Morten Bennedsen y and Kasper Meisner Nielsen z. April 2005

Transcription

1 The Principle of Proportionality: Separating the Impact of Dual Class Shares, Pyramids and Cross-ownership on Firm Value across Legal Regimes in Western Europe Morten Bennedsen y and Kasper Meisner Nielsen z April 2005 Abstract: Recent policy initiatives within theharmonization of European companylawshavepromoted the \principleofproportionality" through proposals that regulate mechanisms opposing a proportional distribution of ownership and control. We scrutinize the foundation for these initiatives by analyzing the use of instruments to separate ownership from control across legal regimes in a sample of 4,000+ publicly traded rms from 14 Western European countries. First, we con rm the negative impact on rm value from disproportional ownership structures previously established in a sample of Asian rms by Claessens et al. (2002). Second, we show that dual class shares have a larger and more signi cant negative e ect on value than pyramids and cross holdings. Third, the impact of disproportionality and the underlying instruments is inversely related to the level of investor protection. Thus, we argue that dual class shares and pyramids may be a substitute for legal protection in countries with inadequate protection of investors. Fourth, we nd no evidence that disproportionality instruments have a signi cant e ect on earnings performance. Finally, we discuss the policy implications of these ndings in relationship tothe process of harmonization of EU company laws. We would like to thank participants at the 'Symposium on Dynamic Corporate Finance and Incentives' at Copenhagen Business School for helpful comments and suggestions. We further like to thank Jonas Herby and Rune Midjord for excellent research assistance. This project has been supported by CEBR ( and the Danish Social Science Research Foundation under the project 'GOCOW'. y Copenhagen Business School and CEBR. Solbjerg Plads 3 C5, 2000 Copenhagen, Denmark. mb.eco@cbs.dk. z Centre for Economic and Business Research and CAM, University of Copenhagen. kn@cebr.dk.

2 1 Introduction The European Commission has initiated a number of policy proposals directed at a ecting the distribution of ownership and control in European companies. 1 A common trend in these initiatives is to promote the so-called \principle of proportionality", which states that it is desirable to have proportional distributions of cash ow rights and control rights in publicly listed corporations. In this paper we scrutinize the foundation for this principle in a sample of publicly traded corporations from 14 Western European countries. The importance of ownership concentration on rm value and performance has been widely analyzed in the last twodecades (Demsetz and Lehn 1985, Morck, Shleifer and Vishny 1988 and many others). In general ownership concentration provides two e ects on the governance ofcorporations, namely an incentive e ect making monitoring of management more e±cient and an entrenchment e ect making it easier for opportunistic owners to behave in a matter that enrich themselves at the cost of other owners. In general corporations with proportional ownership structures seem to create more value than rms in which ownership of control is more concentrated than ownership of cash ow (Claessens, Djankov, Fan and Lang 2002, Cronqvist and Nilsson 2003). Thus, the present evidence suggests that corporations that wish to maximize rm value should obey to the \Principle of Proportionality". We con rm the negative impact of disproportional ownership structures on rm value; however, in addition we take the analysis two steps further: The rst step is to separate the e ect of di erent instruments to create disproportional ownership structures. In particular, we show that creating disproportional ownership structures through the use of dual class shares destroys more value than creating disproportional ownership structures through pyramids, which again destroys signi cantly more value than 1 Backed by the recommendations of the Final Report of the High Level Group of CompanyLaw Experts (Winter et al. 2002) the Commission proposed the so-called Break- Through rule allowing owners of 75 pct. or more of the cash ow rights in a corporation to exercise control even if they possessed less than 50 pct. of the votes. This proposal was later removed due to political obstacles. In the EU Action Plan on \Modernizing Company Law and Enhancing Corporate Governance in the EU" from 2003 it is suggested to prohibit the listing of \abusive" pyramids on the stock exchange. 2

3 cross ownership and other instruments. Hence, we show that it is not disproportionality as such that destroys rm value; rather the reduction comes from the particular instrument used to create a wedge between ownership and control. The second step is to document that the impact of di erent instruments depends on the level of legal protection of outside investors. Theoretically disproportionality instruments trigger both incentive and entrenchment effects. However, the bene t of the incentive e ect depends on the level of investor protection in thefollowing way: If acountryhas astrongprotection of investors, managers have strong incentives to work in the interest of the owners even in the absence of large controlling owners, since the market for corporate control is active and transparent. On the other hand, if investor protection is low, managers may feel little outside pressure and therefore may tend to slack more if they are not monitored closely by large controlling owners. Similarly, the controlling owners' entrenchment opportunities may also be lower in countries with good investor protection. Thus, whereas it is theoretically clear that the impact of a given disproportionality instrument on rm value depends on the level of investor protection, the actual sign of this relationship is an open empirical question. We answer this question by interacting our measures of disproportional ownership structures with the two measures of legal investor protection from LaPorta, Lopez-de-Silanes and Shleifer (1999). We show that the negative e ect of disproportional ownership structures and in particular dual class shares and pyramids is increasing in the level of investor protection, but that the net e ect is insigni cant in countries with a lower investor protection. Hence, we provide evidence for that dual class shares, pyramids and other instruments that separate control from cash ow may substitute investor protection in countries with a low standard of legal protection. The results of the present paper have policy implications for the harmonization of EU company laws. Disproportionality decreases rm value and in particular disproportionality created through the useofdual class shares. Thus, on one side this may provide arguments in favor of policy proposals that make it less attractive to choose dual class shares, pyramids and other instruments. On the other side, we also show that the desirability of such proposals varies signi cantly across countries and in particular it depends 3

