Organizational Design and Environmental Volatility -...

Organizational Design and EnvironmentalVolatility

Heikki Rantakari�

University of Southern CaliforniaMarshall School of Business

This Version: October 28, 2008

Abstract

I investigate the impact of environmental volatility and the cost of informa-tion on the preferred organizational structure, as determined by the allocationof decision rights, the compensation structure of the managers and the degreeof operational integration, such as the use of shared distribution and marketingchannels across the operating divisions. The results are broadly consistent withthe common wisdom regarding the �t between organizational design and the en-vironment, with stable environments generally populated by tightly integratedand centralized organizations and volatile environments populated by looselyintegrated and decentralized organizations. The relationship between decentral-ization and volatility is, however, re�ned in two ways. First, the equilibriumrelationship between decentralization and volatility can be non-monotone, withdecentralization arising as the preferred governance structure in both highlystable and highly volatile environments. Second, even if decentralization andvolatility are generally positively associated in equilibrium, simply decentral-izing decision-making as a response to an increase in volatility will actuallyworsen organizational performance unless the other design parameters are alsoadjusted appropriately.

�Contact: [email protected]. Preliminary and incomplete. This paper is a heavily re-vised version of Chapter 2 of my Ph.D. Dissertation. I would like to thank my advisors RobertGibbons and Bengt Holmström for many helpful discussions and continuous encouragement. Thepaper has also bene�ted from comments by Tony Marino, Ricardo Alonso, the participants of ES-SET2008, ESAM08, 2007 SITE Summer Workshop, and the seminar participants at the Universityof California, San Diego, the University of Michigan, Stanford University and Duke University. Allremaining errors are my own.

1

presented by Heikki RantakariFRIDAY, Nov. 7, 20081:30 pm - 3:00 pm, Room: HOH-302

USC FBE APPLIED ECONOMICS WORKSHOP

�Achieving high performance in a business results from establishing and maintaining

a �t among three elements: the strategy of the �rm, its organizational design, and the

environment in which it operates.�(Roberts, 2004:12)

1 Introduction

The need for a �t among a �rm�s strategy, structure and its operating environment has

been extensively discussed by management and strategy scholars at least since Chandler�s

Strategy and Structure (1962). A question that has received particular attention is how �rms

can reorganize their operations to remain competitive in increasingly turbulent markets.

Building on this literature, this paper analyzes from an agency-theoretic perspective how

environmental volatility and the cost of information about the environment in�uence the

strategic orientation of a �rm and the resulting choice of its organizational structure.

The dimension of strategy that I consider is the choice between (local) responsiveness and

(global) e¢ ciency.1 In short, a �rm can generate value both through customizing its products

and their marketing to meet varying and changing local tastes (increasing customer value)

and through large-scale manufacturing and standardization (reducing production costs). The

resulting strategic challenge faced by all �rms is �nding the right balance between the two.

As observed by Porter (1996:10): "Simultaneous improvement of cost and di¤erentiation is

possible only when a company begins far behind the productivity frontier or when the frontier

shifts outward. At the frontier, where companies have achieved current best practice, the

trade-o¤ between cost and di¤erentiation is very real indeed."

To implement its strategy, the �rm chooses its structure, which I take to consist of three

design parameters: the allocation of decision rights inside the organizational hierarchy, the

compensation structure of the managers and the level of operational integration, such as the

use of shared distribution and marketing channels among the operating divisions. The level

of operational integration re�ects the level of interdependence across the operating units and

directly in�uences the �rm�s ability to realize value through responsiveness and e¢ ciency.

For example, using shared components or undertaking coordinated product development (in-

creases in operational integration) can help to realize cost savings and synergies. To realize

these bene�ts, however, the behavior of the operating units needs to be increasingly coor-

1This particular terminology is most common in the literature on multinational corporations. See, forexample, Bartlett (1986), Prahalad and Doz (1987) and Bartlett and Ghoshal (1989). However, the basictradeo¤ is also present in the general strategy literature, such as the choice between di¤erentiation and costleadership (Porter, 1980).

2

dinated, resulting in a loss of responsiveness.2 Choosing minimal operational integration is

then equivalent to maximizing the potential for local responsiveness by leaving the operat-

ing units free to adapt to changes in local conditions without concern for interactions with

the rest of the �rm. Increasing the level of operational integration increases the potential

e¢ ciency of the operations but reduces the �rm�s responsiveness.

Given the choice of operational integration, how much of the potential value is actually

realized depends on the remaining two choice variables: the allocation of decision rights

and the compensation structure of the managers. Since the level of operational integration

in�uences both the nature and severity of agency con�icts inside the organization, each level

of operational integration is best managed through a particular allocation of decision rights

and a choice of compensation structure. And because the allocation of decision rights and

the choice of compensation structure in�uence the value actually realized at any given level

of operational integration, all three need to be determined simultaneously as the optimal

response to a given environment.

