Msc Business Economics, Finance track Master Thesis
The effect of a CEO succession plan on innovation
Engbers, Max July 2016
Statement of Originality
This document is written by Student Max Engbers who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgements
First of all, I would like to thank my thesis supervisor Prof. Dr. Tolga Caskurlu, Faculty of Economics and Business, University of Amsterdam, for guiding me well through my master thesis. Furthermore, I thank Timothy Quigley of the University of Georgia for sharing the sudden CEO death data with me.
Abstract
In this thesis, I examine the effect of succession planning on innovation by comparing patent-based innovation measures for firms with a succession plan, to firms that do not have a succession plan. I also examine whether the effect of succession planning differs for firms in innovation-intensive industries as compared to other firms. Innovation is measured by patent count and patent citations. To determine whether a firm has a succession plan, I use a criterion of three days as the maximum delay in replacing a suddenly deceased CEO.
Without differentiating for industry type, I find no significant difference between firms with and without a succession plan in terms of innovation. When making a distinction between innovation-intensive and non-innovation-intensive industries, I find a positive, significant effect for the number of patents obtained for companies with a succession plan in an innovation-intensive industry as opposed to a non-innovation industry.
These results suggest that if companies wish to use succession planning to increase their innovation strength, they will need to adjust their succession planning method.
Table of contents
1. Introduction……… 5
2. Literature review……… 7
2.1. CEO turnover……… 7
2.2. Non-direct CEO-related factors………. 7
2.3. Direct CEO-related factors……… 8
2.4. Sudden CEO death……… 10
2.5. CEO succession and innovation……….... 10
2.6. Industry………. 12
3. Methodology………. 14
4. Data collection and summary statistics………. 18
4.1. Sudden CEO death and succession planning data……… 18
4.2. Patent data………. 19
4.3. Industry data……….. 20
4.4. Control variables data………... 21
4.5. Summary statistics……… 21 5. Results………... 26 6. Robustness checks……….... 33 7. Discussion………. 41 References………. 44 Appendix………... 49
1. Introduction
There is extensive literature on innovation, and CEO succession planning has also been studied with relation to net income (Behn et al., 2006) and shareholders’ wealth (Larcker and Tayan, 2012). Even so, the effect of succession planning on innovation has not yet been studied, not to mention that this has been studied in relation to the level of innovation within the industry. The questions that will be answered in this thesis will fill this gap in the literature. The main research question is: Does a succession plan improve future innovation? The second question that will be answered is: Does this differ across industries?
The answer to the first question can change the way that firms understand CEO succession planning. Larcker and Miles (2010) show that companies do not place much importance on succession planning.1 Succession planning means having an immediate successor who can make new future plans directly, and can maintain or even augment firm performance (Behn et al., 2006). Effective planning also means that the level of innovation set by predecessors would be maintained or even augmented if a successor is quickly appointed.
The answer to the second question could imply that, for some firms, succession planning is more important for innovation than it is for other firms. For example, for an industry where product innovations follow each other in rapid succession, it may be more important to quickly name a successor than it would be for industries where product innovation is not the main driver of growth.
The answers to these thesis questions help company boards to decide the level to which they intend to plan CEO succession. If a company’s growth depends on innovation and this thesis shows that succession planning is beneficial for innovation, companies might wish to consider planning for succession. The answer to the second question could indicate that the first question’s solution differs across industries. For instance, companies in less innovation driven industries will be less forthcoming about engaging in succession planning.
The relation with literature in the field of Economics is clear: economists have studied CEO succession and subsequent innovation extensively (Bereskin and Hsu, 2014), and succession planning has been linked to firm performance (Bruce and Picard, 2006). To my knowledge, however, succession planning has not yet been studied with relation to innovation. Therefore,
1 A study by Larcker and Miles (2010) in cooperation with Heidrick and Struggles shows that more than half of the companies (private and public) cannot name an immediate successor.
this thesis contributes to the knowledge with regard to succession planning and firm performance in order for boards to better assess the importance of CEO succession planning for their companies.
This thesis uses a difference-in-differences methodology. Differences in innovation can be measured by examining a group of firms that have succession plans and a group of firms that do not have succession plans. To ensure that a lack of innovation is not already the driver of CEO change, only CEO changes through force (specifically, sudden death) are used to measure the effect of a succession planning on innovation. To define my parameters, I measure innovation as patent count and patent citations. To determine the differences in innovation between industries, I distinguish between and compare innovation-intensive industries and industries that are less innovation-intensive.
Without differentiating for different types industries, I do not find a significant difference between firms with a succession plan and firms without one in terms of innovation. When I make a distinction between different types of industries, I find a positive significant effect for the number of patents obtained for companies with a succession plan in an innovation-intensive industry compared to other firms.
In Section Two, the relevant literature is highlighted and explained. In Sections Three and Four respectively, the methodology and data collection are given in more detail. Section Four ends with a summary of the statistics gathered. The results are presented and interpreted in Section Five, and then tested for robustness in Section Six. Finally, this thesis is concluded and discussed in Section Seven.
2. Literature review 2.1 CEO turnover
CEO turnover affects many aspects of companies, examples of which can be found in the literature. Murphy and Zimmerman (1993) assess general turnover-related changes to CEO turnover, and find that CEO turnover particularly influences the key variables of firms because of poor performance by the preceding CEO. Carpeto et al. (2010) connect CEO turnover to M&A deal activity, and find that deal making at the beginning of a new CEO’s tenure has a positive effect on the company’s short/medium-term performance. Bereskin and Hsu (2012) link CEO turnover and other CEO-related factors to innovation. They find that CEO turnover is positively correlated to the quality and quantity of innovation in the first three- to five-year period after CEO turnover. For Bereskin and Hsu, the quantity and quality of innovation is measured by patent count, number of citations, patents per research, and per development dollar and citation per patent. In their paper, they also link a few CEO-related attributes to innovation and find that CEO-overconfidence, option compensation, and information asymmetries are positively correlated to innovation.
The relevance of this thesis is due to the fact that innovation is very important for a firm’s value. Companies that want to stay ahead of competition need to innovate, which is often measured (among other things) in patent count. Lerner (1994) shows this in his research by measuring the breadth of patents with relation to firm value. He finds that a standard deviation increase in average patent scope leads to a 21 percent increase in a firm’s value. Hirshleifer et al. (2013) emphasize the effect of innovation on stock returns by developing a metric to measure innovation efficiency. They find that innovation efficiency is a good predictor for future stock returns. This indicates the importance of innovation for maximizing firm value.
Many factors have an effect on the level of innovation. This thesis attempts to underscore the effects of succession planning on innovation in case of the CEO’s sudden death.
2.2 Non-direct CEO-related factors
The variety of possible influences on innovation can be divided into two segments: direct CEO-related factors and non-direct CEO-CEO-related factors. Cohen et al. (2013) attribute innovation to a non-direct CEO-related factor. They argue that looking at a firm’s track record is a good way of forecasting a firm’s future innovation. Initially, they show that a firm’s past R&D investment can
have divergent outcomes, but that these outcomes are predictable. If firms were to make better use of their track record, they would be better able to predict future outcomes for patents, patent citation and product innovation.