4 on the degree of investor protection on the national level. Whereas limiting the use of dual class shares and pyramids can increase rm value in Northern Europe, this conclusion is likely not to hold for rms in Southern European countries where these instruments can function as a substitute for investor protection. Hence, our analysis support the view that currently one size does not t all countries with respect totheharmonization of corporate laws and that high investor protection is aprerequisiteforastringentimplementation of the principle of proportionality within the European Union. The paperproceeds as follows: In the nextsection wepresent our sample of Western European rms. In Section 3 we con rm the negative relationship between disproportional ownership structures and rm value in Western European rms. This relationship has previously been con rmed for individual countries and for a sample of Asian rms in Claessens et al. (2002). Section 4 disentangles this negative relationship by separating the e ects of the instruments that have created this disproportionality, showing that disproportionality through dual class shares has aparticular strongnegative impact on rm value. In Section 5 we analyze the connection between disproportionality and investor protection and show that the negative impact of disproportional ownership structure is stronger in countries with good investor protection and weak in countries with bad investor protection. Section 6 analyzes the impact of disproportionality on earnings performance. Finally, the last section discusses the policy implications of our ndings both with respect to creating an internal capital market within the EU and with respect to national codes and principles for good corporate governance. 1.1 Related literature There are a number of studies that have analyzed the impact of disproportional ownership structures on rm value and performance. Claessens et al. (2002) identify empirically the incentive and entrenchment e ects of large shareholders. In asample of1,301publiclytraded corporations in eight East Asian countries they show that ownership concentration in itself increases rm value, but that separation of cash ow and control decreases rm value. They also try to measure the importance of the instruments separating ownership and control rights to shed light on which mechanisms that are driving theresults. However, theresultsare grossly insigni cant and they arethere- 4

5 fore not able to disentangle which disproportionality instrument is associated with the valuation discount. Lins (2003) investigates rm performance and managerial ownership in corporations in 18 emerging markets and nds that rm value is reduced whenever votes is more concentrated than cash ow. Gompers, Ishii and Metrick (2004) analyze a sample of US rms and show that the relationship of rm value to managerial ownership concentration measured with cash ow is positive and concave whereas the relationship to voting concentration is negative and convex. Hence, these ndings are very similar to the Claessens et al. study of Asian rms. Consistent evidence is provided by Gompers et al. by running regressions of sales growth, capital expenditures and R&D spending on managerial ownership. Entrenched managers underinvest whereas managers with high cash ow rights pursue more aggressive investment strategies. Cronqvist and Nilsson (2003) analyze the impact of controlling minority shareholders on rm value and rm performance in a sample of 309 publicly traded Swedish rms. They show that having controlling minority owners, i.e. a disproportional ownership structure, decreases rm value and performance and that this e ect is most signi cant if these controlling minority shareholders are families. Maury and Pajuste (2004) document that rm value decreases if large owners control rms through disproportional ownership structures in a sample of 174 Finnish rms. All these papers provide evidence for that the concentration of ownership and control are associated with both incentive and entrenchment e ects. In thepresentwe con rm this insight in a sample of 4,000+Western European publicly traded rms. Since the European capital markets aredi erent from capital markets in emerging countries and in the US (La Porta et al. 1999), the nding that there are incentives and entrenchment e ects in a large sample of European rms is important in itself. However, compared to the studies above our main contribution is to extent the analysis by decomposing the e ect of disproportionality into types of instruments, i.e. to investigate if instruments like dual class shares, pyramids and cross ownership provide the same impact on rm value, and by analyzing whether legal investor protection matters for these results. There is a huge literature on ownership concentration and rm valuation and performance. Following Demsetz and Lehn (1985) a number of studies 5