The organization I analyze consists of two operating divisions headed by self-interested

division managers and a pro�t-maximizing headquarters, while the environment is charac-

terized by its volatility and the cost of information. The organizational challenge is one

of coordinated adaptation: the division managers �rst acquire information about their lo-

cal conditions, then communicate that information strategically to the decision-maker(s)

and �nally the decision-maker(s) use the information available to them in choosing how the

divisions will respond.

The role of organizational structure is as follows. The choice of operational integration,

by specifying the balance between e¢ ciency and responsiveness, determines how much co-

ordination is needed between the divisional responses. The allocation of decision rights and

the compensation structure of division managers are used to manage information acquisi-

tion, communication and decision-making conditional on the level of operational integration.

With respect to decision-making, I focus on the choice between decentralization (decision-

making authority delegated to the division managers) and centralization (decision-making

authority retained by the headquarters). With respect to the compensation structure, I an-

alyze both the strength of incentives, as determined by the overall sensitivity of managerial

compensation to performance, and the composition of incentives, as determined by the rela-

tive weight placed on division- and �rm-level performance in the compensation contract. For

a given strength of incentives, the composition of incentives determines the balance between

2For example, in the context of multinational corporations, the strategy of local responsiveness used tobe exempli�ed by the highly autonomous operating units of Philips, while the strategy of global e¢ ciencyused to be exempli�ed by the highly centralized operations of Matsushita (now Panasonic). (Bartlett andGhoshal, 1989)

3

how much information is acquired by the division managers and how well that information

is transmitted and used in decision-making. The strength of incentives, in turn, in�uences

the amount of information acquired given the composition of incentives.

The results are broadly consistent with the common wisdom regarding the �t between

strategy, structure and the environment: volatile environments are characterized by orga-

nizations that pursue a strategy of local responsiveness through the combination of loosely

integrated operations, decentralized decision-making and strong divisional incentives, while

stable environments are generally populated by �rms that pursue a strategy of global e¢ -

ciency through the combination of tightly integrated operations, centralized decision-making

and the use of �rm-wide incentives.3 This pattern arises because of three e¤ects. First, the

equilibrium level of operational integration is decreasing in the volatility of the environ-

ment. Intuitively, an increase in volatility increases the value of local responsiveness, which

the organization achieves by reducing the level of operational integration. Second, loosely

integrated organizations are always best managed through a decentralized structure while

centralized decision-making is preferred only when the equilibrium level of operational inte-

gration is su¢ ciently high. Third, the use of �rm-wide incentives is generally increasing in

the level of operational integration under both governance structures because an increase in

interdependence generally leads to an increase in agency con�icts across the divisions.

The results also re�ne and qualify some of these broad patterns, in particular regarding

the relationship between volatility and decentralization. It is commonly argued that a de-

centralized structure is preferred in more volatile environments because it is able to respond

faster and make better use of local information than a centralized structure.4 This logic

is re�ned on two fronts. First, the equilibrium relationship between volatility and decen-

tralization can be non-monotone. This possible non-monotonicity follows from two results.

First, as the cost of information decreases, the relative performance of decentralization in-

creases. Second, the relative advantage of centralization is largest at intermediate levels of

operational integration, where the equilibrium decisional bias under decentralization is the

largest. As a result, for intermediate costs of information, the preferred structure for a very

stable environment is a tightly integrated but decentralized organization, and an increase in

volatility, while leading to a reduction in operational integration, also leads to centralization

of decision-making.

Second, even when decentralization and volatility are positively associated in equilib-

3For classic contributions, see, for example, Lawrence and Lorsch (1967), Galbraith (1973,1977) andMinzberg (1979). In the context of MNCs, see, for example, Prahalad and Doz (1987) and Bartlett andGhoshal (1989). For recent contributions, see, for example, Brickley et al (2003) and Roberts (2004).

4And there exists some empirical evidence supporting a positive relationship between decentralizationand volatility. See, for example, Nagar (2002) and Wulf (2006).

4

rium, volatility is not directly causing decentralization in the present model. Instead, an

increase in volatility causes a reduction in operational integration, which in turn increases

the relative bene�ts of decentralization. Indeed, given the equilibrium level of operational

integration, the rest of the equilibrium structure is independent of volatility. As a result,

the positive association between volatility and decentralization arises only as a part of the

overall adjustment in organizational design in response to a change in the environment. Sim-

ply decentralizing decision-making as a response to an increase in volatility would actually

worsen organizational performance, unless the other design parameters are also adjusted

appropriately.

The remainder of the paper is structured as follows. Section 2 reviews the related liter-

ature and section 3 outlines the model. Section 4 derives the expected performance of the

organization as a function of the environment and the design parameters. Section 5 analyzes

the optimal choice of the design parameters and the link between the organization and the

environment. Section 6 concludes.

2 Related Literature

The model builds directly on the framework developed in Alonso, Dessein and Matouschek

(2008) and Rantakari (2008a). However, instead of focusing on the role that the allocation

of decision rights plays in managing communication and decision-making in organizations,

I focus on the interactions among di¤erent organizational design parameters and their joint

�t with the environment. Because of this integrative nature of the framework, the analysis

of the present paper is related to several di¤erent literatures.

