Minority Expropriation : Study on Tunneling in Norway

Transcription

1 Namhee Matheson Heidi Remmen BI Norwegian School of Management Thesis Minority Expropriation : Study on Tunneling in Norway Hand-in date: Campus: BI Oslo Exam code and name: GRA 1900 Master Thesis Supervisor: Professor Øyvind Bøhren Program: Master of Science in Financial Economics Master of Science in Business and Economics This thesis is a part of the MSc programme at BI Norwegian School of Management. The school takes no responsibility for the methods used, results found and conclusions drawn.

2 i. Content I. CONTENT... II LIST OF TABLES AND FIGURES... IV LIST OF APPENDIX... IV ABSTRACT... V 1 INTRODUCTION BACKGROUND MOTIVATION AND OBJECTIVES RESEARCH QUESTION THESIS OUTLINE THE THEORETICAL AND EMPIRICAL RESEARCH MARKET REACTIONS TO TUNNELING PROXIES FOR TUNNELING Separation of control and cash flow rights Ownership structure Private and Public firms Legal proxies DIFFERENT FORMS OF TUNNELING RESEARCH HYPOTHESIS AND METHODOLOGY TUNNELING MECHANISM The Ability The Incentive The Discretion TESTING FOR TUNNELING DETERMINANT OF TUNNELING Divergence of cash flow right Large Owner s Insider positions Second largest shareholder Regression HYPOTHESIS SUMMARY Testing for Tunneling Determinant of Tunneling DATA DATA DESCRIPTION Data Source Page ii

3 4.1.2 Population and Filtering process Family group identification VARIABLE DESCRIPTION Dependent variable Independent variable Control variables DESCRIPTIVE STATISTICS RESULT AND ANALYSIS REGRESSION RESULT: DOES TUNNELING EXIST? REGRESSION RESULT: DETERMINANT OF TUNNELING ROBUSTNESS TEST CONCLUSION...42 APPENDIX...44 REFERENCES...61 ATTACHEMENT: Preliminary thesis report Page iii

4 List of Tables and Figures Figure 1: Insider ownership and profitability of Norwegian listed firms... 2 Figure 2: Ability and Incentive of Tunneling Figure 3: Ability and Incentive of Tunneling Figure 4: Family group tunneling example Figure 5: Tunneling Flow in family group Table 1: Independent variable description Table 2: Conversion from Individual Owner into Family Table 3: Family Group Table 4: Sub sample description Table 5: Industry shock construction Table 6: Corporate Finance Descriptive Statistics (1) Table 7: Corporate Finance Descriptive Statistics (2) Table 8: Corporate Governance Descriptive Statistics (1) Table 9: Corporate Governance Descriptive Statistics (2) Table 10: Corporate Governance Descriptive Statistics (3) Table 11: Industry Descriptive statistics Table 12: Industry distribution of sample Table 13: Regression result Testing for Tunneling Table 14: Regression result - Determinant of Tunneling Table 15 : Regression result comparison to Stand Alone Table 16 : Inter corporate investment Table 17 : Regression result Robustness test List of Appendix Appendix 1: Tunneling Example Aker Solution Case Appendix 2 : Family group identification process Appendix 3 : Definition of variable Appendix 4: Corporate finance Descriptive statistics Appendix 5: Corporate governance Descriptive statistics Appendix 6: Industry descriptive statistics - Industry Frequency Appendix 7: Industry descriptive statistics-industry ROA Appendix 8 : Industry descriptive statistics-industry Asset Appendix 9 : Industry descriptive statistics-industry Sale Page iv

5 Abstract In this paper we empirically examine to what extent minority expropriation is present among family firms in the Norwegian economy. In particular we investigate a specific type of expropriation known as Tunneling : controlling families transferring resources from companies where they have few cash flow rights to ones where they have more cash flow rights. To investigate whether this phenomenon prevails in the Norwegian economy, we use a general empirical technique for measuring tunneling developed by Bertrand, Mehta and Mullainathan (2002). Based on cross sectional data for 2003, the results suggest a significant degree of tunneling between firms controlled by common family owner. The results also suggest that more tunneling prevails the greater the cash flow rights between two firms diverge and the fewer shares the largest second shareholder holds. *Acknowledgement We would like to thank our supervisor Professor Øyvind Bøhren for invaluable support and feedback on our thesis as well as patience, guidance and understanding in and for our thesis process. We would also thank Professor Bøhren for introducing us to the field of Corporate Governance and awakening our interest and knowledge in this particular area of study. We would also like to thank the Centre for Corporate Governance Research for providing us with the necessary data. Page v

6 1 Introduction 1.1 Background One of the biggest media scandals in 2009 was a transaction that occurred between Aker ASA and Aker Solution. The majority owner of Aker ASA had sold several firms to Aker Solution, a deal seemingly unfavorable to the minority owners of Aker Solution. The state, as the biggest minority owner, argued that the price was well above market value and that the minority shareholders were hurt by the deal. Investigations of the transactions concluded that the majority owner was within his legal rights but that the ethical aspects of the deal were questionable. The Aker Solution case is just one of the many cases where majority owners perform self-dealing actions that provide private benefits for the majority owner at the expense of the minority shareholders 1. Conflicts between majority and minority owners are one of the major themes within the field of Corporate Governance. According to Shleifer and Vishny (1997:1), corporate governance deals with the ways in which suppliers of finance to corporations assure themselves of getting a return on their investment 2. Along with increasing public attention to corporate governance, research on corporate governance has attempted to answer more fundamental questions Is economic value of the firm driven by governance mechanism? and If so, what factors in corporate governance affects the economic value of the firm and how? 1.2 Motivation and objectives The theoretical foundation of Corporate Governance is the Agency Cost Theory. Agency costs are caused by conflicting interests between the firm s stakeholders. These conflicts arise because stakeholders, with deviating interests, don t internalize the utility and wealth of other stakeholders. The main interest of corporate governance is to reduce the agency costs and to ultimately avoid value destructions caused by the agency costs (Jensen et al. 1976; Shleifer and Vishny 1997; Becht, Bolton, and Röell 2002; Tirole 2001). 1 See appendix 1 for relating article by BI Professor Øyvind Bøhren. Another famous International example is the Enron scandal where Thomas (2002) argue that some of the losses sustained by shareholders were as a direct result of related party transactions. 2 There are several definitions available. However the definition by Shleifer and Vishny is most consistent with purpose of our research. Page 1

7 Villalonga and Amit (2006) decompose the overall agency problem into the first and second agency problem. 3 The first agency problem deals with the traditional conflicts between outside shareholders and managers, whereas the second agency problem deals with the conflict between majority and minority stockholders. The second agency problem arises when a shareholder has majority control over the firm s assets and hence has the ability to pursue own interests at the expense of the minority shareholders. As expressed by Shleifer and Vishny (1997:758): Large investors may represent their own interest, which need not coincide with the interest of other investors in the firm. Figure 1 illustrates how the two types of agency costs affect the profitability of Norwegian listed firms as insider ownership increases.(bøhren and Ødegaard 2006). It shows that as insider ownership increases until about 40%, the profitability increases and then decreases again. The graph can be interpreted as net effect of first agency cost and second agency cost: that is, as inside ownership increases up to about 40%, inside owner has incentive to monitor the manager and thus reduce the first agency cost, leading higher profitability. However as ownership increases beyond 40%, the second agency problem becomes dominant and causes lower profitability in the firm. Motivated by empirical evidence and theoretical prediction, our main interest in this study is the effect of second agency cost on firms profitability. Profitability ( Tobin's Q) 1,8 1,7 1,6 1,5 1,4 1,3 1,2 1, Insider Ownership (%) Figure 1: Insider ownership and profitability of Norwegian listed firms *Profitability measured as Tobin s Q ratio (total market value of firm to total book asset value). Data is from *Øyvind Bøhren and Bernt Arne Ødegaard Governance and performance revisited. I International Corporate Governance after Sarbanes-Oxley 3 Even though we make the distinction between the first and second agency problem there also exists a third agency problem: the potential conflict between owners and creditors. (Shleifer and Vishny 1997) How the majority/minority conflict interlinks with owners/creditors conflicts is mentioned in papers by La Porta et al (2000), Berkman, Cole and Fu (2007) and Tang (2008). Page 2

