Dual-stock ownership and the relation to R&D
expenditure and M&A activity in different sectors
by
Luisa Bernardi Lopes 11373210
Supervised by Rafael Perez Ribas
University of Amsterdam Faculty of Economics and Business
Master in Business Economics: Managerial Economics and Strategy 15 ECTS
Statement of Originality
This document is written by Luisa Bernardi Lopes 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.
ABSTRACT
This thesis examines the relationship between voting rights and assets returns through R&D and M&A. I am interested to analyze if the assets returns of R&D/M&A provide useful information to why companies transfer voting rights using a dual-class structure and if they compensate agency costs through it. To achieve this, I manually collect the number of shares per company, the different voting rights each class is eligible to and create a continuous variable similar to a Gini to measure the voting inequality across companies and years. Furthermore, I analyze sectors that have a high number of dual-class companies separately to investigate if there are different motivations among them. The model is divided in two: first I analyze the returns on ROA and then analyze if the returns found in the first part impact voting right inequality. I find that in the overall sample assets returns of R&D/M&A do not provide any information for the voting transfer. When analyzing the sectors separately I find that for the information technology sector the M&A activity seems to compensate agency costs and explain the voting transference to controlling shareholders. I also explore if the choice of R&D/M&A expenditure is impacted by the voting structure adopted the year before by the company. For this, I find that overall R&D is impacted by the voting right structure, but not M&A. When analyzing sectors separately, I find that materials, consumer staples and healthcare sectors are more likely to have a higher R&D if the companies adopt a dual-class structure the year before, while for energy and information technology sectors we see this movement for M&A.
I. Introduction
The dual-class structure has received criticism since the 1920s. Even despite attempts from the U.S. Securities and Exchange Commission (SEC) to prohibit super voting shares, this type of structure has remained. Several renowned companies still choose to adopt it, such as Ford Motor, Google Inc. and the recent IPO of Snapchat in 2017. Companies in the past did not have so many reasons to restrict shareholder rights compared to recent years. With an increase of hostile-takeovers, firms were obliged to respond to it, making use of takeover defenses such as poison pills and restriction of shareholder rights through a dual stock structure.
The rule 19C-4 was implemented in 1988 after changes by SEC. The final version, as stated by Howell (2010), declared that “firms were restricted in the manner in which they could become dual-class”. This way, investors know what type of stocks they are buying and are aware of their voting rights. As mentioned by Gompers, Ishii and Metrick (2003) shareholders tend to accept less voting rights to maximize their wealth. Hence, shareholders believe that buying shares with limited voting right can offer a higher return in the future.
The reason stockowners build their beliefs is indicated by Demsetz (1972) that insider-managers hold higher voting rights to avoid uninformed shareholders to replace the management team for a less experienced one. In addition to this, as DeANGELO and DeANGELO(1984) mentioned, also the fact of spending less resources to explain their decisions to shareholders is helpful to implement a better long-run project and make decisions faster. Indeed, this is a reason why several companies explain the adoption of this type of structure. But do the dual-class companies selected projects could actually explain the voting transfer?
This thesis attempt to fill this gap by testing whether there is a relationship between voting right inequality and asset returns through R&D expenditure and M&A activity. The company selected projects are exemplified by the R&D expenditure choices and M&A activities, which represent the long-term strategy of the firm. The hypothesis is that companies that opt for this strategy would compensate agency costs by offering a higher return on their assets through R&D and M&A by selecting the right choices without the interference of outside shareholders. The intuition is that common shareholders seek for short-term returns and do not want the company to expend high amounts in R&D and M&A. On the other hand, the controlling shareholders seek for long-term performance and use R&D expenditures and M&A activity to pursue it. Therefore, limiting the shareholders decision-making power would be beneficial for the company to implement this type of investments,
which for the controlling shareholders is the most beneficial to the firm. While the common shareholders would benefit from a higher return on assets.
To analyze this hypothesis, I create a continuous variable to measure vote inequality. To create it I had to first identify the dual-class companies though the Institutional Shareholder Services (ISS) – Governance data from Wharton. Secondly, I had to manually collect the number and type of shares for each company. After this process, I identify the voting rights each class was eligible separately and every year, because even when two companies have the same class of stock it does not mean they are eligible for the same voting right. Therefore, it is necessary a separate analysis by company to correctly identify and measure the voting inequality. Also, it is important to mention that companies can stop adopting the dual-class structure or change the voting conditions. Hence, it is important to check the values for every year. Finally, I calculate the inequality level in a similar process used to calculate the Gini variable. I draw a Lorenz curve and measure the area below the 45-degree line, which indicates perfect equality, to the Lorenz curve.
Subsequently, I use the vote inequality variable created to assess if there is a beneficial use of this structure in some sectors that rely on this strategy. To do it, I need to separate the analysis in two. Firstly, I analyze the returns from R&D and M&A in companies’ ROA. I do this first step to identify sectors that offer a higher return on assets through R&D and M&A, because these returns would be the way to compensate agency costs to shareholders. If the company is selecting the right amount of R&D investment and it is successfully implementing it, this would reflect in the returns on assets of the firm. After this, I test if voting rights are impacted by the returns firstly found in R&D and M&A. If the values are positive, there is a transfer of voting rights to the company which the agency cost of doing is compensate by the higher return on assets due to R&D and M&A. Therefore, the explanation used by companies that limiting voting rights enable them to select the right projects could be realistic.
Turning to my results, I find that in the overall sample there is no statistically significant relationship between companies that use a dual-class structure and R&D and M&A activity. Only when analyzing sectors separately I find that for the information technology sector there is a transference of voting rights which compensates agency costs through M&A activity.
Furthermore, I also investigate if there is a direct impact of voting rights into the values agreed of R&D and M&A. I analyze if the R&D that will be spent in the year is impacted by the voting structure adopted by the company the year before. Companies usually approve their budget for the next year during the current year, so the structure adopted can influence
the decision because of the higher flexibility level managers have when they hold a higher voting proportion. I find that, in general, the structure adopted impacts the amount of R&D decided to be spent, but I do not find the same for M&A. When analyzing sectors separately, for R&D I find significant impacts in materials, consumer staples and healthcare sectors. While for M&A the sectors impacted are energy and information technology.
The largest contribution of this paper comes from the difference that most researches use a dummy variable to identify dual class ownership, without being able to differentiate sectors or periods in relation to the level of inequality. Hence, this paper offers useful insights from dual-class companies and their level of voting rights over the years and among different sectors. Also, when comparing the average voting inequality and the number of dual-class companies over the years, I can already identify that the curves are not the same. This is an indication that using a dummy variable does not identify all the movements of the voting inequality. When turning to my analysis, I find that mostly the dummy variable overestimates the impact. Thus, it is an important contribution to the literature to identify the need of correctly measuring the voting inequality over the years.