A number of papers analyze complementarities among various subsets of organizational

design parameters. Milgrom and Roberts (1990,1995) examine complementarities among

di¤erent features of modern production technologies but pay only limited attention to the or-

ganizational structure used to govern that production. Holmström and Milgrom (1991,1994)

analyze the provision of incentives for multiple tasks and extend those results to account for

interactions among the level of incentives, asset ownership and job restrictions.5 However,

their primary focus is the management of the behavior of a single agent operating a �xed

technology and they don�t explicitly analyze the role of decision-making and communication.

This paper builds on both strands, by considering the interdependence between the choice

of operating technology and the choice of organizational structure that is used to manage

that technology.5See also Holmström (1999)

5

The papers most closely related to mine both in approach and content are Friebel and

Raith (2007), Dessein, Garicano and Gertner (2007) and Athey and Roberts (2001), each of

which looks at the simultaneous determination of incentives and decision-making authority

from alternative angles. Friebel and Raith (2007) analyze a resource allocation problem,

where divisional managers need to be motivated to exert e¤ort to generate high-quality

projects and then to communicate that information (truthfully) to the headquarters. Des-

sein, Garicano and Gertner (2007) analyze a synergy implementation problem, where again

the managers need to be motivated to exert productive e¤ort but also have private informa-

tion regarding the costs and bene�ts of implementing synergies. Athey and Roberts (2001)

combine the problem of inducing productive e¤ort with a project selection problem. In

all papers, the basic trade-o¤ is between providing focused incentives to induce e¤ort and

balanced incentives to induce truthful communication and/or appropriate decision-making,

with the allocation of decision rights impacting this trade-o¤.

The incentive provision problem in my setting also faces the basic tension between focused

incentives to motivate information acquisition and balanced incentives to motivate accurate

transmission and use of that information. However, by examining a di¤erent problem, some

of the insights and results di¤er. The strategic nature of communication limits the value of

an uninformed principal relative to Athey and Roberts (2001). By looking at moral hazard

in information acquisition, I can analyze the links among the value of information, the value

of incentive alignment and the allocation of decision rights, an issue that doesn�t arise in

Dessein, Garicano and Gertner (2007). Some of these links are present in Friebel and Raith

(2007), but the frameworks are qualitatively di¤erent and yield di¤erent predictions. For

example, in their model, decentralization is always associated with zero incentive alignment

because if interim reallocation of resources is desired, the headquarters is always in a better

position to do so. In contrast, in my model, decentralization combined with �rm-wide

incentives can outperform centralization. Finally, none of these papers consider the choice

of strategy and the resulting level of operational integration as an endogenous variable.

The role of authority and delegation in managing agency problems is also examined in

a number of other papers. Building on the cheap talk literature that has followed Crawford

and Sobel (1982), Dessein (2002), Harris and Raviv (2005) and Alonso (2007), for example,

examine how the allocation of decision rights can be used to manage the trade-o¤ between

biased decisions and information losses due to strategic communication.6 Aghion and Tirole

(1997) illustrate how delegation can be used as a motivational tool by allowing the agent

to freely use the information he learns. My framework embeds both aspects of the problem

and joins them with the possibility of using monetary incentives, which allows us to examine

6See also Stein (1989) and Melumad and Shibano (1991)

6

the link between delegation and incentives. Rantakari (2008b) examines the impact of noisy

performance measurement on the willingness to delegate and Zabojnik (2002) analyzes the

motivational impact that delegation has on the implementation e¤ort by the agent.

Organizational structures have also been analyzed from various other angles. The paper

closest to mine is Dessein and Santos (2006), who analyze, in a team-theoretic model, the

limitations that the need for coordinated adaptation imposes on task specialization. Coor-

dination in their model is, however, constrained only because information transmission is

exogenously imperfect. Some further perspectives include information processing (for ex-

ample, Marshak and Radner, 1972, Bolton and Dewatripont, 1994), problem-solving (for

example, Garicano, 2000), screening for interdependencies (Harris and Raviv, 2002) and

coordination and experimentation (Qian, Ronald and Xu, 2006).

Finally, while the economic literature on organizational design is still relatively young,

there is a long history of management and strategy scholars that have analyzed the topic of

this paper. As a result, this paper owes an intellectual debt to a long string of contributions,

including Simon (1947), Chandler (1962,1977), Woodward (1965), Lawrence and Lorsch

(1967), Thompson (1967), Galbraith (1973,1977), Mintzberg (1979) and Porter (1980),

among many others, in particular the later works of Prahalad and Doz (1987), Bartlett

and Ghoshal (1989), Nadler and Tushman (1997), Brickley et al (2003) and Roberts (2004).

3 The Model

The organization I consider consists of two divisions, each managed by a strategic division

manager (he), and headquarters (she), who aims to maximize the overall pro�tability of the

organization. This section outlines the payo¤s, available actions and the timing of events in

detail.

