1
Amsterdam Business School
Changing ratio of CEO pay to average production worker
within large U.S. public companies with focus on the
Manufacturing and Information sectors
Name: Lilit Yeritsyan Student number: 10522840 Date: 20-06-2014
Final paper
MSc. Accountancy & Control, specialization Control
Faculty of Economics and Business, University of Amsterdam First Supervisor : Ir. drs. A.C.M. de Bakker
2
Abstract
In 2013 the Securities and Exchange Commission (SEC) initiated a the acceptance of a new rule that requires public companies to disclose the remuneration of its chief executive officer and the median compensation of its employees. For the past 25 years the gap between average employee salary and CEO remuneration increased dramatically. This gap presents a biggest challenge of the modern economies, since it creates social cohesion issues and directly affects the employee motivation. Several reasons behind this increase mentioned in the previous research include size of the company, industry a company operates in, offshore operations, company profitability and earnings, and so on. This research is focused on the differences of the CEO/employee pay gap across industries. Manufacturing and Information sectors are selected for the purpose of inter-sectorial comparison, since the pay ratio is one of the biggest within Manufacturing and one of the smallest within Information sector. Wharton Research Data Services is used to obtain yearly information on S&P 500 large U.S. corporations from each of Manufacturing and Information sectors. Cross-sectional regression analysis is used as a main tool for identifying the magnitude of pay gaps in the selected industries.
3
Table of content
1. Introduction ... 6
1.1. Background ... 6
1.2. Development of CEO relative pay ... 6
1.3. Differences per sector... 7
1.5. Research question ... 8
1.6. Research Method ... 8
1.7. Structure of the Paper ... 9
2. Literature review... 10
2.1. Effects of income inequality ... 10
2.2. CEO Compensation ... 10
2.3. Theoretical framework ... 11
2.3.1. Agency theory ... 11
2.3.2. Managerial power theory ... 12
2.3.3. Technological change and inequality in managerial compensation (paradox of productivity and paradox of wage inequality) ... 12
2.4. The determinants and effects of CEO-employee pay ratio ... 13
2.5. Assessing the impact of financial crisis ... 14
3. Hypotheses development ... 15
3.1. Research Model ... 15
3.2. Hypotheses ... 15
3.3. Control factors ... 17
3.3.1. Compensation Surveys and the Relation Between CEO Pay and Firm Size ... 17
3.3.2. Compensation Surveys and the Relation Between CEO Pay and Firm Performance ... 17
3.4. Industry related factors ... 18
3.4.1. Manufacturing sector ... 18
3.4.2. Information sector ... 18
3.5. Operationalization of the variables in the model ... 20
3.5.1. Dependent Variable ... 20
3.5.2. Independent Variables ... 20
4. Data collection ... 22
4.1. Introduction ... 22
4.2. Development and Use of the Variables in the Model ... 22
4.2.1. Calculating CEO pay ratio ... 22
4.2.2. Independent and Control variables ... 24
4
5. Selection of the right model for panel data analysis ... 27
5.1. Tests performed ... 27
5.1.1. Random Effects Regression vs. Fixed Effects Regression ... 27
5.1.2. Random Effects Regression vs. OLS Regression ... 27
5.1.3. Wooldridge test for Autocorrelation in Panel Data ... 27
5.1.4. Test for Heteroskedasticity in Panel Data... 28
5.2. Log transformation of variables with skewed destruction ... 29
6. Data analysis ... 30
6.1. Descriptive Statistics ... 30
6.1.1. Correlation analysis ... 32
6.1.2. Detection of Multicollinearity in the model ... 33
6.2. Development of Hypothesis Testing Methodology ... 34
6.3. Empirical Results Associated with the Random Effects Panel Model ... 35
6.3.1. Results of Empirical Hypothesis testing ... 37
7. Summary and concluding remarks ... 39
7.1. Summary of the Results ... 39
7.2. Limitations of this study ... 40
7.2.1. Some key factors are left out of the model ... 40
7.2.2. Panel data is unbalanced ... 40
References ... 41
Appendix A. CEO Pay Ratio ( Bloomberg) ... 45
Appendix B. Log transformation of variables with skewed destruction ... 46
Appendix C. Industry related tables ... 48
1. Structure of NAICS ... 48
2. Table 6.3D. Wages and Salaries by Industry ... 49
3. Table 6.2D. Compensation of Employee by Industry ... 50
5
List of Tables
Table 3-1. Description of the factors used in the model ... 19
Table 3-2. Description of independent variables ... 20
Table 3-3. Models representing the hypotheses: ... 21
Table 4-1. Data collection procedure ... 24
Table 4-2. Database size by stage and year ... 25
Table 4-3. Available data per sector ... 26
Table 5-1. Summary of tests performed to choose the right model and check for autocorrelation and heteroskedasticity ... 28
Table 6-1. Descriptive Statistics of the continuous variables ... 31
Table 6-2. Number and relative frequency of companies by sector and gender of the CEO ... 31
Table 6-3. Correlation Matrix ... 32
Table 6-4. VIF score for the three models with total revenue included and without total revenue ... 33
6
1.
Introduction
1.1. Background
Section 953 (b) of the Dodd-Frank Consumer protection and Wall Street Reform Act, known as the “pay ratio provision”, requires companies to disclose the ratio of pay between chief executive officer (CEO) and a median employee. Therefore Securities and Exchange Commission (SEC) was required to compose rules to implement an obligatory disclosure of the ratio between the total compensation of a company’s CEO and the median compensation of all employees. The pay ratio provision is supported by different groups as labour groups, pension funds, unions. On September 18, 2013 the Securities and Exchange Commission (SEC) voted to propose a new rule that would require public companies to disclose the compensation of its chief executive officer (CEO) and the median compensation of its employees. A firm will be able to choose its own methodology to calculate median pay of its employees.
The issue of the compensation of top executive officers is often raised in the academic and business research, because of the drastic increase of CEO compensation over the last 25 years. Many reports show that the gap between salary of an average production worker and CEO compensation (including a salary, bonuses, and shares or call options on company stocks) is continually growing.
This information is important for investors, since various academic studies have found that high pay gap can hurt employee morale, reduce productivity and increase employee turnover. Growing income inequality causes higher rates of health and social problems and harms economic growth (Richard Wilkinson and Kate Pickett, 2011). Many researchers documented a link between income inequality and social tension.
Different academic papers related to this issue have examined the particular variables that may cause top executives’ compensations increases, as well as factors influencing the growing gap between payments to the management and average employees. These factors can be internal or external and may include company size, company earnings and value, stock price, level of company activities outsourcing, company structure and policies, taxation procedures, technological development, and globalization. Previous study also suggests some other factors that can influence increasing ratio between CEO pay and average employee salary; board composition (Hoitash et.al., 2009), internal control material weaknesses (Hoitash et.al., 2012), firm performance (Core et al., 1999).