The related literature date from long time ago, as this is a popular topic. Mostly, researchers try to investigate the return on stocks and if there is a difference between dual-class and single-dual-class companies. There is no consensus about this topic, but more and more the recent papers find positive values. Relating to the R&D literature, several papers study the impact of R&D in companies’ valuation and there is a consensus that it does impact companies positively.
II. Literature Review
This thesis is related to two main strands of literature. Firstly, the research on dual-class structure and shareholder value. Papers investigate the returns stockowners can get from investing in different type of structures. The motivation is to discover if companies with a dual-stock structure have a higher or lower return when compared to single-class.
Some studies correlate a dual stock ownership with lower firm values. Gompers, Ishii, and Metrick (2003) find some evidence that weak shareholder rights caused poor performance during the 1990s. Lins (2003) finds some evidence from emerging markets that firm value is lower for companies with a more concentrated voting structure than cash flow. Li and Zaiats (2016) identify that dual stock companies are correlated to poorer information environment and more likely to make use of earning management tools. However, there is a majority of papers with an opposite view on the matter. Johnson, Moorman, and Sorescu
(2007) reexamine the results found by Gompers et. al (2003) adding a cluster approach to industries and find that there is a “zero long-term abnormal returns for portfolios sorted on governance” (Johnson, Moorman, and Sorescu (2007). As argued by Chemmanur and Jiao (2012) dual stock ownership may increase long-term value for the firm, but only when the company is leaded by high ability managers. Also, Chemmanur and Tian (2013) find that more innovative firms usually have a high number of antitakeover provisions to be protected. Furthermore, Jordan, Kim and Liu (2016) encounter that dual-class structure increases the market valuation of high growth firms. They also find that companies which adopt this strategy have higher sales growth and R&D intensity.
If a causal relationship between poor corporate governance and lower stock return exists, it is expected that the market is negatively surprised by this factor. Hence, investors knowing it would rather buy shares from companies with a high corporate governance. Core, Guay, and Rusticus (2005) study this effect and find that investors and analysts are not surprised by this fact. Given this result, the findings from papers that a negative correlation between dual stock ownership and firm value exists is intriguing.
Secondly, this thesis is related to R&D expenditure and firm value. Several papers state that R&D activity adds value. Davidson and Brooks (2004) found that “undertaking R&D increases the value over the company irrespective of agency costs”. Also Warusawitharana (2015) finds that R&D expenditures contribute significantly to profits and firm value. Hence, it is expected that industries that have a higher expenditure in R&D have a higher return on assets (ROA).
To the best of my knowledge, there is no study that directly tried to measure the reasons companies decide to adopt one structure over another. In this research, I study if companies that have a higher expenditure in R&D and a great amount of acquisitions would be more likely to adopt a dual-class ownership. There is evidence analyzing the data that this hypothesis hold, but no paper tried to measure it. As mentioned by Gompers, Ishii and Metrick (2003) companies with stronger shareholders’ rights have lower capital expenditures and made fewer corporate acquisitions and by Jordan et al (2016) dual stock companies tend to have a higher R&D expenditure.
Important to point out is the fact that most of papers use a dummy variable to identify dual class ownership, while I create a continuous variable to better understand the level of inequality and how it affects the choices. This paper is only focused on the causalities from the different number of voting rights coupled to the stock class and do not take into consideration other corporate governance approaches.
When using a continuous variable, it is easier to understand the different levels of inequality and how it impacts the other variables. Also, the variable provides more information about the company and sectors than a dummy. For example, it is possible to identify sectors that make more use of a dictatorship over this type of structure, while some other sectors can still be more democratic. This approach is not so used in the literature due to the difficulty of gathering the data and analyzing each case separately. The major problem of it is that you limit the number of years and companies studied.
III. Data
A. Sample Construction
This thesis exploits two data sources. First the Institutional Shareholder Services (ISS) – Governance data from the Wharton website to identify the dual class firms. Secondly, Compustat – Capital IQ (North America) from the Wharton website to collect the detailed information about the companies. The data collected ranges from 2006 to 2016, which covers the period before and after the crisis. All the companies are listed in the US Market and all data is recorded in million US dollars.
From ISS, I exclude companies in financial services, real estate and public services sectors.
The total number of companies is 1,551, of which 115 have a dual stock ownership. For industry classification, I use the Global Industry Classification Standard (GIC). Unlike the Standard Industrial Classification (SIC), the GIC offers a better structure to study the high-tech sector.
Companies that did not have any Research & Development expenditure are all moved to a new sector created. In addition to R&D, I also investigate the behavior of incumbent companies that acquire technologies through M&As. The paper takes this approach because it is a popular action among technology companies to acquire smaller ones for the purpose of investing in new resources, or also to have the right in the future to the technology developed by them. It is possible that companies which did not have any R&D had acquisitions during the period studied.
Using the detailed information collected I create a number of variables. RoA, which is the Return on Assets (Net Income(Loss)/lagged Total Assets), the Research and Development book value in t-1 (lagged R&D /second lagg of Total Assets), (lagged M&A/second lag of Total Assets) which defines any value spent related to mergers and
acquisitions. ‘Size’ in 1 is measured as the logarithm of lagged total assets. ‘Leverage’ in t-1 is the lagged value of (Debt in Current Liabilities + Long Term Debt)/Total Assets).
B. Classifying dual shares
Most of the dual stock ownership companies have a fixed number of votes per share and to distinguish the two different classes they use several different labels. The three most common labels are common stock, Class A and Class B. For example, Google Inc.’s Class A grants 1 vote per share, while the Class B grants 10 votes. In this case, Class B is classified as “superior” and Class A as “inferior”. The different companies do not apply the labels consistently; therefore, it is required to analyze each company separately to determine the number of votes eligible to the specific Class.
Figure 1: Voting rights reasoning
This table describes the method used to create the GINI variable. It shows the reasoning between different type of companies and how each definition was quantified.
Furthermore, each company has a different number of votes designed to each class, it can range from 10 to 250 votes in the samples used. It is also common that companies only have the inferior class publicly traded, while company members hold the superior ones. In the sample, 30% of companies are considered exceptions, having a different structure which requires to analyze not only the 10-K file but also the DEF 14A. In those companies, each
Companies Sample
Fixed number of voting rights per share specified in
the 10K
Class "X" has "n" voting rights = "n" votes
Number of votes not completely defined
Class that only can elect "n" number of Board Members = n/total members*voting right
of the class (if eligible)
Class that can only vote in matters as such mergers and acquisitions = 1 vote*voting right of the class (if eligible)
Class that has rights to dividends or to vote for compensation plan = 0 votes
class has a different right and the difference is not only regarding the number of voting, that sometimes is not even showed. To quantify the exceptions, I only take into consideration matters that impact companies’ R&D, such as the constitution of the Board of Directors, mergers and acquisitions, and share buy-back.