6 have found no signi cant relationship between ownership and performance, whereas a second group of studies initiated with Morck et al have found a non-linear relationship as the combined outcome of incentive and entrenchment e ects. Our study has little to say about ownership concentration as such, we focus on the instruments behind the separation of control from ownership. The consequences of disproportional ownership structures have in addition been documented through estimating the voting premium on shares with superior voting rights (see e.g. Zingales 1994, 1995, Nenova 2003 and references herein). This premium hasbeen read as ameasureofprivatebene ts of control. Interestingly, this measure seems to be higher in countries with low protection of investors and lowest in countries with good investor protection. Hence, the ability toextract private bene t may be limited when investor protection is good. This is one part of our general argument described above. With respect to rm value, we argued that dual class shares may also increase the value of better monitoring incentives. We show that the aggregate e ect on rm value of dual class shares is negatively related to investor protection. 2 Data Our data are obtained from two di erent sources. The data on ownership structure and rm organization are primarily obtained from Faccio and Lang's (2002) study of rms in Western Europe. We have extended their data set with rms in Denmark and Sweden. 2 Danish rms were not included in Faccio and Lang's study, while we are able to extend the number of included Swedish rms from 245 to 335implyingthat we have ownership information on 5,521 Western European rms. All ownership variables are de ned accordingly to Faccio and Lang (2002). We merge this data on ownership structure and rm organization with accounting data from Worldscope from 1996 to1998. We use the name of the rm as the identi er between the two data sets. The matching on rm name produced anumber of obstacles duetoshortcomingsin the datasources. To 2 The ownership structures of Danish and Swedish rms are obtained from Greens and SIS Äagarservice, respectively. 6

7 reducethenumber ofmissing rms wechecked for changes in rm name and de-listings. Further, not all listed rms in Europe are included in Worldscope - in particular rms in Spain and Belgium are missing on a large scale, since only 170 out of 604 listed Spanish rms and 94 out of 130 Belgian rms are included. The total number of rms is reduced from 5,521 to 4,410 mainly due to missing rms in Worldscope and to a smaller extent due to shortcomings in the matching procedure and deviations between Faccio and Lang's data and Worldscope. Our analysis seeks to examine the e ects of a disproportional ownership structures on rm value measured by Tobin's Q, where we use the standard de nition that Q equals market value of equity plus book value ofdebt over book value of assets. A potential problem with our measure of Q arises for the rms with dual class shares, since the superior voting shares can either be publicly or privately held. We calculate rm value on the basis of the publicly traded shares and therefore assume that the price of the superior voting shares equals the price on the limited voting shares for those rms with privately held superior voting shares. We thereby assume that the superior voting shares carry zero voting premiums. In Section 6 below we address thispotential valuation biasand show thatit isnotthedrivingforce behind our results in the following sections. In our regressions wecontrol for awiderangeof rm characteristics that are likely to a ect rm performance (see Table 1 for a de nition of the explanatory variables). To control for size e ects we include sales. Similarly, we include leverage to control for rms in nancial distress and both assets tangibility and sales growth to capture that growth rms have higher market-to-book ratios. Finally, we control for the largest owners cash ow stake. Unfortunately not all rms in Worldscope report all of our variables, we therefore exclude 314 rms where either market value, sales, sales growth or assets tangibility were missing. We further exclude 5 rms with assets under1 million dollarsand 4 rms with extremesales growth, since our control variables take extreme values for these few rms. Thus, the empirical analysis is carried out with 4,096 observations. 3 3 We are missing adequate information on the return on assets for 46 of the 4,096 rms, thus in section 6 where we analyze the link between disproportional ownership structures and earnings performance the number of observations is reduced to 4,050. 7

8 The merged sample is a representative sub sample of Faccio and Lang's data with respectto the employment of disproportionality instruments. Table 2 provides summary statistics on country and regional level of both the dependent and explanatory variables that are used in the empirical section. Average Tobin's Q on country level is increasing in the level of investor protection, which is consistent with evidence from La Porta et al. (2002). Further it should be noted that publicly held rms in Central and Southern Europe are larger than rms in the U.K. and Scandinavia; however, the Central and Southern European rms neither grow faster nor perform better than rms in Northern Europe. 3 Incentives and Entrenchment in European Corporations 3.1 The frequencies of disproportional ownership structures in Western European rms Table 3 depicts the use of instruments to separate cash ow and control in Western Europe. It shows the share of rms with dual class shares, pyramidal ownership, cross ownership and other instruments. The latter group is de ned as the residual group, i.e. rms with disproportional ownership structures that do not belong to the three former groups, e.g. rms with restricted voting rights (voting caps) or some form of government control through golden shares. The frequency of these four groups di ers across countries. Dual class shares are widely used in Denmark, Finland, Italy, Sweden, Switzerland and, surprisingly, United Kingdom, whereasthey areabsentin Belgium, Portugal and Spain and almost absent in France. Pyramids are frequently used in all European countries, but less pronounced in Finland and Switzerland. Cross-holdings are rare and only present in Austria, Germany, Italy, Norway, Sweden and United Kingdom. Other instruments to separate cash ow and votes are used by less than 1 pct. of the rms. On the aggregate level disproportionality instruments are highly frequently used in Scandinavia, Ireland, Italy, Switzerland and United Kingdom and less frequently used in France, Portugal and Spain. This nding may at rst glance seems challenging to the traditional view within the 8

