The Impact of Corporate Governance on
Innovation
Master’s Thesis Finance
Corporate Finance
Author:
R. Hidskes (11955686)
Thesis supervisor:
Dr. T. Caskurlu
29-06-2018, Amsterdam
Statement of originality
This document is written by Robin Hidskes who declares to take full responsibility for the content 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 were used to create it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This paper investigates whether improvements in the firm’s corporate governance have effect on innovation. Three measures of innovation are analyzed by their response to a marginally passing or marginally failing shareholder proposal. This establishes a clean causal estimate to handle the endogeneity of governance provisions. I find that a passing proposal gives a significant positive response to the number of patent applications and the market-value of firm innovation. No significant causal estimation is found on the R&D expenses for a passed proposal.
Field keywords: Corporate governance, Innovation, Shareholder Proposal, Patents, Research and Development.
Table of contents
Statement of originality ... 1
Abstract ... 2
1. Introduction ... 4
2. Theoretical Background ... 7
A. Corporate governance ... 7
B. Shareholder proposal ... 8
C. Innovation ... 9
3. Data and descriptive statistics ... 13
4. Empirical section: Hypothesis 1 ... 18
5. Empirical Section: Hypothesis 2 ... 20
A.
Voting percentage on shareholder votes... 20
B.
Fuzzy regression discontinuity design ... 21
C.
Dynamics in the impact of shareholder proposals ... 22
D.
Basic assumptions of a RDD ... 23
E.
Results ... 25
F.
Robustness... 31
6. Conclusion ... 37
7. References ... 40
8. Appendix ... 44
A.
Distribution of Votes ... 44
B.
Pre-existing differences ... 47
C.
Innovation of Passing Governance Proposals ... 49
D.
Innovation per Type of Shareholder Proposal... 50
1. Introduction
Recent literature suggests that changing the internal corporate governance is rewarded by the market and it yields long-term performance improvements (Cuñat et al. 2012). This is line with the literature
suggesting increasing managerial responsiveness to shareholder proposals and other forms of shareholder activism (Ertimur et al. 2010, Harris and Raviv 2008, Renneboog and Szilagyi 2009). However,
Bebchuck (2005) gives contrary arguments by considering shareholder governance proposals a weak form of shareholder monitoring and control. In addition, Kadyrzhanova and Rhodes-Kropf (2011) contribute to the discussion by finding a negative relation between corporate governance and shareholder value. In the midst of these discussions, however, one of the main drivers of firm profitability, innovation, has not been investigated.
In this paper, I investigate whether improvements in the firm’s internal corporate governance have an impact on the level of innovation of a firm. Innovation is measured by the R&D expenses, the number of patent filings and the market-value of firm innovation. At first, it is analyzed whether the occurrence of a shareholder proposal influences innovation. This forms the basis for my first hypothesis: the occurrence of a shareholder proposal increases the innovation in a firm. No evidence supporting the first hypothesis is found. Secondly, I study the innovation response to governance proposals passing or failing by a small margin of votes. This provides the second hypothesis: a passed shareholder proposal increases the innovation in a firm. For this second hypothesis empirical support is found. The results of the following two variables support this hypothesis, namely, the number of patent filings and market-value of firm innovation. However, the results of the R&D expenses do not support the second hypothesis.
For the first hypothesis, I use a Difference-In-Difference estimation (DID) to test whether the occurrence of a shareholder proposal has a positive significant impact on innovation. This estimation analyzes the changes in innovation behavior after the occurrence of a shareholder proposal, to their behavior beforehand. The three innovation variables are regressed over a period of one, three or five years. My sample consists of firms where at least one shareholder proposal occurred from 1997-2014. In the DID, the treatment group consists of the cumulative observations one, three or five years after a shareholder proposal occurred. The same companies are used for the control group. However, only observations prior to the occurrence are included, or observations of at least one, three or five years afterwards. I control for having multiple shareholder proposals at the firm’s annual meeting and for the unobserved, time-invariant differences across firms, industries and states.
The first hypothesis is rejected as the estimates of the DID regression give diverse results. Most estimates show insignificant coefficients and diverse signs over the period of one, three of five years. Therefore, I reject my first hypothesis. These results are in line with the findings of a paper by O’Connor and Rafferty
(2012), suggesting a neither statistically nor quantitatively significant estimate on the relation between innovation and corporate governance.
However, in the second hypothesis, I investigate the innovation response to governance proposals which pass or fail by a small margin of votes. The reason for this emphasis on close-call proposals is to
overcome the endogeneity problem of corporate governance. The essence of this methodology of Cuñat et al. (2012) is that the vote shares with an outcome around the threshold provide a random event. By including a random event in the sample, the endogeneity problem is overcome, as the outcome of a shareholder proposal is unforeseen by the market and uncorrelated with the firm characteristics. By using this method, a clean causal estimate of the relation between innovation and corporate governance is provided.
To quantify this relation, the demonstrated methodology of Cuñat et al. (2012) combines a Regression Discontinuity Design (RDD) with an event study. But, before a RDD is conducted, the identification assumption of McCrary (2008) should be met. At first, the identification assumption is tested by
examining the continuity and smoothness of the distribution of the vote shares. Secondly, the pre-existing differences in innovation and firm characteristics are compared between the firms with a passing proposal and those with a failed one. Multiple regressions are conducted and support the validity of my dataset to conduct a RDD.
My dataset includes all voted shareholder governance proposals from 1997-2014 of firms in the S&P 1500. There is a focus on these types of proposals because the management is unable to constrain these, contrary to management-supported shareholder proposals. In addition, each proposal is treated as an independent event and a dummy variable will be included to deal with multiple votes in one meeting. As stated above, the dependent variables indicating innovation are: the R&D expenses, the number of patent applications and the market-value of firm innovation. This paper compares the values of these variables the three years before a shareholder meeting occurred to the three years afterwards, while estimating the mean of change.
The results for my second hypothesis suggest that the number of patent applications and the market-value of firm innovation increases after the occurrence of a passed shareholder proposal. On average, the mean of change of the number of patent applications is 48.85% higher for passing proposals, three years after the occurrence of a voting. Although the R&D expenses show significance, the estimate does not hold for the additional robustness tests. Due to this insignificant result, the second hypothesis is not supported. However, the results of this paper support the assumption that a passed shareholder proposal increases the innovation in a firm. These results show consistency with the theory proposed by Cuñat et al. (2012).
This paper makes respective contributions to the literature. It is the first study investigating the direct relation between multiple types of innovation and shareholder governance proposals. Several papers investigated the relation between innovation and shareholder activism (David et al. 2001) or the adoption of antitakeover provisions (Meulbroek et al. 2012, Mahoney et al. 1997). These articles gave contrary conclusions of this relation. This paper fills a gap in the literature by the use of extensive methodology. At first, by conducting a clean DID estimation over a period of one, three or five years on three types of innovation. Secondly, I used the unique methodology of Cuñat et al. (2012) to overcome the endogeneity problem and provide clean causal estimates. There is dealt with endogeneity issues by using close-call proposals. As this marginal difference in the vote share causes discrete changes in managerial
responsiveness, it is of importance for the implications of its results. Hereby, this paper provides clean evidence on the relation between corporate governance and innovation.
