – Peck and Sadler (2009) find no support for the claim that CEO pay is higher in firms that use consultants who are also involved in supplying other types of business to the client company. In addition, they find that CEO pay levels are higher in firms with consultants after they control for firm characteristics, but no evidence that governance determinants influence higher CEO pay using simple OLS regression methods.
Following the findings of Armstrong et al. (2010); Core et al. (1999); Cadman et al. (2010) and Chu et al. (2015) with respect to the economic and governance determinants of CEO pay, and contrary to the results of Conyon et al. (2009) regarding the association of (weaker) governance the first hypothesis (H1) in this study is that
both economic characteristics such as firm size, investment opportunities and prior stock returns, and weak governance characteristics as board size, busy board members (that serve on at least two boards), the CEO also being the chairman of the board and the percentage of outside directors that are appointed after the current CEO took its position, are associated with higher CEO pay.
In addition, conform the results of Bebchuk et al. (2011) that some economic and governance characteristics contribute to a higher CEO pay slice relative to the other executives and that this higher CEO power is associated with lower firm value, the following is hypothesized (H2): Weak governance characteristics (i.e. larger board
size, busy directors, old directors and CEO is chairman) provide the largest foundation for CEO power.
Furthermore, the hypothesis (H3) that powerful CEOs increase the probability
of firms hiring compensation consultants, amongst the impact of other firm and governance characteristics, is founded by combining some elements of H1 and H2,
using the results of Chu et al. (2015) and Armstrong et al. (2012) on the economic and governance determinants of compensation consultant use and following the findings from Armstrong et al. (2012); Chu et al. (2015); Cadman et al. (2010); Murphy and Sandino (2010) and Conyon et al. (2009) with respect to the influence of compensation consultant use on CEO pay and the interpretations that CEOs use consultants to extract rents and their compensation structures are reflecting rent-seeking behavior.
3. Methodology and Data
3.1. Methodology
3.1.1. Model specifications
First, analyses are conducted on the dependent variables Total annual CEO
compensation and CEO power to get a clear indication on how these variables are
related to differences in economic and governance factors, independent of the use of compensation consultants. H1 and H2 will be estimated by Ordinary Least Squares
- 10 -
(OLS) models and fixed effects models, similar to prior studies (e.g. Armstrong et al., 2012; Chu et al., 2015; Core et al., 1999), with the natural logarithm of total annual CEO compensation and CEO power as the dependent variables respectively. Consistent with Armstrong et al. (2012) and Chu et al. (2015) the model will be estimated including only the economic variables and then the governance variables are added to assess their incremental explanatory power and importance. The various control variables are based on prior studies and are described in Section 3.2 and provided in Appendix 1. Furthermore, following prior studies (e.g. Armstrong et al., 2012; Chu et al., 2015; Core et al., 1999; Adams et al., 2005; Murphy & Sandino, 2010) the models will be expanded by including entity (industry) and time (year) fixed effects to take into account that there are repeated observations on the same industries and to see whether there are unobserved factors influencing the outcomes. Below, for H1 and H2 only the fixed effects models (1.4, 1.8 and 2.4, 2.8)4 are specified since these are the most comprehensive. For the simple OLS models (specifications 1.1 - 1.3 and 1.5 - 1.7; 2.1 - 2.3 and 2.5 - 2.7),5 the fixed effects indicators 𝛼𝑖 and 𝜆𝑡will be left out entirely or are only partially included.
Specification H1, with economic and CEO characteristics:
𝐿𝑛 𝑇𝑜𝑡𝑎𝑙 𝐶𝐸𝑂 𝐶𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛𝑖𝑡 = 𝛽1𝐿𝑛 𝑀𝑎𝑟𝑘𝑒𝑡 𝑐𝑎𝑝𝑖𝑡 +
𝛽2𝐵𝑜𝑜𝑘 𝑡𝑜 𝑚𝑎𝑟𝑘𝑒𝑡𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝛥𝑅𝑂𝐴𝑖𝑡+ 𝛽5𝑃𝑟𝑖𝑜𝑟 𝑟𝑒𝑡𝑢𝑟𝑛 − 1𝑖𝑡+ 𝛽6𝐶𝐸𝑂 𝑎𝑔𝑒𝑖𝑡+ 𝛽7𝐶𝐸𝑂 𝑡𝑒𝑛𝑢𝑟𝑒𝑖𝑡+ 𝛽8𝑁𝑒𝑤 𝐶𝐸𝑂𝑖𝑡+ 𝛼𝑖 + 𝜆𝑡+ 𝑢𝑖𝑡.
Specification H1, with economic, CEO and governance characteristics: 𝐿𝑛 𝑇𝑜𝑡𝑎𝑙 𝐶𝐸𝑂 𝑐𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛𝑖𝑡 = 𝛽1𝐿𝑛 𝑀𝑎𝑟𝑘𝑒𝑡 𝑐𝑎𝑝𝑖𝑡+ 𝛽2𝐵𝑜𝑜𝑘 𝑡𝑜 𝑚𝑎𝑟𝑘𝑒𝑡𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝛥𝑅𝑂𝐴𝑖𝑡+ 𝛽5𝑃𝑟𝑖𝑜𝑟 𝑟𝑒𝑡𝑢𝑟𝑛 − 1𝑖𝑡+ 𝛽6𝐶𝐸𝑂 𝑎𝑔𝑒𝑖𝑡+ 𝛽7𝐶𝐸𝑂 𝑡𝑒𝑛𝑢𝑟𝑒𝑖𝑡+ 𝛽8𝑁𝑒𝑤 𝐶𝐸𝑂𝑖𝑡+ 𝛽9𝐶𝐸𝑂 𝑐ℎ𝑎𝑖𝑟𝑖𝑡+ 𝛽10𝐵𝑜𝑎𝑟𝑑 𝑠𝑖𝑧𝑒𝑖𝑡+ 𝛽11% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒 𝑑𝑖𝑟𝑒𝑐𝑡𝑜𝑟𝑠𝑖𝑡+ 𝛽12% 𝐵𝑜𝑎𝑟𝑑 𝑏𝑢𝑠𝑦𝑖𝑡+ 𝛽13% 𝐵𝑜𝑎𝑟𝑑 𝑜𝑙𝑑𝑖𝑡 + 𝛽14% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑠 𝑎𝑝𝑝𝑜𝑖𝑛𝑡𝑒𝑑 𝑏𝑦 𝐶𝐸𝑂𝑖𝑡+ 𝛼𝑖+ 𝜆𝑡+ 𝑢𝑖𝑡.