Aghion et al. (2012) find another non-direct CEO-related effect. They study the effect of institutional ownership on innovation, measured by cite-weighted patents, and find a favorable effect for innovation. Acharya et al. (2014) found a positive effect regarding wrongful discharge laws on innovation. Another possible determinant of innovation is the pressure of an external takeover. Sapra et al. (2014) examine different kinds of takeover pressure levels created by anti-takeover laws. They find that innovation is either optimal with unhindered corporate control, or by anti-takeover laws that are strict enough to diminish the number of takeovers effectively. Bernstein (2015) captures the effect of IPOs on innovation, comparing companies that went public to companies that planned to go public but eventually withdrew their filing and remained as a private company. Bernstein shows that companies that stay private tend to be more innovative than companies that choose to go public. The latter companies experience a decline in internal innovation, a drain on their resources for skilled inventors, and a decline in productivity for any remaining inventors.
This shows that there are numerous possible influences on innovation that are not directly CEO-related. However, since the CEO determines the strategy and policy of the company and influences each move that the company makes, it can be argued that every effect stated above is actually CEO-related.
2.3 Direct CEO-related factors
Bereskin and Hsu (2014) study direct CEO-related drivers on innovation, and show that innovation is CEO-related. Moreover, it is argued that CEO-succession causes firms to be more innovative. Bereskin and Hsu conduct a study somewhat similar to the study in this thesis. They examine the influence of a new CEO on firm performance from the perspective of innovation, and find strong empirical results to suggest that new CEOs have a positive influence on innovation. They measure innovation as patent counts, patent citations and patents per R&D dollar. Furthermore, they find that new CEOs found internally in the company outperform new CEOs found externally, in terms of innovation. The crucial difference between this thesis and Bereskin and Hsu’s methodology is that they do not consider firm succession planning. This is,
however, important because firms need to be innovative regardless of what CEO leads the company. Therefore, a succession plan of a company should include that the successor should actively and positively contribute to innovation.
Tian and Wang (2014) also assign the level of innovation to CEO-related factors. They suggest that tolerance for failure of the CEO by company boards actually boosts corporate innovation. They base their research on a venture capital (VC)-backed IPO firm sample, in which they compare firms with failure-tolerant boards to firms with boards that are not tolerant to failure. Manso’s (2011) study also confirms these findings, arguing that the compensation scheme of a CEO determines his/her level of innovativeness. Compensation schemes that are designed to allow for failure in the short term and value success in the long term are the best compensation schemes for stimulating innovation.
Hirshleifer et al.’s (2012) paper suggests that overconfident CEOs are better innovators because they spend more on innovation and are better at exploiting growth opportunities than others. However, the authors stress this is only true for CEOs who manage companies in already innovative industries and that overconfident CEOs that manage companies in less-innovative industries are not greater innovators than their peers. Hirshleifer et al. (2012) and Gallasso et al. (2011) postulate that overconfident CEOs are more innovative than others, measuring the level of overconfidence through a series of stock-option exercises, and innovation in citation-weighted patent counts. The time period for these studies, however, is between 1980 and 1994, making it a little outdated. The authors made use of year-fixed effects, which is also used in this thesis’ empirical approach. Another relevant claim made by the authors is that the effect of CEO-overconfidence on innovation is stronger in more competitive industries.
Pérez-González (2006) examines whether successors who are related to their predecessors perform better or worse than successors who are unrelated to their predecessors. In this instance, a “relation” is someone who is related to a CEO, a founder or large shareholder by blood or by marriage. Pérez-González finds that successors who are related to past CEOs will underperform compared to successors who are unrelated. He also finds that unrelated successors spend more on R&D than successors who are related. Furthermore, Pérez-González points out that the difference in R&D spending alone does not mean that related successors are less innovation driven than unrelated successors, but it does indicate that they might be less innovative than unrelated successors.
In the following section, studies that examine direct CEO-related factors are reviewed. Innovation is subject to many direct and non-direct CEO-related factors. The following section will make clear that succession planning is not clearly attributed to one of these categories, but rather to both.
2.4 Sudden CEO death
Over the past few decades, sudden CEO death has become more popular as a subject for research. The reason for this is that it allows researchers to study the effects of an exogenous shock on a company. Sudden death is exogenous because a company cannot reliably plan ahead for such an issue. If the company fires the CEO, it can be due to all sorts of causes, among others, the CEO not being innovative enough. Jenter and Kanaan (2015) show that CEOs are sometimes fired for factors they cannot control, like bad market performance. Similar literature in the field of Economics claim that firms filter out these exogenous shocks, but Jenter and Kanaan refute this by stating that a decline in industry and market performance increases the probability of a CEO turnover. Bereskin and Hsu (2014), as described earlier, also use sudden CEO death in order to link CEO turnover to innovation, and they find a positive effect between CEO turnover and innovation.
This part of the literature review elaborates on the exogenous shock initiated by the sudden passing of a CEO. As seen above, many studies consider sudden death to be an exogenous shock. These studies elaborate on how this shock enables measurement of effects to a company and is also why this type of shock has been used in this thesis.
2.5 CEO succession and innovation
Although there is limited literature on the subject of CEO succession with relation to innovation, some papers do go into detail. Musteen et al. (2010) examine the effect of a CEO’s attitude toward innovative change in competitive strategies. They find that the more liberal CEOs are more likely to augment innovation and that age also has an influence on innovation level. Older CEOs tend to be more risk-averse and thus will innovate less. Although Musteen et al.’s paper emphasizes the relationship between CEO succession and innovation it does not take CEO succession planning into account.
Larcker and Tayan (2012) consider succession planning, explaining what a good succession plan looks like and stressing the importance of having a viable succession plan. They do not, however, link it to innovation but rather to a firm’s share price. According to the authors, a viable succession plan contains possible candidates to whom the company could turn in case the current CEO is fired or can no longer carry out the tasks needed of him/her. Potential candidates can either be internal or external candidates. In the most favorable situation, the new CEO will take up office whenever the company needs him/her to, but this is not always the case. In some instances, the company will appoint an emergency or interim-CEO prior to hiring the desired CEO. A company that does not have a good succession plan will not have a set of potential candidates in place and will only begin the succession process once the CEO position has become vacant. Due of this, the process of finding the desired CEO can last for months, which is bad for overall firm performance. To reduce uncertainty, a company may disclose detailed information about its succession plan to shareholders. However, Larcker and Tayan point out that a survey conducted by the Institutional Shareholder Service shows shareholders do not value disclosures of information. Only 27 percent of the shareholders (on average) supported a proposal that requires companies to develop and disclose succession plans. Larcker and Tayan give three reasons as to why a good succession plan matters:
1) Some companies are able to appoint a new CEO within a short period of time, where others take weeks to months to do so. This has a detrimental effect on a firm’s share price.