Secondly, the three variables for innovation provide different perspectives on innovation. The R&D expenses and the number of patent applications are variables which show the investment and effort of the management, whereas the market-value of firm innovation represents the response of the stock market to public patent information. In addition, the R&D expense is a direct monetary expense, whereas the market-value of firm innovation is an indirect cost for the firm. The patent applications do demand a monetary contribution within the range of $200 - $300 for a basic application (USPTO Fee Schedule 2018), but as this sample consists of companies included in the S&P 1500, this is not interpreted as a monetary expense. I believe these diverse variables contribute to our understanding of the relation between innovation and corporate governance. Therefore, they provide an important contribution to the finance literature.
Finally, the results of this paper extend the knowledge of the current literature. The contrary arguments whether corporate governance is value-enhancing (DeAngelo and Rice 1983) or value-destroying for a firm and its shareholders (Field and Karpoff 2002) feeds the debate on the principle-agency dilemma. The results of this research support the positive relation between innovation and corporate governance. Because of this, current knowledge on innovation and corporate governance is extended.
The remainder of this article is organized as follows: Section 2 explains the theoretical background of the hypotheses. In section 3, the collected data and summary statistics are described. Section 4 discusses the used methodology, results and robustness tests. Section 5 concludes.
2. Theoretical Background
A. Corporate governance
In the previous years, a shift has been made in the ownership and control structure of firms. The concept of holding shares started in the 1960s and 1970s as a way for people to have a contractual claim to the residual value of a firm. Nowadays, it has developed to a significant monitoring and controlling role for the shareholder. This upcoming of shareholders rights and their increasing activism feed the debate called the principal-agency problem. This dilemma occurs due to divergent interest between the shareholders and management of a firm. This gives implications for the welfare of the organization and its stakeholders as it is difficult to act in the diverging interests of these parties. The role of shareholder activism is
emerging within this discussion. It is questioned whether the increased activism and rights of shareholders are damaging or contributing to firm value.
A paper by Jensen and Meckling (1976) states the agency problem should be considered when a firm’s ownership structure is determined. Especially as this problem effects the investment strategy of a firm. This led to defining the concept of corporate governane by a report of Cadbury (1992) as ‘the system by which companies are directed and controlled’. This arose to describe the role of the shareholder in relation to the responsibility of the company (Friedman 1962, Freeman 1994) and laid the foundation for various theories.
As a response to several corporate scandals in the late 1990s (e.g. Enron, Worldcom and Parmalat), the Sarbanes-Oxley Act was founded in 2002. This law established corporate governance provisions to protect different stakeholders of a firm by rules and procedures for decision making. Herewith, they tried to reinforce the public confidence. The foundation of this legislation and the preceding scandals raised the interest of the public to corporate governance. But, a prevailing law or policy to perfectly balance the control and responsibilities of these various stakeholders remains a widely debated topic.
Examples of currently practiced corporate governance provisions are: dual class shares, cumulative voting and antitakeover provisions. In addition to the agency theory, a debate on corporate governance
provisions specifically is induced. This debate on corporate governance provisions can be exemplified by the use of antitakeover provisions (ATPs), which are policies adopted as a defense against the probability of a takeover. The debate consists of two opposing arguments explaining the adoption of ATPs. The first is the argument of the management entrenchment hypothesis, stating that ATPs restrict shareholders’ rights, entrench CEOs and thus, increase a CEO its power (Field and Karpoff 2002). This theory is opposed to the shareholder wealth theory, suggesting that ATPs increase shareholder wealth by enhancing their bargaining position (DeAngelo and Rice 1983). Herewith, the CEO ownership is reduced.
Not only is the amount of literature on corporate governance increased as a response to the growing public outrage and changing ownership structure within firms (Adams et al. 2010, Yermack 2009, Bebchuk and Weisbach 2010). Additionally, the globalizing environment with more convenient and transparent communication, has enforced this trend. One of the first studies on shareholder protection by La Porta et al. (1998) found a positive relation between shareholder protection and the shareholder’s willingness to financially support a firm. This evidence provided an encouragement for public companies to increase their participation in corporate governance. Supporting the implication of this evidence, the paper of Gompers et al. (2003) suggests that corporate governance effects the capital expenditures of a firm.
Other insights are given by the many studies examining indicators for shareholder activism and their interest in increased control and rights for shareholders. According to Cai and Walkling (2011) and Cziraki et al. (2010), the main indicators of shareholder activism are the size and performance of a firm. In addition, larger firms are more likely to draw public attention, giving shareholders a higher chance to be publicly supported (Rehbein et al. 2004). The fact that shareholders are closely monitoring the
management is supported by the research of Bradley et al. (2010), who states that shareholder activism is higher for underperforming firms.
Moreover, literature identified two channels for shareholders to express their view on the decision-making of the management. One expression of shareholders is the menace of the auctioning of their shares, as this will drop the stock price. This act will negatively impact a firm’s reputation and the management’s compensation (Admati and Pfeiderer 2009). Another way for shareholders to express their discontent is by initiating a shareholder proposal to raise attention on a specific issue. In this note, my research will focus on shareholder proposals as it is an important type of governance provision to shape the firm’s decision-making (Iliev et al. 2015).
B. Shareholder proposal
A shareholder proposal is a way for shareholders, with a share in a public firm, to submit proposals on topics concerning social policy implications (Mathiasen 1994). These proposals will be voted on at the corporation’s annual meeting. For a long time, this mechanism was considered a weak form of
shareholder activism (Black 1990, Bebchuk 2005). This proposition is supported by the studies of Gillan and Starks (2000) and Prevost and Rao (2000), arguing that shareholder proposals have no real
controlling benefits due to their non-binding nature. In addition, Anabtawi (2006) and Bainbridge (2006) argue a shareholder proposal aims to either disrupt the board’s authority unnecessarily, or to act
completely in own interest. Opposed to these views, is the article of Ertimur et al. (2010) who suggest that there is increasing responsiveness of the management to non-binding, majority-vote shareholder
proposals. In line with this statement are the studies of Harris and Raviv (2008) and Renneboog and Szilagyi (2009), who regard shareholder-initiated proxy proposals as a useful tool of corporate
governance for monitoring. The argument addressing the non-binding environment should be considered as the U.S. addresses a non-binding environment, whereas a legally binding shareholder environment is hold in the United Kingdom and most of Europe.
Other important considerations are the requirements for the proponents to impose a proposal. Firstly, the shareholder must own at least $2,000 or 1% of securities entitled to vote on the proposal. It should be submitted in a period within a minimum of 120 days before the date of the company’s annual meeting. The process is regulated by the Securities and Exchange Commission (SEC), which adopted rule 14a8 in 2002 to control the proposals submitted by the proponents.
The proponents of shareholder proposals show a wide variety of individuals, hedge funds, pension funds or other institutional investors. Many studies are provided on the impact of these differing proponents on a firm’s response and the reaction of the public. A study by Cuñat et al. (2012) finds that proposals imposed by institutional activists have a significant impact of 2.1% on the abnormal returns at the day of the vote, whereas individual activists give an insignificant effect of 0.9%. A study by Karpoff et al. (1996) argues that public firms which are larger in size, less profitable and with a relatively high share of institutional ownership are more often the victim of a proposal submitted by an activist.
The literature on the impact of increased shareholder rights varies widely, as some results emphasize the negative effects of increased shareholder rights, whereas others state a significant positive impact on firm performance. Cuñat et al (2012) finds that passing a proposal and adopting one increases the
shareholder’s value by 2.8%. In addition, results highlight how shareholder proposals have a positive effect on deal quality and are able to alleviate the investment decisions of a firm (Kai et al. 2017). Moreover, evidence of Kai et al. (2017) suggests that shareholder voting alleviates the principle-agency dilemma.
