Trading-off Corporate Control and Personal Diversification through Capital Structure and Merger Activity

Transcription

1 Trading-off Corporate Control and Personal Diversification through Capital Structure and Merger Activity Martin Holmen Department of Economics Uppsala University Sweden John D. Knopf* University of Connecticut Stamford, CT U.S.A. Stefan Peterson Storebrand Investments 0114 Oslo Norway Abstract: In this study we use direct estimates of the portfolio diversification of the largest shareholder in a firm to study the impact of shareholder diversification on the firm. For firms where the controlling shareholder is an individual our tests indicate that the owner-managers use debt, dual class shares and corporate control transactions (merger activity) to strategically trade off corporate control and the drawback of poor portfolio diversification. However, for firms where the controlling shareholder is an institution, our results indicate that control but not diversification is important. Key words: diversification, controlling shareholders, mergers, capital structure, votes We are particular grateful to an anonymous referee and Peter F. Pope (the editor) for insightful and useful comments on earlier drafts. *Address for correspondence: John D. Knopf, University of Connecticut, One University Place, Stamford, CT 06901, USA. jknopf@business.uconn.edu

2 1. INTRODUCTION Researchers in corporate finance have focused considerable attention on the ways in which managerial self-interest affects managerial decisions. Jensen and Meckling (1976) point out that managers sometimes set debt below the level which is optimal for unaffiliated outside shareholders. This deviation from the optimal capital structure is primarily due to two important types of non-diversifiable risk in a firm. First, as discussed by Fama (1980), managers have substantial human capital investment in their firms. Second, managers typically have a large equity investment in their firms. Models of managers (controlling shareholders ) behavior frequently account for their exposure to idiosyncratic firm risk. 1 Friend and Lang (1988) find empirical evidence that managers reduce firm debt levels in order to mitigate these two sources of non-diversifiable risk. On the other hand, Stulz (1988) shows how managers, for entrenchment purposes, may use leverage to increase their voting power. Berger, Ofek, and Yermack (1997) and Jung, Kim and Stulz (1996) find empirical evidence that managers avoid leverage for entrenchment purposes. Just as capital structure decisions involve trade-offs between risk and control, so do merger decisions. Firms may make diversifying acquisitions in order to reduce shareholders idiosyncratic risk. Amihud and Lev (1981) and May (1985) find that as CEO ownership increases, firms are more likely to have risk reducing mergers. However, Aggarwal and Samwick (2003) show that managers implement mergers not only to reduce idiosyncratic risk, but also to derive private benefits of control (Jensen, 1986; Stulz, 1990). 2 1 see e.g. Fama, 1980; Fama and Jensen, 1983; Jensen, 1986; Stulz, Knopf, Nam and Thornton (2002) show that firm risk reduction through derivatives is related to managers stock and option holdings. 2

3 In addition, target company managers reduce their firm specific risk when either the acquirer is in another industry or the managers sell their shares and diversify their portfolio. However, if less diversified controlling shareholders place a higher value on control, they should be less likely to relinquish control (Grossman and Hart, 1988) and more likely to entrench themselves. This would lead to less diversified managers being less likely to be a target. Denis, Denis, and Sarin (1997) find evidence that higher levels of managerial ownership lead to greater managerial entrenchment, but they did not directly test the impact of managerial diversification on the likelihood of becoming a target. Previous studies were not able to isolate the marginal impact of diversification after controlling for ownership concentration. As a result, these studies typically use the market value of managerial equity as a proxy for diversification. There are two significant drawbacks to this proxy. First, in order for this to be an exact measure of relative diversification, all shareholders would have to have equal wealth. Second, as the value of equity increases, diversification may decrease, but simultaneously the shareholder becomes more aligned with other shareholders. This makes the interpretation of this proxy inherently ambiguous. In addition, earlier studies use the percentage of equity held by managers as a measure of alignment of interests between managers and shareholders. This variable is also ambiguous. The percentage of equity held is related not only to alignment of interests but also to control. The use of managerial equity holdings as a proxy for diversification, control and alignment of interests has persisted mainly because it is quite difficult to construct a more accurate proxy. There are two major obstacles to constructing a better proxy: First, there is a paucity of publicly available specific information about managerial shareholders personal portfolio holdings outside of the firm. Therefore, there is usually no way to estimate the actual 3

