A multi-dimensional approach to the internationalization-performance relationship during and after the financial crisis

Top management team nationality diversity and the degree

of internationalization

A multi-dimensional approach to the internationalization-performance

relationship during and after the financial crisis

Master Thesis International Economics and Business

Author: Jasper ten Caat

Student number: s2393042

Email: j.ten.caat@student.rug.nl

Supervisor: dr. P. Rao Sahib Co-assessor: dr. K. van Veen

Faculty of Economics and Business University of Groningen

- 2 - Abstract

This paper examines the relationship between the degree of internationalization of a firm and financial performance for a group of 105 multinational enterprises from five European countries. Data is collected to construct measures that capture different attributes of internationalization, in particular the nationality diversity among top management teams and the relative importance of firms’ foreign operations. These measures are employed to investigate whether profitability during and after the crisis of 2008 was greater for companies with higher levels of internationalization. Although there exist strong differences between industries and countries, no convincing evidence of such a relationship can be established.

- 3 - I. Introduction

Perhaps the most conspicuous aspect of globalization since World War II is the increased global and multidimensional interconnectedness among citizens, firms and governments, and this development has had strong implications for many facets of economic life. A prominent feature of globalization is the surge in international trade since the 1990s, in particular between countries that are developed and those that are catching up rapidly (WTO, 2008). At its core, however, the bulk of cross-border trade is not between countries, nor between industries, but rather between individual firms. Specifically, it is the group of only the most productive firms that can afford to engage in international commerce, as is exhibited in the influential Melitz (2003) model and demonstrated empirically in a broad variety of empirical studies (of which an overview is provided in Wagner, 2007).

Evidently, the foreign operations of firms are the outcomes of decisions and actions undertaken by the employees and owners, with the board of directors and the top management team (TMT) having the most powerful tools to direct these operations (Carpenter, Geletkanycz and Sanders, 2004). Previous studies in the field of International Business, of which some will be discussed below, have revealed the importance of the personal characteristics of board members and managers in determining firm structure and performance, both in the domestic market and in the international context. Similarly, given the findings on the relationship between firm productivity and trade mentioned above, it is perhaps not that surprising that firms’ degree of internationalization (DOI) has also been frequently employed in empirical research as an explanatory variable of sales, return on assets and company growth. In this paper, these two apparent determinants of performance are the subject of further analysis, which takes place in the form of an assessment of the relative impact on profitability of both the management-level attributes and the extent to which firms are internationalized.

- 4 -

underpinnings, the empirical literature has provided mixed and even contradictory evidence with regard to the existence of a causal relationship between DOI and firm performance. Several authors have ascribed this inconclusiveness to inadequate measures of internationalization, suggesting that the variables most frequently used in leading academic papers do not truly capture the full extent of firms’ DOI (Hassel, Höpner, Kurdelbusch, Rehder and Zugehör, 2003; Ramaswamy, Kroeck and Renforth, 1996). Although the prevailing methods for gauging DOI serve fairly well as means to illustrate the relevance of overseas markets for firm structures and external revenues, they fail to incorporate other essential forms in which the internationalization of a business can be manifested, in particular those that are internal to the firm such as attitudinal attributes. An example of the latter is the international orientation of the persons in charge of managing and governing a multinational company (Sullivan, 1994).

Clearly, given the central prominence of the internationalization-performance paradigm in the discipline of International Business, the search for suitable measurement units that incorporate the multi-dimensionality of firms’ internationalization levels remains highly relevant and useful (Glaum and Oesterle, 2007). Therefore, in this thesis, the insights outlined above are combined by studying the interactions between TMT characteristics, common measures of firm internationalization and financial performance. Specifically, the focus will be on the role of nationality diversity within the boards of a sample of 105 large European multinational enterprises (MNEs). The aim of this study is to answer the question how this measure of TMT heterogeneity compares to other dimensions of firms’ DOI that are suggested by the literature, and to what extent these measures can explain variations in the profitability of firms. The time period covered in the analysis is marked by strong profit fluctuations and economic turbulence as it includes the financial crisis of 2008 and its immediate aftermath. Observing a potential positive association between DOI and performance would therefore suggest a greater stability and resilience for firms with higher levels of international activeness or orientation.

- 5 -

nationality diversity within top management teams isfacilitated bythe detailed information on European companies’ board members contained in a database developed by Van Veen and Marsman (2008). The gathered data allows for the construction of two distinct variables that take into account multiple features of a firm that determine, and are determined by, internationalization.

These variables are subsequently used both separately and together in regressions that try to explain patterns in firm earnings for the full sample of MNEs, as well as for particular subsets based on industry classification or country of domicile. It is demonstrated that the results of these estimations do not provide convincing evidence in favour of a statistically significant internationalization-performance relationship, neither between DOI and return on assets (ROA), nor between TMT nationality diversity and ROA. Nevertheless, the analysis does yield interesting insights into inter-industry and inter-country differences in firms’ degree of internationalization and the extent of nationality heterogeneity within the boards of these firms. Moreover, it shows considerable disparities between the subsets in terms of the magnitude and sign of the effects examined in this study.

Before these results and their implications are discussed, it is insightful to explore the essential theoretical foundations and empirical evidence on this issue brought forward by the literature in the past. This is presented in the next section. Section III provides the details on the sources and collection of the data, the sample and the variables included in the regression analysis. Section IV presents and reviews the empirical results. The thesis is concluded in Section V with a discussion of the study’s limitations and a brief overview of potentially fruitful avenues for future research.

II. Theory, related findings and hypotheses

- 6 -

rather than dwelling upon the vast body of research in this field that has been published in the past, this section will concentrate on only those essential studies that are directly related to the propositions and methodology adopted in this thesis. First, however, the focus will be on existing research pertaining to the link between TMT nationality diversity and financial performance. This is a strand of academic literature that is considerably less extensive. Therefore, this discussion is broken down into two steps by looking first at the relationship between firm performance and management team heterogeneity in a broader sense, which is then followed by an overview of current knowledge on the nationality diversity within TMTs.