C. Innovation
In the pursuit of growth and welfare, companies are challenged by the decision to invest in innovation activities. Overall, innovation is considered an important driver for economic growth and profitability on a firm-level (Aghion and Howitt 2006, Jaffe 1986). Moreover, innovation received great attention in the literature, due to the quick progress of technological change over the past years. A study of Rubera & Kirca (2017) evaluated the role of innovation on the firm’s welfare and broadens the concept of
of the R&D expenses, number of patent applications, number of granted patents or patent citations are used. An additional measure is the market-value of firm innovation found by Kogan et al. (2010). The current knowledge on innovation suggests that firms more likely to be innovative are larger in size and score higher on the Tobin’s Q ratio (Baumol 2002, Griliches 1990, Scherer 1965, Scherer 1983). Another study of Ferreira et al (2012) and Bernstein (2015) proves that innovation is lower for publicly owned firms relative to private firms.
But, literature on corporate governance and its relation to innovation is scarce. Nevertheless, it is found in a study of David et al. (2001) that shareholder activism increases a firm’s R&D expenses, but not their patents and citations. Moreover, it is examined how innovation is affected by internal mechanisms as managerial contracts (Manso 2011) or external mechanisms as the threat of takeover (Stein 1988). Another form of internal mechanism is examined by Aghion et al (2008), predicting that higher
institutional ownership within a firm has a complementary effect on its level of innovation. Eventually, a study by Sapra et al. (2008) integrated both internal and external governance mechanisms and
demonstrated a U-shaped relation between innovation and the threat of takeover. This finding adds value to the ongoing debate on the market for corporate control in relation to innovation.
On the one hand, it is argued by Jensen (1988) that laws hindering the facilitation of takeovers (e.g. antitakeover provisions) stimulate managerial slack and refrain investment. This argument is in line with the managerial myopia view. This view argues that managers favor short-term profits at the expense of term investment. However, opposing literature proves how antitakeover provisions alleviate long-term contracting, stimulate innovation (Shleifer and Summers 1988) and motivate longstanding
investments (Stein 1988). These opposing studies strengthen the signaling hypothesis proposed by Spence (2002).
The fundament underlying this theory is that management is driven to send a positive signal to the public and other stakeholders to overcome the agency dilemma. A demonstration of this theory is a company which chooses to exorbitantly provide financial support to marketing, not because it is the most efficient investment, but to signal that the company has a good prospect and management has faith in the firm. A study of Lahr and Mina (2016) provides linkage of the signaling theory to innovation, by proving that investment of venture capitalists is based on the number of a firm’s patents. It is found that the company’s innovation influences the probability of receiving investment and thereby it proves innovation signals firm quality (Baum and Silverman 2004, Mann and Sager 2007, Häussler et al. 2012, Audretsch et al. 2012, Conti et al. 2013a,b, Hsu and Ziedonis 2013). This provides knowledge on the relation between corporate governance and innovation. Moreover, these theories claim a higher level of innovation for
firms with more shareholder activism. This, as it is important for the firm’s profitability to send out a positive signal.
Contrary to the view of the signaling hypothesis is the literature of O’Connor and Rafferty (2012), proposing that corporate governance has a minor insignificant effect on the level of R&D expenditures. This view is consistent with the papers of Meulbroek et al. (1990) and Mahoney et al. (1997), which examine the impact of the adoption of antitakeover provisions on a firm’s long-term investment.
According to these studies, the R&D expenses and capital expenditures decline after a firm incorporates antitakeover defenses. Another view on this discussion is given by a study of Atanassov (2013),
examining whether reducing the probability of a takeover has impact on management myopia. The author argues that when the threat of takeover diminishes, innovation reduces as the managerial slack increases. This view supports the management entrenchment hypothesis. This view argues that managements’ self-interest increases with the number of corporate governance provisions in a firm. Eventually, it alters the investment strategy of a firm and harms its shareholders (Walkling and Long 1984).
The literature on the relation between corporate governance and innovation is widely interpretable. Nevertheless, this study expects a positive relation between increasing shareholder activism and innovation. Firstly, there is the upcoming of shareholder activism and the pressure on public firms for growth. At second, shareholders provide financial support to public firms and better protected firms are higher valued (Porta et al. 2002). Moreover, increasing responsiveness of the management, especially to more salient shareholders, is supported by studies of Mitchell et al. (1997) and David et al. (2007). Several studies claim that the level of institutional investors is positively related to the innovation within a firm (Graves 1988, Baysinger et al. 1991, Hansen and Hill 1991, Wahal and McConnell 2000). As a shareholder proposal is regard as useful tool for shareholders to support a public firm (Harris and Raviv 2008, Renneboog and Szilagyi 2009), a positive relation with innovation is expected. This expectation supports the signaling theory and the theory of shareholder wealth.
In summary, there are not many studies specifically focused on the impact of shareholder proposals on innovation. However, the majority of the literature signals a positive relation between shareholder proposals and innovation. The upcoming of shareholder proposals and management responsiveness leads to the following hypothesis:
Hypothesis 1. The occurrence of a shareholder proposal increases the innovation in a firm.
As stated earlier, there is a different role for shareholder-initiated proposals in European firms and those addressed in the U.S.. If an initiated shareholder proposal passes the shareholder voting in Europe, the outcome is legally binding and implementation is mandatory. But, if this would occur in a firm located in
the U.S., the outcome is non-binding. The choice whether or not the shareholder proposal should be implemented depends on the management. It is proved that approximately 40% of the proposals passing in the non-binding environment are implemented (Bizjak and Marquette 1998, Martin and Thomas 1999, Thomas and Cotter 2007, Ertimur et al. 2010). In addition, submitted shareholder proposals in the U.S. which were ignored by the management, caused shareholders to penalize these firms by negative public attention (Ertimur et al. 2012). Moreover, shareholder proposals held in the U.S., cause increases in the stock price around the dates of announcement (Bhagat 1983, Bhagat and Brickley 1984). Furthermore, the voting successes and percentages are significantly lower in the U.K. (Cziraki et al. 2009). Based on the literature, I believe that if a shareholder proposal is proposed in a non-binding environment, the managerial responsiveness is higher. Secondly, I believe that the shareholder proposals are positively received by the public, based on the increasing stock price and vote percentages. In the non-binding environment, submitted proposals receive more attention. In addition, if a shareholder proposes a voting within the non-binding environment, uncertain about the outcome and implementation, more
dissatisfaction with the current managerial’ behavior is found.
