Foreign Investors and Dual Class Shares

Foreign Investors and Dual Class Shares MARTIN HOLMÉN Centre for Finance, University of Gothenburg, Box 640, 405 30 Gothenburg, Sweden First Draft: February 7, 2011 Abstract In this paper we investigate how the lifted ban on foreign ownership on the Stockholm Stock Exchange in 1993 affected the relative valuation of dual class firms. Our sample period is 1985 to 2005. We find that the discount on dual class firms is a post 1993 phenomena. There are no indications that dual class firms were valued at a relative discount before 1993. We find no indications that the post 1993 discount is driven by deteriorating operating performance of dual class firms or a more aggressive use of dual class shares. Furthermore, we are not aware of any evidence suggesting that controlling shareholders in Sweden started to expropriate minority shareholders after 1993. In fact, our proxy for the pecuniary private benefits of control, the voting premium on the high voting A shares, decreased after 1993. Our interpretation of the results is that foreign investors put a discount of dual class firms independent of actual expropriation of minority shareholders by controlling shareholders. Keywords: dual class shares, foreign investors, discounts, voting rights, cash flow rights JEL Classification: G32 The author gratefully acknowledges financial support from VINNOVA (grant 2010-02449) * Corresponding author: Martin.Holmen@cff.gu.se ; Phone: +46 31 786 6442

Foreign Investors and Dual Class Shares Abstract In this paper we investigate how the lifted ban on foreign ownership on the Stockholm Stock Exchange in 1993 affected the relative valuation of dual class firms. Our sample period is 1985 to 2005. We find that the discount on dual class firms is a post 1993 phenomena. There are no indications that dual class firms were valued at a relative discount before 1993. We find no indications that the post 1993 discount is driven by deteriorating operating performance of dual class firms or a more aggressive use of dual class shares. Furthermore, we are not aware of any evidence suggesting that controlling shareholders in Sweden started to expropriate minority shareholders after 1993. In fact, our proxy for the pecuniary private benefits of control, the voting premium on the high voting A shares, decreased after 1993. Our interpretation of the results is that foreign investors put a discount of dual class firms independent of actual expropriation of minority shareholders by controlling shareholders. Keywords: dual class shares, foreign investors, discounts, voting rights, cash flow rights JEL Classification: G32

1. Introduction Before 1993 most Swedish firms had two classes of equity, restricted and unrestricted shares. Most Swedish firms also have dual class shares where the high voting A-shares typically are one share one vote, while the low voting B-shares typically carry 1/10 of a vote. A- and B- shares have equal cash flow rights. Thus, before 1993 many firms had four different shares, restricted and unrestricted A shares and restricted and unrestricted B shares. Only unrestricted shares could be held by foreigners. Furthermore, the proportion of unrestricted shares was limited to 20 percent of the voting rights and 40 percent of the cash flow rights. These restrictions were abolished in January 1993. Foreign ownership on the Stockholm Stock Exchange increased from roughly eight percent in 1990 to almost 40 percent in 1999. This increase in foreign ownership meant that the marginal investor was much more likely to be a foreigner after 1993 compared to the situation before 1993. In this paper we investigate how the lifted ban on foreign ownership on the Stockholm Stock Exchange in 1993 affected i) the relative valuation of dual class firms and ii) the relative valuation of high voting and low voting shares. Our sample period is 1985 to 2005. Dahlquist and Robertsson (2001) investigate foreign investors portfolios of Swedish firms 1993-1997. They document that foreign investors prefer large firm, nondividend paying firms, and firms with large cash positions, liquid stock, and international operations. They also document that foreign investors tend to underweight firms with a dominant owner of voting rights. Dahlquist and Robertsson (2001) do not investigate whether this underweighting by foreign investors affect the relative valuation of firms with a dominant owner of voting rights. Giannetti and Simonov (2006) analyze whether investors in general are reluctant to invest in Swedish firms with poor corporate governance. They find that investors, who generally only receive security benefits not the private benefits of stockholdings (i.e. foreign 1

investors, institutions, and small shareholders), tend to underweight firms where the principal shareholder holds more voting rights than cash flow rights. Giannetti and Simonov (2006) do not investigate whether the reluctance of some investors to invest in firms with dual class shares is related to the relative value of these firms. Their main results are from 2001. In another Swedish study, Cronqvist and Nilsson (2003) investigate dominant owners of voting rights with limited number of cash flow rights, i.e. controlling minority shareholders using the possibilities of dual class shares rather aggressively. They document that controlling minority shareholders are related to lower firm market value in terms of Tobin s q. Their sample period is 1991-1997, i.e. most of their observation stem from the period after the lifted ban on foreign ownership. Thus, we do not know whether this negative relation between the use of dual class shares and firm value existed when the Swedish market was dominated by Swedish investors. Among some Swedish market participants, dual class shares is not considered a major corporate governance problem. In fact it can be argued that the Swedish legislator encourage the use of dual class shares (see Holmén and Högfeldt, 2004). The voting premium on the high voting A shares is also among the lowest in the world. It is lower than in for example the U.S. (Nenova, 2003). The voting premium is usually considered as a proxy for the controlling shareholder s pecuniary private benefits of control. Thus, a small voting premium is taken as evidence that the pecuniary private benefits of control and expropriation of minority shareholders indeed are limited in Sweden. In the second part of this paper we investigate whether the voting premium changed among dual class firms when foreign ownership increased during the 1990s. Few studies have analyzed the voting premium over time 1 and to our knowledge, no study has investigated if the voting premium is related to the degree of international investors on the market. 1 Caprio and Croci (2008) is an exception. 2