TMT heterogeneity and firm performance

- 7 -

Although the pioneering work by Hambrick and Mason (1984) did not include a formal empirical investigation into the validity of their hypotheses, it did provide the field of International Business with a promising and fruitful path for subsequent research. As a result, age, gender, race, educational attainment, cultural background, (international) work experience and a variety of other personal traits have been subject to numerous studies linking their diversity among management teams to firm performance, such as those found in the influential papers by Hambrick, Cho and Chen (1996), Carpenter (2002) and Erhardt, Werbel and Shrader (2003). Using data on 32 major US airlines, Hambrick et al. (1996) study how the degree of top management team heterogeneity determines the competitive moves of a firm. Their results provide evidence supporting the hypothesis that firms with more heterogeneous TMTs (in terms of functions, tenures and educational backgrounds) exhibit a greater propensity to undertake competitive actions of considerable magnitudes. Furthermore, Hambrick et al. (1996) confirm that TMT heterogeneity has a positive impact on firm growth, and show that more diverse top management teams display a superior ability to react and adapt to environmental turbulence. In contrast to the majority of studies exploring management team attributes, Erhardt et al. (2003) focus on non-work-related traits by looking specifically at gender and ethnicity as sources of TMT diversity. They show that a higher percentage of women and minorities in management positions is associated with greater corporate profitability. In a study that is closely connected to this thesis, Carpenter (2002) demonstrates how the relationship between management team diversity and performance is shaped by the extent to which a firm is active beyond the home country. He finds stronger positive effects of tenure and functional background heterogeneity at low levels of internationalization, whereas heterogeneity in terms of educational attainment improves firms’ return on assets more significantly at high levels. Clearly, these findings by Carpenter (2002) are noteworthy considering this thesis’ endeavour to establish whether the effects of nationality heterogeneity on performance diverge in terms of size or sign between industries and home countries with high or low average levels of internationalization.

- 8 -

(2010a) reviews the theories and methodologies that are applied in a sample of sixty upper echelons articles and ascribes part of this inconclusiveness to the divergent mechanisms through which different forms of heterogeneity exert their impact on corporate performance. For instance, since educational background heterogeneity and diversity in ethnicity might well have very distinct causes, their consequences for organizational outcomes are likely to differ as well (Nielsen, 2010a). As a result, predictions of the particular effects of TMT nationality diversity based solely on the results reported in studies investigating other types of management team heterogeneity could be deemed somewhat imprudent.

In much the same way as the empirical explorations are inconsistent, the associated debate on the theoretical underpinnings of the link between TMT heterogeneity and company performance is also unsettled. In spite of the aforementioned advantages of diversity, obstacles to social exchange arising from communication difficulties, biases, discrimination and intra-team conflicts are more likely to occur in heterogeneous groups (Richard, Barnett, Dwyer and Chadwick, 2004; Smith et al., 1994; Tsui, Egan and O’Reilly, 1992). All in all, whether and how the degree of management team diversity affects the performance outcomes of an organization remains a question for which, at present, a decisive answer is lacking. This is particularly true for diversity in terms of team members’ nationality.

Nationality diversity and firm performance

- 9 -

over 20% of foreign members, while this figure is no higher than 7% for Italy, Spain and Portugal. Indeed, Van Veen and Marsman (2008) demonstrate the significance of country-of-origin effects in influencing TMT nationality diversity, which stems from the institutions embedded in the predominant governance regime that either promote or discourage the accessibility for outsiders. The authors also accept the proposition that certain company characteristics, including size in terms of the number of employees, are positively related to nationality diversity. These results provide a compelling incentive to incorporate both country effects and company controls in models that try to explain financial performance on the basis of nationality diversity measures.

- 10 -

enhanced strategy management is promoted through the internationalization of a board. This line of reasoning essentially originates from an assumption, frequently touched upon in the upper echelons literature, which states that a greater degree of observable demographic variation within a team indicates increased cognitive diversity. This, in turn, stimulates a multi-perspective interpretation of the complex environments in which a business operates, and facilitates a team’s attempts to make sense of the circumstances (Kilduff, Angelmar and Mehra, 2000). In the long run, such processes could yield remarkable advantages towards organizational outcomes (Sanders and Carpenter, 1998).

Shifting the focus from theoretical assertions to empirical assessments, it should be noted that studies on the true consequences of nationality diversity are scarce relative to research on other forms of TMT heterogeneity. Nevertheless, there are some notable papers that attempt to specify the direction and magnitude of the relationship between TMT internationalization and corporate performance on the basis of firm-level data. In this respect, the work by Bo Bernhard Nielsen and Sabina Nielsen is most closely related to the analysis performed in this thesis. For example, in a thorough and comprehensive study of 165 large Swiss companies, Nielsen (2010b) shows that a composite index of two measures of management team internationalization (nationality diversity and international work experience) has a positive and significant relation with firms’ propensity to enter markets abroad. This suggests that TMT internationalization serves as a key determinant of firm internationalization, in the sense that a management team that is more diverse in terms of international experiences and nationalities is more likely to guide a company towards foreign expansion. Moreover, foreign market entry is found to have a positive impact on corporate profitability, as measured by a company’s return index. Nielsen (2010b) did not find empirical evidence of a direct linear connection between TMT internationalization and firm performance, however. In a later study on management team diversity, Nielsen and Nielsen (2013) separate the two measures of TMT internationalization and show that nationality diversity by itself does have a positive and significant effect on firms’ return on assets.

Taking into account the theoretical expositions and empirical findings outlined above, the benefits of TMT nationality diversity for firm performance appear to outweigh the associated costs. This notion is reflected in the first of two propositions tested empirically in this study:

- 11 -

Degree of internationalization and firm performance

Nationality diversity among board members is certainly not the only channel through which globalization affects structures and operations at the firm level. Integration of national economies through the reduction of transport costs and trade barriers has increased the importance of regional and global markets for companies, exposing them to foreign competition but also providing them with profit opportunities beyond the borders of their home country. The remarkable growth in international trade is manifested in the increased significance of foreign sales for firms’ overall revenues or profits, and the boom in foreign direct investment and offshoring since 1980 induced MNEs to increase the number of foreign employees and to expand their stock of overseas assets (Hirst, Thompson and Bromley, 2015). Consequently, globalization has affected firms on a multi-dimensional scale.

- 12 -

Asserting that the exact opposite of that statement applies to the related empirical branch of the literature is not too much of an exaggeration, however. After nearly four decades of research, and despite the abundance of studies examining this issue, the significance, sign and shape of the relationship between DOI and financial performance remain strongly disputed (Ruigrok and Wagner, 2004).

Various scholars in the field of International Business have attempted to gauge the degree of internationalization of MNEs by calculating firms’ foreign sales as a percentage of total sales (FSTS). This technique has been prevailing since the early 1970s, but has, in spite of the theoretical unambiguity, generated a wide range of positive, indeterminate and even negative results on the linkage between internationalization and performance (Grant, 1987; Shaked, 1986; Sullivan, 1994). In Sullivan (1994) it is argued that this inconclusiveness may be the consequence of an arbitrary and potentially inadequate measurement of DOI. Sullivan stresses that the use of a single-item unit, such as FSTS, is less likely to fully incorporate the multi-attribute domain in which a firm can become more internationalized. In particular, a firm’s DOI can be observed in terms of its performance, its organizational structure and the attitudinal traits of its management team. Although FSTS can serve as a suitable proxy for the performance aspect of internationalization, it largely neglects the other two dimensions. In an attempt to provide a solution to this methodological problem, Sullivan (1994) collects detailed data on 74 US MNEs in order to construct alternative measures for their DOI. He finds that a more precise and reliable estimation of DOI can be achieved by employing a linear combination of five measures: FSTS, foreign assets as a percentage of total assets, overseas subsidiaries as a percentage of total subsidiaries, top managers’ international experience and the psychic dispersion of international operations. For each MNE in his sample, Sullivan calculates an internationalization index DOIINTS by summing its score on these five variables.