1.2. Development of CEO relative pay
According to a 2005 report by United for a Fair Economy and the Institute for Policy Studies, the average CEO in the U.S. earned 431 times the average pay of a production (i.e., non-management) worker in 2004, up from 301 times in 2003 and 42 times in 1982 (Executive Excess 2005, 12th Annual CEO Compensation Survey). Economic Policy Institute introduces own data according to which since 1978 CEO pay at American firms has risen 725 percent, much faster than the worker pay over the same period and much faster than the stock market growth.
The academic literature of corporate governance examined this phenomenon. Some evidence revealed that the growing CEO pay is a result of an insufficient and weak corporate governance
7
structure. In the contrary some other authors suggested that the compensation of CEO increases with the complexity of the job, firm size, and the growing globalisation. There are also evidences that income growth since 2007 has been very unbalanced as profits have reached record highs and, correspondingly, the stock market has boomed while the wages of most workers (and their families’ incomes) have declined over the recovery period (Mishel et al., 2012, Mishel 2013).
1.3. Differences per sector
The research of FTSE 350 companies conducted by Hay group in 2011 revealed significant variation of the pay ratio between sectors. Pay gap is the smallest in financial services, technology and energy sectors, where the average employee is highly educated, highly skilled and relatively mobile. In contrast, the largest relative pay gap was observed in retail and production companies, where most employees are low skilled, have basic education and are often paid by hour.
Bloomberg makes its own analysis on CEO relative pay ratio per sector based on U.S. government data on worker compensation per industry. Although the Bloomberg.com suggests that there were some limitations in the process of data collection, when some companies did not provide or reveal the average salaries of their employees, nevertheless this data was obtained from Governmental sources and can be considered as correct data with small error margin. The table information in Appendix 1 shows large differences of CEO and other workers’ compensations. It is easy to notice workers’ salary differences across different industries; salaries in clothing and merchandize are under USD 35,000, while workers’ salaries in financial sector are over USD 80,000. The broadcasting services reported salaries at around USD 60,000. Clearly there are differences across sectors of the Economy. Ample theories developed around this topic also documented that the size of gap between CEO and average employee compensation can be different across industries.
We can conclude that the relative pay difference is higher within firms in homogenous industries and in the retail, where employees have low skills, are easy to replace and thus are less powerful relative to management. Similarly, manufacturing firms that employ large number of workers in low-wage countries may also have higher relative CEO pay ratio. As opposed to that in high-technology firms the human resources are valuable, rare, mobile, hard to imitate and have skills and knowledge that make them irreplaceable. Attracting talented specialists who can contribute to success of the company in the financial, technological and energy sectors becomes a great challenge, so the employees in those industries have higher bargaining power.
1.4. Financial crisis
The financial crisis also known as the Global Financial Crisis started in 2007 is considered by many economists the worst financial crisis since the Great Depression of the 1930s.Throughout the world, by the end of 2008, many banks had seen most of their equity destroyed by the crisis that started in the U.S. subprime sector in 2007 (Beltratti & Stulz, 2012) . Since the crisis started in the banking sector various papers examined the relationship of the bank performance and CEO Compensation. There are evidences found that bank performance was related to chief executive officer (CEO) incentives before the crisis (Fahlenbrach &Stulz, 2010).
8
Looking through all the literature on company performance, one cannot ignore the effect of the financial crisis. According to Gabaix et.al (2013) during the crisis, the average total value of 500 biggest companies in the U.S. declined by 17% - which was the sharpest decline in the last forty years. Empirical research suggests that as a result of the financial crisis the average salaries in those companies have declined by 28% (Gabaix et al., 2013).
1.5. Research question
The goal of this research is to analysing magnitude of differences of CEO and average employee pay gap in the light of a specific factor determining this gap.
After intensive literature review it was revealed that the growing income inequality is driven by different factors, such as rising inequality of wages within different sectors, rising inequality of capital, which caused incomes being invested in capital rather that in labor and so on. The pay gap is one of the biggest challenges of the modern economies, since it decreases economic growth in developed countries and may cause social cohesion issues. The CEO and average employee pay ratio has also a great importance for the shareholders, since in the company with high pay gap employees are less motivated and more frustrated.
Another key factor we focus in this research is the sector differences, since there is obviously big difference in the CEO relative pay between different sectors. This conclusion was derived from observing CEO pay ratio table published by Bloomberg (see Appendix 1). Although there are many research papers examining the pay gap phenomenon, only limited academic studies make a comparative examination of CEO relative pay within different sectors. Manufacturing and Information Technologies sectors are selected for the purpose of cross-sectional time-series comparison, since the pay ratio is one of the biggest within Manufacturing and one of the smallest within Information sector. This analysis tries to reveal the magnitude of the difference in the CEO and average workers’ compensation gap between Information and Manufacturing industries.
In the light of all these considerations the research question is:
Is the CEO and average employee pay gap (or pay ratio) significantly different across Manufacturing and Information industries and what was the influence of the financial crisis?
1.6. Research Method
The research is based on secondary data analysis. It includes an in-depth literature review and model estimation using publicly available data on the S&P 500 corporations from both Manufacturing and IT sectors during both pre-crisis period (2006 and 2007) and crisis period (2008 and 2009). Specifically, Wharton Research Data Services is used to obtain yearly information on these companies from 2006 to 2009.
A panel research is conducted among S&P 500, which enables us to see the changes taking place over time and gives us a stable basis for comparison. A panel data allows measuring the changes over time: before the financial crisis and during the financial crisis.
9
A model is developed and estimated through random effect panel regression model to shed light on the research question and examine the impact of other key factors on the pay ratio.
STATA professional statistical software is used to combine and analyse the data and estimate the model.
If the likelihood of entity losses appears to be great as a result of mergers, takeovers or acquisitions, the replacements should be included in the model. Ideally, it is expected to have a balanced dataset, which means that all entities have data for all years. In the reality often the datasets are unbalanced; however we still can run the model (Oscar Torres-Reyna).
1.7. Structure of the Paper
This paper is be divided into 5 Sections. In Section 2 the prior literature on CEO compensation ratio and the pay of the average employee is studied and the hypotheses pertaining to the research question are developed. The methodology of this research and hypotheses testing mechanism is development next in Section 3. In Section 4, the data, the sample choice variables, and the data collection procedure are described. In Section 5 tests are performed and the final model is developed. In Section 6 the relationship between the variables and the compensation ratio of the two industries is analysed. At the end, a summary and conclusions are provided in Section 7.