8 There are many ways in which large owners can expropriate minority owners. We will research on a specific type of expropriation, tunneling, as illustrated in Aker Solution case. The term Tunneling was originally coined to characterize expropriation of minority shareholders in the Czech Republic. According to Johnson et al. (2000) and Glaeser, Johnson and Shleifer. (2001), the expropriation of minority shareholders through the stripping of the firm s assets was so common that it required a new term: tunneling. The assets disappeared from the firm as if removed through a hidden tunnel. Johnson et al. (2000) page 22: defines tunneling as the transfer of resources out of a company to its controlling shareholder (who is typically also a top manager). This definition has been adopted in the majority of tunneling research. This definition is not without weaknesses. Transfer of resources can in its widest interpretation include dividend, loans, salaries, as well as engaging in self-dealing transactions at non-market terms. Since it s a very wide definition; the interpretation and hence the use of the concept throughout the literature spans in a variety of directions. Johnsons definition also lacks a crucial component of tunneling; that tunneling expropriates the minority shareholder. We define tunneling in our paper as: The transfer of profit or resources by controlling shareholders from companies where they have few cash flow rights to ones where they have more cash flow rights through 4 related party transactions, which in effect leads to expropriation of the minority shareholder(bertrand, Mehta, and Mullainathan 2002) In our research we will study tunneling in the Norwegian economy. Norwegian data on non-listed and listed firms extracted from the CCGR database are made available from the Department of Financial Economics. By Norwegian law all limited liability firms have to publish an audited annual report. In addition the company must publish the identity of its CEO and its directors, and the fraction of equity held by every owner. (Berzins, Bøhren, and Rydland 2008:1). Most other nations do not have these kinds of disclosure requirements for private firms. 4 Transactions between the company and another entity, where one of its shareholders/board members/management has ownership stakes, are referred to as connected transactions or related party transactions. Page 3

9 The CCGR data base not only contains accounting data, corporate governance data, but also data on kinship and marriage. The unique data material gives us an exceptional position to conduct our research on tunneling. 1.3 Research question To get knowledge about tunneling in the Norwegian economy, we will investigate following two research questions: 1. To what degree does tunneling exist in Norwegian firm? 2. What are the major determinants for this phenomenon? With these research questions we wish to investigate how widespread and serious the minority-majority conflict is within the Norwegian Economy. By investigating different firm properties and external surroundings we hope to gain more knowledge about determinants for tunneling. 1.4 Thesis Outline Chapter 2 introduces the major theoretical and empirical research that is related to tunneling. Chapter 3 suggest research hypothesis relevant to answering this papers two research questions. A description of the tunneling mechanism is presented to give the reader a more systematic orientation to understanding the concept of tunneling. Based on the understanding of the tunneling mechanism, testing implications and hypothesis are presented alongside a regression model. Chapter 4 introduces the filtering and the sub sample construction process and illustrate the observation from descriptive statistics. Chapter 5 provides result of regression models and includes a discussion of the implications of the previously introduced hypothesis. A robustness test of the base case is also presented at the end of this chapter. In conclusion, Chapter 6, summaries this papers main findings and discuss the limitation with suggestion for future research. Page 4

10 2 The theoretical and empirical research The purpose of this chapter is to investigate the existing research conducted on tunneling and presenting theoretical and empirical insights for our research. The existing literature aims at identifying tunneling and its impact on firm performance and firm valuation. The literature on tunneling can be divided into two subsections: 1. The strand that examines market reactions when a publicly listed firm announce different types of related party transactions. 2. The strand that measures the degree of minority expropriation indirectly through the use of proxies for the degree of expropriation. The sections below give a summary of the two strands of research. The Chapter concludes with a presentation of the different ways of conducting tunneling 2.1 Market reactions to tunneling Most of the existing research conducted on tunneling has been done on publicly traded companies. The main focus of this strand of research is to investigate 1) to what degree valuation of listed firm is affected by related transaction, and 2) whether investor takes the effect of related transaction into account ex-ante. By observing market reactions to announced related party transactions, one can estimate the impact of such a transaction by investigating movements in the market value. This assumes that investors are able to predict the implications of the related party transaction and value the firm accordingly (Cheung, Rau, and Stouraitis 2006). It has been argued that the market will demand a discount ex-ante of firms which they think are more likely to undertake related party transactions. In such a manner they pay a fair price for their stocks given the risk of tunneling. An empirical study by Cheung et al (2006) did not find evidence that the market anticipated expropriation ex-ante. Instead they did find that after the announcement of a related party transaction the company could have a negative abnormal stock returns up to 12 months later. Page 5

11 2.2 Proxies for tunneling Since the different forms of tunneling often is hidden in the firm s regular operations research on tunneling mainly take use of proxies as indicators of tunneling. The most common proxies for tunneling are: 1. The use deviation of cash flow from control rights 2. The use of ownership structure 3. Listing status 4. The use of legal system Separation of control and cash flow rights This strand of theory and research tries to explain how minority expropriation takes place using different forms of ownership structure mechanisms that separate control rights from cash flow rights. The larger the gap between control rights and cash flow rights the greater is the incentive to expropriate. (Grossman and Hart 1988; la Porta et al. 1998; Claessens et al. 2002) These studies argue that this distinction can lead to lower shareholder value. Bebchuk et.al (2000) lists three mechanisms that entrench minority control and enable expropriation of minority shareholders:, pyramids, dual class shares and cross-holdings. For Western European countries, a study conducted by Faccio and Lang(2002) reported that both dual-class shares and pyramids are commonly used. Yet, a study conducted by La Porta, Lopez-de-Silanes and Shleifer (1999) finds little use of these mechanisms for Norwegian listed firms. The main rule in Norway is that every share is granted similar rights in the company, the so called 5 one-vote-one-share principle Ownership structure The literature on ownership structure as a proxy for tunneling deals with the ownership structure in the firm such as the size and distribution of shareholders and ownership types that is present in the firm. Concentrated ownership is normally associated with family ownership. When a family has ownership control, they most of the time also keep insider positions (Claessens, Djankov, and Lang 5 However, the firm can in its articles of incorporation implement dual class shares according to asl 4-1. Based on media coverage, it seems that dual class shares in private firms are used as a way for the founder to keep control, while at the same time distribute wealth to his children. However, we have yet to find any research relating to what kind of relationship the owners of the two classes of shares have to one another. Page 6

12 2000; La Porta, Lopez-de-Silanes, and Shleifer 1999). Almedia and Wolfenzon (2006) suggest that diverting corporate resources are more likely to take place in family business groups. Cronquist and Nilsson (2003) supports this view by arguing that when families are involved in the management of affiliated firms they have larger discretion of manipulation of corporate wealth Private and Public firms Most 6 of the prior research on tunneling has been conducted on publicly listed firms. Reasons for this can be that the effect of transactions (tunneling) on public firms easier can be observed in the company s market values. Public firms are also generally subject to stronger transparency requirements of enclosing company information to the market. This is problematic in a sense that private firms are actually the dominant form of firm in the economy. The exclusion of unlisted firms creates bias in terms of ownership structure and valuation. It also leads to considerable underestimates, as unlisted firms can have direct and indirect ownership in listed corporations. This can lead to a possible underreporting of the measures for ultimate ownership and control. (Claessens et al. 2002) Literature and theories on corporate governance and corporate finance suggest that private firms (as opposed to the listed firms) generally have a) higher ownership concentration (which is presumably major source of second agency problem) and b) are less transparent in conducting transactions, which might have same effect as low legal protection (Shleifer and Vishny 1997; Berzins, Bøhren, and Rydland 2008; McConnell, Servaes, and Lins 2008). Since literally (almost) all research on tunneling have been conducted on listed firms, we have to question whether the knowledge gained from research on listed companies can be applicable on unlisted firms. According to Berzins and Bøhren (2009) the answer is most likely no. They argue that unlisted firms operate under different external conditions than listed firms. They list the main external conditions as: financial market, transparency and regulations. 6 Some studies have been conducted on private firms. Gutierrez and Tribo (2008) examines how multiple large shareholders share control and extract private benefits in closely-held corporations in unlisted Spanish firms. Cheung et al.(2008) incorporates tunneling transactions between listed firms and non-listed subsidiaries in their research on connected transactions in China. Page 7

13 2.2.4 Legal proxies The majority shareholders opportunity and incentive to conduct tunneling is expected to be affected by the laws protecting the minority shareholders and the quality of the enforcement of these laws. Several studies have indicated that there is a great deal of tunnelling in countries with weak legal systems. (Johnson et al. 2000; Friedman, Johnson, and Mitton 2003; La Porta et al. 1998; La Porta et al. 2000) To protect the minority shareholder from tunneling, the Norwegian legal system includes several laws that impose restriction on transactions that can be of conflict of interests between a particular shareholder, board member or management and the company. The Norwegian Limited Liability Companies Act, Aksjeloven (asl), and the Norwegian Public Limited Liability Companies Act Allmennaksjeloven (asal), are the main laws regulating behavior related to connected transactions. In addition the Norwegian Accounting Law, Regnskapsloven (rskl), states disclosure and accounting requirements in relation to connected transactions. The main paragraphs constructed to restrict minority expropriation are: o asl. 6-27(1) and asal 6-27(1) o A member of the board of directors may not participate in the discussion or decision of issues which are of such importance to the board member in question, or to any connected person 7 of said board member that the board member must be regarded as having a major personal or financial special interest in the matter. The same shall apply for the general manager (Norge 2009:44) o asl. 6-28(1) and asal 6-28(1) o The board of directors or other parties who represent the company [ ] must not take any action that confer on certain shareholders or other parties on unfair advantage at the expense of other shareholders or the company (Norge 2009:44) 7 The law considers connected persons to be persons the shareholder has family relations to such as spouse, children, parents and siblings. The parents and siblings (and their spouses) of the shareholder s spouse are also considered connected persons (Norge 2009). Page 8

