An analysis of CEO tenure on firm value in the USA?

(1)

An analysis of CEO tenure on firm value in the USA?

Name: Quincy Vaz

Student Number: 10069372 Bachelor: Economics and Business

Specialization: Finance and Organizations Bachelor Thesis (12 EC)

(2)

2

1. Introduction

Why do organizations act as they do? This question is of key importance to organizational theorists. The upper elections theory composed by Hambrick and Mason (1984) states that people at the top of the organization, in firms known as executives, influence the outcome of an organization. According to this theory is the top management team, which consists of among others the CEO, COO and CFO, playing a big role in influencing the outcome of an organization. They are inspiring, motivating and mobilizing their employees while making important decisions and representing their company all at the same time (Buozinte-Rafanaviciene, Punclizine and Turauskas, 2009). These executives make choices on the basis of their personalized construal’s of the situations they face (Hambrick, 2007). The personalized construal’s, also known as paradigms, are influenced by personalities and functional backgrounds. Managers are often perceived as having their own styles of making investment, financing and other strategic decisions, thereby imprinting their personal marks on the companies they manage (Bertrand and Schoar 2003). These actions are becoming more visible when these managers are implementing crucial changes to carry out the strategy of the company, like aggressive takeovers or high R&D investments. The actions and decisions will influence the firm or organization as a whole and therefore also the firm value (Miller, de Vries and Toulouse, 1982).

But first we’ve to ask ourselves how much individual managers do matter for firm behavior and economic performance. Like already said, Bertrand and Schoar (2003) have found that individual managers do matter in the determination of firm policies since individual managers are often perceived as having their own person specific styles when making investment, financial and strategic decisions. You can observe it by comparing firms’ capital structures, investment decisions and organizational structures, which in turn has an effect on the firm value.

Although individual managers seem to differ amongst each other, you may ask yourself if there are specific managerial positions, like the head of executives, that have influence on the firm value. Further you might also question yourself if these management styles or paradigms may change among their tenure. Studies all over the world have made analysis between the relationship of the executives tenure and firm value. These researches were done by different ways, for instance by comparing two different industries, while others did their research among complete indexes or countries. These researches also differ since different control variables and different estimation methods are being used. Some of these

(4)

4 papers were also having different purposes for doing the research, which makes it difficult to compare the results amongst each other. For instance Bennedsen, Perez-Gonzales and Wolfenzon (2009) who were searching if CEOs actually influence firm value by looking at their deaths or deaths of close relatives families, while others were only focusing on psychological effects. All these factors could have led to ambiguous results on the question whether tenure has an influence on firm value.

Hence there already have been a few efforts made by researchers to investigate the relation between CEO tenure and firm performance. However, there has never been research done while observing a complete index and mainly focusing about tenure. This research will start to fill that gap by investigating the way CEOs in the USA influence the performance within a company along their tenure. That is why we try to answer the following research question:

“Does CEO tenure has influence on the firm value of companies in the USA?”

The research consists of two parts. It starts with a literature review to determine whether individual managers including CEOs affect firm value. Afterwards we’ll also analyze throughout the literature if there already have been found a relation between CEO tenure and firm value. In chapter 3 follows the empirical research. There the data and methodology of this research will be discussed. This study continues in chapter 4 with at first the presentation of descriptive statistics and empirical results. Along the chapter some regressions will be made. After each regression follows an analysis of the results. We conclude this research with chapter five in which a summary and the conclusion of this study will be given.

(5)

5

2. Literature Review

Despite the fact that the key of the research regarding evaluating if characteristics of executives is not new, we try to evaluate and integrate all relevant studies into a new theoretical framework. This will be done throughout this chapter.

2.1 Why individual managers matter.

But first we should ask ourselves why individual managers should matter. As we have already mentioned in the introduction, managers on top of the organizations have influence on their own outcome. This is because they are inspiring, motivating and mobilizing their employees while making important decisions and representing their company all at the same time (Buozinte-Rafanaviciene, Punclizine and Turauskas, 2009). But this is not everything according to Bertrand and Schoar (2003). These two researchers have found that managers not only seem to differ in their approach towards company growth but also in their financial aggressiveness when making investment decisions. Based on these observations from these researchers, we may conclude that individual managers differ from each other. Another way to review if individual managers differ, is by testing the hypothesis that individual managers are perfect substitutes of each other.

Bertrand and Schoar (2003) argue that there are two variants to review this hypothesis. The first one is the standard agency model. This model acknowledges that managers may have discretion inside their firm. They can use this in order to adjust corporate decisions and prioritize their own objectives. The standard agency model also argues that when corporate control is weak, impact of managers increases. That may cause that corporations will adapt suboptimal strategies if the board of executives isn’t properly monitored.

The second variant of manager-specific effects in corporate practices implies that managers vary in their match quality with firms. In this case do managers not impose their discretion on the organization they lead. In this variant are these managers chosen by corporations since they have specific skills. In comparison with the first variant, the second variant will not necessarily mean that differences between managerial behavior will cause suboptimal strategies. However, this variant means in theory that some managers can perform better than others. These two variants are indicating that there are differences among individual managers.

(6)

6 2.2 CEOs

But does the outcome of managerial behavior differs between different positions? This is a crucial question since firms do not randomly appoint or fire top executives randomly. Testing for the influence of one executive on firm outcomes is quite challenging since you’ve to isolate direct effects of individual managers on firms’ outcomes. However Bennedsen, Perez-Gonzalez and Wolfenzon (2009) have found a relationship between firm performance and the impact of CEOs on performance by estimating the effect on CEO deaths and the death of CEOs closest relatives. Like they’ve mentioned in their paper is this an excellent way of analyzing whether outcome of managerial behavior differs because these shocks affect the CEOs ability to perform their jobs. This happens either directly through their own death or indirectly through their loss of their closest relatives. The last situation will give personal sadness that would tend to restrict the CEOs effective ability to execute the professional roles.

Furthermore it is also a good approach since it’s reasonable to expect that beyond its effect on the CEOs personal shock in performance particularly those associated to family members are independent to the managers firm and do not affect firms’ investment opportunities through other channels. Effects on the death of the other members of the board of directors or their direct relatives are also tested, but the researchers didn’t find any robust evidence that the death of individual board members or their immediate family affect the firm profitability. On the other hand the same researchers have found significant declines in firm operating profitability, investment and sales growth. Therefore we can say that this research indicates that CEOs are crucial in the outcome of firm value.

These findings are in line with another research done by Bennedsen, Perez-Gonzalez and Wolfenzon conducted in 2011. Using a variation in CEO exposure from the numbers of days a CEO is hospitalized they’ve observed that hospitalizations lead to significant declines in a broad set of performance measures. Just like in their previous research, they haven’t found any significant declines in performance measures when the other board members were hospitalized.

Other researchers have also found that CEOs are a key determinant of firm performance. Malmendier and Tate (2007) for instance have found that CEO overconfidence and other personnel characteristics affect firm investment decisions. Denis and Denis (1995) find also significant performance improvements when CEOs retires or are forced to resign. This indicates that new management teams improve firms’ prospects.