4 Equations 1.4 and 1.8 correspond to the specifications 1.4 and 1.8 in Table 2. Equations 2.4 and 2.8
correspond to the specifications 2.4 and 2.8 in Table 3.
5 Specifications 1.1 – 1.3, 1.5 – 1.7 correspond to the specifications in Table 2. Specifications 2.1 – 2.3,
2.5 – 2.7 are conform the model specifications in Table 3.
(1.8) (1.4)
Specification H2, with economic and CEO characteristics:
𝐶𝐸𝑂 𝑝𝑜𝑤𝑒𝑟𝑖𝑡 = 𝛽1𝐿𝑛 𝑀𝑎𝑟𝑘𝑒𝑡 𝑐𝑎𝑝𝑖𝑡+ 𝛽2𝐵𝑜𝑜𝑘 𝑡𝑜 𝑚𝑎𝑟𝑘𝑒𝑡𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝛥𝑅𝑂𝐴𝑖𝑡+ 𝛽5𝑃𝑟𝑖𝑜𝑟 𝑟𝑒𝑡𝑢𝑟𝑛 − 1𝑖𝑡+ 𝛽6𝐶𝐸𝑂 𝑎𝑔𝑒𝑖𝑡+ 𝛽7𝐶𝐸𝑂 𝑡𝑒𝑛𝑢𝑟𝑒𝑖𝑡+
𝛽8𝑁𝑒𝑤 𝐶𝐸𝑂𝑖𝑡+ 𝛽9𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑚𝑒𝑑𝑖𝑎𝑛 𝐶𝐸𝑂 𝑝𝑜𝑤𝑒𝑟𝑖𝑡+ 𝛼𝑖+ 𝜆𝑡+ 𝑢𝑖𝑡.
Specification H2, with economic, CEO and governance characteristics:
𝐶𝐸𝑂 𝑝𝑜𝑤𝑒𝑟𝑖𝑡 = 𝛽1𝐿𝑛 𝑀𝑎𝑟𝑘𝑒𝑡 𝑐𝑎𝑝𝑖𝑡+ 𝛽2𝐵𝑜𝑜𝑘 𝑡𝑜 𝑚𝑎𝑟𝑘𝑒𝑡𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝛥𝑅𝑂𝐴𝑖𝑡+ 𝛽5𝑃𝑟𝑖𝑜𝑟 𝑟𝑒𝑡𝑢𝑟𝑛 − 1𝑖𝑡+ 𝛽6𝐶𝐸𝑂 𝑎𝑔𝑒𝑖𝑡+ 𝛽7𝐶𝐸𝑂 𝑡𝑒𝑛𝑢𝑟𝑒𝑖𝑡+
𝛽8𝑁𝑒𝑤 𝐶𝐸𝑂𝑖𝑡+ 𝛽9𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑚𝑒𝑑𝑖𝑎𝑛 𝐶𝐸𝑂 𝑝𝑜𝑤𝑒𝑟𝑖𝑡+ 𝛽10𝐶𝐸𝑂 𝑐ℎ𝑎𝑖𝑟𝑖𝑡+ 𝛽11𝐵𝑜𝑎𝑟𝑑 𝑠𝑖𝑧𝑒𝑖𝑡+ 𝛽12% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒 𝑑𝑖𝑟𝑒𝑐𝑡𝑜𝑟𝑠𝑖𝑡+ 𝛽13% 𝐵𝑜𝑎𝑟𝑑 𝑏𝑢𝑠𝑦𝑖𝑡+ 𝛽14% 𝐵𝑜𝑎𝑟𝑑 𝑜𝑙𝑑𝑖𝑡 + 𝛽15% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑠 𝑎𝑝𝑝𝑜𝑖𝑛𝑡𝑒𝑑 𝑏𝑦 𝐶𝐸𝑂𝑖𝑡+ 𝛼𝑖+ 𝜆𝑡+ 𝑢𝑖𝑡.
Where in the specifications above, 𝛼𝑖 capture the industry fixed effects and 𝜆𝑡 represent the year fixed effects.
Thereafter, for testing H3 the use of compensation consultants will be implemented as the dependent variable of a Correlated Random Effects (CRE) Probit model by using a 0,1 dummy variable which is equal to one if the firm uses compensation consultants and zero otherwise. Furthermore, in this model CEO power is regarded as the core independent variable of interest amongst various control variables. In this model it is investigated by how much the probability that a firm hires a compensation consultant changes with an increase in CEO power. This Correlated Random Effects Probit model is used to examine the impact of CEO power on the use of compensation consultants in particular but will also be used to shed light on the relation of various control variables on the use of compensation consultants. By using correlated random effects, the model intends to also capture time and entity invariant effects. The CRE model can be regarded as a model that intends to obtain bias-corrected versions of ‘fixed effects’ estimators in nonlinear models (Wooldridge, 2009). This method relaxes the assumption of zero correlation between the variables in the model (𝑋𝑖𝑡) and the firm specific effects which are time invariant (𝛼𝑖) – also referred to as unobserved effects or unobserved heterogeneity (Wooldridge, 2002) – which allows for mitigation of omitted variable bias in the model. The coefficients will be unbiased since 𝛽1… 𝛽𝑘 are identical to fixed effects estimates. The estimated effect (θ) of the cluster means (𝑋̅𝑖𝑡)
(2.8) (2.4)
- 12 -
is the difference of the within firm and between firm effects (Schunck, 2013). Year dummies are added to the model to capture fixed effects that are year specific.