9 investor protection literature (see La Porta et al. 1999), since we nd a higher fraction of rms that use instruments toconcentrate control in countries with high investor protection (Scandinavia, Ireland and the UK) than in countries with poor investor protection (Central and Southern Europe). However, as we show in the next section, the use of these instruments differs signi cantly between the Anglo-Saxon countries and the Continental Europe. Apartfrom thedi erences in thefrequency ofthese instruments laid out by Table 3, the legal de nition of each instrument varies from country to country. Even though dual class shares are e ective remedies to trigger a wedge between ownership and control, the potential degree of disproportionality depends on the relative voting power of the superior voting shares (SVS) to the limited voting shares (LVS) and the distribution of the ownership. Dual class shares can be issued without any restrictions in Austria, Ireland and Switzerland whereas a one-share-one-vote ownership structure is required in Belgium and in principle in Norway 4 A majority of the other European countries have a cap on the proportion of the non-voting shares that can be issued. The threshold levels of the nominal capital that LVSs are not allowed to exceed are 50 pct. in Germany, Italy, Portugal and Spain, and 25 pct. in France. Denmark, Finland and Sweden have imposed a minimum voting ratio of one-tenth between SVS and LVS (with potential "grandfather"-rules that provide exemptions for older rms with di erent votingratios when therules where implemented), whereas non-votingshares has been outlawed in the UK since The underlying restrictions and variation in the corporate law among European countries are re ected in the data. In Germany rms with dual class shares often assign no voting power to the limited voting shares, thus the LVS is reduced to a claim on future income rights. 5 Similarly, in Denmark and Sweden a 10:1 voting relation between SLVs and LVSs is the most common, since it is applied by almost all rms with dual class shares in the two countries. 6 In France the rms with dual class shares are primarily privatized state owned enterprises. To sum up, even though dual class shares 4 Departures from the one-share-one-vote principle in Norway require government approval. Our data shows that these exceptions are granted frequently. 5 Becht and BÄohmer (2001) 6 Bennedsen and Nielsen (2004) 9

10 are frequently used in Europe, there is a large variation across nations in design and regulation of these instruments. The valuation of European rms, therefore, does re ect variations in regulation, in the prevalence of the use of disproportionality instruments and in the distribution of ownership. Figure 1 illustrates the importance of the combined e ect of the prevalenceofinstruments and thedistribution ofownership by showingthedistribution of cash ow and votes for the European rms in our gross sample of 4,410 rms. Tosavespacewe group the rmsintothefourregions according to their legal origin (see LaPorta et al. 1999). In Panel A we focus on the largest owner measured by votes, whereas Panel B shows the group of large owners, de ned as the sum of cash ow and votes of all owners who possess more than 10 percent of the votes. For each region we state the number of rms in our sample, the number of rms with dual class shares and the number of rms with other instruments that create disproportionality. We notice - consistently with Barca and Becht (2001), Gugler (2002) and Faccio and Lang (2002) - that most rms in all Western European countries have some form of concentrated ownership, i.e. that the group of large owners in general possesses more than 20 percent of both cash ow and votes. It is also evident that ownership structures in the two Anglo- Saxon countries in general are less concentrated than in the Continental Europe. Hence, this underline, that even though rms in the Anglo-Saxon countries frequently use disproportionality instruments, they do this in a very di erent way than rms in the rest of Europe. In Continental Europe the disproportionality instruments are used to concentrate control in the hands of the largest owners, whereas in the UK and Ireland the typical ownership structure has a more even distribution of ownership and control even in the presence of these instruments. The rms that obey to the proportionality principle are located on the 45-degree line, which we have labeled the \proportionality line" in Figure 1. In all countries, except Sweden, these constitute a majority of all rms. However, the fractions of rms with a disproportional ownership structure vary a lot across countries and regions. Finland, Norway, Portugal, Spain and the UK have relatively few rms with a disproportional ownership structure whereas e.g. Germany and Sweden have a large share of rms with a disproportional ownership structure. 10