Our sample consists of 208 large (with Swedish standard) non financial firms 1985-2005 and contains 2158 firm years. More than 75 percent of the firm year observations have dual class shares. We are able to estimate the voting premium for 458 firm years. In many dual class firms the A-shares are not listed on the stock market and we cannot estimate a voting premium. We first document that dual class shares are associated with lower Tobin s q. However, this in only true for the period after 1992. Before 1993, we cannot find a dual class shares discount. The relative reduction in dual class firm s market value after 1992 cannot be attributed to worse operating performance. Return on Asset (ROA) does not deteriorate for dual class firms. Neither can be attributed to a more aggressive usage of dual class shares. The controlling shareholders average and median difference between vote and equity ownership did not increase. Large voting stakes with limited cash flow rights gave controlling shareholders the possibility and incentives to expropriate minority shareholders also before 1993. However, Swedish investors believed, correctly or incorrectly, that the controlling minority shareholders would not do so. Swedish investors trusted the Swedish controlling minority shareholders and they did not demand an extra risk-premium for dual class firms. With an increased number of foreign investor trust broke down. Foreign investors did not trust the controlling minority shareholders and demanded an extra risk premium on investments in dual class firms. Franks, Mayer, and Rossi (2009) document that UK firms had dispersed ownership already in the beginning of the 20 th century when no minority shareholder protection laws existed in the UK. They argue that this was possible since the stock markets were local, i.e. directors and investors were neighbours. In fact, ownership concentration in the UK actually increased during the 20 th century when all stock listing moved to London. 3

Franks et al (2009) argue that the increased distance between investors and directors led to a breakdown of trust. The reduced relative valuation of dual class firms is associated with a significant drop in the voting premium. The median voting premium is 3.3 percent up to 1992. After 1992, the median voting premium is 0.6 percent. Thus, the A-shares decreased more in value than the B-shares after the lifted ban on foreign ownership. The foreign investors do not enjoy private benefits of control are not interested in paying a premium on the A-shares. Furthermore, A-shares and B-shares carry the same dividend rights. Thus, if there is a substantial price difference between A and B-shares there might be arbitrage opportunities (Schultz and Shive, 2010). Our results suggest that speculation by foreign investors in the price difference between A and B shares might have driven the price difference towards zero. Alternatively, foreign investors increased the illiquidity discount on the A-shares. The A- shares are typically less liquid than the B-shares. We think the contribution of our paper is that our results suggest that the mere indication of corporate governance problems will generate valuation discounts. It does not matter that examples of expropriation of minority shareholders by controlling minority shareholders are scarce in Sweden. And there are no indications that the problems with dual class shares increased during the 1990s. Thus, even if the problems with dual class shares in Sweden are small or non-existent, this cannot be credibly signalled to foreign investors. And it will increase the cost of capital for dual class firms. The question is whether this externality is big enough to suggest a ban on dual class firms. The rest of the paper is organized as follows. The next section outlines our research hypotheses in more details. Section three presents the data and the methodology. The empirical results are presented in section four. The final section summarizes and concludes. 4

2. Research Hypotheses Sweden is number one in the world in terms of the use of dual class shares (La Porta et al., 1998). At the same time, proxies for the pecuniary private benefits of control in Sweden are among the lowest in the world (Nenova, 2003; Dyck and Zingales, 2004). One interpretation of these facts is that private benefits extraction in Sweden does not occur at the expense of non-controlling shareholders and the estimated proxies are likely to understate the value of private benefits substantially (Holmen and Knopf, 2004; Adams and Ferreira, 2007). However, Cronqvist and Nilsson (2003) document that controlling shareholders relying on dual class shares are associated with lower firm value of Swedish firm. Thus, even if private benefits do not appear to be extracted at the expense of non-controlling shareholders, dual class shares are associated with a value discount. Our first hypothesis is based on the conjecture that especially foreign investors do not trust the controlling shareholders who holds majority voting stakes with minority equity stakes not to derive private benefits that are extracted at the expense of non-controlling shareholders. Hypothesis 1: The discount associated with dual class shares increased after 1992 when the ban of foreign ownership was lifted on the Stockholm Stock Exchange. The percentage of foreign ownership on the Stockholm Stock Exchange is reported in table 1. Foreign ownership increased from about 8 percent in 1990 to almost 40 percent in year 2000. After the IT bubble burst foreign ownership decreased somewhat and has stabilized around 35 percent. Our second hypothesis is conditional on two conjectures. First, we must be able to document an increased discount on dual class firms after 1993. Second, we conjecture that discounts on dual class firms are driven by the extraction of pecuniary private benefits by the controlling shareholders. The controlling shareholders hold large voting stakes with limited equity stakes and therefore have the possibilities and incentives to expropriate minority 5

