NBER WORKING PAPER SERIES INTRACOMPANY GOVERNANCE AND INNOVATION Sharon Belenzon Tomer Berkovitz Patrick Bolton Working Paper 15304 http://www.nber.org/papers/w15304 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138 August 2009 We thank Liat Oren for invaluable assistance with the programming of the ownership algorithm and Hadar Gafni for excellent research assistance. We also thank Luca Enriques, Daniel Ferreira, Ronald Gilson, Joshua Lerner, Randall Morck, Daniel Paravisini, Katharina Pistor, David Robinson, John Van Reenen and Daniel Wolfenzon for helpful comments. All remaining errors are our own. The views expressed herein are those of the author(s) and do not necessarily reflect the views of the National Bureau of Economic Research. �� 2009 by Sharon Belenzon, Tomer Berkovitz, and Patrick Bolton. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including �� notice, is given to the source.

Intracompany Governance and Innovation Sharon Belenzon, Tomer Berkovitz, and Patrick Bolton NBER Working Paper No. 15304 August 2009 JEL No. O16,O31,O32 ABSTRACT This paper examines the relation between ownership, corporate form, and innovation for a cross-section of private and publicly traded innovating firms in the US and 15 European countries. A striking novel observation emerges from our analysis: while most innovating firms in the US are publicly traded conglomerates, a substantial fraction of innovation is concentrated in private firms and in business groups in continental European countries. We find virtually no variation across US industries in the corporate form of innovating firms, but a substantial variation across industries in continental European countries, where business groups tend to be concentrated in industries with a slower and more fundamental innovation cycle and where intellectual protection of innovators seems to be of paramount importance. Our findings suggest that innovative companies choose the corporate form most conducive to R&D, as predicted by the Coasian view of how firms form. This is especially true in Europe, where there are fewer regulatory hurdles to the formation of business groups and hybrid corporate forms. It is less the case in the US, where conglomerates are generally favored. Sharon Belenzon Duke University Fuqua School of Business 1 Towerview Drive, Durham, NC United States sharon.belenzon@duke.edu Tomer Berkovitz Columbia University Graduate School of Business New York, NY 10027 tb2122@columbia.edu Patrick Bolton Columbia Business School 804 Uris Hall New York, NY 10027 and NBER pb2208@columbia.edu

1. Introduction Which corporate forms are most propitious to innovation? Is it standalone rms nanced by Venture Capital (VC)? Is it conglomerate incubators, or looser business-group and corporate alliance structures? While GE, Hewlett-Packard, IBM, DuPont or 3M are often mentioned as examples of highly innovative conglomerate rms in the U.S., there have also been many prominent examples of VC-backed standalones in recent years, such as Amazon, Netscape, or Google. Although VC-backed innovation has been less important in Europe, innovative activity in conglomerates such as Philips or Nokia is highly visible, and business-groups in the vein of Ericsson, Zeiss, Alstom and Novartis are also renowned innovators. Economists and business scholars have pointed to the advantages of conglomerates in fund- ing R&D projects with cheaper internally generated funds (Gertner, Scharfstein and Stein, 1994 and Stein, 1997) while at the same time emphasizing the dark side of internal capital markets, in terms of reduced nancial discipline for poorly performing investments (Scharf- stein and Stein, 2000). Seru (2007) nds evidence consistent with this dark side of internal capital markets: publicly traded U.S. conglomerates with above-average reallocation of funds across divisions are less productive innovators than comparable stand-alone rms. Similarly, Guedj and Scharfstein (2004) compare clinical trials in the biopharmaceutical industry and nd that big pharmaceuticals rms engaged in cancer research tend to initiate too many studies but are quicker to terminate unpromising research than smaller (stand-alone) biotech rms. In contrast, Belenzon and Berkovitz (2008a) nd European evidence that innovation tends to be concentrated in business-groups and that business-group a�� liates tend to engage in more innovation than comparable stand-alone rms. While these studies do not necessarily o���er contradictory evidence - as business-groups have a di���erent governance structure than fully integrated conglomerates - they do suggest that the link between governance and innovation is likely to be complex and that more systematic evidence is required to have a better understanding on how governance a���ects innovation. 2