(7)

7 By focusing on CEOs, our aspiration is not to adore them. CEOs aren’t making all the decisions in a company, but they have big influence on the performance of their organization. Overall, no other small group have as much effect on the form and fate of an enterprise as CEOs (Finkelstein, Hambrick and Cannella, 2009, p.5). This is because the CEO is the one who has overall responsibility for the conduct and performance of an entire organization. That being said, while following the findings of the previous researches do CEOs form a key determinant of firm performance and that is why we’re going to focus on effects of CEOs during this thesis.

2.3 Psychological Factors

As already mentioned in the introduction, paradigms may affect the behavior of different managers and therefore also firm value. There has been a lot of research done concerning this question. In these studies has been found that paradigms are formed by two main individual characteristics. These are formed by either psychological or demographical characteristics.

Kaplan, Klebanov and Sorenson (2012) have done the most extensive research in the first area. The data was found by a detailed 4-hour interviews held by a firm specialized in assessing top executives. They have studied individual characteristics of CEO candidates for companies involved in buyout and venture capital transactions. They examined whether these CEO candidates by the time they were hired by the private equity firm or at the time of investment in the company. These psychological characteristics of this research were divided into two categories.

The first category was managerial talent. This means that managers have the capability to naturally outperform others. The second one is about interpersonal abilities and communication. We can interpret that as execution and resoluteness related skills, like teamwork, aggressiveness and listening. For buyout and venture capital firms CEOs success was strongly related to execution, resoluteness and overconfidence-related skills then to interpersonal-related skills. Within their data they also found some demographical information about CEO candidates, including years of management and years of experience in a financial role. Surprisingly none of these demographical two mentioned can be proxied by the assessment characteristics.

While they have created a larger dataset then some previous researchers, with having a variety between public, private and nonprofit; it Kalpan, Klebanov and Sorenson (2012) found their findings still coarse and the private equity firm measures subjective. This since

(8)

8 their results reflect buyout and VC-funded companies that may differ from other types of companies.

2.4 Industry dynamics

Like already said are paradigms of CEOs affected by demographical factors. CEOs demographic backgrounds, which are reflective of their experiences, will be associated with strategic choices (Finkelstein, Hambrick and Cannella, 2009, p.83). One example of demographical factors that may influence CEOs paradigms is tenure. From earlier researches on organizational evolution was concluded that it is likely that CEOs do not behave and think or perform uniformly over their tenures (Henderson, Miller & Hambrick, 2006) and therefore affecting the organization.

The way tenure of a CEO affects firm value depends on the industry dynamism of where the firm is located in. This can be described as the unpredictability of the change of the environment where the company is located. Throughout the literature we have found that there are two types of environmental dynamism. Either the industry or environment is defined as a stable or dynamic one. Henderson, Miller and Hambrick (2006) are giving a clear definition of dynamic industries and stable ones. Dynamic industries are defined as rapid growing, high investing and R&D intensive industries, whereas stable industries are defined as one with stable customer preferences, technologies and competitive dynamics change little. Since this definition is clear, these researchers didn’t make a model with a group of dynamic and stable environments. Instead, they simply choose two industries which had these characteristics, namely the stable food sector and the dynamic computer industry. By comparing the these two groups, they measured that firm value and tenure had different relationships with both groups.

While Henderson, Miller and Hambrick (2006) only made a research by comparing two industries with each other, Bennedsen, Perez-Gonzalez and Wolfenson (2009 and 2012) conducted a research based all limited liability held Danish firms. Which makes the internal validity of the research bigger (Stock and Watson, 2012, p.355). In their paper of 2009, they used the same definition for dividing industries into the two previous called categories. Only difference now is that they did characterize the industries by looking at data instead of theory and literature, which made their research stronger. The characteristics used were profitability, employment growth, R&D and investment growth. In this paper was clearly found that CEO shocks, whether due death or sickness, are higher in dynamic environments than stable once.

(9)

9 This confirms that it is indeed important to take industry dynamics into account when observing CEO effects.

In 2012 Bennedsen, Perez-Gonzalez and Wolfenson made a distinction whether an industry was high growing by looking at the aggregate rate of growth of total assets in the industry during their sample period. Two groups were formed by dividing the firms with high or low growth simply by looking at the median of the group sample. They found in this paper that CEOs in rapidly growing environments at industry level exhibit CEO effects on firm value. Thus these two papers also concludes that CEO effects are different in specific environments.

Another paper made by Gaver and Gaver (1993) divided up growth-firms from non-growth firms by generating an new self-made index. This index was based on the following ratios:

1. Market to book value of the assets 2. Market to book value of the equity

3. R&D expenditures to the book value of assets 4. Earnings/Price ratio

5. Variance of the total return of the firm

6. The frequency that the firm is included in holdings of growth-oriented mutual funds Firms ranked in the first quartile of the index were grouped together as growth firms, while firms at the last quartile of the index were grouped together as non-growth firms. The main disadvantage of this research was that it hadn´t got the purpose to find CEO effects on firm value. Furthermore was the index was self-made, while in the other models the pros and cons are evaluated. This is why we will make use of a combination of these papers during this thesis. Our conclusion by reading these papers is that we need to take industry dynamics into account when we want to know how tenure of CEOs is influencing firm value. In the methodology chapter we will go further into the way how we will characterize these industries.

2.5 CEO tenure in industry dynamics

So we need to account that CEOs effects depends on the environment. But does that also account for CEO tenure? When CEOs are starting in a firm they will start with a lack of knowledge. After a while they’re steadily learning about the environments, jobs and organizations (Hambrick and Fukutomi, 1991). In this period CEOs are alert at their environment and gradually implementing a strategic orientation that fits the firm (Miller,

(10)

10 1991). But after some time are CEOs becoming overcommitted in their approach. Thereby it’s getting harder and harder to weigh and to implement the changes in the environment causing lower performance (Lawrence and Lorsch, 1967).

Besides that are the benefits of internal fit for having the same CEO at the firm only accumulating for a period of ten years (Henderson, Miller and Hambrick, 2006). After some time penalties of external mismatches will occur, which are becoming substantial for CEOs who remain in office for an exceptional amount of time. When CEOs are staying longer at a firm, the chance of entrenchment will arise. Managerial entrenchment occurs when the board of directors gains such power that they are able to use the firm to foster their own interests than the interests of the shareholders. Faleye (2006) shows in his research that classified boards are associated with a significant reduction in firm value when management is entrenching. This is of course bad for the company’s firm value since managers can either enrich themselves or follow a different strategy. The chance that this occurs increases as the board, including the CEO, remains longer in the company.

Weisbach (1988) and Shleifer and Vishny (1989) have found that entrenchment also have other effects. When the board of directors are making manager-specific investments, managers can reduce the probability of being replaced, extract higher wages and making it costly to replace them. This of course has an opposite effect on what shareholders would like to achieve.

