The effects of managerial entrenchment on exploring innovation

(1)

The Effects Of Managerial Entrenchment

On Exploring Innovation

L.E. Dedding

Master’s Thesis to obtain the degree in

Business Economics: Finance Track

University of Amsterdam

Faculty of Economics and Business Amsterdam Business School

Name: Lars Dedding Date: July, 2016

(2)

2

Statement of originality

This document is written by Lars Dedding 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)

3

Abstract

This research tries to establish the causal relationship between managerial entrenchment and explorative innovation. Explorative innovation is regarded as innovation in new fields of technology, where firms do not exploit previous technologies by using these to build new technologies (Balsmeier et al. 2016). Several dependent variables to proxy explorative innovation are used, under which the number of backward citations, the percentage of self-citations, the number of claims and an

originality index which calculates the number of patent categories that make up the total number of a patent’s citations as proposed by Hall et al. (2001). The implementation of state antitakeover legislation between 1980 and 1999 is chosen as proxy for managerial entrenchment. Although the regressions provided some significant results, indicating that managerial entrenchment led to explorative innovation, significant results pointing in the opposite direction were also found. Robustness checks could not aid this ambiguity, leading to a conclusion that there is not sufficient statistical evidence to state a causal relationship between managerial entrenchment and explorative innovation.

(4)

4

1. Introduction

Innovation has always played a key role in economic growth (Solow, 1955), yet there remains

ambiguity regarding the drivers of innovation. Theoretical and empirical results contradict each other in whether and how the level of managerial entrenchment affects innovation (Atanassov, 2013). The agency theory suggests that an entrenched management, which faces a reduced threat of hostile takeover or punishment by the shareholders, is more likely to shirk (Jensen & Meckling, 1976). This would cause the management to reduce the amount of attention to innovative projects, since they are aware that they do not face punishment for doing so. Other directions of theory suggest however, that the threat of hostile takeover reduces innovation. The fear of hostile takeover focusses the attention of management on safe but short-sighted investments instead of investing in long-term riskier innovative projects (Chakrabarty & Sheikh, 2010).

Several empirical researches have studied the effects of the threat of hostile takeovers on innovation. Generally, they made use of quantitative measures of innovation such as R&D-investments (Pugh et al. 1992) and the number of patents (Atanassov, 2013). However, such measures simply measure the quantitative level of innovation, without measuring how truly innovative they are. Balsmeier et al. (2016) introduced measures that were aimed at measuring precisely that. The distinction was made between two forms of innovation: exploration and exploitation. Exploitation was meant with patents that build upon existing technologies and continued with the same form of technology that a firm had previously applied patents for.

Exploration was considered truly innovative, since this regarded patents in new technological classes or classes that the applying firm had limited experience in. Understanding whether managerial entrenchment has an effect on truly innovative patents will contribute to the literature, since it will be able to make a distinction between managements overinvesting on not truly innovative patents and managements that are focused on achieving important explorative innovations.

This research makes use of the measures proposed by Balsmeier et al. (2016) in order to examine the causal relationship between managerial entrenchment and explorative innovation. Such measures are the number of backward citations a patent has, where a smaller number is related to more innovative research since it builds upon less existing technologies. To determine whether firms apply for patents that build upon their own patents, the percentage of self-citations is also used as a measure of exploration. The smaller this percentage is, the more this patent is regarded as

(7)

7

Hall et al. (2001). These patent specifics are all collected from the NBER patent database and will be used as independent variables in separate difference-in-difference regressions. In order to proxy managerial entrenchment, different forms of state antitakeover legislation will be studied. Such laws decreased the threat of hostile takeover, and were enacted in different states in the U.S. in different years, which makes this a useful dataset for a difference-in-difference regression. Firms that were incorporated in states that had enacted such state antitakeover laws at that time were considered part of the “treatment” group. Firms that were incorporated in states which at that time did not have such a law implemented were considered the “control” group.

The studied time period ranges from 1980 to 1999 due to inaccessibility of patent information after 1999. This dataset contains 147737 patents from 956 firms who applied for a patent at least once in this time period. The regressions will control for measures of industry concentration, leverage, profitability, capital expenditure and firm size.

This research is structured as follows. In section 2, the literature review is discussed, where both theoretical and empirical results are mentioned. From this literature review, a hypothesis is derived in section 2.3. Section 3 provides an extensive overview of the variables used in the dataset, the rationale behind these variables and the way they are implemented in the model. Thereafter, the results are discussed in section 4. To verify these results, robustness checks are performed in section 5. Finally, concluding remarks and further discussion is stated in section 6.

2. Literature review

In order to provide an accurate evaluation of the related literature, both theoretical predictions and empirical results will be discussed. Firstly, theoretical support for both a positive and negative relationship between managerial entrenchment and innovation will be provided. Thereafter, related empirical studies will be discussed, together with the distinguishing aspect in which they differ from this research.

2.1 Theoretical predictions

There are several theoretical frameworks that describe the role of managerial entrenchment on the level of innovation. One of these frameworks is the moral hazard view (otherwise known as agency

(8)

8

view), which argues that the possibility of a takeover works as a method of discipline for the management (Jensen & Meckling, 1976). The threat of a takeover reduces the likelihood that the management will shirk and will motivate them to select the most innovative projects with the highest overall value. According to this agency theory, the threat of takeovers therefore works as an

instrument to align the incentives of the management with that of the shareholders. Due to a

shirking management, imposing legislation that will effectively lower the possibility of a takeover will lead to a decrease in innovation.

Other theories lead to an opposite conclusion, one of these being the managerial myopia hypothesis (Chakraborty & Sheikh, 2010). This theory suggests that the possibility of a takeover causes managers to become myopic which leads them to have a short-sighted bias on investments. Instead of

choosing investments that ultimately create the highest value for the firm, managers are less likely to take on net positive value projects if they have long-term payoffs. This conclusion is supported by Manso (2011), who created a theoretical model that established that a restriction in the freedom to choose between all available projects and an increase in punishment for mistakes lowered the likelihood of innovation. Stein (1988) concurs with the managerial myopia hypothesis and adds that the asymmetric information problem between managers and the market is particularly large for such long-term projects, which are harder for the market to fully understand. If the market is incapable of valuing those projects correctly, then they will cause a temporary undervaluation of the firm. Since this increases the possibility of a takeover even more, a myopic management will be less inclined to invest in long-term innovative projects.

The theory of incomplete contracts discussed by Schleifer and Summers (1988) reaches the same conclusion. According to this theory, in the case of a hostile takeover, the acquirer benefits from the long-term returns without having paid for these long-term innovative projects. The higher the

possibility of a hostile takeover, the less keen managers therefore are to invest in long-term projects.