Although he demonstrates the statistical legitimacy of this construct, Sullivan leaves the examination of the relationship between DOIINTS and financial performance to other scholars.

Ramaswamy et al. (1996) take it upon themselves to perform that analysis. They replicate Sullivan’s (1994) approach by using exactly the same data sources to construct the five variables that constitute DOIINTS for the same set of 74 US MNEs. By regressing two

- 13 -

does not exceed a mere 13 percent. Ramaswamy et al. (1996) argue that these findings, which clearly contradict the common theoretical assumptions, are the result of conjectural conclusions regarding the validity of the attitudinal attributes and of errors in psychometric reasoning. In particular, they disapprove of integrating both organizational and managerial components into one single linear combination, as such an exercise disregards the lack of a common scale among these two types of variables and because it has no sufficient foundation in the literature. Consequently, Ramaswamy et al. (1996) comment that although, in general, indices are conceptually and econometrically superior to single-variable measures, DOIINTS still

requires further refinement as it proves incapable of capturing the full complex nature of internationalization. In a more recent review of the literature on internationalization and performance, Hitt, Tihanyi, Miller and Connelly (2006) stress that the difficulties that arise from attempting to account for the full extent of firm’s international expansion can explain why it remains challenging to confirm a positive DOI-performance relationship. A cautious, well-established approach to synthesizing internationalization at all relevant levels of an organization is of pivotal importance for the success of future research.

It is not the goal of this thesis to deliver a comprehensive and uncontested approach for the determination of DOI. Nonetheless, by taking into account the recommendations expressed in the literature, this study aims to examine the subject in an appropriate manner in order to verify the empirical validity of one of the most investigated propositions in International Business:

Hypothesis 2: Firm internationalization is positively associated with firm performance.

III. Data and methods

The use of reliable firm-level data and a solid estimation technique are the key elements of a sound methodology. This section describes the sample choice, data collection, included variables, empirical model and econometric approach that ultimately allow for assessing whether the two hypotheses have any substance in the context of today’s globalized world.

Sample

- 14 -

abroad, for example in the form of foreign sales or overseas production sites. This ensures that a measure of the degree of internationalization can be computed for these firms (Ramaswamy et al., 1996). Second, the data required to construct the variables that are used in the empirical analysis has to be reliable, consistent, and, above all, available for the time period under investigation.

Since Orbis, an extensive database set up by Bureau van Dijk, readily provides the necessary information on financial performance between the years 2006 and 2016 for over 200 million private companies, the data availability constraint is most restrictive for the measurement of firm internationalization and nationality heterogeneity in top management teams. Given that managers’ country of birth is not always reported in accessible business documentation, and taking into account the frequent changes in the composition of TMTs, it is evident that an analysis of nationality diversity in particular involves a considerable data collection effort. Fortunately, I have been granted access to a database introduced in Van Veen and Marsman (2008), which provides information on the nationalities of top managers within the executive and supervisory boards of nearly 350 firms headquartered in one of 15 European countries. For those included companies that have not been the target of a completed acquisition or that have gone bankrupt, the manager-level data is recorded annually for the period 2005-2010 on the basis of information provided in annual reports and on company websites.

- 15 -

intrinsically more interesting than when one limits the attention to predominantly domestic businesses.

Accordingly, the empirical investigation is based on data for a sample that consists of only those European firms that are covered by the database developed by Van Veen and Marsman (2008). Furthermore, in view of the laborious nature of the data collection process, the choice has been made to restrict the sample group to companies headquartered in either the United Kingdom, Germany, the Netherlands, France or Italy. In general, MNEs originating from these five countries differ strongly in terms of nationality diversity (Van Veen and Marsman, 2008; Van Veen, Sahib and Aangeenbrug, 2014), with Dutch and UK companies having a relatively large share of foreign managers within their executive boards, while French and German firms are characterized by more moderate levels of TMT internationalization. In Italy, the ratio of foreign to local managers is among the lowest in Europe. These five countries are therefore selected to ensure that the findings presented below are not merely applicable to firms based in locations where the presence of foreign board members is either common or rare. Significant cross-country variance in nationality diversity allows for more general inference.

In terms of activity type, one important additional restriction was imposed upon the sample group in the form of the exclusion of firms whose primary line of business, according to their Standard Industrial Classification code, is based in the finance sector. The introduction of this constraint implies the removal of banks, insurance companies and real estate investors from the final sample. Four key considerations have contributed to this decision. First, the interpretation and operationalization of key concepts related to firm performance and structure, such as sales, assets and returns, differ fundamentally between finance and non-finance companies. Comparing a typical bank’s income statement and balance sheet to those of, say, a manufacturing firm exemplifies this argument. Interest income and interest expense comprise the main determinants of the total income of a bank, whereas the equivalent items for a manufacturer are sales and the cost of goods sold. Similarly, loans and securities represent the main assets on a bank’s balance sheet, while plants, equipment and inventories embody the most valuable assets of a manufacturing firm. This illustrates the difficulties that arise when one compares, for example, the return on assets for BMW and Commerzbank.

- 16 -

nationalities database report the geographical distribution of sales, but instead refer to income earned abroad. Since sales is essentially a non-negative measure while income is clearly not, this again prohibits a meaningful analysis of the importance of foreign markets for financial companies relative to their non-finance counterparts.

Third, the firm-level data provided by Orbis only covers the last five years for most banks and insurance companies, which is in contrast to the 2006-2016 period reported for non-finance companies and the 2005-2010 period for which the nationalities database contains manager-level information. Evidently, this data timing mismatch will trouble any regression that does not incorporate long lags for the affected variables.

Finally, a series of high-profile mergers in the finance sector during and after the financial crisis of 2008, in particular among the largest banks in Italy, serves as another justification for omitting the companies operating in that business segment. The reasoning behind this latter assertion is twofold. First, a merger frequently involves a drastic and abrupt adjustment in a firm’s organizational structure, which is for example represented in the data by a sudden increase or decrease from one year to another in TMT nationality diversity or other measures of DOI. Second, Orbis does not always provide consistent reporting on firm performance for the periods before and after a merger, leading to a loss of compatibility with the nationalities database.

Together, these four concerns strongly suggest that findings on the presence or absence of a relationship between internationalization and firm performance would be less reliable when financial organizations such as banks were not excluded from the analysis.

- 17 -

Manufacturing (55), Transportation and communications (23), Wholesale and retail trade (8) and Services (7). The latter category consists of the remaining firms that cannot be included in either one of the other four sector types, and contains for instance businesses operating in the fields of publishing, hospitality, enterprise software and human resource consulting. A complete list of the companies included in the final dataset is presented in Appendix I.