14 Basically, this means that a board member, representing a shareholder, is not allowed to partake in decisions where the shareholder or the shareholder s close family members could have personal interests. This is resolved by having the board member leaving the boardroom when such decisions are discussed. Some transactions are considered too big to be approved by the board alone. Transactions which surpass 10 % (AS) or 5% (ASA) of the firms share capital also require approval from the general meeting to be binding for the company (Norge 2009:asl/asal 3-8(1)). Not all decisions involving related party transactions require the same strict impartiality. The recommendation on the law of the Norwegian Limited Liability Companies Act states that one should put less weight on conflict of interest when the company is conducting regular business transactions 8. The argument is that a strict regulation of related party transactions might be out-weighted by the efficiency needs of the firm (Haagensen and Lie 2004). The efficiency transaction hypothesis argues that related party transaction can be beneficial to the firm since they reduces transactions cost, uncertainty and mitigate hold-up problems (Stein 1997; Ryngaert and Thomas 2007). Even though these regulations are designed to protect the company and its minority shareholders, they are far from water-tight. The indirect influence of a partial board member on the other board members can color the boards decision. The notion of what is a normal business activity and what is normal price are also very interpretive, as illustrated by the Aker Solution case. Due to the waste grey area that the majority shareholder can operate under it should be noted that the high legal costs of contesting such a transaction often can lead the minority shareholder either to accept the transaction or to exit the firm instead of confronting. If the minority owner in a listed firm wishes to exit, he can simply sell the stock. In non-listed firms, he has virtually no easy exit strategy. In unlisted firms, the minority owner has according to asl 4-24,4-25 8The Law defined a regular business transaction as : Agreements entered into as part of the normal activities of the company and which are based on a price and other terms and conditions which are normal for such a transaction (Norge 2009:asl/asal 3-8(1)) Page 9

15 the right to demand to be bought out if there exists a serious and enduring clash of interests between the shareholders regarding the operations of the company. This operation requires however both that the minority shareholder institutes legal proceedings, and that the company conducts a general assembly before the minority shareholder can exit the company. The law creates warnings signs and exit strategies for the minority shareholders when a large shareholder passes certain ownership stakes in the firm. For listed firms the Norwegian Securities Act ( Verdipapirhandelloven abbreviated to vphl ) makes shareholder obliged to flag ownership when they pass certain ownership stakes (vphl 4-2). They are also required to make a compulsory bid for all outstanding shares when they pass 1/3 9 of all outstanding votes in the firm (vphl 6-1). The same requirement for bid happens when the ownership stake passes 40 % and 50 % (vphl 6-6.). 2.3 Different forms of tunneling How tunneling is conducted and its impact on the majority owner depends on the resource being tunneled. Johnson et al (2000) separates tunneling into two forms: 1) the transfer of resources from the firm through self-dealing transactions, and 2) financial transactions that discriminate against the minorities. Transfer of resources through self-dealing transactions can include outright theft and fraud. However, most transactions that are considered tunneling are not criminal offences but related party transactions where the majority participant gains at the expense of the minority. Atanasov, Black and Ciccotello (2008) expands Johnsons et al s framework by dividing the first form of tunnelling into two categories: Cash flow tunnelling and Asset tunnelling. They rename tunnelling through financial transactions as equity tunnelling. Whether or not a resource transfer falls into the category of Cash flow tunneling or Asset tunneling depends on how the transaction affects the firm s future cash generating capacity. 9 This percentage was lowered from 40% in the 1997 law to 1/3 in the 2007 law. Page 10

16 Cash flow tunneling removes a portion of the current years cash flow (profit), but it does not significantly affect the firm s future cash generating capacity. Cash flow tunneling hence does not directly affect the balance sheet. Examples of cash flow tunneling are the sale of outputs to an intermediary controlled by insiders for below-market prices; or purchase of inputs at above-market prices. The purchase of services by related party transactions also falls within this category. Excessive executive salaries or perquisites and small-scale sales or purchases of assets also fall within this category. Asset tunneling involves the transfer of long-term (tangible or intangible) assets from or to the firm. They are distinct from cash-flow tunneling because the scale of the transfer has a permanent effect on the firm's future cash-generating capacity. Examples of asset tunneling include overpriced/underpriced purchases/sales of assets, or investments in an affiliated firm on better terms than the affiliate could obtain on its own. Equity tunneling increases the controller's share of the firm's value, at the expense of minority shareholders, but does not directly change the firm s productive assets. Examples of equity tunneling are dilutive offerings, freeze-outs of minority shareholders, loans to insiders (which will not be paid in bad states of the world), equity-based incentive compensation which exceeds market level and insider trading. Page 11

17 3 Research Hypothesis and Methodology Chapter 3 presents the research hypothesis used to answer the two research question; 1) does tunneling exist in Norwegian firm?, and 2) what are the major determinants for this phenomenon? The chapter begins with a description of the tunneling mechanism to give the reader a systematic understanding of the tunneling concept. Based on the understanding of tunneling mechanism, we move on to the testing implication of the first research question, followed by our hypothesis and regression models. Chapter 3 concludes with a hypothesis summary and the regression model presented in this chapter. 3.1 Tunneling Mechanism Having put the various theories in Chapter 2 together, we argue that tunneling is likely to occur when a combination of three factors prevails. These three factors are: (1) the large shareholder has sufficient control to expropriate minority shareholder, (2) the large shareholder has the incentive to tunnel,that is tunneling has positive effect on his net worth, and lastly (3) the firm s environment provides discretion to make easily hide tunneling. These three factors are the foundation of the hypothesis and will be referred to as (1) the ability, (2) the incentive, and (3) the discretion in same order. An explanation of each factor is presented below The Ability To understand first two factors (the ability and the incentive) conceptually, consider two firms A and B owned by common shareholder (denoted as Large Owner for the following). In addition assume Large Owner has 10 ultimate ownership share (or cash flow right) of X% and Y% respectively in firm A and B. Also consider the conditions when large owner tunnels 11 F amount of resource from Firm A to B as described in Figure To measure cash flow fraction, we will use ultimate cash flow right (hereafter simply referred to as cash flow right) so that all the possible structures such as pyramid and cross holdings can transform to simple form as figure above. Voting right, cash flow right will be interchangeable in our case. 11 F can be any form described in 2.3. Different forms of Tunneling. Page 12

18 Tunneling (GRA 1900 Master Thesis) Figure 2: Ability and Incentive of Tunneling In order to make the tunneling feasible, it is easy to understand that Large owner need to have sufficient control in Firm A. This is because the expropriation is initiated by the large shareholder, who can in fact influence the firm for his own interest. Here the meaning of sufficent control is generic. It might be either in the form of formal control given by voting right or in the form of informal control, which may magnify the control right, such as insider ownership or boad seat. We will further describe how informal control affects on tunneling in the following hypothesis. For formal control or voting right, it is of importance to clarify and limit the definition for the following test because there are various thresholds used to define sufficient control. Among the various theresholds, we chose absolute majority greaterr than 50% voting right as a lower bound of the control because of the following two reasons. First absolute majority is absolute threshold for control and thus always entails control in the firm. Secondly and more importantly it is useful to control the effect of first agency problem in light of empirical evidence, which shows that firm s performance increases up to about 40 % due to the reduction of first agency problem (Bøhren and Ødegaardd 2006). By choosing more than 50%, we can therefore test the effect of second agency problem without confusing with the first agency effect. Reflecting on the empirical evidences and logic descibed above, we chose ownership concentration greater than 50% range in our study as formal control to ensure the ability of tunneling. Page 13

19 Tunneling (GRA 1900 Master Thesis) The Incentive Even though sufficient control is a necessary condition for tunneling, it does not necessarily ensure expropriation. Large owner does not have a clear incentive to reduce value in Firm A deliverately if his or her net worth is expected to be unaffected ( or even negative for obvious reason). In view of this argument, we can infer that Large owner has incentive to tunnel when his net worth effect initiated by tunneling is expected to be strictly positive. As an example, assume that Firm A has $ 1,000 earnings and Large owner has CF right 51% and 100% in Firm A and Firm B respectively. If the Large owner is able to tunnel this earning to from Firm A to Firm B, his benefit would be increased to $1,,000 in firm B, relative to $510 if the earnings were kept in firm A. Because of the net gain the Large owner will look for a way to divert them out of firm A to firm B. Figure 3: Ability and Incentive of Tunneling Extending the example to the general case, Large owner would realize ex post benefit and thus have an incentive to tunnel if tunneling generates positive income that is ( 0 ;, when the large owner has more cash flow right in Firm B than Firm A. Reflecting this argument, we will refer to firm A type as Low CF firm and firm B type as High CF firm from now on. The setting described so far is most simplistic setting of all. The argument is now extended into a more realistic setting. So far the owner type of the Large Owner has not been explicitly specified as an individual or any other type of owner. As described in Chapter 2, diverting resources are more likely to take place in family business groups (Almeida and Wolfenzon 2006) and moreover family firm is the Page 14