2) If a good CEO of a company dies unexpectedly, the company’s stock price will falls. What, however, should the board do in case the stock price goes up after an unexpected death? This is something to specify in a company’s succession plan.
3) Most CEO successors are internal candidates. Replacing the deceased CEO with an external candidate takes longer, which has a negative effect on the firm’s share price. Although these reasons relate to a firm’s share price, one can imagine that a delay in the appointment of a new CEO also has a detrimental effect on innovation. The longer there is no CEO, the longer it will take for the firm to make Research & Development (R&D) or new product plans. Since inventing a new product is a long and complex process, the sooner a new CEO is appointed the better it is likely to be for innovation.
Behn et al. (2006), measuring firm performance as future net income, found evidence that a delay in naming a successor for a deceased CEO will have a negative impact on firm
performance. Additionally, they found that firms that name a successor immediately after the passing away of a CEO outperform firms that delay the appointment of a successor for up to two years. Thus, having a CEO succession plan is very important for future firm performance in terms of net income.
There is not a great deal of other literature on succession planning. Shen and Cannella (2003) argue that previous literature suggests succession planning should affect shareholders’ wealth. Their results show that initiating the succession planning process does not cause shareholders to react. At the end of the planning process, reaction depends on whether the process ends in a promotion, or the exiting of the apparent successor. When a promotion takes place, the shareholders react positively. When an exit takes place, the shareholders react negatively. Shen and Cannella state that, according to the results, shareholders prefer relay succession over non-relay inside succession.
Bruce and Picard (2006) state that a successful company can have a well planned out succession but must make sure the “soft issues” in a strategic plan are well thought through. Some examples of these “soft issues” are: long-term goals, communication with stakeholders, stakeholder vision, the role of each stakeholder in the business after CEO succession, and the selection and training of successors to prepare them for their new role. The new CEO only faces these issues if the succession has not been well planned, although according to the authors they are important issues to consider.
The above explains the importance of succession planning and hints at its potential effect on innovation.
2.6 Industry
Although most of the literature above does not differentiate between industries, one can imagine a distinction between them. To illustrate this, a bank may not need an innovative CEO; for a bank’s performance, it may not matter how many patents the bank obtained during a CEO’s tenure. On the contrary, for pharmaceutical and tech companies it is essential to obtain and/or acquire patents in order to develop new products that are successful in the market.
Carlin and Mayer (2003) confirm this notion, suggesting that there is a difference in the level of innovation across industries. They measure innovation by R&D spending, and show that
some industries are more innovation driven, meaning differences in innovation between CEOs may differ across industries.
Different industries may attract characteristically different CEOs. Goel and Thakor (2008) suggest that underinvestment inefficiency caused by a risk-averse CEO will be more likely in some industries than others. This may indicate that different kinds of CEOs can be observed across industries, which affects succession planning across industries. In industries that rely on innovation for their growth, it may be more important for companies to make sure they select an innovative successor and carefully consider their succession planning.
This literature review indicates the existence of a gap, namely in the effect of CEO succession planning on innovation. While both succession planning and innovation have been studied before, the two combined into one study is unique. For this reason, the first hypothesis of this thesis is:
H1: A succession plan contributes positively to innovation following a sudden CEO death.
The above hypothesis has not been studied across industries. Hence, to fill a second gap in the literature the second hypothesis postulates:
H2: The positive effect of a succession plan on innovation following a sudden CEO death is greater in innovation-intensive industries.
It is important to gain knowledge of succession planning in general. Although the U.S. government stimulates companies to plan for succession, it has not yet been widely studied.2 For some firms, it is important to maintain a level of innovation as their business depends on it. Thus, companies benefit from knowledge of the effects of succession planning on innovation. However, this effect may be felt differently across industries.
2 The Social Security Administration (SSA) expresses the importance of succession planning and how to implement it. For source see references.
3. Methodology
Both hypotheses address the effect of a single event: the sudden death of a CEO. To empirically test the hypotheses, I use a difference-in-differences (DID) method with year-firm specific panel data.
Sudden death is unexpected and unpredictable; an exogenous shock. It is not subject to or influenced by changes from within the model described in this thesis. That is why this exogenous shock gives an opportunity to test the effects of succession planning on innovation. The event of a CEO death has been used to test other CEO-related effects than the effect tested in this thesis. For instance, Bennedsen et al. (2007) use CEO death to study family succession and firm performance, and it has also been used to relate CEO turnover to innovation (Bereskin and Hsu, 2014). The difference between these papers and my thesis is that I relate the effect of succession planning and CEO turnover to innovation. The frequent use of this kind of exogenous shock in the literature indicates its suitability for testing CEO-related effects.
Firm- and year-specific data is used to compare innovation from subsequent years to innovation from previous years. In this manner, I can measure innovation across firms and years.
The DID method studies the differential effect of a certain treatment between a treatment and a control group. It computes the effect of a treatment on the dependent variable by comparing the average change over time for the treatment group, to the average change over time for the control group. The dependent variable in this experiment is innovation, which is measured by patent count and patent citations. The treatment group in this case has a succession plan. So the treatment group is the group that has experienced a sudden passing away of their CEO and has a succession plan for his/her replacement. The control group is the group that has also experienced a sudden death of their CEO but does not have a succession plan for his/her replacement.
The advantage of the DID-method is the mitigation of extraneous effects. By using a control group a common trend in the number of patents and patent citations and other changes over time are taken into accounted. Therefore, the DID-method accurately measures the effect of the treatment (i.e., having a succession plan).
