Does international experience of boards in emerging markets enhance firms’
performance? A case study of Brazil
University of Groningen Faculty of Economics and Business Master in International Economics and Business
Student: Sylvia T. Inoue de Oliveira s.t.inoue.de.oliveira@student.rug.nl
S-number: S3005836
Supervisor: prof. dr. van Ees, H.
University of Groningen
Co-Assessor: dr. Rao Sahib, P.
University of Groningen
2 Abstract
This study aims to investigate the benefit of international diversity in the composition of corporate boards in industrial firms in an emerging country. We study the contribution that international experience of corporate board members has on the performance of listed Brazilian industrial firms. We use return on assets (ROA) as dependent variable associated to firms’ performance and three characteristics of international experience as independent variables: foreign nationality, study and work international experiences of national corporate board. We carry out a cross-section analysis and use a sample of 75 listed Brazilian companies, compounded by 614 corporate board members. Possible reverse causality is a common concern on this type of study, and we address this issue by performing two types of regressions:
ordinary least squares and two-stage least squares. The regressions indicate a significant correlation between foreign nationality and firms’ performances. However, no conclusive results are found concerning study and work international experiences of national board members. Inconclusive results are also found after testing if international experience has led to better performance in older firms. In summary, this work intends to shed light on international diversity of the board in an emerging country.
Keywords: corporate board; international experience; nationality, firm performance; Brazil
3 Table of contents
1. Introduction ... 4
2. Literature review and hypothesis ... 7
2.1. Corporate governance and the corporate board ... 7
2.2. Relevance of diversification of the board ... 7
2.3. Corporate board and international experience ... 8
2.4. Board members’ international experience, companies’ performance and Brazil ... 9
3. Data and Methodology ... 11
3.1. Data ... 11
3.2. Model specification ... 12
3.3. Definition of variables ... 14
4. Empirical analysis... 16
4.1. Sample characteristics ... 17
4.2. Specification tests ... 18
4.3. Correlation ... 21
5. Econometric analysis ... 24
5.1. Hierarchical regressions ... 24
5.2. Endogeneity and robustness test ... 24
5.3. Hypotheses tests results ... 25
5.4. Limitations ... 28
6. Discussion and conclusion... 28
7. Future research ... 30
8. References ... 31
9. Appendices ... 38
Appendix 1: List of the analyzed companies ... 38
Appendix 2: Distribution of companies per industry ... 42
Appendix 3: Hierarchical regressions – robustness test ... 43
Appendix 4: Endogeneity tests ... 44
4 1. Introduction
Since the scandals of the 2000’s amongst US companies and the financial crisis in 2007- 2008, attention has been dedicated to corporate governance, as its failures and weaknesses proved to be, to some extent, responsible for the crisis (Kirkpatrick, 2009). Thus, crises have empowered the relevance of studying corporate governance and its mechanisms.
Understanding corporate governance structures is important, as it influences the “efficiency of the market for corporate control and may have a positive impact in the firm valuation and performance” (da Silva & Leal, 2005).
However, when referring to corporate governance, one should keep in mind that the expression is very broad, pertaining to a set of mechanisms or institutions seeking to improve companies’ value and performance (Daily, Dalton, & Cannella, 2003). One mechanism of particular relevance is the corporate board; as stated by Fama and Jensen (1983), it is the
“common apex” in a company’s decision and control. It has the role of helping to solve problems between principals and agents but also, as an indicator for better corporate governance, that may influence the investors’ risk perception (Braga-Alves & Shastri, 2011;
Hermalin & Weisbach, 2003).
In describing a firm’s board of directors, several aspects may be taken into consideration; for instance, the board’s size, presence of independent members, cognitive features of its members. In particular, the Upper Echelons Theory states that experiences, values and personalities of the board members influence their understanding of the situations they face and their decisions, which then affect the companies’ strategy and performance outcomes (Hambrick, 2007). According to Hambrick (2007), demographic characteristics of executives are used as (admittedly imperfect) proxies for their cognitive frames. The demographics of corporate boards is an increasingly active topic of research (Arnegger, Hofmann, Pull, & Vetter, 2014; Pugliese, Bezemer, Zattoni, Huse, Van den Bosch, &
Volberda, 2009).
Of particular interest is the effect of various aspects of the board composition. Several studies have been carried out in this direction and contradictory results have been found. Some works have found a positive correlation between certain characteristics of the corporate board members and the company’s performance (Amaral-Baptista, Klotzle, & Campelo de Melo, 2011; Ameer, Ramli, & Zakaria, 2010; Dahya & McConnell, 2007), whereas others did not succeed on finding significant results (Bhagat & Black, 2002; da Silva & Leal, 2005; Klein, 1998; Santiago-Castro & Baek, 2003).
5 Regarding these conflicting results, it should be mentioned that studies indicate that a country’s characteristics can influence the relationship among the corporate governance variables and the prediction of firm’s performance, indicating that one size may not fit all.
Hence, findings can differ depending on the place and time of data observation (Black, de Carvalho, & Gorga, 2012; Santiago-Castro & Baek, 2003), which means that the setting may also influence the empirical results. In the literature review below, we noticed that a larger number of literature has analyzed US firms (Dahya & McConnell, 2007), whereas little is known about emerging markets (Black, de Carvalho, & Gorga, 2010). Consequently, performing the study in an emerging market (EM) is relevant and will add information to the current literature.