As shown by the studies of Ertimur et al. (2010) and Thomas and Cotter (2007), the rising number of shareholder proposal have increased awareness. They argue that the vote share of a shareholder proposal is important, especially if the vote share passed the majority threshold. Based on these theories and the assumption of hypothesis 1, I expect the following hypothesis:
3. Data and descriptive statistics
This section presents the sample and its characteristics. One of the most important variables within this study is a shareholder proposal of which data is collected of the ISS Governance Database from 1997 to 2014 for companies of the S&P 1500. At first, shareholder proposals which have been withdrawn or omitted, so where eventually no shareholder voting has taken place, are excluded. In addition, the sample is restricted to shareholder proposals solely as the management is not able to omit shareholder-sponsored proposals. Moreover, as the dependent variable is innovation, firms which have never filed for patents or invested in R&D, are excluded. Therefore, companies within the financial industry (SIC code 6000-6999) and the utility industry (SIC code 4900–4949) are omitted. Eventually, the sample size consists of 4,043 governance-related proposals. The ISS Governance Database includes information on the companies’ identification number, the date of the annual meeting, the percentage of votes in favor of the proposal, the purpose of the proposal and its subscriber. All the proposals head towards increased authority and
governance for shareholders. However, the reasoning varies and therefore the proposals are divided over six categories. These categories are proposed by previous literature and include; auditors, board,
compensation, voting, G-index (governance-related) and other. In table I the frequency of governance proposals per year from 1997-2014 are displayed. In addition, the percentage in favor and the average is included. Noticeable is the fact that after 2002 there is a huge increase in the number of proposals. From 2002 onwards, the percentage of approved proposals increased causing an increase in the average vote outcome. As per July 2002 the SEC adopted Rule 14a-8, it is very likely this event is related to the significant jump in the number of shareholder proposals in 2003 and the following years. Overall, the average outcome on shareholder proposals increased over time. Panel B displays the various types of shareholder governance proposals and shows that G-index proposals occurred most. This category also scored the highest mean in favor and the majority of its proposals were approved of. The G-index is founded by Gompers et al. in 2003 to classify governance provisions and anti-takeover laws. This category consists of descriptions as the abolition of poison pills, repellence of a classified board and the adoption of an anti-greenmail. The lowest percentage of approved proposals is the Voting category. Therefore, it is expected that only a few observations of this category will be used for the empirical analysis as this paper is mainly interested in votes around the discontinuity. For the empirical analysis, the sample will mainly consist of shareholder proposals which fell within the G-index- or Other category.
Table I.
Shareholder Governance Proposals
The data of this table is collected by the ISS Governance Database on all shareholder-sponsored governance proposals from 1997 to 2014 for the S&P 1500 companies. In panel A, the prevalence, (percentage of) approved proposals and (percentage of) the average vote share over time is presented. The threshold for approval is 50% for all, except for 30 observations with a threshold of 66.7%, 70% or 80%. The various thresholds are taken in account for computing the (percentage of) approved proposals. In panel B, the governance proposals and their prevalence and vote share are classified by type.
Panel A: Shareholder Proposal Summary Statistics
Year Shareholder Proposals Approved Proposals Percentage Approved
Proposals Average Outcome
Standard Deviation of Vote Outcome 1997 205 33 16.10% 25.56% 19.58 1998 117 16 13.68% 24.26% 20.01 1999 61 8 13.11% 21.87% 18.75 2000 198 53 26.77% 30.01% 22.32 2001 205 53 25.85% 30.53% 22.97 2002 219 82 37.44% 38.75% 23.86 2003 358 135 37.41% 38.25% 23.82 2004 291 81 27.84% 32.78% 25.71 2005 285 84 29.47% 36.68% 24.72 2006 263 82 31.18% 39.84% 22.79 2007 2008 2009 2010 2011 2012 2013 2014 274 235 294 242 164 217 227 188 62 65 101 70 49 76 60 50 22.63% 27.66% 34.35% 28.93% 29.88% 35.02% 26.43% 26.60% 36.45% 41.35% 43.18% 41.16% 43.09% 45.79% 41.43% 41.87% 21.07 22.14 20.97 20.42 20.90 24.84 23.07 23.90 Total 4,043 1,160 27.26% 31.95%
Panel B: Type of Shareholder Proposals – Summary Statistics
Proposal Type Number of Proposals Mean Vote in Favor Percentage Approved
Auditors 47 23.20% 8.51% Board 758 38.97% 26.91% Compensation 1,136 26.02% 9.77% G-Index 1,321 49.18% 49.66% Voting 321 15.03% 3.74% Other 459 45.48% 37.69%
For the 728 firms where at least one shareholder proposal occurred from 1997 - 2014, additional
information is derived from the Compustat database. These 728 proposals correspond to a total of 89,826 firm-year observations. In Table II the descriptive statistics are provided of the innovation values and firm characteristics in the year in which the shareholder voting took place (t).
Table II.
Descriptive Statistics
The sample of 4,043 shareholder proposals resembles a total of 728 firms. In this table the summary statistics of the variables representing innovation and the firm characteristics are represented in the year of the shareholder proposal (t). The numbers of observations change due to missing values of some variables. However, the negative values of the innovation variables are replaced by 0. Panel A represents the summary statistics of the innovation variables derived from Compustat, USPTO and United States Patent Data. R&D/Assets is retrieved from Compustat and calculated by dividing the R&D expenses by its total assets (XRD/AT). The patent filings and market-value firm innovation are obtained from USPTO and the United States Patent Data, both are adjusted by taking the logarithm. The accounting variables are derived from Compustat: Total Assets (AT), Market Value (mkvalt_f), EBITDA (ebitda) and Capital Expenses (CAPX). Furthermore, Tobin’s Q is calculated by taking the market value of assets (AT+mkvalt_f-CEQ) divided by the book value of assets (AT) and the balance sheet Deferred Taxes and Investment Tax Credit (TXDITC). The book to market ratio is the book value of common equity of previous year to the market value of previous calendar year. The return on equity is derived by dividing the net income over common
equity(NI/CEQ) and the return on assets by dividing the net income over total assets (NI/AT). Debt to assets is calculated by dividing total debt by total equity (DT/CEQ). All monetary values are in U.S dollars. All monetary variables are winsorized at 1% and 99%.
Panel A: Innovation Summary Statistics
N Mean Std. Dev. 10th Per. Median 90th Per.
R&D /Assets 2,436 0.0339 0.0465 0 0.0201 0.0914
Log (1 + Patent filings) Log (1 + Market-value firm innovation) 1,772 1,965 3.8709 9.6877 2.0154 0.4098 1.0986 9.0461 3.8712 9.8035 6.3869 10.1371 Panel B: Firm characteristics
N Mean Std. Dev. 10th Per. Median 90th Per.
Total assets ($millions) 3,666 57,509.34 116,792.10 1,566.04 17,547 165,968
Market value ($millions) 3,446 92,291.67 161,628.70 2,984.72 29,822.60 267,180
EBITDA ($millions) 3,664 7,014.07 11,963.93 119,358 2,366 22,200
Capital expenses ($millions) 3,655 2,783.94 5,359.72 41,78 717 9,355
Tobin’s Q 3,132 1.8110 1.101 0.9832 1.4775 3.0126 Book to market 3,446 0.4419 0.4107 0.1095 0.3665 0.8681 Return on equity 3,666 0.1524 0.6279 -0.7215 0.1412 0.3670 Return on assets Debt to assets 3,666 3,332 0.0450 0.9608 0.0983 2.7154 -0.0219 0.0154 0.0497 0.5868 0.1295 2.3491
Innovation is the dependent variable of this study and measured by three different variables. The first variable is the yearly expense on research & development and is scaled by dividing it by the firm’s total assets. The R&D expenses and other financial information are retrieved from Compustat from 1992 to 2017. The second input measure of innovation is the number of patent filings. This is calculated by taking the sum of the number of patent applications by a firm within a year. As this is a count variable, 1 is added and the logarithm is taken, giving log (1 + Patent filings). By taking the logarithm, the variable is normalized and the 1 is added to prevent the value to be 0.