Due to the exogenous shock of sudden CEO death and the advantages of characteristics of the DID-method explained above, the results of this study can be interpreted as causal effects
and not just as a correlation. Given the variables of interest, the hypotheses posed are estimated by using the following equations. For H1:
𝐼𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛 = 𝛽!+ 𝛽!𝑆𝑢𝑐𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑃𝑙𝑎𝑛 ∗ 𝐴𝑓𝑡𝑒𝑟𝐶𝐸𝑂𝐷𝑒𝑎𝑡ℎ + 𝛽!𝐴𝑓𝑡𝑒𝑟𝐶𝐸𝑂𝐷𝑒𝑎𝑡ℎ + 𝛽!𝑅&𝐷 + 𝛽!𝐿𝑜𝑔𝐴𝑠𝑠𝑒𝑡𝑠 + 𝛽!𝐶𝐸𝑂𝐴𝑔𝑒 + 𝛽!𝐶𝐸𝑂𝑇𝑒𝑛𝑢𝑟𝑒 + 𝜀
And for H2:
𝐼𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛 = 𝛽!+ 𝛽!𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑦 ∗ 𝑆𝑢𝑐𝑐𝑒𝑠𝑠𝑖𝑜𝑛𝑃𝑙𝑎𝑛 ∗ 𝐴𝑓𝑡𝑒𝑟𝐶𝐸𝑂𝐷𝑒𝑎𝑡ℎ + 𝛽!𝐴𝑓𝑡𝑒𝑟𝐶𝐸𝑂𝐷𝑒𝑎𝑡ℎ + 𝛽!𝑅&𝐷 + 𝛽!𝐿𝑜𝑔𝐴𝑠𝑠𝑒𝑡𝑠 + 𝛽!𝐶𝐸𝑂𝐴𝑔𝑒 + 𝛽!𝐶𝐸𝑂𝑇𝑒𝑛𝑢𝑟𝑒 + 𝜀
In which: β are the regression coefficients, and SuccessionPlan is the variable indicating whether a company has a succession plan. The variable equals 0 for companies without a succession plan and equals 1 for companies with a succession plan; AfterCEODeath is the variable that indicates whether a company is at a time before or after a CEO death. The variable is 0 for companies in the period before the CEO’s death and 1 for companies after the CEO’s death. R&D is the ratio of R&D expenses and the total assets of a company in the given year. LogAssets is the logarithm of total assets (in million dollars) of a company in the given year. CEOAge is the age of the company’s CEO in the given year. CEOTenure is the number of years the CEO has been in office in the given year. Industry is the variable to indicate whether a company is innovation-intense. The variable is 0 for companies that are not innovation-intense and 1 for companies that are innovation-intense. ε is the error term; SuccessionPlan*AfterCEODeath and Succession Plan*AfterCEODeath*Industry are both interaction variables.
I consider that a company has a succession plan in cases where CEO succession was announced within three days after sudden CEO death. The reason for this is that accessible databases do not contain information about succession planning before 2007, although all sudden CEO deaths were observed before 2007. The justification for my assumption about whether a company has a succession plan can be found in the literature: Behn et al. (2006) argue that, in the event of a sudden executive death, a company must name a successor as soon as possible to assure investors, customers and employees that the company can manage the event. Cannella and Shen (2001) show that many large U.S. firms determine a CEO’s heir well before the actual date
of succession, in order to prepare for a smooth transition when CEO-succession finally occurs. It seems that quickly announcing a successor is important. Given the papers of Behn et al. (2006) and Cannella and Shen (2001) in combination with the fact that succession planning occurs ahead of the succession date, the maximum time lag between sudden CEO death and the announcement of the CEO successor for a company to be qualified as having a succession plan is set at three days. This maximum time lag is set to three days, because I expect this is the maximum time it may take companies with a succession plan to announce their new CEO. Sudden CEO deaths may occur during weekends, and for practical reasons it may take companies until Monday morning before they make the announcement. Naming an interim CEO is often the case when a company is still searching for the right successor. Therefore, I do not consider companies that announce an interim CEO after a predecessor’s sudden death as having a succession plan.
In the equation for H1, SuccessionPlan interacts with AfterCEODeath. This is the variable of interest for H1. For H2, SuccessionPlan interacted with AfterCEODeath and with the dummy variable, Industry. This is the variable of interest for H2. To decide which industries are innovation-intensive, I consulted a study by USPTO (2012) regarding patent data from 2004 until 2008 for each industry category according to the North American Industry Classification System (NAICS).3 The number of filed (and later granted) patents over a period of five years for each NAICS-category was divided by the average payroll employment of that industry category. Therefore, industry employment functions as a reliable measure of industry size and can be used to normalize patent count per industry. An industry category showing a value above the mean of all categories shows that it is an innovation-intensive industry.
The control variables used in both equations are: R&D, LogAssets, CEOAge and CEOTenure. I use R&D (the ratio of R&D expenses and total assets) because a plain R&D expense number does not mean anything if the size of the company has not been accounted. The ratio shows how much a company is spending on R&D relative to firm size. This ratio is commonly used in the literature (Bernstein, 2015 and Hirshleifer et al., 2012). To control firm size, I apply the logarithm of total assets. This approach is also frequently used in literature in the field of Economics, particularly those reporting on innovation (Bernstein, 2015 and Aghion et
3 See USPTO (2012), for a more detailed explanation of the identification of innovation-intensive industries.
al., 2014). Bernstein also uses firm size as a control variable in a regression with scaled patents, and scaled patent citations as dependent variables. Bernstein uses the logarithm of total assets as a control for firm size. The logarithm is used to scale the variable.
Based on existing literature on the subject, the expectation for the variable CEOTenure is that it will have a positive influence on innovation. Although I expect that the sign of this control variable will be positive, Musteen et al. (2010) show that tenure correlates negatively with a manager's attitude toward change and his approach to innovation. On the contrary, Hirshleifer et al. (2012) argue that CEOs with longer tenures are generally more confident and less afraid of making mistakes, which is why they take more risks in terms of innovation.
CEOAge is another control variable for which Musteen et al. (2010) indicate the effect of age on the variable of interest in this thesis. They find that an increase in CEO age decreases innovation. Kim and Lu (2011) measure the effect of CEO age on R&D expenditures and find that CEO age has a negative effect on innovation. One could argue that if R&D expenses drop, then the quantity (patent count) and quality (patent citations) is also likely to drop, which will not lead to further product inventions. Pérez-Gónzalez (2006) does not agree with the previous two papers, and shows that in his sample, age is positively correlated with innovation (measured in R&D expenses). He shows that older successors are actually more likely to be innovative than are younger successors. His sample contained both related and unrelated successors. Related successors were on average 8.15 years younger than the unrelated successors.
The argument here, then, is that CEO tenure has a positive effect on innovation and CEO age has a negative effect of innovation. This seems contradictory information, as older CEOs are more likely to have longer tenures. What might also be the case, however, is that CEOs who are appointed later in their lives are less innovative than CEOs who were appointed at an earlier age.
4. Data collection and descriptive statistics
The data for the variables that I have used in this thesis were collected from different data sources. Data of each of these variables will be explained here:
4.1 Sudden CEO death data and succession planning data
First, I collected data on deceased CEOs. This data consisted of the non-standard database of Timothy Quigley, Graig Crossland and Robert Campbell (2016), who made their database of deceased CEOs available for use in this thesis. This database consists of 239 observations of CEOs who deceased in the period between 1950 and 2007, and comprised information such as company name, CEO name, cause of death, date of death, and the age of the CEO on the day of his/her death. I adopted the same causes of death to define what is meant by “sudden” death, to include heart attack, stroke, traffic accident, aneurism, murder and suicide.
For data on patents, I used a USPTO-based data-base (see Section 4.2). This database contains patent information for the period between 1970 and 2010. As patent information both five years before and after each observation is required, I had to omit any observations made before 1975 and after 2005. This resulted in a total of 122 observations.