Brazil is one of the largest EM countries (Black et al., 2012; da Silva & Leal, 2005).
Since the late 1970s, laws and regulations have been implemented aiming at better corporate governance practices. This objective gained even more relevance after the economic opening in the 1990s, as stated by a report on a Brazilian newspaper “O Estado de São Paulo” (2014).
Due the opening and monetary stabilization, corporate boards broadened their relevance and participation in companies' strategies (Barros Serrão & Borges, 2005; Black et al., 2010). Local institutions – Comissão de Valores Mobiliários (CVM)1, Instituto Brasileiro de Governança Corporativa (IBGC)2, Banco Nacional de Desenvolvimento Econômico e Social (BNDES)3 – were assigned to regulate, control and improve corporate governance practices (da Silva &
Leal, 2005; Da Silveira, 2004). According to these institutions, corporate governance is a key factor for increasing competitiveness and attracting foreign investments to Brazilian economy (Barros Serrão & Borges, 2005; Instituto Brasileiro de Governança Corporativa, 2014).
The IBGC and CVM use regulation and recommendations to stimulate the transition from a family-owned culture to a more independent corporate culture. Most of the Brazilian companies are controlled by the founder's family and family members (Da Silveira, 2002; Da Silveira, 2004). According to a study done by KPMG (2016), only 16% of the companies founded by Brazilian families have an independent president and for most of them, having a
1 CVM: Brazilian securities exchange commission is an organization created in 1976 by the government to control, create, standardize financial reports and communications to investor, and discipline the Brazilian stock market.
It is connected to the Department of Treasure. They have as model the US standards.
2 IBCG: it is a Brazilian NGO created in 1995. It is the national reference for corporate governance best practices.
The organization is responsible to promote training, conferences and networking among business professionals.
It is also responsible to do research on the Brazilian corporate market.
3 BNDES: founded in 1952 is a public federal bank, focused in leveraging development of the Brazilian economy by financing companies’ long term projects.
6 foreigner in their top management teams is not part of their short-term plan. This indicates that, despite the rules, guidelines and recommendations to comply with good corporate governance practices, national companies are reluctant to change their traditional structures.
Generally speaking, the fact that a business is family owned does not promptly indicate a worse performance compared to non-family-owned businesses (González, Guzmán, Pombo,
& Trujillo, 2012; Miller, Le Breton-Miller, Lester, & Cannella, 2007). However, in the Brazilian setting some considerations must be made. Brazilian family-owned companies are related to a shareholding concentration (Saito & Dutra, 2006), where few people have vote power on corporate-board meetings, having the power to decide on a company’s strategies.
Power concentration can indicate two major issues: aggravation of agency problems and possible appointment of family members, who may not be the most qualified persons, to perform leadership roles (Da Silveira, 2002; Martínez, Stöhr, & Quiroga, 2007). The inclusion of an independent professional board member is a good measure to balance power concentration, mitigate agency problems, communicate a commitment to transparency and best governance practices (Anderson & Reeb, 2004; Oxelheim & Randøy, 2003).
Aside from independent board members, one may ask whether or not it is generally valuable for companies to include board members having an international background. Here, an international background is defined as being of a non-Brazilian origin or having work or study experience outside the country. One might conjecture that the diversity in personality, values, network and capabilities that is implied by a board member coming from abroad has a positive impact in Brazilian companies' performances. This diversity could bring a positive impact in the realization of two roles of the board: resource access and strategy advising.
Therefore, in this study we propose to answer the following question:
Does international experience of corporate board members positively impact Brazilian companies' performances?
Our research tries to answer the research question whether Brazilian listed industrial companies that include a board member with international experience perform better or not. In this regard, we contribute to the literature about the international diversity of the board in an EM economy.
7 2. Literature review and hypothesis
2.1. Corporate governance and the corporate board
Corporate governance can be defined as the “determination of the broad uses to which organizational resources will be deployed” and the resolution of organizational (shareholders and stakeholders) conflicts (Daily et al., 2003). Hence, we can say that corporate governance solves conflicts and determines strategies (Prevost, Rao, & Hossain, 2002). There is indication, based on Klapper and Love (2004), that better corporate governance practices are related to better performance in EM. Thus, we can expect that a high level corporate governance may bring more value to Brazilian companies (Melo, Batista, Macedo, & Costa, 2013).
Studies pointed out that corporate governance uses mechanisms to define strategy and solve conflicts; for example, one such internal mechanism is the structure of the corporate board and its monitoring role, that mitigates agency problems (Daily et al., 2003; Prevost et al., 2002). The Upper Echelons Theory states that members of the corporate board, through their decisions, can drive firms’ performance (Hambrick, 2007). Their decisions follow from their experience, values and personalities. These characteristics, according to Hambrick (2007), can be measured by some observable characteristics, such as age, nationality, educational and work backgrounds. In sum, the corporate board, “the apex of corporate governance” (Fama & Jensen, 1983; Gillan, 2006), is known to have the power over many aspects connected to a firm’s action, including corporate strategies, monitoring and influencing important decisions.
Understanding the way, the corporate board interacts with companies’ performance is key to define practices, codes and corporate policies.
Regarding board composition and more specifically the international experience of the board, prior works have found that certain board members’ background characteristics are relevant to enhance firms’ performance (Oxelheim & Randøy, 2003) and influencing, for example, a firm’s internationalization strategy (Oxelheim L., Randoy T., Gregoric A., &
Thomsen S., 2013).