20 Tunneling (GRA 1900 Master Thesis) most dominant firm type in the Norwegian economy (Berzins, Bøhren, and Rydland 2008). In light of these arguments, our unit of analysis will be family 12 not individual owner. A more realistic ownership structure and hence tunneling relationship can also involve group structures as well as bilateral firm relationship. Consider family X as a group owns 4 firms and let their sum of ultimate ownership in each firm, 55%, 65%, 75% and 90% respectively as described in figure 3. Based on the argument in simple setting, tunneling predicts that Large owner has incentive to divert between any combinations of two firms in the group as long as cash flow wedge between two firms are positive. Figure 4: Family group tunneling example For example (figure 4) for the perspective of firm D, the Large Owner would have incentive to tunnel to all the three other firms (A,B,C) in the group because his cash flow right in firm D is lowest of them all. Likewise he would tunnel out from firm C to firm B and A, and from firm B to firm A. From this example, we can infer that the firm with highest cash flow right in the group is the only recipient in tunneling relationship. By the same logic, the firm with lowest cash flow right in the group is only expropriated. The firm in the group with highest (lowest) cash flow rights will be referred to as Max (Min) CF firm in order to distinguish from other type of firms in the group (Mid CF firm). 12 Group of personal owners associated based on kinship and marriage. Page 15

21 Tunneling (GRA 1900 Master Thesis) Figure 5: Tunneling Flow in family group As minimum cash flow right in the group increases, the large owner is expected to internalize more due to increase in loss compared to benefit. Therefore beyond certain ownership point it is expected that the expropriation significantly decreases. Reflecting this argument, we selected a cut-off point of 80% of cash flow right in minimum cash flow firm and assume that beyond this point, the tunneling does not prevail in the group. To gather up two factors (ability and incentive) and generalize the arguments above, tunneling is likely to occur when: 1. A family owns more than two firms which it has ultimate sum of ownership greater than 50%. 2. The maximum ownership in the group is strictly greater than the minimum ownership. 3. The minimum ownership in the group exceeds 80% %. From this point, a group of firms that fulfills the three conditions mentioned above will be referred to family group and thus call three conditions as family group selection criteria. The three criteria will be used to find the sub sample (family group) which will be discussed in Chapter The Discretion Although the combination of the incentive and the ability for Large owner to tunnel could be perceived as creating conditions that could increase the risk of tunneling, theory predicts that these condition are not yet sufficient. Large owner would also have to take into consideration the cost of someone else finding out Page 16

22 about the tunneling transaction. In other word, the possibility of tunneling is intrinsically related to discretion, discretion that others cannot easily restrict (Dyck and Zingales 2004). Such discretion for Large owner to conduct tunneling depends on the firm environment. 3.2 Testing for Tunneling Tunneling is likely to occur in subtle and hard-to-detect ways by large shareholder s intention (Bertrand, Mehta, and Mullainathan 2002). Due to this reason, quantifying the extent of tunneling is proven to be a difficult task. This means that we can only use indirect measures. The only way to measure tunneling is to investigate an observable effect expected to be the result of tunneling. Rooted in corporate governance research, our fundamental ground is that corporate governance mechanism to some extent affects firm s performance. Corporate governance variables are expected to affect the potential and degree of tunneling which again manifests itself in the firm s performance. To be more precise, it is expected that Recipient firm shows higher performance and Tunneled firm shows lower performance relative to the performance that would be expected in the absence of tunneling. If one can 13 measure correctly the performance in the absence of tunneling for each firm or the fundamental earning, then tunneling is simply measured as the difference between the fundamental earning and the observable earning (earning diversion for the following): positive for the recipient in a tunneling relationship and negative for the tunneled firms. Another prediction of tunneling is such that the negative abnormal earning of tunneled firms causes the positive abnormal earning of recipient. This can be verified by testing the causal relationship between abnormal earnings of two group firms. To combine these arguments, tunneling predicts that when earning diversion is measured as the difference of observable earning from fundamental earning the following statements will be observed: 13 It is unrealistic but the purpose of this part is conceptual understanding. We will discuss more possible testing implication later on. For the following discussion we first would like to say that our model is extensively based on the general model for quantifying tunneling suggested by Bertrand, Mehta, and Mullainathan (2002) (denoted as BMM model for the following). Page 17

23 1. Positive earning diversion for recipient firms as a group as oppose to negative earning diversion for tunneled firms as a group. 2. Causal relationship: negative earning diversion for tunneled firm is cause of resulting positive earning diversion for recipient firm. While conceptually appealing, the problem for empirical testing arises in a sense that we cannot measure the fundamental earning perfectly. It is in fact proven as difficult as measuring diversion. Consequently, one needs to use best proxy to the fundamental earning. BMM model suggest that one of the good candidates is the industry movement to which each firm belongs to. A firm s industry affiliation is expected to influence the individual firm s earning to a large extent because different industries operate under different industry specific conditions and general economic conditions. For example will industry specific regulations, macro conditions and competition affect a firm s earning, and hence the firms in the same industry will be exposed to similar condition on their earnings. The BMM model suggests that industry movement is not actual fundamental earning, but more likely to be a major exposure for the individual firm. The BMM model calls industry movements as shock to the firm s performance level. Consistent with the use of term in BMM model, we will also use the term shock referring to industry performance. The important point to take into account is that this proxy works for aggregate industry level base not individual firm. For instance individual firm s performance can deviate from the industry average performance for numerous reasons. Observations of significantly low (or higher) performance for the group of Tunneled firm (or Recipient firm ), after incorporating control variables, can be interpreted as the diversion, caused by some systematic factor suspected to be tunneling. Suppose that the world price of gold rises, causing the gold industry s profits to rise on average. In other words, if the rise in gold prices increases profits in comparable firms by $100, then one can assume that if the reported earning is $90, then $10, on average, has been diverted away in simple setting. As a measure of industry performance, we will use industry median ROA (return on asset) and ROA for firm level performance. Page 18

24 To make comparison, family group defined in previous section is first identified before Max CF firm and Min CF firm are identified among the family groups. In addition, the sample contains firms where family owners don t have ownership stakes in any additional firms. These firms are referred to as Stand_alone. Our primary method will be to compare the difference between sub sample groups: Stand_alone, family group, Max CF firms, and Mid CF firms. If there are significant differences between groups in the direction expected, we will conclude that the data support our hypothesis. With the proxy for fundamental earning and observed earning at hand and construction of sub groups, the next step is to design the empirical model and describe the testing hypothesis for our first research question Does tunneling exist? First it is expected that family group is less sensitive to industry shock than Stand_alone firms. This is because tunneling predicts that earning is to some degree lost during the transaction between firms in the group while stand alone firms have no such influence. Consequently Group firm is expected to be less sensitive to industry shock. Therefore our hypothesis states: Hypothesis A 1): Family group is expected to be less sensitive to its own shock than Stand_alone. Let be a dummy variable for whether firm i is in a group or not. To test this hypothesis, the following regression is estimated: = + _ h + _ h + (1), where Controls are other variables that might affect firm performance and Own_Shock is industry shock measured as industry median. The coefficient b indicates how sensitive firms are, in general, to industry shock. The interaction term asks whether group firms are differentially sensitive to industry performance. If they are less sensitive, as tunnelling would predict, then c should be negative. Page 19

25 The first regression tests for reduced sensitivity of family group relative to Stand-alone. We now turn to testing whether, among family group, Min CF firm show less sensitivity than average group firm and Max CF firm is more sensitive to own shock than average group. Recall the example in figure 2 and 3. We conclude that Max CF firm is only recipient while other can be somewhat expropriated. Hence Max CF firm is expected to be more sensitive to its own shock compared to other in the group. Likewise Min CF firm is only expropriated and thus less sensitive than average to own shock. Therefore we suggest: Hypothesis A 2): Max CF firm is expected to be more sensitive to its own shock than the firms in the same group and Min CF firm is expected to be less sensitive to its own shock than the firms in the same group. Let _ ( _ ) be a dummy variable for whether firm i is in Max CF firm ( Min CF firm ) group or not. We then estimate the following regression for the sample of family group only. = _ _ h + (2) As before, the interaction term measures differential sensitivity. If Max CF firm is more sensitive than average, we would expect c to be positive. Likewise if Min CF firm is more sensitive than average, we would expect d to be negative. The most critical part of test is to verify whether diversion from fundamental earning for tunnelling pair has causal relationship. For testing we assume that Max CF firm ROA responds to industry shock of firms in the group as a whole. We will refer the average of industry shock of the firms in the group other than itself as group shock. We would also expect that Min CF firm does not respond to group shock in the group to confirm not only correlation but also the causal relationship. Thus we suggest that: Hypothesis A 3): Max CF firm is expected to be positively sensitive to group shock while Min CF firm is expected to be insensitive to group shock. Page 20

26 Let _ ( _ ) be a dummy variable for whether firm i is in Max CF firm ( Min CF firm ) group or not and _ h as group shock as defined. We then estimate the following regression for the sample of family group only like the previous test. = + _ h + _ h + _ _ h + _ _ h + (3) It is worth noting that we control for the firm s own shock. This control means that we do not confuse an overlap of industry between firms in the group with the flow of tunneling. A significant and positive coefficient d would suggest that Max CF firm is in fact sensitive to industry shock of other firms in the group. A further prediction of tunneling is that Min CF firm is insensitive to group shock. If this prediction is true, we would expect c to be insignificant. 3.3 Determinant of tunneling In this section we suggest several hypotheses to answer our second research question What are the major determinants for tunneling? Divergence of cash flow right As explained in tunneling mechanism, the divergence of cash flow right fraction in the two firms is main incentive for tunneling. By the same logic we can infer that the incentive is greater when the divergence increases. This is because the large shareholder loses relatively less than the gain he achieves from tunneling. For example in figure 2 and 3, we expect that minimum cash flow firm is more expropriated than firm B while both firms are expected to be tunneled. Hence we expect that: Hypothesis B 1): The incentive of tunneling increases as divergence increases between the cash flow rights in a firm in the family group and maximum cash flow firm in the same group. Page 21