Specification of the Correlated Random Effects Probit model:
𝑃𝑟𝑜𝑏(𝑌𝑖𝑡 = 1|𝑋𝑖𝑡, 𝛼𝑖) = 𝛷(𝛽0+ 𝛽1𝑋𝑖𝑡+ ⋯ + 𝛽𝑘𝑋𝑖𝑡+ 𝛼𝑖 + 𝜆𝑡+ 𝑢𝑖𝑡)
With the conditional independence assumption on Xi = (Xi1, … , XiT) and αi. D( 𝛼𝑖|𝑋𝑖): 𝛼𝑖 = 𝜃𝑋̅𝑖 + 𝜉𝑖, 𝜉𝑖|𝑋𝑖 ~ 𝑁𝑜𝑟𝑚𝑎𝑙(0, 𝜎𝜉2)
Hence, the full model is as follows:
𝑃𝑟𝑜𝑏(𝑌𝑖𝑡 = 1|𝑋𝑖𝑡, 𝛼𝑖) = 𝛷(𝛽0+ 𝛽1𝑋𝑖𝑡+ ⋯ + 𝛽𝑘𝑋𝑖𝑡+ 𝜃1𝑋̅𝑖+ ⋯ + 𝜃𝑘𝑋̅𝑖+ 𝜉𝑖 + 𝜆𝑡+ 𝑢𝑖𝑡)
Where 𝑌𝑖𝑡 is the dependent binary variable, 𝑋1𝑡 is the regressor of interest and 𝑋̅𝑖 are the cluster means of 𝑋𝑖𝑡 which serve as extra control variables. 𝜉𝑖 are the individual unobserved effects and 𝜆𝑡 represents year dummies. 𝑋2𝑡+ ⋯ + 𝑋𝑘𝑡 represent various control variables as discussed in Appendix 1 and which were also used for testing hypotheses 1 and 2.
Specification H3, with key independent variable CEO power and economic, CEO and governance characteristics: 𝑃𝑟𝑜𝑏(𝐶𝑜𝑛𝑠𝑢𝑙𝑡𝑎𝑛𝑡 𝑢𝑠𝑒𝑖𝑡 = 1|𝑋𝑖𝑡, 𝛼𝑖) = 𝛽1𝐶𝐸𝑂 𝑝𝑜𝑤𝑒𝑟𝑖𝑡+ 𝛽2𝐿𝑛 𝑀𝑎𝑟𝑘𝑒𝑡 𝑐𝑎𝑝𝑖𝑡+ 𝛽3𝐵𝑜𝑜𝑘 𝑡𝑜 𝑚𝑎𝑟𝑘𝑒𝑡𝑖𝑡+ 𝛽4𝑅𝑂𝐴𝑖𝑡+ 𝛽5𝛥𝑅𝑂𝐴𝑖𝑡+ 𝛽6𝑃𝑟𝑖𝑜𝑟 𝑟𝑒𝑡𝑢𝑟𝑛 − 1𝑖𝑡+ 𝛽7𝐶𝐸𝑂 𝑎𝑔𝑒𝑖𝑡+ 𝛽8𝐶𝐸𝑂 𝑡𝑒𝑛𝑢𝑟𝑒𝑖𝑡+ 𝛽9𝑁𝑒𝑤 𝐶𝐸𝑂𝑖𝑡+ 𝛽10𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑚𝑒𝑑𝑖𝑎𝑛 𝐶𝐸𝑂 𝑝𝑜𝑤𝑒𝑟𝑖𝑡+ 𝛽11𝐼𝑛𝑑𝑢𝑠𝑟𝑦 𝑚𝑒𝑑𝑖𝑎𝑛 𝑐𝑜𝑛𝑠𝑢𝑙𝑡𝑎𝑛𝑡 𝑢𝑠𝑒𝑖𝑡+ 𝛽12𝐶𝐸𝑂 𝑐ℎ𝑎𝑖𝑟𝑖𝑡+ 𝛽13𝐵𝑜𝑎𝑟𝑑 𝑠𝑖𝑧𝑒𝑖𝑡+ 𝛽14% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒 𝑑𝑖𝑟𝑒𝑐𝑡𝑜𝑟𝑠𝑖𝑡+ 𝛽15% 𝐵𝑜𝑎𝑟𝑑 𝑏𝑢𝑠𝑦𝑖𝑡 + 𝛽16% 𝐵𝑜𝑎𝑟𝑑 𝑜𝑙𝑑𝑖𝑡+ 𝛽17% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑠 𝑎𝑝𝑝𝑜𝑖𝑛𝑡𝑒𝑑 𝑏𝑦 𝐶𝐸𝑂𝑖𝑡+ 𝜃1𝐶𝐸𝑂 𝑝𝑜𝑤𝑒𝑟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖 + 𝜃2𝐿𝑛 𝑀𝑎𝑟𝑘𝑒𝑡 𝑐𝑎𝑝̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖 + 𝜃3𝐵𝑜𝑜𝑘 𝑡𝑜 𝑚𝑎𝑟𝑘𝑒𝑡̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖 + 𝜃4𝑅𝑂𝐴̅̅̅̅̅̅𝑖+ 𝜃5𝛥𝑅𝑂𝐴̅̅̅̅̅̅̅̅𝑖+ 𝜃6𝑃𝑟𝑖𝑜𝑟 𝑟𝑒𝑡𝑢𝑟𝑛 − 1̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖+ 𝜃7𝐶𝐸𝑂 𝑎𝑔𝑒̅̅̅̅̅̅̅̅̅̅̅̅𝑖+ 𝜃8𝐶𝐸𝑂 𝑡𝑒𝑛𝑢𝑟𝑒̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖+ 𝜃9𝑁𝑒𝑤 𝐶𝐸𝑂̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖 + 𝜃10𝐶𝐸𝑂 𝑐ℎ𝑎𝑖𝑟̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖+ 𝜃11𝐵𝑜𝑎𝑟𝑑 𝑠𝑖𝑧𝑒̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖+ 𝜃12% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒 𝑑𝑖𝑟𝑒𝑐𝑡𝑜𝑟𝑠̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖+ 𝜃13% 𝐵𝑜𝑎𝑟𝑑 𝑏𝑢𝑠𝑦̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖 + 𝜃14% 𝐵𝑜𝑎𝑟𝑑 𝑜𝑙𝑑̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅𝑖 + 𝜃15̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅% 𝑂𝑢𝑡𝑠𝑖𝑑𝑒𝑟𝑠 𝑎𝑝𝑝𝑜𝑖𝑛𝑡𝑒𝑑 𝑏𝑦 𝐶𝐸𝑂𝑖 + 𝜉𝑖 + 𝜆𝑡+ 𝑢𝑖𝑡. (3)
Where 𝛽𝑘 are the fixed-effects estimates and 𝜃𝑘 are the cluster mean estimates, with 𝜉𝑖 being the individual unobserved effects and 𝜆𝑡 representing year dummies.