4 diversification of the controlling shareholder. Second, in the United States most companies only have one class of shares. Therefore, any change in equity ownership simultaneously increases control and alignment. This makes it problematic to distinguish one effect from the other. By examining and analyzing panel data collected for Swedish firms in 1988 and 1991, we extend the earlier work of the impact of shareholder diversification on capital structure and merger activity. We contribute to existing research along two dimensions: First, we use a less noisy proxy for diversification. By, using information about controlling shareholders actual portfolio holdings inside and outside the firm, we have constructed a more accurate measure of the actual diversification of the controlling shareholders. As we describe later in the paper, Swedish law is particularly advantageous for estimating individual portfolio holdings. Second, similar to Claessens, Djankov, Fan and Lang (2002), who examined East-Asian firms, we look at dual class voting shares, which are common in Sweden, to separately measure control (votes) and alignment of interest (equity) 3. 4 We believe this approach has allowed us to disentangle the impact of managerial non-diversifiable firm risk from managerial entrenchment motives on firm capital structure and merger activity. 5 Like Friend and Lang (1988), we distinguish between individually controlled (managerial controlled) and institutional controlled firms. For the remainder of this paper we use individual and manager interchangeably. We expect individual shareholders to be affected by both control and diversification interests. This is not the case with institutions. Even if the institutions holdings in the firm are not diversified, the owners of the institutions may be well diversified or 3 La Porta, Lopez-de-Silanes, and Shleifer (1999), and Faccio and Lang (2000) also examine the impact of the separation of votes and equity on managerial incentives. 4 Swedish firms issue two types of shares (A and B) with equal cash flow rights but different voting rights. Typically, A shares are one-share-one-vote while B-shares carry 1/10 vote per share. 5 Managers may also entrench themselves through voting control over ESOPs (see Chang and Mayers (1992)) and non-beneficial holdings (see Farinha (2003)). 4

5 the managers of the institution may not have a personal equity stake at risk. Thus we expect institutions to be concerned primarily with control rather than with diversification. Furthermore, although managerial controlled firms may derive more private benefits of control than their institutional counterparts, institutions also have the opportunity to obtain private benefits. For example, managers may grant themselves excessive perquisites and compensation. Potential private benefits of control available to both individually and institutionally controlled firms include directing capital and corporate resources to outside interests, accruing the power and prestige associated with control, and tunneling funds from one firm to another within a pyramid (Johnson, La Porta, Lopez-de-Silanes, and Shleifer (2000) and Bertrand, Mehta, and Mullanaithan (2002)). In Sweden, however, pecuniary private benefits of control usually appear quite small compared with those in most other countries (see Dyck and Zingales (2004) and Holmen and Knopf (2004)). In our capital structure tests, we find that less diversified institutionally controlled firms set higher debt levels, despite the increased risk associated with more leverage,. For managerial firms, however, the marginal impact of diversification on debt is positive. Furthermore, managerial shareholders maintain control through a combination of high vote shares and high debt levels. Therefore, they are more likely to have relatively less debt than institutional shareholders when they already have high vote shares. In other words, there is a substitution effect for managers that we do not observe for institutions. We interpret this as evidence that both managerial and institutional firms are concerned about control. However, managerial shareholders are more concerned than institutional shareholders about firm specific risk. We also perform two sets of tests on the influence of shareholder diversification on control transactions (merger activity). First we look at the probability of making diversifying 5

6 acquisitions. Second, we examine the probability of becoming a takeover target. If managers implement mergers to reduce idiosyncratic risk (Amihud and Lev, 1983; May, 1995), we expect a negative relationship between controlling shareholder diversification and the probability of a diversifying acquisition. On the other hand, if acquisitions are not implemented to reduce idiosyncratic risk but rather for some other reason unrelated to risk, we would not anticipate any relationship between shareholder diversification and the degree of diversification of a merger. We do not find any significant relationship between shareholder diversification per se and the probability that the firm will make a diversifying acquisition. However, we find weak evidence that the joint effect of a controlling shareholder with a poorly diversified portfolio and a firm with high leverage increases the probability that a firm makes a diversifying acquisition. Finally, we examine whether controlling shareholder diversification is related to the probability of becoming a target. We find that less diversified controlling institutional shareholders are significantly less likely to have their firms taken over. This provides additional evidence that institutional firms are primarily concerned with control. However, for managerial firms, the marginal impact of diversification on the probability of becoming a target is positive. Once again, this is evidence that managerial firms are pulled in opposite directions between the loss of control resulting from becoming a target and the potential diversification benefits of being taken over. In section II we describe our sample and variables. In section III we provide our empirical results. Finally, in section IV we give our conclusion. (i) Sample selection 2. SAMPLE SELECTION AND VARIABLE SPECIFICATION 6

7 We begin with all firms listed on the Stockholm Stock Exchange (A-list, OTC, or unofficial list) for the years 1988 and This includes, with few exceptions, the largest corporations in Sweden. Balance sheet and income statement data are provided by FINDATA. Shareholder data are obtained from Sundqvist (1988, 1991), who reports the major shareholders for all listed firms, together with detailed information on coalition structures as of January each year. A coalition includes voting rights owned by family members and family-controlled firms 7 In the beginning of 1988 (1991), there were 257 (220) firms listed on the Stockholm Stock Exchange, of which 119 (101) were controlled by individuals. The list of individual shareholders who are the largest vote-holders in a listed firm is then compared to the list of the wealthiest Swedes published by Affärsvärlden (1988 and 1991) 8. Affärsvärlden reports the richest Swedish individuals and families with a net wealth of at least 100 million SEK (approximately 15 million USD at the USD/SEK exchange rates in 1988 and 1991). Therefore, firms whose largest shareholder is an individual with wealth of less than 100 million SEK are excluded from our sample. Furthermore, financial firms are excluded. Finally, some firms are excluded from the sample due to missing data. Our final sample consists of 112 individually controlled firms (65 in 1988 and 47 in 1991) and 120 institutionally controlled firms (63 in 1988 and 57 in 1991). The 232 firm year observations are for 157 different firms. Thus, we have an unbalanced panel with two observations for 75 firms and one observation for 82 firms. Our sample comprises 78 percent (71 percent) of the stock market capitalization in 1988 (1991). 6 We limit our study to these two years since these were the two last occasions our source (Affärsvärlden) for market based estimates of the largest shareholders wealth was published. After 1991 it is only possible to collect book values of individual wealth. 7 Bergström and Rydqvist (1990) used a similar principle for grouping shareholders into coalitions. 8 The Affärsvärlden report for wealthy Swedes is equivalent to the Forbes Magazine report for wealthy Americans. However, because privacy laws are not as strict in Sweden, more information about individual wealth is part of the public domain. For example, personal tax forms which include taxable wealth were public information in 1988 and