2.2. Relevance of diversification of the board
As one might expect from the relevance of the corporate board discussed in the previous section, the interest on the corporate board is considerable on corporate governance literature, specially focusing on the agency theory area (Van Ees, Gabrielsson, & Huse, 2009).
8 Regarding the relation between corporate board and firm performance, one central aspect that is studied is diversity. Diversity of corporate boards means the heterogeneous composition of this institution. A person’s gender, age and nationality are demographic variables and easy diversity aspects to observe. Others, such as the contribution of knowledge background and personality are non-demographic and harder to access (Mishra &
Jhunjhunwala, 2013b) but proxies can be used (Hambrick, 2007). The diversity of the corporate board leads to a possibility of having more variety of skills and knowledge amongst its members. Diverse people take different responsibilities, assuming different roles during the decision-making process contributing to a better strategy and possibly to a better result (Hambrick, 2007; Milliken & Martins, 1996; Mishra & Jhunjhunwala, 2013a). Hence, adding new views and opinions can lead to better strategies and decisions (Daily et al., 2003; Van den Berghe & Levrau, 2004).
2.3. Corporate board and international experience
Boards have three main roles mentioned in the literature. The first one is called service role or resource access, when a member of the board contributes to the company’s reputation and can collaborate with their network in order to enhance performance. The second role is control or monitoring; it is related to solving agency problems, as the board member can ensure the shareholders’ interests by improving the monitoring of managers. The third one is the strategy or advisory role: boards are entitled to determine the company’s main strategies by counseling and advising the CEOs (Oxelheim L. et al., 2013; Pearce & Zahra, 1989; van Ees, van der Laan, & Postma, 2008).
All these three roles can be simultaneously present when we talk about corporate boards. To study their influence on a company’s performance we can analyze single characteristics of corporate board members, as has been done in the literature. One example is the analysis of the contribution of international experience of board members. It may bring the contribution of a different corporate culture, possibly more independency (e.g. of the board member in relation to the founder board member) and commitment to higher standards of governance (Oxelheim & Randøy, 2003). Based on previous studies, nationality, relevant work and study experience are proxies to international experience (Oxelheim L. et al., 2013;
Oxelheim & Randøy, 2003). In both studies the international experience indicate commitment of the company to better practices and growth of the business.
9 The Brazilian corporate scenario, although presents formal commitment to improve corporate governance, protect minority shareholders, reveals a strong power concentration in the businesses run by family (Brandão, 2016). The Brazilian corporate market is in its majority compounded by family-owned businesses and this could be a sign of high probability of agency problems (Da Silveira, 2002). Therefore, having a foreign member could be an indication of evolving from the old traditional governance structure to a more professionalized and independent type of corporate governance that will possibly enhance the company’s performance. Furthermore, corporate board members with international experience could add cognitive diversity that may lead to a better firm performance (Milliken & Martins, 1996).
2.4. Board members’ international experience, companies’ performance and Brazil
Brazilian organizations – IBGC, CVM, BM&FBovespa4 – and the federal government have established laws, guides and codes of best practices for companies seeking their acceptance and compliance. Such regulations are mandatory for listed companies and optional, but highly recommended, for non-listed ones. A recent analysis of the corporate governance evolution in Brazil has indicated that government and regulatory organizations understand the relevance of the topic and try to motivate companies to comply with mandatory rules and codes of best practices (Instituto Brasileiro de Governança Corporativa, 2014). According to IBGC’s researches, that can be retrieved on their website5, improvements in overall corporate governance practices and awareness were observed.
In spite of the improvement efforts and awareness, corporate governance and legal system showed to be weak in recent studies (Black et al., 2010; Black et al., 2012). This could influence investors’ risk perception, for example. Additionally, other recent studies about the Brazilian corporate market have shown that the majority of companies has family members as part of the corporate board (Brandão, 2016). The finding is expected according to EM corporate profile, where most of the businesses are run by families, which can increase agency problems (Ameer et al., 2010). Such state of affairs communicates higher risk to shareholders.
In cases where regulations, culture and codes are weak, companies can communicate their commitment to compliance, better performance and a stronger corporate governance system by adding more outside or foreign members to their board (Aguilera & Cuervo-Cazurra,
4 Brazilian stock market
5 http://www.ibgc.org.br/
10 2004; Ameer et al., 2010; Da Silveira, Leal, Carvalhal, & Barros, 2007). Besides communicating a better corporate governance, the international experience of board member as a whole (foreign nationality, international study and work experiences) can provide companies with a unique knowledge, different views and professional backgrounds. This makes corporate board members with international experience a valuable asset to the company (Nigh & Athanassiou, 1999).
Two main roles can enhance performance. First, monitoring by signaling good will, more commitment to a family-independent board and therefore less agency problem (Ameer et al., 2010; Claessens & Yurtoglu, 2013; Pearce & Zahra, 1989). Second, resource or service role, based on the fact that international experience of the board can bring not only important network but also diverse knowledge (Milliken & Martins, 1996), better investment decision, different views and strategies that provides competitive advantage to the company (Mahadeo, Soobaroyen, & Hanuman, 2012; Pearce & Zahra, 1989). The effects of the influence of international experience is expected to be seen on the company’s results, therefore, a higher return on assets (ROA). We measure international experience with three main indicators:
foreign nationality, study and work experience abroad. We can state the following hypothesis:
H1: International experience of the board members enhances a firms' performance in Brazilian listed companies.
