Does distance matter? : the effect of the firm location on CEO turnover

(1)

MASTER THESIS

DOES DISTANCE MATTER?

THE EFFECT OF THE FIRM LOCATION ON CEO TURNOVER

Eva-Christina Höfner

Student-ID: 11086149

Submitted to the

University of Amsterdam, Amsterdam Business School

For the degree of

Master of Science in Business Economics – Finance

Thesis supervisor: Dr. Torsten Jochem

June 2016

(2)

STATEMENT OF ORIGINALITY

This document is written by Eva-Christina Höfner who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

ABSTRACT

This thesis is the first to directly test for the effects of the firm location on CEO departure and replacement types in the U.S.. I argue that firm location influences CEO turnover types since geographic remoteness of the firm’s location is one possible source of friction in the labor market for managers. First, I hypothesize that CEOs in remotely located firms have a lower probability of a forced turnover than CEOs in urban-based firms. Second, at the hand of a smaller labor market, a company insider more often replaces CEOs in remotely located firms. I find that geographic location has no significant effect on differences in CEO departure and replacement types. In line with other researcher’s findings my results indicate, among other things, that forced turnover is inversely related to firm performance. Moreover, firms that have succession plans are more likely to replace the incumbent CEO with an appointed heir apparent.

(4)

3.1.5 Controls ... 11 3.2 SUMMARY STATISTICS ... 13 3.3 METHODOLOGY ... 14 4 RESULTS ... 16 4.1 UNIVARIATE RESULTS ... 16 4.1.1 Departure types ... 16 4.1.2 Replacement types ... 17 4.2 MULTIVARIATE RESULTS ... 18 4.2.1 Departure types ... 18 4.2.1.1 Forced Turnover ... 18 4.2.1.2 Other turnovers ... 19 4.2.2 Replacement types ... 20

4.2.2.1 Company insider replacements ... 21

4.2.2.2 Other replacement types ... 22

5 ROBUSTNESS CHECKS ... 23

6 DISCUSSION ... 24

7 CONCLUSION ... 26

REFERENCES ... 28

(5)

1 INTRODUCTION

This thesis investigates the idea that geographic remoteness influences CEO turnover. In light of the ongoing discussion about the importance of geographic location in finance related topics, this thesis adds to existing literature by connecting two fields of study – geographic remoteness and CEO turnover. This thesis is the first to directly test for the effects of the firm location on CEO turnover in the United States (U.S.). I argue that firm location influences CEO turnover since geographic remoteness of the firm’s location is one possible source of friction in the labor market for managers.

This paper is related to the emerging literature on the relevance of geographic remoteness. Former research has shown that geography explains differences in finance related topics. For instance, rural firms use more debt (Loughran and Schultz, 2006) and pay higher dividends than urban located firms (John et al., 2011). This thesis is also related to CEO turnover literature (see e.g., Coughlan and Schmidt, 1985; Weisbach, 1988; Murphy and Zimmermann, 1992; Parino, 1997; Jenter and Kanaan, 2015) and studies about geographically segmented labor markets (e.g., Kedia and Rajgopal, 2009). Lastly, this thesis complements studies about succession planning (e.g., Weisbach, 1988).

Taking the outcomes of several studies about geographic remoteness and CEO turnover into account and connecting these with findings about geographically segmented labor markets, I hypothesize that CEOs in remotely located firms have a lower probability of a forced turnover than CEOs in urban-based firms. I argue that studies about the “urban wage premium” do not only predict that wages should be significantly higher in large metropolitan areas, they also predict that there is more competition among CEOs in urban areas than among CEOs in non-urban areas since high ability individuals prefer urban areas (see e.g., Francis et al., 2012). If the supply of potential candidates is bigger in urban areas, the risk of firing one CEO and not finding an appropriate new candidate is much lower in urban than in remotely located firms. This implies that CEO turnovers in remotely located firms are less performance sensitive than in urban-based firms. In contrast, badly performing CEOs in remote locations may not be fired as quickly as the supply is much smaller here. Moreover, I hypothesize that a company insider more often replaces CEOs in remotely located firms. If remotely located firms are aware of the smaller labor market and the short supply of high ability individuals, I expect them to replace the leaving CEO more often with a company insider, in comparison to their non-urban counterparts. Thus, I expect them to have better succession planning.

(6)

I perform estimation of 2,063 turnovers over a sample period from 1992 until 2006. In line with academic literature, I define a firm’s location as the location of it’s headquartering, as most strategic managerial decisions are rendered from the firms’ headquarters (see e.g., Davis and Henderson, 2004). I then use several measures of geographic location in order to classify firms as either centrally located or not.

I find no conclusive evidence that geographic location explains significant differences in CEO departure and replacement types between urban firms and their non-urban counterparts. Urban and remotely located firms fire their CEOs equally likely. Above that, I do not find evidence that remotely located firms replace the leaving CEO more often with a company insider. However, in line with other researcher’s findings, my results indicate that forced turnover is inversely related to firm performance. I find that older CEOs and those with longer tenure are fired less often. Moreover, firms that have succession plans are more likely to replace the incumbent CEO with an appointed heir apparent.

My empirical analysis starts with univariate results and continues with multivariate tests that base controls on existing work on the determinants on CEO turnover. In order to determine the robustness of my results, I conduct a number of robustness checks including alternative variable definitions and different sample periods.

The remainder of this thesis is structured as follows. Section 2 summarizes the existing literature surrounding this topic and results in the formulation of the hypotheses. Section 3 describes the sample, data, and the variables as well as the methodology. Section 4 presents univariate and multivariable empirical results on the relation between geographic location and CEO turnover. Section 5 consists of robustness checks. Section 6 provides discussion and the last section concludes.

2 LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT

This section starts with a review of the existing literature. I will summarize empirical studies about the importance of geographic location in corporate governance, followed by research done on CEO turnover, geographically segmented labor markets and succession planning. Taking these studies into account, I derive two hypotheses about the connection between geographic remoteness and CEO departure as well as replacement types.

(7)

2.1 LITERATURE REVIEW 2.1.1 Geographic remoteness

First, this thesis is related to the emerging literature in the relevance of geography in finance and corporate governance related topics. Former research has shown that geographic remoteness explains differences in finance and corporate governance related topics.