11 The distribution ofthe point mass in Figure 1 becomes important when we interpret our empirical results in Section 4 and 5. We can use Figure 1 to illustratethetwomeasuresofdisproportionalityweapplyin theeconometric analysis below; the absolute disproportionality de ned as votes minus cash ow and the relative disproportionality, which equals votes over cash ow. In the gure the absolute disproportionality is the vertical distance from a rm to the proportionality line. Similarly, the relative disproportionality measure is given by the slope on the line from the origin through the plot of votes against cash ow. Equipped with these two measures we can further conclude from Figure 1 that, rms in UK & Ireland and - to a lesser extent - rms in Scandinavia tend to have a smaller degree of disproportionality among rms that depart from a proportional ownership structure. Table 4 tabulates the insight of Figure 1 by showing the average and median ownership structures in the four regions. The largest owner in UK and Ireland tends to hold a smaller cash ow stake than the largest owner in Scandinavian countries, which again holds asigni cantly smaller stake than the largest owner in German speaking and Southern European countries. The same ranking is true for the largest owner's possession of votes across countries. Absolute disproportionality is smallest in the UK and Southern European countries and much larger in Scandinavian and German speaking countries. Notice that the largest owner in the median rm generally holds the same amount of votes and cash ow. 3.2 The Impact of Ownership and Control Concentration on Firm Value We estimate a cross section model of the average of the three yearly observations from 1996 to We do so because Faccio and Lang's data on the ownership structure in each country are not collected in the same year for all countries. Thus, we assume that the ownership structure is constant for the period 1996 to 1998 and focus on the variation between rms. 78 We include country speci c ' xed' e ects to control for country speci c rm invariant heterogeneity. This is in particular important if our basic model 7 For a number of rms we only have one or two yearly observations between 1996 and 1998 of the tangible assets, thus for those we use the average of the available observations. 8 Our results are robust to these assumptions, since they are con rmed when we estimate the cross section model based on yearly observations. 11

12 omits country speci c variables that are correlated with the explanatory variables such as the level of protection of outside investors, which in other studies has been shown a.o. to explain country variation in ownership and capital structures (La Porta et al. 1999). Table 5 analyzes the relationship between ownership concentration, disproportionality and rm value measured by Tobin's Q. In the left side we focus on the largest owner's ownership share whereas in the right side we use the ownership stakes of the whole group of large owners. In this and all other models we control for size, leverage (debt to assets ratio), asset tangibility, sales growth and industry e ects. Table 2 shows the mean and median of these variables on country level. In our sample of rms across Western Europe there seem to be a negative but highly insigni cant e ect of ownership concentration on rm value and rm performance. It is robust to looking at the group of large owners' ownership stake as shown in the right part of the table. Hence, on the aggregatelevel, we cannot conclude anysigni cant linearrelationship between rm value and ownership concentration. Claessens et al. nd a positive and signi cant e ect of ownership concentration in their sample of Asian rms. Naturally our result does not exclude that there could be a signi cant nonlinear relationship as documented by Morck et al. (1988). However, since there is a large literature on these questions we do not pursue it further here. In Table 5, Model 2 we include a dummy for whether a given rm has a proportional ownership structure, de ned as the absence of separatinginstruments and disproportional ownership structures. In both speci cations this proportionality dummy is positive and signi cant at a 5 pct. level. Hence, we see that having a proportional ownership structure generally increases rm value in publicly traded Western European rms. The e ect is economically large: On average rms with proportional ownership structures have app. 15 pct. higher rm value. This is consistent with the evidence for Asian rms provided by Claessens et al., who show that rm value decreases to the extent that ownership is separated from control. Our simple regression model has a satisfactory explanatory power, with an adjusted R 2 around 14 pct. Model 3and 4 look at thedegree of disproportionality. Absolute dispro- 12

13 portionality is signi cant both when we look at the single largest owner and the group of large owners. Again the marginal e ects are large. A 10 pct. increase in the wedge between cash ow and control of the largest owner decreases rm value with 3.2 pct. on average and 2.0 pct. for a similar increase for the group oflargest owners. If we instead use relative disproportionality we notice that the negative impact only is signi cant in the case where we analyze the group of large owners. Hence, our analysis shows that investors both look at whether rms have proportional ownership or not and in the latter case at the actual degree of disproportionality. We sum up in line with Claessens et al. (2002) a.o. that our data support the principle of proportionality since rms with an one-to-one relationship between cash ow and votes indeed seem to have higher valuation relative to rms with disproportional ownership structures. This result raises at least two interestingquestions: Does it matter how rms create disproportionality and are the e ects a ected by the legal environment a rm operates in? We pursue these questions next. 4 Dual class shares, pyramids, cross-ownershipand other mechanisms violating the principle of proportionality In the previous section we established that organizing ownership structures according to the principle of proportionality increases rm value. However, there are many ways to disobey this principle: rms can have dual class shares where thesvss are possessed by asmall group of controlling owners, there can be a chain of corporate ownership (pyramids) concentratingcontrol in the hands of a few ultimate owners, or there can be cross ownership between apair of corporations or there can be voting caps implyingthat none of the owners can have more than a few pct. of the votes. In this section we analyze how the value reduction from disproportional ownership structures depends on which instrument is used to create such disproportionality. There are a number of theoretical contributions analyzing dual class shares. In publicly traded rms with an active market for corporate control, most models have focused on the impact on control ghts of having disproportional ownership structures through the use of dual class shares 13

