Corporate Diversi cation: Good for Some Bad for Others

Transcription

1 Corporate Diversi cation: Good for Some Bad for Others Felipe Balmaceda 1 Centro de Economía Aplicada University of Chile 2 December 26, I would like to thank participants to the LACEA 99 conference and Dartmouth regular seminar for comments on an earlier version, and seminars at University of Chile and Ilades-Georgetown. 2 Republica 701, Santiago, Chile. fbalmace@dii.uchile.cl

2 Abstract In this paper a model based on con icts of interest between shareholders, the CEO and divisional managers is developed to explain why corporate diversi cation is good for some rms and bad for others. It is shown that when the decision to diversify is endogenous, whether diversi cation destroys value depends on the severity of con icts of interests and the complementarities across divisions and not, as usual in the literature that explains value-decreasing diversi cation, on the e ciency of internal capital markets.

3 1 Introduction Corporate diversi cation is a pervasive phenomenon despite the fact that is widely believed to be ine cient. The empirical evidence documents the existence of a diversi cation discount; i.e., the empirical observation that on average diversi ed rms trade at a discount relative to a portfolio of stand-alone rms in the same business segments, 1 yetaround40%ofthemtradeatapremium. 2 The reason why diversi cation is good for some rms and bad for others is still an open question. There is a recent and large body of literature that provides an answer to this question based on the e ciency of internal capital markets. In particular, this literature argues, contrary to Williamsom (1975), that there are circumstances under which internal capital markets allocate funds across divisions in an ine cient way thereby reducing rms overall value. 3 The goal of this paper is to provide an explanation for why diversi cation is good for some rms and bad for others that does not rely on the e ciency of an internal capital market. I consider two units (divisions) that can be operated either as stand-alone rms or as an integrated rm. It is shown that agency problems between the CEO, shareholders and divisional managers induce the CEO, under certain parameter con guration, to choose to integrate two units when the valuemaximizing strategy for them is to stay focused and for others to choose the value-maximizing strategy. Thus, diversi cation is sometimes value-reducing and others value-increasing. The model in this paper is based on agency problems arising from the existence of managerial private bene ts and con icts of interest between the CEO, shareholders and divisional managers. The CEO s private bene ts are a fraction of the rm s total return when his preferred projects are implemented, while a divisional manager s private bene t represents a fraction of his division s return only when his preferred projects is implemented, where the total return is the sum of divisional 1 See, e.g., Berger and Ofek (1995), Lang and Stultz (1994), Rajan, Servaes and Zingales, (2000), and Servaes (1996). 2 The event study evidence also reports a positive returns to diversi cation anouncements through the mid-1970s (Matsusaka, 1993; Hubbard and Palia, 1998). The evidence, however, for the 1980s and 1990s is mixed (Kaplan and Weisbach, 1992; Morck et al., 1990; Hyland, 1999; and Chevalier, 2000) 3 See, e.g., Rajan, Servaes and Zingales (2000), Sharfstein and Stein (2000) Wolf (1999), and Matsusaka and Nanda (2000) and section 6 for a more detailed discussion of this literature. 1

4 returns plus the synergies arising from interdivisional coordination on project choice. Shareholders hire a CEO to choose and implement projects, allocate resources, and choose an organizational structure (centralization vs decentralization) and a diversi cation strategy (diversi cation versus focus). Centralization in this paper is understood as an structure under which the CEO chooses the projects to be implemented in each division, while decentralization as an structure under which each divisional manager chooses the project to be implemented in his own division. In this environment, the consequences of decentralization are that divisional managers exert more e ort, the CEO s productivity is larger and divisional managers productivity is lower. 4 Divisional managers exert more e ort because they work always in their preferred projects, while under centralization they, sometimes, work in their preferred projects. The CEO s productivity is larger because he does not devote time and e ort to project implementation while divisional managers productivity is lower because they devote time and e ort to project implementation. For instance, time to gather information about project characteristics or e ort spent to learn the realized state. Thus, decentralization, understood as delegation of real authority, induces divisional managers to exert more e ort and increases the CEO s productivity, but it decreases divisional managers productivity. The consequences of diversi cation are that divisional managers exert less e ort and the CEO s productivity is lower. The CEO s productivity is lower because he has to manage two divisions instead of one and divisional managers exert less e ort because their expected private bene ts are lower as a consequence of the CEO s lower productivity. Given that in this paper interdivisional coordination on project choices across divisions is achieved only under an integrated rm and it is ensure under centralization, it is shown that the CEO prefers a diversi ed strategy and a centralized structure when the synergies are large, a diversi ed strategy and a decentralized structure when the synergies are neither too large nor too small and a decentralized focused rm when synergies are small. For large synergies, both the CEO and shareholders greatly bene t from interdivisional coordination while for small synergies, both the CEO and shareholders bene t more from being focused and decentralized since synergies cannot outweigh the CEO s lower productivity and divisional managers lower e ort. When synergies are neither too large nor too 4 See, Aghion and Tirole (1997) for the seminal paper on the positive incentive e ects of delegation. 2