Variables and data collection

In order to examine the internationalization-performance relationship, data was collected from various sources to construct the variables that capture TMT nationality diversity, DOI, profitability, firm-level controls and fixed effects.

Dependent variable

Following the example of numerous other papers on the linkage between internationalization and financial performance (Hitt et al., 2006; Sullivan, 1994), this study uses return on assets

(ROA) as the measure of firms’ profitability. ROA is calculated as Profit before taxes_{Total assets} x 100% using the firm-level data provided by Orbis and it is collected for the years 2006-2015. At the time of the data collection, earnings information for the year 2016 was available for only about half of the sample group, and is thus not analysed. The choice for working with pretax profits is based on the assumption that firms are unable to influence the effect of a country’s tax rate on their income. Looking at net (after-tax) profits would therefore be inequitable as it would favour the performance of firms that operate primarily in countries with low tax rates (Elango and Sethi, 2007).

Independent variables

non-- 18 non--

executives or supervisory board members (Van Veen et al., 2014). This implies that the terms ‘TMT’ and ‘board’ can be used interchangeably, and so can the labels ‘manager’ and ‘board member’.

In order for directors to be recorded in this database, they had to be a member of an executive committee or supervisory board overseeing the activities of a listed European company for a term sometime between the years 2005 and 2010. For each of the 105 firms included in the final sample and for each year during which a firm was active, the database provides a list of the company’s board members and their nationalities for the period 2005-2010, enabling a calculation of the number of different nationalities represented in a particular board and their relative prevalence. This data allows for the construction of the variable labelled TMTDiv by using Blau’s index of diversity (Blau, 1977), as recommended by Van Veen and Marsman (2008) and as implemented in many related studies such as Carpenter (2002) and Nielsen (2010b). This measurement method can be deployed productively to capture the dispersion of TMT members in terms of their national background at time t by calculating TMTDiv𝑡 = 1 − ∑(pi,t)

2

, where pi is the percentage of managers with a particular

nationality i. When all board members have the same nationality, the Blau index takes the value of 0. The introduction of an additional or successional foreign board member would then induce this figure to rise towards its limit of 1, as a result of the increase in nationality diversity. This measure of TMT heterogeneity will serve as one of the two main variables of interest explored in this research, and it represents the primary means to capture the attitudinal attribute of internationalization (Caligiuri, Lazarova and Zehetbauer, 2004; Sullivan, 1994).

Although it would be somewhat less complicated to define TMT internationalization as the proportion of foreigners in the board of an MNE, as is done in Van Veen and Marsman (2008), such a technique would neglect the true multi-nationality that might be contained in management teams. For example, it is hard to argue that a six-person team that governs a British company and that consists of, say, three British and three German members is equally internationalized as an equivalent team that consists of three British, a German, a French and a Chinese member. The share of foreigners in those teams would be similar, however. Clearly, the sophistication of Blau’s index should make this the preferred operationalization of TMT heterogeneity in any attempt to estimate the internationalization-performance relationship.

- 19 -

the degree of internationalization (DOI) of a firm. In this thesis, a company’s DOI for a given year t is calculated as the average of its foreign sales as a percentage of total sales (FSTS) and its proportion of foreign assets to total assets (FATA): 𝐷𝑂𝐼𝑡 =

𝐹𝑆𝑇𝑆𝑡+𝐹𝐴𝑇𝐴𝑡

2 . If, for a particular

firm-year observation, one of the two components is not available, DOI is simply approximated by the value of the remaining item. A similar multidimensional strategy for estimating firm internationalization is employed by Nielsen and Nielsen (2013) and in the calculation of the Transnationality Index published by UNCTAD (2007).

A firm’s FSTS embodies the performance dimension of internationalization and demonstrates the degree to which that firm depends on foreign markets. FATA, on the other hand, is a measure that can be used to evaluate the internationalization of the structure of the firm and it captures the extent to which a firm relies on overseas production (Gomes and Ramaswamy, 1999; Sullivan, 1994; Thomas and Eden, 2004). The two items FSTS and FATA are closely related in a statistical sense, as illustrated by a pairwise correlation coefficient equal to 0.77.

- 20 -

12% of its revenues in overseas markets. More insights regarding industry and inter-country differences related to firms’ DOI are presented in the next section.

Data unavailability and time considerations induced me to forego the use of alternative DOI measurement units proposed by Sullivan (1994), such as top managers’ international experience and the psychic dispersion of international operations. Nevertheless, it is not unreasonable to assume that TMT internationalization and FATA can serve as decent alternates for these omitted proxies of DOI.

Control variables

The choice for control variables is based on economic reasoning and the practices of the studies most closely related to this research. First, firm-level controls are essential elements of any regression in which performance is the dependent variable, because the incorporation of these controls facilitates an isolation of the relationship that is of primary interest. Following the expositions in Carpenter (2002), Nielsen (2010b) and a great number of other relevant papers, firm size is included since it is assumed to have a strong impact on both TMT characteristics and financial performance. For instance, a larger firm is more likely to operate in multiple markets, which expands the scope for a diversified workforce and raises the probability of non-nationals entering the board (Van Veen and Elbertsen, 2008). In this thesis, firm size is defined as the log of the total number of employees. Using total assets to compute firm size did not result in estimation outcomes notably different from those presented below. Firm leverage is another variable put forward by the literature as a factor influencing managerial decision-making and financial performance, because a dependence on debt stimulates executives to signal credibility by boosting profitability (Agrawal and Knoeber, 1996). Therefore, the estimated model will include a control variable calculated as the ratio of long-term debt to total capital (i.e. shareholders’ equity). This mirrors the approach adopted by Nielsen (2010b) and Nielsen and Nielsen (2013). Both firm-level control variables are constructed based on data provided by Orbis for the years 2006-2015.

- 21 -

(WhoRet) and Services. The significance of industry effects in explaining cross-sectional variation in internationalization and performance is well-established in the empirical literature (Tallman and Li, 1996). Manufacturing serves as the base industry in the regression model analysed below because most firms in the sample are allocated to this sector.

A demonstration of the critical impact of home country institutions, both formal and informal, on the connection between companies’ DOI and profitability is the central contribution of a meta-analysis performed by Marano, Arregle, Hitt, Spadafora and Van Essen (2016) of 359 studies exploring this relationship. In a similar vein, Elango and Sethi (2007) stress the crucial importance of the country-of-origin effect, which manifests itself through the specific advantages or disadvantages experienced by firms from a particular country. For example, governance modes, strategic choices and competitive capabilities are major firm attributes determining internationalization and performance. Nonetheless, all three elements are (partly) shaped by country-based factors such as cultural values, physical resources and economic policies (Elango and Sethi, 2007). These findings convincingly illustrate the necessity to control for country effects. The country-level control variables used in this thesis are operationalized as dummy variables taking the value of 1 if a firm is headquartered in a particular nation. Since the data covers companies from five different countries (United Kingdom, the Netherlands, Germany, France and Italy), the same number of dummy variables are generated (firm_UK, firm_NL, firm_DE, firm_FR and firm_IT). In the regression analysis, the indicator variable for France is excluded as this country is home to the majority of covered firms and therefore represents the reference category. The classification of co-headquartered companies in terms of their country of origin follows the procedure adopted by Van Veen and Marsman (2008). This implies, for example, that Unilever and Shell are both treated as Dutch firms. The reader is referred to Appendix I for an overview of the included firms and their respective home countries.