shareholders. Furthermore, the controlling shareholders typically hold a substantial fraction of the high voting A-shares. Minority investors mainly trade in the low voting B-shares. Thus, any discount on dual class firms should be associated with a larger price difference (voting premium) between A and B shares. Hypothesis 2: The voting premium on high voting A-shares increased after 1992 when the ban on foreign ownership was lifted on the Stockholm Stock Exchange. Based on Schulz and Shive (2010) the opposite hypothesis can be derived. For investors who generally enjoy only dividend benefits and not private benefits, the A and B shares are equivalent. And a substantial price difference between the two could be used for speculation, i.e. trading strategies that would be profitable would the price difference decrease (Schulz and Shive, 2010). If we conjecture that especially institutional traders (domestic and foreign) would engage in such trading, then increased foreign ownership, mainly institutional, would increase the frequency of such trading strategies. And these activities would lead to a smaller price difference between A and B shares, i.e. a reduced voting premium, ceteris paribus. 3. Data and Methodology 3.1 Sample We start with an unbalanced panel dataset containing accounting and stock market data for the largest non-financial Swedish firms listed on the Stockholm Stock Exchange 1985-2005. The accounting data is collected from the Findata Trust database. The sample contains the vast majority of the largest non-financial public firms in this time period. Some large firms that were only listed for one or two years before delisting are not included in the sample. The sample consists of 208 firms and 2158 firm years. The sample characteristics are reported i 6

table 2. On average we have 100 firms per year and it corresponds to roughly 70 percent of the stock market capitalization. Roughly 2/3 of the sample firms have dual class firms. The accounting and market data is combined with ownership data from Sundqvist (1985-1993), Sundin and Sundqvist (1994-2002), Fristedt, Sundin, and Sundqvist (2003) and Fristedt and Sundqvist (2004-2006). 2 This source reports the 25 largest owners in all listed firms as of January each year. Sundin and Sundqvist provide detailed information on coalition structures and families in a wide sense. Thus, if two families are known to cooperate, their shareholdings are aggregated by Sundqvist et al. We have followed their definitions of ownership coalitions. 3.2. Descriptive Statistics In Table 3, we report summary statistics. The median firm has a Tobin s q of 1.123. Tobin s q is defined as the sum of the market value of equity and the book value of total debt divided by the book value of total assets. Furthermore, the median firm has assets with a book value of 1552 million SEK (Size), has financed 21.3 percent of the total assets with long-term debt (Leverage), has 56.4 percent short term assets (Liquidity), is 46 years old (Age), and 11.8 percent Return on Assets (Profitability). Return on Assets is defined as Earnings Before Interest Taxes, and Depreciation (EBITDA) divided by total assets. The sample is split according to the existence of dual class shares. Dual class firms have significantly lower Tobin s q, are significantly larger (Firm Size) and older (Firm Age), and have lower Leverage than one share one vote firms. The median dual class firm has lower Profitability than the median one share one vote firm. Mean differences tested on the natural logarithm of Tobin s q, Firm Size, and Firm Age. Median differences tested by means of Wilcoxon Ranksum test. 2 These publications are referred to as Sundqvist et al below. 7

In panel B we divide the sample in two ways. First according to whether the firm has dual class shares or not. And secondly, according to whether the observation is from before or after 1993 when the ban of foreign ownership was lifted. In the 1985-1992 period the average dual class firm has Tobin s q equal to 1.255 while the average single class firm has Tobin s q equal to 1.438. Even if this difference is large the difference is only significant at the 10 percent level since the distribution is positively skewed and the cross-sectional variance is large. The median dual class firm has Tobins s q equal to 1.083 and the median single class firm has Tobin s q equal to 1.088. This difference is insignificant. However, after 1993 the man and median differences are significant at the 1 percent and 5 percent level, respectively. The dual class firms have mean and median Tobin s q of 1.485 and 1.145, respectively. The single class firms have mean and median Tobin s q of 1.783 and 1.205, respectively. Both dual class and single firms have higher Tobin s q after 1993. We collect the voting premium for 458 firm year observations 1985-2000. In many dual class firms the A-shares are not traded and it is not possible to estimate the voting premium. The voting premium is estimated as the yearly average of the closing price of the A share minus the closing price of the B share divided by the average of the closing price of the A share and the B share. We report statistics on the voting premium in table 4. The average and median voting premium is equal to 7.3 percent and 1.7 percent respectively. During the 1985-1992 period the average and median voting premium is 10.7 percent and 3.3 percent respectively. During the 1993-2000 period the average and median voting premium is 4.1 percent and 0.6 percent respectively. These differences are statistically significant. We also report the average and median voting difference (Voting Diff) for these 458 firm years, i.e how many more votes do the A shares carry compared to the B shares. The Swedish corporate law requires that A and B shares have the same dividend rights but a 8

voting differential of utmost 10 to 1 for A shares over B shares. Firms incorporated under a previous corporate law are allowed to use a voting differential of 1000/1. All firms that had differential of 1000/1 in 1985 have changed to 10/1 during our sample period. Ericsson was the last listed firm to do so in 2004. The median Voting Diff in our sample is 10 in both time periods. However, due to 1000/1 differential in some firms the average is 146 before 1993 and 90 after 1993. Furthermore, we report fraction of A shares of total shares (Fraction A). Before 1993 this fraction is roughly 0.55. After 1993 the fraction of A shares decreased to roughly 0.44. The difference is statistically significant. The change might be due to the tendency of dual class firms to make Seasoned Equity Offers (SEOs) of B shares only (see Holmen and Högfeldt, 2004). Finally, we report the difference between the fraction of cash rights minus the fraction of cash flow rights held by the largest shareholder (Vote-Equity). Among these 458 firm years, the Vote-Equity is 14.9 percent on average. The median is 10.9 percent. There are no significant differences between the two periods. 3.3. Regression models For the main tests we rely on OLS regressions. We adjust for heterskedasticity and correlated observation with the Huber-White-Sandwich estimator for variance. Fixed effect panel regressions are not possible since there is very little within variation in the dual class share indicator variable. Similarly, in the regressions with the voting premium as dependent variable there is very little within variation in some of the independent variables. Furthermore, studies of ownership structure and firm value are always plagued with endogeneity and selection problems. However, since we are mainly interested in how the relation between dual class share and firm value changes 1993 these problems are reduced, i.e. we assume that the endogeneity and selection problems are roughly the same before and after 1993. Tobin s q as a measure of firm value can also be problematic. However, the problems should be the same before and after 1993. Finally, any measure of corporate ownership structure has several 9