8 (ii) Ownership Variables When the largest shareholder is a manager, he has non-diversifiable human capital invested in the firm, in addition to his equity holdings. This gives the owner-manager an added incentive to reduce firm risk below the desired level of other unaffiliated shareholders. An institution is typically less concerned about the diversification of its portfolio. An institution usually does not have a direct human capital investment in the firm. Furthermore, even when the institution itself is not diversified, the individuals who control the institution may be diversified. For this reason, we distinguish managerial from institutional control, for many of our tests. For a firm to be classified as individually controlled it must meet two criteria: First, there must be a traceable individual or family who directly or indirectly ultimately controls the largest voting block of shares. Second, the individual or family must also personally own equity that can benefit from the control so that the individual s portfolio diversification is directly affected by the firm. 9 The individuals may own these shares directly or through a corporations or institution. For example, if a family controls an institution that controls Corporation A which in turn controls Corporation B, the family would be classified as controlling Corporation B. All other types of largest shareholders are classified as institutional, including for example, non-profit foundations controlled by families. Although family controlled non-profit foundations which control the largest voting block of shares meet the first criterion, they do not meet the second because they do not personally own the equity and therefore have no access to the capital in the foundation. Also included in the institutions groups are state or community 9 Most of our individual owners are insiders (104 out of 112 observations). For the eight observations where the largest individual shareholder is not an insider, for two observations we have found a clear family representation by husband or wife. In two more observations we have found a clear group representation. That leaves four 8

9 governments, associations (e.g. unions), and public corporations (without an individual in ultimate control). 10 For differentiating managerial (individual) and institutional control, we use the following dummy variable, MANAGER 1,if the largest shareholder is anindividual 0,if the largest shareholder is an institutio n Table 1 divides the sample according to ownership categories. The sample is fairly evenly split between manager-controlled (48.3 percent) and institution-controlled firms (51.7 percent). The founder or his family controls about half of the managerial firms (56 out of 112). 11 Using the market value of a manager's equity as a proxy for his diversification, Friend and Lang showed that as the market value of a manager's equity increases, firm debt decreases. Their proxy was only indirectly correlated with diversification. In contrast, for our proxy for diversification, we use direct estimates of the actual portfolio holdings of the managers of firms in our sample. For each firm in our sample, we estimate the diversification of the largest shareholder. We use different estimation procedures for individual and institutional shareholders. For individuals, we collect wealth data from Affärsvärlden (1988, 1991). In order to approximate the net wealth of an individual, Affärsvärlden has carried out interviews and exploited various official data sources such as: annual reports, real estate registers, tax authority records and various commercial data bases. Due to the principle of public access to official records observations where we can t find an obvious inside connection although there might be one. We have rerun all of our tests without those four observations and the results are essentially unchanged. 10 In Sweden, there are no legal restrictions to family control of financial institutions and, at least in 1988 and 1991, families often controlled financial institution. Today, few financial institutions are controlled by families. 11 Although we haven t reported the results, we didn t find any significant explanatory power by distinguishing among the various types of individual owners. 9