I complemented this database with hand-collected data on the date when the deceased CEO took office, and I combined this with the date of death to calculate the tenure of the deceased CEO. In cases where the tenure of the deceased CEO was less than five years, I collected data on the preceding CEO(s) in order to complete at least a five-year time window before the date of death. I extended the database with hand-collected data on the name of the successor, the date the successor took office, the nature of the assignment (interim or fixed), the age of the successor when taking office and the date the successor resigned. The tenure of the successor was calculated from the date when the successor took office and resigned. In cases where the tenure of the successor was less than five years, I collected additional data on his/her subsequent successors in order to at least complete a five-year time window after the year of CEO’s death.
For observations before 1992, no standard databases were available that contained the necessary information. Therefore, I hand-collected this data from news sources, typically online archives of newspapers such as The New York Times, The Wall Street Journal, The Chicago
Bloomberg, and Business wire. For observations in 1992 and later, I first collected data via the
ExecuComp database. Quigley, Crossland and Campbell’s database contained gvkey and permno codes so that I could easily match the data from ExecuComp to their database. For any data that was still missing, I manually collected as much as possible from the above-mentioned news sources.
To determine whether a company had a succession plan, the number of days between the CEO death and the date the successor was announced was determined. If this delay was more than three days, I concluded that the company did not have a succession plan. I applied this approach because there is no succession planning data available (in accessible databases) for observations from before 2007, and all of the analyzed sudden CEO deaths in this study are from before 2007. In addition, in cases where an interim-CEO was announced within the three-day period, I concluded that the company did not have a succession plan. I consider an interim-CEO an ad hoc solution not being based on a succession plan.
In case an observation in the database could not be completed with all required data, I excluded the observation from my study. At this point, the sample was reduced to a total of 81 companies.
4.2 Patent data
To measure innovation, I use Kogan, Papanikolaou, Seru and Stoffman’s patent dataset. This dataset contains USPTO-data on the patent number, the patent filing date, the patent issue date, the permno code – which is also in the dataset containing sudden CEO deaths – and the number of times the patent is cited between the years 1926 and 2010. I dropped any data before 1975 because firm level patent data are incomplete before this date. Patent-based measures are widely reported in the literature as instruments for measuring innovation (Hall, Jaffe and Trajtenberg, 2001, and Bereskin and Hsu, 2014). The USPTO checks every patent application on the basis novelty and non-obviousness, and demands the applicant to make his/her inventions public within a reasonable amount of time once the application has been approved. Since the mid-80s, U.S. firms have become more active in patenting ideas for new products early on and in defending their intellectual property. This development has taken place because of the establishment of the Court of Appeals for the Federal Circuit, as well as several other court rulings in important patent lawsuits (Hall and Ziedonis, 2001, and Hall, 2005).
Another way to use patent data to measure innovation could be through R&D expense. Companies, however, only rarely disclose information about their R&D budget and the projects on which they are spending this money. This causes uncertainty in the ability of R&D expense serving as a proper measure of innovation. Furthermore, patent count by itself is not a proper measure for the innovation strength of a company (Griliches, 1990). Therefore, I use both patent counts and the number of citations as a measure for innovation. The number of citations per patent is an indication of the quality of the patent, where quality is defined as the potential of the patent to protect a new product.
The simplest alternative would be to use basic patent counts and patent citations, but total patent counts and total citations change over the years and across technology classes. This change can occur due to the importance of a technology class at a point in time, or to a change in the patent system. This jeopardizes how informative these basic measures are, which is why I use scaled patent counts and citations. To calculate these more advanced measures, I follow Hall et al. (2001). ScaledCitations is the sum of the scaled citation counts that a company has filed (and is later granted), and that are scaled by the average number of citations of patents in the same technology class that is filed for (and later granted) in the same year. Accordingly, ScaledPatents is constructed. Patent count is scaled by the average number of patents filed for (and later granted) by companies in the same year and in the same technology class.
4.3 Industry data
To find a reliable equation for the second hypothesis, I must first distinguish innovation-intensive industries from non innovation-innovation-intensive industries. The USPTO (2012) conducted a study regarding patent data between 2004 and 2008 to identify innovation-intensive industries. The USPTO calculates the ratio of the number of patents over a period of five years for a NAICS industry category distinguished by average payroll employment (see also Section 3). Thus, the industry employment functions as a measure of industry size and normalizes patent count for each industry. Industries showing a value above the industry mean are indicated as innovation-intensive industries. I used the CRSP/Compustat merged database to obtain the NAICS category codes of companies in the dataset.
4.4 Control variables data
I retrieved data for the control variables from a number of databases. The annual R&D-expenses and total assets of the period from five years prior (t-5) to 5 years after (t+5) the year of the CEO death was collected from Compustat – Capital IQ. As I have already explained in the methodology, these control variables are widely used in the existing literature in the field of Economics. Subsequently, I matched the control variables with the companies in the dataset using gvkey and year. Companies for which total assets were not reported for the complete time window of t-5 until t+5 were omitted from the sample. Any R&D expenses that reported no value were set to zero.
For observations from 1992 onward, the ExecuComp database was used to extract data about the tenure and age of the CEO. Tenure was computed by subtracting the year the CEO took office from the year that the CEO passed away. The age of the CEO is a variable given in the ExecuComp database. For observations before 1992, data on CEO age and tenure were collected from various news sources (see also Section 4.1). Typically, articles announcing a CEO taking office or resigning from his/her position also gave the age of the CEO. At this point, the sample consisted of 46 companies.
4.5 Summary statistics
In Table 1 below, I summarize the statistics of the treatment and the control group five years before sudden death of 46 companies’ CEOs. These 46 companies are evenly divided between the treatment and control group. The variables ScaledCitations, ScaledPatentCount, R&D-ratio and LogAssets are all winsorized at a 5 percent level on the upper side. Industry, CEOAge and CEOTenure are not winsorized as they were all collected by hand and the descriptive statistics do not show numbers that I would consider as odd values.
What is noteworthy is that both groups within the sample produced below average quality patents in the period before the CEO’s death (0.15 and 0.16 for firms with and without succession plans, respectively). A value of one for ScaledPatents means a firm will produce an average number of patents. Both groups in my sample produce a below-average number of patents (0.79 and 0.71 for firms with and without succession plans, respectively). These ScaledCitations and ScaledPatentCount values are small when compared to numbers reported by Bernstein (2015) who also uses ScaledCitations and ScaledPatents to measure patent activity.
I observe a significant difference at the 5% level for the independent variable of Industry. For the treatment group, this variable is 0.11 higher than it is for the control group, indicating that the treatment group contains more innovative companies than the control group. Furthermore, I observe a significant difference at the 10% level for the control variable of CEOAge, with the CEOs of the treatment group being 2.16 years older than the CEOs of the control group. None of the other variables differ significantly.
In Table 2, I summarize the statistics of the treatment and the control group analogously
to Table 1, and for the five-year period after sudden CEO death. I observe no significant mean difference for patent-related variables (ScaledCitations and ScaledPatents) for either group. For the period after CEO death, the variable of Industry shows a significant difference of 0.13 at the 5% level, with companies in the treatment group being more innovative than companies in the control group. This is in line with my observation for the five-year period before CEO death. The difference in CEO age is also significant at the 5% level. Although for both groups the successors are normally younger than the deceased CEOs, the successors of the control group are on average 2.15 years younger than the successors of the treatment group.