A full list of subjects is available in the Appendix (Table A1). I do not consider, for instance, issues such as right to dividends or compensation plan. An overview of the voting rights reasoning is displayed below in Figure 1.
C. Measuring voting right inequality
After determining the number of votes each class is eligible to, I calculate some measures of voting rights inequality per company. Creating a one-dimensional variable that characterizes this inequality is critical to simplify the research and to better understand changes of company structure during the years. Moreover, it enables us to compare companies in different industries and their level of inequality.
The first one is the Gini index, which derives from the Lorenz curve. Figure 2 shows the Lorenz curve of voting rights from Google Inc. exemplified before.
The Lorenz curve is a graphical device used to represent distributional inequality. It was developed in 1905 by Max Lorenz and it is mostly used to represent income inequality and wealth distribution. In this case, I use it to represent the distributional inequality of voting rights. To construct it I analyze the number of shares per class and the number of voting rights granted. The Lorenz curve in this case is a function of the cumulative of ordered classes (the one with less voting rights to the highest one) mapped onto the corresponding cumulative proportion of the total number of shares per class and it is represented by the gray line in Figure 2.
The line of 45-degree (dotted lined) represents a company that only has one class of stock or every class has the same voting right, meaning that it is perfectly equal. If the cumulative votes resemble the 45-degree line, then all stakeholders have the same rights and it only depends on the number of shares held. The companies with a dual class structure in their majority will be located below the 45-degree line or will be equal to it when both classes have the same right. For those companies the area below the perfect equality line and the line created for each company is measured. Companies close to the value of 1 are more unequal and labeled as a “dictatorship” type of company, while companies closer to 0 are more equal and resemble more a democracy.
To measure the area in between the dotted line and the Lorenz curve is used a simple calculation of the triangle area ((Base*Height)/2) and trapezoid ((Base 1 + Base 2)/2 * Height). Since the Lorenz curve is a cumulative function, both axes range from 0 to 1. The total area below the 45-degree line is 0.5. After this, it is possible to calculate the areas below the Lorenz curve. In the case of Google Inc., we have two different trapezoids areas to calculate. First, to the Class A with the first base equals to zero, the second of 0.2 and the height of 0.8, totalizing an area of 0.08. The second trapezoid we have the first base of 1, second of 0.2 and height of 0.2, totalizing 0.12. Hence, the total area below the Lorenz curve is the sum of the trapezoids areas calculated, in this case 0.2.
The Gini Coefficient is a measure from 0 to 1. Area A is represented by the area between the 45-degree line and the Lorenz curve. Area B is the area below the Lorenz Curve. Hence, the Gini Index in this case for Google Inc. during 2006, is Area A / (Area A + Area B) = 0.3 / 0.3 + 0.2 = 0.6.
The second measure of inequality is the percentage of voting rights held by the top 5% shareholders. Analyzing again Figure 1, it is notable that more than 20% of owners of Google Inc. shares have nearly 80% of the voting rights, while the most stock owners (8o% of shares) only hold the remaining 20% voting rights. The percentage held by the top 5% is 17%.
Figure 2: Lorenz Curve of Voting Rights at Google Inc.
This graph represents the voting structure from Google Inc. in 2006. The dotted line represents the perfect linear voting right line. The gray line represents the Lorenz curve, which is the structure used by the company. In this case, nearly 80% stockowners have 20% voting rights, and the others 20% have 80% of the voting rights.
The difference between the two variables is critical. The former is more precise to analyze the middle of the curve, while the tails of the distribution are not so well measured.
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 % of v ot in g rig ht s % of shares
On the other hand, the latter variable is useful to measure the tails, adding a robustness test to the regression. Moreover, the database has a dummy variable, where 1 equals to companies with a dual stock ownership structure and 0 to companies without it. This way it is also possible to identify companies that have a dual ownership structure but the Gini Index of 0, meaning they have a democratic structure. Furthermore, it is possible to compare the gains of using a continuous variable to measure the voting inequality when compared to the use of a dummy.
D. Descriptive Statistics D.1 Dual-stock ownership
A dual-stock ownership enables insiders to hold a higher number of voting rights, and therefore a higher control over the company’s decision. In theory, it is expected that industries that can profit from maintaining control would most likely adopt this type of structure. As some researches of this topic show, dual class IPOs are more common used in three type of companies: (1) industries in which value can be created by seeking long-term goals and that short-term trends are not so relevant; (2) family owned companies and run by founding entrepreneurs; and (3) firms with large private benefits of control (Chemmanur & Jiao, 2006).
Table 1 summarizes the number of companies per sector in the sample and it is divided between companies that have a dual stock ownership and companies that do not. It is noticeable that companies without it are predominant in number. The sector with more companies in the sample is Information Technology, even though the one with more dual stock companies is Consumer Discretionary, which includes media, followed by Information Technology.
The findings are consistent to the database used by Howell (2009), which is the largest sample of United States dual-class firms (1,096 companies studied over 20 years). Both studies show that the main industries that adopt a dual class structure are media (such as The New York Time and News Corporation), IT services (such as Facebook and Google) and consumer discretionary. The IT services industry is the one most heard of in recent years for using this structure. Mark Zuckerberg, in 2016, defended its use pointing out that if Facebook had adopted a single-class the firm would have been sold out to Yahoo back in 2006 and it would not be such a valuable company as it is now.
Table 1: Summary statistics (averages)
Companies with Dual Stock Ownership GIC
Sector Sector Classification
# of
Comp. Size Leverage
Adj. ROA Adj. R&D Adj. M&A 10 Energy 1 10.302 0.555 0.009 - 0.008 15 Materials 3 8.482 0.248 0.051 0.005 0.031 20 Industrials 17 7.405 0.179 0.050 0.044 0.030 25 Consumer Discretionary 51 7.980 0.296 0.034 0.006 0.019 30 Consumer Staples 15 7.961 0.258 0.081 0.009 0.029 35 Health Care 5 7.704 0.257 0.025 0.101 0.019 45 Information Technology 20 7.933 0.084 0.071 0.075 0.048 50 Telecommunication Services 3 8.644 0.408 (0.012) - 0.016
Companies without Dual Stock Ownership GIC
Sector Sector Classification
# of
Comp. Size Leverage
Adj. ROA Adj. R&D Adj. M&A 10 Energy 123 8.376 0.263 0.007 0.015 0.016 15 Materials 121 7.750 0.279 0.045 0.016 0.030 20 Industrials 253 7.740 0.221 0.053 0.023 0.035 25 Consumer Discretionary 271 7.447 0.249 0.055 0.015 0.017 30 Consumer Staples 79 8.340 0.291 0.077 0.009 0.031 35 Health Care 233 7.320 0.212 0.048 0.067 0.050 45 Information Technology 333 7.085 0.137 0.030 0.093 0.041 50 Telecommunication Services 23 8.391 0.467 0.024 0.010 0.023
This table is a summary of the sample used in this paper. It is divided in two: companies which adopt the dual stock ownership and companies which do not. It contains the averages of size, leverage, ROA, R&D and M&A per sector during the period studied. Data are annual from 2006 to 2016.