These effects only happen when a CEO has time to adapt and to learn from his or her environment, which is the case in stable industries. Like previously defined these are industries where the customer preferences, technologies and competitive dynamics change little over time(Henderson, Miller and Hambrick, 2006).

Henderson, Miller and Hambrick (2006) have also found that the relationship between CEO tenure and company performance has a reverse U-shaped relationship. The firm performance initially will have a top at the outset of the tenure, which later-on decreases. This confirms the findings of Allgood and Farrell (2003) who’ve found that the hazard of an executive during the first years increases and later on decreases. Hazard can be defined as the risk willingness of a CEO. With less willingness to go for risk, the potential returns decreases, which ultimately has therefore affect the firm value (Investopedia, 2015).

Since we expect a reverse U-shaped relationship between firm value and tenure, there should be a peak between these relationships. In dynamic industries the top of the firm value will lie in the first years of the CEOs’ tenure, since these new CEOs can align their paradigms very quickly (Henderson, Miller and Hambrick, 2006). The company performance also

(11)

11 declines earlier in dynamic industries then stable once, since benefits of internal fit are limited and the penalties imposed by external mismatches increase quickly. Besides that does the major environmental rotation make that yesterday’s knowledge becomes useless (Miller and Shamsie, 2001). On the other hand, in stable industries the top will be at ten years. Like already mentioned in the previous section, CEOs need to have certain time to adapt in a stable industry. As the environment doesn’t change a lot are the benefits of learning spread over a bigger period, since the knowledge acquired today has a bigger chance that it also will apply for tomorrow. Hence, the decline of CEOs performance will occur earlier in dynamic environments (Henderson, Miller and Hambrick, 2006).

2.6 Hypothesis

Based on the literature research do we expect a relation between CEO tenure and firm value. Following the literature investigated are external dynamics also playing a role. Therefore we have to compose two different hypothesis between dynamic and stable environments.

• Hypothesis 1A: We will expect a reverse U-shaped relationship between CEO tenure and firm value. The firm value will initially increase among tenure, but later on decrease. • Hypothesis 1B: We will also expect that the relationship between CEO tenure and

company performance will have a reverse U-shaped relationship in dynamic industry firms, although we foresee that the performance will be higher at the outset of a CEOs tenure compared with the firms in stable environments. Firm value will decrease earlier on compared to the situation of the firms in stable environments.

Although Kaplan, Klebanov and Sorenson (2012) named their outcome measures to estimate whether the effect of CEO characteristics and abilities matter for firm value subjective and coarse. Furthermore they also took other performance measures than the researchers who have been investigating demographical measures over time, namely OROA or Tobin’s Q compared to CEO was hired successfully. We still have to take the findings of Kaplan, Klebanov and Sorenson (2012) in our minds since there might occur some form of bias in our research. This because we do not take any psychological factors in this thesis into account. We have to have to pay our special attention into it since some of the demographical characteristics mentioned in the research were proxied by the psychological characteristics. Therefore hypothesis two is added, which is:

(12)

12 • Hypothesis 2: There is a possibility that we won’t find any relationship between firm

value and CEO tenure since we’ve omitted psychological characteristics in our research model. Then, has to be examined in a future research whether psychological characteristics are influencing firm value

When both hypothesis 1A and 1B are rejected, we will accept hypothesis 2. This means that we need to continue to examine whether psychological characteristics are influencing firm value. However, hypothesis 1A and 1B are based on the earlier mentioned researches in sections 2.4 and 2.5. In the next chapters we will test out our hypothesis through an empirical research.

(13)

13

3. Research Method:

In this chapter the research method will be discussed. All variables, the data and methodology will be described which are used for answering the research question.

3.1 Data

This empirical research is based on 758 of firms which are included on the S&P 500. The reason why we’ll take this index is since it includes 80% of the US market capitalization (Us Spindices, 2015). On top of that, the S&P 500 gives broader coverage than other indexes. When comparing the S&P index to the Dow Jones Industrial Average Index, the S&P has a wider range since the Dow Jones index only contains 30 companies. We also had to conclude that the S&P was a better index when comparing it with the NASDAQ since the NASDAQ is part of the S&P.

The S&P 500 is an index which contains the top 500 companies on market capitalization in the USA. The index as a whole covers around 80% of the total market capitalization in the USA. Just like Bertand and Schoar (2003), you might argue that this focus on big firms might give a bias in our results. Especially in consideration of that a specific individual might be more influential in a smaller organization. In these type of firms, personal involvement is more required by top managers in the day-to-day activities. On the other hand, you might argue that managers with more noticeable styles are more likely found in larger firms (Bertrand and Schoar, 2003). This last argument is stronger since a lot of researchers made usage of the S&P 500.

To analyze the impact of CEO tenure and firm value, we have used four main databases. The first database is COMPUSTAT. Here we collect the EBIT, total assets, market value and amount of sales for our dataset. This dataset had many gaps in the data, so that is why we filled the gaps where possible with data from YCHARTS and Orbis. EXECUCOMP, the forth database we have used in our thesis, contains all the CEO specific data. Here we got the CEO tenure, CEO compensation and CEO age.

From these 758 firms previously mentioned, we’ll exclude 96 of firms since these are financial institutions (SIC 60) and public utilities (SIC 49). Just like Gaver & Gaver (1993) mentioned, these type of firms are regulated corporations and therefore may be consistently different from unregulated corporations. We then restrict our attention to the subset of firms for which at one specific head of head of the company can be observed and since we’re

(14)

14 building our research on the S&P 500, which is a publicly traded index, we cannot include any data without any observable market value.

Furthermore will our dataset have a time span of twelve years, just like Schoar and Zuo (2011) and Bennedsen, Perez-Gonzelez and Wolfenzon (2009) had in their research. The timespan will be as close as possible from the current year, namely from 2001 until 2013. For each firm satisfying these restrictions, we keep all observations. Including all these restrictions in our dataset, our final sample size of the balanced data-panel consists of 661 corporations and a total sum of 7084 CEO year observations. The statistical analysis is carried out using the software program Stata. In addition there are there 1334 CEOs identified in the sample.

3.2 Methodology

3.2.1 Environmental dynamism:

Like already mentioned in the literature overview, we’ve to make a separation between dynamic and stable firms. Environmental dynamism can be described as the unpredictability of the change of the environment. This type of dynamism should be limited since it’s hard to predict and also increases the uncertainty for key organizational members (Dess and Beard, 1984). In the literature were Henderson, Miller and Hambrick (2006) the only ones who really focused on the relationship between company performance and CEO tenure. Unfortunately, they’ve composed a research only comparing two different industries, which makes their research less consistent those who test over all industries (Stock and Watson, 2012, p.108). Therefore we’ll test in this thesis the external validity of their findings. However, they do have formed good criteria to characterize the difference between dynamic industries and stable once. Rapid growing, high investing and R&D intensive industries are defined as dynamic once, while in the stable industry customer preferences, technologies and competitive dynamics changes little. Another disadvantage of the methodology of this paper is that they didn’t pick out the dynamic and stable industries with data but throughout literature. While observing comparable researches, we’ve found some industry characteristics picked out from data which we can adopt in our thesis.