Loughran and Schultz (2005) examine the impact of geographic location on liquidity for U.S. rural- and urban-based companies. They find, after adjusting for size and other factors, that rural firms trade much less, are covered by fewer analysts, and are owned by fewer institutions than urban firms. Furthermore, trading costs are higher for rural firms, and volume that can be attributed to market wide factors is lower for rural firms. Loughran and Schultz (2006) find that rural firms wait longer to go public, are less likely to conduct seasoned equity offerings and have more debt and less equity in their capital structure than otherwise similar urban firms. Francis et al. (2007) find that the location of corporate headquarters significantly affects the firm’s bondholders. They show that firms in remote areas experience greater costs of debt mainly because it is more difficult to monitor insider activities in such firms. Kedia and Rajgopal (2009) find that the location of firms’ headquarters explains variation in broad based options grants after controlling for industry effects and firm characteristics. Moreover, remote location explains higher dividend payments and a preference for regular dividends over repurchases and special dividends (John et al., 2011). Knyazeya et al. (2013) examine how local director labor markets affect board composition choices. They show that proximity to larger pools of local director talent leads to more independent boards for all but the largest quartile of the S&P 1500.

2.1.2 CEO turnover

Second, this thesis is closely related to papers on CEO turnover.

Eisfeldt and Kuhnen (2013) provide a competitive assignment model that can be used to understand the dynamics of CEO turnover. Most other research focused on the connection between CEO turnover and firm performance. The major result those studies indicate is that CEO turnover is inversely related to firm performance – CEOs are more likely to be dismissed when stock price and accounting performance is bad than when it is good (e.g., Coughlan and Schmidt, 1985; Weisbach, 1988; Murphy and Zimmermann, 1992; Parino, 1997; Jenter and Kanaan, 2015). While the relation between CEO turnover and firm performance is statistically significant, its economic significance is quite small. Most studies

(8)

find that moving from the top to bottom decile of performance increases the probability of CEO turnover in publicly traded firms by about 4% (Brickley, 2003).

While both stock price and accounting performance have predictive power in explaining CEO turnover, the age of the CEO is more important. Prior studies document that turnover is positively related to the age of the CEO – the older the CEO, the higher the probability of a turnover event (e.g., Coughlan and Schmidt, 1985; Weisbach, 1988; Jensen and Murphy, 1990). Murphy (1999) finds evidence that the probability of a CEO leaving office is nearly 30% higher when the CEO is over the age of 64 than when he is younger.

Earlier work (e.g., Brickley, 2003) also emphasizes the importance of tenure in predicting CEO turnover. Evidence suggests that tenure is negative and significantly associated with forced turnover – CEOs with longer tenure are fired less often (e.g., Peters and Wagner, 2013). Similar to the findings about the connection between CEO age and turnover, a longer tenure results in a higher probability of a turnover due to retirement reasons.

Moreover, researches have paid considerable attention to the role of boards of directors in internal corporate governance, focusing on their role in monitoring and firing badly performing CEOs. Jensen (1993) points out that when the CEO also holds the position of the chairman of the board, internal control systems fail since the board cannot effectively perform its key functions – namely evaluating and firing CEOs. This lack of independent leadership makes it more difficult for the board to replace a poorly performing CEO. Hence, the probability of a CEO turnover is less performance sensitive in firms where the CEO also holds the title of the chairman than in those where there is a clear separation. Several researches have found evidence that the probability of a turnover for chairman-CEOs is significantly lower than for their non-chairman counterparts (e.g., Goyal and Park, 2002). Considering time changes of CEO turnover, one can see that annual CEO turnover has increased over time (e.g., Kaplan and Minton, 2010). Turnover is 15.8% from 1992 to 2007, implying an average tenure as CEO of less than seven years. In the more recent period since 2000, total CEO turnover increases to 16.8%, implying an average tenure of less than six years.

Lower-level managers from the same firm usually replace CEOs. However, outside replacement has become more common in recent years (Murphy, 1999). The likelihood of an outside replacement is inversely related to prior firm performance and is most likely when the CEO was forced out. Parrino (1997) finds empirical evidence that outside CEOs are more often hired after forced turnovers.

(9)

2.1.3 Geographically segmented labor markets

Third, this thesis is related to papers about geographically segmented labor markets.

The presumption of the finance literature is that the market for CEOs is national in nature (e.g., Kedia and Rajgopal, 2009). Most models about finance related topics that do not consider the role of geography in the modeling of the CEO labor market implicitly make the same assumption. However, researchers have shown that geography plays an important role in the market for CEOs. Yonker (2015) finds that firms hire locally five times more often than expected if geography were irrelevant for the matching process. This local matching bias is widespread even among the largest U.S. firms.

The economics literature provides extensive evidence indicating that firms in large metropolitan areas pay their workers significantly more than their non-urban counterparts, a phenomenon typically known as the “urban wage premium”. This gap is especially large for highly skilled individuals, such as CEOs (e.g., Gould, 2007; Bacolod et al., 2008). According to Francis et al. (2012), there are two main explanations for these findings. The first one is that high ability individuals are attracted to urban areas with a wealth of employment opportunities, where there are more easily matched with better jobs and paid the value of their marginal product in competitive labor markets (e.g., Helsley and Strange, 1990, 1991; Glaeser et al., 1992). The second is that urban areas provide workers with positive externalities, such as business connections (e.g., Gould, 2007; Christoffersen et al., 2009; Yankow, 2006). These studies predict that especially for high ability individuals such as firms’ CEOs, who tend to be attracted to big cities and therefore more likely to benefit from their positive externalities, wages should be significantly higher in large metropolitan areas. 2.1.4 Succession planning

Fourth, this thesis is related to studies about succession planning.

CEO succession as a process has received little explicit attention in the finance literature. Most studies address the determinants of CEO turnover, the choice between an inside and an outside successor, and the practice of combining the CEO and chairman position (see e.g., Weisbach, 1988; Parrino, 1997; Agrawal et al., 2006). Vancil (1987) is one of the few researchers who focused directly on CEO succession. He suggests that one common pattern of succession is relay succession. One way a firm can ensure that the new CEO has all the skills required is to implement a formal succession process, a relay succession. In a relay succession, the heir apparent (president and/or COO) is often selected a few years before the incumbent CEO is expected to step down. Succession planning is believed to help

(10)

organizational governance structures prepare for executive turnover and train qualified candidates for appointments to future manager positions (Behn et al., 2005).

Using a sample of the 800 firms included in the Forbes 1991 annual compensation survey, Naveen (2000) finds that four years before CEO turnover, 48.7% of firms have an heir apparent. However, 68.7% have an heir apparent immediately before the turnover occurs. After controlling for other factors affecting turnover, the probability of turnover is higher for firms that have succession plans than for those who do not have succession plans. Furthermore, firms that have succession plans are more likely to replace the incumbent CEO with the appointed heir apparent. More specifically, 84% of firms that do have an heir apparent select the heir apparent as the next CEO.