Comparing the characteristics of the different structures, we have that companies are very similar in size in both sub-groups, but with dual-class companies on average being larger. This is mostly driven by the Energy and Materials sector, excluding both the sizes are practically the same with dual-class being slight larger. Turning to leverage, companies with dual stock, on average, have a higher value. Those findings are also in line with the database from Howell (2009). This is a good indication that the sample used in the paper is a good representation of the dual stock population.
In comparison to this, when looking to ROA most of the sectors are practically the same, with companies without it having a higher average, besides Information Technology sector that has 2.3 times higher ROA. Also when analyzing the Telecommunication Services sector it is possible to notice that companies with dual stock ownership have a negative ROA, while without a positive one. When checking R&D we can notice that, on average, companies without the dual structure spend more on it, with two exceptions: Industrials and Health
Care. For M&A, companies with a single-class structure also make more use of this action, with two main exceptions: Consumer Discretionary and Information Technology sectors.
Table 2 shows a comparison of the inequality level among different sectors and the number of observations in the total sample. Industrials have the highest mean among all sectors, with a range going from 0 to 0.91. It is followed by Health Care and Consumer Discretionary, with the latest having companies ranging for the whole interval (0 to 1). Having such a broad interval shows that there are companies very diversified within the same sectors, one being more democratic with others being closer to a dictatorship. The Materials sector is more to the opposite, its average is below 0.5, so most of the companies resemble a democracy. Also, the maximum value in this segment is 0.54, therefore there is not so extreme type of companies. The IT sector has a mean below 0.4, but analyzing its range we can notice that there are companies with a zero level of inequality and in the opposite side there are companies with a maximum of 0.82, being almost completely unequal.
Table 2: Gini Variable Details
GIC Sector Sector Classification Obs. Mean Std.
Dev. Min. Max. 10 Energy 2 0.08 0.02 0.06 0.09 15 Materials 18 0.40 0.17 0.04 0.54 20 Industrials 161 0.49 0.19 0.00 0.91 25 Consumer Discretionary 445 0.42 0.26 0.00 1.00 30 Consumer Staples 132 0.34 0.24 - 0.63 35 Health Care 47 0.44 0.27 0.14 0.87 45 Information Technology 154 0.38 0.23 - 0.82 50 Telecommunication Services 26 0.27 0.23 - 0.58
This table is a summary of the Gini Variable. It shows the number of observations in the sample, the mean, standard deviation, minimum and maximum value.
E. Sectors new sub-division and periods division
This thesis analyzes the impacts of R&D and M&A per sector. For the GIC taxonomy there are 11 different sectors, as mentioned before I am not taking into consideration 3 of them, hence being left in total with 8 sectors. The sectors with the highest number of dual stock companies were then opened per industry and the industries that had a value higher than the average of dual stock of the sample were defined as a new sub-sector. The other industries were left in the main sector.
To easily work with the data, the period of 11 years was sub-divided into three different periods: 2006 – 2008, 2009 – 2012 and 2013 – 2016.
The 3 sectors with the most number of dual stock companies are: Consumer Discretionary, which has new sub-sectors for Household durables, Textile, Apparel & Luxury Goods, Hotels, Restaurants & Leisure, and Specialty Retails; Consumer Staples, which has new sub-sector for Food Products; and Information and Technology, which has sub-sectors for Internet Software & Services and IT Services. The Telecommunication Services sector did not have enough observations, so it was coupled together with IT Services.
To easily work with the data, the period of 11 years was sub-divided into three different periods: 2006 – 2008, 2009 – 2012 and 2013 – 2016. This choice was done to separate the period before crisis and after it. Also, it was preferred to group more years in each period to have a higher number of observations and a better understand of what is happening.
IV. Empirical Strategy
I attempt to test the relationship between companies that decide to use a dual stock ownership structure and their R&D expenditure. To analyze this the paper makes use of a model divided into two different regressions.
The first part of the model analyzes if for a specific company “i”, at sector “s” and at period “t” if R&A and M&A affect the Return on Assets, the dependent variable. Size, leverage and fiscal year are used as control variables in the model. The model estimates a different return for R&D and M&A per sector and period. Hence, each sector has 3 different returns in the sample studied. To control for unobserved heterogeneity that is constant over time, I use a fixed effect model and I cluster for company.
The model is set out below.
!"#$% = ( + *+%,∗ !&/$%0++ *1%,∗ 2&#$%0++ *3%,∗ 4567$%0++ *8%,∗ 97:.$%0++
*<%,5. =5>?@AB7@C + D$%, (1)
I am interested in the impact of R&D and M&A in ROA to then apply it to the second part of the model. The intuition behind is that expenses in R&A and M&A could result in a higher ROA depending on the sector and period. The different βFGH found in the first part of the model are used in the second regression to identify if there is any transfer of voting rights.
In the second part of the model, the dependent variable is the Gini Index, the one-dimensional continuous variable created using the voting right. It uses the βFGH from the first part to analyze if there is a compensation of agency costs through R&D expenses and M&A
activity to shareholders of dual stock companies. As in the first equation, size, leverage and fiscal year are used as control variables. Also, to control for unobserved heterogeneity that is constant over time, I use a fixed effect model and I cluster for company.
I first regress to all companies in the sample adding cluster to industry. I do it because the industry distribution of companies with a democratic or dictatorship structure are statistically and economically different, as mentioned by Johnson, Moorman, and Sorescu (2007). I also regress separately for the sectors that most use the dual-class structure to investigate if they have any particularity. The second part of the model is defined as below.
I5J5 KJL7M = ( + N+ *+%, + N1 *1%, + N3∗ 4567$%0++ N8∗ 97:.$%0++ N<5. O7C5PL + D$% (2)
The theory used is that the return on R&D and/or M&A compensates agency costs for shareholders of companies considered not democratic. If this holds, there is a transfer of voting rights. On the contrary, there is no transfer of voting rights. In summary:
If N$ > 0: a positive value to the companies, which implies a “dictatorship”. If N$ < 0: a negative value to the companies, which implies a “democracy”.
I hypothesize that companies with a dual stock ownership will have a positive γR and companies, which do not use this strategy, will have a negative γR , implying that there is no transfer of votes so there is no need for the companies to compensate through R&D a higher RoA. The idea is that companies that offer less voting rights in their structure are doing it because they have better knowledge of internal projects that can take longer to offer return on the investment than outsiders and therefore they need to compensate for the lack of democracy offering a higher value of return.