alternatives. We keep our measure as simple as possible and include an indicator variable for dual class firms. We perform robustness tests on some of these issues below. In our regressions related to Hypothesis 1 Tobin s q is the dependent variable. Besides the dual class share indicator variable (Dual) we include firm size (Size), Liquidity, Leverage, and firm age (Age) as explanatory variables. We also include year and industry dummies. Thus, we end up estimating the following regression model 93 where Dual93 is equal to one if the firm has dual class shares and the observation is from 1993 or later, and zero otherwise. Instead of including one dummy variable for all observations after 1993 we include a dummy variable for each year. We cannot include both due to perfect linearity. Including one dummy for all observations after 1993 instead of separate year dummies does not change our main results. In our regressions related to Hypothesis 2 the Voting Premium is dependent variable. To capture the effect of the lifted ban on foreign ownership in 1993 we include a variable 1993 which is equal to one if the observation is from 1993 or later, and zero otherwise. We include Voting Diff, Vote-Equity, and Fraction A as independent variable since all these variable should affect the voting premium (see Nenova, 2003). We also include the same firm characteristics and year dummies as in the Tobin s q regressions. 1993 10

4. Empirical Results 4.1. Firm Value In table 5 we report the OLS regressions with Tobin's q as dependent variable. We report coefficients with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. In Model 1 Dual is negatively significant at the 1 percent level. The coefficient is -0.1078. Since we use the natural logarithm of Tobin's q as dependent variable this indicates that dual class firms are valued at roughly a 10 percent discount compared to single class firms. Leverage and Firm Age are negatively significant in both models. In model 2 we include an interaction term of the Dual indicator variable with a dummy being equal to one for the years 1993 to 2005 (Dual_Post1993). 3 This variable is negatively significant at the 1 percent level. Moreover, Dual per se is now insignificant. Thus, the negatively significant relation between dual class shares and Tobin's q is only present after 1993. The coefficient on Dual_Post1993 is -0.1199, i.e. dual class firms are associated with almost a 12 percent discount after 1993. These results are in line with Hypothesis 1. In order to investigate to what extent the post 1993 discount on dual class firms is driven by lower profitability we replace Tobin's q with our Profitability measure in model 3 and 4. Profitability estimated by Return on Assets (ROA) and defined as EBITD divided by total assets. In model 3 Dual is negative but insignificant. When we include the Dual93 interaction term, Dual becomes negatively significant at the 10 percent level. Dual_Post1993 is insignificant. Thus, there is some weak statistical evidence that dual class shares are associated with lower profitability. However, there are no indications that the profitability of dual class firms deteriorated after 1993, which would have explained the discount post 1993. 3 A Post1993variable per se is not included since we already include year dummy variables. 11

In the ROA regressions Firm Size and Age are positively significant while Liquidity and Leverage are negatively significant. Thus, after controlling for industry and year effects larger and older firms have higher ROA. We interpret the reversed sign on Firm Age, compared to the Tobin's q regressions, as older firms being "cash cows", i.e. they generate relatively stable and high cash flows but have little growth opportunities. We think the significance of Size in the ROA regressions can be interpreted in a similar manner. 4.2. Voting Premium In table 6 we report the regressions with the Voting Premium as dependent variable. We report coefficients with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. In model 1 the Post1993 indicator variable is significantly negative indicating that the average price difference between A and B shares decreased after 1992. This result makes us reject our second hypothesis. Instead the results appear to be in line with Schultz and Shive (2010). The fraction of A shares of total shares (Fraction A) is also negatively significant. Thus, the more A shares the lower the voting premium. This is intuitive since a given value of private benefits will be distributed among a larger number of A shares (see e.g. Nenova, 2003). The Voting Diff and Vote-Equity variables are insignificant. This insignificance is surprising especially for the Voting Diff variable. However, there are very few dual class firms that have Voting Diff different from 10 to 1. Thus, it is difficult to pick up an effect even if it indeed exists. In model 2 we include firm characteristics and in model 3 we include year dummies. The main results do not change, i.e. the Post1993 variable is negatively significant. Among the firm characteristics it is only the Leverage that is significant. It is positive. Thus, a high voting premium is associated with more leverage. This result can be interpreted in terms 12

of Stulz (1988). In Stulz (1988) managers increase leverage to entrench themselves and their protect private benefits of control. Such a behavior would result in a positive relation between leverage and proxies for private benefits of control. 4.3. Relation between Tobin s q and the Voting Premium In order to shed on the question whether a voting premium is indeed driven by a premium on the high voting A-shares, or rather a discount on the low voting B-shares, we estimate regressions with Tobin s q as dependent variable and the Voting Premium as independent variable. The results are reported in table 7. We include the same control variables as in the previous regressions and use the 458 firm years for which we have the voting premium. We report coefficients with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. In model 1 Voteprem is negatively significant, i.e. a higher voting premium is associated with lower Tobin s q suggesting that the price difference between A- and B-shares are more likely to be associated with a discount on the B shares, rather than a premium on the A-shares. The results in Model 2 suggest that this relation does not change around 1993. However, when we include year dummies in model 3 the significant relation between Tobin s q and the Voting premium vanishes. Thus, the significant negative relation in Model 1 and 2 appears to be driven by a spurious relation over time, i.e. time periods with lower firm values appear to be related to higher voting premiums, and vice versa. 4.4. Robustness Tobin's q: The approximation of Tobin s q as the sum of the market value of equity and book value of total debt divided by the book value of total assets might be associated with measurement errors. We therefore replace Tobin s q with the market to book ratio of equity in the regressions in table 5. We results are qualitatively the same. 13