2.2 Empirical research

Alike the theoretical predictions, there is no consensus on the empirical relationship of the research question. Empirical research on the effects of corporate governance on innovation has been

addressed often yet the relationship between the threat of takeover and innovation has empirically led to both positive and negative results. In this part the most relevant researches will be discussed, and their methodologies and results will be compared.

(9)

9

For instance, Chakraborty & Sheikh (2010) study the relation between antitakeover amendments and long-term investments. As a proxy for the antitakeover amendments, they make use of the G-index as proposed by Gompers et al. (2001), which is based on the number of antitakeover amendments which increases the freedom of the management while decreasing the shareholder rights. Besides the G-index, Chakraborty & Sheikh also control for the level of entrenchment within the

management using the Entrenchment Index, or E-index. Higher E-indexes were denoted to firms who, for instance, had implemented supermajority requirements for voting rights, poison pills and golden parachutes. Using Ordinary Least Squares, they found a significant negative relationship between both the G- and E-index and their proxy for long-term investments, which consisted out of R&D-investments and capital expenditures. This result is in line with the managerial myopia

hypothesis, which states that a decrease in the possibility of takeovers lowers long-term investments such as innovative R&D-investments.

Pugh et al. (1992) performed a similar study as Chakraborty & Sheikh (2010) yet yield the opposite result. Pugh et al. (1992) discuss the effect of shareholder-approved amendments which decrease the likelihood of hostile takeovers. They test the effect of the number of shareholder-approved amendments with regard to R&D-investments and capital expenditures, relative to total assets. Pugh et al. (1992) make use of an event study approach to determine this effect. They find that firms who implemented shareholder-approved amendments had significantly higher R&D-expenses and capital expenditure relative to firms who did not.

However, in the research of Chakraborty & Sheikh (2010) and Pugh et al. (1992) R&D-expenses and capital expenditures are used as a proxy for innovation, while they simply measure the amount of money that is spend on innovation, and not the actual quality and quantity of innovative projects. Large sums of money can be spend on R&D, without the actual accomplishment of innovation. Whether a project turns out to be successful or not is not taken into account, which causes this proxy to fail in differentiating between useful innovation and unusable overinvestment by the

management.

While not perfect, the patent count can be used as a better measure for innovation, as used by Atanassov (2013). Atanassov studied the effects of changes in legislation on anti-takeover provisions (the so-called Business Combination Laws), which were implemented between 1984 and 1991 among different U.S. States. According to Atanassov (2013), this particular anti-takeover legislation change is widely regarded as supporting managerial entrenchment the most among other anti-takeover laws, and is widely used in related researches as proxy for a decrease in the threat of hostile takeover (Catan & Kahan, 2014). Since this legislation was implemented in various states in various years, it

(10)

10

was possible to study its effects using a difference-in-difference-method. Both the number of patents as well as the number of citations per patent were used as dependent variables herein. The number of patents provides a quantitative measure of innovation, in an attempt to remove the bias of overinvestment which otherwise would have been measured using R&D as dependent variable. To determine the importance of the patents, the number of citations a patent received is also used. Citations are references to other patents, which the new patent builds upon. The rationale behind this is that highly cited patents apparently were of sufficient interest to firms to keep investing in that particular patent. Atanassov (2013) acknowledges that the use of patents also comes with its

drawbacks, since not all industries and firms apply for patents. With regard to the patentability of particular projects and the cost of patenting, some firms decide not to apply for a patent. The difference-in-difference-regressions find that the implementation of the Business Combination Laws led to a significant decrease in innovation in comparison to state who did not (yet) implemented this legislation. These results are in line with the moral hazard view theory. There are two main points of critique to be noted on Atanassov’s research, concerning the identification of managerial

entrenchment and the identification of innovation.

Firstly, Catan and Kahan (2014) and Karpoff and Wittry (2016) criticized the use of the Business Combination Law as a proxy for managerial entrenchment. They discuss that this change in legislation cannot be used on its own since there are important antitakeover instruments not discussed in this law. Catan and Kahan (2014) state that, for instance, the use of poison pills is of far greater importance to firms than the antitakeover amendments that were covered in the Business Combination law. The legislation regarding poison pills therefore cannot be left uncontrolled for, and therefore the empirical results of Catan and Kahan (2014), which do control for these laws, do not show the same significance as the results of Atanassov (2013). Several other important changes in antitakeover legislation were also not taken into account, according to Karpoff and Wittry (2016). They provided a new framework to study antitakeover legislation by controlling for the five most important and most widely used antitakeover laws and, similarly to Catan and Kahan (2014), did not find the same significant results as Atanassov (2013). Karpoff and Wittry added another point of critique, by noting that Atanassov (2013) overestimated the level of antitakeover protection provided by the Business Combination laws. They contributed by stating that underestimating the level of antitakeover protection implemented before the Business Combination laws, could lead to distorted results by improperly coding the level of antitakeover protection.

Secondly, the other aspect of criticism on Atanassov’s research regards the use of patents and patent citations as a measure of innovation. Balsmeier et al. (2016) researches the relationship between the independence of the board and innovation. To assess the quality of innovation, they discuss whether

(11)

11

all patents are equally innovative, or that specific measures of patents can be used to form a more suited proxy to determine which patents truly aid innovation. Balsmeier et al. (2016) distinguish between two forms of innovation: exploration and exploitation. Herein, exploration is meant with new inventions of which patents are applied for in a new field of technology or a new field of technology for that firm. Exploiting innovation is described as innovation that keeps on improving existing patents or existing fields of technology. Balsmeier et al. (2015) created several variables to act as a proxy for exploring innovation, which is the kind of innovation they regard as truly

innovative. One of these variables is the number of backward citations per patent. Although

Atanassov made use of forward citations per patent, Balsmeier et al. (2016) looks at citations from a different perspective. They state that the larger the number of backward citations a patent has, the less innovative this patent is, since the large number of other patents that the new patent builds upon indicates that this market is likely to be crowded and mature. Besides the number of backward citations, the number of citations is also taken into account. A patent with a large number of self-citations points toward a patent in an industry in which the assignee apparently is already

experienced, indicating exploitation instead of exploration. Balsmeier et al. (2016) find that an increase in governance quality, due to an increase in board independence, led to an increase in the exploiting of existing technologies. As proxy for the increase in board independence, they made use of the implementation of the Sarbanes-Oxley Act and found that it led to an overall increase in the number of patents but an overall decrease in exploring innovation.

Similarly to Atanassov (2013), this paper intends to make use of changes in antitakeover legislation. However, as Catan and Kahan (2014) and Karpoff and Wittry (2016) mention, in order to increase the predictive power of the model, other legislation besides the Business Combination law need to be taken into account. Therefore, this research will make use of the five most important and most widely used antitakeover laws, provided by Karpoff and Wittry (2016). These five laws are the Business combination laws, Control share acquisition laws, Directors’ duties laws, Fair price laws and Poison pill laws. A detailed explanation is provided in the data section and appendices A, B and C. As dependent variables, the innovation variables as described in Balsmeier et al. (2016) will be

implemented. By doing so, this paper should be able to determine how a change in managerial entrenchment can affect exploring innovation.