The industry characteristics of Bennedsen, Perez-Gonzalez and Wolfenzon (2009 and 2011) were for instance derived from their data. The most important factors they’ve taken into account in their paper of 2009 were industry levels of R&D and industry investment-rates as their criteria. Unlike Henderson, Miller and Hambrick (2006) was this research done

(15)

15 over an entire private limited liability market, namely firms in Denmark. These firms were grouped in two groups, one below or above the median of each relevant variable.

One of the measures applied by Bennedsen, Perez-Gonzelez and Wolfenzon (2009) to compare industry dynamism is the growth of industry investment-rates. This is determined by calculating the growth of total assets in a year. It is a good way to measure industry dynamism since the strategy in a more dynamic and rapid growing environment aims more to invest more into the firm then when it’s in a stable environment. Potentially there will be more emphasis on reducing costs in the last mentioned situation, rather than in expanding operations, which makes the aforementioned strategy more appropriate for a dynamic industry (Bennedsen, Perez-Gonzalez and Wolfenzon, 2011). Therefore we will include this measure in this research.

With relative R&D expenditures you can measure the technological instability (Dess and Beard, 1984). Bennedson, Perez-Gonzalez and Wolfenzon (2009) however did not take relative R&D expenditures in their paper. Instead they grouped an industry in the “high” R&D group if there was reported any research and development activity in the industry and qualified “low” when there wasn’t any research and development activity measured in the industry.

Gaver and Gaver (1993) on the other hand did use relative R&D expenditures by taking the ratio of R&D expenditures to book value of the assets. However, one main disadvantage of using R&D is that it’s not relevant for all the industries. Yet using R&D ratios is seen in the literature as a good measure since this ratio clearly shows the technological instability in the industry, but also defines research intensity growth. Moreover the invention of new technologies may cause that an industry spends a lot on R&D. This indicates that the industry is changing.

Another alternative is R&D expenditure to sales ratio. This ratio is more useful to compare the effectiveness and efficiency of R&D expenditures between companies in the same industry (Wikinvest, 2015). Therefore this ratio does not show the technological instability in the industry. For that reason we make usage of the R&D expenditures to book value of the assets ratio.

Unlike Gaver and Gaver (1993) we choose not to split the firms by taking the top 25% and bottom 25% factor scores. Instead we’ll characterize the industries in two groups, one above and one under the median. This is what Bennedson, Perez-Gonzalez and Wolfenzon (2009) did in their research. We have chosen to establish a separate regression for each characteristic, in this case two (one for R&D to asset and one for industry growth). Therefore

(16)

16 composing two other regressions, one for dynamic and one for stable industries so in the end we will carry out four regressions in total.

3.2.2 Depended variables:

The regressions made during this thesis are made with use of the program Stata. As measures of the firm value are two dependent variables in the estimated model. These are the OROA and Tobin´s Q. Although Return on assets (ROA) is a widely used measure of accounting performance to compare firms (Carpenter, 2002), operating return on assets (OROA) however is a better measure since it’s unaffected by differences in firms’ capital structure decisions. OROA is measured as the ratio of earnings before interest and taxes (EBIT) to the book value of assets compared to the income-based ratio of ROA (Bennedsen, Perez-Gonzalez and Wolfenzon, 2007). These are the reasons why we include OROA as a depended variable.

Using both an accounting measure and a non-accounting measure gives better results, therefore a separate regression with another dependent variable will be made. One possible concern is that the systematic differences in rate of return on assets across managers may not reflect actual differences in performance but rather differences in aggressiveness of accounting practices or willingness to "play with the books" (Bertrand and Schoar, 2003). In order to address this concern, we use an alternative accounting measure of performance that is less subject to accounting manipulations and better captures true performance, which is the natural Tobin’s Q.

The natural logarithm of the ratio of the total market value of firm to the total assets is called the Tobin’s Q. There are several reasons why many researchers are making usage of the Tobin’s Q, not only because it’s a simple derived and COMPUSTAT covers all data necessary, but also since it’s explains cross-sectional differences in investment and diversification decisions (Chung and Pruitt, 1994). When the Tobin's Q ratio is between 0 and 1, it costs more to replace a firm's assets than the firm is worth, also known as undervalued. A Tobin's Q above 1 means that the firm is worth more than the cost of its assets, this means that the company is overvalued.

OROA = _{𝑬𝑬𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 𝑨𝑨𝑨𝑨𝑨𝑨𝑨𝑨𝑻𝑻𝑨𝑨}𝑬𝑬𝑬𝑬𝑬𝑬𝑬𝑬 and Tobin’s Q = 𝑬𝑬𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 𝑴𝑴𝑻𝑻𝑴𝑴𝑴𝑴𝑨𝑨𝑻𝑻 𝑽𝑽𝑻𝑻𝑻𝑻𝑽𝑽𝑨𝑨_{𝑬𝑬𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 𝑨𝑨𝑨𝑨𝑨𝑨𝑨𝑨𝑻𝑻𝑨𝑨}

By having two separate regressions with one market-to-book ratio, makes our research also stronger as market-to-book also measures not only the current earnings but also

(17)

17 the future once (Gaver and Gaver, 1993). You might formulate that by adding the Tobin’s Q, we also add the growth potential of the firm in our research.

3.2.3 Independent variables

In this thesis we will investigate the relationship between tenure and firm value. Like already said at the hypothesis we need to take industry dynamics into account in our model by separate them by the R&D to assets ratio and the growth of total assets. But there are also some other variables which we should take into account in our models in order to have more unbiased results, as omitted variable bias may occur (Stock and Watson, 2012, p. 222). This is since firm value is not only specified by CEO tenure.

But it is obvious that the first independent variable included in our model will be the tenure of a CEO itself. The tenure will be measured by counting the years a CEO has been in office. During the implementation of the research we’ll also look if the square of tenure will be significant and if it should include the regression or not. This will be an important measure, since without the square root there isn’t anything in the regression analysis what will determine the curve. Like said by Finkelstein (1990) and Bertrand and Schoar (2003) are managers nowadays changing quite often from industry. CEOs need relatively less industry and firm specific knowledge and have to rely more on general management skills. Therefore we don’t take into account that the CEOs are in the same industry in the tenure. CEO tenure is going to be measured by counting the years a chief executive had been in office as a CEO within the specific company. Unless otherwise noted, all predictors were updated annually and lagged by 1 year.

The first other independent variable different from tenure that will be included in our regression model is the factor time. During time certain technological developments, innovation or competitive rivalry can take place which effects the firm value over time (Geletkanycz and Hambrick 1997). Also may time correlate with tenure since the further we go into time, the greater the chance is that the tenure of a CEO increases. The time factor will be measured as the numbers of years passing by within the sample period.