5 small, then the CEO chooses a decentralized multidivisional rm because synergies are large enough to outweigh divisional managers lower e orts and the CEO s lower productivity due to overload, but not large enough to compensate for the CEO s lower productivity due to that he has to spend time and e ort implementing projects. In addition, it is shown that when synergies are not large enough, the CEO bene ts more from diversi cation than shareholders do and thereby, the CEO chooses to diversify when the valuemaximizing strategy is to remain focused. Thus, conditional on diversi cation taking place, when synergies are not large enough diversi cation destroys value. This result is driven by three forces: the wrong organizational structure e ect, the pure disaligned preference e ect, and the empirebuilding e ect. The wrong organizational structure e ect arises when the CEO chooses a centralized multidivisional rm despite that a decentralized one is value-maximizing. The intuition being that under decentralization the CEO works with positive probability in a project that does not yield any private bene ts to him, while this is not true for shareholders. This implies that the CEO s bene ts from decentralization are lower than shareholders bene ts. Thus, from shareholders point of view synergies are not large enough to outweigh divisional managers lower e ort and the CEO s lower productivity, yet they are large enough from the CEO s point of view. The pure disaligned preference e ect is also the outcome of disaligned preferences on project choices, but in this case the value-maximizing structure is adopted. This e ect arises because under a decentralized focused rm, shareholders bene t from any project implemented by divisional managers, while the CEO bene ts only when certain type of projects are implemented. The empire building e ect arises because the CEO s decision to diversify depends on the private bene ts yielded by a diversi ed rm relative to the private bene ts yielded by a focused rm, while the optimality of diversi cation depends on the returns yielded by diversi ed rm relative to the return of a pool of focused rms. This implies that even if the incentives are perfectly aligned, the CEO s prefers diversi cation more often than it is optimal. This e ect while trivial it has been ignored by the agency literature because the decision to diversify is usually assumed to be exogenous or made by shareholders. Agency theory as developed by Jensen and Meckling (1976) suggests that managers make decisions that increase their utility while potentially decreasing the rm s value because they are not full 3

6 residual claimants. In this context, there are two di erent types of agency problems that provide explanations for why a conglomeration strategy is adopted or why managers diversify their rms. The rst is based on the idea that managers diversify their idiosyncratic risk resulting from having undiversi ed positions in their own rms. The evidence on this is, however, mixed; some authors nd that managers with more stock ownership acquire divisions in business that allow to lower the risk, while others nd evidence of less diversi cation in rms with more managerial stock ownership. But, more importantly, nothing prevents a manager from diversifying using the stock market. Thus, this is not a good explanation for why managers diversify their rms. The second type of explanation is based on the idea that managers derive private bene ts of control from managing more diversi ed rms (Jensen, 1986; Stulz, 1990). Reasons for this range from prestige from managing larger rms, entrenchment through speci c human capital investments to the idea that larger rms provide larger pay. These theories, although based also on agency problems, fail to explain when diversi cation takes place and why diversi cation is good for some rms and bad others. In fact, they can only explain why diversi cation can be value-reducing in the case that a diversi ed strategy is adopted. In addition, the model produces a number of predictions concerning the interaction between diversi cation strategy, organizational structure and rm s value that are more subtle. The model predicts that; (i) rms that pursue diversi cation that is not su ciently related are traded at a discount (See, e.g., Ravenscraft and Scherer (1987), Kaplan and Weisbach (1992), and Fang and Lang (2000)); (ii) the more the CEO understands his rm s opportunities, the smaller the set of parameters under which diversi cation destroys value; (iii) there are both, a causal link between diversi cation and value and measurement problems (See, e.g., Berger and Ofek (1995), Lang and Stultz (1994), Whited (2000)); (iv) when cash- ows under a multidivisional rm are abundant relative to a focused rm, diversi cation is less likely to reduce value, while when cash- ows are scarce, the opposite occurs; and (v) the more unequal are units opportunities ex-post, the larger the set of parameters under which diversi cation destroys value (See, e.g., Rajan, Servaes and Zingales, (2000)); (vi) value-reducing diversi cation occurs more often when rms are centralized. The next section, section 2 presents the model. Section 3 derives the CEO and divisional managers project choices and optimal e ort choices under each organizational structure and diversi- 4

7 cation strategy. In the next section, section 4, the CEO s preferred diversi cation strategy and organizational structure is derived. In section 5, I discuss the empirical evidence on the existence of a diversi cation discount and show how the model can explain why diversi cation is good for some rms, but bad for others. In section 6, I study how an e cient internal capital market can destroy value, the empirical evidence and alternative models. In the next section, section 7, the related literature is brie y discussed. Finally, in section 8, concluding remarks are presented. 2 The Basic Model I consider a rm composed of shareholders, a risk-neutral CEO and two risk-neutral agents or divisional managers. Shareholders hire a CEO to implement and work on projects, choose an organizational structure and a diversi cation strategy, and allocate resources across divisions. Whereas divisional managers are hired to implement and work on projects. 5 The CEO must jointly choose the organizational structure (centralization vs decentralization) and diversi cation strategy (multidivisional rm vs focused rm). In a centralized organizational structure the CEO chooses the projects to be implemented, while in a decentralized one this decision is fully delegated to divisional managers. 6 Each division faces 2 projects, called and in what follows. A project s return depends on divisional manager s e ort, the CEO and divisional manager s productivity and, in a multidivisional rm, the project combination across divisions. A unit s return when operated as a focused rm is Hh i ¼ i e i, i =1; 2, where¼ i is the return for which unit i by itself can be accountable for, e i is divisional manager i s e ort, H is a parameter that captures the CEO s productivity and h i is a parameter that captures divisional manager i s productivity. When both divisions units are operated as an integrated rm total return when project combination (n i ;n j ) 2f( ; ) ; ( ; ) ; ( ; ) ; ( ; )g is implemented is (h 1 ¼ 1 e 1 + h 2 ¼ 2 e 1 + s (n i ;n j )(h 1 e 1 + h 2 e 2 )) H, wheres (n i ;n j ) is the synergy arising 5 I assume that there is no con ict of interest within a division; that is, a divisional manager s preference fully captures the preferences of the members belonging to his division. 6 Notice that in the terminology of Aghion and Tirole (1997), here, real and formal authority are the always the same. 5