Divisional pro�ts and alternative governance structures: The organization consistsof two (symmetric) divisions, i and j: The pro�tability of each division depends on both

how well the activities of the division are aligned with its local conditions and how well

the divisions are coordinated with each other. Given the decisions di and dj regarding the

operations of divisions i and j, respectively, the ex post pro�t of division i is given by

�i (�i; di; dj) = K (�)� � (dj � di)2 � � (�i � di)2 ;

where �i � U��; �

�indexes the locally optimal decision for division i; with �i and �j

7

independently distributed and � > 0 measuring the volatility of the environment. The align-

ment of the division with its local conditions is then measured by � (�i � di)2 ; while thealignment with the other division is measured by � (dj � di)2 : The realized pro�ts of theorganization are given by �i + �j:

The �rst choice variable for the organization is the level of operational integration � 2[0;1); which measures the level of interdependencies across the organization�s activities, asdetermined by the extent to which the organization uses shared components, manufacturing

facilities, distribution networks, sales forces, coordinated product development e¤orts and

the like across the divisions. The bene�ts of operational integration come from the poten-

tial reduction in operating costs that results form the elimination of duplicated assets, the

increased scale of the remaining operations and the realization of potential synergies across

activities. These bene�ts are captured by an increasing and continuous function K (�) : To

fully realize these bene�ts, however, the behavior of the divisions needs to be increasingly

coordinated. For example, the e¢ ciency of a common sales force is compromised if con�ict-

ing demands are placed on it by the two divisions, introducing just-in-time manufacturing

and inventory management relies on smooth functioning of the supply chain and the value of

standardized production facilities is reduced if the products manufactured require con�icting

customization. This induced value of coordination is captured by � (dj � di)2 :To summarize, the argument is two-fold. First, the level of operational integration, which

essentially re�ects the underlying con�guration of productive assets, is a choice variable.7

Second, when choosing its level of operational integration, the organization faces the fol-

lowing trade-o¤: an increase in operational integration increases the potential e¢ ciency of

operations, but this reduces the �exibility of the divisions in responding to local conditions.

The second choice variable for the organization is the allocation of authority over the

divisions, captured by the right to make the decisions di and dj. I consider two alternative

arrangements. Under centralization, the headquarters retains control of both decisions, while

under decentralization, control over the divisions is delegated to their respective division man-

7As observed by Porter (1990:17):"A �rm faces an array of options in both con�guration and coordinationfor each activity. Con�guration options range from concentrated (performing an activity in one location andserving the world from it - e.g., one R&D lab, one large plant) to dispersed (performing each activity in eachcountry). In the latter case, each country would have a complete value chain. Coordination options rangefrom none to very high. For example, if a �rm produces its product in three plants, it could, at one extreme,allow each plant to operate with full autonomy �e.g., di¤erent product standards and features, di¤erent stepsin the production process, di¤erent raw materials, di¤erent part numbers. At the other extreme, the plantscould be tightly coordinated by employing the same information systems, the same production processes,the same parts, and so forth."

8

agers. I will use superscript g 2 fcent; decg to denote the two governance structures.8 ;9

Division managers: Each division is headed by a risk-neutral division manager (managersi and j, respectively). Their behavior, which consists of information acquisition, communi-

cation and, in the case of decentralization, decision-making, is controlled through the third

choice variable, which is their compensation structure. I assume that manager i is o¤ered a

linear incentive contract

Ti (�i; �j) = Ai + sii�i (�i; di; dj) + sij�j (�j; di; dj) ;

where (sii; sij) are the weights placed on the pro�tability of the two divisions. We can

normalize this contract by rewriting it as

Ti (�i; �j) = Ai + � (s�i (�i; di; dj) + (1� s)�j (�j; di; dj)) ;

where � = sii + sij measures the strength of incentives and s 2�12; 1�measures the de-

gree of incentive alignment.10

To relate this formulation to the composition of incentives, note that this contract is iden-

tical to a contract that would place a weight 2� (1� s) on �rm-wide performance �i+�j andan additional weight � (2s� 1) on divisional performance �i: The closer s is to 1=2; the moreweight the contract places on �rm-wide performance and, as a result, the more aligned the in-

terests of the divisional managers are both with each other and with the organizational goals.

Timing of events: The organizational design parameters (g; �; s; �) are used to man-

age the unfolding of events summarized in �gure 1. First, the division managers invest

in acquiring information about the local conditions faced by their divisions. In particular,

manager i acquires a signal ti of the realized state �i that is correct with probability qi and

a random draw from U��; �

�with probability 1 � qi at a personal cost of C

��; qi

�. The

8For brevity, we will only analyze symmetric divisions. Because of the assumed symmetry, asymmetricgovernance structures do not arise as equilibrium governance structures. For asymmetric divisions, thequalitative logic of Rantakari (2008a) regarding the relevant tradeo¤s continues to hold. Results are availablefrom the author on request.