(12)

12

2.3 Hypothesis

This research combines papers that look at legislation that increased managerial entrenchment from different angles in order to create the best possible proxy for this. Previous research by Atanassov (2013) found that an increase in managerial entrenchment led to a decrease in the number of patents, and concluded that therefore, innovation declined. The research of Karpoff and Wittry (2016) and Catan and Kahan (2014) reconsidered this research, and after controlling for several other changes in legislation, found no such significant relationship. Therefore, the empirical relationship between managerial entrenchment and innovation remains ambiguous.

However, Balsmeier et al. (2016) studied the relationship between innovation and another form of corporate governance, board independence. They stated that patents were not sufficiently

appropriate to proxy true innovation, and looked at several variables that did proxy exploration adequately. Their results showed that the increase in board independence (followed by the

introduction of the Sarbanes-Oxley Act) decreased the number of truly innovative patents. This effect was significantly more pronounced for firms that had relatively poor levels of corporate governance, and they conclude that firms with high levels of corporate governance tend to innovate relatively more in the form of exploitation (Balsmeier et al. 2016).

This makes it possible to hypothesize the effect between true innovation and managerial entrenchment measured by changes in antitakeover legislation. Therefore, the hypothesis is as follows:

H1: An increase in the level of managerial entrenchment increases exploring innovation

The reasoning for this is that the increase in board independence that Balsmeier et al. (2016) studied, can be interpreted as an increase in the level of corporate governance, while the changes in

antitakeover legislation can be interpreted as a decrease in the level of corporate governance. If the results of Balsmeier et al. (2016) hold, then the changes in antitakeover legislation should increase true innovation in the form of exploration.

3. Methodology & Data

The hypothesis will be tested using a difference-in-difference regression, similar to Atanassov (2013), Balsmeier et al. (2016) and Catan & Kahan (2016). The longitudinal dataset used in this research

(13)

13

ranges from 1980 to 1999 and contains 147737patents from 956public U.S. firms which were granted at least one patent in that timeframe. In order to assess how firms’ innovation altered following the changes in legislation, these data were collapsed to individual firms per year level. This provided a dataset with 6188 firm years, which is a comparable number of observations relative to the dataset used by Balsmeier et al. (2016).

The dataset contains innovation data of the NBER patent database, dating from 1963 to 1999. The full dataset is not used since most studied changes in legislation were implemented from 1982 onwards and several control variables are not accounted for before 1980. The information on the patents was merged via CUSIP to firm-specific financial data from Compustat. Karpoff & Wittry (2016) provided the years in which the state anti-takeover laws were effectively implemented, this data was merged on state of incorporation. Data on the Herfindahl-Hirschman Index (HHI) originated from the U.S. Census from 1982 to 1997. The HHI is merged on industry using the 4-digit SIC-codes between 1980 and 1994. Due to a change in the measurement of industries, the HHI between 1995 and 1999 is merged on 6-digit NAICS-codes.

The following section will thoroughly explain the variables used, and the reasoning behind this. Then, descriptive statistics are provided of the key variables and discussed. After that, the difference-in-difference model will be specified and explained.

3.1 Dependent variables: measuring managerial entrenchment

As a proxy for the level of managerial entrenchment, the expected possibility of a hostile takeover will be used. Since 1968, legislation has been implemented in different states in the U.S. which affected antitakeover provisions and therefore the expected possibility of hostile takeover

(Atanassov, 2013). The number of antitakeover provisions increased rapidly in the following years, decreasing the possibility of a hostile takeover. On June 23, 1982 the U.S. Supreme Court declared the legislation up until then invalid in the Edgar vs MITE corp. case, since it covered firms outside of state jurisdiction (Karpoff & Wittry, 2016). In the literature, legislation up to the 1982 is referred to as first-generation anti-takeover laws.

Since then, new legislature in the form of second-generation anti-takeover laws has been enacted, covering only the state of incorporation. Within these laws, Business Combination laws have been one of the most common type of second-generation laws that has been used in finance literature as a proxy for a decrease in the possibility of hostile takeover (Catan & Kahan, 2014). These laws impose

(14)

14

a disruption for a number of years on asset sales and mergers between the firms protected by the Business Combination Laws and large shareholders when they reach a certain threshold of ownership. In effect, this decreases the likelihood of hostile takeovers.

As noted section 2.2, Business Combination Laws on their own are not able to serve as a good proxy for lowering the likelihood of hostile takeovers since not all anti-takeover instruments are covered within these laws. For instance, a firm still has the possibility to withhold a hostile takeover by taking on poison pill takeover defenses (Catan and Kahan, 2014). Doing so gives them the possibility to sell shares to all other shareholders besides the acquiring party at a discount, preventing the hostile takeover from happening. Thus, whether a firm has implemented poison pills or not is of great importance when looking into their anti-takeover provisions as well (Catan & Kahan, 2014). Since poison pills are not covered under Business Combination Laws, it is necessary to also take the legitimacy of poison pills into account.

Poison pill laws are by far not the only other form of anti-takeover legislation that has been

implemented in the United States since the U.S. Supreme Court ruling in 1982. Since then, 43 states have effected a minimum of 157 second-generation anti-takeover laws (Karpoff & Wittry, 2016). In order to cover as much anti-takeover provisions as possible, this paper will take the five most prominent changes in anti-takeover legislation into account, following the methodology of Karpoff and Wittry (2016). The laws that will be used are Business combination laws, Control share

acquisition laws, Directors’ duties laws, Fair price laws and Poison pill laws. A description of these laws and the date of effective implementation can be found in Appendix A and B, respectively. The dates in which these laws were implemented in each of the 43 states were provided by Karpoff and Wittry (2016).