The third measure of innovation is the market-value firm innovation, an output measure of innovation. This metric was founded by Kogan et al. (2010) and is proposed as a measure of the economic importance of each innovation as it is based on the response of the stock market to the information on the newly
granted patents by the USPTO. The value is calculated by taking the market price of the value of the granted patents for firm f within year t, according to the stock market response,
(1)
where sm represents the stock market, j the values of the patents granted and 𝑃𝑓, 𝑡 the number of patents
issued to firm f in year t. Eventually, this metric is scaled by the firm size and this leads to the following equation,
(2)
𝐵𝑓𝑡 represents the book value of assets for firm f in year t and m is based on the stock markets’ patent
value (sm). As the market-value of firm innovation gives extreme values, the logarithm is taken. To prevent a value of 0, 1 is added, giving: log(1 + Market-value firm innovation).
The latter two variables are retrieved from the United States Patent Data which were collected for the paper of Kogan et al. (2010). As this dataset is only available until 2010, data on these variables is retrieved from 1992 to 2010. Besides the United States Patent Data, information on these two variables is obtained from the USPTO patent-level database. This is necessary to link the companies’ identification numbers, patent numbers and corresponding data correctly. Panel A in Table II provides information on these three innovation variables and their number of observations, mean, standard deviation, 10th, 50th (median) and
90th percentile. The variables are winsorized at 1% and 99% to minimize the impact of outliers.
In panel B, the financial characteristics of the firms in my sample are displayed. The choice of the firm characteristics displayed within this panel is based on previous literature on firm characteristics and innovation. At first, measurements of firm size as: total assets, book-to-market value and capital expenses, are used. This, as a study by Hausman et al. (1984) proposed that larger firms innovate relatively more. In addition, performance is considered and displayed by the market value, EBITDA, Tobin’s Q, return on equity and return on assets. Furthermore, Stein (2003) discussed that firms with a higher ratio of debt to
assets invest less. Another factor which could affect innovation is competition, based on a study by Aghion et al. (2005). This variable is not presented in Table II as there is controlled for by the industry-by-year fixed effects based on the SIC codes.
To merge the various datasets, the CRSP database and CRSP-Compustat merged database were used to combine the United States Patent Data, USPTO, ISS Governance database and Compustat. This, as the various identification codes change over time, making merging more problematic. As the CRSP and CRSP-Compustat merged database provide historical identification codes, a solution is provided to deal with the changing and different identifiers of each database.
For further regressions on the dependent variables, the difference in innovation for period t of firm f is measured. Thus, the dependent variables of innovation present the change in mean of the period before (t-1, t-3, t-5) and after (t+(t-1, t+3, t+5) the shareholder voting took place. Thus, by subtracting the mean of the period before the shareholder proposal of the mean of the period afterwards, the mean of change is calculated.
4. Empirical section: Hypothesis 1
As the first hypothesis is considered a preparation and endorsement for the second hypothesis, the empirical analysis on this section is briefly discussed. The assumption made in the first hypothesis is based on the direction of the current literature and I will test whether the occurrence of a shareholder proposal has a significant impact on innovation. To investigate this effect, a Difference-In-Difference estimation (DID) is used. This estimation strategy will consider the changes in the behavior of firms after the occurrence of a shareholder proposal for a period of one, three or five years to their behavior prior to the shareholder proposal. The treatment group consists of observations of the period (t+1, t+3, t+5) after a shareholder proposal occurred. The control group uses observations of the same firms, however, solely the years prior to the shareholder proposal are included and the observations at least a period (t+1, t+3, t+5) after a shareholder proposal occurred. If multiple shareholder proposals occur, all observations until the period after the last shareholder proposal took place, is taken as treatment.
I control for the unobserved, time-invariant differences across firms and industries. In addition, there are tests to diminish the chance to interpret the results alternatively. At first, there are no measurable
differences in the firm characteristics of the treatment and control group. As identical firms function as a control group, this assumption is derived. Secondly, by controlling for industry-by-year and firm fixed effects, there is controlled for economic changes or external events. Nonetheless, as my sample consists of firmsincorporated in the U.S., different policies can be adopted or modified over time per state. Therefore, there is controlled for the location state-by-year fixed effect. The incorporation of a firm’s location is used to control for this fixed effect, although I am aware that not all firms are located in their state of incorporation. Finally, no significant differences are found in the trends of the dependent
variables before the occurrence of a potential shareholder proposal between the two groups. These results suggest that the control group is valid as a counterfactual for the treatment group and findings are not explained by endogeneity. Table III shows results on the impact of the three variables on innovation for a period of one, three or five years after the occurrence of a shareholder proposal.
Table III
Effects of Shareholder Governance Proposals on Innovation
This table reports estimates from firm-level regressions of measures of innovation. It shows whether a firm increases its innovation after a shareholder proposal (t) occurred. Only firms which ever voted on a shareholder meeting are included. The dependent variables are: R&D/Assets (Column 1), log (1+ Patent filing) in column 2 and the log (1 + Market-value firm innovation) in column 3. All values are relative to firm size. The cumulative years after the occurrence of a shareholder proposal (t) are summed. For example, (t+5) refers to the values from the period between (t) and (t+5). Each row resembles a different dependent variable and each entry is separately regressed. The sample includes firm-year observations from 1992 to 2017. There is controlled for firm fixed effects (FE), state-of-location-by-year FE, and four-digit standard industrial classification industry-by-year FE. The standard errors (in parentheses) are clustered at the firm level. Significance at the 10%, 5% and 1% levels is indicated by *,**, and ***, respectively.
(1) R&D/Assets
(2)
Log(1 + Patent filings)
(3)
Log(1 + Market-value firm innovation) Shareholder proposal, (t+5) -0.0005 (0.0005) 0.0457 (0.0662) 0.0250*** (0.008) Shareholder proposal, (t+3) -0.0021 (0.0034) -0.0053 (0.1985) 0.0558 (0.0435) Shareholder proposal, (t+1) -0.0004 (0.0004) -0.0442* (0.0267) -0.0006 (0.0068) Observations 58,610 37,893 39,062 R² 0.7183 0.7560 0.6388
The first column on R&D expenses does not report any significant coefficients for the rows presenting the periods of one, three and five years after the shareholder proposal occurred. These results indicate there is no significant change in the R&D expenses after the occurrence of a shareholder voting. The regressions in column 1 all give negative coefficients and the coefficient of determination, reflecting the goodness of fit between the model and the data, is acceptable. In column 2, the impact of a shareholder voting on the amount of patent applications is explored. The row displaying the impact one year after the event, gives an estimate of -0.0422, which is 4.22% and statistically significant at the 10% level. This estimate indicates that the year after a shareholder proposal occurs, the number of patent filings decrease by 4.22%. This result does not support my first hypothesis. In column 3, my coefficient of interest for the period of five years is 2.5% and statistically significant at the 1% level. This means that, after the shareholder voting, the market value of firm innovation increases by 2.5% over a period of five years. Contrary to the significant estimate in column 2, this evidence provides support for my first hypothesis. Overall, I conclude that the estimates in Table III do not have evidential power to support the first hypothesis. As the regressions give various results on the relation between innovation and corporate governance, this assumption is not supported. Therefore, I reject the assumption made in hypothesis 1. Although my first hypothesis could have provided an endorsement for the following hypothesis, I believe its rejection does not provide a conclusive answer on the second hypothesis. Though, the results of hypothesis one are taken in account when interpreting the results of the second hypothesis.
5. Empirical Section: Hypothesis 2
This section discusses how to use a fuzzy regression discontinuity design (RDD) for an event study. I use this test to estimate the impact of a passed shareholder proposal on the innovation of a firm. Previous literature often uses an OLS regression and includes various control variables for the estimation. However, these regressions do not deal with the endogeneity problem of governance structures and the adoption of shareholder provisions. This RDD will provide a clean estimate of the impact of a shareholder proposal, by testing the impact on close-call proposals. This research will apply the empirical dynamic regression discontinuity model of Cellini et al. (2010) and combine this with an event study as presented in the paper of Cuñat et al. (2012).