Table 1
Summary statistics
This table presents the summary statistics for the variables used in this study over a five-year period before the CEOs’ deaths. The sample consists of all companies that suffered a sudden death of their CEO in the period between 1975 and 2007. Company and CEO data is extracted from databases (Compustat - Capital IQ, ExecuComp) or manually collected from news sources. Patent data is extracted from the USPTO-database. The treatment group (companies with a succession plan) consists of 23 companies, each contributing five years of data (five years before the death of the CEO). The control group (companies without a succession plan) also consists of 23 companies. ScaledCitations is the relative number of citations for one company in one year. ScaledPatents is the relative number of patents for one company in one year. The variables ScaledCitations, ScaledPatents, R&D-ratio and LogAssets were all winsorized at a 5 percent level on the upper side. Industry, CEOAge and CEOTenure were not winsorized as these were all manually collected, and the descriptive statistics do not show any numbers that could be considered as odd values. In the last column of the table, the difference in mean between the treatment and control groups is given for each variable. *, **, and *** indicate the statistically significant differences with means of 10%, 5% and 1% respectively.
Observations 5 years before CEO death with succession plan
N=115 (23 companies)
Observations 5 years before CEO death without succession plan
N=115 (23 companies)
Mean Median SD Min Max Mean Median SD Min Max Diff. ScaledCitations 0.15 0 0.27 0 0.85 0.16 0 0.29 0 0.85 -0.01 ScaledPatents 0.79 0 1.72 0 6.86 0.71 0 1.65 0 6.86 0.08 Industry 0.35 0 0.48 0 1 0.24 0 0.43 0 1 0.13 ** R&D ratio 0.03 0.01 0.04 0 0.14 0.02 0.00 0.04 0 0.14 0.03 Log Assets 5.32 5.40 2.26 0 9.02 5.14 5.02 1.81 0 8.49 0.18 CEOAge 60.88 59 11.2 4 39 86 58.72 58 11.50 34 83 2.16 * CEOTenure 12.96 10 11.8 8 1 52 13.21 12 10.14 0 39 -0.25
Table 2
Summary statistics
This table summarizes the statistics for the variables used in this study in the five-year period after the CEO’s death. The sample consists of all companies that suffered a sudden death of their CEO between 1975 and 2007. Company and CEO data is either extracted from databases (Compustat - Capital IQ, ExecuComp) or manually collected from news sources. Patent data is extracted from the USPTO-database. The treatment group (companies with a succession plan) consists of 23 companies, each contributing to five years of data (five years before the death of the CEO). The control group (companies without a succession plan) also consists of 23 companies. ScaledCitations is the relative number of citations for one company in one year. ScaledPatents is the relative number of patents for one company in one year. The variables ScaledCitations, ScaledPatents, R&D-ratio and LogAssets were all winsorized at a 5 percent level on the upper side. Industry, CEO age and CEO tenure were not winsorized as these were all manually collected and the descriptive statistics do not show any numbers that could be considered as odd values. In the last column of the table, the difference in means between the treatment and control group is given for each variable. *, **, and *** indicate statistically significant differences with means of 10%, 5%, and 1%, respectively. Differences in means may not be correct because of rounding up to two decimals.
Observations 5 years after CEO death with succession plan
N=125 (25 companies)
Observations 5 years after CEO death without succession plan
N=105 (21 companies)
Mean Median SD Min Max Mean Median SD Min Max Diff. ScaledCitations 0.13 0 0.26 0 0.85 0.13 0 0.26 0 0.85 0.01 ScaledPatents 0.67 0 1.63 0 6.86 0.72 0 1.95 0 6.86 -0.05 Industry 0.35 0 0.48 0 1 0.22 0 0.42 0 1 0.13** R&D ratio 0.02 0 0.04 0 0.14 0.02 0.01 0.03 0 0.14 0.00 Log Assets 5.63 5.82 2.35 0 9.02 5.46 5.48 1.81 0 9.02 0.18 CEOAge 55.37 55 9.32 28 74 53.2 53 8.28 36 72 2.17** CEOTenure 3.03 3 1.76 1 10 2.66 2 1.40 1 5 0.37**
Furthermore, I observe a significant difference at the 5% level in CEO tenure of both groups. Tenure is 0.37 years less for the successors of the control group, indicating that the successors in this group went out of office earlier than those in the treatment group. One possible explanation is that more interim-CEOs are appointed in the control group. Moreover, it may additionally be possible that without succession planning a successor is forced to resign earlier, the company goes out of business more frequently, or the company is acquired more frequently. None of the other variables differed significantly.
5. Results
Tables 3 and 4 present the results of the regressions. In Table 3, I examine the effect of succession planning on innovation; in Table 4, I examine that effect between firms in innovation-intensive industries and firms in other industries.
In Table 3, I show the results of all regressions. For regressions (1) to (3), ScaledPatents are regressed on the interaction variable SuccessionPlan*AfterCEODeath, AfterCEODeath and all control variables, including firm and year fixed effects. In regressions (4) to (6), ScaledCitations are regressed on these variables. The regressions are made for specific time windows: Regressions (1) & (4), (2) & (5) and (3) & (6) compare two years before and after the year of death – three years before and after and five years before and after the sudden CEO death. In all of the time windows, the year of death is left out because no fair judgment can be made to whom patents and patent citations obtained in that year can be attributed. The variable SuccesionPlan is not added to the regressions because of collinearity issues. SuccessionPlan does not change within the firm over the years. Therefore, when firm fixed effects are applied to the regressions, they are collinear with SuccessionPlan. The same is true for the variable of Industry; therefore, Industry is omitted as well.
Regression (1) reflects the time window of two years before and after the sudden passing away of a CEO; the coefficient of SuccessionPlan*AfterCEODeath shows that firms having a succession plan do not generate a significantly greater or lesser number of scaled patents in years subsequent to the CEO’s death compared to companies that do not have a succession plan in place.
Regressions (2) and (3) reflect the time windows of three and five years before and after a CEO’s death, respectively. I find no significant effect in these regressions, nor do I discover a trend between them. The coefficient for this variable is positive for all three of the time windows, but it fluctuates. Therefore, I conclude that succession planning does not induce a significant effect on the quantity of patents obtained.
Table 3
Succession planning and innovation
This table shows the effect of succession planning on innovation in the event of the sudden death of a company’s CEO. The interaction variable is the variable that interacts SuccessionPlan with AfterCEODeath. This interaction variable is equal to one if a company is in the treatment group and in the period after the sudden death of its CEO. The assumption is that companies that appoint a successor within three days after the death of their deceased CEO will have a succession plan. Innovation is measured by the variables ScaledPatents and ScaledCitations. These dependent variables are patent counts and patent citations per patent, and are normalized by the average number of patents or citations per patent, obtained by a firm in the same technology class. For all other variables see Table 16 in the Appendix. The estimated time windows are t-2 to t+2 for regressions (1) and (4), t-3 to t+3 for regressions (2) and (5) and t-5 to t+5 for regressions (3) and (6), with t0 being the year of the CEO’s death. For the time windows estimated in these regressions, the year of CEO death is left out as no proper assessment can be made to which CEO, patents and related citations of that year can be attributed. Constants were included in the regressions but are not reported. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.