10
2.
Literature review
2.1. Effects of income inequality
Growing income inequality can have a big impact on social, economic and political developments within OESO countries. Income inequality is associated with morality across countries. Various studies have examined the impact of the income inequality and health, criminality, morale in high income countries. The conclusions are that equal societies have better health and lower criminality (Hales, Howden-Chapman, Salmond, Woodward & Mackenbach, 1999; Kaplan, Pamuk, Lynch, Cohen & Balfour, 1996; Lynch et al. 2001; Wilkinson & Pickett, 2011; Avendano, 2011).
The impact of the CEO and median employee pay gap is also significant within organizations, since high gap can reduce productivity and increase employee turnover. High CEO pay gap can also work against certain firm competitive moves, especially when these actions require greater levels of coordination, foresight, and integration of diverse perspectives (Gnyawali et al., 2008). In this sense it is important to determine the factors identifying the pay gap. This chapter includes the components and effects of the CEO – employee pay ratio.
2.2. CEO Compensation
One of the components determining the pay gap is the size of CEO compensation. Efficient contracting point of view of economic theory suggests that CEO compensation is composed to attract and motivate top executives and any differences in pay levels could be the result of different economic characteristics, such as the industry a company operates in, size of the company (Frydman/Jenter, 2010) and high competition because of entry of foreign businesses (Hubbard & Palia 1995; Cuñat & Guadalupe, 2009b). The critics mainly argue that top executives high compensation is not always in shareholders’ best interests.
The academic literature and business research consistently raise the issue of overpaid top executives, often with a keen focus on the large and growing gap in the compensation of chief executive officers (CEOs) relative to rank-and-file employee pay (Faleye, et.al, 2012). The most research is conducted in the U.S., since the ratio of CEO to average worker compensation there is the highest in the world and is still increasing. According to a 2005 report by United for a Fair Economy and the Institute for Policy Studies, the average CEO in the U.S. earned 431 times the average pay of a production (i.e., non-management) worker in 2004, up from 301 times in 2003 and 42 times in 1982 (Executive Excess 2005, 12th Annual CEO Compensation Survey).
Murphy (1999) documented that there has been an explosion in academic research on executive compensation after 1990. The modern history of CEO pay was supplemented by the agency the-ory that set differences between the CEO (agent) interests and shareholders (principals) interests. Early studies documented the relation between executive pay and company performance and size (Coughlan and Schmidt, 1985; Murphy, 1985, 1986; Jensen and Murphy, 1990a). Executive pay was linked to earnings, management and investment decisions, capital structure, dividend policies, mergers, and diversification (Murphy, 1999).
There is also growing literature suggesting that U.S. board of directors’ operations are becoming increasingly ineffective (Core et al., 1999). In fact, some studies found a correlation between
11
CEO compensation and the board composition. For example Boyd (1994) found out a positive relation between CEO compensation and the percentage of outside directors. Lambert et al. (1993) documented that CEO has higher compensation when he appoints actual board members himself. Other studies examined the relationship between ownership structure and the level of CEO compensation (Core et.al., 1999).
In the contrary some authors suggest that the compensation of CEO increases with an augmented executive skill requirements and depends on the complexity of the job, firm size, risks and growth opportunities (Core et al., 1999). At the same time agency theory suggests that a firm better performance increases the CEO’s power to bargain a better pay (Faleye et al.,2012). Other authors (Merz and Yahsiv (2007), Papanikolaou (2007), Bazdrech, Belo and Lin (2008), and Parlour and Walden (2008) document a positive link between average CEO compensation and firm value.
2.3. Theoretical framework
The rapid increase in CEO compensation over the past 30 years has raised many theories about the determinants and effects of executive pay. At one end of the spectrum the high level of executive compensation is seen as a result of the complexity of the job. On the other hand many other researchers argue that CEO compensation has increased sharper than the compensation of other executives (Frydman, Jenter, 2010). There is relationship between CEO pay and weak corporate governance and internal control (Jensen, 1993, Crystal, 1991), which leads to the conclusion that U.S. board of directors are ineffective (Core, 1999). Many other explanations arose in recent years and the debate is too extended to be highlighted in this paper. Nevertheless, main theories are briefly summarized within the framework of this research.
2.3.1. Agency theory
An agency relationship is defined as one in which one or more persons (the principal(s)) engage another person (the agent) to perform a service on their behalf which involves delegating some decision-making authority to the agent (Jensen and Meckling, 1976; Ross, 1973). The agency di-lemma occurs when the interests of principals and agents diverge, when a conflict of interests arises and agent is motivated to act in his own best interests. An important comment is that the agent and the principal not only have different interests, but also asymmetric information. Since the agent is more informed, the principal cannot be sure that the agent is acting in principal’s best interests.
The principal can limit divergence by establishing relevant incentives for the agent and by setting a good monitoring and control mechanisms. Further the principal may pay to the agent an ap-propriate compensation if the agent acts in the best interests of the principal. Various pay mechanisms may be used to avoid the conflict of interests, such us commissions, performance related bonuses, stock options or profit sharing.
The sum of the principal's monitoring expenditures, the agent's bonding expenditures, and any remaining residual loss are defined as agency costs (Hill & Jones, 1992).
12
2.3.2. Managerial power theory
According to managerial power theory weak Corporate Governance allows top executives to determine their own compensation regardless their performance, and gives possibility to influence the board of directors. This theory , developed by Bebchuk & Fried (2004) is called “managerial power hypothesis” and suggests that most of the compensation occurs through forms of pay that are less transparent or difficult to value, such as stock options, pensions, and severance pay. Later Kuhnen & Zwiebel (2009) created a dynamic model to examine the relationship between observable and hidden compensation and other variables. They suggested that observable compensation (salary) and hidden compensation (perks, pet projects, pensions, etc.) serve different roles for management and have different costs, and both are used in a certain balance (Kuhnen & Zwiebel, 2009).
Morse et al. (2011) also documented that powerful CEO-s design the incentive part of their pay themselves (2011). They designed a model in which powerful CEO-s are able to opportunistically alter the focus of their incentive pay and in this way they can manipulate their compensation. The rigging is greater when the human capital intensity of the CEO and uncertainty about the firm’s future prospects are high. They also found out that firm performance is decreasing in accordance to the amount of rigging, which explains the lack of sensitivity to negative performance. In their model they documented that stronger governance and internal control lower contract rigging by powerful CEOs.