- 22 -

Estimation method

Collecting the data on the variables listed above for the 105 European MNEs resulted in the construction of a panel of data that can be described as relatively short and wide. Three broad types of models can be employed to analyse panel data, namely pooled, fixed effects and random effects models. Although a fixed effects model takes into account the time-invariant heterogeneity across individual companies by allowing for the presence of firm-specific intercepts, the use of such an estimation technique is less powerful when the time period under consideration is rather short (Nickell, 1981). Moreover, the explanatory power of time-invariant controls and relatively static variables such as DOI is strongly diminished as the fixed effects replace their impact on the dependent variable (Van Veen et al., 2014). Given the theoretical significance of the dummies and the key role that slowly-changing variables play in this study, this is an undesirable outcome. Consequently, a fixed effects model is not deemed to be conducive to the efficiency of this research. Similarly, assuming that the large, listed MNEs included in the sample are selected randomly, as is required when one uses a random effects model, is neither truly appropriate in this case. In light of these considerations, the choice has been made to estimate the relationship between internationalization and performance by utilizing a pooled OLS method. Robustness checks have indicated that no coefficients diverge significantly when one alternates between the pooled and random effects models.

A typical issue related to panel data that is not addressed by applying pooled OLS is the presence of nonzero error correlation, detected over time, for the individual companies (Hill, Griffiths and Lim, 2011). Unobserved firm characteristics can explain the existence of a constant component in that firm’s error term. Using cluster-robust standard errors is a suitable procedure to guarantee the validity and consistency of the least squares estimates and to allow for reliable inference when heteroskedasticity and autocorrelation are expected. The statistical (in)significance of coefficients reported in the next section is therefore based on these corrected standard errors.

- 23 -

(1) or bias the estimation results. With respect to the operationalization of the dependent variable, histograms and formal Jarque-Bera tests indicated that there was no need to transform or rescale the data on firm performance. A normal distribution of the regression residuals already emerged from the models that used no transformations.

Model

Having determined the most appropriate regression methodology, it is now insightful to explore the specification of the model that is estimated in the empirical analysis. This model is given by equation (1):

𝑅𝑂𝐴𝑖,𝑡 = 𝛽1+ 𝛽2𝑇𝑀𝑇𝐷𝑖𝑣𝑖,𝑡−1+ 𝛽3𝐷𝑂𝐼𝑖,𝑡−1+ 𝛽4𝐹𝑖𝑟𝑚𝑠𝑖𝑧𝑒𝑖,𝑡+ 𝛽5𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖,𝑡

+𝛽6𝑀𝑖𝑛𝐶𝑜𝑖+ 𝛽7𝑇𝑟𝑎𝑛𝑠𝐶𝑜𝑚𝑖 + 𝛽8𝑊ℎ𝑜𝑅𝑒𝑡𝑖+ 𝛽9𝑆𝑒𝑟𝑣𝑖𝑐𝑒𝑠𝑖

+𝛽₁₀𝐹𝑖𝑟𝑚_𝑈𝐾𝑖+ 𝛽₁₁𝐹𝑖𝑟𝑚_𝑁𝐿𝑖 + 𝛽₁₂𝐹𝑖𝑟𝑚_𝐷𝐸𝑖+ 𝛽₁₃𝐹𝑖𝑟𝑚_𝐼𝑇𝑖

+𝛽14𝑌2006𝑡+ 𝛽15𝑌2007𝑡+ 𝛽16𝑌2009𝑡+ 𝛽17𝑌2010𝑡+ 𝛽18𝑌2011𝑡+ 𝜀𝑖,𝑡

In the equation, the subscripts i (= 1, 2, …, 105) and t (= 2006, 2007, …, 2011) refer to the particular firm and year under investigation, respectively. The first row consists of the firm-level measures, including the two variables that capture internationalization of the management team and of the organization as a whole. Positive and significant estimates for β2 and β3 would suggest empirical validation of the two hypotheses assessed in this thesis.

Industry, home country and year effects are accounted for by the dummies that comprise, respectively, the second, third and fourth row of the formula. The error term is denoted by ε.

In accordance with related papers (Carpenter, 2002; Erhardt, Werbel & Shrader, 2003; Nielsen, 2010), the variable representing TMT nationality diversity is lagged by one year to incorporate the delay that exists between the time at which managers enter a firm’s board and the time at which their decisions actually influence firm performance. Similarly, a one-year lag for firms’ degree of internationalization is employed to account for the notion that it takes time before the effect of skills and strategies developed by internationally operating

Table 1. Pairwise correlation matrix

- 24 -

businesses can be observed in their profitability (Elango and Sethi, 2007; Grant, 1987). Although one could make a case for extending these lags beyond one year, the absence of TMT and DOI data for the period before 2005 prevents this study from doing so. Moreover, a short lag can be justified properly given the relatively sluggish nature of the two internationalization variables, which results in high year-to-year correlations for TMTDiv and DOI (Ramaswamy et al., 1996). The inclusion of lags also serves as a means to avoid problems arising from reverse causality, as advocated by Hambrick (2007) and executed by Nielsen and Nielsen (2013).

IV. Results

The estimation of equation (1) for the period 2006-2011 allows for an exploration of the effects of TMT nationality diversity and firm internationalization on financial performance during an era marked by major economic upheavals. Besides the regression outcomes and a discussion thereof, this section also provides an analysis of the raw data, which offers preliminary and intriguing insights into cross-sectional variance and the developments of key MNE characteristics throughout the studied time interval.

Presentation and discussion of descriptive statistics

Mean values that describe the five firm-level variables are exhibited in Table 2. From the standard deviations reported for the full sample, it is evident that there are substantial fluctuations observable for the different measures, in particular with respect to return on assets, TMT heterogeneity and leverage. In order to determine whether these sizable variances stem from cross-categorical divergence or are instead the result of strong changes occurring over time, Table 2 splits the full sample in terms of industry type and home country, and subdivides the full time period by year as well.

- 25 -

and retail trade. Furthermore, the management teams of MNEs operating in these two sectors consist primarily of native members. With regard to the firm-level control variables, the logarithmic transformation used to compute firm size somewhat disguises the fact that there are extraordinary dissimilarities between industries in terms of the number of employees. On average, the eight retailers that entered the sample provided a job to over 165000 workers, whereas the equivalent figure for the seven companies selling services is no larger than 45000. A high indebtedness (i.e. leverage), on the other hand, appears to be a distinguishing feature of the companies assigned to the Transportation and communications category.