Dependent variable = ScaledPatents Dependent variable = ScaledCitations
(1) (2) (3) (4) (5) (6) SuccessionPlan*AfterCEODeath 0.045 (0.338) 0.014 (0.192) 0.096 (0.153) -0.024 (0.059) -0.016 (0.043) -0.003 (0.037) AfterCEODeath 0.463 (0.611) -0.040 (0.357) -0.327 (0.241) -0.034 (0.071) -0.007 (0.059) -0.008 (0.052) R&D-ratio -6.772 (4.944) -2.124 (4.842) -1.531 (3.017) -0.921 (0.925) -0.550 (0.661) -0.062 (0.469) Log Assets 0.145 (0.203) -0.030 (0.094) 0.072 (0.061) 0.043 (0.045) -0.001 (0.014) 0.014 (0.011) CEOTenure -0.017 (0.012) -0.016* (0.009) -0.011 (0.007) -0.002 (0.002) -0.001 (0.001) -0.001 (0.001) CEOAge -0.004 (0.012) -0.005 (0.009) 0.006 (0.010) -0.002 (0.002) 0.002 (0.001) 0.001 (0.001) Observations 184 276 460 184 276 460 Adjusted R2 0.315 0.240 0.148 0.230 0.108 0.113
Firm FE Yes Yes Yes Yes Yes Yes
Regressions (4) to (6) are the same as regressions (1) to (3), except for the dependent variable. For regressions (4), (5) and (6), the dependent variable is ScaledCitations. Observing the interaction variable SuccessionPlan*AfterCEODeath in these regressions, I find no significant effect of the variable of interest on ScaledCitations in any of the time windows. The coefficient is negative for all regressions, but since they are insignificant I conclude that succession planning does not have a significant effect on the quality of patents obtained in the period after sudden CEO death.
Given these insignificant effects of succession planning on the variables that I use to measure innovation, I conclude that companies do not have to engage in the current methods of succession planning for purposes of innovation.
I do not find a significant effect of the variable AfterCEODeath on the scaled number and citations of patents. However, my results suggest that sudden CEO death may be negatively correlated with the scaled number of patents, especially considering the filing of a patent takes time and may cause a delay effect.
My results are not in line with Bereskin and Hsu’s (2012) results, for they find that CEO turnover implies a significant increase in the quantity and quality of innovation. However, the authors do measure innovation with the number of patents, number of citations, patents per R&D dollar and citations per patent for a three- and five-year period after a CEO’s death. On the contrary, I measure innovation by the number of scaled patents and scaled citations.
At this point, it is hard for me to conclude which method is more suitable for measuring innovation. I argue that for company performance it is important to be innovative and stay ahead of one’s competitors. To determine whether a company is more innovative than its competitors, it is better to use scaled numbers over absolute numbers.
The coefficients of the control variables are insignificant, except for CEOTenure in regression (2). As CEOTenure is insignificant in all other regressions, I cannot determine the size and direction of the control variables CEO tenure with certainty. The coefficient of the R&D ratio in the regressions of ScaledPatents or ScaledCitations is negative for all time windows. This means that, for an increase in R&D-ratio, the quantity and quality of patents will decrease.
Except for the middle time window, the coefficients of LogAssets in the regressions of ScaledPatents and ScaledCitations are positive and insignificant. The coefficient for LogAssets in regressions (2) and (4) is negative. Given the insignificance of the coefficients for LogAssets,
I cannot draw any conclusions here. The observation that the sign of the coefficients are mostly positive indicates that larger companies will produce more, and qualitatively better, patents.
The coefficients of CEOTenure are all small, negative, and except for one, all insignificant. The coefficient of CEOTenure in regression (2) is significant at the 5% level, and comes to -0.016. This implies a decrease of 1.94 percent (=-0.016/0.823, where 0.823 is the average number of scaled patents in the three years preceding sudden CEO death). This could indicate that a CEO is a little less “productive” at a certain point in its tenure, but it could also be an isolated incidence.
The coefficients of CEOAge in the regressions are very small, and the direction fluctuates. As they are also insignificant, I conclude that CEO age has no effect on innovation. To summarize, for hypothesis one, A succession plan contributes positively to innovation
following a sudden CEO death, the results show no statistically significant evidence that a
succession plan contributes positively to innovation in a period of one to five years following a sudden CEO death. Innovation was measured by a scaled patent count and scaled patent citations. Neither do the results show a statistically significant negative effect of succession planning on innovation. This indicates that for my sample succession planning has no effect on company innovation in the one- to five-year period after a CEO’s sudden death. These findings do not take the possible effects of industry type into account, which I will study when testing my second hypothesis.
Table 4
Industry, succession planning and innovation
This table shows the effect of succession planning on innovation in innovation-intensive industries in the event of a sudden death of the company’s CEO, compared to less innovation-intensive industries. The interaction variable is the variable that interacts Industry with SuccessionPlan and AfterCEODeath. This interaction variable is set to 1 if a company is in the treatment group, in an innovation-intense industry and in the period after the sudden death of its CEO. The assumption is that companies that appointed a successor within three days after the death of their deceased CEO had a succession plan. Innovation is measured by the variables of ScaledPatents and ScaledCitations. These dependent variables are patent counts and patent citations per patent, and are normalized by the average number of patents or citations per patent, which are obtained by a firm in the same technology class. For all other variables see Table 16 in the Appendix. The time windows estimated are t-2 to t+2 for regressions (1) and (4), t-3 to t+3 for regressions (2) and (5) and t-5 to t+5 for regressions (3) and (6), with t0 being the year of a CEO’s death. For the time windows estimated in these regressions, the year of the CEO’s death is left out as no proper assessment can be made to which CEO, the patents and the related citations of that year should be attributed. Constants were included in the regressions but are not reported. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.