We have also run the regressions without taking the natural logarithm of Tobin s q-. The results are qualitatively the same as the ones reported in table 5. Dual Indicator variable: In table 8 we report regression results where the Dual indicator variable has been replaced with the largest shareholder s Vote-Equity ratio. We also include Equity per se since a certain difference between Vote and Equity might have different implications depending on the Equity fraction per se. In model 1, there is no significant relation between Vote-Equity and Tobin s q. In model 2 we introduce a Vote-Equity93 interaction variable that is equal to the largest shareholder s Vote-Equity from 1993 and afterwards, and zero otherwise. The interaction term is negatively significant at the 10 percent level. Vote-Equity per se is insignificant. Thus, there is a statistically weak indication that the relation between firm value and the actual use of dual class shares becomes more negative after 1993. Furthermore, the results with the Dual indicator variable in table 5 is actually statistically stronger suggesting that it is the mere existence of dual class shares that are associated with a discount after 1993, not the actual use of them. Equity per se is negatively significant at the 10 percent level. Thus, it appears as if Equity captures a negative entrenchment effect rather than a positive incentive effect. Alternative breaks: We have tried alternative dates for the change in the relation between dual class shares and firm value. We do not find any indications of a break before 1993. However, we can more or less choose any year during the 1990s and get qualitatively similar results as the ones reported above. This result might be due to the gradual increase in foreign ownership during the 1990s (see table 1). The alternative explanation is that the attitudes towards dual class shares among Swedish investors changed gradually during the 1990s. To the best of our knowledge, very few Swedes were aware of the internationally unique separation of voting rights from cash flow rights in Sweden until the La 14

Porta et al papers (1997, 1998, and 1999) documented that Sweden is number one in the world in terms of the use of dual class shares and number two in terms of pyramid structures. As far as we know, the first working paper versions of the La Porta papers appeared around 1995. This almost coincides with the lifted ban on foreign ownership in Sweden. Thus, any changes in the relative valuation of dual class firms can be attributed to more foreign investors, changed attitudes among Swedish investors, or both. 4 Furthermore, the biggest economic fraud ever in Sweden, the Trustor case, happened in 1997. And it did involve a block transfer of the high voting A shares. Thus, changed attitudes towards dual class shares among Sweden investors could have been triggered by this event. Selection problems: Finally, we run treatment effect regressions to control for potential selection problems. If certain types are firms are more likely to have dual class shares and lower Tobin s q our results so far might be driven by a spurious correlation. In the treatment equation the existence of dual class shares is regressed on the fraction of dual class firms in the industry the firm belongs to (Dual Class, Ind), firm Size and firm Age. Dual Class, Ind should be highly correlated with Dual and uncorrelated with Tobin s q. Most large and old firms have dual class shares and we therefore include firm Age and Size in the Dual equation. The model is estimated by maximum likelihood. Estimating the model with the Heckman two-step procedure generates similar results. The results are reported in table 9. Instead of using a Dual93 interaction term we now split the sample in a pre 1993 and post 1993 period. Using the Dual93 interaction term would require GMM techniques. Observations from 1993 are included in the post 1993 subsample. The estimated Dual indicator variable is insignificant in the Pre-1993 period and 4 It is of course questionable whether Swedish investors read academic working papers. 15

negatively significant at the 1 percent level in the post 1993 period. Leverage is negatively significant in both periods while Size is negatively significant in the pre 1993 period. In the Dual equation the Dual Class, Ind is positive and highly significant. Firm Age is also positively significant. However, when we control for Dual Class, Ind and firm Age, firm Size is not statistically significant. The Wald test rejects the null hypothesis that the equations are independent in the pre 1993 period. However, the null hypothesis is not rejected in the post 1993 period. 5. Summary and Conclusion In this paper we investigate how the lifted ban on foreign ownership on the Stockholm Stock Exchange in 1993 affected i) the relative valuation of dual class firms and ii) the relative valuation of high voting and low voting shares. Our sample period is 1985 to 2005. We find that the discount on dual class firms is a post 1993 phenomena. There are no indications that dual class firms were valued at a relative discount before 1993. We find no indications that the post 1993 discount is driven by deteriorating operating performance or a more aggressive use of dual class shares. Furthermore, we are not aware of any evidence suggesting that controlling shareholders in Sweden started to expropriate after 1993. We actually document that the voting premium on the high voting A-shares, a proxy for the pecuniary private benefits of control, significantly decreased post 1993. Our general interpretation of the results is that foreign investors put a discount of dual class firms independent of actual expropriation of minority shareholders by controlling shareholders. We think our results have two general implications. First, even if there are few indications of dual class shares leading to clear expropriation of minority shareholders in Sweden, the mere existence of dual class shares leads to value discounts and higher cost of 16

capital for dual class firms. This might have implications for corporate investment levels and economic growth. Second, many corporate governance studies are based on data spanning only one or a couple of years. Our results suggest that the relation between corporate governance characteristics and firm value and behaviour might change over time. In particular, changes to the investor base might affect the relation between corporate governance characteristics and firm behaviour and value. 17