In this paper we o���er a more complete picture of the link between governance and innovation by looking at a cross-section of innovating rms in the U.S. and in 15 European countries. In addition, we provide a broad conceptual framework that allows us to compare the relative strengths of standalone rms, conglomerates and business-groups in fostering innovation. Building on existing theories of conglomerate internal capital markets, we emphasize the relative strengths of conglomerates in terms of cheaper internal R&D funding, better winner- picking, and lower stigma of failure, versus the stronger intellectual property rights for innova- tors in stand-alone rms.1 Business-groups occupy a middle ground compared with conglomer- ates and standalones. These are fuzzier and more diverse organizational forms with potentially complex controlling stakes holding together a set of independently incorporated companies. Business groups may be entirely composed of privately held entities or may combine both privately held and publicly traded companies. Business groups may take the form of pyrami- dal structures, where a single controlling company has direct or indirect controlling stakes in multiple subsidiary companies, or business alliances, where the companies in the alliance are connected through interlocking stakes. Business groups are rare in the U.S. and somewhat rare in the U.K., but they are om- nipresent elsewhere in the world. Business groups have been associated with poor minority investor protection and ine�� cient rent extraction by controlling shareholders (see Johnson, La Porta, Lopez-de-Silanes, and Shleifer, 2000, and Bertrand, Mehta, and Mullainathan, 2002). They are often seen as a consequence of poor legal investor protections in emerging market countries, and also to some extent in developed civil lawcountries. Accordingly, we model business groups as in Almeida and Wolfenzon (2006) and allow for ine�� cient diversion of funds by controlling shareholders from rms a�� liated with a business group. Due to this ine�� cient diversion business groups involve greater ine�� ciencies than conglomerates or standalones, other things equal. However, business groups also have strengths in terms of cheaper internal R&D funding and a lower stigma of failure relative to stand-alone rms, as well as strengths in terms 1We mainly build on Gertner, Scharfstein and Stein (1994), Stein (1997) and Scharfstein and Stein (2000). 3

of stronger intellectual property rights for innovators relative to conglomerates. An important basic premise of our analysis is that innovating rms tend to set up their corporate governance to implement an e�� cient environment to foster innovation. Thus, when the observed choice of corporate form is a conglomerate or a business group we assume that this is the revealed preferred choice of organization. In other words, a basic premise of our analysis is a Coasian perspective on rm organization. Thus, to the extent that a particular corporate form is especially well suited for innovative activities, we expect this corporate form to be more prevalent among innovative rms in a given industry than among non-innovating rms. We base our analysis on a comprehensive data-set of both private and publicly traded American and European rms, and match all corporate patents granted by the United States Patent and Trademark O�� ce (USPTO) and the European Patent O�� ce (EPO) to these rms. We also single out rms that publish their research in academic journals. We have thus identied about 64.000 rms that hold at least one patent from the EPO or USPTO, or have published at least one scientic article in a science journal. Of these 64.000 rms, about 60% are American, 11% German, 8% British, 4% French, and 5% are Italian. We matched 880.000 USPTO, 615.000 EPO patents, and 205.000 scientic publications. 75% of USPTO patents and close to 40% of EPO patents are held by American corporation. Germany appears to be the most innovative European country while holding 12% and 20% of USPTO and EPO patents, respectively. For scientic publications, about 70% of the articles are published by U.S. corporations, 10% German, and 3% French and British. Several surprising ndings emerge from our analysis. First and foremost, while the overall distribution of USPTO patents by ownership structure in the whole sample is 18% standalones, 74% conglomerates, and 8% business-group a�� liates, in the U.S. this distribution is skewed towards standalones and conglomerates, with 20% and 76% respectively of patents falling in these categories, and only 4% of patents being held by business group a�� liates. In Europe, on the other hand, 20% of the patents are held by business-group a�� liates, and 11% and 68% 4