2.2 HYPOTHESIS DEVELOPMENT

This thesis is the first to directly test for the effects of the firm location on CEO turnover, more explicitly departure (forced, retirement, unclassified) and replacement types (company insider, company outsider but industry insider, industry outsider). Therefore, my thesis attempts to extend prior research by connecting four fields of research – geographic remoteness, CEO turnover, geographically segmented labor markets and succession planning. Taking the outcomes of the literature review about geographic remoteness and CEO turnover into account and connecting this with the findings about geographically segmented labor markets, I argue that there is a difference between CEO turnover in urban and remote located firms – CEOs in remotely located firms have a lower probability of a forced turnover than CEOs in urban-based firms.

Hypothesis I: CEOs in remotely located firms have a lower probability of a forced turnover than CEOs in urban-based firms.

Several arguments lead to this hypothesis. First, the studies about the “urban wage premium” (e.g., Yankow, 2006; Gould, 2007) do not only predict that wages should be significantly higher in large metropolitan areas. They also predict that there is more competition among CEOs in urban areas than among CEOs in non-urban areas since high ability individuals prefer urban areas (Francis et al., 2012). As a result, there are more high ability individuals competing for jobs in urban areas. Hence, the supply of potential candidates is higher in urban areas. But also the demand for high ability individuals is higher in urban-based firms since firms usually located in urban areas (e.g., Davis and Henderson, 2004). If firms have a bigger pool of potential candidates to choose from, the risk of firing one CEO and not finding

(11)

a appropriate new candidate is much lower in urban than in remotely located firms. This implies that forced CEO turnovers in remotely located firms are less performance sensitive than in urban-based firms. If a CEO in a remote located firm performs badly, the firm might not fire him as fast as if it was located in an urban area.

However, the higher supply of CEOs is not the only argument that leads to this hypothesis. Remotely located firms might not only have problems to find new candidates, they also face higher search costs. They might be forced to e.g. engage heir search companies or hire compensation consultants that help them design attractive compensation packages that make remotely located firms competitive. In order to avoid these costs, firms may choose to settle with an average performing or even low talent CEO as the costs overweight the pros of having a better performing CEO.

If remote managerial labor markets are so much worse for firms than their urban counterparts, there might, however, be other outcomes that not necessarily result in firing a CEO. Hence, firms may engage in other activities to deal with the tightness of the local managerial labor market. While the ultimate punishment for bad performance – firing – might be less of an option for boards, they may provide greater pay-to-performance sensitivity in their pay. So if managers perform badly, they might not be fired right away but instead be punished more in their pay – relative to their urban counterparts. Moreover, some firms would never even locate in the first place in a rural location where they cannot easily replace badly performing managers. Another potential outcome might be that the internal promotion system may work differently in urban and remotely located firms. There might e.g. be more tournaments among lower rank managers.

Lastly, one should also keep in mind that different types of managers exist. There are a lot of studies about the endogenous matching process between firms and managers. These studies (e.g. Ackerberg and Botticini, 2002; Bandiera et al., 2011) show that there is strong evidence that particular types of managers are matched with particular types of firms. It might be possible that risk-seeking CEOs avoid rural firms in sleepy towns, but risk-averse CEOs and those who enjoy the quite life, seek out to go to those rural firms. This means that managers are sorting firms in advance. As risk-averse managers are never taking enormous amounts of risk by nature, they need to be less often fired than risk-seeking managers.

While these other outcomes might all be possible, it is important to keep these in mind and address these issues by, at least partially, controlling for them. Since I only have data about CEO departure and replacement types, I will not investigate the possible greater pay-to-performance sensitivity. Moreover, in the data sources I use there is no data available about

(12)

why firms locate in certain areas. Nevertheless, I will control for one measurement about tournaments among lower rank managers, via the CEO Pay Slice. Above that, I control for CEO age and CEO gender that partially control for the risk-averseness of managers.

While the first hypothesis relates CEO turnover, geographic remoteness and geographically segmented labor markets, the second hypothesis also takes succession planning into account. Though, the arguments for deriving the second hypothesis are strongly connected.

If remote located firms are aware of the smaller labor market and the short supply of high ability individuals, I expect them to replace the leaving CEO more often with a company insider than their urban counterparts. Facing a smaller labor market, I also expect them to have more succession planning than urban-based firms. Thus, I hypothesize that they plan their succession more in advance than their urban counterparts.

Hypothesis II: A company insider replaces CEOs in remotely located firms more often than in urban-based firms.

In order to have a heir apparent at hand, internal succession is less costly, easier to perform and more successful than trying to attract outsiders in remotely located areas. This hypothesis is strongly connected with the findings of Parrino (1997). He finds empirical evidence that outside CEOs are more often hired after forced turnovers. If CEOs in remotely located firms have a lower probability of a forced turnover than their urban counterparts, they might more often use insiders to replace the resigned CEO.

If the second hypothesis is true, the first hypothesis might also be insignificant. It could be that remote firms are just as likely to replace a CEO as urban firms, although they have a smaller labor market, since they are much more active in succession planning and raising internal talents.

3 DATA AND METHODOLOGY

This section starts with a description of the different databases that are used in this thesis. It also contains precise definitions of all variables and summary statistics for the complete dataset. In addition, this section also discusses the methodology that is used to test the two stated hypotheses.

(13)

3.1 DATA

3.1.1 Data sources

The data comes from four main data sources. The Eisfeldt and Kuhnen database (2013) is used to identify CEO departure and replacement types. Information about the firm’s headquarters as well as firm specific accounting data is taken from the Compustat database. S&P’s ExecuComp database is used to determine CEO characteristics, such as information on CEO age, tenure and compensation. Data on geographic information, most importantly information about all U.S. zip codes and their corresponding latitude and longitude coordinates, is downloaded from the website of the U.S. Census bureau (USCB, 2015). All variable definitions are summarized in Table I.

[Table I approximately here]

3.1.2 CEO turnover

To measure CEO turnover, I make use of the dataset from the research of Eisfeldt and Kuhnen (2013). They have compiled an accessible database including 2,113 CEO departures with all CEO turnovers of publicly traded companies in the U.S. from 1992 until 2006. Using the ExecuComp variable CEOANN, they classify a turnover event in year t if the name of the CEO in year t is different from the name of the CEO in year t+1. They classify CEO turnover as either forced, due to retirement or unclassified. Instances where the press reported that the CEO was fired or left the company due to policy differences with, or pressure from, the board or from shareholders, are classified as forced departures. Similar to Jenter and Kanaan (2015) and building on the procedure proposed by Parrino (1997), CEO departures are classified as planned retirements if they have been announced at least six month before the departure. All other events (e.g. acceptance of another position) are labeled unclassified departures. Of the 2,113 CEO departures, 29.25% are the result of a planned retirement decision, 15.52% are forced and 55.23% are cases that do not fit into any of those two categories and are therefore labeled unclassified.