3.1.2. Endogeneity issues
For the first and second hypothesis holds that one key assumption of OLS regression is that the independent variables are uncorrelated with the error term. If correlation exists, then these OLS estimates will be incorrect. This omitted variable bias problem is the major difficulty of observational data. Advantage can be taken of the longitudinal design of the database to eliminate some of this unobserved heterogeneity by designing a fixed effects model as mentioned above. First, a set of dummy terms will be defined for which holds that it is equal to one if the observation comes from company i and zero otherwise or equal to one if the observation is from year t and zero otherwise. These dummy variables allow for fitting a term for every individual or year. Since there are multiple observations per individual over several years, the combined entity and time fixed effects regression model eliminates omitted variable bias arising from both unobserved variables that are constant over time and from unobserved variables that are constant across entities (industries in this study). Essentially, trying to explain variation within individuals is the objective. By adding these dummy variables to the OLS models, the fixed effect models are ready for use. Furthermore, HAC, clustered standard errors at the firm level will be used to allow for correlation within a cluster/entity, but assume uncorrelated error terms across clusters. Including these clustered standard errors will provide for valid standard errors in the model.
For the third hypothesis holds that the binary dependent variable requires a more advanced approach than the fixed effects model to get rid of endogeneity problems. These issues are taken care of by allowing the time invariant fixed effects to correlate with the independent variables that vary over time and differ per firm via so called conditional independence. Therefore, in order to eliminate time invariant fixed effects (firm specific effects) which mitigates omitted variable bias, a Correlated Random Effects Probit model is used.
3.1.3. Reverse causality/Simultaneity
For the hypothesis that powerful CEOs increase the probability of firms hiring compensation consultants however, it might be the case that the relative power of the CEO affects the use of compensation consultants but it could well be the case that the
- 14 -
use of compensation consultants influences the relative power of the CEO to the other executives, e.g. trough change in executive compensation levels on recommendation of consultants. This problem could be solved by using an instrument that is correlated with the independent variable of interest but uncorrelated with the error term in the model. No appropriate instruments are used in previous studies to deal with reverse causality. Hence, no instrument is readily available to use in this study. Using a compensation consultant or not is voluntary and up to firms to decide. Therefore, it is not the result of an exogenous or random assignment as already mentioned by Armstrong et al. (2012) and Conyon et al. (2009). CEO power is also not exogenous since it depends on the proportion of CEO pay to the pay of the other executives. These pay packages have similarities and are well discussed rather than randomly assigned. Since this is a limitation of this research, there might be a niche here for future studies to jump into and test this hypothesis with a different method of research.
3.2. Data and descriptive statistics
3.2.1. Sample
Data on compensation consultants comes from the private database of Torsten Jochem6 and was collected by automatically scraping 10-K filings with an algorithm over the period 2006 – 2013. It contains data on a number of firms and who provides their compensation consultant services in a given year (if any). The data is adjusted to be able to differentiate between specialist firms, multi-service firms and related spin-offs if necessary. This is done according to the classification in the Chu et al. (2015) paper. As in Armstrong et al. (2012) and Chu et al. (2015), firms with fiscal years ending before December 2006 are excluded to ensure that the effects of the disclosure requirements set by the Securities and Exchange Commission (SEC) had entered. These demanded to disclose which compensation consultants provided advice to the firm. Furthermore, the sample contains data until 2013 since the data on compensation consultants was only available up to and including 2013.
The data on compensation consultants is merged with data from ExecuComp and Compustat North America based on a unique identifier – which is assigned to each company – and fiscal year to exclude firms from the private database that are not also
6 Dhr. Dr. T. (Torsten) Jochem is an assistant professor of Finance at the UvA, Amsterdam Business
in ExecuComp. Furthermore, including these control variables for economic and governance characteristics that may affect executive pay, is to ensure that the compensation data and CEO characteristics are more similar to prior studies such as Armstrong et al. (2012) and Chu et al. (2015). In addition, all firms that are in ExecuComp/Compustat North America, but are not also in the private database, will be marked as not-using compensation consultants. These firms will be the “zero” part of the dummy variable whether a firm is using a compensation consultant or not.
Stock market returns are collected from the CRSP/Compustat Merged database and are also merged with the other data based the unique firm identifier and fiscal year. The returns are annualized since all data contains yearly observations.
Firms that do not occur for three or more consecutive years are dropped from the sample since those are likely to not contribute significantly to the model. The final sample consists of 1,580 unique public US firms and 9,644 firm-year observations from 2007-2013.
3.2.2. Variables
Following Bebchuk et al. (2011), CEO power (measured by the CEO pay slice) in this study is computed using data from Compustat’s ExecuComp database for the period 2007 to 2013. This main variable of interest is based on the total compensation to each executive, including salary, bonus, other annual pay, the total value of restricted stock granted that year, the Black and Scholes value of stock options granted that year, long-term incentive payouts, and all other total compensation.7 CEO power is equal to the percentage of the total compensation to all executives that goes the CEO (Total CEO compensation normalized by the total compensation of all executives in a particular year). According to Core et al. (1999) demand for managerial talent and according to Armstrong et al. (2011) and Cadman et al. (2010) the level of executive compensation might highly differ depending on industries. Therefore, the industry median CEO power – based on the four-digit standard industrial classification (SIC) group – will serve as a control variable for determining CEO power.
Armstrong et al. (2012) and Chu et al. (2015) serve as the main examples when determining the variables for economic and governance characteristics. The following controls for firm characteristics that may influence compensation levels will be used.
- 16 -
The natural logarithm of market capitalization at the beginning of the fiscal year is used to control for firm size (Log Market Cap); Book-to-market ratio at the beginning of the fiscal year which serves as an inverse measure of the firm’s investment opportunities (Book-to-market); return on assets for the prior fiscal year (ROA), which is derived by normalizing operating income after depreciation (Compustat OIADP) for the prior fiscal year by total assets at the end of the prior fiscal year; also the change in return on assets (ΔROA) between the current fiscal year and the prior fiscal year is taken into account; and the annualized raw stock return of the prior fiscal year is also included (Prior return -1).