8 when project combination (n i ;n j ) : To capture in a simple form the bene ts from interdivisional coordination, I assume that s ( ; ) = s and s ( ; ) =s ( ; ) =s ( ; ) =0. This assumption captures well the idea that interdivisional coordination in a given set of projects is a necessary condition to get a positive return from synergies and that coordination could be more bene cial when is in one type of project, project ; than in other type, project. 7 The CEO s productivity parameter H is equal to 1 in a decentralized multidivisional rm and M in centralized multidivisional rm, while it is 1 in a decentralized focused rm and M in a centralized focused rm, with M 2 [0; 1] and 2. This is meant to capture two things: (i) the CEO s cost from overseeing two instead of one division this explains why within a given organizational structure the CEO s productivity parameter is smaller in a diversi ed rm; and (ii) the idea, in a reduced form, that implementing projects diverts personal productive resources away from other tasks. For instance, project implementation requires to gather some information in order to distinguish projects. This explains why the CEO s productivity parameter within a diversi cation strategy is smaller in a centralized rm. Divisional manager i s productivity parameter h i is equal to 1 in a centralized organization and m 2 [0; 1] in a decentralized one for i =1; 2. This is meant to capture, in a reduced form again, the idea that the task of implementing projects diverts personal productive resources away from other tasks. Furthermore, a divisional manager s disutility of e ort e i is 1 2 e2 i. Furthermore, it is assumed that m M; that is, it is more costly in terms of productivity for the CEO to investigate projects than for divisional managers. This captures the idea that the manager closest to the project (division) is more likely to know more about its prospects and thereby it has to use less personal resources to implement a project. Implemented projects yield di erent private bene ts to the CEO and divisional managers. They yield to divisional manager i afraction' i of the return that accrues only to division i when his preferred project is implemented and 0 otherwise, while they yield to the CEO a fraction Á of the return that each division by itself is accountable for plus a fraction Á of the synergies when his 7 It can be shown that if s ( ; ) =s, most of our results holds, but the algebra gets unnecessarily complicated. In the last section, I discuss how our results change when s ( ; ) =s is assumed. 6

9 preferred project is implemented and 0 otherwise. This implies that divisional managers do not get any private bene ts from synergies. This is meant to capture the idea that s in uences the overall rm s performance, but cannot be traced back to any particular division and thereby cannot bene t divisional managers, and that divisional managers have less incentives to coordinate project choices than the CEO. 8 This is consistent with the evidence that coordination is harder to achieve under a decentralized structure. 9 The table below shows in the rst entry the fraction of the return, Hh i ¼ i e i, that accrues to divisional manager i as a private bene t for each project combination and each state, while the rst element in the second entry is the fraction of division i s return that accrues to the CEO as a private bene t and the second one is the fraction of return from division j s return that accrues to the CEO as a private bene t. State ( ; ) ( ; ) ( ; ) ( ; ) probability 1 ' i ; (Á; Á) ' i ; (Á; 0) 0; (0;Á) 0; (0; 0) p 2 0; (Á; Á) 0; (Á; 0) ' i ; (0;Á) ' i ; (0; 0) 1 p. This shows that in state 1 divisional manager i prefers project andinstate2 project ; while the CEO prefers always project and thereby ( ; ) over ( ; ), ( ; ) and ( ; ) : Notice also that the CEO prefers combination ( ; ) over ( ; ) and ( ; ) and these two over combination ( ; ) while shareholders prefer ( ; ) over ( ; ), ( ; ) and ( ; ) ; but they are indi erent between combinations ( ; ), ( ; ) and ( ; ) : This is meant to capture the existence of con ict of interests between the CEO and shareholders. Furthermore, the severity of the con ict between shareholders, the CEO and divisional managers is captured by p; for as p goes to 1, shareholders, the CEO and divisional managers con ict become less severe since divisional managers choose the CEO and shareholders preferred project combination more often and the CEO s preferred combination is more likely to be divisional managers preferred combination. ² Assumption 1: ' i 2 [0; 1], Á 2 [0; 1] for i =1; 2. 8 It could be assumed that divisional managers also get private bene ts from synergies, yet this does not change the results and over complicates the algebra. 9 See, e.g., the discussion in Argyres (1995) on a common technology adoption at IBM (centralized) versus GM (decentralized) and the management of strategy literature (Hill and Hoskisson, 1987 and the references therein). 7

10 Finally, it is assumed that the CEO and divisional managers reservation utility is zero and both have limited liability that it is also normalized to zero. The timing of decision is as follows: At stage 1, shareholders hire the CEO to undertake the tasks mentioned above. At stage 2, the CEO hires divisional managers and chooses the organizational structure and diversi cation strategy. At stage 3, divisional managers choose e ort. At stage 4, states are realized and learned by divisional managers and the CEO. At next stage, stage 5, under centralization the CEO chooses the projects to be implemented, while under decentralization divisional managers choose the projects to be implemented. At the nal stage, returns are realized and compensation, if any, takes place Project Choice and Divisional Managers E ort Allocations. Suppose rst that the CEO chooses a centralized multidivisional rm and thereby he chooses the projects to be implemented. Given the CEO s preferences over projects, he implements project in both division independent of the realized state since the implementation of project yields no private bene ts to him. Thus, the CEO s expected payo in a centralized multidivisional rm, denoted by U2 c in what follows, is given by U2 c = Á M (¼ 1e 1 + ¼ 2 e 1 + s (e 1 + e 2 )) ; while divisional manager i s expected payo, denoted by u c i2 in what follows, is u c i2 = ' i pm ¼ ie i 1 2 e2 i : M Note that ' i ¼ i is multiplied by p because only in state 1 the CEO chooses divisional manager i s preferred project and m has been suppressed since divisional manager i does not spend any resources on project implementation. 10 The timing in which divisional managers choose e ort after seeing the projects implemented yields the same results. 8