I expand my model using the Top 5% variable to test for the tails of the distribution and analyze if the result found using the Gini Index is consistent. Moreover, I run the model using a dummy variable for dual stock ownership to evaluate if there are any gains on expanding the analysis to a continuous variable besides the better knowledge of the sectors.
V. Results
The results are divided in four: a description of the inequality voting rights over the years, the return on assets regression, the voting rights regression and a new model to test for direct relationship between voting inequality and R&D/M&A. The first part explains the difference of voting rights over the years in the sectors comparing it to the number of companies. With
this analyzes we can already identify if the distributions are different and if there are any gains of using the continuous variable instead of a dummy.
The second one is related to the first part of the model, where I identify the returns from R&D and M&A for each sector and period. In this section I present the distribution of the returns and the average of the return during each period. I also compare with the descriptive analytics from the data section and identify if the sectors, which are more famous to adopt the dual stock ownership, have a higher return on R&D and M&A.
The third part presents the regression of the returns explained in the former section and analyzes if the returns are used as a mean to compensate the agency costs from the voting right structure. I try to identify if there is a difference among the three sectors that most use the dual-class ownership, if R&D or M&A is more important to the structure or if in general companies adopt this type of structure more often due to the expenditures in R&D and M&A. The fourth and last part analyzes if companies that concentrate voting rights increase R&D expenditure/M&A activity. The idea is to identify if instead of R&D and M&A being the reason for companies to adopt the structure, that both are impacted by the different voting structures.
A. Inequality voting rights over years
Figure 3 shows a comparison of the inequality level among sectors during the years, having the year of 2006 as baseline and equals to 100. We can notice that sectors usually decrease the average level of inequality from 2006 to 2016, with some sectors going into a different direction as Materials, Industrials and Consumer Discretionary.
In Figure 4, we have the number of companies during the year for each sector, also using the year of 2006 as baseline and equals to 100. We can notice that when comparing to 2015 there is not a major decrease in the number of companies with this type of structure. The largest difference is for Consumer Discretionary, from 44 to 38. It is important to compare here with 2015 since some companies did not have their balance sheet released by the moment this research was done, meaning that the number can be lower because some companies did not have their data available, and not because companies changed their structure or shut down.
It is also noticeable that Information Technology started with a higher value of inequality, and after the crisis in 2009 only decreased this value over the years. In 2011 two companies are not part of the sample anymore, Keithley Instruments because it was sold to another company, and SRA International Inc. which stopped using a dual stock ownership, two
companies which have a inequality value higher than the average. In 2012 two new companies start adopting the structure, being one of them Facebook. When checking Figure 4 we can see that over the year the Information Technology sectors drops every year, while the level of inequality is almost stable. This is due to the fact that the companies have a similar inequality right.
A sector which has vert abrupt changes is materials. It is hard to draw conclusions about the whole sectors because there are only 3 companies that adopt a dual class structure. In 2008 and 2009 the inequality value was so high because Greif was the only company and it has a high Gini value. In 2010 the level goes down because of a new company adopting the structure, Mosaic. This company has a lower Gini value when compared to the other two, hence the average is impacted by it. In 2015 Mosaic stopped using the structure, and once again Greif was the only company.
Another sector with a lot of changes is Telecommunication Services. In 2010 Sprint Corporation stopped adopting the structure, the company had a very low value for gini, which explains the increase of the average value for this sector in the year. In 2011 one of the companies, Telephone & Data Systems Inc. kept using a dual stock ownership, but both classes were eligible to the same voting rights. Therefore, the company had a gini equals to 0, bringing the average for the sector to lower levels.
Figure 3: Average Gini over the years
This graph illustrates the average value of the Gini Variable growth having 2006 as base 100. -20.00 40.00 60.00 80.00 100.00 120.00 140.00 160.00 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
Energy Materials Industrials
Consumer Discretionary Consumer Staples Health Care
It is important to point out that the sectors with a high number of dual-class companies do not have such abrupt changes, this is due to the fact that companies which start to use a dual-class structure and the ones that stop even each other out over the years.
From Figure 4, we can notice that Healthcare was the only sector which increased the number of companies compared to 2006, but the voting rights had a decrease when compared to 2006. This implies that more companies in the sectors started to adopt this type of structure, but with a lower level of inequality.
Figure 4: Number of dual-class companies over the years
This graph illustrates the number of companies with a dual stock ownership from 2006 to 2016 for different sectors, having 2006 as base 100.
When comparing both Figure 3 and 4, we can notice that the movements from the curves are not the same. This already shows that a dummy variable does not contain all the information as the Gini does. Hence, it is important to create a continuous variable to correctly capture the differences of voting rights over the years.
B. Return on Assets Regression
The return on R&D and M&A vary from industry to industry. Some sectors rely more on R&D expenditure than others. As for example the Life Science, which is part of Healthcare denomination, Aerospace (part of Materials) and Information and Technology Communication sectors. This type of industries is fast moving pace and they need to keep update with the new developments. Also, Life Science and Information and Technology rely
-20 40 60 80 100 120 140 160 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
Energy Materials Industrials
Consumer Discretionary Consumer Staples Health Care Information Technology Telecommunication Services
on creating new products and developing new technologies. Hence, they need to keep a high level of R&D expenditure over the years. Those sectors also rely on M&A activities, as such buying smaller companies that already did research in a topic to acquire the product developed or just for the knowledge.
On the other hand, some sectors as Energy and Telecommunication Services, are more Capex intense. The companies do not need to innovate as much as other sectors, but they do need to provide a high-level service through a good infrastructure. Hence, R&D is not considered a productive investment.
The main results from the first regression are shown below. Figure 5 presents the distribution of the R&D returns. The axis x is a combination of industry and period. The numbers shown represent one industry and the following two data points are related to the industry over the period. For example, number 4 represents the energy sector in period 1. The following data points represent period 2 and 3, respectively. Table 3 summarizes the industry and the following data points are always the subsequent periods.
Table 3: Description of axis X (Figure 5 and 6)
1 Sectors without R&D 4 Energy
7 Materials 10 Industrials
13 Consumer Discretionary 16 Household Durables
19 Textile, Apparel & Luxury Goods 22 Hotels, Restaurant & Leisure 25 Specialty Retails
28 Consumer Staples 31 Food Products 34 Health Care
37 Information Technology 40 Internet Software & Services 43 IT & Telecommunication Services This table describes the variables of axis X from Figure 5 and Figure 6.
In Figure 5 we can identify the sectors which their ROA are more impacted by R&D intensity. The household durables, materials, health care and information technology have their betas very close to zero, meaning those industries do not have a high impact in ROA due to R&D. It was expected that health care and information technology would have a positive impact, because the two sectors have high R&D expenditure and are very innovative.