As for the governance variables, Bebchuk et al. (2011) is followed by including a dummy variable for whether the CEO is also chairman on the board of directors (CEO chair) which is derived from the ISS - formerly RiskMetrics database - because some studies argue that agency problems increase when the CEO is also the board chair (Yermack, as cited in Core et al., 1999). Consistent with prior studies, other governance controls that are related to excess CEO compensation are also included such as the number of directors on the board (Board size), % outside directors (% of the board members who are classified as independent), % board busy (serving on at least two boards), % board old (directors that are at least 69 years old) and % of outside directors appointed by CEO (fraction of board members classified as outsiders who were appointed after the current CEO’s term started (Core et al., as cited in Armstrong et al., 2012).
Furthermore, Armstrong et al. (2012) and Chu et al. (2015) serve as examples for including CEO characteristics as those may also influence compensation levels and hence, the level of power a CEO will have. Accounted for are the number of years the CEO has been in his current role of chief executive officer (CEO tenure); the age of the CEO in years (CEO age); and whether the CEO was appointed to its position during the current fiscal year (New CEO). The latter is included because in the first year of a CEO’s tenure, firms often provide relatively large compensation packages to establish a certain level of equity incentives and to make the CEO whole for abandoned compensation from his or her prior employer.
For the use of compensation consultants, a 0,1 dummy variable is included (Compensation consultant) which is equal to one if a compensation consultant is used by a firm and zero otherwise. Also, an industry median 0,1 dummy variable is included to capture the influence of compensation consultant use by peer firms.
For all ratios hold that they are winsorized at the 1% and the 99% level when implemented in the regression models.
3.2.3. Descriptive statistics
Descriptive statistics for the full sample are reported in Appendix 1. Table 1 presents differences in descriptive statistics for firms that use compensation consultants and those that do not. In addition, Kruskal-Wallis p-values are reported. Total annual CEO compensation in the full sample ranges from $0 to $137,207,000 with a mean of $5,830,173 and a median of $4,043,865. Similar to the Armstrong et al. and Chu et al. studies, the natural logarithm of total annual CEO compensation is used in the empirical analyses because of the highly (right) skewed distribution of pay. The minimum CEO power in the sample is 0% which corresponds with the CEOs in the sample that are receiving 0 total annual compensation. In the regressions these outliers are winsorized at the 1% and the 99% level to limit their influence on the model. The number of directors on the board ranges anywhere from 4 up to a maximum of 34 members. The ratio’s that have restricted intervals between 0 and 1 are CEO power, the percentage of outside directors on the board, the percentage of board members that serve on at least two boards, the percentage of directors that is at least 69 years of age and the percentage of directors that is appointed after the CEO took its position. The CEO tenure varies between zero years and 51 years. The youngest CEO in the sample is 32 years of age and the oldest CEO is 97 years old.
Companies that use compensation consultant seem to differ substantially from firms without consultants on governance and CEO characteristics as can be derived from Table 1. The total annual CEO compensation at firms using compensation consultants is on average higher than at firms that do not use compensation consultants, as is clear from Table 1. This corresponds with the findings of Chu et al. (2015) that CEO pay levels at consultant using firms are economically and statistically significantly higher than the pay levels at non using firms. On average larger firms use the services of compensation consultants as is in line with the univariate tests from Chu et al. (2015) and Armstrong et al. (2012). Furthermore, other firm characteristics such as stock return of the prior year, return on assets and investment opportunities are not considerably different at firms with consultants. These findings are very much like the findings of Armstrong et al. (2012) and Chu et al. (2015). Companies that use compensation consultants tend to have larger boards and more directors that serve on at least two
- 18 -
boards. These elements are associated with weaker governance. However, it is more likely that they have a higher percentage of independent board members which is an indicator of stronger governance. CEOs generally have longer tenure at firms without consultants. Also, firms without consultants tend to have both a larger percentage of outside directors appointed by the CEO and older boards. These elements are claimed to indicate weaker governance. This could indicate that consultant use is higher in firms with weaker governance as found by Armstrong et al. (2012). From Table 1 it seems that CEO power and industry median CEO power are higher in firms that employ compensation consultants.
Table 1 - Descriptive Statistics (2007-2013)
This table shows the summary statistics for the firms that used and did not use a compensation consultant during the fiscal year. The sample embraces a period from 2007 – 2013. The Kruskal-Wallis p-values are displayed to report whether the mean ranks (for all variables the shape of the distributions of the two groups is not equal) between the firms that used and did not use compensation consultants are significantly different. For all variables but ROA and New CEO, this difference is significant at a 5% level.
No consultants used Consultants used
Firm-level data: full sample Mean Median Std. dev. Mean Median Std. dev. Kruskal-Wallis p-value
Firm characteristics
Total executive compensation 10,702,852 6,184,315 14,552,466 16,607,991 12,082,340 17,359,373 0.000
ROA 0.098 0.082 0.098 0.093 0.085 0.091 0.807
Log (Market capitalization) 7.349 7.101 1.540 8.080 7.897 1.729 0.000
Prior return (-1) 0.103 0.067 0.454 0.124 0.091 0.463 0.029
ΔROA -0.006 -0.001 0.059 -0.002 0.000 0.063 0.000
Book-to-market ratio 2.002 1.056 3.218 2.213 1.141 4.556 0.000
CEO characteristics
Total CEO compensation 3,893,055 2,074,545 6,575,295 6,371,517 4,625,640 6,628,059 0.000
Log (CEO compensation) 7.665 7.644 1.109 8.397 8.441 0.884 0.000
CEO power 0.360 0.356 0.136 0.387 0.393 0.108 0.000
Industry Median CEO power 0.373 0.373 0.049 0.387 0.384 0.041 0.000
CEO tenure 7.709 5.000 9.023 5.717 4.000 6.704 0.000 CEO age 57.216 56.500 8.403 56.030 56.000 6.651 0.000 New CEO 0.064 0.000 0.245 0.070 0.000 0.256 0.316 Corporate governance Board size 8.573 8.000 2.391 9.637 9.000 2.386 0.000 % Outside directors 0.730 0.750 0.119 0.798 0.818 0.104 0.000 % Board busy 0.155 0.125 0.168 0.238 0.222 0.178 0.000 % Board old 0.277 0.250 0.199 0.220 0.200 0.165 0.000 CEO is chairman 0.528 1.000 0.499 0.560 1.000 0.496 0.010
% Outsiders apptd. by CEO 0.275 0.143 0.308 0.240 0.111 0.288 0.000
From the correlation results provided in Appendix 2 it can be seen that total annual CEO compensation is highly correlated with firm size and the change in return on assets between the prior and the current fiscal year. Furthermore, compensation levels are significantly positively correlated with board size, CEO being the head of the board of directors and the percentage of busy board members. These correlations are consistent with higher CEO compensation levels in firms with weaker governance. The use of compensation consultants is positively associated with total annual CEO compensation, CEO power, Compensation consultant use by peer firms from the same industry and firm size. Furthermore, positive correlation exists with the CEO also being the chairman of the board, a large number of directors and the percentage of busy directors on the board. This suggests that large firms and firms with weaker boards are employing compensation consultants.