11 It readily follows from this that divisional manager i s e ort under a centralized multidivisional structure, denoted by e c i2 ; is given by ' i pm¼ i. Thus, the CEO s expected payo when a centralized multidivisional is adopted becomes U c 2 = ÁM 2 p 2 '1 ¼ ' 2 ¼ s (' 1 ¼ 1 + ' 2 ¼ 2 ) : (1) When a centralized focused rm is adopted, the CEO s preferred project continues being project and divisional manager i s e ort is e c i1 ' ipm¼ i : Thus, the CEO s expected payo when a centralized focused rm, denoted by U c i1,is U c i1 = Á' i pm 2 ¼ 2 i : (2) The parameter has been suppressed since there is only one division and thereby there is no overload. Consider next the case in which the CEO chooses a decentralized multidivisional rm. In this case divisional managers choose to implement their preferred projects always; that is, project in state 1 and project in state 2 because the implementation of project in state 1 and project in state 2 yields 0 private bene ts. Thus, the CEO s expected payo in a decentralized multidivisional rm, denoted by U c 2 in what follows, is given by U d 2 = Ámp (¼ 1e 1 + ¼ 2 e 1 + ps (e 1 + e 2 )), while divisional manager i s expected payo, denoted by u d i1 in what follows, is u d i1 = ' m i ¼ ie i 1 2 e2 i. Ám Note that in this case ' i ¼ i is not multiplied by p because independent of the realized state divisional managers always implement their preferred project and m has been added to capture the decrease in productivity from the use of personal resources in the process of project implementation. Notice also that M has been suppressed from U d 2 because under decentralization the CEO does not 9

12 implement projects and thereby does not need to take time or resources away from other productive taskstobeabletoimplementprojects. It readily follows from this that divisional manager i s e ort under a decentralized structure, denoted by e d i2,isgivenbyed i2 = ' i m¼ i. Thus, the CEO s expected payo when a decentralized multidivisional rm is adopted becomes U d 2 = Ám2 p 2 '1 ¼ ' 2 ¼ ps (' 1 ¼ 1 + ' 2 ¼ 2 ) : (3) Lastly, when a decentralized focused rm is adopted, divisional manager i also implements his preferred project always, thereby his e ort is e d i1 ' im¼ i. Thus, the CEO s expected payo when a decentralized focused rm is adopted, denoted by U d 1i,is U d 1i = Ápm 2 ¼ 2 i. (4) In this case the parameter M has been suppressed since the CEO focuses only on one division and specializes on the productive task only. It readily follows from the analyzes above that the consequences of decentralization are that divisional managers exert more e ort, the CEO s productivity is larger and divisional managers productivity is lower. Divisional managers exert more e ort because they work always in their preferred projects, while under centralization they work in their preferred projects only when state 1 is realized. The CEO s productivity is larger because he does not devote time and e ort to project implementation while divisional managers productivity is lower because they devote time and e ort to project implementation. For instance, time to gather information about project characteristics or e ort spent to learn the realized state. Thus, decentralization, understood as delegation of real authority, induces divisional managers to exert more e ort and increases the CEO s productivity, but it decreases divisional managers productivity. In what follows, I call the decrease in divisional managers e ort from centralization, divisional managers negative incentive e ect of centralization. The consequences of diversi cation are that divisional managers exert less e ort and the CEO s productivity is lower. The CEO s productivity is lower because he has to manage two divisions instead of one and divisional managers exert less e ort because their expected private bene ts 10

13 are lower as a consequence of the CEO s lower productivity. In what follows, I call the decrease in divisional managers e ort from diversi cation, divisional managers negative incentive e ect of diversi cation. 4 Organizational Structure and Diversi cation Strategy In this section, I solve the CEO s joint determination problem consisting on choosing the organizational structure and diversi cation strategy. First, I derive conditions under which centralization is the CEO s chosen structure. To simplify the analysis and focus on the main trade-o s, it is assumed from now on that divisions are symmetric; that is, ¼ 1 = ¼ 2 = ¼ and ' 1 = ' 2 = '. It readily follows from equations 1 and 3 that the di erence between the CEO s expected private bene ts in a centralized multidivisional rm and a decentralized one, denoted by 4U 2,is = 2Á ¼ M e c 2 e d 2 + e d 2 (M mp) + s M e c 2 e2 d + e d 2 M mp 2 = 2Á'¼p ¼ M 2 m 2 + s M 2 m 2 p (5) 2 Notice that 4U 2 has two terms: the rst one corresponds to the di erence in private bene ts from divisional returns and the second one captures the di erence in private bene ts from synergies. The rst term is negative because of divisional managers negative incentive e ect of centralization and because the CEO s productivity is lower under centralization. Given the assumptions made so far, the second term may be either positive or negative. what follows, I assume that this term is positive; 11 To make the problem interesting in i.e., M 2 m 2 p 0. That is, the lower divisional managers e ort is outweighed by the fact that under centralization coordination is ensure while under decentralization interdivisional coordination takes place only when state 1 is realized in both divisions, which occurs with probability p 2. Thus, as the synergy parameter s increases 11 Notice that if M 2 m 2 p 0 and M m; centralization is always the preferred structure, while if M 2 m 2 p 0 and M m; decentralization is always optimal. Thus, these two cases are ruled out because they are less interesting toanalyzethantheonepursuedinthetext.theotherpossiblecaseistheoneinwhichm 2 m 2 p 0 and M m, yet this parameter con guration can never occur. 11