A. Voting percentage on shareholder votes
First, a framework is presented to display how to extract the value of a shareholder proposal by only including shareholder proposals with a vote share close to the cutoff point. At the annual meeting, shareholders are voting on a shareholder proposal of firm f at time t. The total vote share this proposal receives (vft) and its threshold (v*) determine whether a proposal passes or fails. This is indicated by,
𝐷𝑓𝑡= 1(𝑣𝑓𝑡≥ 𝑣∗) (3)
The significant impact of a passed shareholder proposal compared to a failed shareholder proposal is tested in this research. The definition of a passed shareholder proposal is that the voting percentage is over the 50% (𝑣𝑓𝑡 ≥ 𝑣∗), if the cutoff point is 50%. If so, a failed shareholder proposal is one in which the vote share
is beneath 50%. In my sample, 30 observations have a different cutoff point than 50% and this is taken in account.
The purpose is to discover the abnormal impact of a passing proposal on the outcome variable, innovation (yft) which gives,
𝑦𝑓𝑡 = 𝑘 + 𝐷𝑓𝑡𝜃 + 𝑢𝑓𝑡 (4)
The coefficient of interest is 𝜃 and this reflects the impact of a passing proposal on the outcome variable – for example, the logarithm of the number of patents filings. The term 𝑢𝑓𝑡 includes all other determinants
of the outcome (E(𝑢𝑓𝑡) = 0). However, one of the main problems in corporate finance is the endogeneity
problem and it is very likely of 𝐷𝑓𝑡 to be related to the error term (𝐸(𝐷𝑓𝑡, 𝑈𝑓𝑡) ≠ 0). Thus, it is expected that the coefficient of interest (𝜃) is biased.
Therefore, a method proposed in the paper of Cuñat et al. (2012) is used to deal with the endogeneity problem on the voting percentage. This methodology creates an exogenous random variable by only taking the vote shares lying in an arbitrarily small interval around the threshold. The intention behind this is that the outcome of these vote shares, which nearly passed or failed, is seen as a random event and
uncorrelated with any firm characteristics. Moreover, based on the article of Ertimur et al. (2010) the difference between slightly passing or failing, cause a discrete change in the probability of
implementation. By using this clean estimate to deal with the endogeneity problem, the estimates are better able to test the unbiased responsiveness of the management and stock market in innovation. In addition, according to the article by Cuñat et al. (2012) there is a different impact on the responsiveness of the shareholders and management, if the voting outcome is already known. In a shareholder voting with a percentage of votes in favor of 85%, it is more likely the outcome was already known, than in one with votes in favor of 55%.
B. Fuzzy regression discontinuity design
Since the RDD methodology was proposed in an article of Thistlethwaite and Campbell in 1960, it is widely used. This is mainly due to the fact that it determines the assignment to the treatment group, according to a rule or other forcing variable. In my study this is the voting percentage. Eventually, this leads to a flexible and robust estimation of the impact of a shareholder proposal. This design compares the responses of the control group to those of the treatment group. Lee (2008) argues that adding a random component to the voting creates two random samples of a treatment (pass and Dft = 1) and control group (fails and Dft = 0). Due to the non-binding environment there is no clear cut-off point of receiving treatment or not. Therefore, a fuzzy RDD is used, as the threshold does not clearly determine treatment exposure or not. But, the passing or failing of a shareholder proposal may create a discontinuity in the probability of treatment. It is more or less an encouragement to receive treatment, instead of a clear cutoff.
An important precaution for a RDD is that the voting percentage should be continuous. The discontinuous effect around the cutoff point of the dependent variable 𝜃 is the effect of interest. In addition, a polynomial is added to obtain the underlying relation between 𝑦𝑓𝑡 and 𝑣𝑓𝑡 .
A polynomial for observations is included on both sides of the threshold and this provides a new equation, namely, 𝑌𝑓𝑡 = 𝐷𝑓𝑡𝜃 + 𝑃𝑟(𝑣𝑓𝑡, 𝛾 𝑟) + 𝑃 𝑙(𝑣𝑓𝑡, 𝛾 𝑙) + 𝑢 𝑓𝑡 (5)
C. Dynamics in the impact of shareholder proposals
For this test, it is important to consider that the occurring shareholder proposal has impact on the years following the event. It is decided to take a range of three years around the event to provide a valid measurement of innovation. Thus, the mean of innovation three years (t+1, t+2, t+3) after the occurrence of a shareholder proposal, is compared to the mean of innovation three years prior to the event (t+1, t+2, t+3). Another feature which should be considered is the occurrence of multiple votes within one meeting or within the three following years. This is resolved by treating each shareholder proposal independently. The framework of a dynamic regression discontinuity design of Cellini et al. (2010) is followed. In addition, year fixed effects are included to control for different time periods. As the firms within this dataset only have one shareholder meeting per year, there is no need to add fixed effects for the meeting date. This gives the following equation,
𝑌𝑓,𝑡+𝜏= 𝐷𝑓𝑡𝜃𝜏 + 𝑃𝑟(𝑣𝑓𝑡, 𝛾 𝑟) + 𝑃 𝑙(𝑣𝑓𝑡, 𝛾 𝑙) + 𝑢 𝑓𝑡, 𝑡 + 𝜏 (6)
This equation presents the dynamic structure of the regression and a polynomial of order four is used on both sides of the cutoff point. The aim of this research is to calculate the coefficient 𝜃, measuring the average causal effect of a shareholder proposal at time (t) on outcomes (𝑡 + 𝜏). The terms of Intent to Treat (ITT) and Treatment On the Treated (TOT) designed by Cellini et al. (2010) are interchangeable as there is one shareholder proposal per year and no other proposal occurring after (t).
D. Basic assumptions of a RDD
At first, there is tested whether the vote shares of the shareholder proposals are continuous around the cutoff point. The essence is to test the validity and generality of my approach. A histogram on either the distributions of vote shares of G-index proposals and all non G-index or Other proposals (Auditors, Board, Compensation, Voting and Other) are presented in Figure I.
Figure I. Distribution of the vote share in favor for G-index proposals and all the other proposals. The left histogram
represents the G-index proposals (N = 1,321) and the right histogram represents the Other shareholder proposals (N = 2,722) from 1997 to 2014.
The formal density test for smoothness on the vote share is presented in Figure I. and demonstrates a smooth distribution. Only a slight discontinuity in the vote share for the G-index proposals is shown. This should be considered for further tests, nonetheless, additional tests are performed to check whether the identification assumption of McCrary (2008) holds. Another conclusion extracted from Figure I. is that there are more G-index proposals having a vote share around the discontinuity. As this research focuses on observations closely around the cutoff point, the G-index proposals will mostly influence the outcome. Additional preparational tests on the smoothness of the vote share are shown in Appendix A. by figures IA.1. – IA.5.
Table IV.