14 (see Grossmann and Hart 1988, Harris and Raviv 1988, and many others). In closely held rms Bennedsen and Wolfenzon (2000) show that having a one-share-one-voteruleincreases incentivestocollaboratewith other owners and this may increase rm value. A priori there is no reason to expect that di erent disproportionality mechanisms work through the same channels. When dual class shares are used, the ultimate owners have a direct contact with a given rm. On the other hand if chains of corporate ownership are used, the agents representing the ultimate owners may have di erent constituencies perhaps re ecting compromises between con icting interests on a higher level of the pyramid. There are few studies of pyramidal ownership (the main exemption is Almeida and Wolfenzon 2004 who analyze the dual question why pyramids arise and what determines the structure of a pyramid). Hence, theory is weak on providing a de nite answer to how disproportionality created through pyramids a ects rm value. Table 6 provides the rst empirical answer to this question. Model 5 explains rm value (Tobin's Q) as a function of the particular instrument used to create disproportionality and our standard control variables. Notice that the impact of the presence of dual class shares on rm value in our sample is negative, large and very signi cant even on a 1 pct. level. The e ect is economically large: The numbers corresponds to that rm value of an average European rm with dual class shares is app. 20 pct. smaller than an average rm with a proportional ownership structure. In Model 6 and 7 we look at the interaction e ects of dual class shares and the degree of disproportionality on rm value. Given that rms havedual class shares the degree of disproportionality has a signi cant negative impact on rm value when weusetheabsolutedisproportionalitymeasure. However, thenegative impact of the interaction of dual class shares and relative disproportionality is insigni cant. Thus, we have con rmed that dual class shares on average seem to destroy rm value in the European rms in our sample. This is consistent with the argument that ownership concentration is positive, since controlling owners internalize signi cant cash ow, but that dual class shares entrench owners since they possess signi cant control without internalizing the cash ow to a similar extent. At least our data suggest that this is the way it is 14

15 perceived by the external capital market. Similar to the value reduction ofdual class shares, pyramids have a negative and statistically signi cant e ect on rm value in our sample. The coe±cients aresmaller than those for dual class shares; however, theeconomic consequences are still large, since on average the rm value of European rms with pyramidal ownership structures is around 8 pct. lower than for rms with proportional ownership structures. The interaction e ect between our two measures of disproportionality and pyramidal structure is generally insigni cant. Dual class shares and pyramids, thus, seem to have a di erent impact on rm value. Usinga F-test we strongly reject the null hypothesis that the e ect of dual class shares and pyramids are identical. Hence, the two coef- cients are both economically and statistically di erent: Dual class shares have a signi cantly stronger negative e ect on rm value than pyramids. Cross-holdings have a positive e ect on rm value, which is signi cant on a 10 pct. level. Even though the impact is much less statistically clear than for the two previous instruments, there is some indication that cross holding on average can increase rm value. The interaction e ects between cross-holdings and our two measures of disproportionality vary in signs and are highly insigni cant. Finally, others instruments have a negative e ect on rm value, but this e ect is highly insigni cant, as are the interaction e ects with the two disproportionality measures. Note by looking at Table 4, that rms with cross holdings and other instruments are few in numbers. If weanalyze thegroup of large owners theseresults donot change. The economicand statistically impact ofdual class shares and pyramids arevery similar to the analysis of the single largest owner. Again we can reject the null that the two instruments have the same impact on a 1 pct. level. The e ect of cross holding is still positive and signi cant on a 10 pct. level. Other instruments have an insigni cant e ect. In sum this section has shown that the type of instrument through which disproportionality is created is indeed important for the e ect of disproportionality on rm value. Dual class shares and pyramids have a large negative impact on rm value. In addition dual class shares have a signi cantly larger negative e ect on rm valuethan pyramids. This insightdelivers important policy implications, which we will return to in the nal section below. 15

16 5 Investor protection and the principle of proportionality Today there are a large number of studies analyzing the impact of investor protection on various topics within corporate nance and rm organization (see survey by La Porta et al. 2000). In this section we analyze if investor protection alsoplays arole in explaining the impact of particular disproportionality instruments on rm value. Tomotivate this analysis, thinkaboutthe cost and bene ts ofdual class shares. The bene t of dual class shares is that it makes it easier for controlling owners to be informed and involved in management and it provides them with more power to maketheir arguments heard by the daily management. Thus, dual class shares can reduce the opportunistic behavior of the management in rms where there are a signi cant separation of ownership and control. The cost of dual class shares is the increased opportunities for controlling shareholders to extract rent from non-controlling shareholders. This may induce management tooperate the rm in away that caters more tothe controlling than tothe non-controlling owners or it may induce a disproportional distribution of surplus from otherwise e±cient rm operation. Good investor protection reduces ceteris paribus the management's ability to engage in opportunistic behavior contrary to the interest of the owners. Thus, in countries with a high level of investor protection there is a reduced bene t of disproportional ownership structures. Similarly, the controllingowners' ability toextract rent from non-controllingownersincreases in countries with low investor protection. This is most evident in transition economies, where insiders' insu±cient ability to commit not to divert outsiders' investments has been a serious obstacle for increased foreign investment in these economies (see Shleifer and Vishny 1997 a.o.). However, even in rms where non-controlling shareholders are well protected there will always be a legal scope for controlling shareholders to expropriate rent to themselves, for instance through pecuniary and/or non-pecuniary bene ts or by in uencing business decisions in a way that promote their own interests, e.g. doing business with rms wherethe controlling owners have a speci c interest. Weconcludethatsince both costs and bene tsofdual class shares are a ected by the degree of investor protection in a given country, 16