In order to make use of these second-generation laws, five separate dummy variables are created which equal one when a patent was applied for by a company that was incorporated in a state that had one of those five antitakeover laws implemented at that time, and zero when not. Since the dataset starts in 1980, when the level of antitakeover was deemed relatively high by Karpoff and Wittry (2016), first-generation antitakeover legislation also needs to be taken into account. Appendix C provides the effective dates on which first-generation antitakeover laws were implemented in 38 separate states. Another dummy variable named First Generation Antitakeover Laws is generated that equals one when a patent was applied for by a company that was incorporated in a state that had implemented this form of legislature at that time up until 1982. After 1982, First Generation

Antitakeover Laws is set equal to zero, in the same way to States that did not implement any of the

(15)

15

Balsmeier et al. (2016) cover two other relevant court decisions, which deemed the Business

Combination Laws and the Control Share Acquisition Laws officially constitutional. The legality of the Control Share Acquisition Law was not confirmed up until then, because in the Dynamic Corporation

of America v. CTS Corporation case, the District Court and the Court of Appeals ruled that the Control

Share Acquisition Law was in conflict with the federal Williams Act (Choate, 1987). However, on April 21, 1987 this was overruled by the U.S. Supreme Court who deemed Control Share Acquisition Laws constitutional. A similar ruling by the U.S. Court of Appeals in Amanda Acquisition Corp v. Universal

Foods Corp. concerning Business Combination Laws, deemed them legal on May 24, 1989 (Cuomo,

1989). Since the legality of these laws was only confirmed after these court rulings, another dummy variable is created. This dummy is set equal to one when the court ruling has passed, it is then multiplied by the concerned antitakeover legislation dummies in order to control for these laws from the time that they were deemed constitutional. In the following regressions, these variables are labeled Control Shares Court Ruling and Business Shares Court Ruling.

The creation of these dummy variables makes it possible to perform a difference-in-difference regression (Catan and Kahan, 2014), since the longitudinal dataset contains “treated” and “control” firms. In this difference-in-difference framework, “treated” firms are those which are incorporated in a state which had implemented any of the first- or second-generation antitakeover laws at that time. “Control” firms are those who are incorporated in a state which did not had implemented any of the first- or second-generation antitakeover laws at that time. Using such dummy variables in a

difference-in-difference framework is in line with previous related research from Atanassov (2013), Catan and Kahan (2014), Karpoff and Wittry (2016) and Balsmeier et al. (2016).

3.2 Independent variables: measuring innovation

This research aims to establish the relationship between managerial entrenchment and truly innovative patents, so-called exploring innovation instead of exploiting innovation (Balsmeier et al., 2016). As described in the literature review, previous research has mainly focused on exploiting innovation, by using quantitative proxies for innovation. For instance, Atanassov (2013) made use of the number of patents that firms applied for, as well as the number of citations their patents

received. When applying for a patent, firms are legally bound to cite previous patents that hold information upon which their patents builds. This prohibits the patenting firm to claim knowledge that already has been patented (Hall et al. 2001). Atanassov (2013) made use of the number of

(16)

16

patents and the number of citations per patent as a measurement of innovation. Similarly, Chakrabarty and Sheikh (2010) used the level of investment in R&D as a proxy for innovation. Balsmeier et al. (2016) noted that such measures only look at the size of innovation and overlook whether a patent is truly innovative or not. Manso (2011) distinguishes between exploitation and exploration, where exploitation refers to patents which are applied for in a mature market in which the patents build upon well-known technologies. Contradictory, exploration refers to patents applied for in new technological fields, or in fields previously unknown to the firm. This research will make use of several variables as proxy for exploration, in line with Balsmeier et al. (2016).

The first variable that will be used is the number of backward citations per patent, similarly to Atanassov (2013) as previously mentioned. However, Atanassov (2013) made use of the number of citations as measure for how valuable a patent was, where a patent with a higher number of citations was denoted as a more valuable patent. This paper will look at the information that citations provide regarding the links between patents, as mentioned by Hall et al. (2001). Since patenting firms are obligated to report citations to already issued patents, a high numbers of citations can therefore be linked to a patent in a more mature industry, which corresponds to exploitation rather than exploration (Balsmeier et al. 2016). The number of backward citations is provided by the NBER patent database (Hall et al. 2001), and the average number of backward citations per firm per year is calculated in Stata and named citations.

Another measure suggested by Balsmeier et al. (2016) is the number of citations a company makes to patents owned by themselves, otherwise known as self-citations. A large number of self-citations points toward research in areas familiar to the firm, and therefore indicates exploitation, while a small number of self-citations indicates exploration (Sørenson & Stuart, 1999). The NBER patent database provides two measures of the number of self-citations, described as a lower bound and an upper bound (Hall et al. 2001). The NBER calculated this for each assignee, by counting the number of citations which were assigned to the same assignee and dividing this by either the total number of citations that the assignee made (lower bound) or by the total number of citations that have an assignee code (upper bound). The reason that the lower bound is considered as such is that data on citations is only available from 1969 onwards, so self-citations before 1969 are not taken into account. In this dataset, the mean value of the lower- and upper-bound are respectively 15.5% and 17.9%. Since the true value of the percentage of self-citations lies between the lower- and upper-bound, the average value per firm per year is calculated and named selfcitation.

The third variable Balsmeier et al. (2016) propose is the total number of claims that a firm made per year. They argue that if a firm desires immediate results, it is likely that it will put more effort in the

(17)

17

patenting process. Since the number of claims is connected with the effort that a firm puts into its patents, Balsmeier et al. (2016) argue that firms with more claims are more likely to have innovative patents. Lanjouw and Schankerman (2004) concur with this statement and claim that a high number of claims points toward a more novel and potentially more profitable innovation, which is in line with exploration. The total number of claims per firm per year originate from the NBER patent database, they are labeled claim.

The last variable that will be studied is the originality index, which originated from the NBER patent database (Hall et al. 2001). Similarly to the first measure, the originality index is also based upon a patent’s backward citations. This index takes the origin of the backward citations into account, and measures the fraction of total citations that originate from separate patent classes. It is calculated as follows:

𝑂𝑟𝑖𝑔𝑖𝑛𝑎𝑙𝑖𝑡𝑦𝑖= 1 − ∑ 𝑠2𝑖𝑗 𝑛𝑖

𝑗

Where s denotes the fraction of backwards citations made by patent i in patent class j, out of ni

patent classes (based on the 3-digit patent classes from the NBER patent database) (Hall et al. 2001). High scores of originality are rewarded to firms that cite patents from a large array of patent classes, while low scores are given to firms that cite patents from a narrow set of patent classes. According to Hirschleifer et al. (2015), this measure of originality serves as a proxy for innovative originality since originality is measured by how wide the range of information used is to innovate. In existing

literature, the variety of knowledge used to come up with a new patent is often chosen to proxy innovation coming from the view that innovation can originate from joining existing technologies in such a way that creates new technologies (Hirschleifer et al. 2015). The NBER patent database provides the originality index, which is constructed by Hall et al. (2001) and is named original. These four measures will be used separately as independent variables in the difference-in-difference regression. If the hypothesis were to be confirmed, the directions of the coefficients of citations and

selfcitations would be negative, and the hypothesized directions of the coefficients of claims and original would be positive. Patents and the measures that rely upon them usually have a two-year lag

from the moment of application to the grant date (Atanassov, 2013). In this period, the patent application has to progress through the U.S. Patent Bureau procedure. Since the timing of the actual innovation and the application date are closely related, the application year will be used as relevant date. To control for this lag, the dependent variables will have to be regressed using lags in line with Atanassov (2013).