respectively are held by standalones and conglomerates. A similar pattern emerges for EPO patents and scientic publications. There is also a wide variation in organizational form across European countries. While Germany appears to resemble the U.S. in terms of concentration of innovation in conglomerates (80%), French and Italian rms are clearly skewed towards business-group a�� liates, with only 35% and 22% respectively of their patents being held by conglomerates. Second, we distinguish between business-group a�� liates that are wholly-owned by their parent company, which are more likely to be fully integrated, and a�� liates that have minority shareholders. The presence of minority shareholders in an a�� liate company establishes a sepa- ration between the a�� liate and its parent company in terms of retained prots, patent holdings, and resource reallocation. Business groups with partially owned a�� liates are therefore more decentralized corporate structures. Remarkably, in France 20% of innovation takes place in a�� liates with minority shareholders, while only 1% of innovation takes place in such a�� liates in the U.S. and Great Britain. Third, the U.S. distribution of innovating rms is also heavily skewed towards publicly traded rms. Indeed, the fraction of patents held by publicly traded American corporations is above 60%, while this fraction is only about 40% in Europe. In some large European countries this picture is even more extreme. In France, only 9% and 18% of USPTO and EPO patents, respectively, are held by publicly traded rms, while in Italy these gures drop to as low as 3% and 7% respectively. Also, publicly traded innovating rms in Europe are more likely to belong to a business group, with more than 40% of innovating listed rms belonging to a business- group in France and Germany, compared with only 2% and 1% respectively in the U.S. and Great Britain. Fourth, we nd a substantial variation in organization structure across industries. Business groups tend to be concentrated in industries where innovation takes a longer time, is more un- certain, but has a higher payo��� when it succeeds (such as pharmaceuticals and biotechnology). These innovative activities may be more vulnerable to hold-up problems and require greater 5

protection of the intellectual property rights of the innovator. In contrast, conglomerates are more prevalent in industries with rapid, incremental, innovations (such as computer hardware and telecommunications) where the ability of conglomerates to identify the relevant innova- tion and to quickly redeploy assets may give this organizational form an edge over business groups. We nd that European rms are strongly skewed towards business groups in the latter industries, while U.S. innovating rms take the conglomerate form in more or less the same proportion across all industries. Our ndings are broadly consistent with a Coasian view on rm organization form, which is to say that rms choose the corporate form that is best suited for fostering innovative activ- ities. The alternative view in the literature is that organization form is chosen by entrenched managers, or controlling shareholders, to suit their best interests at the expense of minority shareholders and overall rm e�� ciency. By this latter view it is not obvious that one should see any systematic di���erence in organization form in a given industry and country across inno- vative and non-innovative rms. We examine whether innovative rms are disproportionately represented in a given organization category, using sales of about 3.5 million rms as our benchmark. We nd that innovation is disproportionately concentrated in business-groups in the pharmaceuticals industry, and disproportionately concentrated in conglomerates in the telecommunications industry. This pattern is especially true in Europe, where there are fewer regulatory hurdles to the formation of business groups and hybrid corporate forms. It is not the case in the U.S., where conglomerates are generally preferred. The remainder of the paper is organized as follows. Section 2 presents the theoretical framework, section 3 overviews the data, section 4 reports the main empirical ndings and section 5 concludes. 2. A Simple Model We set up a model with at least two business units, which can be stand-alone entities, divisions in a conglomerate rm, or units linked in a business group, with one parent company owning 6