14 centralization is the preferred structure since the private bene ts from interdivisional coordination outweigh divisional mangers negative incentive e ect of centralization and the CEO s lower productivity. It readily follows from this that centralization is the preferred organizational structure when a diversi ed strategy is adopted if and only if s s c (p), where s c (p) ¼ m 2 M 2 (M 2 m 2 p). > < 0: In a diversi ed rm centralization is more likely to be the CEO s preferred organizational structure as the CEO s productivity increases, divisional managers productivity decreases, project return and the congruence of interest parameter decrease. That both, a decrease in m and in increase in M; make centralization more likely is due to that the former results in a lower divisional managers productivity under decentralization and the latter in an increase in the CEO s productivity under centralization. That centralization is less likely to be the CEO s preferred structure as p increases follows from that divisional managers are more likely to implement the CEO s preferred project and interdivisional coordination on project without CEO s command takes place more often. Lastly, that an increase in ¼ makes centralization less likely to be the preferred structure is due to that divisional returns are larger in a decentralized structure. Consider next the case in which a focused strategy is adopted. Then, the di erence between the CEO s expected private bene ts in a centralized focused rm and in a decentralized one, denoted by 4U 1 ; is given by Á'pM 2 ¼ 2 Á'm 2 ¼ 2. Because m M, under a focused strategy decentralization maximizes the CEO s payo. The intuition being straightforward. These results up to here are summarized in the next proposition. Proposition 2 (i) Under a diversi ed strategy centralization is the chosen structure for all s s c (p); and (iii) under a focused strategy decentralization is always the chosen structure. 12

15 Next I derive conditions under which the CEO prefers a diversi ed strategy over a focused one given that he is choosing his preferred organizational structure. Because under a focused strategy decentralization is always the preferred organizational structure, the CEO chooses to diversify when 4U 21 2max ª U2 c;ud 2 U d 1 > 0. It readily follows from equations1,3and4,that 8 < 4U 21 = : Á ¼M 2 e c 2 e d 1 + e d 1 (2M mp) +2Me c 2s if s s c (p), Ámp ¼ 2 e d 2 e d 1 + e d 1 (2 ) +2pe d 2 s otherwise. (7) Thus, if the optimal values e c 2 ;ed 2 and e d 1 are plugged in 7, 4U 21 > 0 if and only if s>s c (p; ), while 4U 21 0 if and only if s s c (p; ), where 8 < s c (p; ) : ¼( 2 2) 2p if p>p c, ¼((m ) 2 2M 2 ) 2M 2 otherwise, and p c M2 ( 2 2). (m ) 2 2M 2 The intuition being straightforward. The second term in 4U 21 is positive because it corresponds to the private bene ts from synergies that take place only under integration. Whereas the rst term is negative due to divisional managers negative incentive e ect of diversi cation and the CEO s lower productivity from being overloaded. Thus, as s increases, diversi cation is more likely to be the CEO s preferred diversi cation strategy. This leads to the following result. Proposition 3 A multidivisional rm is adopted for all s>s c (p; ) while a focused one is adopted for all s s c (p; ). As expected s c (p; ) increases with the overload parameter, the CEO s productivity parameter M and the congruence of interest parameter p. Furthermore, it decreases with m when centralization is the preferred structure. The reason being that an increase in m increases the CEO s private bene t from a decentralized focused rm, while it does not increase the CEO s private bene t from a centralized multidivisional rm. 13

16 γ 1 U c 2 () s Decentralized Focused Decentralized Multidivisional U d () s 2 Centralized Multidivisional U d 1 U d ( s = 0) 2 U d ( s = 0) 2 s c ( p, γ) ( p) s c S L s Figure 1: So far I have derived the CEO s preferred organizational structure within each diversi cation strategy and conditional on that the preferred diversi cation strategy. In the next proposition I combine these two results together to get the following. Proposition 4 (i) A centralized multidivisional rm is adopted for all s>max fs c (p; ) ;s c (p)g; (ii) a decentralized multidivisional rm is adopted for all s c (p; ) <s max fs c (p; ) ;s c (p)g; and (iii) a decentralized focused rm is adopted for all s s c (p; ). These results can be seen in gure 1 where s c (p; ) and s c (p) aredepictedinapicturethathas in the x-axis the synergy level s andinthey-axis the CEO s expected payo in each of the relevant cases. Notice that the upper envelope (dark line) represents the CEO s private bene ts when he chooses the strategy and structure combination that maximizes his private bene ts. 14