10 (Offentlighetsprincipen), the journalists at Affärsvärlden have access to all papers and files that arrive into an agency or are finished by a civil municipal servant. 12 We are unaware of any other country, with the exception of Finland, 13 where it is possible to construct such accurate estimates of individual wealth. Annual reports were used to find the book value of private companies. Due to the offentlighetsprincipen, manager-held companies annual reports are publicly available. Private companies are subject to the same accounting standards as public firms. 14 Private companies were then given valuations similar to public companies based upon size and line of business. Real estate values were estimated by using recent valuations completed by independent appraisers or by approximating the value by the amount of assessments and rental revenues. Due to the offentlighetsprincipen, all Swedes tax returns are publicly available. Additionally, since Sweden levies wealth taxes, taxable wealth is also publicly available The offentlighetsprincipen has been part of the Swedish constitution since Although it has been amended the basic principles have never been changed. It states that all official records collected by the government must be handled in the following manner (the word paper stands for information, on paper or electronic, and the word agencies includes courts): First, a paper arrives into an agency, or a paper is finished by a civil or municipal servant. Second, this constitutes the paper a common public paper, and as such, it is irrevocably archived for eternity (with exceptions stated in a separate law). Third, the paper s existence is registered. If some part of it is classified (e.g. for national security reasons), that is flagged in the register. Fourth, anyone may, anonymously and without giving any reason, immediately read the paper without any cost, and get copies against a fee without undue delay. Fifth, the offentighetsprincipen is part of the right to print and distribute daily papers; the constitution s extremely clear wording allows a publisher to get a copy of a public paper and print it. No one, not even the government or parliament or the original author, can stop that printing (Johannison, 1981). With modern data processing techniques the information has become readily available to the general public. 13 Finland has similar laws regarding access to public information. 14 Listed firms are subject to tougher disclosure rules than non-listed firms. 15 The general principle of the wealth tax is that all wealth is taxable even foreign assets are taxed. This means that all stocks, bonds, bank-deposits, cash, cars, boats, machines, animals, and real estate are taxable. A special taxable value is assigned to all Swedish real estate. It should represent 75 percent of the market value with two years lag, i.e. taxable value 2004 should represent 75 percent of the estimated market value in In 1988 and 1991 listed stocks (in Sweden or abroad) were valued at 75% of market value. OTC traded stocks were valued at 30 percent of the market value. Non traded stocks, private firms and partnerships were valued at book values. Some assets are however, not taxable. Insurance other than life insurance is not taxable. Other examples of non-taxable assets are art and coin collections (if they are not part of a business inventory). Furthermore, furniture and household utensils are not taxable. Most debt is tax-deductible, i.e. the wealth tax is levied on net wealth. In 1988 (1991), net wealth below 400,000 SEK (800,000 SEK) was not taxable (Bratt et al., 1987; Rabe, 1991). 10

11 Affärsvärlden used taxable wealth to refine their estimates of the market value of total wealth. 16 Finally, Affärsvärlden carried out interviews to check the reliability of their estimates. Affärsvärlden only reports the total wealth estimate, not the components of the sum. Affärsvärlden approximated the indebtedness of the large shareholders by basing the loan values on the length of the time that the stock-holdings were in the coalition's possession. If the founder was still the major shareholder, then he or she is regarded as being debtless. Taxable income and wealth were used to refine indebtedness estimates. For example, if taxable wealth was substantially below the taxable value of an individual s known assets, the difference was attributed to debt. Also, if taxable income was below known income from dividends and salary, the difference was assumed to be interest payments. Debt can then be estimated given an assumed interest rate. Affärsvärlden aggregates the wealth of family members. 17 Thus, our wealth estimates generally apply to a family, not to a single individual. Our source for equity ownership (Sundqvist, 1988 and 1991) also aggregates family ownership. 18 In 76 observations in our sample of manager controlled firms, the largest shareholder is defined as a family. Thus, our measure of portfolio diversification for managers generally applies to families, not to simple individuals. For individually controlled firms, we calculate diversification as, 16 The wealth tax creates incentives to hide wealth in offshore accounts. However, hiding wealth in Sweden is illegal and studies have shown that Sweden has a very high rate of tax compliance (La Porta, Lopez-De-Silanes, and Shleifer (1999) and Dyck and Zingales (2004)). 17 Affärsvärlden s wealth estimates do not include family controlled foundations. For example, Peter Wallenberg was reported as the 64 th (68 th ) wealthiest Swede in 1988 (1991) even though the Wallenberg group was the largest shareholder on the Stockholm Stock Exchange. 18 We do not know whether each family has been defined the same way by Affärsvärlden and Sundqvist. 11

12 marketvalueof individual ' s equity in the firm DIV net wealth of individual Where the market value of individual s equity in the firm is adjusted for pyramid structures. For institutionally controlled firms we approximate wealth by summing the market value of all of the holdings of the institutions in firms listed on the Stockholm Stock Exchange. Admittedly, in some cases, this underestimates the portfolio holdings of institutions. Furthermore, we do not account for the correlation between assets. Nonetheless, we believe that this proxy is superior to previously used measures of diversification. For institutions, we calculate diversification as, market value of institutio n' s equity in firm DIV market value of all of institutio n' s listed shares In Table 2 panel A, we report the fraction of the controlling owner's wealth invested in the firm (DIV). It is (0.519) on average (median) for managerially controlled firms and (0.045) for institutionally controlled firms. The median managerial owner has 50 percent of his wealth invested in the firm. The median institutional owner has less than 5 percent of its wealth invested in the firm. For managerial firms, DIV ranges from a minimum of to a maximum of A DIV of 1.47 implies that the value of the individual's shares in the firm is almost 50 percent larger than his wealth. Hence, the individual has leveraged himself in order to purchase some of his shares. Thus, managerial owners appear willing to hold highly leveraged and poorly diversified personal portfolios in order to maintain control. Our other ownership variables are described below. In Sweden as opposed to the U.S., most companies have A and B shares, where A shares have significantly higher voting power per share than B shares, but are entitled to the same cash flow. For voting control of the ultimate largest shareholder, largest shareholder' s number of VOTE total votes outstanding votes 12