a controlling block in a separately incorporated rm. 1.Technological Assumptions There are N 2 business units with liquid assets Wi 0 and a new R&D project. The R&D project has xed costs f 0 at date 1 and a probability of success (a) at date 2. The probability of success depends on how well the project is managed, or more formally, on which action a 2 f0 1g the projects manager chooses. The projects value to the business unit conditional on success is given by v(a) 0. A successful R&D project may also generate spillovers to other business units worth s(a) 0. When the research is unsuccessful (with probability (1 (a))) it yields no nancial return. The project brings expected private benets or costs to the managerial team undertaking it. The benets may come in the form of improved human capital and knowledge, better future career prospects, enhanced reputation in the event of success. The costs may be the stigma of failure when the research is not successful. Without loss of generality we normalize the net benets arising from simply undertaking the project to zero. We denote by b 0 the private benets that come with success and by 0 the private costs associated with the stigma of failure. We shall assume that: b1 (1 1) b0 (1 0) so that the expected net private benets to the manager are highest when the manager chooses a = 0. Note that the latter inequality reduces to (b + )( 0 1) 0 and therefore holds if and only if 1 0 0: We also assume that (v1 + b) 1 (1 1) f 0 (v0 + b) 0 (1 0) f These inequalities together with 0 mean that the innovative activity is equivalent to taking a higher risk, higher expected payo��� actions. In other words, R&D activities are high 7

risk, but also high reward activities. The above inequalities also mean that the R&D activity is worth undertaking if and only if the manager can be induced to choose the high risk, high reward, action a = 1. To simplify the analysis we also restrict attention to parameter values such that b1 (1 1). The managers expected private costs from the stigma of failure when he chooses a = 1 are then greater than the private benets from a successful innovation. Under this assumption the manager has to receive some form of nancial incentive to be willing to undertake the risky innovative activity. 2.Governance Assumptions and Firm boundaries We distinguish between three main types of rms: standalones, conglomerates (with wholly- owned a�� liates) and business-groups with a�� liates owned in part by minority shareholders. A standalone rm is essentially a rm with a single business unit. Such a rm is run by a manager with an equity stake su�� cient to align incentives and is monitored by a large blockholder, venture capitalist (VC), or creditor, who plays the role of an active monitor as in Holmstrom and Tirole (1994). A standalone rm is viable only if it is able to raise su�� cient funds to nance the research, while preserving the managers incentives to conduct the research and the monitors incentive to monitor. We assume that only a fraction of the rms value v(a) can be pledged to raise external funds. The parameter 2 (0 1) represents the fraction of R&D cash-ow that is capitalizable by the rm. It provides a simple measure of the nancial development of a country. The higher is the more nancially developed a country is. We assume that capital markets are perfectly competitive for the pledgeable part of rm value and for simplicity we normalize the equilibrium required risk-adjusted return to zero. Following Holmstrom and Tirole (1994) we also assume that the timing of the contractual relation between investors in the rm, the VC monitor, and the entrepreneur is as follows: 1. following an injection of funds in the rm by investors, the monitor moves rst by choosing 8

whether to engage in costly monitoring or not 2. the R&D project manager moves second by choosing whether to take action a = 0 or a = 1, having observed whether the active monitor is monitoring or not 3. the R&D project either succeeds or fails 4. if it succeeds the proceeds from the project are distributed to all investors and the man- ager. Suppose that the active monitor decides to monitor the manager and incurs the monitoring cost m 0. Then as in Holmstrom and Tirole (1994), we assume that the managers net private benets are lowered by a fraction (1 ), where 1. The manager must be incentivized to choose a = 1. If his nancial stake in the project, , is large enough he will choose a = 1 even though his private benets are higher for a = 0. Formally, the managers stake must be such that the following incentive compatibility constraint holds: (v1 + b) 1 (1 1) (v0 + b) 0 (1 0) or V + (b + ) 0: where V (v1 1 v0 0) and = 1 0. 2 If there is no monitoring, on the other hand, the incentive compatibility constraint is: 3 V + (b + ) 0: Monitoring is benecial if two conditions are satised. First, it must be the case that the rm cannot get su�� cient nancing in the absence of monitoring. That is, the rm should not be able to invest in R&D, by using all its internal funds Wi and raising (f Wi) from outside investors, while at the same time granting su�� cient 2Note that under our assumptions V is strictly positive and is strictly negative. 3Since 0 the incentive constraint is clearly harder to satisfy with no monitoring than under monitoring. 9