17 5 The Value Consequences of Diversi cation There is a large empirical literature asking the following question: what are the consequences of diversi cation for shareholders value? For the most part, the evidence is one that is unfavorable to diversi cation, especially if one focuses on unrelated diversi cation and data after around, say, 1980s. The evidence shows that conglomerates are traded at a discount relative to a portfolio of stand-alone rms and during the 1980s a trend toward value increasing refocus has been highly documented. 12 The most common measure to study the value e ect of diversi cation in recent work is the wellknown diversi cation discount. As developed by Lang and Stulz (1994) and Berger and Ofek (1995), the diversi cation discount compares the stock price of a diversi ed rm to the imputed values for its individual segments, where these imputed values are obtained from comparable focused rms that operate in the same industries as the diversi ed rms segments. These authors nd substantial mean discounts, on the order of 15%, which they interpret as evidence of value destruction by diversi ed rms. This work has been extended to a variety of other sample periods and countries by Servaes (1996), Lins and Servaes (1999, 2000), and Rajan et. al., (1999). However, a number of other papers have questioned further the idea that the diversi cation discount is the result of value destruction. Campa and Kedia (1999), Hyland (1999), Whited (2001) and Chevalier (2000) all argue in one way or another that the discount is the result of uncontrolled endogeneity; for rms with poor returns are the ones most likely to diversify. Thus, despite the fact that the evidence suggests, after taking into account the endogeneity problems, that the mean discount is not as large as the rst wave of papers found, one cannot be completely sure that there is no causal link from diversi cation to value See, e.g., Wernerfelt and Montgomery (1988), Bhagat, Shleifer and Vishny (1990b), Morck, Shleifer and Vishny (1990), Kaplan and Weisbach (1992), Liebeskind and Opler (1993), John and Ofek (1995), Comment and Jarrell (1995), Berger and Ofek (1996, 1999), and Denis, Denis and Sarin (1997). In contrast, Matsusaka (1993) nds positive event returns for diversifying acquirors in the 1960s. 13 The evidence in Chevalier 2000, however, reinforces the view that there is no causal link between diversi cation and value. She nds that although the stocks of acquirors tend to drop upon announcement of a diversifying transaction, looking at the combined return to acquirors and targets in such deals generally nd it to be either close to zero or slightly positive. See, also, Matsusaka, (1993); Hubbard and Palia, (1998); Kaplan and Weisbach, (1992); Morck et 15

18 In this section I use the results derived above to show that: (i) there exist a causal link between diversi cation and value; that is, diversi cation may destroy value for some rms; and (ii) that the failure to control for endogeneity problems favors the existence of a diversi cation discount. To show that there is causal link between value and diversi cation, the value-maximizing organizational structure and diversi cation strategy must be derived. Given the optimal e ort and the chosen projects, shareholders value from a centralized focused rm is ¼ c M 2 p'¼ 2 while under a decentralized focused rm is ¼ d 'm 2 ¼ 2. Thus, under a focused strategy decentralization is value-maximizing since m 2 >M 2 p: Firm s value under a centralized multidivisional rm is while under a decentralized one is ¼ c 2 = 2'M 2 ¼ 2 p (¼ + s), (8) ¼ d 2 = 2'm2 ¼ 2 ¼ + p 2 s. (9) It readily follows from equations 8 and 9 that the di erence between rm s value in a centralized multidivisional rm and a decentralized one, denoted by 4¼ 2,isgivenby = 2 ¼ M e c 2 e d 2 + e d 2 (M m) + s M e c 2 e2 d + e d 2 M mp 2 = 2'¼ ¼ M 2 p m 2 + sp M 2 m 2 p (10) 2 Notice that, as 4U 2, 4¼ 2 has two terms: the rst one that corresponds to the di erence in divisional returns and the second one that captures the di erence in the returns from synergies. The rst one is negative because divisional managers negative incentive e ect of centralization and because the CEO s productivity is lower than divisional managers productivity. The second term is positive because divisional managers negative incentive e ect of centralization is outweighed by the fact that under centralization interdivisional coordination is ensure while under decentralization that takes place only when state 1 is realized in both divisions, which occurs with probability p 2. Thus, centralization is the value-maximizing structure when a diversi ed strategy is adopted if and al., (1990); and Hyland, (1999). 16

19 only if s s f (p), where s f (p) ¼ m 2 M 2 p p (M 2 m 2 p). (11) The following comparative statics can be easily derived. Proposition (p) > @sf (p) < 0 > 0 for p>~p: 14 The intuition for m, M and ¼ isthesameasfors c (p) : Notice that in contrast to the case for s c (p) ; the threshold s f (p) is not monotonically increasing in p: This di erence arises because the di erence in private bene ts from divisional returns p¼ M 2 m 2 decreases as p increases, while the di erence in divisional returns ¼ M 2 m 2 p increases as p increases. This di erence arises because under decentralization the CEO gets private bene ts that are proportional to divisional returns only when project is implemented, while shareholders bene t from the implementation of both, project and. This explains why divisional returns in a decentralized rm are m 2 ¼ and the CEO s private bene ts are pm 2 ¼. Given that under a focused strategy decentralization is always the value-maximizing organizational structure, it is value-maximizing to diversify when 4¼ 21 2max ¼ c 2;¼ d 2ª 2¼ d 1 > 0: It readily follows from equations 8, 9 and ¼ d 1 that 8 < 4¼ 21 = : 2 ¼M e c 2 e d 1 + e d 1 (M m) + Me c 2s if s s f (p), 2m ¼ e d 2 e d 1 + e d 1 (1 ) + p 2 e d 2 s otherwise. Thus, if the optimal values e c 2 ;ed 2 and e d 1 are plugged in 12, 4¼ 21 > 0 if and only if s>s f (p; ), while 4¼ 21 0 if and only if s s f (p; ), where 8 < s f (p; ) : ¼( 2 1) p 2 if p>p f, ¼((m ) 2 M 2 p) M 2 p otherwise, ³ and p f ( m)2 2M 1 2 2M ( m) 4 4M The intuition being straightforward. The second term in 4¼ 21 is positive because it corresponds to the private bene ts from synergies that take place only under integration. Whereas the rst term 14 ~p = 2m M 1 M m 2 M (12) 17