The dataset of Eisfeldt and Kuhnen (2013) does not only include reasons for the departure but also the replacement types as either from inside the company, from outside the company but inside the industry, or from outside the industry. Most of the replacements, 70.75%, are company inside replacements, 8.19% are company outsiders from the same industry, and 21.06% are industry outsiders.

(14)

Without using this dataset of Eisfeldt and Kuhnen, who spend a lot of time in hand-collecting data, it would not be possible to do such in in-depth analysis. Furthermore, this dataset is a reliable source, since it went through a rigorous peer review process and is widely used also in other research (Jenter and Lewellen, 2010; Yonker, 2015).

3.1.3 Geography

Consistent with existing academic literature (Davis and Henderson, 2004: Pirinsky and Wang, 2006; John et al., 2011), I define a firm’s location as the location of its headquarter, since most strategic managerial decisions are expected to be rendered from the firm’s headquarters. I obtain the headquarters locations for the companies in the Eisfeldt and Kuhnen dataset from Compustat. Since not all headquarters locations are reported in the Compustat database, I hand-collect 135 missing zip codes by looking up the location of the firm’s headquarter on publicly available information, e.g. the company’s website or Google maps. After collecting this data, I match the zip codes to GPS data, more explicitly the latitude and longitude data form each firm’s headquarter. This information is available at the U.S. Census Bureau’s Gazetter city state files (2015).

I use several measures for geographic location. First, I consider the effect of central location similarly to Loughran and Schultz (2005). Firms are classified as urban if they are headquartered in a top ten metropolitan area according to the population statistics compiled by the 2000 Census. These include New York City, Los Angeles, Chicago, Washington-Baltimore, San Francisco, Philadelphia, Boston, Detroit, Dallas, and Houston. To classify a firm as urban, I look up all zip codes that belong e.g. to the New York metropolitan area. If the zip code of the firms’ headquarters is in this zip code database, it is classified as centrally located. Second, I look at the log of distance to the closest major metropolitan area, distance to a top ten metropolitan area. Firms located farther away firm large cities are expected to have a lower probability of a forced turnover. Third, I use a refinement of the distance measure. I define another variable, distance to a top 49 area, similarly identified on the basis of the 2000 Census. Firms located near medium-sized cities could still be similar to urban-based firms and therefore have a higher probability of a forced turnover, which is not captured by the original distance measure. Distances are computed by the vincenty command in Stata. To compute the distance to a metropolitan area, I calculate the distance from the firms’ headquarter to the center of a metropolitan area. To compute the exact distance, I use the vin output from the vincenty command rather than the loc or hav result, as this result assumes the earth to be ellipsoid whereas the other two assume the earth to be a perfect sphere, which it is not. Lastly, I use the measurement “rural”. A company is defined as rural

(15)

if its’ headquarter is 100 miles or more from the center of any of the 49 U.S. metropolitan areas of 1 million or more people according to the 2000 Census. Rural companies are e.g. located in most of Alabama, western and southern Taxes and all of Alaska and Hawaii. 3.1.4 Measurements for firm performance

In accordance with other researchers’ work (e.g. Huson et al., 2001), I measure the relative performance by filtering out shocks that affect all firms in the same industry. I use two different measures for firm performance: (1) the one-year-lagged industry-adjusted level of accounting performance and, (2) the one-year-lagged industry-adjusted stock return. The reason to use a one-year-lagged performance measure is that the turnover could have possibly occurred because of a bad performance of the CEO the year before the dismissal. Additionally, Fee et al. (2015) also argue to use lagged performance measures to estimate CEO turnover. They find that the widely documented negative relation between firm performance and CEO turnover is extremely robust and highly significant when using the one-year-lagged performance measurements. Optimally, a measure of the performance over a period of e.g. one year prior to the announcement date of a fired CEO would be preferable (see e.g. Peters and Wagner, 2013). However, this information is not available to me - neither in the Eisfeldt and Kuhnen dataset nor in the other databases I use. I define the industry-adjusted level of accounting performance as return on assets (ROA), calculated as the ratio of net income (Compustat item NI) to total assets (Compustat item at). To control for industry factors of affecting the firm’s earnings, I subtract the median of the corresponding measure for the population of the firms listed on Compustat in the same two-digit Standard Industrial Classification (SIC) code. I define industry-adjusted stock return as a market measure of firm performance. I obtain stock return data by calculating the dividend and stock-split adjusted annual return using Compustat and subtracting the median of the corresponding measure for the population of firms in the same way as already described above. To minimize the influence of outliers in the data, I winsorize ROA at the 95th_{percentile, stock return at the}

99th_percentile.

3.1.5 Controls

CEO characteristics

To capture the effect of the normal succession process in the likelihood of turnover I create a CEO age dummy. The dummy variable CEO age >=60 equals one if the departing CEO is older than 59 and zero otherwise. This information is downloaded from the ExecuComp database (ExecuComp item present age). The actual age in the year of the turnover is then

(16)

computed afterwards as the item age has limited data available. Using ExecuComp item date became CEO and date left as CEO, I compute CEO tenure in months. Data about the CEO’s gender is available at the ExecuComp item gender. I download information about the CEO’s total compensation from ExecuComp, item TDC1. I winsorize this variable at the 99th percentile.

Following Bebchuk et al. (2011), I define the CEO Pay Slice (CPS) as the percentage of the total compensation to the top five executives that goes to the CEO. Since the CEO is not in office the entire turnover year, I compute the CPS one year prior to the turnover event to avoid data mismatches. Data mismatches could e.g. occur if the CEO left the company in the middle of the year and therefore the CEO did not receive the fully annual pay. Then, the CPS would be very low since the other top four executives did receive the full payment. The measure is based on the total compensation to each executive, including salary, bonus, other annual pay, the total value of restricted stock options granted that year, long-term incentive payouts, and all other total compensation (as reported in ExecuComp item TDC 1). For some firm years more than five executives are listed in ExecuComp. In such cases, I use only the five executives with the highest compensation (similar to Bebchuck et al., 2011)1. Similarly, I exclude firms that report compensation for fewer than five executives to ensure that CPS remains comparable across firms. Yet, this lack of reporting is very uncommon.