Firm age may strain managerial action due to managerial inertia forces (Forbes &
Milliken, 1999), which indicates that older firms will may benefit from the change that a member of the corporate board with international experience will bring. Additionally, a study found that age showed to be positively correlated to firms’ performance (Oxelheim & Randøy, 2003). We expect to find that international experience of members of the corporate board impacts older firms rather than younger ones, leading to the second hypothesis:
H2: International experience of board members has a greater positive impact on older firms’ performance than it has on younger ones.
11 3. Data and Methodology
3.1. Data
The data used on this study is firstly retrieved from the Bureau van Dijk database (Orbis, 2016). The sample consists of public information on Brazilian listed industrial companies as non-listed companies provide incomplete public information about their board members and financial reports. We do not take into account financial institutions. One reason is that they should be analyzed separately as they are under different and sterner regulations than non- financial companies. We also excluded public organizations, due to the fact that their job positions are occupied by people who were employed based on qualification exams, which is a different procedure than the one followed by private companies.
According to the parameters above, we came up with 410 listed companies. However, when we restricted the search to industrial companies, excluding financial and public institutions, the sample size reduces to 281. From this preliminary sample we exclude some of the ‘gas, water, electricity’ firms that have a joint administration with governmental institutions. In Brazil, some of companies of this sector are greatly influenced by the government when it comes to the appointment of members of the corporate board. Such utility companies also follow a specific legislation, which may make them very similar to public companies. Companies with incomplete or missing information are also excluded from the sample. Therefore, the final sample consists of a reminder of 75 industrial companies and it 614 board members.
Bureau van Dijk (Orbis, 2016) retrieves information from annual and official reports of the listed Brazilian companies. These reports comply with the International Accounting Standards Board and federal laws (KPMG Brazil, 2009; Law 11.638, 2007; Law 6.404, 1976), which provides a reliable database of financial information. However, we encounter inconsistent and inaccurate information regarding the composition of the board. The critical limitation is the data availability for former board compositions. For instance, less than 10 companies provide information for years prior to 2015. That is why we have tried to find information from years prior 2015 on each company through their websites. However, only weak and not satisfactory information about former board members can be found. There is no standard amongst the companies’ websites. As a result, we decide to concentrate our effort in a cross section analysis providing an indication of the interaction between the international experience of the board on the performance of Brazilian companies.
12 Even for 2015, we have to complement our data base and thus, obtain more accurate information as described in the following: we access each company’s website and retrieve information on corporate board structure, names of the members of the corporate board, corporate board members’ curriculum and age (when available), meeting reports and minutes.
This way we can build our final database with more accurate information on who are the members of the corporate board in 2015, and their background profiles. When the company’s website lacks information, we also access other resources such as Reuters, Bloomberg and Forbes’ websites. The list of analyzed companies and their respective websites will be available at the Appendix 1.
3.2. Model specification
To test our hypotheses, we employ a model that has been developed for a similar study on Nordic firms in 2003, which is consistent to the empirical equations models seen on the work of Hermalin (2003). To test our hypotheses, we use ordinary-least square (OLS), but also two-stage-least square (2SLS) regressions in order to cope with possible endogeneity issues.
Our model considers almost the same variety of independent variables to minimize possible specification biases (Oxelheim & Randøy, 2003). Overall, we benefit mostly from the work done by Oxelheim and Randøy (2003) by using most of the independent variables they indicate on their research, as such variables are also relevant for our model.
Our goal is to isolate the effect the board composition has on ROA (our dependent firm’s performance variable) by controlling for several determinants of firm performance.
Thereby, we follow Oxelheim and Randøy (2003), who separate their regressors in two categories to facilitate the understanding of the groups of variables. One group is named
‘international corporate governance’ that comprehends corporate governance variables, which indicates the company’s connection with foreigners’ markets. For example, this could have an impact on having foreigners in the corporate board. The other group of independent variables is ‘general corporate governance’ with corporate governance variables that are pertinent to this type of study, but they do not necessarily indicate a connection to international markets.
Following Oxelheim and Randøy (2003), we control for international corporate governance variables such as foreign ownership, foreign listing and foreign subsidiary.
Regarding general corporate governance variables, we control for board size (Amaral-Baptista et al., 2011; Yermack, 1996), firm age (Forbes & Milliken, 1999), firm size (Da Silveira, Leal,
13 Carvalhal, & Barros, 2010; Daily, Johnson, Ellstrand, & Dalton, 1998; 1999) and board independence that is a central element of corporate governance in Brazilian corporate governance practices and, for that, much discussed (Black et al., 2012).
One potential problem of our type of research, indicated by past studies (Da Silveira et al., 2010; Hermalin & Weisbach, 2003; Oxelheim & Randøy, 2003), is the endogeneity due to reversed causality. This occurs when the dependent variable may have a casual impact on one or more explanatory variables (Hill, Griffiths, & Lim, 2008; Verbeek, 2008). In our case, the effect is the following: firms with higher ROA may be more attractive for executives to join as board members, in particular individuals with special qualifications, such as international experience. There are some techniques to treat this potential endogeneity problem. One often used methodology is to employ 2SLS regressions, which is taken into account (Black, Kim, &
Jang, 2006; Oxelheim & Randøy, 2003).