20 c γ 1 π 2 () s Decentralized Focused Decentralized Multidivisional d π 2 () s Centralized Multidivisional π d 1 d π ( s = 0) 2 d π ( s = 0) 2 s f ( p, γ) ( p) s f S L s Figure 2: is negative due to divisional managers negative incentive e ect of diversi cation and the CEO s lower productivity from being overloaded. Thus, as s increases diversi cation is more likely to be the CEO s preferred diversi cation strategy. The discussion so far is summarized in the next proposition, which parallels the results in proposition 4: Proposition 6 (i) A centralized multidivisional rm is value-maximizing for all s>max fs f (p; ) ;s f (p)g; (ii) a decentralized multidivisional rm is value-maximizing for all s f (p; ) <s max fs f (p; ) ;s f (p)g; and (iii) a decentralized focused rm is value-maximizing for all s s f (p; ). These results can be easily seen in gure 2 where s f (p; ) and s f (p) are depicted in a picture that has in the x-axis the synergy level s and in the y-axis 4¼ 21. Notice that the upper envelope (dark line) represents shareholders value when the value-maximizing strategy and structure combination is chosen. Diversi cation destroys value when the CEO chooses to diversify and diversi cation yields a 18

21 lower rms value than a pool of focused rms in the same business segments. 15 Thus in what follows I derive conditions under which the CEO chooses to diversify when being focused is the value-maximizing diversi cation strategy. According to propositions 4 and 6, the CEO chooses to diversify if and only if s>s c (p; ) while diversi cation is value-maximizing if and only if s>s f (p; ). This, implies that if s c (p; ) < s f (p; ), there is s range of values of s for which diversi cation destroys value because the CEO chooses to diversify for synergy levels that are lower than the minimum synergy level that makes diversi cation value-maximizing. Thus, in the next lemma conditions under which s c (p; ) <s f (p; ) are obtained. Lemma 7 s c (p; ) <s f (p; ) for all p: Proof. Suppose that (m ) 2 2M 2 > 0: Then, the rst to notice is that p c <p f. It is easy to show that p c p f = M 2 2 (m ) 2 2M ( m)2 2 2M ³ 2M 2 ( m) 4 4M if and only if (m ) 2 M This holds always because m M. Given this, it readily follows after some simple algebra that 4s (p; ) de ned as s f (p; ) s c (p; ) is equal to 8 >< 4s (p; ) = >: ¼( 2 (2 p) 2(1 p)) 2p 2 if p>p f ; ¼( 2 (2m 2 M 2 )+2M 2 (1 p)) 2M 2 p if p c <p p f ; ¼(m ) 2 (2 p) 2M 2 p if p p c : Because p M 2 m 1 and m M; it readily follows from this that 4s (p; ) > 0 for all p. Thus, s c (p; ) <s f (p; ) for all p. 15 Most papers that deal with the diversi cation discount at a theoretical level argue that a diversi cation discount exists when rm s value is lower than the sum of focused rm s value in the same business segments. This, however, is not true unless when that holds a rm chooses to diversify. If the shareholders choose the strategy this will never be the case since, diversi cation will not take place when it is not value-maximizing and the diversi cation discount will be only due to a measurement problem. 19

22 Given that s c (p; ) <s f (p; ) for all p, the next proposition follows from the de nition of 4¼ 21, propositions 4 and 6 and lemma 7. Proposition 8 Diversi cation destroys value for all s 2 (s c (p; ) ;s f (p; )) : This proposition establishes that if diversi cation takes place; that is, if s>s c (p; ), thenitis value-reducing for small values of the synergy parameter s, s s f (p; ) and value-increasing for larger values of s; s > s f (p; ). This explains why rms that diversify into sectors that are not related aremuchmorelikelytobetradedatadiscount. Infact,theAmericancorporateworldwitnessed during the 1990s a movement from unrelated diversi cation to related diversi cation and away from conglomerates. For instance, Fan and Lang (2000) documents that both vertical relatedness as well as complementarity of rm s segments have increased over time and that multi-segment rms have decreased in number. In 1979, the proportion of multi-segment rms was 46 %, while this was only 20 % in 1996, which is consistent with the refocus trend that occurred during the 1990s. 16 To understand the intuition behind this proposition it is useful to analyze di erent ranges for the synergy level s. Consider rstthecaseinwhichs s c (p; ), then the CEO chooses a decentralized focused rm which is the value-maximizing organizational structure and diversi cation strategy (region C plus F in gure 3). This occurs because synergies are not large enough to compensate for the CEO s lower productivity from overseeing two divisions instead of one and divisional managers negative incentive e ect of diversi cation. Consider next the case in which s>s f (p; ). In this case the CEO chooses a diversi ed strategy which is the value-maximizing strategy (region A plus B in gure 3). The reason being that synergies are large enough to compensate for the CEO s lower productivity and divisional managers negative incentive e ect of diversi cation. This is not to say that the CEO s choices are value-maximizing since he may be choosing a value-reducing structure. To see this notice that s f (p) >s c (p) for all p; 17 that is, there is a region given by (s c (p) ;s f (p)) 16 See, for instance, Ravenscraft and Scherer (1987) and Kaplan and Weisbach (1992), who nd that many unrelated acquisitions are later divested. 17 This follows from that m 2 M 2 p>m 2 M 2 and p M 2 m 2 p <M 2 m 2 p. 20