References Adams, R., and D. Ferreira (2007) One share, one vote: The empirical evidence, ECGI Finance working paper 177/2008. Caprio, L., and E. Croci (2008) The determinants of the voting premium in Italy: The evidence from 1974 to 2003, Journal of Banking and Finance, 32, 2433-2443. Cronqvist, H., and M. Nilsson (2003) Agency Costs of Controlling Minority Shareholders, Journal of Financial and Quantitative Analysis, 38, 695-719. Dahlquist, M., and G. Robertsson (2001) Direct foreign ownership, institutional investors, and firm characteristics, Journal of Financial Economics, 59, 413-440. Dyck, A. and L. Zingales (2004) Private benefits of control: An international comparison, Journal of Finance 59, 537-600. Frank, J., C. Mayer, and S. Rossi (2009) Ownership: Evolution and regulation, Review of Financial Studies, 22, 4009-4056. Fristedt, D., A. Sundin, and S. I. Sundqvist (2003) Owners and Power in Sweden s Listed Companies, Dagens Nyheter, Stockholm. Fristedt, D., and S. I. Sundqvist (2004-2006) Owners and Power in Sweden s Listed Companies, Dagens Nyheter, Stockholm. Gianetti, M., and A. Simonov (2006) Which investors fear expropriation? Evidence from investors portfolio choice, Journal of Finance, 61, 1507-1547. Holmén, M. and J. D. Knopf (2004) Minority shareholder protection and the private benefits of control for Swedish mergers, Journal of Financial and Quantitative Analysis, 39, 167-191. Holmén, M., and P. Högfeldt (2004) A law and finance analysis of initial public offers, Journal of Financial Intermediation, 13, 324-358. La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. W. Vishny (1997) Legal determinants of external finance, Journal of Finance, 52, 1131-1150. La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. W. Vishny (1998) Law and Finance, Journal of Political Economy, 106, 1113-1155. La Porta, R., F. Lopez-de-Silanes, and A. Shleifer (1999) Corporate ownership around the world, Journal of Finance 54, 471-517. Nenova, T. (2003) The value of corporate voting rights and control: A cross-country analysis. Journal of Financial Economics 68, 325-351. Schulz, P., and S. Shive (2010) Mispricing of dual-class share: Profit opportunities, arbitrage, and trading, Journal of Financial Economics, 98, 524-549. 18

Stulz, R. (1988) Managerial control of voting rights: Financing and the market for corporate control, Journal of Financial Economics, 20, 25-54. Sundqvist, S. I., (1985-1993) Owners and Power in Sweden s Listed Companies, Dagens Nyheter, Stockholm. Sundin, A., and S. I. Sundqvist (1994-2002) Owners and Power in Sweden s Listed Companies, Dagens Nyheter, Stockholm. 19

Table 1: Foreign Ownership on Stockholm Stock Exchange Year % Foreign ownership 1985 8 1986 8 1987 6 1988 7 1989 7 1990 7.7 1991 12.3 1992 18.0 1993 21.3 1994 28.3 1995 29.6 1996 31.6 1997 31.6 1998 34.6 1999 39.0 2000 39.0 2001 34.6 2002 33.5 2003 33.1 2004 33.9 2005 35.3 20

Table 2: Sample In this table, we provide statistics on the number of sample firms. The sample consists of Swedish non-financial firms listed on the Stockholm Stock Exchange 1985-2005 contains 208 firms and 2158 firm years. Year 1. Number of Sample Firms 2. Percentage of market cap. 3. Number of dual class firms 4. Percentage of Sample Firms having dual class shares 1985 80 60.3 61 76.2 1986 89 70.7 69 77.5 1987 90 66.9 74 82.2 1988 86 60.1 75 87.2 1989 93 73.1 82 88.2 1990 93 70.1 81 87.1 1991 90 71.2 81 90.0 1992 91 72.8 81 89.0 1993 94 78.4 82 87.2 1994 110 81.1 89 80.9 1995 118 78.3 94 79.7 1996 132 81.2 102 77.3 1997 143 75.6 100 69.9 1998 142 70.5 99 69.7 1999 130 53.0 89 68.5 2000 120 70.5 81 67.5 2001 104 49.1 68 65.4 2002 100 50.5 66 66.0 2003 90 49.3 64 71.1 2004 87 49.6 61 70.1 2005 76 51.2 55 72.4 21