To perform the 2SLS regression, that treat endogeneity problems, one needs to find instrumental variables that are not weak to define the explanatory variable (Hambrick, 2007).
In this study we use foreign ownership as instrumental variable. Therefore, we first use an OLS regression, followed by a 2SLS where foreign ownership is instrumental variable to substitute for international experience of the board (foreign nationality). To do a 2SLS regression we need at least one instrumental variable (Hill et al., 2008).
The OLS model is as follows:
Firm Performance = ∝ +𝛽1∗ Foreign Nationality + 𝛽2∗ Study + 𝛽3∗ Work + 𝛽4∗ Board Independence + 𝛽5∗ Board Size + 𝛽6∗ Firm Size + 𝛽7Firm Age + 𝛽8∗ Foreign Listing + 𝛽9∗ Firm ownership +
𝛽10∗ Foreign Subsidiaries + 𝜀
Where α is the intercept which is unknown and constant, indicating the value of ROA or firm performance when the independent variables assume value zero. βn (n=1,2,3…n) are the coefficients for each of our independent variables. Finally, 𝜀 is the standard error term. We provide a deeper interpretation of each variable in the next section.
For the use of 2SLS, let us give a brief overview of the intention behind this regression model, (e.g. Verbeek, (2008); Hill et al. (2008)). As in the usual OLS regression, we start with a linear statistical model, which gives a dependent variable y as a linear combination of variables 𝑥1, ..., 𝑥𝑘 (regressors), plus an error term 𝜀. The coefficients of the linear combination, denoted 𝛽1, ..., 𝛽𝑘, are the parameters we would like to estimate. Now, suppose we believe one
14 of the regressors (say, 𝑥1) is endogenous, that is, has non-zero correlation with the error term.
In that case, regular OLS would produce biased and inconsistent estimators of the beta coefficients, because the effect of changes of 𝑥1 on y is affected by the correlated error term.
This problem can be bypassed when instruments for the problematic regressor x1 are available. Instruments are new variables 𝑧1, 𝑧2, ..., 𝑧𝑟 that, albeit possibly correlated to 𝑥1, are uncorrelated with the error term epsilon. When they are available, we can carry out a 2SLS regression by following the two stages:
1) we produce an OLS regression of 𝑥1 (foreign nationality) using the instruments (firm age and foreign ownership) as regressors. We thus obtain an estimate 𝑥1∗ for 𝑥1;
2) we then go back to the original regression equation, but replacing 𝑥1 by 𝑥1∗, and again apply an OLS regression to obtain estimators of the beta coefficients.
3.3. Definition of variables
The dependent variable firm performance is defined by the return on assets (ROA) from the respective analyzed year-end, in our case 2015. ROA is an accounting measure that is available for all firms in our sample, while not enough data for calculating Tobin’s Q is present.
This accounting measure was also used in another study regarding board composition in Latin American companies (Santiago-Castro & Baek, 2003). We consider it to be a reliable and a standard ratio, because Brazilian listed companies have, by law, their results audited and reported in a standard way.
As explanatory variables for international experience, we mean either foreign nationality or international study and work experience of nationals6. To arrive at their final ratios (total number of people with international background divided by the number of board size), we considered the board composition at the year-end of our observation and used the following assumptions:
First, our data collection suggests that average age of the corporate board members is between 40 and 60 years. We assume, based on the rationale of Oxelheim et al. (2013), that geographical mobility 20 to 40 years ago was not as frequent as today, so it is most likely that the board members of our sample have studied in their country of nationality. Hence, whenever there was no information about international study background we assumed value zero.
6 People with Brazilian nationality.
15 Second, we took into account that studying and working abroad mean that the person has studied and/or worked in a country that is not of his/her nationality7. Third, in cases where a board member has double nationality, we counted zero for foreign nationality whenever the first nationality was Brazilian. Therefore, foreign nationality indicates the ratio of foreigner members present on the corporate board. Study and work indicate the ratio of national people who have studied and worked outside Brazil respectively in any time prior 2015. Additionally, we assumed zero when there was no information about international work experience.
About the foreign ownership, due to the absence of data about the holders of the companies’ equity, we consider 1 when on the company’s website there was mention of an international ownership and zero otherwise. We also used dummy variables for foreign subsidiaries and foreign listing, 1 when the company presented subsidiaries abroad and 0 otherwise. We can expect that having subsidiaries or foreign listing may influence the company to have a foreign member on the board (Luo, 2005; Oxelheim L. et al., 2013; Oxelheim &
Randøy, 2003), or even invest on sending national board members to study abroad or to work abroad in one of the subsidiaries, using these dummies we can separate the effect of these characteristics on having a member with international experience.
Regarding the ‘general corporate governance’8, the board size is the number of board members at the end of the sample year. The same rationale is for the board independence, we obtained the number of independent board members at the year-end of our analysis and calculated the ratio between the total number of independent board members and the number of total members of the board (board size).