4. Analysis
4.1. Empirical Results
4.1.1. Economic and governance determinants of CEO compensation
Table 2 presents the results from the tests on the determinants of CEO compensation. The first models (1.1 - 1.4) are estimated only with economic and CEO characteristics. The specifications in the table include simple OLS (1.1 and 1.5), only year fixed effects (1.2 and 1.6), only industry fixed effects (1.3 and 1.7) and both year and industry fixed effects (1.4 and 1.8). In all models the standard errors are robust or clustered at the two digit industry level if applicable. The models of interest are the fixed effects models with year and industry effects included (1.4 and 1.8) since these are the most advanced and mitigate the issue of omitted variable bias. The predicted signs are based on the findings of Core et al. (1999), Armstrong et al. (2012) and Chu et al. (2015) with respect to economic, CEO and governance determinants of CEO compensation.
As can be seen from specifications 1.4 and 1.8, the majority of the model is explained by economic determinants such as firm size and prior return. Coefficient signs on most of the economic are consistent with the expectations. The influence of firm size is positively and significantly related to the compensation of CEOs as is consistent with prior studies (e.g. Armstrong et al., 2012; Core et al., 1999; Chu et al., 2015). I.e. if the Market capitalization of the firm goes up by 1%, the compensation of the CEO increases with 0.34% ceteris paribus.
- 20 -
Table 2 - Total CEO compensation regressions (2007-2013). Economic and governance
determinants.
This table provides the regression results for the economic and governance determinants of annual CEO compensation. The natural logarithm of total annual CEO compensation is the dependent variable. All other variables are as provided by Appendix 1. All ratios in the above specifications are winsorized at the 1% and the 99% level to exclude the influence of outliers. Model specifications 1.1-1.4 represent estimates of the model including only economic variables and CEO characteristics. For the specifications 1.5-1.8 governance variables are added to estimate the model and assess their incremental explanatory power. For both the specifications with and without the governance characteristics pooled OLS (1.1 and 1.5), year fixed effects (1.2 and 1.6), industry fixed effects (1.3 and 1.7) and both year and industry fixed effects (1.4 and 1.8) are reported. Standard errors are all clustered at the two-digit SIC industry level if industry fixed effects are included in the specification.
Robust t-statistics are reported in parentheses. Statistical significance at the 1%, 5% and 10% level is reported as follows ***, **, *. Log(CEO compensation) Pred. sign (1.1) (1.2) (1.3) (1.4) (1.5) (1.6) (1.7) (1.8) Intercept 5.543*** 5.492*** 5.448*** 5.428*** 4.715*** 4.758*** 4.406*** 4.479*** (53.04) (50.65) (29.46) (29.07) (43.75) (42.33) (21.82) (20.99) Log (Market capitalization) + 0.339*** 0.338*** 0.343*** 0.341*** 0.270*** 0.268*** 0.272*** 0.268***
(34.98) (34.40) (20.04) (20.07) (25.60) (24.89) (14.67) (14.75) Book-to-market ratio - 0.001 -0.001 0.036*** 0.034*** -0.005 -0.006* 0.021** 0.020** (0.27) (-0.12) (2.70) (2.67) (-1.39) (-1.65) (2.08) (2.01) ROA + -0.094 -0.097 -0.397 -0.385 0.099 0.099 -0.071 -0.059 (-0.61) (-0.62) (-1.30) (-1.25) (0.66) (0.65) (-0.23) (-0.19) ΔROA ? 0.143 0.133 0.146 0.138 0.229 0.218 0.270 0.263 (0.73) (0.67) (0.62) (0.58) (1.11) (1.05) (0.94) (0.90) Prior return -1 + 0.102*** 0.077*** 0.098*** 0.068** 0.118*** 0.096*** 0.116*** 0.091*** (4.12) (2.76) (3.76) (2.58) (5.21) (3.79) (4.61) (3.63) CEO age + 0.001 0.000 0.001 0.000 0.001 0.001 0.002 0.001 (0.74) (0.16) (0.43) (0.09) (1.09) (0.37) (0.76) (0.36) CEO tenure ? -0.005*** -0.007*** -0.004* -0.006** -0.012*** -0.012*** -0.009* -0.009* (-3.83) (-4.48) (-1.88) (-2.38) (-4.46) (-4.65) (-1.89) (-1.97) New CEO + -0.027 -0.032 -0.027 -0.030 -0.006 -0.015 -0.009 -0.017 (-0.80) (-0.93) (-0.79) (-0.90) (-0.19) (-0.47) (-0.25) (-0.50) CEO is Chair + 0.076*** 0.096*** 0.067** 0.088*** (4.53) (5.44) (2.28) (2.78) Board size + 0.033*** 0.033*** 0.046*** 0.046*** (7.03) (7.08) (3.96) (4.00) % Outside directors - 0.903*** 0.820*** 1.095*** 1.007*** (10.78) (9.61) (7.54) (6.70) % Board busy + 0.096* 0.057 0.094 0.052 (1.85) (1.07) (0.83) (0.45) % Board old + 0.946*** 0.965*** 0.761*** 0.785*** (18.12) (18.48) (8.39) (8.56) % Outside directors apptd.