nancial incentives to the entrepreneur to choose a = 1 even though there is no monitoring. Therefore, a rst condition is that: (1 )v1 1 f Wi for any such that V + (b + ) 0 or [1 + (b + ) V ]v1 1 f Wi: Second, the rm must be able to get nancing when there is monitoring. The active monitor has an incentive to monitor and incur the monitoring cost m as long as his stake in the rms prot is large enough that V m. In exchange for the stake = m V , which just ensures that he will monitor, the active monitor is willing to invest an amount equal to his expected equilibrium net return: v1 1 m. Therefore, the standalone rm is viable under monitoring if (1 )v1 1 f Wi v1 1 + m or, substituting for and if: (1 + (b + ) V m V )v1 1 f Wi m( v1 1 V 1): Rearranging, this condition reduces to: [1 + (b + ) V ]v1 1 f Wi + m: Clearly, for large enough, and m and small enough this condition will be satised given our assumption that v1 1 f: A conglomerate rm operates n 2 business units that are wholly-owned subsidiaries. It has corporate headquarters that make investment and R&D decisions for all divisions. A key 10

feature of conglomerate rms is that prots of subsidiaries are channeled to headquarters, who report consolidated earnings to the rms shareholders. Divisional managers typically cannot be rewarded based on their divisions prot, and their incentive pay-component is generally based only on conglomerate prots and stock price. Moreover, divisional managers dont own the intellectual property they create, unless they negotiate a contract which explicitly gives them the right to any innovation they create. 4 On the other hand, divisional managers may be closely monitored by headquarters. Another advantage of undertaking research in a conglomerate is that failure can be hidden to some extent inside the rm, so that there is a lower stigma of failure. Conglomerates also develop winner picking skills, and benet from R&D spillovers on other divisions. For our formal analysis we assume that conglomerate rms are composed of only two divi- sions, A and B. We also simplify the governance of conglomerate rms to the CEOs role in picking winner R&D projects and to the CEOs active monitoring of division managers. We model winner picking by assuming that a project selected by headquarters has a probability of success (a), where 1. We model CEO monitoring by letting the CEO monitor division managers j = A B at cost and thereby lowering division managersprivate benets by a fraction (1 ). If the divisional managers research is successful, however, it is owned by the rm. Thus, division managers have lower powered incentives to pursue R&D in conglomerates. 5 In our model, a divisional manager can only extract a small share of the direct value of the innovation by, for example, threatening to leave and pursue his research at another rm. Thus, the divisional managers incentive constraint for choosing action a = 1 in a conglomerate is: (v1 + b) 1 + (1 1) (v0 + b) 0 + (1 0) or V + (b + ) 0: 4Such contracts are not common and even when a contract grants intellectual property rights to the innovator it is di�� cult to enforce. 5Divisional managers may get a share of conglomerate prots, but due to moral hazard in teams problems, they will have lower powered incentives than if they received a share of their own divisions prots. For simplicity, we ignore the nancial incentives from shares in conglomerate prots. 11

Finally, conglomerates have the nancing advantage of drawing on the sum of liquid assets of both divisions, (WA + WB), to fund a divisions R&D project. To summarize, in our model a conglomerate picks the most promising R&D project from the divisions, thus ensuring a probability of success of 1 in equilibrium. It nances the project entirely with its own funds (WA +WB) and it incurs monitoring costs . Thus, the conglomerates expected return is given by: (1 )v1 1 f + WA + WB provided that it can induce the divisional manager whose R&D project is picked to choose action a = 1. Should be so low that (b + ) V then the division manager will choose a = 0, in which case we assume that the conglomerate will not pick any R&D projects. A business-group of n 2 rms is similar to a conglomerate rm. It has a controlling hub akin to corporate headquarters and multiple divisions. The key di���erences with a conglomerate are that: i) the business units are not wholly owned subsidiaries, but rather are independently incorporated companies controlled by the groups majority shareholders through a controlling stake ii) earnings of business units are not consolidated, and ownership of intellectual property remains with the innovating business unit iii) R&D project selection is more decentralized, so that the groups hub plays less of a winner picking role. The group mainly plays the role of an internal capital market to fund R&D investments. We denote by ��� the ownership share of the group and (1 ���) the share owned by minority shareholders. An important potential source of ine�� ciency in business groups is the separation of ownership and control of business units. This separation can give rise to ine�� cient diversion of business unit earnings by the controlling shareholders. We model this diversion as in Almeida and Wolfenzon (2006) by allowing the controlling shareholders to divertd of earnings from 12