Pre-existing Differences in Firm Characteristics as a Function of a Passed or Failed Vote Outcome
This table indicates whether passing or failing shareholder proposals are consistently corresponding to firm characteristics prior to the voting. In panel A, the variables representing innovation are tested and in panel B other firm characteristics. The cumulative years after the occurrence of a shareholder proposal (t) are summed. For example, (t-5) refers to the values from the period between (t-5) and (t). Each row resembles a different dependent variable and each entry is separately regressed. In column 1, there has not been controlled for a polynomial in the vote percentage, so it examines the average impact of a passed proposal to a failed proposal. In column 2, a polynomial of order four is included on the vote percentage on both sides of the cutoff point, thus, it examines the impact around the threshold. There is controlled for year fixed effects and standard errors (in parentheses) are clustered at the firm level. Significance at the 10%, 5% and 1% levels is indicated by *,**, and ***, respectively.
Panel A: Innovation (1) (2) R&D/Assets, (t-5) R&D/Assets, (t-3) R&D/Assets, (t-1) 0.0038 (0.0036) 0.0041 (0.0035) 0.0044 (0.0034) 0.0030 (0.0036) 0.0032 (0.0035) 0.0035 (0.0035)
Log(1 + Patent filings), (t-5) -0.5122**
(0.2393)
-0.4750** (0.2335) Log(1 + Patent filings), (t-3)
Log(1 + Patent filings), (t-1)
-0.4839* (0.2478) -0.4352* (0.2543) 0.4293* (0.244) -0.3551 (0.2536) Log(1 + Market-value firm innovation), (t-5)
Log(1 + Market-value firm innovation), (t-3) Log(1 + Market-value firm innovation), (t-1)
0.0082 (0.0184) 0.0022 (0.0221) -0.0025 (0.0271) 0.0088 (0.0188) 0.0039 (0.0232) 0.0073 (0.0286)
Panel B: Firm characteristics (1) (2)
Tobin’s Q, (t-3) 0.0058 (0.084) -0.0087 (0.0825) Capital expenses/Assets, (t-3) 0.0040 (0.0034) 0.0041 (0.0034) Return on equity, (t-3) Debt/Assets, (t-3) -0.0275 (0.0212) -0.2347 (0.1661) -0.0227 (0.0211) -0.2048 (0.1458)
Polynomial in the vote share no yes
Secondly, there needs to be evaluated whether prior to the year of the shareholder proposal (t), there were significant changes in the characteristics of firms which pass or fail a shareholder proposal. The assumption is that firms marginally passing, relative to those marginally failing, should not have systematic differences in firm characteristics. In panel A of Table IV it is measured whether, prior to the shareholder proposal, the firms passing a proposal score significantly different on the innovation variables than the firms rejecting a
proposal. In column 1, a polynomial on the vote percentage is not included and examines the average difference in innovation before the proposal. The whole sample is regressed. In column 2, a polynomial on the vote percentage of order four is included and examines the impact of the underlying relation. There can be concluded that there are no significant differences for the five, three and one year(s) before the
occurrence of a shareholder proposal in the R&D expenses and market-value of firm innovation. However, the number of patent filings does present a significant difference in the period before a proposal. Column 1 shows significance on a 5% level in the five years prior and significance at 10% is given for the three and one year(s) prior to the shareholder meeting. However, by including a polynomial in the vote share of order four, the significance level diminishes on the 1 year period prior to the meeting. But as the period of three years is the focus of this hypothesis, it should be considered there is a pre-existing difference in the logarithm of patent applications, between failing or passing shareholder proposals at a 10% significance level. This evidence provides a threat to the identification strategy for a RDD and should be weighed when further conducting tests and interpreting results. One of the reasons this specific methodology is used, is to overcome endogeneity and the possible correlation in firm characteristics and votes received. In Appendix B. Table IB.2. presents the insignificant results on the logarithm of patent applications for one, three and five years before the occurrence of a shareholder proposal. This table includes a sample of only the vote shares ranging from 40% to 60%. This table provides support that the identification assumption of the close-call proposals is hold.
In panel B, other firm characteristics as Tobin’s Q, capital expenditures, return on equity and debt to assets are measured in the years before the meeting. There are no significant differences witnessed on these firm characteristics, neither in column 1 nor column 2. Additional tests on pre-existing differences are derived in Appendix B. and provide evidence that there is no prior selection on the firms passing or failing a proposal. I conclude that my estimates support the identification assumption.
E.
Results
Table V presents the difference in the innovation between firms where a proposal passes to one where it fails. Innovation is expressed by R&D expenses, the number of patent filings and the market-value of firm innovation. The dependent variable gives the difference in innovation between the period prior to the shareholder proposal and the period afterwards. The difference is composed by subtracting the mean of the innovation variable the three years prior to the event from the mean the three years after the event. Due to the endogeneity problem, the emphasis is on the sample at the smallest interval around the threshold.
Table V
Innovation around the Majority Threshold
This table gives the results on the difference in innovation between passed proposals and failed proposals. The difference is calculated by subtracting the mean for the 3 years before (t-1, t-2, t-3) to the 3 years after the meeting takes place (t-1, t-2, t-3). The dependent variables are: R&D/Assets, log(1+ Patent filings) and the log(1 + Market-value firm innovation). All Market-values are relative to firm size. In column 1, the estimates are based on the whole sample. In column 2, the sample is restricted to nonclose votes beyond 10 points around the cutoff point (0-40% and 60-100%). Column 3 represents the sample restricted to close votes within 10 points around the cutoff point (40-60%). In column 4, the sample is restricted to the close votes within 5 points around the threshold (45-55%), and so forth. Column 7 introduces a polynomial in the vote share of order four, one on each side of the threshold, and uses the full sample. All columns control for year fixed effects; standard errors (in the parentheses) are clustered by firm. Significance at the 10%, 5%, and 1% levels is indicated by *, **, and ***, respectively. Note that the number of observations may change due to missing or incomplete values for some of the variables.
Panel A: R&D/Assets (1) All Votes (2) Nonclose (3) 40-60% (4) 45-55% (5) 48-52% (6) 49-51% (7) Full Model Pass Observations R² 0.0036* (0.0021) 2,193 0.0052 0.0036 (0.0046) 1,570 0.0098 -0.0026 (0.0036) 623 0.0504 0.0006 (0.0052) 314 0.1019 -0.0090 (0.0093) 139 0.1997 -0.0095 (0.0139) 80 0.2613 -0.0049 (0.0044) 2,193 0.0366 Panel B: Log(1 + Patent Filings)
(1) All Votes (2) Nonclose (3) 40-60% (4) 45-55% (5) 48-52% (6) 49-51% (7) Full Model Pass Observations R² 0.8523 *** (0.1085) 2,302 0.0461 1.3678*** (0.1822) 1,696 0.0460 0.5001** (0.2057) 606 0.0260 1.1366*** (0.2837) 307 0.0187 1.4468*** (0.4404) 145 0.6029 2.5141*** (0.6342) 85 0.6168 0.4885** (0.2336) 2,302 0.4795 Panel C: Log(1 + Market-value firm innovation)
(1) All Votes (2) Nonclose (3) 40-60 (4) 45-55 (5) 48-52 (6) 49-51 (7) Full Model Pass Observations R² 1.7951*** (0.3355) 2,440 0.0211 3.3018*** (0.6427) 1,772 0.0337 0.4724 (0.6719) 668 0.0312 0.6636 (0.9315) 353 0.0042 1.785 (1.5284) 154 0.6179 2.9849 (2.1428) 87 0.6969 0.5735 (0.7843) 2.440 0.6424
In panel A, the difference in R&D expenses is examined. Note that the number of observations may change due to missing or incomplete values. In addition, if a firm did not invest in R&D the 3 years before and the 3 years after the meeting, its observations are erased of the sample. Column 1 estimates the difference between proposals passing or failing and shows there is a significant difference at a 10% level. The estimate shows an average impact of 0.0036, that is 0.36%. In column 2, votes that passed or failed by a margin of more than 10% around the cutoff, and column 3, votes that passed or failed within a margin of 10% around the cutoff, insignificant results are presented. While the margins around the cutoff points are narrowing from column 4 to 6,the coefficient sign changes but the estimates remain
four in the vote share, one on each side of the threshold. A polynomial flexibly captures the underlying relationship between any variable continuously effected by the vote share and the outcome variable. Therefore, changes in the estimate by including a polynomial, should be considered. However, the
estimate still remains insignificant. Solely the marginal significance in column 1, does not provide enough support for my second hypothesis. Especially as the emphasized sample, close-call proposals, does not show significance.