23 on which the CEO, when a diversi ed strategy is adopted, chooses a centralized structure despite that a decentralized one is value-maximizing. Thus, if s f (p) >s>s c (p) s f (p; ), theceo chooses a centralized multidivisional rm despite that a decentralized multidivisional rm is valuemaximizing (region B in gure 3). Thus, diversi cation does not destroy value in the sense that when units are integrated under one rm are valued less than when they are operated as focused rms, yet the CEO s chosen structure is not value-maximizing since rms value when a diversi ed strategy is adopted is larger under a decentralized multidivisional rm. Lastly, consider the case in which s 2 (s c (p; ) ;s f (p; )) (region D plus E plus G in gure 3). In this case it follows from proposition 4 that the CEO chooses to diversify while it follows from proposition 6 that the value-maximizing diversi cation strategy is to remain focused. Thus, diversi cation destroys value for those rms that have s 2 (s c (p; ) ;s f (p; )). To better understand why diversi cation destroys value when s 2 (s c (p; ) ;s f (p; )), itisuseful to re-write 4¼ 21 as a function of 4U 21. It follows from equations 7 and 12 that 8 1 >< Á 4U21 Á (2 p) ¼me1 d if s>s f (p), 4¼ 21 = 1 Á 4U21 Á4¼ 2 Á (2 p) ¼me1 d if s c (p) <s s f (p), >: 4U21 Á (2 p) ¼me1 d if s s c (p). 1 Á (13) Notice that the value of 4¼ 21 depends on which organizational structure is in place. When s>s f (p) and s s c (p) ; the CEO s choice of structure is value-maximizing, while when s c (p) < s s f (p) ; the CEO chooses a centralized structure despite that a decentralized one is valuemaximizing; that is 4¼ 2 < 0 for s c (p) <s s f (p). This implies that when the CEO s chooses to diversify and s c (p) <s s f (p), there is gain that is not realized equal to Á4¼ 2 : The intuition is as follows. Centralization is the CEO s preferred organizational structure because the private bene ts from synergies outweigh divisional managers negative incentive e ect of centralization and the CEO s lower productivity. Yet, centralization is not value-maximizing because the return from synergies are not large enough to outweigh the negative e ects of centralization. The reason being that the CEO does not get any private bene ts when project is implemented, while shareholders bene t from the implementation of both projects, and. I call this e ect wrong organizational structure e ect of diversi cation. 21

24 Furthermore, it follows from equation 13 that 4U 21 may be positive while 4¼ 21 is negative even when the value-maximizing structure is adopted. To understand why this occurs, it is useful to analyze the term (2 p) ¼me d 1 : This term can be split in two terms ¼med 1 and (1 p) ¼me d 1, where the rst one captures what I call the empire-building e ect and the second one captures what I call the pure disaligned preferences e ect. The empire building e ect arises because the CEO s decision to diversify depends on the private bene ts yielded by a diversi ed rm relative to the private bene ts yielded by a focused rm, while the optimality of diversi cation depends on the returns yielded by diversi ed rm relative to the return of a pool of focused rms. This implies that even if the incentives are perfectly aligned; i.e., p =1, the CEO s prefers diversi cation more often than it is optimal. This e ect while trivial it has been ignored by the agency literature because the decision to diversify is usually assumed to be exogenous or made by shareholders. 18 The second e ect results from disaligned preferences on project choices between the CEO and shareholders and the CEO and divisional managers. For the CEO does not derive any private bene ts from project, while shareholders and divisional managers do. Thus, shareholders value in a focused rm is positive when either project or is implemented, while the CEO s private bene t is positive if and only if state 1 is realized; i.e., if and only project is implemented. Given that project is implemented with probability p, the CEO s private bene t from a focused rm is a fraction Á of ¼me d 1, while shareholders value is ¼me d 1 : It is interesting to know how the range of values of s under which a diversi cation destroys value changes with the main parameters of the model. Proposition 9 4s (p; ) is strictly increasing in ¼ and, strictlydecreasinginp, non-decreasing in m and non-increasing in M. That 4s (p; ) is strictly decreasing in p follows immediately from the fact that as p increases shareholders, the CEO and divisional managers preferences get more aligned and therefore, the 18 It could be argued that this e ect is fully driven by the assumption that the CEO is not allow to own shares on a rm other than the one he works on. This argument, though correct, does not change this result since the second e ect is immune to this critique. If the CEO is allowed to own shares in a focused rm other than the one he is the CEO, the term (2 p) ¼me d 1 becomes (1 p) ¼med 1. 22

25 s f s f ( p,γ ) s f ( p) A G s c ( p) π ( mγ 2M ) 2M s c ( p,γ ) F D E B A π ( m M ) M C C 2 πm ( γ 1) 2 M πm ( γ 2) 2 M p c p f M m p Figure 3: A+B: Value-maximizing diversi cation; C+F: Value-maximizing focused strategy; G+D+E: Value-decreasing diversi cation; B+G+F: Value-decreasing centralization CEO is more likely to choose the organizational structure that maximizes rm s value and less likely to diversify when that strategy is not value-maximizing. That 4s (p; ) is strictly increasing in ¼ follows from that this makes the return from the implementation of project more attractive relative to the implementation of project for shareholders, but not for the CEO since he gets no private bene ts from project. That 4s (p; ) does not decrease with m and does not increase with M follows from that an decrease in m andanincreaseinm decrease the range of values of s on which the CEO chooses a value-decreasing organizational structure; that is, s f (p) s c (p) decreases. The discussion around the diversi cation discount also revolves around the question of whether diversi cation destroys value or is just a measurement problem. As I already showed diversi cation destroysvalueforthose rmsthathaveasynergylevels2 (s c (p; ) ;s f (p; )) ; but it also true that even when diversi cation does not destroy value a diversi cation discount, as measured empirically, 23