13 In case of a pyramid structure, we consider the individual s controlling interest in the larger firm. For example, an individual owns a controlling interest of 51% in firm X, and Firm X owns 10% of firm Y, we consider the individual to control 10% of the vote in Firm Y. For the equity ownership of the largest shareholder in the firm, UNADJUSTED _EQUITY market value of firm equity of largest shareholder total market value of firm equity For the equity ownership of the largest ultimate shareholder adjusted for pyramid structures (i.e. for firm Y which is controlled by firm X this means the ultimate owner s equity fraction in firm X times firm X s equity ownership in firm Y), market value of firm equity of EQUITY total market value of ultimate shareholder firm equity For measuring the separation of ownership from control, VOTE UNADJUSTED _ VOTE _ TO_ EQUITY UNADJUSTED _ EQUITY For the separation of ownership and control adjusting for pyramid structures, VOTE VOTE _ TO_ EQUITY. EQUITY The ability of a shareholder to act in his own interest is influenced not only by the percentage of votes that he controls, but also the concentration of ownership of other voting shares. An outside blockholder typically is not able to extract private benefits of control and has less non-diversifiable equity invested in the firm than the largest shareholder. Consequently, he 13

14 may serve as a monitor of the largest shareholder for himself and other shareholders. For outside block holders, BLOCK10 1, when there exists an outside 10 percent voting block, 0, otherwise Table 2 panel A provides some summary statistics for ownership characteristics of our sample. The mean VOTE for the entire sample is 53.6 percent, while the mean UNADJUSTED_EQUITY is only 38.3 percent. In our sample, the largest shareholders control the firm by using high vote shares despite the fact that they own significantly less than half the equity. If UNAJUSTED_EQUITY is adjusted for pyramid structures and defined as ultimate owner s net investment in the firm (EQUITY), the difference is even more pronounced. On average, EQUITY is 28.3 percent. Pyramid structures are more common among institutionally controlled firms. Panel C shows that 62 percent (32 percent) of the institutionally (manager) controlled firms are part of a pyramid structure. The vote-to equity ratio (VOTE_TO_EQUITY) for the median ultimate owner is The difference between managerial and institutional firms VOTE_TO_EQUITY is driven by the higher frequency of pyramid structures among the institutional firms, not use of dual class shares. Note that the median UNADJUSTED_VOTE_TO_EQUITY ratios are not significantly different. The mean difference of UNADJUSTED_VOTE_TO_EQUITY is driven by outliers. In our sample, eight of the institutionally and one of the individually controlled companies maintained control with high vote shares that had 1000 times more voting rights than the low vote shares. 19 In Sweden, this construction is not allowed today, but old firms are allowed to keep their old 19 All our tests below have been rerun without these nine observations (unreported). It does not change the results. 14

15 corporate charters. The limit today is 1/10 vote per share for lower vote shares. As the last listed firm on the Stockholm Stock Exchange, Ericsson abolished a 1/1000 vote differential in Panel C shows, that the vast majority of managerial controlled firms (93 percent) and a substantial majority of institutionally controlled firms (77 percent) have dual-class shares. Additionally, almost half (47.8 percent) of the firms are part of a pyramid structure. 20 The combined effect of dual-class shares and pyramids (VOTE_TO_EQUITY) is stronger for institutionally controlled firms. (iii) Firm Variables To measure the use of debt in a firm s capital structure, we compute the following ratio, LEVMARKET book value of debt market valueof equity book valueof debt We also measure the leverage ratio using book value of equity as follows, LEVBOOK book value of debt book valueof equity book valueof debt We define a number of other firm characteristic variables: For size: 21 LSIZE log( book value total assets) For operating (accounting) performance: earnings before interest and taxes ROA total assets 20 This is consistent with La Porta, et al. (1999) examination of 27 of the world s richest counties. Sweden ranks first in the use of dual-class shares and second after Belgium in the frequency of pyramids. 21 Total Assets are deflated into 1991 prices by the consumer price index. 15

16 For dividends paid: total dividends paid PAYOUT accounting earnings If dividends are paid when accounting earnings are negative, PAYOUT is set to 1. For asset collateral: net plant, property and equipment RPPEAP book valuetotal assets For equity risk: VOLATILITY yearly standard deviation of daily share price changes For diversification of a firm s operations, SINGLESEG 1, when the firm is active in only oneindustry 0, otherwise Table 2, panel B, shows that the managerial controlled firms have significantly more debt than institutionally held firms, based on the book value of total assets. Also, managerial controlled firms are significantly smaller and younger. VOLATILITY is significantly larger for managerial controlled firms while PAYOUT is significantly lower for managerial controlled firms. This appears to be inconsistent with Farinha (2003), who finds that entrenched managers pay higher dividends to overcome agency problems. Furthermore, although Grinstein and Michaely (2005) find that institutions avoid firms that don t pay dividends, of the dividend paying firms they are more likely to invest in the ones that pay lower dividends. The differences between ROA, RPPEAP, and SINGLESEG (panel C) for managerially and institutionally controlled firms are statistically insignificant. 16