The general control variables firm size, in line with its use in previous studies, is measured by the natural logarithm of total turnover of the sample year-end, as the size amongst our large firms can vary. It is important to have this variable, as the companies differ in size and a previous study have shown a positive correlation between firm size and firm performance (Wallace & Naser, 1996). So this variable would help to separate the impact of firm size and, for instance, board member nationality on firms’ performance. The other control variable is firm age, which due to “inertial forces that constrain organizational action” may relate negatively to a firm’s performance (Forbes & Milliken, 1999, p.501); we measure firm age by the natural logarithm of the number of the years between 2015 and the company’s founding year. The total size of the sample is sufficient to perform a statistical analysis, although its
7 We assumed zero in cases with lack of information about board members’ background.
8 See section 3.2 for the meaning of this expression
16 showed to be limited when considering companies per sector. Therefore, to not compromise the degrees of freedom, we ruled out the possibility of using industry dummies variables for industry sector. We provide a table with the distribution of the companies per industry in Appendix 2.
4. Empirical analysis
This section will present the statistic description of our sample, the specification tests and the results of our regressions. First we introduce the table of acronyms and short definition for all independent variables that were discussed on Section 3.
Table 1 - Variables acronyms and descriptions
Variable Acronym Definition
- International Experience Nationality NAT
Member with foreign nationality; ratio between the number of foreign members on the board to the total size of the corporate board.
Study STD
Study experience abroad; ratio between the number of national members on the board who have any international study experience to the total size of the corporate board.
Work WRK
Work experience abroad; ratio between the number of national members on the board who have any international work experience to the total size of the corporate board.
- General Corporate Governance Board
Independence BIND Ratio between the number of independent board members to the total size of the board.
Board Size BSIZE The total number of board members; only effective members were accounted. Alternate board members were not accounted.
Firm Size FSIZE The natural logarithm of each company's 2015 turnover.
Firm Age FAGE The natural logarithm of the difference between 2015 (year of observation) and the incorporation date of each company.
- International Corporate Governance
Foreign Listing FLIST Dummy variable, 1 for companies listed at least in one foreign stock market.
Firm
Ownership FOWN Dummy variable, 1 for companies with foreign ownership or partnership and zero otherwise.
Foreign
Subsidiaries FSUB Dummy variable, 1 for companies with one or more foreign subsidiaries and zero otherwise.
17 For our dependent variable of firm performance, we kept the acronym for return on assets (ROA).
4.1. Sample characteristics
Table 1 shows the summary statistics of our sample. The sample has 75 listed companies. Despite all companies are classified as very large on the Bureau van Dijk database, we can observe variation among them regarding the variable FSIZE, which measures firm size based on a company’s turnover, ranging from 8.97 to 41,727.9 million of USD. The table shows that the mean of 9.9% for the board members are foreigners, we can say that it is a slightly smaller mean compared to what Oxelheim and Randøy (2003) found for their sample (12.6%) of Nordic firms, just counting Anglo-American board members on a time from 1996 to 1998.
We did not expect to have foreigners as majority in the corporate board, as the great part of Brazilian companies are family-owned businesses, who prefer having family member occupying top management positions.
Study (STD) and work (WRK) experience show higher percentages, which is not surprising for very large companies being likely to engage in international business. Members who have studied abroad has a particularly high mean (24.1%), this value is not very low, thus one could claim that this is due to the fact that study mobility has been increasing nowadays.
However, for our study it was not possible to discriminate the time when the degree was obtained, therefore, further analysis is necessary to obtain enough descriptive information about the graduation time and the type of degree. The percentage of foreign work experience (WRK) is on average 12.8% for the board’s in our sample
18 Table 2 - Summary statistics
Variables Obs Mean Std. Dev. Min Max Dependent Variable
ROA 75 2.929 8.569 -28.472 26.734 International Experience
NAT 75 0.099 0.180 0 1
STD 75 0.241 0.194 0 0.75
WRK 75 0.128 0.143 0 0.60
General Corporate Governance
BIND 75 0.225 0.204 0 0.875
FSIZE 75 13.987 1.468 9.102 17.547
BSIZE 75 8.187 2.639 2 16
FAGE 75 3.461 0.812 1.099 4.963 International Corporate Governance
FLIST 75 0.093 0.293 0 1
FOWN 75 0.080 0.273 0 1
FSUB 75 0.467 0.502 0 1
We can also observe a yet low, but positive mean for ROA (2.93%). Regarding this aspect, the year of our observation is 2015, in which corruption scandals involving Petrobras, some big construction companies and the Federal government came to the surface. This is likely to have resulted in a negative impact on the Brazilian economy (The Economist, 2016).
Regarding the size of the corporate board, we found values varying from boards with minimum 2 and maximum 16 members, with a mean of 8.2 members. The mean is between 5 to 11 members that is the range recommended by the IBGC’s recommendations (Instituto Brasileiro de Governança Corporativa, 2014). But there are some companies that have not complied to the IBGC’s recommendation.
Regarding the dummies for international corporate governance variables, we see that only few companies are listed in foreign stock market (9.3% of our sample) and 8.0% showed to have a foreign ownership. On the other hand, almost half has a foreign subsidiary.
4.2. Specification tests
As we are using OLS regression, some specification tests have to be done. The common tests for this type of regression are discussed as follow.
19 Testing for Heteroskedasticity: One way to test for heteroskedasticity is by looking at the scatterplot of residuals, which is more of a visual instrument. The graph indicates that heteroskedaticity may be present. By performing the Breusch-Pagan test we found p-value of Chi-squared equals to 0.067, which is slightly above the significance level of 5% indicating that there may be a possibility to have heteroskedaticity on our sample. We use robust standard errors in our regressions implanted with Stata9 to correct for this possible problem. The robust regression corrects for possible heteroskedaticity by estimating the standard errors by using the Huber-White sandwich estimators (Wooldridge, 2010). Furthermore, such robust standard errors can deal with sampling problems about failure to meet assumptions, for instance, minor problems about normality and heteroscedasticity (UCLA - Institute for Digital Research and Education, 2013).