Although managers do change a lot within industries, we still have to take industry into account as a separate dummy variable. This since in industries there are industry specific factors. These factors are for example time factors, the business cycle, industry variance, competitive rivalry and supply instability factors which has an effect on firm value (Henderson, Miller and Hambrick, 2006). We will these factors into account by creating a

(18)

2-18 digit SIC dummy variable in the regression. By doing so, we cancel out the abnormal or disappointing returns in certain industries. We will only report the adjusted industry figures.

Another variable that we will include in the regression equation is firm size. This can correlate with firm returns since at bigger firms might occur size advantages like for example economies of scale, product differentiation or bigger brand loyalty hold for bigger firms. Like Allgood and Farrel (2000) we use the logarithm of sales as a firm size measure. This means that a 1% change in firm size, or sales, changes the firm value with (firm size)*0,01. We add firm size as a logarithm because we want to reduce skewness in our model and to linearize the relationship between firm size and firm value. Size is also an important feature as CEOs might stay longer within a firm than CEOs in a smaller firm or the other way around.

When a CEOs total compensation is higher, the willingness for a CEO to outperform will get bigger. Therefore the possibility for abnormal returns within a firm with higher compensation packages will increase. By taking the logarithm of the total compensation of a CEO, which is the salary, bonus and value of stock options granted in a year we can take into account of this measure in our model. This variable will be just like firm size expressed as a natural logarithm in the model. On the other hand, the willingness of the CEO to stay within a firm will increase if he gets a higher compensation, therefore tenure should be positively correlated with the total compensation package.

Furthermore we also take into account to the age of CEOs by looking at the birth cohort of a CEO. The cohort will be ten years, just like Bertand and Schoar (2003) did in their research. We’ve to take into account the age of a CEO, since apparently CEOs from older generations tend to appear more conservative in the decision-making in for example capital expenditures (Allgood and Farrel, 2003). These executives are therefore less investment or risk-taking, which is a common strategy when CEOs are getting older. This influences the firm value. We’ll also add the cohort of CEO age, because we expect that age also has a positive relationship with tenure. Therefore CEO cohort will be added in our regression.

(19)

19 3.3 The final regression model

OROA = β0 + β1*(CEO Tenure) + β2 *(CEO tenure)2 + β3*log (size) + β4*time variable (i) +

β5*Log (CEO compensation) + β6*(CEO Cohort) + β7*(industry) + ε

Whereby ε = the error term

Log (Tobin’s Q) = β0 + β1*(CEO Tenure) + β2 *(CEO tenure)2 + β3*log (size) + β4*time

variable (i + β5*Log(CEO compensation + β6*(CEO Cohort + β7*(industry) + ε

(20)

20

4. Results

This chapter provides an overview of the results found in the previously described method of data collection. This chapter starts with the descriptive statistics in 4.1. The regression analysis will be presented and discussed in part 4.2.1 and 4.2.2. Lastly, in part 4.3 robustness checks will be described.

4.1 Descriptive variables and correlation mix

The total number of companies included in this analysis is 661. Overall, there are 7083 firm-year or CEO-firm-year observations over the period 2001-2013. Over this period, 1356 CEOs were observed, which had throughout the sample period an average tenure of 7 years, which is comparable with the average stay of managers in firm Henderson, Miller and Hambrick found in their data, namely 6 years (2006). Another observation which is comparable in with their research is the skewness of CEO tenure distribution. Most of the CEOs are leading the firm for less than fifteen years. Furthermore we also have to mention that concerning tenure our sample tenure matches with the literature. Most values lay between one and 36 years, just like (Henderson, Miller & Hambrick, 2006).

As shown in table 1, for the variables log (CEO compensation), the R&D variables and Firm size the number of observations are being decreased, this is due several reasons. First of all in the sample period are some CEOs observed without any CEO compensation package. Taking the logarithm of zero is not possible, hence the number of observations decrease automatically. On top of that it was sometimes impossible to find the R&D expense of certain firms. Therefore Stata dropped the number of observations automatically.

A normal distribution of a continuous random variable is a key assumption in statistics when using these variables in composing regressions. For all variables in our regression derived in section 3.4 whether this assumption holds. This does not hold for the total compensation variable, expressed in thousands the sales size variable. Therefore the histograms for both variables are provided on the next page to give more information.

(21)

21

Table 1 Descriptive statistics

Variable Obs. Mean Std. Dev Min Max

Tenure 7083 6.64 6.486 0 51

Tenure2 ₇₀₈₃ _86.161 _190.402 ₀ ₂₆₀₁

CEO age 7083 55.641 6.778 28 85

CEO cohort 7083 5.564 0.678 2.8 8.5

CEO Compensation 7083 9698.539 10393.91 0 243558.1

Log (CEO Compensation) 7071 8.77 1.217 -6.908 12.403

EBIT (Millions) 7083 2094.716 6147.042 -80053 130622

Total Sales (Millions) 7083 14886.06 31235.49 -2758.73 474259

Total Assets (Millions) 7083 35854.29 148026.9 40.463 3270108

Total Market Value (Millions) 7083 19440.71 38846.17 3.38 626550.4

Firm Size log (Sales) 7079 8.68 1.338 2.716 13.07

R&D 6997 346.73 1024.639 0 12183

R&D/BV 6997 0.283 0.054 0 0.887

OROA 7083 0.103 0.095 -1.093 0.858

Tobin’s Q 7083 1.449 1.339 0.00008 15.531

Log (Tobin’s Q) 7083 -0.544 1.089 -9.446 2.743

Figure 1A. Distribution Total Compensation Figure 1B. Distribution Log (Total Compensation)

The distribution of the Total Compensation package of CEOs is given in figure 1A. Figure 1A is definitely not bell-shaped like a normal distribution should be, but heavily skewed in a positive way. To get of this skewness, the logarithm can be taken. The result we see in figure 1B, where we indeed can observe a more bell-shaped distribution when the logarithm is taken over that variable. We still observe some negative values. Logarithms are not accepting negative values. Winsorizing might be a possible solution for taking care of these negative values, since this distribution is negatively tailed. Winsorizing means that we will modify one or more data points at the end of the tails of the distribution to the next lowest or higher values within the distribution that are not suspected to be outliers. We do

(22)

22 this in order to prevent that we make wrong conclusions out of our data set due typographical errors, measurement errors or contaminated distributions (Hawkins, 1980). We only will winsorize throughout our dataset at a 99% level.

Figure 2A. Distribution Sales Figure 2B. Distribution Log (Sales)

Just like the distribution of total compensation package of CEOs, the sales variable distribution is also heavily skewed in a positive way and therefore not normally distributed. This is why we took also in this case the logarithm from this variable. Unlike the logarithm distribution in figure 1B, this distribution isn’t heavily tailed in a negative way. Nevertheless we still need to winsorize this variable, in order to prevent that we make wrong conclusions out of our data set due typographical errors, measurement errors or contaminated distributions.