17 it is not possible to theoretically determine the exact aggregate impact on the value reduction of dual class shares. et al. We use the two standard measures of investor protection from La Porta (1998); anti-director rights and creditor rights. The anti-director rights index summarize 6 speci c decision rights granted to minority shareholders by the corporate law, whereas the creditor right index measures the presence of 5 protective rights granted to creditors in case of bankruptcy. Both indencies are increasing in the level of legal investor protecting. It is worth emphasizing that we expect the interaction of anti-director rights with thedisproportionality measurestobemost important and that the two indices are positively correlated. In Table 7 we have interacted the two measures of investor protection with the degree of disproportionality. To simplify the presentation of the results wedonot report the control variables and industry e ects in Table 7 and 8. Hence, weuse the same control variables as in Table 5and 6. We start by including the interaction of anti-director rights with the proportionality dummy. The interaction e ect is positive and highly signi cant whereas the proportionality dummy becomes negative and marginally insigni cant. Thus, the e ect of proportional ownership structures is increasing in the level of investor protection, but insigni cant in countries with low levels of investorprotection. A simplef-testofthenete ect ofaproportional ownership structure reveals that the positive e ect kicks in when theanti-director rights score is 3 or higher, whereas the e ect is insigni cant for scores below 3 9. We obtain similar results when using the creditor rights index, but Model 2(c) shows that thisismainly dueto the positive correlation between anti-director and creditor rights. When we include both terms only the interaction with the anti-director index is signi cant. Further, these results are robust towards our measure of disproportionality, since we get identical results when we interact the investor protection indices with the absolute degree of disproportionality. The negative e ect of disproportional ownership structures isincreasing in thelevel of investor protection, butthe e ect is only present for countries with a relatively high level of investor protec- 9 The F-test of the net e ect of proportional ownership structures with an anti-director rights score of 2 yields a F-value of 0.61 which is grossly insigni cant, whereas the F-value when the score equals 3 is 7.20, which is signi cant at the 1 percent level. 17

18 tion. For countries with a high level of investor protection (UK, Ireland and Scandinavia) the negative e ect of disproportional ownership structures are economically important; rms with a disproportional ownership structure and anti-director rights score of 5 have a 20 pct. lower rm value than rms with a proportional one. Table 8 extents the analysis of the e ect of disproportionality into the use of particular instruments. Again we have for expositional convenience skipped the presentation of the control variables, which are identical to the ones used throughout the analysis. We start by including the interaction of the two investor protection indices with the four groups of instruments separately in Model 5(a) and (b). In the top of Table 8 we notice that the direct e ect of dual class shares and pyramids disappear. All instruments enter positively but the e ects are highly insigni cant. The interaction of dual class shares and anti-director rights is negative and signi cant at the 1 percent level. Similarly, the negative e ect of pyramids on rm value is increasing in the level of investor protection. The net e ect of dual class shares is negativeand signi cantfor investorprotection scoresat 2or above, whereas the net e ect of pyramids becomes signi cantly negative when the score is 3 or higher. The coe±cient on the interaction e ect of dual class shares is almost twice as large as the coe±cient on pyramids. We use a F-test to test whether the net impact of dual class shares and pyramids is statistically di erent. We nd that the negative e ect of dual class shares is signi cantly higher that the e ect of pyramids in countries with ahigh level of investor protection. 10 Model 5(c) con rms the insight from Table 7, that the anti-director rights index is the relevant measure for the interaction of the level of investor protection and disproportional ownership structures. The residual group of rms with cross ownership and other disproportionality measures does not seem have a strong negative impact on rm value, since the net e ect in Table 8 is grossly insigni cant. As mentioned previously, there are few rms and many di erent instruments in these residual groups, hence, we will not put to much emphasize on these results except from noticing that they are consistent with our main hypothesis about in- 10 The F-test of a di erent net e ect of dual class shares and pyramids yields an F-value of 1.23 at an anti-director rights score of 2, whereas the F-value equals 4.12 when the score equals 3, which is signi cant at the 5 percent level. Increasing the level of investor protection will expand this di erence further both statistically and economically. 18