(18)

18

3.3 Control variables

In order to minimize omitted variable bias, firm-specific and industry-specific control variables will also be used in the difference-in-difference regression. Compustat provides annual data per firm, which can be used to create variables based on firm characteristics in line with existing literature. The variable firmsize is computed by taking the logarithm of total revenues, in order to take care of the skewness of this variable (Balsmeier et al. 2016). One can predict larger firms, with higher revenues, to have more funding available that can be spend on innovation. Following Atanassov (2013), a variable is generated to control for profitability (_{𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠}𝐸𝐵𝐼𝑇𝐷𝐴𝑖𝑡

𝑖𝑡), since relatively more profitable firms have more investment opportunities. Since financial constraints affect the

investment choices of a firm and therefore the flexibility of the firm, this paper controls for capital

expenditure (𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝐸𝑥𝑝𝑒𝑛𝑑𝑖𝑡𝑢𝑟𝑒_{𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠} 𝑖𝑡

𝑖𝑡 ) and the mean industry leverage (

𝐿𝑜𝑛𝑔 𝑡𝑒𝑟𝑚 𝑑𝑒𝑏𝑡𝑖𝑡

𝑇𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠𝑖𝑡 ), based on 4-digit SIC industry codes. Another reason for which leverage is taken into account is that firms with high leverage are more likely to have reduced free cash flows which increases the opportunity that the management will be replaced (Atanassov, 2013). Since this is directly connected to the level of managerial entrenchment, this is an important factor to account for. Atanassov (2013) argues that firm-specific leverage is likely to have endogeneity problems, while the mean industry leverage will not be manipulated by firms themselves which makes it a more suitable proxy. Since high leverage can have a negative impact on the firm’s flexibility, one can expect it to have a negative relationship with true innovation, since it can force companies to invest in saver, exploiting innovations. In order to manage other firm specific characteristics that can influence the results, I control for firm fixed effects in all regressions.

Industry specific characteristics can also play an important role in firms’ innovative behavior. Giroud and Mueller (2010) found that an increase in competition within an industry can keep management in an industry from shirking. Following the moral hazard view as described above, this can lead to an increase in patenting, and therefore a possible increase in truly innovative patents. However, Schumpeter (1942) claimed that less competitive industries are more innovative, since they are able to benefit more from the profits that are gained by the innovation, whereas the firms in competitive industries would have those profits competed away earlier. As a proxy for market competition the Herfindahl-Hirschman index (HHI) is often used (Atanassov (2013), Giroud and Mueller (2010), Ali et al. (2009)). This index measures industry competition by adding up the squared market shares of that industry. In an industry with a high concentration, where a few companies have large market shares,

(19)

19

the HHI gives large scores. An industry with a relatively large number of small companies obtains a low score, indicating a highly competitive market.

Market shares per company for each industry are necessary in order to calculate the HHI. Atanassov (2013) and Giroud and Mueller (2010) calculated these market shares using the proportion of industry sales, based on the Compustat dataset. However, Ali et al. (2009) discuss that the use of the market shares based only on companies from Compustat does not function well as a proxy for the HHI. Since Compustat only lists public firms, they make use of the HHI as gathered by the U.S. Census of Manufactures, for which every U.S. firm is obligated to report to. They state that when using the Compustat HHI, they find results contradictory to several theoretical productions regarding market concentration. The Compustat HHI has a correlation of only 0.13 with the U.S. Census HHI, for which the results are in line with theoretical predictions. Hence, this paper makes use of the HHI based on the U.S. Census in the same manner as Ali et al. (2009). Since the U.S. Census does not make data available from the Census earlier than 2002, co-author S. Klasa of Ali et al. (2009) provided the U.S. Census HHI ranging from 1982 to 2002. These figures can be used to proxy the industry

concentration from 1980 to 1999, in line with Ali et al. (2009). This data has been merged to the companies via Standard Industry Codes (SIC) up to 1994, and via the North American Industry Classification System (NAICS) codes from there on due to a shift in industry measurement. While the positive aspect of using the U.S. Census HHI is obvious, it does come with a negative aspect. The U.S. Census only provides information regarding industry concentration on the manufacturing sector, thus with SIC codes ranging from 2000 to 4000. This means that patents from other sectors are discarded, which leaves a smaller yet more accurate dataset. This need not necessarily be a problem, since the majority of patenting activity is registered through this sector (Balasubramanian and Sivadasan, 2011). Excluding SIC codes below 2000 and above 4000 also removes patents from the financial, utilities and computer sector which operate under specific regulation (Amore et al. 2013). Excluding these sectors is in line with Atanassov (2013).

3.4 Model Specifications

This research will make use of difference-in-difference regression in order to establish the empirical relationship between managerial entrenchment and innovation. The difference-in-difference regression is based on the changes in first-generation- and five separate second-generation

antitakeover legislations which were implemented in different years in separate U.S. states. Dummy variables are created which equal one when a state has implemented this legislation on that time,

(20)

20

and zero when otherwise. This creates treated and non-treated firms, which allows the difference-in-difference regression to establish a causal relationship, not mere a correlation. This form of

difference-in-difference regression is widely used in related research (Atanassov (2013), Karpoff and Wittry (2016), Balsmeier et al. (2016)).

The regression model that will be used has the following form:

log (1 + 𝑦_{𝑖𝑘𝑠(𝑡+𝑛)}) = 𝛼 ∗ 𝑙𝑒𝑔𝑖𝑠𝑙𝑎𝑡𝑖𝑜𝑛_𝑠𝑡+ 𝛽 ∗ 𝑙𝑒𝑔𝑖𝑠𝑙𝑎𝑡𝑖𝑜𝑛_𝑠𝑡∗ 𝑐𝑜𝑢𝑟𝑡 𝑟𝑢𝑙𝑖𝑛𝑔_𝑡+ 𝛾_𝑖𝑡+ 𝛿𝑘𝑡+ 𝜂𝑡+ 𝜃𝑖+ 𝜀𝑖𝑘𝑠𝑡

where i indexes firms, k indexes the industry, s indexes the state of incorporation, t indexes the time in years and n indexes the number of years the dependent variable will be lagged. The dependent variable is represented by 𝑦𝑖𝑘𝑠(𝑡+𝑛), in which y stands for the four measures of innovation: citations,

selfcitations, claims and original. In accordance with Atanassov (2013), lags will be used to mitigate

the fact that the year of application and the grant year are on average two years apart. Out of the independent variables, legislationst is the vector of antitakeover legislation dummies consisting out of

Business Combination Laws, Control Share Acquisition Laws, Directors’ Duties Laws, Fair Price Laws, Poison Pill Laws and the First Generation Antitakeover Laws. Two separate court rulings deemed the Business Combination Laws and the Control Share Acquisition Laws constitutional in 1989 and 1987,

respectively. Therefore, the concerned antitakeover legislation dummies are multiplied with another dummy variable that equals one when the court decision has passed. A vector of firm-specific control variables is also included, indexed by αit. Industry-wide control variables such as industry competition

and industry leverage are indexed with βkt. This regression will also control for year-fixed and

firm-fixed characteristics with γt and δi respectively. Lastly, the error term is indexed with εikst.