Table 3: Descriptive statistics large Swedish non-financial firms 1985-2005 In this table, we provide summary statistics for the 208 firms and 2158 firm years in our sample. The sample is split by whether the firm has dual class shares or not. Tobin s q is defined as the sum of market value of equity and book value of debt divided by the book value of total assets. Firm Size is defined as the book value of total assets in Million SEK. Leverage is equal to the value of long-term debt divided by the book value of total assets. Liquidity is measures as. Firm Age is the number of years since the firm was founded. Profitability is measured as Earnings Before Interest, Taxes, and Depreciation (EBITD) divided by Total Assets. Median difference tested by means of Wilcoxon- Ranksum test. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level. Panel A: Total sample characteristics and differences between Dual Class and Single Class Firms Total Sample, N=2158 Dual Class Firms, N=1654 Single Class Firms, N=504 Difference Mean Median Mean Median Mean Median t-test Ranksum test Tobin s q 1 1.473 1.123 1.401 1.115 1.709 1.158-4.23*** 3.20*** Firm Size 1 10543 1552 12364 1576 4569 1504 2.77*** 1.80* Leverage 0.237 0.213 0.242 0.221 0.221 0.187 2.26** 3.97*** Liquidity 0.541 0.564 0.545 0.564 0.526 0.559 1.34 1.53 Firm Age 1 57 46 59 50 49 21 9.43*** 8.39*** Profitability 0.141 0.118 0.141 0.101 0.142 0.117 0.32-2.59*** 1 mean difference tested on the natural logarithm of these variables. Panel B: Two-by two comparison of Tobin s q sorted by dual class and single class firms and 1985-1992 and 1993-2005, respectively. Dual Class Firms 1985-1992, n=604. Single Class Firms 1985-1992, n=108. Dual Class Firms 1993-2005, n=1050. Single Class Firms 1993-2005, n=396. Tobin s q 1 Dual Class Firms Single Class Firms Difference Mean Median Mean Median t-test Ranksum test 1985-1992 1.255 1.083 1.438 1.088 1.69* 1.29 1993-2005 1.485 1.145 1.783 1.205 3.32*** 2.43** Difference -2.31** -3.36*** -1.75* -1.94* Mean difference tested on the natural logarithm Tobin s q. 22

Table 4: Voting Premium In this table we report the voting premium for the 51 sample firms with listed A and B shares 1985-2000. The number of firm years is 458. The sample is split in two time periods, 1985-1992 and 1993-2000, respectively. The voting premium is estimated as the yearly average of the price of the high voting A share minus the price of the low voting B share divided by the average of the price of the A share and the B share. Statistics on the voting difference (Voting Diff), fraction of high voting A shares (Fraction A), and the largest shareholders vote to equity difference (Vote-Equity) are also reported. The fraction of A shares are estimated as the number of A share divided by the total number of shares. Median difference tested by means of Wilcoxon- Ranksum test. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level. Total Sample, N=458 1985-1992, N=222 1993-2000, N=236 Difference Mean Median Mean Median Mean Median t-test Ranksum test Voting Premium 0.073 0.017 0.107 0.033 0.041 0.006 4.41*** 5.33*** Voting Diff 117 10 146 10 90 10 1.93* 0.76 Fraction A 0.494 0.485 0.544 0.548 0.447 0.439 4.18*** 4.16*** Vote-Equity 0.149 0.109 0.133 0.103 0.148 0.123-1.09 1.31 23

Table 5: Regressions with firm performance measures as dependent variables In this table we report OLS regressions with the natural logarithm of Tobin s q and profitability, respectively, as dependent variables. Tobin s q is defined as the sum of market value of equity and book value of debt divided by the book value of total assets. Profitability is measured as Earnings Before Interest, Taxes, and Depreciation (EBITD) divided by Total Assets (ROA). Coefficients are reported with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. Dual is equal to one if the firm has dual class shares, and zero otherwise. Dual_Post1993 is equal to one if the firm has dual class shares after 1992, and zero otherwise. Firm Size is defined as the book value of total assets in Million SEK. Leverage is equal to the value of long-term debt divided by the book value of total assets. Liquidity is measures as. Firm Age is the number of years since the firm was founded. Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level. Model 1 Ln(Tobin s q) Model 2 Ln(Tobin s q) Dual -0.1078-0.0169 (-3.98)*** (-0.37) Dual_Post1993-0.1199 (-2.19)** Ln(Size) -0.0048-0.0037 (-0.62) (-0.59) Liquidity 0.0302 0.0274 (0.67) (0.61) Leverage -0.4471-0.4487 (-5.93)*** (-5.94)*** Ln(Age) -0.0267-0.0268 (-2.64)*** (-2.66)*** Year Dummies Industry Dummies Adj R 2 0.2582 0.2594 N 2158 2158 Model 3 ROA -0.0028 (-0.50) 0.0082 (5.98)*** -0.0371 (-4.21)*** -0.1039 (-5.81)*** 0.0046 (2.33)** 0.1530 2158 Model 4 ROA -0.0107 (-1.91)* 0.0104 (1.13) 0.0082 (5.97)*** -0.0368 (-4.16)*** -0.1038 (-5.80)*** 0.0046 (2.33)** 0.1529 2158 24

Table 6: Regressions with the voting premium as dependent variable In this table we report OLS regressions with the voting premium as dependent variable. The voting premium is estimated as the yearly average of the price of the high voting A share minus the price of the low voting B share divided by the average of the price of the A share and the B share. The voting premium is estimated for 51 sample firms with listed A and B shares 1985-2000. The number of firm years is 458. Coefficients are reported with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. 1993 is equal to one if the observation is from 1993 or later, and zero otherwise. Voting Diff is the number of votes per A share divided by the number of votes per B share. Vote- Equity is the largest shareholder s fraction of votes minus fraction of cash flow rights. Fraction A shares is defined as the number of A share divided by the total number of shares. Firm Size is defined as the book value of total assets in Million SEK. Leverage is equal to the value of long-term debt divided by the book value of total assets. Liquidity is measures as. Firm Age is the number of years since the firm was founded. *** Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level. Model 1 Model 2 Model 3 Post1993-0.0742 (-3.27)*** -0.0815 (-3.46)*** -0.0685 (-2.50)** Voting Diff 0.0005 (1.48) 0.0004 (1.01) 0.0004 (1.01) Vote-Equity 0.065 (0.71) 0.040 (0.45) 0.0267 (0.30) Fraction A -0.1823 (-2.63)** -0.1775 (-3.37)*** -0.1750 (-3.56)*** Ln(Size) 0.0082 (1.27) 0.0065 (1.08) Liquidity -0.0027 (-0.05) 0.0462 (0.71) Leverage 0.1945 (2.29)** 0.2696 (2.62)** Ln(Age) 0.0125 (1.38) 0.0153 (1.51) Year Dummies Industry Dummies No No Adj R 2 N 0.2837 458 0.3322 458 0.3575 458 25