After running the Breusch-Pagan test we plot the graph of residuals below, from where we cannot identify heteroskedaticity problem of major concern:
Graph 1: Residuals scatter plot
Testing for Multicollinearity: The test used was the Variance Inflation Factor (VIF). On our regression the value found as 1.39. It is known that there is no consensus regarding the cut-
9 Stata is a data analysis and statistical software that is used on this research.
-30-20-10 01020
Residuals
-5 0 5 10 15
Fitted values
20 off value and the value for our model is a moderated low value. Furthermore, following Oxelheim and Randøy (2003), there is no indication of collinearity among the variables when VIF<10. Then we can assume that for this sample multicollinearity is no concern. Before presenting the regressions, we demonstrate that there is no multicollinearity amongst the variables presenting a correlation table (see Table 2), after the statistic description.
Testing for normal distribution: we performed the tests for kurtosis and skewness.
Based on these tests we could not reject the null hypothesis that ROA has a normal distribution.
The Jarque Bera test indicates the rejection of the null hypothesis of normality. However, this test is more accurate to larger samples. As we know that the natural logarithm of ROA could correct for a possible normality problem, we created the variable, one problem is that if we use it, we would lose 19 observations, which could bias our regression by using a sample with critical small size. Another option to test normality is the use of histograms. We draw the histograms for ROA and ROA_ln (natural logarithm of ROA). Depicted from the graphs 2 and 3, we see that ROA has a better normal shape than ROA_ln. Additionally, we use – as mentioned before – the robust standard errors regression in Stata, which can correct for minor problems about normality and heteroscedasticity. Therefore, we continue using ROA for our regressions.
Graph 2: ROA histogram
0
.02.04.06.08
Density
-40.00 -20.00 0.00 20.00 40.00
ROA
21 Graph 3: Natural logarithm of ROA histogram
4.3. Correlation
Table 2 shows the correlation values for the analyzed variables. First we find no correlation coefficient higher than |0,5|, which would typically mean a large correlation (Pallant, 2005). Thus, the correlation table did not show a multicollinearity problem.
Looking at the correlation between the dependent variable (ROA) and the independent variables on Table 2, we can say the following: we observe that the three international experience variables (NAT, STD, WRK) of our study showed positive correlation to ROA (respectively 0,16; 0,07; 0,14). The presence of independent members on the board (BIND) also shows a positive correlation (0,24) with ROA, which was also expected, as independent members can indicate good will, commitment to good practices that will collaborate to the firm’s performance. On the Brazilian case, board independency showed to have a stronger correlation with ROA.
Between the independent variables, firm age (FAGE) showed a negative correlation with all three international experience variables. We can speculate that this finding can be related to the fact that these older companies may be more closed to have a foreign member on
0.1.2.3.4.5
Density
-4 -2 0 2 4
ROA_ln
22 their board and when they have, maybe the majority of the board being family members may limit the participation of the outsider. On the other hand, firm size (FSIZE) showed positive correlation with all three international experience variables, indicating that some change could be happening on the firm’s culture, for example, the younger generation of family members are accessing the board and, differently from older member, they benefited from global mobility easiness. Regarding foreign ownership (FOWN), it indicates a strong positive correlation to foreign nationality (NAT).
Table 2 - Variables correlation
ROA NAT STD WRK BIND BSIZE FSIZE FAGE FLIST FOWN FSUB
ROA 1.00
NAT 0.162 1.00
0.164
STD 0.074 -0.178 1.00 0.528 0.126
WRK 0.140 -0.041 0.356* 1.00 0.230 0.725 0.002
BIND 0.241* -0.183 0.194 0.239* 1.00 0.037 0.116 0.096 0.039
BSIZE -0.010 0.207 0.214 0.010 0.065 1.00 0.931 0.074 0.065 0.932 0.579
FSIZE -0.052 0.033 0.166 0.147 -0.025 0.449* 1.00 0.657 0.780 0.156 0.209 0.833 0.000
FAGE -0.280* -0.102 -0.092 -0.109 0.019 -0.075 0.207 1.00 0.015 0.382 0.434 0.351 0.872 0.525 0.074
FLIST -0.253* -0.103 -0.038 -0.096 -0.091 0.047 0.255* 0.021 1.00 0.029 0.380 0.749 0.415 0.439 0.688 0.028 0.859
FOWN 0.066 0.401* 0.059 0.001 -0.102 0.223 0.268* -0.096 -0.095 1.00 0.571 0.000 0.617 0.997 0.385 0.055 0.020 0.415 0.419
FSUB -0.217 -0.035 0.164 -0.056 0.013 0.148 0.437* 0.280* 0.343* 0.020 1.00 0.061 0.769 0.159 0.635 0.909 0.207 0.000 0.015 0.003 0.867
* at 5% confidence level
5. Econometric analysis
5.1. Hierarchical regressions
Hierarchical regressions are made to observe the behavior of each additional variable on the model, by doing this we can check whether or not the additional variable adds value to the model. If the dependent variable becomes better explained by the additional independent variable (e.g. if it increases the value of R-squared), one should keep it in the model. We perform a sequence of ten OLS robust regressions, adding one independent variable at a time.