17 (iv) Merger Activity (Corporate Control Transactions) Variables We have two merger activity variables. Our first variable is constructed to distinguish whether firms became target companies during our sample period. To reduce endogeneity problems, we used the three years after we observe the controlling owner s wealth, i.e for the 1988 observations and for the 1991 observations. We define the following dummy variable: TARGET = 1, if firm was a target during sample period, =0, otherwise. Using the same time frame as for our TARGET dummy variable, we have also constructed a variable for firms that have made acquisitions during our sampling period... This variable is related to the extent of diversification of a merger. For this we have a dummy variable: DIVERSIFYACQUISITION = 1, if acquired firm has different 2-digit industry code, = 0, otherwise. Panel C indicates that there are no significant differences in the corporate control variables between managerial controlled and institutionally controlled firms. 3. EMPIRICAL RESULTS In this section we shall test how controlling shareholders use leverage, dual-class shares and mergers to tradeoff the benefits of control and the drawback of poorly diversified portfolios. We also test whether the identity of the controlling shareholder, individual or institution, influences the trade-off between control and poor diversification. (i) Leverage 17

18 In this series of tests, we will perform regression analysis to investigate how the largest shareholder s portfolio diversification affects leverage. Ceteris paribus, as shareholders increase their investment in a firm decreasing their diversification they prefer the firm to take less risk through the reduction of leverage. In Table 3, we first present ordinary least squares regression results for various specifications, using LEVMARKET as the dependent variable. To limit endogeneity problems, all independent variables are from beginning of the year, while the dependent variable is from year s end. We draw cross-sectional data from two separate year s observations (1988 and 1991) of partly overlapping firms. The 232 firm-year observations are related to 157 different firms. Pooling firm-year observations treats each observation as independent, which underestimates standard errors and overstate reported t-values when firm values are correlated from year to year. The Huber-White Sandwich estimator for variance used in all OLS regressions and logit regressions below relaxes the assumption of independence of the observations. As an alternative, using the fact that two observations are drawn from the same firm, we report results for fixedeffects panel regression procedures in panel B. From the results of specification M1 in Table 3, it appears that DIV has no significant influence upon the LEVMARKET. However, by comparing model M1 to M2-M4, consistent with Berger, Ofek and Yermack (1997), Friend and Lang (1988), Jung, Kim and Stulz (1996), and Mehran (1992), it is evident that accounting for whether a firm is managerial controlled contributes to the significance of DIV on firm capital structure. In M2-M4 where we include MANAGER*DIV to control for the marginal effect of managerial control on diversification, DIV is positive and significant. This means that less 18

19 diversified institutional owners have more debt. This result supports the Stulz (1988) argument that institutional owners use leverage to entrench themselves. However, the negative sign for MANAGER*DIV in M2-M4 is negative and significant. To determine the overall impact of DIV on leverage for managerial controlled firms, we must add the regression coefficients of M4 for DIV and MANAGER*DIV, which are and , respectively. This sum, , is negative but not significant. This indicates that, compared to institutions, as managerial owners invest more of their wealth in the firm, they decrease firm leverage. This supports the hypothesis that managerial owners, when making the leverage decision, trade-off the risks to their equity holdings and firm-specific human capital and the benefits of managerial entrenchment. (ii) Differential Voting Rights As in other studies, we find a positive relationship between EQUITY and a firm s debt for our OLS regressions reported in Table 3. This supports the hypothesis that, as the percentage of equity ownership increases, a shareholder s interests become more aligned with other shareholders, so that more debt is assumed to increase the firm s value. However, common shares have cash flow and voting rights; therefore, this result also supports the Stulz (1988) argument that increased levels of debt are used to increase voting control. There is not any significant relationship between EQUITY and firm debt for the within (fixed) effects regression results reported in Panel B. Although theoretically it makes sense to use a fixed effects model, there is a practical problem when examining insider ownership: As Zhou (2001) points out, typically from year to year, there is too little within-firm variation of insider ownership for a fixed effect procedure to find a significant relationship, even if one exists. 19

20 To give additional insight into these alternative hypotheses, we shall now turn to the issue of the separation of cash flow rights and voting rights. When a manager decides to increase firm debt in order to concentrate voting control, he simultaneously increases the risk of his nondiversifiable equity through the increase in leverage. Dual-class voting shares give a manager an option that allows him to maintain control without the negative consequences of leverage. A high ratio between vote and equity ownership enables a shareholder to control the firm s actions while only assuming part of the economic consequences of these actions. Thus, an investor whose primary concern is control will prefer a high vote-to-equity ratio and a high debt ratio. This will minimize the amount of capital needed to control the firm. In contrast, an investor who has a long-term interest in the firm and wants to minimize the risk of his portfolio can issue dual-class shares and keep the high-vote shares. This will only marginally dilute his control but will decrease his risk. We expect a positive relation between the controlling shareholders vote-toequity ratio and the firm s debt for firms controlled by well-diversified institutions. However, we expect a negative relation between vote-to-equity ratio and debt for firms controlled by poorly diversified individuals. In M3 and M4 we introduce the influence of the separation of votes from equity. For M3 and M4, VOTE_TO_EQUITY is positive and highly significant for the entire sample. The positive sign indicates that, for the entire sample, controlling shareholders minimize the amount of capital needed to control the firm by using a high vote-to-equity ratio and a high debt ratio. However, as with DIV, distinguishing between managerial and institutional controlled firms gives additional insights. MANAGER*VOTE_TO_EQUITY is negative at the 10 percent level. The negative relationship indicates that institutional owners use more debt than poorly diversified managerial owners, who issue dual-class shares and keep the high voting stock for 20