19 vestor protection and disproportionality instruments being substitutes. Our analysis provides evidence for that proportional ownership structures increases rm value only in countries where the degree of legal investor protection is high. Hence, our data show that investor protection and disproportional ownership structures are substitutes: If there exist inadequateprotection ofoutside investorsthe bene ts ofdisproportional ownership structures, i.e. the reduced agency problem between owners and managers, out weights the costs, i.e. the increased agency problem between controlling and non-controlling owners. If investor protection on the other hand is strong, then there seem to be less bene t of violating the principle of proportionality and investors therefore require a larger discount on such rms. This nding has important policy implications for the harmonization of European company laws, since countries in Northern Europe (UK, Ireland and Scandinavia) generally have a higher level of investor protection than countries in Central and Southern Europe (See Table 2). To substantiate this important aspect of our ndings we further divided our 14 countries into four regions: UK & Ireland, the Scandinavian countries (Denmark, Finland, Norway and Sweden), Southern European countries (Belgium, France, Italy, Portugal and Spain) and countries inspired by German legal system (Germany, Austria and Switzerland). According to La Porta et al. there are many similarities within each of these four regions with respect to the content and implementation of relevant corporate and accounting laws, i.e. these four regions represent di erent legal systems. La Porta et al. emphasize that these four regions have very di erent levels of investor protection with outside owners in UK and Ireland being best protected followed by outside owners in Scandinavia. Outside owners in Southern Europe are less protected than in Scandinavia but in general slightly better protected than outside owners in countries with a German legal origin. Supportive of the evidence reported above we nd the that negative e ect of disproportional ownership structures is signi cant in UK, Ireland and Scandinavia, whereas it is insigni cant in Southern European countries and countries with a German legal system. 11 This patterns holds for all 11 To save space we have decided not to report these results, but they are available from the authors upon request. 19

20 our measures of disporportionality and for the decomposition into instruments that created this disproportionality. Consistently, we also nd that dual class shares have a signi cantly larger negative e ect on rm value in Northern Europe. Thus, on one side we nd statistically and economically important e ects on rm value in Northern Europe, but no e ect in Central and Southern Europe. We conclude thissection byemphasizingthat regional and cross-country variations are important. Dual class shares and pyramids destroy rm value in Northern Europe but not necessarily so in the Central and Southern European regionswhereinvestor protection traditionally islower. Thissupport the argument that dual class shares and pyramids (and possible also other disproportionality instruments) substitute good investor protection in countries with low protection of outside investors. As we will discuss in the nal section, this have important policy implications for the harmonization of European capital markets. 6 The impact of disproportionality instruments on earnings performance Our analysis has so far focused on the impact of disproportionality on rm value across instruments and legal regimes. It is an interesting question if disproportionality in general and dual class shares in particular only has an e ect on rm value or if it also a ects rms' earnings performance. In this section we, therefore, studythe impactofdisproportionality on the earnings performance of the rms measured by return on assets. Table 9 shows the impact of our various measures of disproportionality on earnings performance. In general the e ect of disproportional ownership structures disappears when we use return on assets as our endogenous variable. The e ects are close to zero and highly insigni cant. However, it is important to notice that our performance models based on RoA have very little explanatory power. The R 2 is only around 3 pct. in our performance analysis, compared to R 2 around 14 pct. when we analyze Tobin's Q. Hence, it is di±cult to tell if the lack of signi cance is due to a real lack of e ects from ownership structures to performance or whether it is coursed by our weak performance model. 20

21 Table 10 repeats the analysis where we separate the e ects of di erent instruments by using earnings performance as the endogenous variable. In general the e ects, which are very clear when we use Tobin's Q as endogenousvariable, become highlyinsigni cant in these models. Thus, wecannot provide any evidence of dual class shares or pyramids having a signi cant negative impact on earnings performance in the Western European corporations. Table 9 and 10 reveal that it does matter which measure we use for the prosperity of the rms. In general disproportionality and in particular dual class shares seem to have a large and signi cant negative impact on rm value measured through Tobin'sQ. Contrary tothis, noneofour modelshas shown any signi cant impact on rm performance measured through return on assets. Wenow discuss anumber ofpotential arguments explainingthese ndings. First, as mentioned above this could be due to that the two types of models have di erent quality. In general we have a higher explanatory power in our rm value models than in our performance models. However, the explanatory power in the earnings performance models varies a great deal and even in the best models we do not nd any signi cant e ects of disproportionality or its instruments. Second, a possible explanation could be that they re ect a valuation bias if we misrepresent the 'true' value of the controlling ownership blocks. We have calculated Tobin's Q on the basis of the stock price from trades of minority holdings. Thus, if the controlling blocks are traded at a premium we this provides a negative bias on the total market value of the rm. However, our rst counter argument against this concern is that we should expect the same e ect for any given concentration of control. However, we found above that the di erent mechanisms gave statistically di erent e ects on rm value. In particular both at the European level and at the disaggregated level for countries with high investor protection, dual class shares have a statistically larger negative impact on rm value than pyramidal ownership structures. Our second counter argument against the claim that our valuation results are driven by a valuation bias focuses on dual class shares. When rms have dual class shares there is generally attached a voting premium to the 21