Table 2 also indicates that firms headquartered in the UK or in the Netherlands achieved the best operating results, as well as the highest scores on TMT nationality diversity and DOI. On average, the German and French MNEs are the largest organizations while also being the least dependent on debt. Low internationalization levels characterize Italian firms.

Table 2. Means of firm-level variables

Means

ROA TMTDiv DOI Firm size Leverage

Full sample

(standard deviation in brackets)

6.356 (6.126) 0.390 (0.238) 0.677 (0.242) 4.795 (0.475) 0.816 (0.888) Industries

Mining & Construction 8.188 0.415 0.728 4.703 0.578

Manufacturing 7.010 0.421 0.777 4.812 0.668

Transport & Communications 3.962 0.300 0.470 4.730 1.415 Wholesale & Retail trade 5.143 0.344 0.459 5.224 0.735

Services 7.105 0.454 0.788 4.652 0.492 Countries United Kingdom 9.891 0.548 0.724 4.787 0.971 Netherlands 7.519 0.622 0.784 4.523 0.773 Germany 5.299 0.228 0.673 5.005 0.688 France 4.581 0.386 0.690 4.929 0.722 Italy 6.001 0.167 0.474 4.458 1.068 Years 2005 0.372 0.614 2006 10.578 0.384 0.633 4.575 0.707 2007 8.969 0.383 0.653 4.770 0.719 2008 6.332 0.400 0.669 4.808 0.903 2009 4.997 0.399 0.679 4.810 0.855 2010 7.059 0.404 0.688 4.806 0.767 2011 6.806 0.697 4.819 0.873 2012 5.555 0.698 4.821 0.877 2013 5.351 0.699 4.816 0.794 2014 5.136 0.694 4.811 0.870 2015 4.773 0.695 4.809 0.746

- 26 -

Year Year

% TMTDiv

All in all, disaggregating the sample averages by sector or country of origin seems to yield a suggestive perception of the existence of some positive relationship between performance and the two dimensions of internationalization. Notwithstanding this interesting suspicion, the historical trends in the relevant variables are more contradictory. The bottom segment of Table 2 shows how average ROA falls during the period 2006-2009, then increases sharply in 2010, and subsequently decreases continuously until the end of the recorded term. Such distinct downward trends are practically absent for DOI, which rises rather quickly up until 2012 and levels out afterwards. Likewise, TMTDiv goes up by a considerable margin between the years 2005 and 2010.

A more detailed examination of the trends in the three main variables of interest is presented graphically in Figure 1. In contrast to the measures of performance and firm internationalization, the size of the differences in TMT nationality diversity between the five countries remained rather stable, and no swaps in the relative positions took place during the covered era. On the other hand, a marked convergence between all countries except Italy aptly summarizes the developments observable for DOI. This is explained by the stagnation in

0 2 4 6 8 10 12 14 16 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Return on Assets UK NL DE FR IT 2005 2006 2007 2008 2009 2010 TMT nationality diversity 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 2005 2007 2009 2011 2013 2015 Degree of Internationalization Figure 1. Trends in average ROA, TMTDIV and DOI, segregated by home country.

Year

- 27 -

internationalization for Dutch firms, accompanied by notable increases for the firms from Germany, France and the UK. Finally, regarding profitability, there are stark discrepancies between the experiences of companies from the different countries. The average ROA for MNEs from the United Kingdom and Italy declined strongly, both in absolute and relative terms, until 2009 and again after 2011. For the firms from the other three countries, return on assets remained more or less stable after 2011.

What can explain the steadiness of nationality diversity differences and the partial convergence in DOI depicted by the two bottom panels of Figure 1? The literature provides some persuasive answers to this question. In a study of the fundamental determinants of nationality diversity in corporate boards and the corresponding role played by country-of-origin effects, Van Veen and Elbertsen (2008) attribute differences in TMT heterogeneity among companies to differences in dominant types of governance regimes. In particular, they argue that employee representation and the commonness of ownership by small groups of large domestic shareholders prevents outsiders and foreigners from entering German boards. In contrast, Dutch and UK boards are hypothesized to be more accessible to people from other countries as employees have no influence on board members’ recruitment. Moreover, the absence of blockholders facilitates the entry of non-British directors in UK organizations. Since governance regimes are very persistent, these and similar structural aspects impacting TMT composition might largely explain the lack of fluctuations in (rankings of) nationality diversity. The significance of country-specific institutional contexts therefore serves as a principal reason why a wide variety of nationality diversity levels is discernible in Figure 1 and Table 2.

- 28 -

expensive once firms achieve high levels of foreign spread (stage 3). They start to run the risk of over-internationalization (i.e. the costs exceed the benefits) and entering more foreign markets is no longer advantageous (Ruigrok et al., 2007). This rationale might be exemplified by the stagnancy of Dutch companies’ DOI since 2005.

Estimation results

Whether and how the investigated measures of internationalization affect ROA is answered formally by estimating equation (1). Before doing so, the isolated impacts of the two variables TMTDiv and DOI are analysed in separate regressions, which allows to examine in what way the observed performance-internationalization relationship is shaped by the choice of the particular dimension that is incorporated.

Table 3 presents the results on the DOI-ROA connection that are derived by estimating equation (1) without including the variable TMTDiv. Since both the firm internationalization and performance data are collected for a period of ten years, additional dummies are introduced to account for year effects after 2011. The predicted coefficients for these and the other types of indicator variables, listed in the first column of results, confirm the basic understandings obtained from the descriptive statistics. Moreover, the effects of the global economic downturn on the profits of MNEs are clearly demonstrated by the significantly negative coefficient for the 2009 year dummy, as well as by the large positive coefficients corresponding to the years before the crisis (recall that 2008 serves as the reference period). Surprisingly, adding the firm-level measures, including the degree of internationalization, barely increases the proportion of the variance in ROA explained by the right-hand side variables. Despite its magnitude, the coefficient for DOI is not significant at any of the conventional confidence levels (s.e. = 1.849). Only leverage exerts a statistically significant impact on profitability, but the sign of this relationship contradicts the prevailing theoretical expectations. Nonetheless, a negative leverage-performance connection is reported repeatedly in the empirical literature that is focused specifically on recessions (Erkens, Hung and Matos, 2012; Mitton, 2002).