I include an additional control for high equity ownership. Comparable to Jenter and Kanaan (2014), I define high CEO equity ownership if the CEO owns more than 5% of all outstanding shares the year prior the turnover year. This information is downloaded from the Execucomp database (item SHROWN_EXCL_OPTS_PCT). Denis et al. (1997) report that the probability of top executive turnover is negatively related to stakes held by officers and directors. More powerful CEOs should be better able to defend themselves against “unfair” dismissals in bad times, weakening the effect of performance on forced turnovers.

I identify chairman-CEOs from ExecuComp’s annual title and annual CEO flag fields. The variable CEO is chairman equals one if the CEO also serves as the chairman of the board.

Firm characteristics

Similar to Naveen (2006) and building on the procedure of Kini and Williams (2011), I identify a firm as having an heir apparent, and therefore a relay succession, if the firm has a president and/or chief operating officer (COO) as a top five highest paid executive distinct

1_{In my sample period, firms were required to disclose information concerning the amount and the type of}

compensation paid to the CEO, CFO and the three most highly compensated executive officers (SEC rules). Some firms voluntarily report the compensation for more executives than required.

(17)

from the CEO or chairman the year before turnover. This is computed by combining the Compustat items annual title and annual CEO flag as well as information about executive payment from the item TDC1. If a firm has a president and/or COO who is not the CEO in the year prior to the turnover and is one of the top five highest paid executives, I classify the firm as having an heir apparent.

As a measure of firm size I use (the log of) total assets (Compustat item at). I control for growth opportunities by including Tobin’s Q (see Table I for variable definition). I winsorize both variables at the 99th_percentile.

3.2 SUMMARY STATISTICS

Summary statistics including all variables used in this thesis are presented in Table II.

[Table II approximately here]

As displayed in Panel A, there are 2,063 CEO turnovers that are investigated. The original dataset of Eisfeldt and Kuhnen (2013) included 2,113 CEO turnovers. Thus, 50 observations are left out. This is due to two main reasons. First, not all of the firms in the Eisfeldt and Kuhnen dataset are located in the U.S.. Second, and of more importance, some of the zip codes are not included in the GPS data from the U.S. Census Bureau. As the number of observations only differs slightly from the original dataset, the distributions of the departure types in Panel A and the replacement types in Panel B are almost the same as already described in Section 3.1.2.

Panel C contains geography statistics. Of the 2,063 firm-year observations and corresponding 1,378 unique firms, 43% of all firms (593) are classified as urban. Therefore, 43% of all firms are located in a top ten metropolitan area based on the 2000 Census. These findings are in line with other researchers’ findings (e.g., Loughran and Schultz, 2005). Most publicly listed firms’ headquarters are located in large metropolitan areas. An average firm is located 175.81 miles from the center of any of the closest top ten areas. As the pool of cities located nearer gets bigger when including not only the biggest top ten but also the biggest top 49 areas, the average distance becomes a lot smaller. An average firm is located 35.99 miles from the center of any of the closest top 49 areas. The minimum of both geographic distance measures is, of course, zero, since some firms are located in the very center of the areas. Three firms in the sample are located in Hawaii. This explains the maximum of over 2,000 miles, as the closest city from Honolulu in Hawaii is San Francisco, which is 2,380.46 miles

(18)

away. As expected, only a small amount of firms, namely 6% of the sample, is classified as rural.

Panel D presents firm performance measures. The median industry-adjusted ROA is 2%. The median industry-adjusted stock return equals 5% one year preceding CEO turnover. For the forced turnover subsample the median industry-adjusted ROA is approximately 1%, the median industry-adjusted stock return is even negative, namely minus 4%.

Panel E includes CEO characteristics. The average CEO is approximately 60 years old. In my sample, the youngest CEO is 29, the oldest 86. More than half of the sample is older than 60 which explains that the dummy variable, CEO>=60, is 53%. Furthermore, 99% of all CEOs are male. Very similar to Bebchuk et al. (2011), who use a comparable sample period, I find that the average CPS equals 35% and its standard deviation equals 13%. The summary statistics of Bebchuk et al. (2011) are computed based on a panel data set of 12,011 firm-year observation that represent 2,015 different firms and 3,256 different CEOs between 1993 and 2004. In this time period they find that the average CPS was 35% and its standard deviation equals 11.4%. Comparable to Peters and Wagner (2013), I find that 69% of the CEOs do also hold the title of the chairman.

Firm characteristics are given in Panel F. In line with the findings of Naveen (2000) as well as Kini and Williams (2011), I find that 65% of all firms have an heir apparent the year before the CEO turnover.

3.3 METHODOLOGY

This subsection discusses the main methodology I employ to evaluate the impact of firm locations on CEO turnover.

Hypothesis I: CEOs in remotely located firms have a lower probability of a forced turnover than CEOs in urban-based firms.

To test the first hypothesis, I will use ordinary least square regressions. The dependent variable is equal to one if a CEO turns over and is forced to do so and zero otherwise. The main independent variables are urban, the two distance measures and rural. To capture the effect of firm performance, I include two performance measures, industry-adjusted return on assets and industry-adjusted stock returns. To test the predictions of my hypothesis, I interact the geography variables with the performance measurements. Beta1 measures whether urban firms are more likely to have a forced CEO turnover. Beta2 captures the effect of performance on CEO turnover. Beta3 is whether urban firms are more likely to have a forced

(19)

turnover when also measuring performance. Thus, beta1 should be positive and beta2 should be negative. As a result and in line with my hypothesis, I expect the interaction term to be negative.

Controls include both CEO (age, tenure, gender, ln(total compensation), high equity ownership, CPS and chairman-CEO dummy) as well as firm characteristics (ln(assets), Tobin’s Q) as described in Table I.

Forced CEO Turnoverit = β₀+ β₁*Urbani + β₂*Performanceit + β₃* Urbani*Performance_it +

β₄*Controls_it + u_it

Forced CEO Turnoverit = β₀+ β₁*Distancei + β₂*Performanceit

+ β₃*Distancei*Performance_it + β₄*Controlsit + uit

Hypothesis II: A company insider replaces CEOs in remotely located firms more often than in urban-based firms.