2.3.3. Technological change and inequality in managerial compensation (paradox of productivity and paradox of wage inequality)
There are various explanations of the growing gap of the levels of compensations between chief executives of a company and average employees. An empirical literature study reveals a positive relationship between increased compensation gap and the technological change (Violante, 2002; Guvenen and Kuruscu, 2009; Autor, Katz, and Kearny, 2008; Acemoglu, 2002, Lusting et al., 2011). Lusting et al (2011) documented that the technological change leads to high productivity and the growth of capital of the company. The increased accumulation of organizational capital, resulting from the growth composition change, improves the manager’s outside option in successful firms (Lusting et al., 2011).
Doms, Dunne and Troske (1997) documented that plants using more innovative technologies hire better skilled workers and pay higher wages compared with the firms that use less advanced technologies. They concluded that plants with high skilled employees adopt new technologies more often, so adoption of the newest technologies can have influence on median or average employee pay.
There is a great interest in the academic world to the phenomenon of the labour wages, technological change and the productivity growth. The results are conflicting and they bring about two paradoxes: paradox of productivity and paradox of wage inequality. Brown and Campbell (2002) documented the following changes in U.S. wage structure:
13
A decline in manufacturing employment (especially high-wage manufacturing employment), which results from globalization, increasing level of automation, and shifts in consumption demand.
A decline in unionization, which increases inequality by reducing the wages of production workers as their power declines.
A decline in the rate of growth of the college-educated population, which leads to growing inequality by decreasing their relative supply and pushing their wages higher. Skill-based technological change which increases the demand for high-skill workers
relative to low-skilled workers. (Brown, Campbell, 2002).
There is also growing empiric literature suggesting that the technological changes also caused the shift in demand for skilled, higher educated labour, and so the innovative firms hire more skilled workers and pay higher wages than less advanced firms. This can result in the development of wages inequality. Therefore we can conclude that not an absolute amount of CEO compensation, but the human resources availability and scarcity in different sectors may determine the CEO-employee pay ratio.
Another trend which could influence the increasing pay gap is the globalisation and outsourcing of the non –skilled and scarce resources to low-waged country. But since the outsourcing is an external factor and increasingly takes place in all industries, the effects of it is not be included in this study. The study is based on U.S. companies and the CEO pay is compared with the compensation of average American employee.
2.4. The determinants and effects of CEO-employee pay ratio
Another group of researchers linked the ratio between CEO compensation and average employee pay to the technological innovation and sector in which a firm operates as well as to the resources used. According to Barney (1991) resources that can provide competitive advantage to a firm (1) are valuable, (2) are rare, (3) are difficult to imitate, and (4) have qualities that make them irreplaceable. In a high-tech firm, the ability to be technologically innovative is vital and can be considered as a resource that is valuable, rare, hard to imitate, and difficult to substitute (Grant, 1991). We can expect that in innovative firms CEO pay should be linked to the innovation efforts. There is evidence that the pay of scientists and engineers in high-technology firms is also often linked to meeting innovation goals such as achieving significant technological project milestones (Balkin& Bannister, 1993; Riggs, 1983).
There are evidences that the relative pay difference is higher within firms in homogenous industries, where employees are presumably more interchangeable and therefore less powerful relative to management. In contrast, relative pay difference or gap declines with employee unionization and capital intensity (Faleye et.al 2013). Faleye et al. (2013) documented that CEO– employee pay ratios depend on the balance of power between the CEO (relative to the board) and ordinary employees (relative to management).
14
2.5. Assessing the impact of financial crisis
Most executive pay packages contain four basic components: a base salary, an annual bonus tied to accounting performance, stock options, and long-term incentive plans, including restricted stock plans and multi-year accounting-based performance plans (Murphy, 1999). A long literature on the compensation of CEOs shows that CEO compensation depends on stock return and the composition of pay can differ per sector (Barro and Barro, 1990). In particular, the share of pay in the form of stock and options for bank CEOs is lower than in other industries (e.g., Adams and Mehran, 2003; Houston and James, 1995). Based on that we can conclude that the CEO pay has been declined in the Manufacturing ant Information sectors during the financial crisis as a result of decreased stock options, which decrease the CEO relative pay gap.
Based on the previous literature review we can also conclude that in the industries where average employee has lower skills and thus can easily be replaced, the salaries tend to be lower (Grant, 1991; Faleye et.al, 2013). If this is the case, financial crisis would cause many employees lose their jobs, lose power relative to their management and would result in a sharper compensation declines for these industries.
In contrary fast technological development and growing competition between the companies make skilled personnel more valuable and scarce. If our suggestions are correct then the financial crisis should have caused a greater decrease in the CEO and average employee gap in the Information sector, as opposed to the Manufacturing industry.
15
3.
Hypotheses development
3.1. Research Model
The research attempts to estimate the possible differences between the CEO and employee compensation gaps within the Manufacturing and IT sectors and the impact of crisis on these sectors. Therefore, a causal model is built with the company level ratio of the CEO and employee compensation gaps as a dependent variable. Independent variables are sector, describing belonging of the company to either manufacturing or IT sectors, and crisis, showing the specific impact of crisis on the compensation gap. It is also assumed that the crisis had different impact on the compensation gap in manufacturing sector than in the IT sector. A causal relationship between sector and the crisis is introduced in the model to capture this possible effect.
In addition, a set of control factors described in the empirical research are introduced in the model to approximate the causal relationship present in the U.S. Publicly traded IT and Manufacturing companies.
Graph 1 presents the causal relationship and the impact of the crisis moderator. It describes the sector as an independent variable, ratio as the dependent variable and crisis as the moderator.
3.2. Hypotheses
Literature review suggests that employees can bargain higher wages if they cannot easily be replaced. This happens in cases when they have rare, valuable skills that are hard to imitate. This type of skills are obtained with extensive training and education, and are mainly used in high tech industries, where human presence is crucial and cannot be substituted even by very advanced machinery. In high tech industries where innovative technologies are used and constantly upgraded, the employees need to be trained regularly to have up to date knowledge. On the contrary, in homogeneous industries, such as Manufacturing, employees are less skilled and can easily be replaced (the effect of globalization and outsourcing opportunities make this phenomenon even more considerable). Employees in Manufacturing have lower bargaining power relative to management. High employee wages in one industry and low employee wages in the other yield different CEO/employee wage ratios across industries.
Sector (IT or
Manufacturing) Relative Pay
Crisis
16 IC
IP
MP
MC
Based on these findings, first research hypothesis can be defined as:
Hypothesis 1: The CEO relative pay in the manufacturing industry is traditionally higher than in the Information sector.
It was discussed that the CEO pay has been declined during the financial crisis as a result of decreased value of stock options, which resulted in a decrease in the CEO relative pay gap. Based on this the hypothesis 2 is as follows:
Hypothesis 2: The financial crisis caused an decrease of the CEO relative pay, in general.