27 3.3.2 Large Owner s Insider positions The previous hypothesis describes how Large owner s incentive to tunnel changes. This section describes how the ability side of Large owner affects tunneling. As discussed, sufficient control to make tunneling occur can be either formal control or informal control. We argue that informal control can also affect on tunneling through the effect on Large owner s ability to tunnel. One of the most widely discussed informal controls is insider ownership. Morck et al. (1988) argue that powerful insiders may entrench themselves and expropriate wealth from outside owners. Insider positions give discretion over day-to-day business which particularly increases the risk of cash flow tunneling. Legally, this can be linked to the relaxation in the regulation of related party transactions concerning regular business operations. We will examine how insider ownership of the largest family owner affects on tunneling in the case where family has CEO or board seats. Therefore, Hypothesis B 2): When the large shareholder holds insider positions, greater degree of tunneling is expected Second largest shareholder The existence of a minority shareholder with a significant ownership stake can function as a prevention mechanism of tunneling. If the majority shareholder is monitored it will reduce his or her incentives and/or ability to conduct expropriation. By the same logic as the tunneling argument, a second shareholder s incentive to monitor also depends on the cost-to-benefit ratio he or she is facing. If a minority shareholder holds a marginal stake in the firm it s assumed that the costs of monitoring will be too high to ensure monitoring relative to the gains. (Jensen and Meckling 1976)(Demsetz and Lehn 1985). Consequently Pagano and Röell (1998) predict that the size of the equity stake of the second largest shareholder affects this shareholders incentive to monitor the majority shareholder, hence affecting the degree of tunneling. While empirical research suggests the positive effect of size of the equity stake of the second shareholder, the legal framework indicates minimum share required to contest the largest shareholder (according to Bloch and Hege (2001) contestability). If a Page 22

28 shareholder or a group of shareholder holds more than 10 % of the voting shares in a limited liability company (5% in a listed company) they can: demand an extraordinary general meeting (asl 5-6) demand an inquire regarding specific aspects of the management or the account (asl 5-25) bring a claim for damages against the management or other owners (asl 17-4) Since our unit of analysis is family not individual, we define the second largest shareholder as a personal owner who is outside of largest family and has largest share among non-family shareholders. Based on the arguments above we expect that the ownership share of the second largest shareholder affects the largest shareholder opportunity to expropriate. Therefore, Hypothesis B 3): The existence of a large non-controlling family member shareholder works as a corporate governance mechanisms. Thus reduction in the ownership share of second largest shareholder increases the possibilities of tunneling Regression For the hypothesis related to the determinants of tunneling, we will estimate the following regression. Let be the independent variable representing each hypothesis and the rest of variables consistent with the previous regression equations. We then estimate the following regression for the sample of family group. = + _ h + _ h + (4) A significant coefficient c would suggest that the independent variable actually influence the sensitivity of its own shock. The expected signs of each independent variable are presented in table 1. Page 23

29 Table 1: Independent variable description Hypothesis Factor Description Abbreviation Expected Sign B 1 B 2 B3 Cash flow wedge Wedge of cash flow right between respective firm and maximum cash flow firm in the same family group CF - Share of the second The share of the second largest Second + largest shareholder shareholder(largest personal owner outsider of the controlling family) Insider CEO Controlling family has CEO. CEO - position Board Controlling family has chair. Board - Seat 3.4 Hypothesis summary Testing for Tunneling Hypothesis A 1): Family group is expected to be less sensitive to its own shock than Stand_alone. Hypothesis A 2): Max CF firm is expected to be more sensitive to its own shock than the firms in the same group, and Min CF firm is expected to be less sensitive to its own shock than the firms in the same group. Hypothesis A 3): Max CF firm is expected to be positively sensitive to group shock while Min CF firm is expected to be insensitive to group shock Determinant of Tunneling Hypothesis B 1): The incentive of tunneling increases as divergence increases between the cash flow rights in a firm in the family group and maximum cash flow firm in the same group. Hypothesis B 2): When the large shareholder holds insider positions, greater degree of tunneling is expected. Hypothesis B 3): The existence of large non-controlling family member shareholder works as a corporate governance mechanisms. Thus a reduction in ownership share of second shareholder increases the possibilities of tunneling. Page 24

30 4 Data 4.1 Data Description Data Source In this thesis we use data from the CCGR database. The database provides accounting, corporate governance data and ownership data (containing ownership ID and share in each company) for all Norwegian limited liability firms. With this data we will do cross sectional data analysis for year We selected this year as it appear as the most neutral year, avoiding the effects of 9/11 (2001), the tax reform of 2005 (Norge 2004),and any affects associated with the current financial crisis Population and Filtering process The CCGR database for accounting, corporate governance variable contains a sample of 155,996 firms in For a firm to be able to tunnel the firms have to be active. Hence we exclude non operating firms that 1) do not have any asset (total asset) and 2) operating income in We also exclude financial firms because they are subject to strict regulations which reduce the risk of tunneling. Since it is critical for our test to use the industry performance, we exclude the firms with missing industry code. Having filtered non-operating firms and financial / missing industry firm we are left with 127,306 firms. For ownership database, the CCGR database in 2003 does not provides ownership ID for all the reported firms contained in accounting/ corporate governance data: the number of observed firms in ownership database is 132,670 firms. Hence we further filtered the firm which miss the ownership data and obtained 111,539 firms. We refer to this sample as overall economy sample. From the overall economy sample the sample in our interest is the firms in which largest family sum of ownership (item_15302 in the database) is strictly greater than 50%. We observed 71,832 firms which meet this criterion as opposed to 39,707 firms for the rest. Further we exclude 1,897 firms 14 we could not find the largest family owner identification. As a result we get 69,935firms. This is our testing sample and for the following we refer it as family firms. 14 See the description in Appendix 2: Family group identification process. Page 25

31 4.1.3 Family group identification As described in chapter 3 briefly, we need first to construct sub samples from family firm: family group and Stand_alone. To identify any links by common family owner, family identification and its ownership stake in each firm is critical information for testing. However we could not get the family membership data regarding which owners construct family groups due to the data restriction. Since this information is key information for this study, we used indirect measures to estimate the match between ownership ID and family group membership 15. As a result, we classified individual owner of 80,801 observed in family firm into 53,064 families with its own family ID. The table 2 shows number of family members in the family we assigned and the corresponding number of families. The figures below should be used with caution: they should not be interpreted as the entire family membership since we consider only family firm sample to construct family membership. The majority of family firms are composed of only one person, 67% of sample families. Families with two to seven family members constitute the majority of the remaining sample while families with more than seven members constitute just few cases. Table 2: Conversion from Individual Owner into Family Number of Number of % of Number of % of Family Members Families Total Family Personal Owners Total Owner 1 35, % 35, % 2 10, % 21, % 3 4, % 12, % 4 1, % 7, % % 2, % % % 7 to % % Total 53, ,0% 80, ,0% Once family membership is defined, we identify the largest family owner for each firm (for detail see Appendix 2). Then we grouped the firms according to common largest family owner. Table 3 presents the overview of the result. We define the 15 See the description in Appendix 2: Family group identification process. Page 26

32 firm which one family owns only one firm as Stand_alone. The rest of firm we define as Group firm as opposed to Stand_alone. Among Group firm, we identify Family group according to family group selection criteria, described in page 16. For example if we observe Group firm where all the firms in the group have same cash flow right owned by the largest family,or minimum cash flow right in the group is greater than 80%, we exclude them from the Family group sample. As a result the sample is reduced from 26,821 firms in Group firm to 8,879 firms in Family group. Among Family group, we group the maximum cash flow firm as Max CF firm and minimum cash flow firm as Min CF firm. The observations for Max CF firm and Mid CF firm are 2,821 firms and 2,778 firms respectively. Table 4 summaries the sub sample selection and corresponding sample size. For the test we will use four sub samples: Stand_alone, family group, Max CF, and Min CF. Table 3: Family Group #of Family Member to 14 Total % of Total # of Firm Owned 1 Stand_alone 30,633 8,653 2,499 1, ,114 62% 2 7,062 3,206 1, ,316 19% 3 2,367 1,290 1, ,496 8% 4 1, ,712 4% ,520 2% ,020 1% Over 7 1, ,757 4% Group Firm 12,632 6,032 4,413 2,247 1, ,821 38% Total 43,265 14,685 6,912 3,269 1, , % Table 4: Sub sample description Sample Name Description Observation Filtered population Applied the filters described in page Family Firm Largest family ultimate sum ownership >50% Stand Alone 1+ Family owns only one firm Group Firm 1+ Family owns several firms Family Group 3+ Apply Family group selection criteria (p.21) Max CF 4+ Maximum cash flow firm in the family group Min CF 4+ Minimum cash flow firm in the family group Page 27