9We thus assume that delegation is credible, an aspect which can be supported through both resourceallocation and legal standing. For example, in the context of MNCs, there are several examples where thenational subsidiaries have utilized both their direct control over national resources and their separate legalincorporation to explicitly disobey the instructions of the parent organization. One of the more �agrantcases of disobedience was the refusal by Philips North America to adapt the internally developed V2000videocassette standard, and instead decided to sell a VHS product supplied by Matsushita, the archrival ofPhilips.10A symmetric contract turns out to be optimal because of the assumed symmetry of the divisions, so

assuming symmetry of compensation contracts from the beginning is without loss of generality.

9

The level of operational integration, theallocation of the decision rights and the

incentive structure are chosen

Division managers invest in information acquisition

Division managers communicate with thedecisionmaker(s) through one round of

simultaneous cheap talk

Decisions are made

Payoffs are realized

(i) Timing of events

i j

ti tjdi dj

mj

mi

HQ

pi pj

i j

ti tjdi dj

mjmi

HQ

pi pj

(A) Centralization (B) Decentralization

(ii) Alternative governance structures

Figure 1: The timing of events and the alternative governance structures.

manager does not learn whether the signal is correct or not, so that upon observing a signal

ti his posterior expectation about the local state is given by Ei (�ijti) = qiti:11 Forecastingthe results, the value of accuracy in terms of expected pro�ts will be quadratic in qi: As a

result, I de�ne pi = q2i and will refer to pi as the quality of primary information. The cost

of acquiring information is given by

C��; pi

�= ��

2

3(pi + ln(1� pi)) ;

where � parameterizes the (marginal) cost of information and �2=3 is the ex ante variance of

local conditions. This particular functional form simpli�es the solution to the information

acquisition problem without altering the qualitative nature of the solution.12 As a normal-

ization, I assume that the marginal cost of information is proportional to the volatility of the

environment. This assumption helps to provide a benchmark solution for the information

acquisition problem and to separate the e¤ects of volatility��and cost of information (�)

on the organizational design.

11I use this particular acquisition technology for two reasons. First, for purely technical reasons, thesmooth posterior facilitates the analysis. Second, for descriptive purposes, having a belief over the reliabilityof information appears for many settings more plausible than knowing whether a piece of information iscorrect or not.12As discussed in an earlier version of the paper, the logic of the results generalizes to any well-behaved cost

function C��; pi

�with, for � > 0; Cpi

��; pi

�� 0, Cp2i

��; pi

�> 0; lim

pi!1Cpi

��; pi

�=1; and C�pi

��; pi

�� 0:

10

Given the acquisition technology, the division managers choose pi to maximize their ex-

pected payo¤, E�Ti��gi ; �

gj

�� C

��; pi

�: I assume that pi is observable but not veri�able

to the organizational participants.13 Having acquired their private information, the divi-

sion managers strategically communicate their information to the decision-maker(s) through

one round of simultaneous cheap talk. In the case of centralization, communication occurs

vertically to the headquarters, while in the case of decentralization, communication occurs

horizontally between the division managers. Finally, after communication, the decision-

maker(s) choose their decisions conditional on their information. In the case of centralization,

the headquarters makes decisions to maximize E��centi + �centj

�conditional on the messages

(mi;mj) that she received in the communication stage. In the case of decentralization, the

division managers make decisions to maximize E�T deci

��deci ; �

decj

��conditional on their in-

formation, which consists of their private signal ti and the messages (mi;mj) exchanged in

the communication stage.

The organizational design problem and the cost of incentives: In the beginningof the game, the headquarters chooses the organizational design parameters to maximize

her payo¤. As a �nal observation, note that for the problem to be interesting and for the

governance structure to matter, it needs to be that incentive provision is costly to the head-

quarters. This result follows because if the organization faced no cost of providing incentives,

then the optimal solution would be to set � = 2 and s = 1=2; which would duplicate the full

pro�t margin for both divisional managers and thus achieve the pro�t-maximizing solution.

To simplify the analysis, I take a reduced-form approach to modeling this cost of incen-

tives and simply assume that the headquarters faces a cost �2

3G (�) of providing an incentive

contract of strength �; with G� (�) ; G�� (�) � 0; while being agnostic as to the particular

source of this cost. The scaling of the cost function with volatility turns out to be a convenient

normalization for isolating the various interactions that are present in the model.14 Finally,

given the cost of incentives, we can write the design problem as the headquarters choosing

the governance structure g; the compensation structure (s; �) and the level of operational

integration � to

maxg;�;s;�

Xk=i;j

�E (�gk (�; s; �) jp

gk (�; s; �))� C

��; pgk (�; s; �)

��G

��; ��:

13Observability of acquisition simpli�es the analysis signi�cantly while not altering the qualitative results.Some results on the case of non-observable information acquisition are available from the author on request.14This assumption, together with a discussion of the potential sources for the cost of incentives are discussed

in Appendix C.