Sectors that are positively affected are textile, apparel & luxury goods, with the higher return, followed by food products, consumer staples and IT services. The highest return in fashion and luxury goods can be explained by the necessity of always have a new product and innovate to attend the high expectations of consumers. The sector that is most negatively impacted is specialty retails, followed by internet software & services and consumer discretionary with a lower impact. The internet software & services had a high negative impact in the first Period, pre-crisis, but after this the following two periods were positive and increasing. A higher return for this sector was expected, especially since it includes the most popular companies that adopt a dual-stock ownership, such as Google and Facebook.
Figure 5: R&D Beta distribution over industries and periods
This graph shows the R&D Beta distribution over different sectors and periods. The Betas are defined from regression (1) (see Section IV for details). Beta is defined as the impact of R&D in RoA.
Figure 6 shows the distribution for the M&A returns. Axis X has the same denomination illustrated in Table 3. It is noticeable that the impact of M&A is much lower than R&D. The explanation being companies do not have the possibility of make acquisitions or mergers every year and it has a limited number of companies in the market available for this procedure. Additionally, the graph below is more volatile than the R&D distribution of betas.
-5 0 0 50 100 Be ta R &D 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46
Industry Taxonomy and Period
From Figure 6, we can identify the sectors that their ROA is more impacted by M&A intensity. The higher Betas are in textile, apparel & luxury goods, followed by consumer discretionary, energy and companies without R&D. Apparently M&A is being used as a substitute, with companies opting either to develop their products in house, or look for other companies to acquire the knowledge. The lower returns are in consumer discretionary, followed by household durables. The sector of consumer discretionary is both in the highest return and lowest, due to a difference of periods. This helps to support the theory that not all acquisitions are profitable or that it takes time to notice all synergies. Additionally, the only sector which experienced negative returns in all periods is healthcare, which was expected to have a higher value due to the popularity of small companies being acquired in the health tech sector. This could be explained by the long time that it takes to launch a new product in this sector, or to find a new technology.
Figure 6: M&A Beta distribution over industries and periods
This graph shows the M&A Beta distribution over different sectors and periods. The Betas are defined from regression (1) (see Section IV for details). Beta is defined as the impact of M&A in ROA.
Comparing both graphs, we can identify differences among sectors. For the energy sector M&A shows to be a better investment than R&D. For the textile, apparel & luxury goods it seems that both R&D and M&A are strategies used and both offers its returns to the sector.
-1 0 1 2 3 Be ta M&A 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46
Industry Taxonomy and Period
The internet software sector had a similar return for both R&D and M&A, being negative in the first period and positive for the remaining. This could be an impact from the moment before crisis.
From both Beta distributions, we can see that most industries do not have a constant high or low impact in ROA caused by R&D or M&A. What we can notice is that there is a period where the impact was higher, as an important product development, which was launched, or a valuable acquisition/merger.
A full table of the significance level for the returns of Figure 5 and 6 is attached in the Appendix (Table A2 to A4).
Figure 7 illustrates the average of R&D and M&A returns in each period.
Figure 7: Average of R&D and M&A returns during Periods
This figure shows the average of the returns of R&D and M&A found in the first regression. It represents the average impact in all the companies studied in ROA. The graphs are divided per periods to analyze the differences across time.
Period 2 has the average of its M&A returns equals to zero, while R&D reached its highest point of 0.5. This is the period after crisis, when companies were negatively impacted. Acquiring new companies during the period was more complicate, since most companies were suffering from the recession. Furthermore, companies that could manage to
0 .1 .2 .3 .4 .5 0 .1 .2 .3 .4 .5 Period 1 (2006 - 2008) Period 2 (2009 - 2012) Period 3 (2013 - 2016)
Avg. of R&D Returns Avg. of M&A Returns Graphs by RECODE of DataYearFiscal (DataYearFiscal)
differentiate themselves from the others, with the help of R&D, could benefit from it in the period after-crisis. In Period 1 the both returns are very low, but it is the only one that the average of M&A is more noticeable. In Period 3 the impact of R&D was positive and high, but still lower than in Period 2. For M&A, the average Beta was slightly higher than zero. In general, we can conclude that R&D activities offer a higher return, on average, to companies when compared to M&A.
C. Voting Rights Regression
The second part of the model is related to the voting rights. This section is divided in two. Firstly, I present the overall analysis, clustered by industry. The second part I open the industries that have a higher number of dual-class companies and I analyze the result separately for each of the three sectors: consumer discretionary, consumer staples and information technology.
C.1) Clustered by the new industry taxonomy
Table 4 shows the impact of returns in the voting right allocation of companies, clustered by the new industry taxonomy. This table exemplifies the overall impact of R&D and M&A in all sample companies, correctly clustered by different sector specifications.
For all the dependent variables, the R&D impact is negative, but not significant in the 10% level. Analyzing M&A, the gini and top-5% variable both have a negative return, while for the dummy variable a positive one. All returns are also not significant in the 10% level. We can notice that the impact of the Betas are lower for the GINI and Top 5% variable regressions when compared to the dummy. This can be explained by the fact that a dummy variable does not contain any information about the inequality level of a company. Hence, when using a fixed effect model the dummy variable will become collinear, while the continuous measures will provide more variation and a more accurate result.
When analyzing the interaction term of R&D and M&A it is also not significant. Besides this, the gini variable has a positive value, while the top-5% and dummy a negative one.
Leverage and size seem to not be relevant for the voting right allocation. This makes sense when analyzing the sample and not having any very large difference between the averages of dual-class and single-class companies. The periods are relevant for the gini and dummy variable, but in the robustness variable it seems to not have an impact.
As none of the variables besides some of the control variables, we reject the hypothesis that there is a compensation of agency costs through R&D and M&A. Therefore, the R&D expenditure and M&A activity would not explain the use of a dual stock ownership structure.
Table 4: Estimated Effect of Returns on Voting Rights
Voting Right Allocation
GINI Variable (1) Dummy Variable (2) Top-5% Variable (3) Beta R&D -0.000176 -0.000487 -0.0000256 (0.000147) (0.000459) (0.0000987) Beta M&A -0.00244 0.00359 -0.00171 (0.00745) (0.0161) (0.00387) R&D-M&A 0.0000457 -0.00000928 -0.0000853 (0.000264) (0.000577) (0.000136) Period 2 -0.00176** -0.00420* -0.000469 (0.000829) (0.00226) (0.000481) Period 3 -0.00303* -0.00781** -0.00112 (0.00171) (0.00387) (0.000848) Leverage -0.000631 0.00803 0.000888 (0.00434) (0.0130) (0.00219) Size 0.00194 0.00309 0.00114 (0.00131) (0.00305) (0.000727) Constant 0.00151* 0.0467** 0.00601 (0.00913) (0.0208) (0.00510) N 14,237 14,237 14,237 Adj. R-square 0,002 0,002 0,002
Industry Cluster Yes Yes Yes
Company Cluster Yes Yes Yes
Fixed Effects Yes Yes Yes
***, **, * represent statistical significance at the 1%, 5% and 10% levels, respectively. Standard errors in parentheses are clustered by the new sector category created and by company. The top panel presents the regression coefficients of interest. Column (1) has the results for the Gini variable. Column (2) has the results for the dummy variable. Column (3) has the results for the Top 5% variable.