We used robust OLS regressions, as discussed in Section 4.3, because it helps to correct for minor problems about normality and heteroscedasticity.
The regressions are presented in Appendix 3. Based on the results we conclude that regression (10), which uses all the variables of our model, is the best fit with the highest R- squared (24,7%) amongst all the equations. Additionally, no huge difference among the coefficients of each variable was observed. Hence, the model should have all the independent variables of our model (Section 3.2) to explain firm performance (ROA).
We note that FLIST, throughout equations (3) to (10), is negatively significant correlated to firm performance at 5% of significance level, which is not expected (Oxelheim &
Randøy, 2003). Being negatively correlated means that being listed on a foreign exchange market impacted negatively on ROA. BIND, as expected (Da Silveira, 2004), showed to be positively correlated to firm’s performance at 5% confidence interval thorough all regression that is appeared; it remained relevant even after we included our explanatory variables for international experience.
FAGE also shows to be significant and negatively correlated to firm’s performance (ROA). Older firms have a higher negative impact on ROA based on the managerial inertia possibility due to a low motivation to change (Cohen & Levinthal, 1990). Finally, as expected (Oxelheim & Randøy, 2003), NAT, shows positive and significant correlation at 5%
confidence level.
5.2. Endogeneity and robustness test
As mentioned before, reverse causality is present on works regarding board composition and firms’ performance (Hermalin & Weisbach, 2003; Oxelheim L. et al., 2013;
Oxelheim & Randøy, 2003). This problem can be reduced by using diverse techniques, some
25 examples are as follows: performing a lagged analysis (Ameer et al., 2010), basing the study on an exogenous country effect like a change in the law that makes companies to comply not completely voluntarily (Dahya & McConnell, 2007), or performing a 2SLS regression that corrects for endogeneity problems (Black et al., 2006; Oxelheim & Randøy, 2003). We chose the last approach to our study, which was the method that best fitted to our cross section analysis. The 2SLS regression is a statistical technique to reduce the bias caused by endogeneity (e.g. if the independent variable and the error term are positively correlated, the increase on their values would overestimate the value of the independent variable). In cases where endogeneity conferment to be present, OLS models are not reliable.
One important part of regressing using 2SLS technique is defining instrumental variables (or instruments). These are variables that have no direct effect on the dependent variable, is exogenous and is not weakly correlated to the (possible) endogenous variable. For the 2SLS regression, one or more instruments can be used as long as they respect the previous conditions, as long as we have at least one instrument to each endogenous variable (Hill et al., 2008). Therefore, we based our choice of instrument on the correlation table (Table 2), where we see a positive and significant correlation between FOWN (instrumental variable) and NAT (endogenous variable). Hence, we firstly regressed our model using OLS and then the 2SLS using FOWN as described in Section 3.2. Later we perform the Durbin and Wu-Hausman tests (tests of endogeneity), the results are reported in the Appendix 4. From the tests we see that the p-values for Durbin and Wu-Hausman tests are high, meaning that we can reject the null hypothesis that the variables are exogenous, hence we can continue using OLS regressions for our work.
5.3. Hypotheses tests results
Table 3 shows the results for both regressions (OLS and 2SLS). The OLS regression shows an explanatory power (R-squared of 24.7%), which is acceptable compared to other works in board composition and firm’s performance, see Oxelheim & Randøy, (2003) and Ameer et al., (2010). The results, at 5% significance level, indicate that NAT and BIND have a significant positive correlation with ROA, as we expected (Ameer et al., 2010; Oxelheim &
Randøy, 2003). FLIST and FAGE show a significant negative impact on ROA. FAGE being negatively related to ROA was expected, some reasons can be given to explain this correlation, such as financial constrains that may prevent firm’s to have access to the necessary funds to
26 cover its opportunity costs (Loderer & Waelchli, 2010). A negative correlation between FLIST and ROA was also expected as the costs associated to foreign listing can be high, impacting performance negatively (Mittoo, 1992). Regarding the other international experience explanatory variables, STD and WRK, no significant correlation is found our study.
Table 3: OLS and 2SLS regressions (OLS - robust)
(2SLS - robust)
VARIABLES ROA ROA
NAT 10.35** 3.101
(4.816) (10.12)
STD 2.735 1.030
(5.100) (4.907)
WRK -1.247 -0.360
(5.869) (5.381)
BSIZE -0.483 -0.361
(0.319) (0.329)
BIND 11.39** 10.46**
(5.055) (4.747)
FLIST -5.806** -6.146**
(2.898) (2.847)
FAGE -2.952** -3.024**
(1.462) (1.389)
FSIZE 1.116 0.970
(1.376) (1.186)
FOWN -1.846
(3.581)
FSUB -2.403 -2.193
(2.797) (2.568)
Constant -0.791 1.513
(18.32) (15.96)
Observations 75 75
R-squared 0.247 0.230
F test 0.005 0.001
Robust standard errors in parentheses;
*** p<0.01, ** p<0. 05, * p<0.1
Based on these findings (Table 3), we obtain the indication that having one more member with foreign nationality contributes positively to industrial Brazilian firms’
performance. On the other hand, work and study abroad do not show a significant correlation.
Hence, our first hypothesis (H1) is partially confirmed.