33 Variable description Dependent variable There are many different measure of performance. In our research most of firms are un-listed firms, which indicate that the commonly used market based performance measures such as Tobin s Q ratio and ROE cannot be used. Consequently we will use accounting based measure: return on assets (ROA) as our measure of performance. ROA is extracted from CCGR database and defined as earning before interest after tax (EBI) divided by total asset Independent variable To measure industry shock, we used median ROA of the industry to which each firms belong based on NAIC code. The industry median was constructed in family firm sample as a whole because this is the reference performance of each industry in the economy. This variable is referred to as Own_Shock. To measure group shock, we measured the average of Own_Shock of firms in the group except itself. For example, consider the family group described in Table 5. Since each firm has its industry code, Own_Shock can be easily indentified. Once Own_Shock is identified for all the firms, we then compute the average of Own_Shock for firm B to firm D as Group_Shock for firm A,and firm A to firm C as Group_Shock for firm D. It is worth nothing that we exclude its own shock from computing group shock because group shock and own shock is the distinct independent variables in the regression test. Table 5: Industry shock construction Firm Name Family Share Industry (Example) Own shock A 55% Service Median ROA of Service Industry B 60% Transport Median ROA of Transport Industry C 75% Manufacturing Median ROA of Manufacturing Industry D 90% Construction Median ROA of Construction Industry Group shock Average of Own_Shock of firm B,C and D Average of Own_Shock of firm A,C and D Average of Own_Shock of firm A, B and D Average of Own_Shock of firm A, B and C 16 See the Appendix 3 for details on how each variable is constructed from the data source. Page 28

34 For hypothesis B 1), CF wedge ( CF ) is defined as difference between shares in respective firm s largest family share and family share of maximum CF firm in the same group. We will measure hypothesis B 2) insider positions as two dummy variable CEO and Board. CEO variable takes the value of 1 if controlling family has CEO and 0 otherwise. Likewise Board variable takes the value of 1 if controlling family has board seat. In our study, the second largest shareholder is defined as the personal owner who is not a member of the largest family in a given firm and has the highest share among the non family owners. For hypothesis B 3) we simply measure the ultimate share of the second largest shareholder in a given firm ( Second ). While the other variables were readily available from initial source, this variable is found by matching ownership data and largest family. For example if the firm has five personal owners and three of them are the members of controlling family, we then know that rest two owners are outside of controlling family. If among two outside owners one has 10% share and 5% share, the former is defined as the second largest shareholder and his or her share as Second variable Control variables To control for other firm characteristics that can influence the dependent variable the following control variables is introduced in our regression: firm size ( Size ), firm age ( Age ), leverage ( Leverage ), firm growth ( Growth ). There is a general consensus that firm size affects firm performance. Large firms have a tendency to be the most successful businesses because expansion (increase in size) is often a result of profitability. Two proxies are used as indicators of firm size: total asset and operating revenue. To measure we take log value of total assets ( Size (asset) ) and operating revenue ( Size (sale) ). The age of a firm is expected to influence the firm s performance in two ways. Older firms are expected to have higher ROA because older firms normally have more sales and hence higher profits. Age might also be linked to performance due to a self-selection bias: older firms might be presents simply because they are Page 29

35 successful (Schulze et al. 2001). Older firms also generally have depreciated their assets more than younger firms leading to higher ROA. To control this effect on ROA, we measure the age of a firm as log of age ( Age ). Leverage can both have a positive and negative impact on the firm s performance. Jensen (1986) argues that debt will reduce the firms potential agency costs by lowering the available free cash flow in the firm. Leverage can also make managers more efficient since they must meet debt repayments (Stiglitz 1985). McConnell and Servaes (1995) argue that the same debt repayments hinder the managers to invest in profitable investments because of constraints (e.g., covenants) associated with the debt. Leverage is expected to correlate negatively with ROA since more profitable firms can finance more from earnings. We define leverage as total debt divided by total asset ( Leverage ). Firm value may be related to the firm s investment opportunities. High growth firms tend to be more profitable than low growth firms (Maury 2006; Cooper, Gulen, and Schill 2008). We use the proxies for growth as sales growth ( : g_sale ). 4.3 Descriptive statistics Appendix 4 presents summary statistics of key corporate finance variables for each sub group (Panel A: Family firm, Panel B: stand-alone firm, Panel C: Family Group, Panel D: Min CF firm, and Panel E: Max CF firm). Appendix 5 reports statistics for the corporate governance characteristics. Table 6 to 10 presents histograms of median values of key variable in order to make the comparison between sub groups easier. Table 6 shows that family group tends to be larger than stand alone. A median family group recorded total asset of 3,6 million NOK in 2003 and sales of 3,0 million NOK. In contrast median stand alone firms recorded total asset of 1,2 million NOK and sales of 1,8 million NOK. On the other hand we did not observe noticeable size difference in term asset among sub samples of the family group. In terms of sales the median Max CF firm recorded almost half of sales of the median Min CF firm. Page 30

36 Table 6: Corporate Finance Descriptive Statistics (1) 4,00 3,50 3,00 2,50 2,00 1,50 1,00 Family StandAlone Group 4,0 3,5 3,0 2,5 2,0 1,5 1,0 Group Min CF Max CF 0,50 0,00 A S g_s D D to A 0,5 0,0 A S g_s D D to A *A : Total Asset, S: Sales, g_s: growth of sales from 2002 to 2003, D : Total debt and D to A : Leverage ratio defined as total debt to total asset (for description of variables, see Appendix 3 and for full statistics, see Appendix 4) *The value shown is million NOK except g_s and D to A as of 2003 * Missing column represents value of zero. Table 7: Corporate Finance Descriptive Statistics (2) Age ROA Max CF Min CF Group StandAlone Family 0,00 2,00 4,00 6,00 8,00 10,00 12,00 Table 7 presents that the median age and ROA. The median family group firm is one year older than stand alone, while median Max CF firm is oldest among them all. We will control for these difference with control variables described previously. Descriptive statistics for ROA will be further discussed in association with industry sector descriptive statistics. Page 31

37 Table 8: Corporate Governance Descriptive Statistics (1) 120,00 100,00 80,00 60,00 40,00 20,00 0,00 Sum of Largest family ultimate share Sum of Largest family direct share Second CEO Holding Largest share CF WEDGE Family Stand Alone Group Min CF Max CF * Missing column represents value of zero. * Second largest share has median value of zero for all the sub samples except Min CF because sum of largest family ultimate share has median value close to 100 for those. * CF wedge applies only for family group. As shown in table 8, the median stand alone firms recorded sum of largest family ultimate share of 100 as opposed to 90 of the median family group. Consistent with this the second largest share is lower and CEO holding is higher for the median stand alone firm compared to the median family group. Among the family group firms, the difference is driven by the method how we classify Min CF and Max CF. Table 9: Corporate Governance Descriptive Statistics (2) 3,50 3,00 2,50 2,00 1,50 1,00 0,50 Family Stand Alone Group Min CF Max CF 0,00 # Family Owner # Family chair # Owner Board Size As illustrated in Table 9 the median number of family members is one in all five groups. This is with the observation in table 2 and 3 suggesting that majority of family group consists of one person. Median family group recorded three owners compared to two owners in median stand alone firms. We expect that this tendency is somewhat related to the size difference of two groups. Page 32

38 Table 10: Corporate Governance Descriptive Statistics (3) Family CEO Family Chair CEO and Chair Family Stand Alone Group Min CF Max CF *The histogram presents the percentage of firms in which controlling family has 1) CEO, 2) Chair, and 3) CEO and Chair both. Table 10 presents that it is quite common for family to have CEO or board seat. The firms in which controlling family has both CEO and board seat are majority: 76%, 49%, 41%, and 65% for stand alone, family group, Min CF, and Max CF respectively. In contrast the firms with neither family CEO nor family chair are 3%, 12%, 14%, and 4% in same order (see Appendix 5). We expect that group differences reflect higher controlling family ownership in stand alone and Max CF firms (median value 100%) compared to family group (90%) and Min CF firm (65%). Table 11: Industry Descriptive statistics 12,0 10,0 8,0 6,0 4,0 2,0 0,0 10,0 9,0 8,0 7,0 6,0 5,0 4,0 3,0 2,0 1,0 0,0 Asset Family Firm Asset Stand- Alone Asset Group Note: The graph shows median of ROA and asset of three groups. The line chart (primary axis) represents median ROA of the group for corresponding industry sector in horizontal axis. The histogram (second axis) represents median asset in million NOK. Industry sectors are 1: Agriculture, forestry, fishing and mining, 2: Manufacturing and chemical products, 3: Energy, 4: Construction, 5: Service, 7: Trade, 8: Transport and Total for total sample. We classified the industry sector as reference of (Berzins, Bøhren, and Rydland 2008)For detail industry analysis based NAIC code, see appendix. Page 33