To test the second hypothesis, I use the same regression method as for the first hypothesis. The dependent variable, inside replacement, is equal to one if a company insider replaces a CEO and zero otherwise. The main independent variables will be the same as in the first regression equation, namely measurements about the geographic location of the firm’s headquarters. I include a dummy variable for succession planning – heir apparent. Since I hypothesize that a company insider replaces the leaving CEO more often in remotely located firms than in urban-based firms, I also include interaction terms in this regression, similar to the interaction terms used to test the first hypothesis. I interact the geography variables with the heir apparent dummy. Beta1 measures whether urban firms are more likely to have an inside replacement. Beta2 measures whether inside replacements are more likely to occur if there is a succession planning in place. Finally, beta3 is whether urban firms are more likely to have an insider replacement when they also have succession planning than rural firms when they have succession planning. Hence, I expect beta1 to be negative. Rural firms are more pro-active than urban firms. Beta2 should be positive if succession planning does what it is supposed to be doing. Finally, beta3 could be negative if rural firms indeed used their successor more often than urban firms during a turnover. But conditional on firm having a succession plan, an urban firm may just follow through with the successor as often as a rural firm does. Thus, the coefficient could also be zero. So if the coefficient is negative, it would help with the main idea – rural firms need to use the successor option more often than urban

(20)

firms as they don’t find a fitting outsider – but it is not threatening to the main idea that distance mattered if it was zero.

Control variables are the same as the ones added in the regression to test the first hypothesis. Inside Replacement_it = β₀+ β₁*Urbani + β₂*Heir apparent_it + β₃*Urbani*Heir apparent_it

+ β₄*Controls_it + u_it

Inside Replacement_it = β₀+ β₁*Distancei + β₂*Heir apparent_it

+ β₃*Distance_i *Heir apparent_it + β₄*Controls_it + u_it

To capture the influence of year and industry effects, all regressions include year- and industry-fixed effects. Moreover, all regressions are clustered at the firm level. Within a given firm, the turnover variable may well be correlated across years.

4 RESULTS

My empirical analysis starts with univariate results and continues with multivariate tests that base controls on existing work on the determinants on CEO turnover. Both subsections in this part, univariate and multivariate results, are divided into two parts. While the first part discusses results for the first hypothesis, the second presents results for the second hypothesis respectively.

4.1 UNIVARIATE RESULTS

The first set of results is based on univariate tests of means of CEO departure and replacement data for centrally and remotely located firms and is presented in Table III (departure types) and IV (replacement types). The tests illustrate the significance and magnitude of the differences in CEO turnover measures.

4.1.1 Departure types

[Table III approximately here]

In order to divide my sample into comparable sub-samples, I categorize firms based on the four geographic classifications – urban, distance top ten area, distance top 49 area and rural (for exact definitions see Section 3.1.3). First, I compare urban and non-urban firms computing a two-sample t-test of means of the different departure types – forced, retirement and unclassified. Second, I compare those firms nearer to a top ten metropolitan area and those further away. Here, I use the median of 52.43 (see Table II, Panel C) to classify a firm

(21)

as either closer (yes) or further away (no) from any of a top ten metropolitan area. I continue with the same procedure to cluster firms as either nearer to a top 49 metropolitan area (yes) or further away (no). Equally to the subset before, I use the median of 16.46 (see Table II, Panel C) to classify the two samples. Fourth, I compare rural and non-rural firms. Lastly, I check urban against rural firms. In all five cases, the sub-samples are approximately of the same sample size.

As one can see in Table III, there is no clear pattern that points in the direction of my first hypothesis – CEOs in remotely located firms do not have a lower probability of a forced turnover than CEOs in urban-based firms. I expected more significant results as those would have given a hint that there are differences in CEO departure and replacement types between firms nearer to large metropolitan areas and firms further away. Although there are some differences in the samples, most differences are not statistically significant. Only for the rural and non-rural classification as well as the comparison between urban and rural firms there is a noteworthy statistically significant difference. First, non-rural firms have a higher portion of forced turnovers – 16 % of all turnovers are classified as forced. In comparison, only 9% of turnovers are classified as forced for the rural sample. The difference is 7% and statistically significant at the 10% level. Second, when comparing urban and rural firms, the difference is almost the same. Within urban firms, 15% of all turnovers are forced. In rural firms, as stated above, only 9% are forced. This difference is 6% and, as in the previous case, statistically significant at the 10% level. In all other cases, the two-sample t-test does not result in statistically significant results.

4.1.2 Replacement types

[Table IV approximately here]

In order to tabulate Table IV, I used the same sub-sample characterization as for Table III. This is based on the simple dummy-variable classification for urban and rural firms and a classification of two samples for the two continuous variables (distance top ten area, distance top 49 area) based on the respective sample median. The last table compares urban and rural firms.

The results from the univariate tests for the replacement types are comparable to the ones of the departure types. Most of the time, there are differences between the particular samples but the differences are too small to be statistically significant. This time, the most distinct differences are for firms nearer to a top ten metropolitan area and those further away. If a

(22)

firm’s headquarter to the closest top ten metropolitan area is below the sample median, 68% of the CEOs are replaced by company insiders. For those firms whose headquarters are above the sample median, there are 6% more company insiders who replace the CEO (74%). The difference is significant at the 1% level. This finding points in the direction of my second hypothesis – a company insider replaces CEOs in remotely located firms more often than in urban-based firms. Additionally, there is another statistically significant result in the category distance to top ten area. As shown in Table IV, those firms located close to a top ten metropolitan area do replace a CEO more often with an industry outsider than firms located remotely. The difference is 4% and is statistically significant at the 10% level.

As stated above, univariate tests illustrate that there is not an obvious pattern that points in the direction of my two hypotheses. Univariate tests, however, omit a number of determinants that could be correlated with both location and CEO turnover. Therefore, multivariate analysis is used in the following to formally evaluate the magnitude of the location effect after accounting for other covariates.

4.2 MULTIVARIATE RESULTS

In this section, the results from the regressions proposed in the previous section are displayed and discussed. This section is divided into two main parts. The first part consists of the effect of the firm location on CEO departure types and therefore tests the first hypothesis. Since I mainly investigate the effect of the firm location on forced turnover, these regression results are more extensively discussed as the effect on retirement and unclassified turnovers. The second part discusses regression results on the effect of the firm location on replacement types after a CEO turnover event occurred. Hence, the second hypothesis is tested in this part. 4.2.1 Departure types

4.2.1.1 Forced Turnover

Table V, Panel A and B, presents the results of the ordinary least square regressions of a dummy variable indicating forced CEO turnover on the four geographic variables, firm performance as well as controls. As Panel A uses ROA, Panel B uses stock return as a proxy for firm performance. For the two sets of regressions, I report the results for the full sample period of 1992 until 2006. In both panels and in all columns I use CEO (CEO age >=60, CEO tenure, CEO is male, ln(total compensation), high equity ownership, CEO Pay Slice, CEO is chairman) and firm characteristics (ln(total assets), Tobin’s Q) as controls.