Observing the distributions of the two performance measures in our sample, it’s not necessary to create logarithms over the OROA performance measure it includes negative values. Taking the logarithm over negative values are not accepted, which means a potential significant loss in information when we decide to take the logarithm over the OROA. Since, at the Tobin’s Q are no negative values observed, we’ll decide to take the logarithm over it. When we examine the distribution of the Tobin’s Q we observe, just like the distribution of sales and total compensation, a heavily skewed distribution in a positive way. When taking the logarithm we find a bell shaped distribution. Therefore it’s necessary take the logarithm over the Tobin’s Q in our regression analysis. The figures can be found in appendix A.

On page 23 we have added a correlation matrix, since correlation is an important aspect when interpreting the results from a regression. The definitions and descriptions of the abbreviations of all variables can be found in appendix B. We’ve used the command “pwcorr” in Stata since by using this command we’ll exclude missing values and only calculates the correlation of the data inserted in Stata. Given the correlation matrix, we need

(23)

23 to look at the most relevant and important values. The most important are those of the included variables in the regression model.

The correlations matrix shows that a lot of variables are significantly negatively or positively correlated with CEO tenure. For both dependent variables, CEO tenure gives a significant correlation, which indicates that when tenure increases firm performance increases. With respectively a positive and significant correlation of 0.4032 between cohort and tenure, indicates that CEOs from older generations have a bigger likelihood to stay longer in a position, whether this is due their age or their experience.

Also a crucial feature is that Tobin’s Q is negatively correlated with time. Normally due technological progress this should be positively. But due the global crisis in 2008, which insured for more competitive dynamics but also for a negative investing climate throughout all industries, caused that market values dropped significantly within the sample period.

With respect to the two dependent variables, Tobin´s Q and OROA, the correlation matrix appears to give a significant correlation of 0.5578. This means that firms with higher profits usually have larger profits.

(24)

24

Table 2 Correlation matrix

The individual correlation is statistically significant at: ***1%, **5% or *10% significance level (1-tailed). Below the correlations are the p-values given

Year BV Sale MV OROA Tobin’s Q RDBV IND Coh Ten TC Ten2 Log(s) Log(TC)

Year 1.000 BV 0.0566*** 1.000 (0.0000) Sale 0.098*** 0.3991*** 1.000 (0.0000) (0.0000) MV 0.0746*** 0.3089*** 0.685*** 1.000 (0.0000) (0.0000) (0.0000) OROA 0.0524*** -0.1121*** -0.0006 0.1188*** 1.000 (0.0000) (0.0000) (-0.9628) (0.0000) Tobin’s Q -0.0889*** -0.1591*** -0.1288*** 0.1018*** 0.5578*** 1.000 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) RDBV -0.021** -0.0841*** -0.1001*** 0.0145 -0.1063*** 0.2829*** 1.000 (0.078) (0.0000) (0.0000) (0.2238) (0.0000) (0.0000) IND 0.0129 0.138*** 0.029** 0.0272** -0.0449*** 0.0172 -0.0776*** 1.000 (0.2765) (0.0000) (0.0146) (0.0222) (0.0002) (0.1483) (0.0000) Coh 0.061*** 0.0563*** 0.0987*** 0.0691*** 0.0613*** -0.0752*** -0.1258*** -0.0597*** 1.000 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Ten 0.0334*** -0.0305** -0.0233** 0.0158 0.0428*** 0.0917*** 0.0124 0.0242** 0.4032*** 1.000 (0.0049) (0.0103) (0.0498) (0.1840) (0.0003) (0.0000) (0.2963) (0.0415) (0.0000) TC 0.0355*** 0.1187*** 0.209*** 0.2544*** 0.0106 -0.0002 -0.0166 0.0007 0.0832*** 0.0488*** 1.000 (0.0028) (0.0000) (0.0000) (0.0000) (0.3734) (0.9852) (0.1635) (0.9534) (0.0000) (0.0000) Ten2 0.0075 -0.0017 0.0047 0.0365*** 0.03** 0.0732*** 0.0025 0.0499*** 0.3646*** 0.9077*** 0.0405 1.000 (0.5275) (0.8852) (0.6914) (0.0021) (0.0115) (0.0000) (0.8304) (0.0000) (0.0000) (0.0000) (0.0006) Log(s) 0.1683*** 0.3391*** 0.6642*** 0.5349*** 0.0556*** -0.2679*** -0.2597*** -0.0238** 0.1493*** -0.0678*** 0.3058*** -0.0261** 1.000 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0280) Log(TC) 0.1071*** 0.0536*** 0.1594*** 0.0988*** 0.0331*** -0.09*** -0.039*** -0.0651*** 0.0904*** -0.0273** 0.5902*** -0.0399*** 0.3006*** 1.000 (0.0000) (0.0000) (0.0000) (0.0000) (0.0053) (0.0000) (0.0010) (0.0000) (0.0000) (0.0218) (0.0000) (0.0008) (0.0000) Log(Q) -0.0687*** -0.4481*** -0.1627*** 0.1027*** 0.526*** 0.7481*** 0.2637*** -0.109*** -0.0612*** -0.0129 0.1033*** 0.0755*** -0.2824*** -0.0524*** 1.000 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.2758) (0.0000) (0.0000) (0.0000) (0.0000)

(25)

25 4.2 Regression analysis, winsorize

This part starts with an overview of results calculated by the method Ordinary Least Squares (OLS). The research method that we’re going to matches with the literature discussed in the previous chapters within this thesis. After discussing an overview of the results calculated by OLS are we going to discuss the implications of these results. As previously described, the sample period runs from 2001 to 2013. This sample period consists of 661 corporations and a total sum of 7084 CEO year observations.

4.2.1 Ordinary least squares

The hypothesis for company performance is tested and performed with two types of depended variables in order to make our research stronger. These two performance measures are the Tobin’s Q and the OROA. Subsequently, we found in the literature that it’s necessary to split the firms based on their industry dynamism. Therefore we had to split up our sample size into two industry pools, which can be found in appendix C. Noticeable is that the food industry, SIC 20, is in the dynamic industry when it comes to asset growth. When it’s characterized by the R&D to asset ratio food industry is part of the stable industry. The first result is ambiguous compared to the theory of Henderson, Miller and Hambrick (2006) since they clearly argued in their paper it belongs in the stable industry. The computer industry, with SIC 35 is in both industry dynamism measures placed in the dynamic group. This is in line with the theory out of the previously called researchers.

Furthermore are the two industry pools, again one for an industry situated in a dynamic environment and one for a stable one, having different hypothesis. Our expectation is that there is a reverse u-shaped relationship between firm value and CEO tenure. The decrease of company performance will outset around the tenth year of the tenure of the CEO in a stable environment and in an earlier stage in a dynamic environment. When these two conditions are found in our regression analysis, both hypothesis 1A and 1B from part 2.6 are accepted. If not, hypothesis 2 is accepted, which will mean more research would be done about this subject.