21 themselves. Substituting low-vote shares for debt allows the individual owners to only marginally dilute their control and simultaneously decrease risk. Also, it is interesting to note in regressions M3 and M4 that MANAGER becomes positive and significant at the 5% level. This indicates that manager-controlled firms, after accounting for diversification and voting control, actually have more debt than institutions. This is not surprising, since individuals have limited wealth. These individuals are more likely than institutions to have to use debt to maintain control, controlling for their use of high-vote shares and concerns about diversification. M4 brings in the influence of outside shareholders. Consistent with other researchers results, there is a positive and significant relationship between BLOCK10 and LEVMARKET. For all of the regressions, we include the control variables ROA, LSIZE, RPPEAP, SINGLESEG, year and industry dummies. These are similar to the control variables used by Farinha (2003), Friend and Lang (1988), Titman and Wessel (1988) and Berger, et al. (1997). We find a negative relationship between SINGLESEG and LEVMARKET. Not unexpectedly, this implies that as a firm s operations are more diversified, the firm has more debt. We also find a positive relationship between LSIZE and LEVMARKET, although the empirical results from other studies are inconsistent. 22 The negative and significant relationship between LEVMARKET and ROA is consistent with other studies; so is the (insignificant) positive relationship that we find between LEVMARKET and RPPEAP. 23 Farinha (2003) suggests that firms, with entrenched managers who use directly held and non-beneficial shares, may pay higher dividends to compensate outside shareholders for the 22 For example, Friend and Lang (1988) found a positive relationship and Gupta (1969) found a negative relationship between the debt ratio and LSIZE. 21

22 agency costs associated with entrenched management. Therefore, since the dividend decision may be intertwined with the leverage decision, we have included dividend payout (PAYOUT) in our regressions. However, we don t find any significant relationship between PAYOUT and leverage. Although we don t fully report the results, we also performed regressions with the causality reversed. That is, PAYOUT is used as the dependent variable. With this reversal, we did not find any significant relationship between DIV and PAYOUT. We check the robustness of our results by running panel data regressions. The Hausman test rejects the assumption that the firm-specific effects are uncorrelated with the regressors. We therefore estimate the panel regressions using fixed effects. Fixed effect estimates adjust for the possibility that unobservable firm-specific factors influence the level of leverage in each individual firm and are equivalent to estimating OLS models and including an indicator variable for each firm. In Table 3, panel B, we show that our diversification results hold for fixed effect estimation. The levels of significance are lower, however. DIV, MANAGER, and MANAGER*DIV drop to 10 percent significance (from one percent in OLS). For the other ownership variables, the fixed effect estimates are statistically insignificant. As we have already discussed, this is not surprising, since the ownership variables generally vary little within firms from 1988 to 1991 (see Zhou, 2001). Although we believe that using market values is the theoretically correct method, it is useful to examine whether our results hold using the book value of equity to compute our 23 Kester (1986), Baskin (1989), and Friend and Lang (1988) found a negative relationship between the debt ratio and ROA as we found in this study, and Friend and Lang (1988) found a positive relationship between RPPEAB and debt ratio as we found in this study. 22

23 leverage ratio. 24 As shown in Table 3, Panel B, our OLS results for book and market measures of leverage are virtually the same, but there are some differences for the fixed effect estimates. MANAGER and MANAGER*DIV go from the 10 percent level of significance for market value leverages to the 5 percent level of significance for the book value estimates, while DIV goes from the 10 percent level of significance to insignificance. 25 VOTE_TO_EQUITY goes from insignificance to the 10 percent level of significance in the book value regression. (iii) Diversifying Acquisitions Controlling shareholders may also reduce their personal exposure to the idiosyncratic risk of the firm by making diversifying acquisitions. If controlling shareholders use acquisitions to reduce idiosyncratic risk, less diversified controlling shareholders should be likely to execute more diversifying mergers. In Table 4, we report the results for our tests of the relationship between the diversification of the controlling shareholder and the extent of diversification provided by an acquisition. The dependent variable is an indicator of the firm making a diversifying acquisition on the Stockholm Stock Exchange during the three years after we observe the controlling owner s total wealth. 26 In Model 1, we do not find any significant evidence that diversification plays a role in acquisition policy. In Model 2, where shareholder diversification is interacted with firm leverage, shareholder diversification is negative and significant at the 10 percent level. This 24 We have also defined leverage in terms of long term debt only (unreported) as suggested by, e.g., Rajan and Zingales (1995). Our results are robust to this alternative specification of leverage. 25 Differences between market value and book value results could be due to accounting rules and standards. However, Sweden has been shown to rank number 1 in the world in terms of reliability of accounting rules and company disclosures by La Porta et al (1998). Also, during our study period, the purchase method of accounting for acquisitions was standard with Goodwill written off over a maximum of 10 years. Furthermore, the use of the pooling method (no goodwill resulting in lower book value of equity) was severely limited. Finally, Berger, Ofek and Yermack (1996) also found slightly weaker results using book values. Therefore, we do not believe that differences in book value and market value results are due to any deficiencies in Swedish accounting methods. 23