Dependent variable = ScaledPatents Dependent variable = ScaledCitations
(1) (2) (3) (4) (5) (6) Industry*SuccessionPlan *AfterCEODeath 0.362 (0.227) 0.446 (0.268) 0.318* (0.160) 0.071 (0.071) -0.011 (0.052) 0.042 (0.061) AfterCEODeath -0.432 (0.518) -0.029 (0.318) -0.333 (0.230) -0.032 (0.063) -0.001 (0.062) 0.000 (0.047) R&D-ratio -7.860 (5.118) -3.542 (4.848) -2.316 (3.186) -1.117 (0.934) -0.537 (0.014) -0.191 (0.476) Log Assets (0.204)0.165 (0.093)-0.017 (0.061)0.076 (0.046)0.048 (0.014)0.001 (0.011)0.015 CEOTenure -0.016 (0.012) -0.016* (0.007) -0.011 (0.007) -0.002 (0.002) -0.001 (0.001) -0.001 (0.001) CEOAge (0.012)-0.004 (0.010)-0.006 (0.010)0.006 (0.002)0.002 (0.001)0.002 (0.001)0.001 Observations 184 276 460 184 276 460 Adjusted R2 0.323 0.253 0.156 0.237 0.108 0.116
Firm FE Yes Yes Yes Yes Yes Yes
In Table 4, the regressions (1) to (3) ScaledPatents is regressed on the interaction variable Industry*SuccessionPlan*AfterCEODeath, AfterCEODeath and all of the control variables that include firm and year fixed effects. In regressions (4) to (6) ScaledCitations is regressed on these variables. The regressions are made for specific time windows. Regressions (1) & (4), (2) & (5) and (3) & (6) respectively compare the two years before and after, the three years before and after, and the five years before and after the sudden death of the companies’ CEOs in the dataset. For all the time windows, the year of death is left out, because no fair judgment can be made to whom the patents and patent citations that are obtained in that year should be attributed. The variable SuccesionPlan is not added to the regressions due to collinearity issues. SuccessionPlan does not change in the firm over the years. Therefore, when firm fixed effects are applied to regressions, they are collinear with SuccessionPlan. As the same is true for the variable of Industry, I also omitted this variable.
I find that the coefficients of the interaction variable in the regressions for ScaledPatents are relatively large and significant at a 10% level for the largest time window. In that time window, the regression coefficient is equal to 0.318, implying that the number of scaled patents is 42.29 percent (=0.318/0.752, where 0.752 is the average number of scaled patents over five years preceding sudden CEO death), which is higher for companies with a succession plan and in an innovation-intensive industry than it is for other firms. These results imply that succession planning is only beneficial for the quantity of scaled patents in the largest time window for firms in an innovation-intensive industry.
For ScaledCitations, I find that the coefficients are relatively small, insignificant and with a changing direction. These results imply that succession planning does not contribute positively to the quality of the patents in any of the time windows.
The coefficients of AfterCEODeath is negative, or zero, for all regressions. For ScaledPatents, the value fluctuates with the different time windows. Therefore, I conclude that although this variable has no significant effect, within my dataset the successors seem less innovative than their predecessors. I observed more or less the same response in the regressions given in Table 3.
I will not discuss the control variables because they are elaborated on in Table 3 and the effects observed are insignificant except for CEOTenure in regression (2). But that finding for CEOTenure is also in Table 3.
Regarding hypothesis two, The positive effect of a succession plan on innovation
following a sudden CEO death is greater for innovation-intensive industries, my results provide
statistically significant evidence that the presence of a succession plan contributes positively to scaled patents for innovation-intense companies after sudden CEO death. It should be noted, however, that the effect is only significant at a 10 percent level. Consequently, this conclusion should be treated with caution. In terms of the quality of patents, I do not find that having a succession plan positively contributes to firms in innovation-intensive industries when compared to other firms. Therefore, hypothesis two cannot be rejected because having a succession plan is beneficial for firms in innovation-intensive industries in terms of the scaled number of patents produced.
6. Robustness checks
In this section, I will describe my tests and their results in order to assess the robustness and reliability of my results.
A potential threat for the exogeneity of my results is that the assignment of firms to the treatment group (having a succession plan) is not random. Based on a criterion, I decided whether to assign a company to the treatment group (having a succession plan) or to the control group (not having a succession plan). This criterion was that the number of days it takes a company to replace its deceased CEO should be three or less for a company to be considered having a succession plan.
As this assignment is not random, the treatment variable is potentially correlated to other effects not measured in the model, called endogeneity. To address this issue, I perform a Regression Discontinuity Design (RDD). RDD is a method that is gaining popularity, because of its ability to examine a sample with a possible selection bias. The RDD method closely compares observations that lay around the criterion value. The intuition is that companies around the criterion value are considered to be very similar; it assumes that a company replacing its CEO in five days may not be that different from a company replacing its CEO in three days. By performing the RDD the Average Treatment Effect (ATE) can be calculated, where ATE is a measure to compare means between observations assigned to the treatment group and observations that were not assigned to the treatment group. I am using a bandwidth of three days around the criterion value, based on the specifics of my sample.
Table 5 shows the outcome of the performed RDD. The assignment variable in the RDD regressions is NumberOfDaysUntilSuccession and indicates the number of days it takes a company to replace the deceased CEO. In all regressions, the criterion value for the NumberOfDaysUntilSuccession assignment variable is equal to 3 days. The bandwidth applied is plus and minus 3 days. In regressions (1) and (2) the dependent variable is ScaledPatents and in regressions (3) and (4) the dependent variable is ScaledCitations. The ATE compares observations below and above the criterion value within the bandwidth set. Covariates included are AfterCEODeath, R&D ratio, LogAssets, CEOTenure and CEOAge in all regressions, and the variable Industry is added as a covariate to regressions (2) and (4). Firm fixed effects were also included for all RDDs. Year fixed effects are not included, because an RDD cannot be conducted for time-series (or panel) data.
Table 5
Regression Discontinuity Designs
This table shows the results for the RDD, with the assignment variable is NumberOfDaysUntilSuccessor (indicating the number of days it takes a company to replace the deceased CEO), the criterion value is set at 3 (companies that replace the deceased CEO within 3 days belong to the treatment group), the range estimated is the criterion value plus and minus 3 days, the covariates included are AfterCEODeath, Industry (only in regressions (2) and (4)), R&D-ratio, LogAssets, CEOTenure and CEOAge. Firm fixed effects were also included. ATE is the Average Treatment Effect. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.
Dependent variable = ScaledPatents Dependent variable = ScaledCitations
(1) (2) (3) (4)
ATE (0.691)0.047 (0.729)-0.119 (0.063)0.048 (0.095)0.076
Observations 297 297 297 297
Firm FE Yes Yes Yes Yes
Table 5 shows no significant effects for the RDD, implying that the ATE does not differ significantly between firms with and without succession plans. These findings are in line with the main results discussed in Section 5. This shows that my results most likely do not suffer from endogeneity problems.
In order for my findings to be robust, the assumption that I made when constructing the variables AfterCEODeath and SuccessionPlan, must be tested. The assumption for AfterCEODeath is that a new CEO can and change patent filing from the beginning of his/her tenure. However, it may be that research in developing new products is planned at a substantial amount of time ahead of the actual patent filing date. This suggestion is also supported by literature, which argue that R&D-projects can endure for several years (Veryzer, 1998). If that is the case, it may take a while before patent filing that should be assigned to the new CEO is observable. This may have an impact on this thesis’ results. Therefore, I will test this by regression analysis. Table 6 presents the results of this analysis.