(23)

Concerning the control variables, I find that, unsurprisingly and in line with other researchers’ findings, CEOs delivering poor performance are more likely to be fired or, putting it differently, CEOs delivering good performance are less likely to be fired. For example, as one can see in Panel B column (3), the coefficient of the industry-adjusted stock return is -0.1565 and significant at the 1% level. In all regressions, the coefficients of the firm performance measurements are negative and most of the time significant. In accordance with other researchers’ empirical work (e.g. Peters and Wagner, 2013), I find that older CEOs and those with longer tenure are fired less often. Tenure and especially age are an important determinant of turnover, as one can see from the significance level of the coefficients. The higher the CEO Pay Slice, the percentage of the total compensation to the top five executives that goes to the CEO, the higher the risk of a forced turnover. Turnover risk for CEOs who also hold the title of the chairman is significantly lower than for their non-chairman counterparts. Lastly, bigger firms fire more often than their smaller counterparts. The variables that capture the gender of the CEO and the Tobin’s Q variable do not have a statistically significant effect on the probability of forced turnover, in both Panels A and B. Of main interest are the geographic variables (urban, log(top ten area), log(top 49 area) and rural) and the interaction terms of those variables with the two firm performance proxies – ROA and stock return. The multivariate results, however, do not considerably differ from the univariate results discussed in the previous section. Almost all coefficients turn out to be insignificant. An interpretation is therefore redundant. All geographic variables are insignificant. Only when using the distance measure log(top 49 area), so the log of one plus the distance to the top 49 metropolitan area is the distance in miles to the closest top 49 area, the interaction terms are significant, both at the 10% level in Panel A (coefficient equals 0.1699) and B (coefficient equals 0.0271). Yet, the interaction terms point in the opposite direction as expected. They imply that firms further away from a top 49 metropolitan area are more likely to fire CEOs. Given the fact that the log(top 49 area) coefficient is insignificant, the geographic impact is probably not as severe after all.

In sum, I do not find evidence for my first hypothesis. The economic impact of geography on forced turnover is insignificant for the four measures of geographic location.

4.2.1.2 Other turnovers

Retirement turnover

Table VI, Panel A and B, shows ordinary least square regressions of a dummy variable indicating retirement CEO turnover on four geographic variables, firm performance as well

(24)

as controls. Identical to Table V, Panel A uses ROA and Panel B uses stock return as a proxy for firm performance. This table also uses the same controls as already used in Table V.

[Table VI approximately here]

The results concerning the control variables are quite obvious. The older the CEO and the longer in office, the higher is the probability of a retirement turnover. As for the regression using stock return as firm performance measurement, I find that Tobins’s Q is inversely related to retirement turnover. All other controls – CEO gender, CEO Pay Slice, CEO is chairman, ln(total assets) – are insignificant in the stated tables.

Comparable to the results in Table V, the four geographic variables are insignificant in most cases. Only the geographic variable log(top 49 area) is significant in both panels. The magnitude of the coefficient (0.0235) implies that the reason for a turnover in remote firms is more likely due to retirement reasons. The coefficient is significant at the 5% confidence interval.

Unclassified turnover

Results of ordinary least square regression of a dummy variable indicating unclassified CEO turnover on four geographic variables, firm performance and controls are presented in Table VII. The approach is identical to the previously stated regressions.

[Table VII approximately here]

As expected, unclassified turnovers are the sort of turnovers hardest to predict. The R-squared is the lowest for regressions when the dependent variable is the dummy variable unclassified turnover. This result is not surprising, as unclassified turnovers are defined as turnovers that are neither forced or due to retirement. This implies that more than one sort of reasons fall into this category. According to Eisfeldt and Kuhnen (2013) events like unexpected retirements, the acceptance of another position and vaguely described health problems are labeled unclassified departures. Given these reasons, I refrain from interpreting these results in more detail.

4.2.2 Replacement types

Table VII illustrates the effects of geography and succession planning on the three different CEO replacement types – company insider, company outsider but industry insider and industry outsider. In Panel A the dependent variable is company insider, an indicator equal to

(25)

one if the CEO successor is a company insider. The dependent variable in Panel B is company insider, a dummy variable equal to one if the CEO successor is from outside the company but inside the industry. Finally, the dependent variable in Panel C is industry outsider, an indicator equal to one if the CEO replacement type is from outside the industry. Similar to the regressions beforehand, the first regression results in Panel A are of major interest since the second hypothesis is tested here – a company insider replaces CEOs in remotely located firms more often than in urban-based firms. All regressions in this table use the identical controls as the regressions about CEO departure types.

4.2.2.1 Company insider replacements

Table VII, Panel A shows ordinary least square regressions of the dummy variable company insider on four geographic variables, an heir apparent dummy as well as controls. As defined in Table I, the heir apparent dummy is an indicator equals to one if the firm where a turnover arise has a president and/or COO as a top five highest paid executive prior to the year of turnover. The main variables of interest are the geographic variables, the heir apparent dummy and the interaction terms between those.

[Table VIII approximately here]

The marginal effect of the heir apparent dummy on inside replacement is positive and significant at the 1% level in all four columns. This is in line with my expectations and prior findings of other researchers (e.g. Naveen, 2006). A firm that has a CEO successor at hand chooses the company insider to replace the CEO rather than taking someone from outside the company.

Analyzing the coefficients of the geographic variables is, again, difficult. Only one of them, log(top ten area), is significant at the 5% level. All other geographic variables are insignificant. The positive sign of the significant coefficient (0.0242) implies that remote firms do choose a company insider more often than urban-based firms. As all other coefficients are, however, insignificant, the effect is probably not that big. The interaction terms of the geographic variables and the heir apparent dummy are, with the exception of one, also insignificant. With a p-value of about 0.06 for the rural category, and an increase of 21.18% in the probability to hire a company insider after turnover, this category might capture the remoteness category of a firm best. Given that the rural measure itself does not pick up any result, the impact is probably sizeable for all firms and/or firms are dealing with it somehow that it does not show up in equilibrium.

(26)

An interesting result regarding the controls is that the older the leaving CEO, the higher the probability that the new CEO is from inside the company. All coefficients in the four columns are positive and significant at the 1% level. The coefficients of the variable total compensation in t-1 implies that the more the leaving CEO earned, the higher the chance that a company insider replaces him. If the CEO is also the chairman of the board, it is also more likely that a company insider replaces the CEO. Also, bigger firms tend to choose a relay successor. One reason could be that bigger firms have, in comparison to their smaller counterparts, a bigger pool of potential candidates to pick from.

In summary, I do not find evidence to accept my second hypothesis. 4.2.2.2 Other replacement types

Company outsider and industry insider

The dependent variable in Panel B is company outsider and industry insider. The regression approach is the same as in Panel A.