11

4 Expected Pro�tability and Organizational Design

The �rst step in the analysis is to derive the expression for the expected pro�ts of the

organization conditional on the design parameters (g; �; s; �) : The solution follows through

backward-induction. I begin by discussing the equilibrium decisions (section 4.1), followed by

the communication equilibrium (4.2) and the resulting expected pro�tability of the divisions

conditional on the quality of primary information (4.3). These three steps are identical

to Alonso, Dessein and Matouschek (2008) and Rantakari (2008a), with the exception of

introducing imperfect primary information. I revisit these results here because the insights

underlying their determination play a key role in understanding the rest of the solution,

as the decision-making and communication equilibria determine how good use is made of

the information generated by the division managers.15 The remainder of section 4.3 then

discusses how the value of information depends on organizational design and completes the

expression for the expected pro�ts by analyzing the information acquisition problem. Section

5 analyzes the choice of the design parameters and the relationship between overall design

and the environment.

4.1 Equilibrium decisions

In the decision-making stage, the decision-maker(s) use the information available to them to

maximize their individual payo¤s, conditional on the accuracy�qgi ; q

gj

�of the signals (ti; tj)

obtained at the divisional level regarding the local conditions and the messages (mi;mj)

sent regarding the realization of those signals by the division managers in the communica-

tion stage.

Centralization: The information available to the headquarters (P ) consists of the mes-sages mi and mj sent by the division managers. She solves

maxdi;dj

E (�i + �jjmi;mj) ;

solution to which is given by

dcenti (s; �) =(�+2�)EP (�ijmi)+2�EP (�j jmj)

�+4�; where EP (�ijmi) = q

centi E (tijmi) :

15The discussion is, however, by necessity very brief. The reader interested in further details is advised toread either of the mentioned papers.

12

Conditional on the information available to the headquarters, these decisions are, by as-

sumption, pro�t-maximizing. If � = 0; the divisions are fully independent and the head-

quarters chooses di = EP (�ijmi) and dj = EP (�jjmj) : As the level of operational integration

increases, the increasing importance of coordination makes the decisions increasingly insensi-

tive to information regarding the division�s own local conditions while becoming increasingly

accommodating to the needs of the other division. As � ! 1; the decisions become per-fectly coordinated, with di = dj =

EP (�ijmi)+EP (�j jmj)

2:

Decentralization: The information available to manager i consists of his private signalti and the messages exchanged in the communication stage. Given the compensation con-

tract (s; �) ; he solves

maxdi�E (s�i + (1� s)�jjti;mi;mj)

and similarly for manager j: The equilibrium decisions are given by

ddeci (s; �) = s�(s�+�)

Ei (�ijti) + �(s�+2�)

Ei [Ej (�jjtj) jmj] +�2

(s�+�)(s�+2�)Ej [Ei (�ijti) jmi] ;

where Ei [Ej (�jjtj) jmj] = qdecj E (tjjmj). As long as s > 1 � s; the division managersplace, from the organization�s perspective, excessive weight on the performance of their own

division when making decisions. I will refer to this suboptimal use of available information

as the quality of decision-making, taking both the accuracy of the signals and the accuracy of

their transmission as given. The key feature of the solution, which later plays an important

role in understanding the choice of governance structure, is that the payo¤ consequences

of this bias are non-monotone in the level of operational integration. In particular, despite

the potential own-division bias of division managers, the equilibrium decisions converge to

the pro�t-maximizing decisions both when � ! 0 and when � ! 1: In the �rst case, thedivisions are fully independent and, as a result, no coordination is needed between the divi-

sions. In the second case, since the divisional payo¤s are fully dependent on coordination,

the managers are willing to coordinate their behavior even absent any �rm-wide incentives.

However, whenever some interdivisional con�ict is present and � is interior, the own-division

bias leads the managers to choose decisions that exhibit too much adaptation (and too lit-

tle coordination), with the bias being largest when the relative importance of coordination

�=(� + �) is intermediate (when the tension between adaptation and coordination is the

largest). Finally, the impact of incentive alignment is immediate: the quality of decision-

making is monotonically improving in the relative weight placed on �rm-wide performance

and converges to pro�t-maximizing when only �rm-level incentives are used.

13

4.2 Equilibrium communication

The communication stage is modeled as a cheap-talk game between the privately informed

division managers and the decision-maker(s). Knowing how the equilibrium decisions depend

on the beliefs of the decision-maker(s), the division managers send simultaneously non-

veri�able messages regarding their local information in an attempt to induce more favorable

decisions.16 Of course, in equilibrium, such attempts to mislead the decision-maker(s) are

futile and only lead to garbling of information. As a result, the equilibrium characterized by

a partition structure, where the division managers are able to reveal only that their signal

lies within a certain interval of the state space.