- 29 -

Table 3. Regression results for DOI model with ROA as dependent variable (2006-2015)

Full sample Full sample High DOI industries Low DOI industries High DOI countries Low DOI countries Only post-crisis period Only fixed

controls

All independent variables

MinCo, ManufB _and Services

TransComB _and WhoRet

UK, NL and FRB _DEB _{and IT Period 2010}B_-2015

DOI -2.008 2.404 -7.007** _{-1.315 -4.259}* _-1.740 Firm size -0.225 0.420 -0.116 1.517 -0.563 -0.142 Leverage -0.929** _-1.798*** _{-0.405 -0.416 -2.625}*** _-0.772* MinCo 0.876 0.248 0.564 -0.382 2.346 -1.159 TransCom -2.973*** _-2.719** _-3.939** _-1.944** _-3.407*** WhoRet -2.680** _-2.851* _{0.311 -4.137}** _-1.837** _-3.801** Services 0.662 1.270 1.644 -1.471 10.279*** _0.320 Firm_UK 5.195*** _5.031*** _5.395*** _2.007* _4.884*** _4.113*** Firm_NL 2.682** _{1.284 -0.513 3.874}** _{2.017 0.538} Firm_DE 1.100 0.630 0.945 0.180 0.779 Firm_IT 1.929 0.926 0.042 -0.242 0.379 0.282 Y2006 3.482*** _2.793*** _2.751* _3.061*** _3.482*** _1.233 Y2007 2.675*** _2.169*** _2.342*** _1.917*** _2.134*** _2.245*** Y2009 -1.326** _-1.061** _-1.890*** _{0.179 -0.481 -1.825}** Y2010 0.752 0.888 0.472 1.310 0.920 0.871 Y2011 0.514 0.614 0.632 0.003 0.304 0.876 -0.308 Y2012 -0.763 -0.834 -0.769 -1.559 -1.266* _{-0.036 -1.767}*** Y2013 -0.957 -0.637 -0.382 -1.525 -0.984 0.146 -1.561*** Y2014 -1.171 -0.947 -0.883 -1.423 -1.447 -0.008 -1.875*** Y2015 -1.511* _{-1.297 -1.571 -0.829 -1.473 -0.856 -2.205}*** Constant 5.175*** _8.590** _{2.836 7.697 -0.095 12.708}*** _9.562** R2 _{0.21 0.22 0.21 0.34 0.23 0.36 0.19} N 951 823 570 253 506 317 550 *** p<0.01; ** p<0.05; * p<0.1

- 30 -

with high average levels of internationalization (Mining and construction, Manufacturing and Services). Again, however, the predicted effect is not significant. In contrast, the coefficient of interest is negative and significant for the subsamples of MNEs from industries and countries with a low mean value of DOI. Especially for retailers and businesses operating in the Transportation and communications sector, the results indicate that return on assets is expected to fall considerably when these firms expand their operations abroad.

The final column of Table 3 looks exclusively at ROA for 2010 and the five years thereafter to gauge the post-crisis situation. In comparison with the assessment of the full time period including the crisis, the results differ little.

Replacing DOI by TMTDiv in the equation used in the regression analysis generates the outcomes presented in Table 4, which describe the shorter period 2006-2011. The coefficients corresponding to the indicator variables for industry, home country and year are more or less Table 4. Regression results for TMTDiv model with ROA as dependent variable (2006-2011)

Full sample Full sample High TMTDiv industries Low TMTDiv industries High TMTDiv countries Low TMTDiv countries Only fixed controls All independent variables MinCo, ManufB and Services TransComB _and WhoRet UKB _{and NL FR}B_{, DE and IT} TMTDiv 1.022 3.994 -7.565** _{4.817 -0.518} Firm size -1.088 -0.895 -0.574 0.253 -1.724** Leverage -1.705*** _-2.790*** _{-0.798 -0.551 -2.308}*** MinCo 3.061** _2.968** _3.301** _{2.876 3.377}* TransCom -2.621** _{-0.977 -3.025 -0.661} WhoRet -2.107** _{-1.076 -0.155 -1.825 -0.108} Services 0.937 0.865 1.246 -2.708* _2.759 Firm_UK 7.146** _7.241*** _7.484*** _5.077*** Firm_NL 3.493** _{2.025 0.379 5.451}** _-4.396** Firm_DE 1.429 1.269 2.345 -1.254 1.203 Firm_IT 3.038* _2.783* _{2.636 0.907 2.554} Y2006 3.085*** _1.896** _1.855* _1.926* _3.389*** _1.079 Y2007 2.703*** _2.223*** _2.480*** _1.726*** _3.925*** _1.577*** Y2009 -1.311** _-1.739*** _-2.521*** _{0.160 -1.184 -1.958}*** Y2010 0.763 0.354 -0.016 1.235 0.734 0.230 Y2011 0.522 0.317 0.400 0.104 0.356 0.279 Constant 4.041*** _10.418** _{8.938 8.981 7.819 14.450}*** R2 _{0.29 0.35 0.39 0.35 0.31 0.28} N 553 544 382 162 182 362 *** p<0.01; ** p<0.05; * p<0.1

- 31 -

similar to those exhibited in Table 3, with the main notable differences relating to the significant positive effects of being headquartered in Italy and being active in Mining and construction. The inclusion of the firm-level measures somewhat improves the explanatory power of the model, even though it is again only the control variable leverage that has a significant (negative) impact on ROA. TMT nationality diversity is positively associated with performance for the full sample of firms, yet its coefficient is not statistically relevant (s.e. = 2.587).

Some large, positive, but statistically insignificant effects of the management-level dimension of internationalization are also reported in estimations using only the MNEs allocated to industries or countries with above-average levels of TMT diversity. On the other hand, for firms in the industries labelled Transportation and communications and Wholesale and retail trade, the results indicate that the relationship between TMTDiv and ROA has a significantly negative sign.

All in all, these stark dissimilarities between different sets of industries and home countries in terms of the sign and significance of the evaluated relationships seem to characterize both dimensions of internationalization. Whether this is also true when one combines the two measures in a single linear model, defined by equation (1), is answered in Table 5. Compared to the isolated effects calculated in the separate regressions, the coefficients corresponding to TMT nationality diversity and DOI do not change sign in the estimation for the complete sample, neither do they become statistically significant (s.e. TMTDiv = 2.817, s.e. DOI = 2.302). In fact, the impact of TMTDiv is almost completely diminished. Firm internationalization captured by DOI is negatively associated with performance between 2006 and 2011, which is an observation analogous to the findings for the longer period displayed in Table 3.