It was also discussed that economic downturns cause many employees lose their jobs, which results in surplus of labor and decreased demand for it. As a result, employees lose bargaining power and employee compensation could decline. In addition to this technological development and growing competition between the companies make skilled personnel more valuable and scarce. So the economic downturn leads to the situation when low skilled employees lose jobs and get paid less, while the compensation for high skilled personnel may not get affected at all (or decline slightly). If our suggestions are correct then the recession should have caused a greater decrease in the CEO and average employee pay gap in the Information sector, than in the Manufacturing industry.
Our third hypothesis, therefore, is the following:
Hypothesis 3: The financial crisis caused a greater decrease of the CEO relative pay in the Information sector, than in the manufacturing industry .
Graph 2, where IP (Information Pre-crisis) represents the pre-crisis relative pay ratio in Information sector; IC (Information Crisis) is crisis relative pay ratio in Information sector ; MP (Manufacturing Pre-crisis) is pre-crisis relative pay ratio in Manufacturing; and MC (Manufacturing Crisis) is the crisis relative pay ratio in Manufacturing, gives the graphical presentation of all three hypotheses.
IT Manufacturing Sector
Relative Pay
17
3.3. Control factors
3.3.1. Compensation Surveys and the Relation Between CEO Pay and Firm Size
Early studies documented the relation between executive pay and company performance and size (Coughlan and Schmidt, 1985; Murphy, 1985, 1986; Jensen and Murphy, 1990a; Bakker et al.) According to Bakker et al. (1988) there are numerous studies that examine the relationship of executive pay to firm size. They explain that on one hand larger firms may employ better quali-fied CEOs, which require larger compensation package, but on the other hand CEO pay is not only based on the abilities of CEO. Bakker et al. refers to Murphy (1985), who documented in one of his previous studies that if the value of the firm remains constant a sales growth of 10 percent will increase the salary and bonus of its CEO by about 2-3 percent. They determined the compensation/sales elasticity’s estimated for five years and for five industry groups, have been stable across time and industries. Moreover, the correlation between size and compensation is very high; R-squares for the 1983 regressions, for example, are 0.60 (Manufacturing), 0.53 (Retail Trade), 0.67 (Utilities), 0.68 (Banking), and 0.69 (Insurance).
Rosen (1981, 1982) found a relationship between CEO pay and scarcity of managerial talent. If higher CEO talent is more valuable in larger firms, then larger firms should offer higher levels of pay and be able to get better skilled CEO-s according to the rules of efficient labor market (Rosen 1981, 1982).
Based on this finding we can conclude that the size/pay relation is causal and the CEO pay is not based only on his/her abilities. Another conclusion is that CEO-s can increase their pay by in-creasing firm size or firm earnings, even if this increase in size reduces the firm’s market value. Following Bakker et al. (1998), Murphy (1995), and more recent researchers such as Chen et al. (2006), Aebi et al. (2012), we can measure the size as the natural logarithm of total assets of the firm.
Based on these studies we can acknowledge that the control variables for firm size are logarithm of Total Assets and logarithm of Total Revenue of the firms, estimated for two industries (in our case): Information Technologies and Manufacturing. Revenue indicates the total money that the company generates without subtracting the costs. Revenue does not give any indication of oper-ating income.
3.3.2. Compensation Surveys and the Relation Between CEO Pay and Firm Performance
From agency theory it is known that various mechanisms are used to avoid the conflict of inter-ests between agent and principal. Some of those mechanisms are pay mechanisms such as com-missions, performance related bonuses, stock options, profit sharing. Corporate governance ex-perts consider that management’s interests are better aligned with those of shareholders if man-agers’ compensation increases when shareholders gain and falls when shareholders lose (R. Fahlenbrach, R.M. Stulz, 2010). They cited Murphy (1999): ‘‘Stock ownership provides the most direct link between shareholder and CEO wealth.’’
18
Another control variable used by Fahlenbrach and Stulz (2010) is tested in this study, namely the Return on assets. The Return on Assets (ROA) percentage shows how profitable a company's assets are in generating revenue. It is calculated as:
ROA = EBIT / Total Assets
EBIT is showing operational profit of the company, which seems more relevant indicator of the profitability, the performance of the company and the change of its capital structure.
3.4. Industry related factors
In addition to the preceding economic and governance factors, the compensation literature emphasizes the importance of industry membership for labor market benchmarking purposes (Armstrong, Ittner, Lacker, 2012). Following the previous studies the industry fixed effects on the executive compensation and the ratio between CEO pay and the median employee pay is tested. Two industries-Information and Manufacturing is compared to each other: Since actual employee wages information is calculated from public governmental data such as Bureau Labor Statistics (BLS) and The Bureau of Economic Analysis’s (BEA), industries are determined by the sector classification used by the U.S government organisations. The system used is North American Industry Classification System (NAICS). NAICS industries are identified by a 6-digit code. In this research the sectors of Manufacturing and Information are included.
3.4.1. Manufacturing sector
The Manufacturing sector is divided into manufacturing of durable goods such as wood products, machinery, electronic equipment, computers, and nondurable goods, such as food, tobacco, paper products, chemical products, and textile. The most significant difference with formal SIC classification is the creation of the Computer and Electronic Products Manufacturing sub sector, which includes production of computers, computer peripherals, communications equipment. The sub sector was created because of the economic importance and rapid growth of these sectors. Another important activity, such as Publishing, has been removed out of manufacturing sector, and added to the Information sector. Another important change in the traditional SIC codes is the recognition of an Information sector. The two-digit codes of manufacturing sector are 31-33.
3.4.2. Information sector
Information sector includes companies that create, or provide the means to distribute information and data processing services. Industries included in this sector are publishers, previously included in the manufacturing sector in the SIC; software publishers, previously included in services; broadcasting and telecommunications producers and distributors, previously included in utilities and transportation; and motion picture and sound recording industries, information services, and data processing services, previously included in services. According to the www.naics.com there are 34 industries included in this new sub sector, 20 of which are new. Some of the new industries include paging, cellular and other wireless telecommunications, and satellite telecommunications. The two-digit sector code of information sector is 51.