39 In addition to general descriptive statistics for overall industry, we further present the descriptive statistics by industry. The purpose is first to get snapshot on how each group performance is difference from industry norm and second to see whether any group is particularly concentrated on certain industry. Appendix 6 to 9 presents industry distribution of each sub group, ROA by industry, asset by industry and sales by industry respectively. Table 11 and table 12 present the summary from those appendixes. Most interestingly median ROA of family group is lower than those of stand alone firms in all industry sectors. In previous section simple description on corporate finance variables showed that median family group is larger and older than median stand alone firms. The description on control variable predicts that larger and older firm tends to have higher performance than smaller and younger firm. Considering these observations and theoretical predictions, lower median ROA of family group signals that our hypothesis might fit to the data. Table 12 suggests that there is no noticeable difference in industry distribution except industry sector 4: construction and 5: service. Stand alone firms are relatively more concentrated in the construction sector and less so in the service sector in comparison to family group. Table 12: Industry distribution of sample Multi Family 2,12 % 7,55 % 0,13 % 9,69 % 48,34 % 22,85 % 4,50 % 4,16 % StandAlone 2,15 % 7,62 % 0,13 % 11,88 % 44,78 % 23,43 % 4,77 % 4,68 % Group 2,31 % 7,31 % 0,23 % 5,67 % 54,96 % 20,79 % 4,72 % 3,21 % Min CF 2,48 % 8,32 % 0,25 % 7,20 % 51,04 % 21,53 % 4,18 % 4,36 % Max CF 1,95 % 5,42 % 0,25 % 5,28 % 56,86 % 19,74 % 3,90 % 5,81 % Note: The table shows the percentage of each industry to total observation in respect group sample. The industry sector is same as Figure 3 and Multi represents multi industry. Max CF is the maximum CF firm in family group while Min CF is the minimum CF firm in family group For the regression analysis, we excluded multi industry firms from the sample because we cannot compute the own shock and group shock variable without knowing weight on each industry of multi industry firms. Table 12 shows that multi industry firms are not populated in particular sub group: overall 3,2% to 5,8% of sample size of each group. By excluding multi industry and unspecified Page 34

40 industry firms, the sample reduced from 8,879 firms to 8,522 firms for stand alone and from 2,778 (2,821) firm to 2,639(2,635) firms for Min CF (Max CF) group. Page 35

41 5 Result and analysis 5.1 Regression result: Does tunneling exist? First we test the first prediction of tunnelling: whether group firms would be less sensitive to their own industry shocks than Stand_alone. Column (1) in table 13 displays our basic result. The general sensitivity of firms to industry performance is, as expected, positive and significant. More importantly the interaction term which captures the differential sensitivity of group firms is negative as tunnelling predicts. The result can be interpreted as follows. If industry shock (measured as industry median which each firm belongs to) increases by one unit, it leads to about 1.19 unit increase in ROA of stand alone firms. For a group firm, the same degree of shock leads to 0.67 unit smaller increase in ROA, or only 0.52 unit sensitivity to one unit industry shock. This suggests that some part of profitability in a group firm is lost. In short, the data supports the first prediction. Table 13: Regression result Testing for Tunneling (1) (2) (3) Own_Shock 1,191 (,450)***,437 (,154)***,664 (,139)*** Group_Shock - - -,365 (,169)*** Group* Own_Shock -,669(,073)*** - - H_CF* Own_Shock -,532 (,132)*** - L_CF* Own_Shock -,019 (,130) - H_CF* Group_Shock - -,479 (,149)*** L_CF* Group_Shock - - -,040 (,148) Size (Asset) 3,215 (1,738)*** 1,715 (,389)*** 1,734 (,391)*** Size( Sale) 8,417 (1,481)*** 1,417 (,334)*** 1,374 (,334)*** Age 4,139( 2,125)** 3,459 (,560)*** 3,516 (,560)*** Leverage 5,470 (,044)*** -2,349 (,086)*** -2,350 (,086)*** Group* Leverage (a) -8,755 (,190)*** - - g_sale,001 (,003),000 (,000),000 (,000) Constant -171,889 (20,774)*** -48,506 (4,934)*** -46,976 (4,961)*** F 1997,403 *** 131,861*** 116,742*** Adjusted R 2,244,110,109 # Observation 49,382 8,522 8, Own_Shock is industry median of the industry each firm belongs to. Group_Shock is average of Own_Shock of the firms in the same family group except itself. Group is binary variable whose value is 1 for family group firms and 0 for stand alone firms. H_CF (L_CF) is binary variable whose value is 1 for maximum (minimum) cash flow position in family group and 0 for otherwise. Size and Age are log value of observation as of g_sale is sales growth measured as ratio of sale in 2003 to sale in Leverage is total debt to total asset. 2. Standard errors are in parentheses. *** indicates that coefficients estimates are significant at the 1% level according to the student test, ** at 5%, and * at 10% level. 3. (a) Add the variable to control the different coefficient sign for two groups. Page 36

42 Regarding control variables: Size and Age, the regression result is consistent with our prediction that larger and older firms tend to perform better on average. Leverage variable suggests more interesting result: two groups of firms show opposite coefficient signs. We have already discussed the conflicting effect of leverage on firm s performance and inconclusive net effect. Here the result suggests that for stand alone firm, positive leverage effect prevails while negative leverage effect is dominant in family group. In our test, growth measure does not provide significant effect on ROA. These tendencies are consistent throughout the rest of the tests. The second prediction provides a more stringent test: within group firms, Maximum CF firm should show greater sensitivity than average group firm. Column (2) in table 5 shows that Maximum CF firm is more sensitive to its own shock than group firm on average, indicated as positive and significant coefficient in interaction term (H_CF*Group_Shock). The result implies that one unit increase in industry shock leads to about 0,43 unit increases in ROA for a group firm on average. For Maximum CF firm, it leads to 0,53 unit greater increase, or about 0,96 sensitivity to one unit industry shock in total. Combining the finding in the first test, the result suggests that Maximum CF firm is only slightly less sensitive to industry shock than stand alone firm (1,19 sensitivity). The coefficient on L_CF interaction term indicates that Minimum CF position is indifferent from other position in the group firm when it comes to its own sensitivity. The result also suggests that for the rest of the group firms one unit increases in own industry shock leads 0,4 unit increase in ROA, which is slightly less than group as a whole (0,67) due to the exclusion of Maximum CF firm. Therefore data supports the second prediction as well. Column (3) shows the result of third prediction: Maximum CF firm would be positively sensitive to group shock. The negative coefficient on group shock indicates that group firm on average is negatively related to each other s shock. More importantly interaction term and group shock coefficient term indicates that Maximum CF firm is positively associated with group shock: on average 0,1 unit increase as a response of one unit industry shock in other firm of the group as a whole. Minimum CF firm shows indifferent sensitivity to group shock, which confirm the tunnelling flow direction from down to the top of CF hierarchy. Page 37

43 5.2 Regression Result: Determinant of Tunneling The results of first part of the test generally support the prediction of tunnelling. Base on the empirical support on the existence of tunneling, we move on to the second part of the test for finding what affects the degree of tunnelling. From Chapter 4, our hypotheses suggest three factors affecting on the extent of tunnelling: CF wedge, second largest owner, and insider position of largest family. Table 14 show both univariate and multivariate test for three independent variables: column (1) CF wedge, column (2) second largest owner, column (3) / (4) insider position measured as family CEO / family board seat. The last column (5) shows the multivariate regression result for three independent variables altogether. Testing for CF wedge and second largest owner supports our prediction. The negative interaction term for CF wedge suggests that as more CF right in the group firm deviates from Maximum CF right in the group, more tunnelling is prevailed. We can infer that one percent CF wedge change causes 0,005 unit decreases in its sensitivity to own industry shock. Considering average CF wedge is about 12% (see Appendix 5: Panel C.), CF wedge on average drives the sensitivity 0,06unit down for one unit industry shock. The result for testing second largest owner variable is also consistent with our hypothesis that as the share of the shareholder outside of largest family increases, tunneling is less prevailed. The positive coefficient on the interaction term supports the tunneling prediction: one percent increase in outside shareholder s share causes 0,01unit increase in its sensitivity to one unit own industry shock. Considering average second largest share is about 0,6% in group firm (see Appendix 5: Panel C.), second largest share on average contributes the sensitivity increase by 0,11unit for one unit industry shock. On the other hand insider position does not support our prediction that insider position increases ability of large family and as a result more tunneling would be expected when large family has insider position. The result shows that Family CEO contributes to reduce tunneling while Family board member plays no significant role. The result might be because the regulations restricting insiders in conducting related party transactions in Norway is effective. Alternatively it might be that controlling families have already sufficient control in terms of voting right and thus whether to have insider position does not particularly make difference. Page 38