C.2) Results per sector
This section shows the second part of the model divided by sectors. The regressions were only possible for the sectors in which the new industry classification was sub-divided. For the others, there were not enough observations.
Table 5 shows the results for the Consumer Discretionary sector. This sector includes the textiles, apparel & luxury goods which had the higher returns in R&D and M&A. Moreover, it is the sectors which have more firms with a dual-stock ownership. As in the previous section, the results are not relevant. And as before, the R&D returns are all negative, while
the M&A returns are negative for both gini and top-5% variable, while for the dummy it is not.
In conclusion, it seems that for consumer discretionary companies, R&D and M&A do not compensate agency costs and are not a useful tool to explain the adoption of a dual stock ownership. This type of industry relies in short-term trends, therefore you can verify the return of your R&D and M&A expenditures very quickly and therefore this is not a reason to adopt a protective model through dual-class.
As before, the Dummy regression has higher values for the coefficients than the other two models. And period 2 and 3 seem to be relevant for the adoption of the voting structure.
Table 5: Estimated Effect of Returns on Voting Rights for Sect. 25 - Consumer Discretionary
Voting Right Allocation
GINI Variable
(1) Dummy Variable (2) Top-5% Variable (3)
Beta R&D -0.000176 -0.000487 -0.0000256 (0.000147) (0.000459) (0.0000987) Beta M&A -0.00244 0.00359 -0.00171 (0.00745) (0.0161) (0.00387) R&D M&A 0.0000457 -0.00000928 -0.0000853 (0.000264) (0.000577) (0.000136) Period 2 -0.00176** -0.00420* -0.000469 (0.000829) (0.00226) (0.000481) Period 3 -0.00303* -0.00781** -0.00112 (0.00171) (0.00387) (0.000848) Leverage -0.000631 0.00803 0.000888 (0.00434) (0.0130) (0.00219) Size 0.00194 0.00309 0.00114 (0.00131) (0.00305) (0.000727) Constant 0.00151* 0.0467** 0.00601 (0.00913) (0.0208) (0.00510) N 14,237 14,237 14,237 Adj. R-square 0,002 0,002 0,002
Company Cluster Yes Yes Yes
Fixed Effects Yes Yes Yes
***, **, * represent statistical significance at the 1%, 5% and 10% levels, respectively. Standard errors are the in parentheses. The table shows the regression for the sector number 25: Consumer Discretionary. The top panel presents the regression coefficients of interest. Column (1) has the results for the Gini variable. Column (2) has the results for the dummy variable. Column (3) has the results for the Top 5% variable.
Table 6 shows the results for the Consumer Staples sector. Once again R&D and M&A are not relevant. Moreover, in this case the top-5% variable has opposites signs for both returns. In this segment, we have famous family-owned companies, as such as Walmart, and this can drive the use of voting right allocation in this sector more than the type of innovation they make use of.
Table 6: Estimated Effect of Returns on Voting Rights for Sector 30 – Consumer Staples Voting Right Allocation
GINI Variable (1) Dummy Variable (2) Top-5% Variable (3) Beta R&D 0.000257 0.000473 -0.0000375 (0.000718) (0.00174) (0.000273) Beta M&A -0.00151 -0.0395 0.00773 (0.0397) (0.0851) (0.0288) R&D M&A 0 0 0 (.) (.) (.) Period 2 0.00267 0.00253 0.000997 (0.00719) (0.0166) (0.00377 Period 3 -0.000528 -0.00550 0.00160 (0.00299) (0.00347) (0.00297) Leverage -0.00722 -0.0140 -0.00183 (0.00784) (0.0187) (0.00236) Size -0.000341 0.000290 -0.000326 (0.000617) (0.000939) (0.000509) Constant 0.0528*** 0.151*** 0.0207*** (0.00590) (0.0125) (0.00346) N 870 870 870 Adj. R-square -0.004 -0.005 -0.005
Company Cluster Yes Yes Yes
Fixed Effects Yes Yes Yes
***, **, * represent statistical significance at the 1%, 5% and 10% levels, respectively. Standard errors are the in parentheses. The table shows the regression for the sector number 30: Consumer Staples. The top panel presents the regression coefficients of interest. Column (1) has the results for the Gini variable. Column (2) has the results for the dummy variable. Column (3) has the results for the Top 5% variable.
Table 7 shows the results for the Information Technology sector. Different than the others, M&A is statistically significant in the 10% level and positive for the three methods
used. As in the other results, the dummy variable presents higher impact than the other two. This implies that a higher M&A intensity would signal a higher change of a dictatorship.
When analyzing R&D returns we notice that they are not significant, but they are all positive and the dummy overestimates the impact. The interaction term, is statistically significant in the 10% level but negative. This implies that companies which make use of both methods are more likely to have a democratic structure. For the dummy variable it is not relevant, even though the impact is also negative.
In conclusion, companies in the information technology sector that have a higher M&A would more likely have a dual stock ownership. This makes sense because usually shareholders care more about big events, such as M&A, than internal projects linked to R&D and therefore would most likely vote for those. Because of a higher activism in this type of events, companies that rely on M&A would rather protect themselves. This is also explained by the fact that shareholders care more about short-term returns, while company owners think about the long-term and making their companies more successful. A typical example is the case of Yahoo and Facebook in 2006, where Yahoo wanted to buy Facebook but due to the dual-class ownership Facebook was not sold and shareholders could experience even a higher value for their shares then they would receive with the sale. In this case, shareholders were only seeking the short and right investment, while Facebook owners were betting in the long-term success of the company.
In summary, when analyzing the whole sample the values are not statistically significant. But when comparing each industry separately, R&D impact is also not statistically significant for the three sectors. M&A activity is only significant for the information technology, and the impact is positive. Hence, for information technology a higher M&A expenditure would increase the probabilities of a voting transference, and therefore the use of a dual stock ownership.