In all our regressions we’ll make usage of robust variance estimates in order to account for heteroscedasticity and unobserved differences across firms and CEOs (White, 1980). Table 3.1A starts to show the ordinary least squares output for a firm which lies in a stable environment (change into stable). In column 1a and 2a we only regress firm value on tenure. In column 1b and 2b we add tenure squared in the regression. We’ve to add that in our regression in order to create a possible reverse u-shaped relationship between firm value and

(26)

26 tenure. We see that there is a reverse u-shaped relationship observable in column 1b, but unfortunately not in column 2b.

Observing only these two regressions could give inaccurate conclusions since the regression might be subjected to omitted variable bias. That is why columns 1c and 2c repeats the previous regressions with the control variables mentioned in the previous chapters based on the literature. We can see that after adding the control variables, the R2 rises substantially. R2 means the fraction of the sample variance of the dependent variable that is explained by the regressors. Hence, that increase means that the fit of the regression has been improved. Normally the adjusted R2 is a better variable since the R2 always increases when a variable is added within the regression and the adjusted R2 not. Due the large sample size, the differences between these two features are small. Therefore the adjusted R2 is not notated in the tables. The observations of table 3.1A about the increasing R2 after adding control variables are in line with the other regressions of 3.1B, 3.2A and 3.2B. That is why we continue to report only the observations of the regressions of the regressions with the control variables.

But simply observing and comparing the different R2 is not enough. The F-statistic can help us to test joint hypothesis about regression coefficients (Stock and Watson, 2012, p.263). At the bottom of all the tables contains the F-statistic values and p-values for the overall regression that all the slope coefficients are zero. Under the null hypothesis, none of the regressions explains any of the variation in the dependent variable. Except for the regressions in columns 4b, 6a and 6b we can observe that the p-values are significant at a 1% level. This means that our model has a strong model and therefore we can continue to pay attention to tenure and tenure2 coefficients.

4.2.2 Robust results

As earlier described is our main goal of this thesis to check whether we can find a reverse u-shaped relationship between firm value and CEO tenure. Therefore we need to perform t-tests in order to evaluate the significance of the tenure and tenure2 regressors. Moreover we can tell with the information from these regressions whether firm performance in a dynamic environment is much higher at the outset of a CEOs tenure compared with firms situated in a stable environment. We determined that the performance of CEOs in a stable environment should decrease after ten years (Henderson, Miller and Hambrick, 2006).

Clearly observable in table 3.1A column 1c and 2c are that the coefficients of tenure and tenure2 give a reverse u-shaped relationship for both performance measures. This is in

(27)

27 line with the regression in table 3.1B, where we conduct the same regression but then with firms in a dynamic environment. However, in column 1c we have to mention that the coefficient tenure2 is insignificantly small, while the coefficient is significant in column 2c. Therefore we need to reject our u-shaped relationship hypothesis for the Tobin’s Q performance measure when firms are divided due to the dynamism of asset growth. In contradiction we’ve found when using the OROA as a performance measure significant results for tenure and tenure2 coefficients when regressing with the control variables. Moreover we’ve found that in the dynamic group the performance of CEOs with longer tenures decrease earlier then CEOs in a firm situated in a stable environment. This is in line with hypothesis 1B. On the other hand there is still a part which is not in line with hypothesis 1B since performance is not higher at the beginning of the tenure of a firm. Therefore we’ve to reject hypothesis 1B for now, although we’ve found that firm value decreases after ten years a CEO has been in office.

Tables 3.2A and 3.2B are using the same variables and data for calculating the coefficients. The only difference in these regressions is that industry dynamism is based on the median of the R&D to book value ratio instead of using asset growth. Furthermore we regress again with using robust standard errors. We can find in columns 5c, 6c, 7c and 8c a 1% significant reverse u-shaped relationship for all the regressions containing control variables. The coefficient for tenure2 in column 6c is still significantly negative, but we have to note that the number was less than five decimal places. Therefore we can say that hypothesis 1A is accepted.

Moreover we can observe that firm performance is higher in column 7c than 5c at the outset of a CEOs tenure. This means that firm performance is higher when a firm is situated in a dynamic environment than a stable one. However, hypothesis 1B is still rejected when using the Tobin’s Q as a performance measure since firms in a stable environment firm performance is decreasing earlier.

When using OROA as a performance measure we have to conclude that we won’t reject the 1B hypothesis. This because we observe that the performance of dynamic firms are higher at the outset and are decreasing earlier during a CEO tenure then when the firm is situated in a stable environment. Therefore both hypothesis 1A and 1B hold when we use the OROA as a performance measure and the R&D as a measure to indicate/divide industry dynamism. However, the top of the business performance in a stable environment after a period of ten years of a CEO is not found. The top lies in the fifteenth year and therefore we can say that our findings in column 6c and 8c are in line with the theory.

(28)

28 Overall, the ordinary least squares regressions give support concerning the 1A hypothesis that there is a reverse u-shaped relationship between tenure and firm performance. Furthermore there is some support concerning hypothesis 1B. Further investigation on the results on the winsorize check will tell us more whether we should or shouldn’t reject the 1B hypothesis.

(29)

29

Table 3.1A OLS

Stable asset

Independent variable Log (Tobin's Q) OROA

1a 1b 1c 2a 2b 2c Constant -0,4994 -0,5967 -0,1092 0,0827 0,0767 -0,0490 (0,0337) (0,0440) (0,3087) (0,0024) (0,0033) (0,0262) Tenure 0,0208*** 0,0500*** 0,0312*** 0,0006*** 0,0024*** 0,0020*** (0,0036) (0,0080) (0,0071) (0,0002) (0,0006) (0,0006) Tenure2 -0,0011*** -0,0003 -0,0001*** -0,0001*** (0,0003) (0,0002) (0,0000) (0,0000)

Log (Total Compensation) 0,2374*** 0,0112***

(0,0291) (0,0028) Cohort 0,0114 0,0108*** (0,0301) (0,0028) Year -0,0132*** 0,0011*** (0,0048) (0,0004) Log (Sales) -0,1822*** -0,0004 (0,0229) (0,0021) N 2727 2727 2717 2727 2727 2717 F-statistic 33,47*** 25,02*** 47,91*** 6,43*** 7,32*** 29,61*** R2 _0,0125 _0,0181 _0,3861 _0,0018 _0,0053 _0,2355

***, **, and * denote significance at the 1, 5, and 10 percent levels, respectively. Only for the independent variables tenure and tenure2

, we have conducted a one sided t-test instead of a two sided t-test. Following the hypothesis , the expected outcome of the coefficient of tenure had to be positive while for tenure2 negative.