19 Table 3-1summarizes the variables used in the model. Table 3-1. Description of the factors used in the model
Factors Description Source / Database Why important Theoretical framework
CEO Compen-sation
Most executive pay pack-ages contain four basic components: a base salary, an annual bonus tied to accounting performance, stock options, and long-term incentive plans. For the model are used the CEO officer pay figures reported to the Securities and Exchange Commis-sion. Compustat Execu-comp; Wharton Research Data Services To calculate depended
variable (pay raio) Coughlan and Schmidt (1985); Murphy (1985); Jensen and Murphy (1990a); Bakker et.al (1988); Rosen (1981, 1982)
Industry 2 sectors: Manufacturing (NAICS code 31-33) and Technologies ( NAICS code 51) are studied
Compustat Execu-comp; Wharton Research Data Services; www.naics.com Independed variable which possibly deter-mine the pay ratio
Brown and Campbell (2002);
Economic Policy Insti-tute study (EPI);
Average Employee Pay
Includes total average compensation of employ-ees per sector.
The Bureau of Economic Analysis. Bureau of Labor Sta-tistics
To calculate depended
variable (pay raio) Method of calculation developed by Economic Policy Institute
Total
Assets Includes total assets of the company Compustat North America databases; Wharton Research Data Services
A control variable to test the relationship between company size and pay ratio
Bakker et al. (1998), Murphy (1995), Chen et al. (2006), Aebi et al. (2012),
Revenue
Includes total revenue of the company for each year
Compustat North America databases; Wharton Research Data Services
A control variable to test the relationship between company performance and pay ratio
R. Fahlenbrach, R.M. Stulz (2010);
Murphy (1986); EBIT Companies earnings
be-fore interest and taxes Compustat North America databases; Wharton Research Data Services
A control variable to test the relationship between company performance and pay ratio, used for the calculating ROA
R. Fahlenbrach, R.M. Stulz (2010);
Faleye et al.,2012 ROA Return on Assets Calculated by dividing
EBIT on Total Assets Is a control variable to test the relationship between company performance and pay ratio
R. Fahlenbrach, R.M. Stulz (2010)
Crisis Financial crisis of
2007-2009 Years 2007-2009 Dummy variable, to calculate impact of the crisis between the sectors
Fahlenbrach &Stulz, 2010; Gabaix et al, 2013
20
3.5. Operationalization of the variables in the model
3.5.1. Dependent Variable
The gap between CEO and employee compensations is determined by calculating the ratio of CEO pay to average employee salary.
GAP is the dependent variable representing the company level compensation gap of CEO and
employee compensations for manufacturing and IT sectors that were at least once included in the S&P 500 list during the years of 2006 to 2009.
3.5.2. Independent Variables
Dummy variables are created describing the sector to which company belongs to, the pre-crisis period of crisis period. We also assume that there could be the case that the effect of crisis on the pay gap may depend on the sector of the industry.
Dummy variables pertaining to the pre-crisis and crisis periods: The period of 2006 to 2009 is divided into two categories: pre-crisis (2006 and 2007) and crisis (2008 and 2009). A separate dummy variable is developed for each category as follows:
1. Variable PR_CRIS describes the pre-crisis period. It is equals 1 and 0 for the pre-crisis period and other periods, respectively.
2. Variable CRIS describes the crisis period. It is equals 1 and 0 for the crisis period and other periods, respectively.
Note that the PR_CRIS variable is eliminated for model estimation purposes so as to avoid the “dummy variable” trap.
The table below describes the rest of the independent variables used in the model. Table 3-2. Description of independent variables
Name Description
MAHUF dummy variable showing whether the company belongs to the manufacturing or the IT industry. MAHUF has two values: 1 and 0 for manufacturing and IT industry, respectively.
GENDER dummy variable showing whether the company CEO is male or female.
GENDER has two values: 0 and 1 for male and female, respectively.
TA continuous variable measured in USD and shows the company total assets in million USD
EBIT continuous variable measured in USD and shows the company earnings before interest and tax in million USD
TREV continuous variable measured in USD and shows the revenues in million USD ROA is a continuous variable showing company return on assets. It is calculated as
the ratio of EBIT and TA.
NEMP continuous variable showing the number of employees in the company
CRISMANUF interaction variable showing the difference of the effect of crisis on manufacturing and IT sectors. It is derived by multiplying CRIS and MANUF variables.
21
To test the hypothesis were developed three models which are represented in the table below: Table 3-3. Models representing the hypotheses:
Hypothesis Model Base categories
Model
1 The relative pay in CEO the manufacturing industry is traditionally higher than in the Information sector. IT sector (MANUF = 0); and female CEO (GENDER = 0). Model
2 The financial crisis caused an decrease of the CEO relative pay, in general
pre crisis period (CRIS = 0); IT sector (MANUF = 0); and female CEO (GENDER = 0).
Model
3 The crisis caused a financial greater decrease of the CEO relative pay in the Information sector than in the manufacturing industry.
pre crisis period (CRIS = 0); IT sector (MANUF = 0); and female CEO (GENDER = 0).
Model 1 does not include the crisis and models 2 and 3 pertain to both crisis and pre-crisis situa-tions.
22
4.
Data collection
4.1. Introduction
The research includes all of the S&P 500 corporations from both Manufacturing and IT sectors. Wharton Research Data Services is used to obtain yearly information on these companies from 2006 to 2009. The years of 2006 and 2007, 2008 and 2009 are used as pre-crisis period and crisis period, respectively.
This study focuses on the CEO and average employee relative gap between two sectors - Information and Manufacturing. The research includes all of the S&P 500 corporations from each of Manufacturing and Information sectors. Wharton Research Data Services is used to obtain yearly information on these companies from 2006 to 2009. The years of 2006 and 2007, 2008 and 2009 are used as pre crisis period and crisis period, respectively. The Compustat database for North America provides with the information about the biggest U.S. companies. The CEO compensation information is obtained from Compustat Execucomp database, which includes compensation information of executives of North America and Canadian larges companies.
The average worker compensation information is obtained from the U.S. governmental databases, such as The Bureau of Economic Analysis’s (BEA) and the Bureau of Labor Statistics (BLS). The gap between CEO and employee compensations is determined by calculating the ratio of CEO pay to average employee salary. The detailed description of the calculation, limitations associated with those and methods for overcoming these limitations are described below.
4.2. Development and Use of the Variables in the Model
4.2.1. Calculating CEO pay ratio
For the calculation of CEO pay ratio it is preferable to use median employee compensation. Median expresses middle value of the variable, which gives a good representation of central tendency if values are clustered (or skewed) toward one end. In this case average can be influenced by the few values, while median still represents a central tendency.
Unfortunately the median salary information is not available, so the average employee compensation per industry was used as a proxy to calculate the compensation ratio. The gap between CEO and employee compensations is determined by calculating the ratio of CEO pay to average employee salary based on the following steps:
1. the average total yearly compensation (AYC) of employees except the CEO
2. the annual total compensation of the CEO, based on figure reported to the Securities and Exchange Commission (SEC).