A natural measure of the accuracy of information transmission is the expected variance of

the beliefs of the recipient over ti conditional on the received messagemi; E (E (tijmi)� ti)2 :We can characterize this variance (and so the coarseness of the most informative partition)

through a single coe¢ cient 'gi (s; �) ; which depends on the particular governance structure

g; the degree of incentive alignment s and the level of operational integration �: I will refer

to 'gi (s; �) 2 (0;1) as the quality of communication (given the accuracy of the managers�private signals ti and tj). As '

gi (s; �) increases, communication becomes more accurate and

becomes perfectly informative as 'gi (s; �)!1:

Centralization: In the case of centralization, communication is from the division man-

agers to the headquarters (vertical). The quality of communication is given by

'centi (s; �) = s�+��(2s�1) 2 [1;1) ; with

@'centi (s;�)

@�;@'centi (s;�)

@s< 0:

If � = 0 (so that there is no operational integration); then the equilibrium is fully infor-

mative. In this case, the headquarters makes decisions that are fully responsive to local

information without any concern for coordination, thus replicating the preferences of the

division managers for any s. However, whenever � > 0 and s > 1=2; the own-division bias of

the division managers leads them to prefer decisions that are more adapted to local condi-

tions than what the headquarters will actually implement. This vertical con�ict is increasing

in the level of operational integration (for a given s) and, as a result, the quality of vertical

communication is monotonically decreasing in �: The quality of communication is naturally

improving in degree of incentive alignment and becomes perfect when s! 1=2 for all �.

16Simultaneous communication is preferred over sequential communication. The results are available fromthe author on request.

14

Decentralization: In the case of decentralization, communication is between the divisionmanagers (horizontal). The quality of communication is given by

'deci (s; �) = �+s(1�s)�(s�+�)(2s�1) 2 [0;1) ; with

@'deci (s;�)

@�> 0;

@'deci (s;�)

@s< 0:

In contrast with the solution under centralization, the quality of communication under de-

centralization is at its worst when � = 0 and improving in the level of operational integration.

This result re�ects the di¤erential use of information under the two governance structures.

Under decentralization, the division manager is in control of the operations of his division.

As a result, instead of needing to persuade the headquarters about the adaptive needs of

his division, he is only trying to persuade the other division manager to be more accom-

modating to what he plans to do. As the level of operational integration is increased, the

division managers become increasingly responsive to each other�s behavior, which reduces

the horizontal con�ict between the managers and thus allows for more accurate communica-

tion. The impact of incentive alignment is naturally the same as under centralization, with

communication becoming perfectly informative as s! 1=2 for all �.

Finally, note that so far we have only characterized the accuracy of information transmis-

sion and not its value. Intuitively, the value of accurate transmission will depend on both the

governance structure and the level of operational integration. In the case of centralization,

communication is needed for the headquarters to be able to adapt her decisions to local

conditions. As a result, the value of accurate communication will be highest for low levels of

operational integration, when the decisions will be most responsive to that information. In

the case of decentralization, the division managers already have access to local information

but communication is needed to coordinate the responses. As a result, the value of accurate

communication will be highest for high levels of operational integration.

The net e¤ect of the value and endogenous accuracy of communication is such that for

any s > 12; the loss due to strategic communication is almost always increasing in the level

of operational integration under both governance structures, with centralization generally

more dependent on accurate communication.17 In contrast, the loss due to biased decision-

making under decentralization was non-monotone in �: The next subsection analyzes in more

detail how these distortions depend on the organizational structure and thus impact relative

organizational performance.

17The exception occurs under decentralization when s is large. In this case, for high �; a further increasein � can lead to a su¢ cient improvement in decision-making that the loss due to strategic communicationactually decreases.

15

4.3 Expected pro�ts

Having derived the equilibrium decisions�dgi (s; �) ; d

gj (s; �)

�and the equilibrium quality

of communication�'gi (s; �) ; '

gj (s; �)

�; and for now assuming that the quality of primary

information generated by the structure (g; �; s; �) is given by�pgi (:); p

gj (:)�; we can write the

expected pro�ts of the organization as

E

Xk=i;j

�gk(s; �; �)

!=Xk=i;j

�K (�)� ��

2

3+ pgk(:)

gk (s; �)

�2

3

�:

The �rst component, K (�) � ��2

3; gives the expected pro�ts conditional on a given level

of operational integration � and no information acquisition (pgi (:) = pgj (:) = 0). The second

component, pgk(:)gk (s; �)

�2

3; measures the impact that informative local signals have on the

expected pro�tability in terms of improved adaptation to local conditions, where pgk(:) � 0 isthe quality of information acquired by manager k and gk (s; �) > 0 measures the (marginal)

value of that information. This expression captures the basic logic underlying the solution.

Given its level of operational integration, the organization is able to improve its responsive-

ness by either inducing more information acquisition or by using the existing information

better. The next steps are then to analyze how the value of information gk (s; �) depends

on the organizational structure and to endogenize the quality of primary information, pgi (:) :

4.3.1 Value of information: gi (s; �)

The value of information is determined by the improvement in the equilibrium responsiveness

of the organization that is caused by a given increase in the quality of primary information.