- 32 -

Table 5. Regression results for TMTDiv-DOI model with ROA as dependent variable (2006-2011)

Full sample High TMTDiv / DOI industries Low TMTDiv / DOI industries High TMTDiv / DOI countries Low TMTDiv / DOI countries MinCo, ManufB and Services TransComB _and WhoRet

UK, NL and FRB _DEB _{and IT}

TMTDiv 0.036 0.969 -5.451* _{0.423 -1.269} DOI -2.220 2.036 -6.892* _{-2.034 -3.784} Firm size -0.480 0.323 -0.636 1.341 -0.935 Leverage -1.564*** _-2.403*** _-1.150** _-1.056* _-2.384*** MinCo 2.900* _3.517** _{1.980 5.987}** TransCom -1.317 -2.469 -1.004 WhoRet -1.621 -0.133 -3.135 -0.103 Services 1.906 2.822 -1.481 13.274*** Firm_UK 7.364*** _8.090*** _3.724*** _7.104*** Firm_NL 2.087 -0.616 6.064*** _2.717 Firm_DE 0.990 1.900 -1.313 Firm_IT 1.715 1.242 -1.723 0.405 Y2006 2.144** _2.326* _2.384** _2.832** _0.912 Y2007 2.081*** _2.364*** _1.757*** _2.024** _2.273*** Y2009 -1.227** _-2.158*** _{0.262 -0.718 -1.823}** Y2010 0.676 0.214 1.334 0.683 0.932 Y2011 0.499 0.462 0.177 0.224 0.939 Constant 9.091** _{2.179 13.037}** _{0.163 13.577}*** R2 _{0.30 0.35 0.46 0.32 0.47} N 457 310 147 280 177 *** p<0.01; ** p<0.05; * p<0.1

B = base category in subsamples TMTDiv & DOI are lagged by one year

Table 6. Robustness checks for TMTDiv-DOI model with ROA as dependent variable (2006-2011)

DOI quadratic DOI sum DOI sum + OSTS Random effects model Hausman-Taylor model TMTDiv -0.020 1.369 1.422 -0.293 -0.145 DOI -2.835 -1.412 -0.813 0.886 1.999 DOI2 0.579 Firm size -0.444 -0.043 -0.418 -0.510 -1.077 Leverage -1.562*** _-1.306*** _-1.450*** _-1.554*** _-1.492** Constant 9.066** _{5.545 7.994 7.308 8.874} R2 _{0.31 0.34 0.36 0.29 n.a.} N 457 399 350 457 457 *** p<0.01; ** p<0.05; * p<0.1 TMTDiv & DOI are lagged by one year

- 33 -

Robustness tests

In order to check how the basic estimation results would change if another model specification or variable operationalization method would be utilized, Table 6 demonstrates the coefficients that are generated by a number of divergent regressions for the full sample of MNEs. First, a term calculated as the square of DOI is added to the model in order to examine the empirical scope for a quadratic relationship between internationalization and performance. Evidence supporting the validity of such a(n) (inverted) U-shaped model is found by a number of researchers (Capar and Kotabe, 2003; Elango and Sethi, 2007; Gomes and Ramaswamy, 1999; Ruigrok and Wagner, 2003). Although the signs for DOI and its square term indeed indicate a U-shaped relationship with ROA, the coefficients are not significant. Therefore, in this study, the quadratic model is deemed to be of little value.

Instead of taking the average of the two components, it is also possible to operationalize DOI as the sum of FSTS and FATA, which is an approach previously employed by Carpenter (2002) and Nielsen (2010b). Table 6 demonstrates that this calculation technique for DOI does not alter the general findings. Alternatively, an additional measure of the structural attribute of firm internationalization can be incorporated by including in the calculation of DOI the ratio of overseas subsidiaries to total subsidiaries (OSTS). This method is proposed by Sullivan (1994), and Ramaswamy et al. (1996) provide empirical evidence of the validity of OSTS in the estimation of an internationalization-performance relationship. For MNEs that are the global ultimate owner of a corporate group, it is feasible to determine the number of domestic and foreign subsidiaries from the information on firms’ organizational structure provided by Orbis, allowing for the computation of OSTS. Adding this measure to the other two components of DOI does not lead to any noteworthy changes in the estimates.

- 34 -

Discussion

Evidently, the estimation results do not provide any support for the pair of hypotheses linking the two examined internationalization attributes, TMT nationality diversity and firms’ DOI, to financial performance. In none of the evaluations that use data on the full sample is it possible to identify a positive and significant relationship between TMTDIV or DOI and return on assets. The bulk of the variation in profitability during the economically turbulent period 2006-2011 seems to be absorbed by the dummies that are included to account for industry, country and year effects. Besides this general insignificance of the main variables of interest, a second conclusion derived from the three tables containing regression results is that there are strong differences between sets of industries and countries of origin regarding the sign of the relationship. By itself, this latter finding might be a principal explanation for the statistical irrelevancy observed in the full sample, in the sense that it is not unlikely that positive and negative effects pertaining to distinct subsamples cancel out. If this is the case, the heterogeneity in the set of analysed MNEs is beneficial to the extent that it facilitates the inspection of cross-sectional divergence, but the large within-sample variance jeopardizes attempts to establish significant relationships that can be generalized to a broad set of firms. Before considering the theoretical rationalizations and interpretations of these results, it is enlightening to recall the inconclusiveness of the empirical literature with respect to the consequences of both TMT nationality diversity and DOI for financial performance. Although the two hypotheses tested in this paper were formulated based on the plurality of studies reporting positive associations, various papers cited above did not show such a clear-cut connection. Nielsen (2010b), for instance, was unable to establish a direct link between nationality diversity in management teams and corporate performance, and the empirical validation of a DOI-profitability relationship has been one of the most prominent concerns of International Business scholars for over thirty years (Glaum and Oesterle, 2007). Claiming that this research is not unique in finding inconsistent internationalization-performance relationships would therefore be somewhat of an understatement.

- 35 -

First, although upper echelons theory stresses the advantages of TMT heterogeneity in complex and uncertain environments, these benefits might not be applicable to either the phase of the business cycle or the type of diversity studied in this paper. A diversified board could well promote superior strategic decision-making and thus performance in the context of a highly competitive industry or era, yet the suddenness of the crisis might have curtailed managers’ scope to alleviate tensions and mitigate losses through the implementation of strategies that take time to implement. This notion relates closely to the finding by Carpenter and Fredrickson (2001) that TMT heterogeneity is negatively associated with organizational outcomes in the face of high volatility. Even if it could be established that some forms of TMT diversity enhance financial performance during and shortly after a crisis, it remains questionable whether this also holds true for variety in terms of managers’ national background (Nielsen, 2010a). While the knowledge and networks of foreign managers might represent a valuable asset in dealing with country-specific issues or in the process of cross-border expansion, nationality heterogeneity does not necessarily preserve its value in a period when demand is down nearly everywhere and firms are less eager to spread their operations geographically. During such times, diversity in terms of educational or functional backgrounds might prove much more beneficial (Greening and Johnson, 1996).

Still, the coefficients for TMTDiv were not merely insignificant in the estimations – in some regressions they were even significantly negative. In particular, for MNEs operating in industries with low average levels of nationality diversity, an increase in the variety index corresponds to a subsequent decrease in ROA. It is not inconceivable that this finding reflects an influential theory of heterogeneity introduced by Blau (1977), which states that intra-team conflicts and communication problems are most likely to occur in moderately heterogeneous teams, resulting in deteriorated organizational outcomes (Richard et al., 2004). In contrast, in highly diversified TMTs the benefits of heterogeneity typically outweigh the costs as the dominance of the largest subgroup is dwindled and discrimination is reduced (Blau, 1977). This theory might therefore also explain the positive coefficients in the estimations using the subsamples with above-average TMTDiv.