Table 7: The relation between Tobin s q and the voting premium In this table we report OLS regressions with the natural logarithm of Tobin s q as dependent variables. Tobin s q is defined as the sum of market value of equity and book value of debt divided by the book value of total assets. Coefficients are reported with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. Voteprem is the voting premium on listed A shares. It is estimated as the yearly average of the price of the high voting A share minus the price of the low voting B share divided by the average of the price of the A share and the B share. Voteprem93 is equal to the votepremium if the observation is from 1993 or later, otherwise, it is zero. Size is defined as the book value of total assets in Million SEK. Leverage is equal to the value of long-term debt divided by the book value of total assets. Liquidity is measures as. Firm Age is the number of years since the firm was founded. Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level. Model 1 Model 2 Model 3 Voteprem -0.2931 (-2.08)** -0.3446 (-2.19)** -0.1172 (-0.74) Voteprem93 0.2165 (0.84) -0.0871 (-0.36) Ln(Size) 0.0073 (0.21) 0.0075 (0.22) -0.0082 (-0.24) Liquidity 0.3654 (1.46) 0.3862 (1.47) 0.5103 (1.61) Leverage -0.2521 (-1.18) -0.2612 (-1.24) -0.2837 (-1.33) Ln(Age) 0.0656 (1.59) 0.0657 (1.58) 0.0755 (1.69)* Year Dummies Industry Dummies No No Adj R 2 N 0.2624 458 0.2620 458 0.3131 458 26

Table 8: Alternative Regressions with Tobin s q as dependent variable In this table we report OLS regressions with the natural logarithm of Tobin s q dependent variable. Tobin s q is defined as the sum of market value of equity and book value of debt divided by the book value of total assets. Coefficients are reported with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. Vote-Equity is equal to the difference between largest shareholder s voting fraction and equity fraction. Vote-Equity_Post1993 is equal to the difference between largest shareholder s voting fraction and equity fraction. if the observation is from 1993 or later, and zero otherwise. Firm Size is defined as the book value of total assets in Million SEK. Leverage is equal to the value of long-term debt divided by the book value of total assets. Liquidity is measures as. Firm Age is the number of years since the firm was founded. Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level. Model 1 Ln(Tobin s q) Model 2 Ln(Tobin s q) Equity -0.1272 (-1.96)* -0.1139 (-1.75)* Vote-Equity -0.0087 (-0.11) 0.1704 (1.41) Vote-Equity_Post1993-0.2704 (-1.83)* Ln(Size) -0.0081 (-1.12) -0.0074 (-1.03) Liquidity 0.0275 (0.66) 0.0262 (0.62) Leverage -0.4587 (-6.10)*** -0.4560 (-6.03)*** Ln(Age) -0.0332 (-3.30)*** -0.0326 (-3.26)*** Year Dummies Industry Dummies Adj R 2 N 0.2530 2158 0.2538 2158 27

Table 9: Treatment Effect Regressions In this table we report treatment effect (TE) regressions with the natural logarithm of Tobin s q as dependent variable. In the treatment equation the existence of dual class shares is regressed on the fraction of dual class firms in the industry the firm belongs to, firm size and firm age. The model is estimated by maximum likelihood. Tobin s q is defined as the sum of market value of equity and book value of debt divided by the book value of total assets. Coefficients are reported with t-statistic in parenthesis. The t-statistics are corrected for heteroskedasticity and correlated observation with the Huber-White Sandwich estimator. The sample is split into a pre 1993 and post 1993 period. Vote-Equity is equal to the difference between largest shareholder s voting fraction and equity fraction. Vote-Equity_Post1993 is equal to the difference between largest shareholder s voting fraction and equity fraction. if the observation is from 1993 or later, and zero otherwise. Firm Size is defined as the book value of total assets in Million SEK. Leverage is equal to the value of long-term debt divided by the book value of total assets. Liquidity is measures as. Firm Age is the number of years since the firm was founded. Significant at the 1% level; ** Significant at the 5% level; * Significant at the 10% level. Pre-1993 Post-1993 TE Ln(Tobin s q) Dual TE ln(tobin s q) Dual Dual 0.0037 (0.08) -0.2253 (-2.91)*** Ln(Size) -0.0282 (-3.90)*** 0.0134 (0.46) 0.0066 (0.82) -0.0013 (-0.07) Liquidity -0.1025 (-0.98) 0.0253 (0.51) Leverage -0.3248 (-2.58)** -0.4782 (-5.30)*** Ln(Age) -0.0051 (-0.46) 0.1162 (1.95)* -0.0156 (-0.97) 0.2318 (5.38)*** Dual Class, Ind. 2.9602 (5.70)*** 2.8727 (8.68)*** Year Dummies Industry Dummies No No No No p-value indep eq. N 0.0004 712 0.1876 1446 28