Using dummies to account for the different changes in legislation, this effectively creates a difference-in-difference regression. Doing so leads to the establishment of a causal relationship instead of a mere correlation. This is because the change in innovative ability of a firm, due to change in legislation, is controlled for by comparing it with firms that did not face such legislation. This controls for trends in the measures of innovation, by taking the difference in innovative ability out of the difference in separate states. The statistical power of this model is even larger due to the fact that firms can be part of both the treatment and control, at separate times, when the law is implemented at that time or not (Atanassov, 2013). Besides controlling for such trends, statistical noise is reduced further by controlling for non-observable time- and firm-specific fixed effects. Due to the skewness of the variables, the dependent variables’ functional form is transformed to a logarithm. This is in line with Atanassov (2013) and Balsmeier et al. (2016). Also, in order to control

(21)

21

for autocorrelation between firms, the regressions make use of robust standard errors (Balsmeier et al. 2016)

3.5 Descriptive Statistics

In order to mitigate biases due to outliers, winsorizing at the 0.5 percent level was performed on the measures of profitability, mean industry leverage, capital expenditure and firm size. I end up with a dataset consisting of 956 firms and 6188 firm years. Descriptive statistics are provided in table I, which is divided into three parts. The first part denotes the characteristics of the dependent

variables, the second part denotes the dependent variables in the functional firm that will be used in the regressions and the third part denotes several control variables.

In the first part of table I it is noteworthy that in particular, the count variables citations and claims have a relatively broad range with means of respectively 11.51 and 13.83 and maximum values of 219.7 and 111. By taking the logarithm of these variables, as is done in the second part, the skewness of these variables is lowered.

In the last sector, statistics of several control variables are shown as well as the patent count,

although it is not featured in these regressions. Still, the number of patents is featured in this table in comparison, since it has been used throughout previous literature. The average number of patents per firm per year is 23.87, with a maximum value of 1363 patents. Furthermore, the Herfindahl-Hirschman Index shows that the dataset consists of both very concentrated as well as very competitive industries, with ranges of 14 to 2999. Also, both profitable firms and firms with a negative EBITDA are featured in this dataset.

Since this research uses a similar dataset as used by Balsmeier et al. (2016), that research makes the most useful comparison. Assessing the dataset, it appears that the data selection has been

successful. The values of the control variables such as firm size, capital expenditure and research &

development, are quite similar. Although Balsmeier et al. (2016) made use of a dataset that ranged

(22)

22

4. Results

Four difference-in-difference-regressions are performed in order to determine the relationship between managerial entrenchment and the four measures of innovation. The regression results on the dependent variables citations, self-citations, claims and the originality index are shown in tables II, III, IV and V respectively. The regression results and their interpretation will be discussed next.

4.1 The effect of antitakeover legislation on the number of backward citations

In table II the regression results of the regression model, introduced in the methodology section, on the number of backward citations is presented. Specifically, the dependent variable used in columns (1) to (4) is the logarithm of one plus the number of backward citations, lagged between one and four years. These lags are used in order to control for the fact that the application date of a patent is in our dataset, while the patenting procedure takes on average two years to grant the patent (Atanassov, 2013). The coefficients of interest are the first eight, out of which the first six represent the implementation of first- and second-generation antitakeover laws, while the latter two represent the effective dates of two second-generation antitakeover laws, based on specific court rulings. In all regressions, firm- and year-fixed effects are included, as well as robust standard errors.

In line with the hypothesis, the implementation of antitakeover legislation should be accompanied with a decrease in the number of backward citations. This would indicate that the reduced possibility of a hostile takeover would provide the management with a larger incentive to invest in explorative rather than exploiting innovation. However, only the dummy variable First Generation Antitakeover

Laws shows high significance in column (2), where the dependent variable is lagged for two years.

This indicates that the first-generation antitakeover legislation, which was effective until 1982, had a negative effect on the number of backward citations which indicates explorative innovation.

Unfortunately, it is not possible to control whether this effect holds for the years thereafter since the lagged variable surpasses the dataset that starts in 1980. This is also the reason that the number of observations declines from 4390 to 3044.

Also, in column (3) the coefficient of the Business Combination Laws in combination with the court decision in 1989 is negative and significant at the 10 percent level. The economical meaning of this is that the implementation of the Business Combination Laws only lowered the number of backward citations when it was effectively implemented in 1989. One can tell because the coefficient of the

(23)

23

Business Combination Laws alone is not significant, while it is significant when it is multiplied with a dummy variable set equal to one from 1989 onwards. The negative coefficient of this variable indicates explorative innovation, similar to the first-generation antitakeover legislation.

In column (1) there are contradicting results, positive results for the implementation of Poison Pill Laws and negative results for the implementation of Fair Price Laws and Directors’ Duties Laws. These are the results for the one-year lag of the dependent variable, while there is on average a two-year lag before patents that were applied for are granted. According to Atanassov (2013) the

implementation of the laws could be expected beforehand, causing the laws to have an effect on innovation before the two years. If one of these three laws was expected upon, then this could give this results an economic meaning. Since this is unknown, it is not possible to give these results a causal interpretation.

Within the control variables, the most significant results are found at the logarithm of total revenues, or firm size. This shows a positive and highly significant relationship between total revenues and the number of backward citations, indicating that firms with higher revenues generally have more backward citations. This is in line with theoretical expectations of exploitation, which is a less risky way for firms to make steady profits (Balsmeier et al. (2016). The effects of other control variables are ambiguous, for instance that of the mean industry leverage. This is significant and positive in column (3) yet significant and negative in column (4). This could be due to the way of measurement, since it is measured for the entire industry to sort out endogeneity problems, yet therefore perhaps loses its statistical power. Also, the Herfindahl-Hirschman Index gives negative values which are significant only in column (3). This is line with the economic view of Schumpeter (1942), which states that a decrease in competition increases innovation. In particular, explorative innovation is increased which makes sense since firms can reap the benefits for a longer period of time if the patent is truly innovative. Overall, the statistical evidence of the effect of antitakeover legislation on the number of backward citation is mild, yet negative. This is a slight indication of explorative innovation.

4.2 The effect of antitakeover legislation on the percentage of self-citations

In table III the regression results are given of the regression model, introduced in the methodology section, on the percentage of citations. Herein the dependent variable is the percentage of self-citations, and columns (1) to (4) represent lags of this variable ranging from one to four year. Since

(24)

24

lower percentages of self-citations indicate explorative innovation, negative coefficients are expected in order to confirm the hypothesis.