by CEO + 0.327*** (6.04) 0.296*** (5.46) 0.261*** (2.82) 0.229** (2.47) Sample size 9,176 9,176 9,176 9,176 9,176 9,176 9,176 9,176
(Adj.) R-squared (in %) 36.15 36.59 41.74 42.15 41.77 42.15 46.84 47.18
Year fixed effects No Yes No Yes No Yes No Yes
Industry fixed effects No No Yes Yes No No Yes Yes
Furthermore, the company’s stock return of the prior year is also positively and significantly related to CEO pay which is in line with the findings of Armstrong at al. (2012) and Core et al. (1999) and implies – ceteris paribus – a 6.8% increase in pay if the stock return over the past year has increased by one percentage point. Contrary to what prior studies find on the influence of investment opportunities - as measured by the Book-to-Market ratio - in this study it is positively and significantly related with CEO pay levels at the 5% level. This implies that when a firms investment opportunities are low, CEOs are compensated more for their services. According to prior studies, lower Book-to-Market ratios reflect investments opportunities since factors that are not (yet) captured in the book value of the firm’s assets are reflected by the market valuation of the company and its future cash flows (Armstrong et al., 2012). In contrast to prior studies the coefficient for CEO tenure is significant (albeit at the 10% level). The sign indicates that the longer the CEO is in its position, the lower its compensation. This is contrary to what one would expect when a CEO remains longer in its position. The CEO might sign improved contracts for which his compensation is expected to increase.
When the governance characteristics are added to the model (specifications 1.5 - 1.8), the explained variance in the model increases from 42.15% (in specification 1.4) to 47.18% (in specification 1.8). A conducted F-test on the joint significance of the added governance determinants points out that this increase in explanatory power is highly significant (Prob. > F = 0.0000). The majority of governance determinants are statistically significant with the exception of % Board busy. Pay tends to be significantly higher (8.8%) if the CEO is also chairman of the board – ceteris paribus – which is in line with findings of prior studies (e.g. Core et al., 1999; Chu et al., 2015). The number of directors on the board, the percentage of directors that are at least 69 years old and the percentage of outside directors that were appointed after the CEO took his position are also positively and significantly related to CEO compensation. All of these variables imply that firms with weaker governance tend to provide their CEOs with higher compensation which supports the results of prior studies. An alternative interpretation is that the CEO is able to extract additional pay from the company when governance is weak as mentioned by Core et al. (1999). E.g. if the number of directors increase with one person, CEO pay tends to increase by 4.6%. Furthermore, the percentage of outside directors on the board implies to have a positive relation with the compensation of CEOs but this is contrary to its expected sign. Since it is anticipated that the higher the amount of outside directors on the board, the better the governance,
- 22 -
CEO pay is expected to decrease if % outside directors increases with one percentage point.
The majority of the results described above is in line with findings from prior studies and the expectations according to hypothesis 1 of this study. Therefore it can be stated that total annual CEO compensation is mostly dependent on economic determinants. This is understandable since large firms with good firm performance and high sales are able to compensate their CEOs and executives more generous than smaller firms, even if large firms show signs of weak governance and smaller firms meet requirements of strong governance. Nevertheless, weaker governance would allow CEOs to benefit more from extraction of extra pay or even carry out more power within the firm which is hypothesized by various prior studies (Core et al. 1999; Chu et al. 2015; Armstrong et al. 2012). Hence, conform Core et al. (1999) greater agency problems seem to positively influence the compensation levels of CEOs.
4.1.2. Economic and governance determinants of CEO power
Next the determinants of CEO power are examined in Table 3. For these estimations, the same control variables and the same design are used as in hypothesis 1 with the addition of Industry Median CEO power. Since the demand for managerial talent and therefore the rewards of the CEO might be industry specific, the median CEO power (based on four-digit SICs) is added to capture the effect of CEO power at comparable firms from the same industry. The latter is accountable for the largest determination of CEO power8. When estimated only with economic and CEO characteristics CEO power tends to increase with 0.84 percentage points if median CEO power in the peer industry increases by one percentage point which can be seen in Table 3. This relation is consistent with the predicted sign of the coefficient and the results from Bebchuk et al. (2011). Furthermore, from Table 3 it can be retrieved that both stock return from the prior year and Book-to-Market ratio have the predicted significant relations with CEO power. Ceteris paribus, a one percentage point increase in prior year’s stock return implies that the relative power of the CEO to other executives increases with 0.001 percentage points.
8 Industry Median CEO power (the power a CEO has in peer firms from the same industry) explains the
most of the variation in CEO power amongst the other economic determinants. It shows an adjusted R² of 12.41% when included solely in an untabulated model.
Table 3 - CEO power regressions (2007-2013). Economic and governance determinants.
This table shows the regression results for the economic and governance determinants of CEO power. CEO power - winsorized at the 1% and 99% level to mitigate influence of extreme values - is the dependent variable. All other variables are as provided by Appendix 1. All other ratios in the above specifications are winsorized at the 1% and the 99% level to exclude the influence of outliers. Model specifications 2.1-2.4 represent estimates of the model including both economic variables and CEO characteristics. For the specifications 2.5-2.8 holds that governance variables are added to estimate a more extended model and to assess their incremental explanatory power. For both the specifications with and without the governance characteristics pooled OLS (2.1 and 2.5), year fixed effects (2.2 and 2.6), industry fixed effects (2.3 and 2.7) and both year and industry fixed effects (2.4 and 2.8) are reported. Standard errors are all clustered at the two-digit SIC industry level if industry fixed effects are included in the specification.