3. the ratio of the two amounts (also referred as the relative pay or the gaps of CEO and employee compensations).
As a first step the average compensation of employers needs to be determined. Unfortunately the information on the actual compensation is not available for any particular firm. Since there is
23
no universal standard for all firms to report the staff costs, most companies do not disclose average or median worker pay. The methodology used to calculate the pay ratio is based on the one developed and used by Economic Policy Institute (EPI) for a study focused on income inequality trends within U.S. companies (Mishel & Sabadish, 2012).
In 2013 the same methodology was used by Bloomberg to calculate compensation ratios for S&P companies. The ratio of CEO compensation to the annual average wages and compensations of workers in the key industries is computed for each of the largest S&P firms.
Industries were determined by the North American Industry Classification System (NAICS) which is used by the U.S. government agencies.
EPI developed a measure of the annual compensation of a typical domestic worker in the key industry of each company in their sample as follows: the average hourly earnings of production and non-supervisory workers in an industry were identified and then were converted to a full-time, full-year salaries plus benefits, which makes the measure of annual compensation. The available information that could be used for calculating average worker's compensation disclosed by The Bureau of Economic Analysis (BEA) includes two databases: hourly wages for production and non-supervisory workers for each industry and total compensation (TC) of all workers by industry. To come up with average worker compensation first was used the data for total compensation of all workers by industry from The Bureau of Economic Analysis’s National Income and Product Account (NIPA) Table 6.2D (Compensation of Employee by Industry). This table gives total compensation of all workers by industry for years 2005-2012. Then the data on wages and salaries that corresponds to this industry data was obtained from NIPA Table 6.3D (Wage and Salary Accruals by Industry). This table provides total wage and salary disbursements (TW) to all workers in each industry for 2005-2012. By using these two datasets based on EPI methodology an industry-specific compensation-to-wage ratio was created by dividing total compensation (TC) by total wage and salary accruals (TW) in each industry. This ratio shows what percent of total compensation is attributable to wages.
The survey was conducted on a monthly basis, so there can be 12 different hourly wages mentioned in the survey for any full year. To overcome this problem EPI used the average of all twelve salaries reported during each month to get one number for the average salary during a particular year.
Then, because the data used are average hourly earnings and the CEO compensation data are presented as annual numbers, the final industry-level typical worker compensation data are multiplied by 2.080 (average worked hours per year). The hourly wage per industry (HW) was multiplied by total hours (2.080) worked by one employee to get average yearly wages (AYW) of an employee in the specific industry:
The total yearly wages then were increased by industry-specific compensation-to-wage ratios to reflect not only the salary, but also actual benefits and bonuses received by workers (compensation). So after simply multiplying AYW by this ratio we get the average yearly compensation (AYC) of an employee in a particular industry:
24
The second step includes determining CEO compensation information. This is readily available in Compustat Execucomp database. Most executive pay packages contain four basic components: a base salary, an annual bonus tied to accounting performance, stock options, and long-term incentive plans (including restricted stock plans and multi-year accounting-based performance plans) (Murphy, 1999). All components are included in the calculation. The data used for the calculation of relative pay ratio include the most recent chief executive officer pay figures reported to the Securities and Exchange Commission. In our model we use the company data as from 2006, because in 2006 the Securities and Exchange Commission (SEC) adopted new disclosure requirements concerning, among other items, executive compensation. The amendments to the compensation disclosure rules provide investors with a clearer and fuller picture of the particular executive officer compensation. The new rules were designed to improve tabular presentation and to offer material qualitative information regarding the manner and context in which compensation is awarded and earned (Beltratti & Stulz, 2012) .
And finally as the third step, the ratio was calculated by dividing the CEO compensation by estimated industry-specific employee compensation.
4.2.2. Independent and Control variables
The independent and control variables are collected from Compustat North Amerrican databases. For the study were used S&P 500 biggest companies, selected with the ticker number I0003. All datasets are connected in STATA with unique GVkey company number.
Sectors are identified with NAICS codes: the two-digit code of manufacturing sector are 31-33, and for the Information sector is 51. The following company information was obtained: company name, industry, total revenue, total assets, earnings before interest and tax, number of employees and gender of CEO. Table 4-1 summarizes data collection procedure.
Table 4-1. Data collection procedure
List information col-lection procedures
Describe how problems are identified, e.g., in-complete information and errors
Describe pre-liminary analysis
Describe how the in-formation is stored and retrieved
Average
Compensa-tion of Employees Checked during the calculat-ing in process (used Micro-soft Excel)
Compared with the calculation made by Bloomberg
Average compensation tables per sector per year (incl. years 2006-2009) in Excel
CEO Compensation Checked the completeness of the database, if incom-plete the information is obtained from the Company Annual Report
Random check with annual reports
Excel file format, the cal-culation of the ratio made in Excel: Ratio = CEO Compensation/Average Employee Compensation Independent and
Con-trol variables Check the completeness of the database and the consis-tence of the information; The comparison of the same information in the different sources has to be done
Check the outlin-ers, check the company names with the S&P list; Pilot analysis in STATA is imple-mented
Data downloaded from WRDS database in Excel format
25
4.3. Cleaning and Reorganization of the Data for Analysis
The data collected from different sources was put combined, reorganized, and cleaned using STATA statistical software. After combining a company information from Compustat with the available CEO compensation information from Execucomp, database included 815 companies in 2006, 903 companies in 2007, 867 companies in 2008 and 838 companies in 2009. The average number of companies for four year is 856, the total number of observations is 3423.
Afterwards the data was merged with the S&P 500 list. Outliers, companies with CEO to average employee pay ratio less than 1, were excluded from the database. For example, companies such as Apple, Facebook and Microsoft pay to their CEO-s a formal yearly salary of $1.These companies were eliminated from the model, since those are exceptions that can influence the total result. From the remaining 953 observations companies with missing information about assets, company earnings before interest and tax, return on assets and the number of employees were excluded. Appendix D contains the list of the companies eliminated from the total database. Table 4-2 illustrates data processing stages.
Table 4-2. Database size by stage and year
Database stage description Number of companies by year Number of Observations
2006 2007 2008 2009 Average Total Initial database after combining
company information with available CEO compensation information per industries Information and
Manufacturing
815 903 867 838 856 3423
Database after merging with the database of the companies listed in the IT and manufacturing sectors of S&P 500
254 257 252 246 252 1009
Database for the model estimation that excludes companies with GAP less than 1
204 254 250 245 238 953
Final database for the model estimation that excludes companies with missing information on company assets, company earnings before interest and tax, ROA, and the number of employees.