24 is contrary to existing theory and empirical evidence (Amihud and Lev, 1983; Aggarwal and Samwick, 2003; May, 1995). However, the interaction term DIV*LEVMARKET is positively significant at the 10 percent level. When the controlling shareholder holds a poorly diversified portfolio and the firm is highly levered, there is increased probability that the controlling shareholder will merge the firm with a firm in another industry to reduce the idiosyncratic risk of his portfolio. 27 Firm size significantly increases the probability of a diversifying acquisition. All other control variables are insignificant. 28 (iv) Relinquishing Control When a firm is taken over, the current controlling shareholders yield control to the acquiring firm. However, the controlling shareholders portfolio becomes more diversified. Therefore, if the controlling shareholders primary concern is diversification, we expect that if controlling shareholders are less diversified, they are more likely to be taken over. However, if the controlling shareholders are most interested in control, and hold less diversified personal portfolios to maintain control, we expect the opposite relationship (Grossman and Hart, 1988). In Table 5, we report results for logistic regressions where the dependent variable is an indicator of whether the firm was taken over during the three years after we observe the controlling owner s total wealth. In model M1, DIV bears no significant relationship with the probability of the company being taken over. 26 The results in Table 4 should be interpreted with caution, since there are only 17 diversifying acquisitions. 27 The interaction terms between managerial control and portfolio diversification and the controlling owner s vote-toequity ratio, respectively, show no explanatory power in explaining the probability of a diversifying acquisition (unreported) 28 The BLOCK10 dummy is not included since it perfectly predicts that no diversifying acquisition will occur, i.e. there is no observation where the BLOCK10 dummy is equal to one and the firm made a diversifying acquisition. 24

25 The VOTE_TO_EQUITY coefficient is negative and highly significant in M1. As we stated before, a high VOTE_TO_EQUITY coefficient is associated with managerial entrenchment. With high VOTE_TO_EQUITY, managers can control firms with relatively less equity. Therefore, consistent with Denis, et al (1997), we find that firms with more entrenched managers are less likely to become target companies. Once again, our results are sensitive to whether an individual or institution controls the firm. In M2, when we include the cross terms MANAGER*DIV and MANAGER*VOTE_TO_EQUITY, DIV becomes negative and significant at the five percent level. We interpret this as support for our suggestion that less diversified owners are less likely to give up control through a takeover. The MANAGER*DIV interaction term is positive and significant at the 5 percent level, indicating that this effect is weaker for managerial controlled firms (the sum of the coefficients for DIV and MANAGER*DIV is positive). This gives additional support to our suggestion that firms controlled by individuals trade-off diversification and control, while institutions are primarily concerned about control. We have included a number of control variables in this regression. When there is a large outside blockholder (BLOCK10), the firm is more likely to be taken over. This is not unexpected, given the large premiums associated with being a target and that outside blockholders do not receive private benefits of control. Many of the takeovers are also made by the outside blockholder. We find no significant relation between LEVMARKET and the probability of the firm being taken over. Interaction terms between leverage and DIV, MANAGER, and VOTE_TO_EQUITY, respectively, are also insignificant (not reported in tables). Our other control variables LSIZE, ROA, and SINGLSEG are also insignificant. 25

26 However, EQUITY is negative and significant at the 1 percent level. Large equity holders are less likely to relinquish control, despite the high premiums paid to target shareholders. 29 (v) Robustness Tests: Alternative sample and proxies for total wealth A central premise of this paper is that we use a better proxy for the personal diversification of managers than previous researchers, because we have a more accurate measure of managerial wealth. As we described in our data description section, by taking advantage of Swedish public information access laws and personal interviews, Affärsvärlden (1988, 1991) collected extensive data about managers personal portfolios. Unfortunately, Affärsvärlden has not published this data since This raises two robustness issues. First, is the Affärsvärlden data superior to other currently available wealth estimates? Second, do the relationships found using a sample from 1988 and 1991 still hold today? In order to address these questions, once again exploiting Swedish public information laws, we have collected data for two additional individual wealth proxies for 1991 and When calculating these proxies for 1991, we use the same sample of firms as for our previous tests. For the 2002 sample, we start with all non-financial firms listed on the Stockholm stock exchange in January Then we identify all firms with an individual in control (largest 29 Although we do not provide tables of the results, for a robustness test we defined the dependent variable in tables 4 and 5 (conglomerate merger and control change, respectively) over 5 and 10 year periods. Although the signs for all the key variables remain the same, the results become progressively weaker as we expand the number of years. Unfortunately, in Sweden compared to countries such as the United Kingdom and United States, The available sample size for mergers in a particular year is much smaller. Inevitably, by extending the time period for mergers, we must tradeoff the advantages gained by a larger sample and loss of power by increasing the time-lag between the variables. Therefore, our results although supportive, should be viewed cautiously. Specifically, for our sample, apparently any gains from an increased sample period are offset by too much of a time lag between the observation of the dependent and independent variables. 30 We choose 2002 such that we would be able to replicate the tests in table 4 and 5, i.e. we need three years of data after the year for which we collect wealth proxies. 26