The most significant results are once more those of the First Generation Antitakeover Laws dummy variable. Table III shows negative coefficients in columns (1) and (2), while columns (3) and (4) are again omitted due to the limited years in the dataset before the 1982 U.S. court decision that ruled all first generation antitakeover laws unconstitutional. Similarly, negative and significant coefficients can also be found in column (3) and (4) of the Fair Price Laws dummy variable. The implementation of the Fair Price Laws led to a decrease of the percentage of self-citations three and four years after the date of implementation. Both these significant results point in the direction of the hypothesis. However, column (3) also displays a positive coefficient on the Business Combination Laws dummy. This indicates that the implementation of the Business Combination Laws increased the average percentage of self-citations, indicating exploitation rather than exploration. However, this result is significant only at the 10 percent level, and the dummy variable indicating when the Business Combination Laws truly became effective (BC*Court Ruling) shows no significance. Therefore, the reliability of this result is questionable.

Regarding the control variables, profitability shows positive and significant results in all four columns. This could indicate that more profitable firms tend to have relatively more self-citations, and

therefore make use of exploitation more. Similarly to the positive and significant coefficients found at firm size in section 4.1, exploitation can be used as a less risky way for firms to make steady profits and thus creating high revenues (Balsmeier et al. (2016).

4.3 The effect of antitakeover legislation on the number of patent claims

In table IV the regression results are given of the regression model, introduced in the methodology section, on the number of patent claims. Herein the dependent variable is the percentage of self-citations, and columns (1) to (4) represent lags of the dependent variable ranging from one to four year. Since higher number of patent claims indicate explorative innovation, positive coefficients are expected to be found in order to confirm the hypothesis. However, contradictory results are found. In this table the Business Combination Laws dummy as well as the effective date of the Business Combination Laws (BC*Court Ruling) exhibit significant positive coefficients in column (2). This is a result in conformation with the hypothesis. However, column (1) states a similar positive and

(25)

25

significant coefficient at the Business Combination Laws dummy while displaying a significant negative coefficient at the effective date of the Business Combination Laws.

Although the dummy variable of the effective date of the Control Share Acquisition Laws (CS*Court

Ruling) also shows a significant positive coefficient in column (2). This is refuted by significant

negative results of Fair Price Laws in column (2), (3) and (4) as well as the dummy variable for First

Generation Antitakeover Laws in column (2).

Table IV shows a number of variables with significant results that vary in the direction of their coefficient. This denotes that the effect of antitakeover legislation on the number of patent claims is ambiguous which makes its effect on explorative innovation unclear.

4.4 The effect of antitakeover legislation on the originality index

In table V the regression results of the regression model, introduced in the methodology section, on the originality index is presented. Specifically, the dependent variable used in columns (1) to (4) is the logarithm of one plus the number of backward citations, lagged between one and four years. The originality index is suggested by Hall et al. (2001) and is measured by the number of patent classes out of which a patent’s citations consist. Higher scores of this index represent exploring innovation, thus a positive coefficient is expected to be found in order to confirm the hypothesis.

These significant positive values are found in column (1), which represents a dummy for the effective implementation date of the Control Share Acquisition Laws (CS*Court Ruling), as well as in column (3) for the implementation of Poison Pill Laws. Similar to the reasoning in section 4.1, it is not possible to know whether the significant result in column (1) has economic meaning. This is because the average lag between application and grant date of a patent is on average two years. While it could be that firms expect some laws to be implemented, this is not known for sure, creating uncertainty in the results. In these results it is noticeable that the First Generation Antitakeover Laws dummy is omitted for column (2), (3) and (4), as well as the dummy variable for the effective date of the Control Share Acquisition Laws (CS*Court Ruling) in column (4). Due to the long time lags in these models, several years from the beginning from the dataset are not accounted for, which causes multicollinearity issues. With regard to the control variables, the profitability measure has a significant and negative coefficient. This in line with previous findings, that indicates that more profitable firm tend to have less explorative innovations.

(26)

26

5. Robustness checks

The results of section 4 indicate that there is no strong statistical evidence to confirm a causal relationship between managerial entrenchment and explorative innovation. In fact, the results even show contradicting coefficients within a regression. To see whether these results hold when using different functional forms or implementing other variables, several robustness checks are performed.

5.1 Altering the functional form of dependent count variables to Poisson

Firstly, the count variables citations and claims are normally adjusted for their skewness by taking the logarithm of one plus these variables in the regressions. One alternative way of the functional form of the regression for a count variable is by using a Poisson model as proposed by Atanassov (2013). The log-linear model is used in the main regressions, since Poisson models remove data from the dataset that have only one observation. Since these observations add statistical power, they were used in the main regressions in line with existent literature. Table VI compares the regression results of the log-linear model with the Poisson model for the dependent count variable citations.

Specifically, in column (1) and (2) the regression results of the two year lagged dependent variable is presented, in log-linear form and in Poisson form respectively. In column (3) and (4) the regression results of the three year lagged dependent variable is presented, in log-linear form and in Poisson form respectively. The two and three year lags were chosen since they showed significant results with economical meaning, therefore it is useful to know whether these results are robust to the use of a Poisson model.

In columns (1) and (2), the two functional forms for the two-year lag are being compared. There is a noticeable drop in the number of observations and in the number of firms used, this is due to the fact that a Poisson regression omits firms that only show up once in the dataset. It is also noticeable that the main significant result hold, that is, that the First Generation Antitakeover Laws dummy variable has a significant negative effect on the number of backward citations. When comparing columns (3) and (4), the coefficient of BC*Court Ruling that was significant at the ten percent level in column (3), has lost its significance in column (4). This could be due to the lower number of

observations, which reduced the statistical power of the model. However, most control variables hold their significance and more importantly the direction of their coefficient such as the

(27)

Herfindahl-27

Hirschman Index, profitability and the mean industry leverage. On the basis of these observations

one can say the results are robust to estimation with Poisson estimation.

5.2 Increasing the dataset to control for First Generation Antitakeover laws

Another robustness check will control whether the First Generation Antitakeover Laws dummy maintains its significance when the data has an earlier beginning date. In the regressions presented in section 4, this variable was omitted in column (3) and (4) due to the fact that all these laws were deemed unconstitutional in 1982, and the dataset started in 1980. Column (3) and (4) represent time lags of three and four years, which yield only values of zero for the dummy variable. The problem with the dataset is that there is reliable industry concentration information only from 1980 onwards (see section 3 for more information on the industry concentration variable). For the following robustness test, the industry concentration control variable is omitted, which enlarges the dataset with four more years of data since the NBER patent data is available from 1976 (Hall et al. 2001). By having these additional years of data, it is possible to verify if the First Generation Antitakeover Laws dummy remains significant with a lag of three and four years. Table VII shows the effect of

antitakeover laws on the percentage of self-citations. The percentage of self-citations is chosen to be presented here, since the First Generation Antitakeover Laws had the most significant results on the percentage of self-citations. Comparing table VII to table III, it is immediately noticeable that the number of observations of table VII is much higher than those of table III. This is not only due to the fact that the regression results of table VII have a five year longer time range, it also does not exclude firms from outside of manufacturing. Although the number of observations is increased, the

statistical power is also somewhat decreased since industries with specific legislation are also take into account without being able to control for this form of industry-specific legislation.