26
Table 4-3 shows how the data is divided per sector per year within the both Manufacturing and Information sectors. We can conclude that the Manufacturing sector is much larger than the Information sector.
Table 4-3. Available data per sector
Available data per sector /
Years 2006 2007 2008 2009 Average Total number of observations Manufacturing 154 191 186 184 179 715
Information 41 51 56 55 51 203
27
5.
Selection of the right model for panel data analysis
There are different models developed for panel data analysis. Broadly they can be grouped in: fixed effects, random effects, random parameters, polled regression. There are different techniques, with allow to evaluate whether the statistical method used corresponds to the data, or no. The following tests are performed to determine the appropriate model for panel data: Hausman specification test, Breusch-Pagan Lagrange multiplier (LM) test, Wooldridge test for autocorrelation in panel data, and test for heteroskedasticity in panel data.
5.1. Tests performed
5.1.1. Random Effects Regression vs. Fixed Effects Regression
Hausman specification test (1978) is used to determine which panel data model is more appropriate
to use. Fixed-effects model is used to explore the relationship between predictor and outcome variables within entity and to analyze the impact of variables that vary over time. Unlike the fixed-effects model, in the random effect model the variation across entities is assumed to be unrelated with the independent variable (Torres- Reyna, using Stata 10.x). Green (2008) explained that “the crucial distinction between fixed and random effects is whether the unobserved individual effect embodies elements that are correlated with the regressors in the model, not whether these effects are stochastic or not”.
Torres-Reyna explained in his manual for STATA that if there are reasons to believe that differences across entities have some influence on the dependent variables, then it is preferable to use random effects model, which allows to include time invariant variables, such as sector. In our model the sector plays a role of explanatory variable. Hence, our estimation should be based on a random effects model. We have performed the Hausman test to eliminate any doubts about the preference of the random effects model over the fixed effects model. The test shows that the calculated p-value (0.633) is higher than the critical p-value of 0.05, which means that we fail to reject the null hypothesis that random effects model is more appropriate for the panel data.
5.1.2. Random Effects Regression vs. OLS Regression
Breusch-Pagan Lagrange multiplier (LM) test is conducted to helps us to decide between a random effects regression and a simple OLS regression. The null hypothesis in the LM test is that variances across entities is zero. The results of the test reject the null and conclude that random effects method is more appropriate than OLS regression. There is evidence of significant differences across companies; therefore we should run a random effects regression.
5.1.3. Wooldridge test for Autocorrelation in Panel Data
Another test which was performed is Wooldridge test for autocorrelation (2002) in panel data. Autocorrelation or serial correlation means a similarity of values separated from each other by a given time lag. The result of the test shows that there is no first-order autocorrelation in the model.
28
5.1.4. Test for Heteroskedasticity in Panel Data
Random variables in our model could be heteroscedastic if there are sub-populations that have different variability. The presence of heteroscedasticity in the model can invalidate statistical tests of significance that assume that the modelling errors are uncorrelated and normally distributed. The simple test shows that there is no heteroscedasticity in the model.
The table 5-1 summarises the tests performed to choose a right model for panel data and the results of the tests.
Table 5-1. Summary of tests performed to choose the right model and check for autocorrelation and heteroskedasticity Test description Random Effects Regression vs. Fixed Effects Regression Random Effects Regression vs. OLS Regression Presence of Autocorrelation in the Model Presence of Heteroskedasticity in Panel Data
Test Name Hausman test Breusch and Pagan Lagrangian
multiplier test for random effects Wooldridge test for autocorrelation in panel data Test for heteroskedasticity in panel data Null
Hypothesis Ho: difference in coefficients not systematic
Ho: Var(u) = 0 Ho: no first-order
autocorrelation
Ho:
homoskedasticity
P-value Prob>chi2 =
0.6330 Prob > chibar2 = 0.0000 Prob > F = 0.2347 Prob > chi2 = 1.0000
Conclusion We failed to reject the null hypothesis and concluded that random effects model is more appropriate than fixed effects model for the panel data
We rejected the null and conclude that random effects is more appropriate than OLS regression. There is evidence of significant differences across companies, therefore we should run a random effects regression. We failed to reject the null hypothesis and concluded that there is no first-order autocorrelation in our model. We failed to reject the null hypothesis and concluded that there is no
heteroskedasticity in our model.
29
5.2. Log transformation of variables with skewed destruction
The distribution of variables of compensation gap, GAP, is significantly right skewed (Graph 5A; Appendix B). The log transformation can be applied to make the distribution less skewed and data more interpretable. After logarithmic transformation (see Graph 5B, Appendix B ) the compensation gap is much closer to normal distribution. Therefore, the estimated model is based on log-linear model, where the dependent variable is a log-transformed variable named Log (GAP). Natural logarithm is used for the log transformation.
Graphs 5C, 5E and 5G (Appendix B ) clearly show that variables Total Assets (TA ), Number of Employees (NEMP) and revenue (REV) are also significantly right skewed, but after logarithmic transformation (see Graphs 5D, 5F and 5H, Appendix B) the variable are much closer to normal distribution. Natural logarithm is used for the log transformation in all cases.
The continuous variables EBIT and ROA have some negative values and therefore, are not suit-able for logarithmic transformation, so no transformation was applied for this varisuit-ables. Graph 5J (Appendix B) shows that the distribution of ROA is close to normal distribution, so transforma-tion is also not needed.
Therefore, the estimated model is based on a mix of log-log and log-linear models, where the dependent variable is a log-transformed variable named Log (GAP).
30
6.
Data analysis
6.1. Descriptive Statistics
The table 6-1 below shows the descriptive statistics values for the Manufacturing and Information sectors before and during the financial Crisis. The years of 2006 and 2007 are used as pre-crisis period, the years 2008 and 2009 are used as crisis period.
The median CEO relative pay ratio in Information sector was 146 before the crisis and decreased to 122 during the crisis. The maximum pay ratio in the sector was 1536 before the crisis and 1361 after the crisis. In the manufacturing sector the maximum pay ratio was much lower: 864 before the crisis and 914 during the crisis. Although the median ratio in manufacturing sector was higher than in the Information sector: 206 before the crisis and 171 during the crisis. We can also conclude that the median CEO pay ratio is decreased in both sectors as a result of the financial crisis.
In Information sector the standard deviation is much higher than in manufacturing sector: 272 against 158 before the crisis and 216 against 139 during the crisis. There are no significant differences between the sectors regarding the median company size in terms of total assets (TA) and performance (EBIT). Number of employees did not decreased during the crisis in Information sector, while decreased slightly in Manufacturing. The table shows a significant increase of maximum number of employees in manufacturing sector, while the maximum number of employees remain stable in Information sector.