Table 7: Estimated Effect of Returns on Voting Rights for Sect. 45 – Information Technology
Voting Right Allocation
GINI Variable (1) Dummy Variable (2) Top-5% Variable (3) Beta R&D 0.0149 0.0242 0.00831 (0.0115) (0.0224) (0.00582) Beta M&A 0.758* 1.241* 0.403* (0.394) (0.746) (0.219) R&D M&A -0.482* -0.909 -0.270* (0.291) (0.558) (0.156) Period 2 -0.0783** -0.127* -0.0413* (0.0393) (0.0745) (0.0223) Period 3 -0.0371** -0.0591* -0.0185* (0.0168) (0.0321) (0.00982) Leverage -0.00126 -0.00619 -0.00107 (0.00512) (0.0110) (0.00182) Size 0.00315 0.00749 0.00194* (0.00209) (0.00489) (0.00110) Constant 0.0284 0.0487 0.0108** (0.0173) (0.0387) (0.00491) N 3,174 3,174 3,174 Adj. R-square 0.026 0.013 0.040
Company Cluster Yes Yes Yes
Fixed Effects Yes Yes Yes
***, **, * represent statistical significance at the 1%, 5% and 10% levels, respectively. Standard errors are the in parentheses. The table shows the regression for the sector number 45: Information Technology. The top panel presents the regression coefficients of interest. Column (1) has the results for the Gini variable. Column (2) has the results for the dummy variable. Column (3) has the results for the Top 5% variable.
D. Concentration of voting rights and its direct impact
All the results presented could not explain changes in voting rights, besides to the information technology sector but still with a low level of significance. Now I attempt to test if companies that concentrate voting rights increase R&D expenditure/M&A activity.
To test for it I estimate !&/%/2&#% as a function of I5J5%0+ and other measures, as ROA, leverage, size and period. The model is as shown below:
!&/$% = ( + *+%,∗ I5J5$%0++ *1%,∗ !"#$%0++ *3%,∗ 4567$%0++ *8%,∗ 97:.$%0++ *<%,5. O7C5PL + D$%, (3)
2&#$% = ( + *+%,∗ I5J5$%0++ *1%,∗ !"#$%0++ *3%,∗ 4567$%0++ *8%, ∗ 97:.$%0++ *<%,5. O7C5PL + D$%, (4)
The reasoning is that companies’ R&D expenditure is related to the voting structure adopted by the company the year before. I use the same logic to M&A activity.
The results are sub-divided in two: first I present the overall results clustered by the new industry taxonomy and then the results per sectors. Also, the first tables are always related to the R&D expenditure followed by the M&A activity table.
D.1 Clustered by the new industry taxonomy
Results are presented in Table 8 for R&D and Table 9 for M&A.
From the results below, we can notice that only R&D seems to be impacted by the voting inequality variables. In Table 8 we see that the three different variables are significant, with gini and top-5% being 5% statistically significant and the dummy variable in the 10% level, meaning that the expenditure value spent in the current year is correlated to the voting structure used the year before. The values for R&D are always positive, concluding that companies that have a unequal voting structure the year before are more likely to have a higher expenditure in R&D the following year. This makes sense since managers do not need to seek approval from outside shareholders.
When focusing on the other variables, we can notice that ROA is only significant for M&A, with a 1% significance level, while for R&D it is only significant for the gini variable, but with a 20% level of significance only. It is important to point out the different signs as well. For M&A a higher ROA impacts the M&A activity positively, while for R&D it is negative. When companies have a higher return on their assets, they are more likely to make use of M&A strategies than R&D.
Table 8: Estimated Effects of Returns on R&D Expenditure
R&D Expenditure Reasoning
R&D (1) R&D (2) R&D (3) Gini 0.0158** - - (0.00743) - - Dummy - 0.00730* - - (0.0318) - Top-5% - - 0.0370** - - (0.0169) ROA -0.0132* -0.0132 -0.0132 (0.00798) (0.00798) (0.00798) Period 2 0.00224* 0.00222* 0.00222* (0.00115) (0.00114) (0.00115) Period 3 0.0132*** 0.0132*** 0.0132*** (0.00215) (0.00215) (0.00215) Leverage -0.0149** -0.0149** -0.0149** (0.00718) (0.00717) (0.00717) Size -0.0398*** -0.0398*** -0.0398*** (0.00436) (0.00436) (0.00436) Constant 0.348*** 0.348*** 0.348*** (0.0318) (0.0318) (0.0318) N 8,782 8,782 8,782 Adj. R-square 0.157 0.158 0.158
Industry Cluster Yes Yes Yes
Company Cluster Yes Yes Yes
Fixed Effects Yes Yes Yes
***, **, * represent statistical significance at the 1%, 5% and 10% levels, respectively. Standard errors in parentheses are clustered by the new sector category created and by company. The top panel presents the regression coefficients of interest. Column (1) has the results for the Gini variable. Column (2) has the results for the dummy variable. Column (3) has the results for the Top 5% variable.
The periods seem to also have an impact in it. Period 2 is significant and positive for R&D, while it is not significant for M&A and negative. Hence, during this period R&D activity was more important. This is in line with Figure 7, which the return on R&D is higher than M&A. Therefore, during the crisis companies were mainly focusing on R&D. When turning to Period 3, after the crisis, it is significant in the 1% level for both R&D and M&A and positive. Hence, the period after the crisis companies could increase their R&D expenditure and M&A activity.
Table 9: Estimated Effects of Returns on M&A Activity
M&A Activity Reasoning
M&A (1) M&A (2) M&A (3) Gini 0.00124 - - (0.0183) - - Dummy - -0.000573 - - (0.00964) - Top-5% - - 0.0304 - - (0.0305) ROA 0.0675*** 0.0675*** 0.0674*** (0.0120) (0.0120) (0.0120) Period 2 -0.000832 -0.000812 -0.000961 (0.00295) (0.00295) (0.00295) Period 3 0.0236*** 0.0236*** 0.0235*** (0.00387) (0.00387) (0.00387) Leverage -0.162*** -0.162*** -0.162*** (0.0212) (0.0212) (0.0212) Size -0.0436*** -0.0436*** -0.0435*** (0.00729) (0.00729) (0.00729) Constant 0.396*** 0.396*** 0.396*** (0.0554) (0.0554) (0.0553) N 13,107 13,107 13,107 Adj. R-square 0.0.34 0.0.34 0.034
Industry Cluster Yes Yes Yes
Company Cluster Yes Yes Yes
Fixed Effects Yes Yes Yes
***, **, * represent statistical significance at the 1%, 5% and 10% levels, respectively. Standard errors in parentheses are clustered by the new sector category created and by company. The top panel presents the regression coefficients of interest. Column (1) has the results for the Gini variable. Column (2) has the results for the dummy variable. Column (3) has the results for the Top 5% variable.
For R&D and M&A leverage is significant in 5% and 1% level, respectively. For both the impact is negative, meaning that companies with a higher leverage are less likely to spend in R&D or in M&A. This makes sense because companies with a higher leverage have more difficulties to access the market and therefore to have the cash necessary to such transactions.
Size is also significant for both in 1% level. Again, for both the impact is negative. Hence, larger companies are more likely to spend less in R&D and M&A expenditure. For