In panel B, the difference in the logarithm of the number of patent filings is examined. If a firm did not apply for any patents the 3 years before and the 3 years after the meeting, its observations are erased of the sample. Column 1 presents a significant difference between the 3 years prior to the meeting compared to the 3 years afterwards for a passed proposal. A significant estimate of 0.8523 is shown and this means a 85.23% higher amount of patent applications. This percentage is based on the mean of change. Thus, on average, the difference in the number of patent applications is 85.23% higher for passed proposals than failed proposals. Besides the fact that the coefficient is based on the value of difference, it should be considered that the logarithm of patent applications is an irrational number, giving a relative value. Considering column 2 and 3, the nonclose vote shares give a higher coefficient and significance than the close vote shares, 1.3678 and 0.5001 respectively. Furthermore, the standard errors of the nonclose sample is smaller and the coefficient of determination is higher. This estimate could direct that proposals passing with a great majority have relatively more impact. However, considering column 4, 5 and 6, the coefficient increases while remaining significant. Column 6 displays a coefficient of 2.5141, or an approximate increase of 251% , on the number of patent applications. Thus, for proposals passing with 1%, the amount of patent applications is 2.5 times as high as for the failed proposal with 1%. In addition, the goodness of fit shows that the regression models of column 5 and 6 fit the data well. In column 7, the total sample is presented while including two polynomials of order four in the vote share on both sides of the cutoff. According to this estimate, the average effect of a passing proposal is 48.85% higher than for a failed one. Again, it should be considered this percentage is taken relative to the difference between the period before and after the event. As column 6 and 7 are significant at a 1% level, evidence supporting the second hypothesis is found.
Panel C shows the market-value of firm innovation. According to the whole sample, presented in column 1, there is a significant difference. The estimate of 1.7951, or 179% approximately, is significant at a 1% significance level. Column 2 presents a significant estimate of 3.3018, or 330% approximately, at a significance level of 1%, whereas column 3 provides an insignificant smaller coefficient of 0.4724. The estimates remain insignificant for column 4 – 7. As the market-value of firm innovation is based on the stock price response to innovation, it indicates a significantly high response for proposals passing with a
vast majority (60-100%) to those exceedingly failing (0-40%). However, Table IB.1. in Appendix B. proves there are no significant differences between the samples presented in column 2 and 3. This suggests these results are generalizable to a considerable number of firms in this sample. Though, not to all.
Although Table V provides a transparent depiction of the data in my sample, it does not justify the optional impact of various proposals within one meeting. Table VI analyzes whether there is a critical number of passing proposals within a meeting to obtain a significant impact on innovation. To conduct these regressions, various dummy variables are included for the number of passed proposals within a meeting. My sample ranges from one to ten governance proposals voted on within one meeting.
Therefore, there is chosen to include dummies with a maximum of six passed proposals in one meeting. This, as there is a vast fall in the sample size after six passed proposals which limits the statistical power of the regression. The remaining observations are excluded.
Table VI
Innovation of Passing Governance Proposals
This table presents the effect of one to six passing proposals at a meeting. Six different dummy variables, capturing the number of passed proposals at the meeting, are included. The rows account for six different dummy variables, capturing the number of passed proposals at a meeting. The dummy variable ranges from one passed proposal at a meeting to a maximum of six passed proposals at a meeting. The dependent variables are: R&D/Assets, log(1+ Patent filings) and the log(1 + Market-value firm innovation). All values are relative to firm size. The dependent variables, capturing the difference in innovation between the 3 years before (t-1, t-2, t-3) and the 3 years after (t+1, t+2, t+3) are given in columns 1, 2 and 3. The difference in innovation is calculated by subtracting the mean for the 3 years before the meeting from the mean of the 3 years after the meeting. All columns include a polynomial in the vote share of order four on each side of the threshold and display the full sample of passed proposals. Note that the number of observations may change due to missing or incomplete values for some of the variables. All columns control for year fixed effects, standard errors in parentheses are clustered by firm. Significance at the 10%, 5%, and 1% levels is indicated by ∗, ∗∗, and∗∗∗, respectively.
(1) R&D/assets
(2)
Log(1 + Patent filings)
(3)
Log(1 + Market-value firm innovation)
One vote passed -0.0022
(0.0023)
0.1403* (0.0830)
0.1294 (0.1748)
Two votes passed -0.0026
(0.0025)
0.0774 (0.1503)
-0.0253 (0.2681)
Three votes passed 0.0030
(0.0034)
-0.0323 (0.1646)
-0.2300 (0.4656)
Four votes passed 0.0003
(0.0049)
-0.5987** (0.2381)
-0.0469 (0.6155)
Five votes passed -0.0031
(0.0089)
-0.0234 (0.3043)
-1.1098 (1.4538)
Six votes passed 0.0049**
(0.0022) 0.3805*** (0.1240) 0.7746*** (0.2704) Observations 15,343 10,105 10,937 R² 0.0356 0.4797 0.6419 Number of firm-meetings 2,186 2,286 2,429
Column 1 displays the difference in innovation by the R&D expense. The first row, including a dummy variable for one passed proposal at a meeting, gives a negative insignificant estimate of 0.0022. The dummy variables displaying two passed votes to five passed votes remain insignificant and their signs differ. In the sixth row, the first significant estimate is shown by an estimate of 0.0049 at a 5% significance level. This shows that the firms passing six proposals at a meeting, increase the R&D expenses afterwards by 0.49%. This increase is relative to the firms with shareholder meetings in which less than six shareholder votes passed.
The number of patent filings is displayed in column 2 and gives a significant estimate of 0.1403, approximately 14%. This estimate is significant at a 10% level. While the number of passed votes increase over the rows, the coefficient changes in sign. The fourth row gives a highly significant estimate which is significant at a 1% level. The estimate presents an approximate decrease of 60% for shareholder meetings in which four proposals passed. The last row, including a dummy for six passed proposals at a meeting, give a highly positive significant estimate of 0.3805, approximately 38,05%. Overall, the coefficient of determination is acceptable and represents a good fit of the model with the data.
Considering the varying outcomes of the 1st, 4th and 6th rows, I believe it is difficult to justify assumptions
based on these results.
The third column reports the difference in the market-value of firm innovation. Except for the estimate for six passed proposals at a meeting, column 3 does not provide any significant estimates. The 6th row has an estimate of 0.7746 that is significant at a 1% level. This gives an approximate increase of 77% for six passed proposals at a meeting, relative to the change of mean.
Based on the results of each column on the dummy variable accounting for six passed votes at a meeting, a significant increase in the innovation is shown. Due to the high significance in row 6 (5% level in column 1, 1% level in column 2 and 3), there can be concluded that there is a critical number, namely six passed proposals, at which a substantial increase in innovation is obtained.