This time, the marginal effect of the heir apparent dummy is negatively related to the dependent variable. Only in column (4) the coefficient is significant, at the 1% level. This means that firms that do have a CEO successor do not choose a company outsider from the same industry to replace the CEO. This outcome is expected, as Panel A already showed that firms who have a successor choose the successor rather than someone from outside the company.

Concerning the geographic variables and the interaction terms, there is no obvious pattern. Only one of those variables is significant. The coefficient of urban is positive with a value of 0.0457 and significant at the 5% level. The positive sign implies that urban firms choose a company outsider and industry insider to replace the CEO.

Industry outsider

The last panel, Panel C, shows ordinary least square regressions of the industry outsider dummy on the four geographic variables, the heir apparent dummy as well as controls.

In all four columns, the heir apparent dummy is negative and significant at the 5% level, expect for the second column, where the coefficient is significant at the 1% level. Only one of the geographic variables is significant at the 10% level, log(top ten area), in column 2. The negative sign of the coefficient implies that firms farther away from the top ten metropolitan areas are less likely to choose an industry outsider as successor. However, the interaction

(27)

term of the geographic variable with the heir apparent dummy does not turn out to be significant.

5 ROBUSTNESS CHECKS

In this section, I perform several tests to determine the robustness of the effect of a firm’s location on CEO departure and replacement types. I conduct a number robustness checks including alternative variable definitions and a different sample period. Alternative variable definitions are defined in Table IX.

[Table IX approximately here]

Alternative variable definitions

As shown in previous tables, the main findings hold with different measures of geographic location – the geographic location of a firm’s headquarter does not have a significant effect on CEO turnover. So far, I used four different proxies of geographic location. To check the robustness of this outcome, I use an alternative definition of remote location. So far, I classified a firm as rural if it the firm is headquartered 100 miles or more from the center of any of the 49 U.S. metropolitan areas of 1 million or more people according to the 2000 Census. Now, I redefine this category. A firm is classified as remotely located if the firm is headquartered 75 miles or more from any of the 49 U.S. metropolitan areas. This new variable is named remote.

In previous regressions, I controlled for CEO age by including a dummy variable equal to one if the incumbent CEO is older than 59 – CEO age >=60. Now, I use an alternative definition – CEO retirement age, an indicator equal to one if the CEO is between 63 and 66 years old to account for likely retirements.

As can be seen from Table X, the main results continue to hold after I include the two new variables. The effect of the geographic variable remote on forced turnover as well as company inside replacement is insignificant. The same pattern can be seen for the included interaction terms. The CEO retirement dummy variable is, in accordance with previously stated outcomes, negative and significant at the 1% level when the dependent variable is forced turnover.

(28)

Different sample period

My findings are based on the 1992 until 2006 sample period due to availability of data on CEO turnover. The sample period is changed in Table XI, which reports the results of regressions for the period from 1992 until 1999. Panel A and B present regressions where the dependent variable is forced turnover. Thus, they present results for testing the first hypothesis. Panel C includes regression results for the second hypothesis. The effects of location on CEO departure and replacement types continue to hold. Using a different sample period therefore does not affect the stated results. In unreported tests, the main results remain consistent when I use different sub-periods.

[Table XI approximately here]

6 DISCUSSION

The presented results leave room for discussion. This section highlights and discusses some further thoughts and limitations of the presented results.

Both univariate and multivariate results indicate that geography does not play an important role for CEO departure and replacement types. There is no noteworthy difference in CEO turnover types in centrally and remotely located firms. Although contradictory to the predictions and studies of the “urban wage premium”, the simplest explanation why I do not find significant results would be that geographic location does have no effect at all on CEO turnover. Urban and remotely located firms fire equally likely and succession planning is the same in both urban and non-urban firms. Meanwhile, there are a lot of other reasons that could result in not finding an effect.

First, and of major importance, are possible measurements issues of the firm’s geographic location. There is a possibility that the remoteness measures might not be capturing what I expected them to capture. In order to check the robustness of the presented results, different measurements of geographic remoteness are worth incorporating. One could e.g. proxy the remoteness of a firm by measuring the area population and classifying firms by clustering population into certain sub-clusters. Firms would then be classified as remote when the area population falls under a predetermined threshold. Using this measurement allows that the area population is considered. Besides that, there are many other proxies of geographic remoteness. Similar to Knyazeva et al. (2013), one could measure the local pool of prospective CEOs by the density of companies headquartered within a predetermined mile radius of a sample firm’s headquarter. Similar to the aforementioned approach, a firm is then

(29)

classified as remote if the density of companies nearby is especially low. If the density is low, the pool of potential candidates is a lot smaller than for firms where the density is high. One refinement of this measure includes measuring nearby firms in the same industry by classifying firms in the same SIC-code. Another approximation several other authors (e.g. John et al., 2011) used who studied the effect of geographic location is the distance to the closest major airport. Firms located farther away from airport hubs are then expected to be less accessible to prospective candidates. While these three suggested alternative measurements are conceivable, there are many other measurements worth looking at.

Second, there might also be measurement issues in the CEO turnover dataset from Eisfeldt and Kuhnen (2013) that mainly influence the forced turnover classification. As stated in Section 3.1.2, the authors classified forced-out departures as instances where the press reported that the CEO was fired or left the company due to policy differences with, or pressure from, the board or from shareholders. Although this approach is straightforward, it leaves room for misclassifications. One might argue that the press does not get full information that circulates within a firm. CEOs are almost never openly fired from their positions and, even with a clear classification scheme, the distinction between voluntarily or forced departure can be ambiguous. Hence, there might be misclassifications in the dataset. Some departures that were classified as departures due to retirement were actually forced because the firm, the board of directors and/or the CEO decided not to inform the press. Third, if the labor market is so much worse for remotely located firms, firms may engage in other activities to deal with the tightness of the local managerial labor market that not necessarily result in firing the CEO. Boards of remotely located firms may provide greater performance sensitivity in their pay in comparison to their urban counterparts. If a CEO performs badly, he is then not punished by being fired but by getting less money. Moreover, some firms would never even locate in a rural location where they cannot easily replace badly performing managers. Above these reasons, firms may also use other measures to overcome the lack of CEO talent in the labor market.

Fourth, and another reason why I did not find an statistically significant effect can be attributed to possible omitted variables that could be correlated with both departure and/or replacement types as well as location. Although I controlled for a range of variables, including CEO and firm characteristics that limit the concern that other omitted variables are biasing the results, I cannot completely exclude the possibility that other factors partially explain the results. Firm risk may e.g. be correlated with CEO turnover and location. One might argue that firm risk is positively correlated with CEO turnover and also influences the

Does distance matter? : the effect of the firm location on CEO turnover

MASTER THESIS