Robust t-statistics are reported in parentheses. Statistical significance at the 1%, 5% and 10% level is reported as follows ***, **, *. CEO power Pred. sign (2.1) (2.2) (2.3) (2.4) (2.5) (2.6) (2.7) (2.8) Intercept 0.048*** 0.045*** 0.056** 0.054** -0.013 -0.009 -0.008 -0.004 (3.21) (3.00) (2.33) (2.26) (-0.80) (-0.52) (-0.33) (-0.15) Log (Market capitalization) + 0.001 0.001 0.001 0.001 0.000 -0.000 -0.000 -0.001 (1.51) (0.84) (0.71) (0.36) (0.24) (-0.26) (-0.02) (-0.31) Book-to-market ratio - -0.001*** -0.001*** -0.002*** -0.002*** -0.001** -0.001* -0.002*** -0.002*** (-2.86) (-2.66) (-3.41) (-3.24) (-2.00) (-1.85) (-4.67) (-4.55) ROA + 0.008 0.015 0.011 0.018 0.015 0.019 0.018 0.023 (0.44) (0.80) (0.31) (0.50) (0.82) (1.06) (0.50) (0.64) ΔROA ? 0.029 0.027 0.032 0.030 0.033 0.031 0.034 0.032 (1.03) (0.95) (1.61) (1.52) (1.24) (1.15) (1.63) (1.55) Prior return -1 + 0.012*** 0.011*** 0.012*** 0.001*** 0.012*** 0.011*** 0.012*** 0.010*** (4.89) (3.76) (4.84) (3.44) (4.96) (3.93) (4.85) (3.57) CEO age ? -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 (-0.64) (-1.22) (-0.43) (-0.72) (-0.77) (-1.20) (-0.44) (-0.63) CEO tenure + 0.001*** 0.001*** 0.001* 0.001 -0.000 -0.001 -0.001 -0.001 (4.87) (3.80) (1.99) (1.57) (-1.20) (-1.39) (-0.66) (-0.74) New CEO ? -0.006 -0.007 -0.006 -0.007 -0.004 -0.005 -0.004 -0.005 (-1.21) (-1.40) (-1.10) (-1.29) (-0.87) (-1.09) (-0.82) (-1.04) Industry median CEO power + 0.852*** 0.854*** 0.839*** 0.841*** 0.785*** 0.787*** 0.779*** 0.781***
(30.34) (30.72) (20.87) (21.26) (28.18) (28.41) (21.50) (21.39) CEO is Chair + 0.006** 0.007*** 0.006* 0.008** (2.43) (3.10) (1.94) (2.46) Board size + -0.003*** -0.003*** -0.003*** -0.003*** (-5.03) (-4.97) (-3.43) (-3.42) % Outside directors - 0.130*** 0.122*** 0.138*** 0.130*** (11.59) (10.69) (7.01) (6.49) % Board busy + 0.028*** 0.024*** 0.025*** 0.028*** (4.12) (3.54) (2.83) (3.17) % Board old + 0.017** 0.020*** 0.029** 0.025* (2.52) (2.91) (2.27) (1.91) % Outside directors apptd. by
CEO + 0.042*** (5.78) 0.039*** (5.41) 0.043*** (3.20) 0.041*** (2.99) Sample size 9,198 9,198 9,198 9,198 9,198 9,198 9,198 9,198
(Adj.) R-squared (in %) 12.81 13.15 13.49 13.84 15.29 15.47 15.60 15.73
Year fixed effects No Yes No Yes No Yes No Yes
Industry fixed effects No No Yes Yes No No Yes Yes
- 24 -
The negative coefficient for the inverse measure of investment opportunities
(Book-to-Market ratio) shows that if the investment opportunities increase by one percentage
point, CEO power increases with 0.002 percentage points. Therefore it seems that positive firm performance and future prospects are enhancing the relative power of the CEO. When governance variables are added to the model (specifications 1.5 – 1.8) the coefficient signs of the economic determinants stay the same although they change somewhat in magnitude. The predictions that weaker governance is positively related to the relative power of the CEO are consistent with the results in specification 1.8 and with the interpretation that a CEO could use its power and influence to increase CEO compensation with respect to the compensation of the other executives as is hypothesized by Bebchuk et al. (2011). When the CEO is also chairman of the board, CEO power increases with 0.008 percentage points at a 5% significance level – ceteris paribus – which corresponds to the result of Bebchuk et al. (2011). If the percentage of directors that serves at least on two or more boards increases with one percent point the relative power of the CEO increases with 0.028 percentage points. Furthermore, %
Outside directors apptd. by CEO also shows a positive relation with CEO power.
Contrary to the expectations Board size shows a negative coefficient which is significant at the 1% level and the percentage of outside directors on the board shows a positive coefficient which is significant at the 1% level. Both findings are contradictory to their relation with weaker governance (a larger number of directors on the board) and stronger governance (more outside directors on the board) as is stated in prior studies (Armstrong et al., 2012; Core et al., 1999). When CEOs can apply their power and influence to extract compensation or to influence extreme decision making this is perceived as a sign of weaker governance.
Based on the above described results the hypothesis that CEO power is largely founded on determinants of weaker governance is not confirmed. CEO power is mostly determined by the proportions of relative CEO power to the other executives in peer industry firms. However the majority of governance characteristics support the influence of weaker governance on CEO power and only a few show contradictory results.
4.1.3. CEO power and the use of compensation consultants
Appendix 3 provides the results for the tests on the determinants of compensation consultant use. In this Correlated Random Effects Probit model the main variable of interest CEO power shows no significant relation with the use of compensation consultants which is not in line with the univariate analysis from Appendix 2. Higher relative CEO power to the other executives could not really be interpreted as a governance issue but more as a CEO characteristic. However, the industry median of CEO power seems to have a significantly positive association with firms employing compensation consultants. E.g. for a given firm, a one percent point increase in the median of industry CEO power increases the predicted probability of the firm using a compensation consultant with 3.95, ceteris paribus. Also the use of compensation consultants by peer firms from the same industry shows a large significant relation with firms using consultants. These findings correspond to their predicted signs based on statements of prior studies that there might be differences in CEO pay, consultant use, economic performance and governance characteristics based on industry specific differences. In addition, the results imply that the probability of employing compensation consultants increases with firm size, which is in line with the univariate analysis of Table 1 which implies that firms employing compensation consultants are on average larger. Furthermore it is conform the findings of Armstrong et al. (2012) and Chu et al. (2015) that larger firms use compensation consultants. I.e. if for a given firm the market capitalization increases with 1 percent, all else equal, the predicted probability of hiring a compensation consultant increases with 0.16.
None of the governance characteristics in Appendix 3 seem to have a significant relation with consultant use by firms except for CEO also being chairman of the board. Except for the CEO is Chair variable, these results are not conform the expectations formed based on prior studies, which showed that firms with weaker governance were more likely to increase the probability of firms employing consultants to justify their pay levels. These findings are also not in line with the expectations from the univariate analyses in Table 1 and Appendix 2. These tables show that board size and the percentage of busy board members – which are assumed to be weak governance indicators according to Armstrong et al. (2012) and Core et al. (2012) – have positive relation with the use of compensation consultants.
The estimated coefficients of the cluster mean variables are not so important to discuss extensively since these are extra control variables that merely model the