44 The results in univariate test were all consistent with multivariate test as well. One noticeable finding is that CF wedge gains more significant and magnitude of the effect is two times larger than univariate case. Second largest share also shows slight increase in magnitude. Table 14: Regression result - Determinant of Tunneling (1) CF (2) Second (3) CEO (4)Board (5) Multivariate Own_Shock,675 (,141)***,474 (,138)***,608(,145)***,612(,157)***,672 (,178)*** CF *Own_Shock -,005(,003)* ,010 (,004)*** Second* Own_Shock -,011(,003)** * - -,016 (,004)*** CEO*Own_Shock - - 5,451(1,049)*** - 5,605 (,1,072)*** Board *Own_Shock ,009(,063)- -,083 (,065) Size (Asset) 1,648(,385)*** 1,881(,387)*** 1,827(,385)*** 1,713(,384)*** 2,005 (,390)*** Size( Sale) 1,309(,334)*** 1,118(,333)*** 1,345(,333)*** 1,234(,331)*** 1,297 (,336)*** Age 3,644(,560)*** 3,849(,558)*** 3,526(,562)*** 3,738(,559)*** 3,558 (,564)*** Leverage -2,355(,086)*** -2,348 (,086)*** -2,349(,086)*** -2,355(,086)*** -2,338 (,086) Growth,000(,000),000(,000),012,000(,000),000(,000),000 (,000) Constant -46,33(4,883)*** -47,14(4,889)*** -52,36(5,079)*** -46,21(4,888) -54,46 (5,101)*** F 147,830 *** 149,101 *** 150,670 *** 147,435*** 107,864 *** Adjusted R 2,108,109,110,108,112 # Observation 8,522 8,522 8,522 8,522 8,522 Table 15 : Regression result comparison to Stand Alone (2)Second (3) CEO (4) Board Own_Shock 1,154 (,539)*** 1,125(,708)*** 1,204 (,521)***,010 CF *Own_Shock Second* Own_Shock,008 (,017) - - CEO*Own_Shock -,116(,567) - Board *Own_Shock - -,295 (,781) Size (Asset) 3,115(2,204) 3,143 (2,207) 3,139 (2,204) Size( Sale) 10,716 (1,844)*** 10,710(1,850)*** 10,771 (1,844)*** Age 3,909(,2,533)* 3,849 (2,529)* 3,815 (2,530)* Leverage 5,472(,047)*** 5,472(,047)*** 5,472 (,047)*** Growth,008(,017),004 (,008) 004 (008) Constant -202,016(24,598)*** -202,439 (24,582)*** -203,140 (24,609)*** F 1,913,802*** 1,913,763*** 1,913,782 Adjusted R 2,247,247,247 # Observation 40,860 40,860 40,860 *CF is cash flow wedge between each firm and maximum cash flow firm in the same family group. Second is the share of second largest shareholder. CEO (Board) is binary variable whose value is 1 when family has CEO (Board members) and 0 for otherwise. *Other variables / description are same as in table 10. To confirm the finding we did similar regression test for Stand alone firm as shown in Table 15 except for CF wedge. All the test results show no significant result in sample of stand alone firms. This confirms that our prediction captures unique variances in group firms. Page 39

45 To summarize, the results gives support for that a large cash flow wedge makes tunneling more attractive (incentive) and that less contesting by other shareholders makes tunneling more prominent (discretion), but also that informal control through insider positions (ability) seems restricted. 5.3 Robustness test Although the findings in previous section support the predictions of the tunneling hypothesis, other possible explanations need to be considered. We believe that alternative explanations that ought to discuss: the dividend earning from the shares held in each other can be prominent alternative explanation. Hence we did first part of regression test again considering inter-corporate investment between group firms. One might worry that the results merely arise from the possible inter-corporate investment between the pair of companies resulting in owning shares in each other (Bertrand, Mehta, and Mullainathan 2002). If this is the case, the sensitivity of one firm in the pair to the other s performance would then mechanically arise through the dividend income earnings from the shares held. To take into account this possibility, we first check the group composition which CCGR data provides (See Appendix 3 for variable description.). Table 16 shows the group identity of family group sample. It is expected that the cases where maximum CF firm is parent of the group are what we have to concern. This is because the dividend payment flow is same direction as tunneling from low cash flow firm ultimately to maximum CF firm. Hence we found the family group where maximum CF frim is parent and removed all the family group member firms from the sample in order to re-test. As table 16 presents, the family group in which Max CF firm is parent in corporate group is 548 cases. By this process, the sample is reduced to 6,837 firms from initial 8,522 firms. Table 16 : Inter corporate investment Min CF Max CF Mid CF Group Total Associated Independent JC Parent Subsidiary Total Page 40

46 Table 17 shows the results of regression test for Does tunneling exist? with the reduced sample. The result is robust after applying new sample. The sign and magnitude of coefficient remains similar to those in base case for all three regression test. Table 17 : Regression result Robustness test (1) (2) (3) Own_Shock 1,200 (,463)***,439 (,179)***,640 (,163)*** Group_Shock - - -,266 (,199) Group* Own_Shock -,671 (,081)*** - - H_CF* Own_Shock -,501 (,156)*** - L_CF* Own_Shock -,079 (,151) - H_CF* Group_Shock - -,417 (,177)*** L_CF* Group_Shock - - -,026 (,175) Size (Asset) 3,317 (1,818)*** 1,406 (,388)*** 1,376 (,388)*** Size( Sale) 8,746 (1,544)*** 1,875 (,455)*** 1,843 (,458)*** Age 4,160 (2,199)** 3,554 (,648)*** 3,601 (,648)*** Leverage 5,471 (,045)*** -2,441 (,092)*** -2,442 (,092)*** Group* Leverage (a) -8,763 (,194)*** - - g_sale,001 (,003),000 (,000),000 (,000) Constant -177,925 (20,774)*** -50,953(5,751)*** -46,419 (5,775)*** F 1930,196 *** 118,618 *** 104,871*** Adjusted R 2,244,121,121 # Observation 47,697 6,837 6, Own_Shock is industry median of the industry each firm belongs to. Group_Shock is average of Own_Shock of the firms in the same family group except itself. Group is binary variable whose value is 1 for family group firms and 0 for stand alone firms. H_CF (L_CF) is binary variable whose value is 1 for maximum (minimum) cash flow position in family group and 0 for otherwise. Size and Age are log value of observation as of g_sale is sales growth measured as ratio of sale in 2003 to sale in Leverage is total debt to total asset. 2. Standard errors are in parentheses. *** indicates that coefficients estimates are significant at the 1% level according to the student test, ** at 5%, and * at 10% level. Page 41

47 6 Conclusion Our research is motivated by stream of corporate government research, pursuing to expand the knowledge on how corporate governance mechanism affects on firms profitability. In this paper we attempt to understand how controlling owners expropriates minority owners in the Norwegian economy where private firms controlled by family are dominant form (Berzins, Bøhren, and Rydland 2008). In particular we investigated a specific type of expropriation known as Tunneling : controlling families transferring resources from companies where they have few cash flow rights to ones where they have more cash flow rights. To explore the research topic, we investigated two research questions 1) to what degree tunneling prevails among the family firms, and 2) what are the main determinants of this phenomenon? We used general empirical technique for measuring tunneling developed by Bertrand, Mehta and Mullainathan (2002). With data provided by CCGR data base, we did a cross sectional analysis for Regarding the first question, the results suggest a significant tunneling between firms controlled by common family owner. Data showed that the family groups are on average about less sensitive to industry shock than stand alone. On the other hand, maximum cash flow right firms show on average higher sensitivity than the average firms in the same family groups. This suggests that, as tunneling predicts, some part of profitability in a family group is lost, and the lost profitability is more prevalent in the lower cash flow right firms in the group. The result also suggests that maximum cash flow right firms on average positively respond to the group shock as opposed to negative response for the rest of firms in the group. This confirms that the resources are transferred from the low cash flow right firms to the high cash flow firms. Regarding the second research question, the results suggest that the more tunneling prevails as the greater the cash flow rights between two firms diverge and the fewer shares the largest second shareholder holds. We however found little significance for insider positions in terms of family CEO and family chair. The results gives support for that a large cash flow wedge makes tunneling more attractive (incentive) and that less contesting by other shareholders makes tunneling more prominent (discretion), but also that informal control through insider positions (ability) seems restricted. Page 42

48 In our knowledge, our findings are with four major weaknesses due to data limitation and limited scope of the research. First we indirectly identify the family groups. Consequently we might have underestimated the real size of the family groups. Second we are considering tunneling through connected transactions between companies as legal entities and not transactions through between a company and an individual. This creates a bias in our findings because tunneling also can occur between a company and a sole proprietorship. Due to lack of data these transactions remain undetected and will lead to underestimation in our results. Lastly validity of our finding is highly dependent on how precisely a firm s industry is measured. If we are mismeasuring the firm s industry, this mismeasurment would lead firm to appear less sensitive to their industry shock(bertrand, Mehta, and Mullainathan 2002), particularly problematic for family group firms. Due to limited resources and data, we could not conduct the further investigation on this matter. Consequently there is possibility of this alternative explanation for our findings. Our findings also gives indication that tunneling occur, but not how tunneling occur. In chapter 2.4 different forms of tunneling were presented. Since we were unable to directly observe how tunneling occur in real life, large parts of how tunneling occur remains a dark spot. We leave these unsolved questions for the future researchers to investigate further. Page 43

49 Tunneling (GRA 1900 Master Thesis) APPENDIX Appendix 1: Tunneling Example Aker Solution Case (Bøhren 2009) Page 44