a unit in which they have a stake ��� at a deadweight cost of k(d) = 1 2 d2. Group owners then choose d to maximize: ���(v d) + d 1 2 d2 where denotes the business units earnings. As is easily established, the optimal amount of diversion and the deadweight cost of diversion are then respectively: d(��� ) = 1 ��� and k(d(��� )) = (1 ���)2 2 : The earnings retained in the rmv ( 1 ���)can be shared between the division manager and all the owners. Thus division managers can be incentivized as in standalones based on the non-diverted earnings. Moreover, the business unit managers private benets can be reduced through monitoring by the groups controlling owners. As the literature on business groups has emphasized, the diversion of earnings by controlling shareholders reduces the e�� ciency of business groups relative to stand-alone rms or conglom- erates. 6 The pervasiveness of business groups around the world (with the important exceptions of the U.S. and Great Britain) suggests, however, that there is also a countervailing e�� ciency improving role of business groups. 7 We argue that the e�� ciency gain of business groups is due to the stronger intellectual protection given to innovators in business units a�� liated to business groups, who can retain (partial) ownership of their patents. To summarize, business groups combine the R&D incentive benets of stand-alone rms with the internal capital market benets of conglomerates. Given equal R&D incentives, however, 6There can also be diversion of funds from conglomerates with separation of ownership and control. For simplicity we ignore this form of diversion in our model. 7While, some commentators simply explain the existence of business groups as an ine�� cient outcome caused by managerial entrenchment, others point to the value adding role of business groups as providers of a substitute source of funding for investment in environments where external capital markets are ine�� cient due to inadequate investor protection. Thus, the prevalence of business groups in emerging market countries is seen as a necessary but transitory consequence of the underdevelopment of securities markets in these countries. However, as our evidence highlights, business groups are also prevalent in advanced countries, with developed securities markets and strong investor protections. Moreover, as we show, these groups tend to be both very competitive and innovative. 13

business groups are less e�� cient than conglomerates, as they involve both ine�� cient diversion of funds and less e���ective winner-picking. 3.Assumptions on external legal environment Besides (the fraction of the value of innovations that is capitalizable) we introduce three other key parameters to capture the external legal environment: an intercompany dividend tax, the R&D project managers bargaining power, and the stigma cost of failure. As we argue in the next section, the main comparative advantages of the U.S. relative to the EU are that U.S. securities markets free up a higher and . Moreover, U.S. entrepreneurial culture and bankruptcy laws also result in a lower . Against these comparative advantages the EUs main advantage is a lower . 3. Model Analysis We divide the model analysis into two parts. We begin by considering the main tradeo���s in organizational form keeping the external legal environment xed. In a second step we do a EU-US comparative analysis by analyzing how changes in the parameters ( ) a���ect the equilibrium distribution of R&D across organizational forms. 3.1. Main tradeo���s within a given legal environment We consider in turn the choice between undertaking R&D in a stand-alone unit or inside a conglomerate rm, the pros and cons of a business-group and a conglomerate structure, and nally the choice between stand-alone and a�� liation to a business-group. 3.1.1. Stand-alone rm vs. Conglomerate: the costs and benets of internal capital markets Consider the funding of R&D projects in a conglomerate with two business units, A and B. To facilitate comparisons with a stand-alone rm we assume that only one of the two units has a potential research project, say division A. We also assume that the conglomerate has 14