(30)

30

Table 3.1B OLS

Dynamic Asset

3a 3b 3c 4a 4b 4c Constant 0,0553 0,0274 1,7372 0,1093 0,1091 -0,0111 (0,0230) (0,0306) (0,1703) (0,0021) (0,0028) (0,0245) Tenure 0,0127*** 0,0208*** 0,0276*** 0,0005*** 0,0005 0,0017*** (0,0021) (0,0053) (0,0044) (0,0002) (0,0005) (0,0005) Tenure2 _-0,0003* _-0,0005*** _0,0000 _-0,0001*** (0,0002) (0,0001) (0,0000) (0,0000)

Log (Total Compensation) 0,0029 -0,0013

(0,0124) (0,0012) Cohort -0,0936*** 0,0062** (0,0214) (0,0028) Year -0,0106*** 0,0004 (0,0034) (0,0004) Log (Sales) -0,1482*** 0,0091*** (0,0124) (0,0018) N 4356 4356 4350 4356 4356 4350 F-statistic 36,34*** 18,39*** 58,13*** 4,7*** 2,38** 90,94*** R2 _0,0071 _0,0077 _0,4081 _0,0011 _0,0011 _0,1622

(31)

31

Table 3.2A OLS

Stable R&D

5a 5b 5c 6a 6b 6c Constant -0,5437 -0,5982 1,0291 0,0935 0,0930 0,1053 (0,0342) (0,0464) (0,3084) (0,0022) (0,0029) (0,0245) Tenure 0,0174*** 0,0324*** 0,0311*** 0,0003 0,0005 0,0013*** (0,0029) (0,0075) (0,0061) (0,0002) (0,0006) (0,0005) Tenure2 _-0,0005*** _-0,0006*** _0,0000 _0*** (0,0002) (0,0002) (0,0000) (0,0000)

Log (Total Compensation) 0,1435*** 0,0065***

(0,0337) (0,0018) Cohort -0,0504 0,0000 (0,0312) (0,0022) Year -0,0082* 0,0003 (0,0049) (0,0003) Log (Sales) -0,2461 -0,0088*** (0,0275) (0,0018) N 3033 3033 3018 3033 3033 3018 F-statistic 35,13*** 17,97*** 78,85*** 2,35 1,19 65,39*** R2 _0,0093 _0,0108 _0,4664 _0,0007 _0,0007 _0,3579

(32)

32

Table 3.2B OLS

Dynamic R&D

7a 7b 7c 8a 8b 8c Constant 0,1059 0,0832 1,5787 0,1018 0,0988 -0,0817 (0,0202) (0,0268) (0,2131) (0,0024) (0,0032) (0,0218) Tenure 0,0193*** 0,0269*** 0,0308*** 0,0010*** 0,0019*** 0,0023*** (0,0021) (0,0054) (0,0051) (0,0003) (0,0007) (0,0007) Tenure2 _-0,0003 _-0,0005*** _0,0000 _-0,0001*** (0,0002) (0,0002) (0,0000) (0,0000)

Log (Total Compensation) 0,0083 -0,0010

(0,0129) (0,0013) Cohort -0,0543*** 0,0139*** (0,0210) (0,0030) Year -0,0144*** 0,0010*** (0,0033) (0,0004) Log (Sales) -0,1028*** 0,0135*** (0,0117) (0,0018) N 3986 3986 3985 3986 3986 3985 F-statistic 81,6*** 40,8*** 41,14*** 12,43*** 6,67*** 18,58*** R2 _0,0192 _0,0196 _0,2185 _0,0035 _0,004 _0,1252

(33)

33 4.3 Winsorize checks

When making the regressions, we will make usage of winsorizing. This means that we will modify one or more data points at the end of the tails of the distribution to the next lowest or higher values within the distribution that are not suspected to be outliers. We do this in order to prevent that we make wrong conclusions out of our data set due typographical errors, measurement errors or contaminated distributions (Hawkins, 1980). We only will winsorize throughout our dataset at a 99% level. The results can be found in appendix E, which contains table 4.1A, 4.1B, 4.2A and 4.2B. The data of the tables is based on whether the industry dynamism is characterized by assets or R&D. Each category has just like in the previous part a different table for firms situated in a dynamic or stable environment.

The results are consistent with what we´ve found in the previous tables. The tenure coefficient is significantly positive at 1% level when we are creating a simple regression between tenure and firm value. When we’re adding the tenure2 coefficient we can clearly observe a reverse u-shaped relationship between tenure and firm performance. Most of the time this u-shaped relationship is significant at a 1% level. There are some exceptions with a 5%, 10% significance level or no significance at all. When we add the control variables into the model, we see that these exceptions decrease. These exceptions are caused by omitted variable bias since the R2 and the F-statistic values increases. For regressions containing the control variables we find again a significance level of 1%, which means that we can continue to pay attention to tenure and tenure2 coefficients.

In table 4.1A we do the same regression as in 3.1A. The same also applies for tables 3.1B, 4.1B, 3.2A, 4.2A, 3.2B, 4.2B Only difference now is that the results are based on winzorising. The variables that we’ve winzorised are log total compensation, log Tobin’s Q, log of sales and the OROA depended variable. These variables were still heavy tailed, sometimes even after we took the logarithm of the variable. The tenure2 coefficient is not significantly like in column 1c. Therefore we need to reject the 1A hypothesis from section 2.5, just like we did for the regression of column 1c. When we take instead of the Tobin´s Q the OROA as a performance measure, we find a significant reverse u-shaped relationship between tenure and firm value. We also won’t reject 1B hypothesis since when the firm is located in a dynamic environment, the firm performance is higher at the beginning of a CEOs tenure compared with firms situated in a stable environment. The year of drop is again higher than found in the literature, at the twenty-first year instead of the tenth year for firms in a stable environment.

(34)

34 We also find some consistent observations when we compare the results from tables 4.2A and 4.2B with 3.2A and 3.2B. Column 13c and 15c give the regressions with as firm performance measure the Tobin´s Q. In tables 3.2A and 3.2B we´ve found that the 1A hypothesis wasn´t rejected. However, hypothesis 1B was rejected since the peak for firm performance while lies at longer tenures for firms with in a dynamic environment (42 years) than firms in a stable one (28 years). When we examine the same tables, but then taking the OROA as the firm performance measure, we identify the same findings as noted in section 4.2.2. The peak lies in shorter tenures in the dynamic firm pool than the stable one. Also we find that firm performance is higher at shorter tenures in the firms situated in the dynamic pool than the stable one. Therefore for these regressions hypothesis 1B is not rejected.

Further research, needs to be conducted in the future, since we´ve found some ordinary findings in the control variables. One of the findings was really ordinary is the fact the time regressor is negative in almost all of the regressions with the dependent variable Tobin´s Q. Like already said, Tobin´s Q is a measure for market value and OROA was a book value measure. Our expectation was that the time regression coefficient year was positive for both measures. But while conducting the regression in our sample, the time regression variable was negative when Tobin´s Q was being used and positive when the OROA measure was being used. This can indicate that we are dealing with omitted variable bias. After all, the crisis of 2008 have lead that stock prices fell. But this didn´t cause that book values of companies fell more. Therefore more research should be done on this matter.

Not only the time regression coefficient had some ordinary findings. Also the sales variable. Apparently this variable had a negative relationship with firm value, while we had predicted this should be positive. There might be another better variable for size. There was no possibility during to research to further investigate this manner.

An analysis of CEO tenure on firm value in the USA?