When comparing the results between table VII and table III, it is immediately noticeable that the coefficients of the First Generation Antitakeover Laws dummy are no longer significant. This shows that the negative relationship between first generation antitakeover legislation and the percentage of self-citations is not robust to omitting industry concentration. Looking at the other significant coefficients, we notice in both column (2) and (3) that there are significant coefficients for the

legislation dummies that point in opposite directions. The results still find a negative significant effect of Fair Price Laws in column (3) in both tables. Also, a significant negative effect is reported on the coefficient of the Control Share Acquisition Laws dummy, while the coefficient of the effective date of these laws (CS*Court Ruling) report a significant positive effect. These ambiguities on the results of

(28)

28

the legislation were also reported in section 4.3, where the effect of the antitakeover legislation on the number of claims is presented. Although the ambiguity reported in table VII could be due to the lower statistical power of the model that does not control for industry concentration, the overall results do not point in a clear direction. That result is maintained in the robustness checks.

6. Conclusion/Discussion

Conclusion

The effects of the implementation of antitakeover legislation on innovation have shown ambiguous empirical results (Atanassov, 2013). This research aimed to aid this field of research, by researching the effect of managerial entrenchment on explorative innovation. Balsmeier et al. (2016) made the distinction between two forms of innovation. Patents that were applied for in mature industries, in which the applying firm was already granted previous patents were denoted as exploiting patents. Exploitation was named as such, since they exploited existing and familiar technologies to form new technologies. To Balsmeier et al. (2016), exploiting patents were not considered truly innovative patents, since they simply build upon existing technologies instead of creating pioneering ideas out of nowhere. Patents that did fall in that category were named explorative innovation, and considered truly innovative. These patents were applied for in categories previously unknown to the firm, or were not linked to other patents owned by the firm.

In order to proxy managerial entrenchment, the implementation of state antitakeover laws is chosen in line with extant literature. In order to create the best possible proxy, several critiques on

previously used variables in the literature were taken into account (Catan and Kahan, 2014). This research makes use of data ranging from 1980 to 1999. It accounts for the first-generation state antitakeover laws which were implemented in several states and were disregarded after a U.S. Court decision in 1982. From there onwards, second-generation antitakeover legislation was implemented. Since there were several forms of this type of legislation, and not one all-encompassing law, the five most important second-generation antitakeover laws were taken into account. These forms of legislation were implemented at different times across different U.S. states (Karpoff and Wittry, 2016). The implementation of these laws was modeled using dummy variables equal to one if the firm owning that patent was incorporated in a state in which the antitakeover legislation was enacted at that time. Doing so created a model in the form of a difference-in-difference regression, with firms incorporated in a state that had enacted antitakeover legislation as the “treatment” group

(29)

29

and firms incorporated in a state that did not had enacted antitakeover legislation at that time as the “control group”. Several control variables were added, such as measures for industry concentration, firm size, profitability and leverage. Besides that, firm-and year-fixed effects were used in order to decrease any bias due to unobservable characteristics.

The regressions were performed on four measures of explorative innovation, as proposed by Balsmeier et al. (2016): number of citations, percentage of self-citations, number of claims and an

originality index as suggested by Hall et al. (2001). The regression results provided inconclusive

statistical evidence to state that an increase in managerial entrenchment increased explorative innovation, although this was hypothesized. Even though the regression results on the dependent variables citations (table II) and self-citations (table III) pointed to some extent in the direction of exploration, there remained significant variables with opposite coefficients. This was even more obvious from the regression results of claims and originality index. Robustness checks were performed by altering the functional form and by enlarging the dataset, however the ambiguous effects remained.

Discussion

Both positive as well as negative significant coefficients were reported, indicating that the increased level of managerial entrenchment could have led to either an increase in exploration or an increase in exploitation over time. These conflicting results shows similarities with the concluding remarks of Balsmeier et al. (2016), which make use of the same dependent variables and states that true innovation can also take place within exploiting patents. He argues that besides making truly

innovation patents from exploration, this can also happen under exploitation (Lerner et al. 2011). The rationale behind this is that in exploitation, firms have more experience in the categories in which they apply for a patent, this experience causes an increase in likelihood of creating valuable patents. Following this theory, the results pointing in the direction of exploitation could still end up leading to truly innovative patents.

The reason that a causal relationship between managerial entrenchment and explorative is difficult to establish might be due to the nature of explorative patents. They are inherently more difficult to measure and more time can pass before they become measureable (Balsmeier et al. 2016). This in effect makes it even more difficult to assess whether antitakeover legislation has an effect on true innovation.

Nevertheless, a concise relationship between managerial entrenchment and explorative innovation is not evinced by this research. However, this paper has its shortcomings that further research could try

(30)

30

to improve, causing them to find a causal relationship. For instance, the reason that the results show both positive and negative results could be because of omitted variable bias. If some control

variables were unaccounted for, while having an impact on the dependent variables, this could have led to biases. The existent literature states several possible variables that this research was not able to control for, due to data limitations. For instance, Atanassov (2013) argues that the presence of large shareholders mitigates the problem of a shirking management when antitakeover legislation is enacted. This is caused by the increased level of monitoring, for which such block holders have larger incentives. The increase in monitoring could also affect the understanding of the innovative projects that firms are working on, increasing the likelihood that they will initiate such projects in line with the asymmetric information problem regarding long term projects (Stein, 1988). Also, the use of patent-based variables has its remarks since not all firms choose to apply for patents due to practical reasons or because the inventor prefers to rely on secrecy (Atanassov, 2013). Since not all

explorative inventions are accounted for this way, another measurement for explorative innovations might be better fitting.

Another suggestion for future research could be to find a measure of industry concentration that is available before 1980. Doing so enables one to control for industry concentration and make use of a longer dataset. With this dataset the Herfindahl-Hirschman Index as measures by the U.S. Census of Manufacturers is a limiting factor regarding the dataset, due to its unavailability before 1980. If another measure of industry concentration can be calculated, the effect of first-generation state antitakeover legislation on true innovation can be studied in a better way. This could be an important factor to add, since Karpoff and Witrey (2016) deemed this form of antitakeover legislation as

The effects of managerial entrenchment on exploring innovation