The determinants of this value (for signal ti) are illustrated in �gure 2. If � = 0 (zero

operational integration), then the value of information is simply the value of adaptation, �;

which is realized by division i:

When � > 0; this value is reduced for two reasons. First, division i becomes less respon-

sive to information relative to � = 0: I denote this reduction in value by �gi (s; �) and it is

composed of three e¤ects. First, even under pro�t-maximizing decisions and perfect trans-

mission of information, the decisions will be less adaptive because of the positive value of

coordination. This e¤ect is given by �FBi (�) and it is increasing in �: Second, under decen-

tralization, the equilibrium decisions are biased (as long as s > 1=2), which also reduces the

value of information. This e¤ect is captured by ��gi (�; s) and, as argued in section 4.1, this

loss is non-monotone in �: Third, some of the information is lost due to strategic commu-

nication. This e¤ect is given by �gii (s; �)V ('gi (s; �)) ; where the �rst component measures

16

ϕ (β,s)Ψ (β,s) α ϕ (β,s)_ _=value of adaptation negative externality information imposes on division j due to β > 0

reduction in the value of information to division i due to β > 0

Λ (β) + ∆Λ (β,s) + Γ (β,s)V(ϕ (β,s))

loss due to the reduction in theresponsiveness of profitmaximizingdecisions due to β > 0

loss due to biaseddecisionmaking

accuracy of communicationvalue of accurate communication

loss due to strategic communication

ig

ig

jg

iFB

ig

iig

ig

Figure 2: Determinants of the value of information

the value of accurate communication and the second component measures the equilibrium

quality of communication. As argued in section 4.2, this loss is generally increasing in �

under both governance structures.

Second, and which is another key feature of the solution, while information is always

valuable to the division doing the acquisition, it also weakly reduces the pro�tability of the

other division. Intuitively, the more accurate information a division generates, the more

adaptive it will be. When the level of operational integration is positive, this increase in

adaptiveness requires increasing accommodation by the other division, which happens at

the expense of its own adaptiveness. I denote this negative externality by �gj (s; �) and it

is increasing in � under both governance structures, re�ecting the increasing importance of

coordination.

The net e¤ect of these forces and their dependency on (g; s; �) is illustrated in �gure

3, which plots the value of information for both centralization and decentralization as a

function of (s; �), together with the di¤erence between the two.18 The value of information

is monotonically decreasing in � under both governance structures. When s = 1=2 (so that

there is no agency problem in the use and transmission of information), this decrease in

value is due to the reduced adaptiveness of the pro�t-maximizing decisions. For s > 1=2; the

value of information is further reduced because of the loss of information due to strategic

communication and, in the case of decentralization, biased decision-making. Because both

of these losses are increasing in the level of interdivisional con�ict, the value of information

is also monotonically decreasing in s under both governance structures.

Panel (iii), in turn, illustrates which governance structure makes relatively better use of

the existing information for a given choice of (s; �).19 The key result to understanding the

18Because the value of information depends only on equilibrium decisions and communication, it is inde-pendent of �; which only a¤ects the amount of primary information.19See Alonso, Dessein and Matouschek (2008) and Rantakari (2008a) for a more detailed elaboration of

17

0

1

(i) Value of informationunder centralization

(ii) Value of informationunder decentralization

(iii) Difference in thevalue of information

Ψcent(s,β) Ψdec(s,β) Ψcent(s,β)Ψdec(s,β)

β/(α+β)1

1/2

s 0

1

β/(α+β)1

1/2

s0

1

β/(α+β)1

1/2

s

Figure 3: The relationship between organizational design and the value of information.

equilibrium choice of governance structure is that centralization makes better use of existing

information if and only if the level of operational integration is su¢ ciently high and the

degree of incentive alignment is su¢ ciently low. The intuition for this result follows from two

observations. First, recall that centralization is more dependent on accurate communication

than decentralization, while the loss due to strategic decision-making under decentralization

is non-monotone in � (with the equilibrium decisions converging to pro�t-maximizing when

� ! 0 or � ! 1). As a result, the relative advantage of centralization is largest for

intermediate levels of operational integration (when the quality of decision-making is at its

worst). For su¢ ciently low levels of operational integration, the improvement in the quality

of decision-making is never su¢ ciently large to outweigh the increased loss due to strategic

communication. Second, incentive alignment improves the quality of decision-making at a

faster rate than it improves the quality of communication. As a result, as incentive alignment

is increased, the remaining agency losses are increasingly due to strategic communication,

making decentralization increasingly attractive.

4.3.2 Managerial information acquisition: pgi (:)

The �nal step of the game is to solve for the equilibrium level of information acquisition.

Using the notation from above, de�ne

egi (s; �) = s (�� gi (s; �))� (1� s)�gj (s; �) > 0this result.

18

as the perceived value of information, re�ecting the di¤erential weights placed by the man-

ager on the performance of the two divisions. Given (�; s; �) ; manager i solves

maxpi�piegi (s; �) �23 � C ��; pi� ;

which gives us, using the assumed functional form C��; pi

�= ��

2

3(pi + ln(1� pi)) ; the

equilibrium quality of primary information

pgi (�; s; �; �) =�egi (s;�)�egi (s;�)+� :

The properties of the solution are summarized in the following proposition:

Proposition 1 Properties of pgi (�; s; �; �):

(i) ; @pgi (�;s;�;�)

@�> 0;

@pgi (�;s;�;�)

@�< 0;

@pgi (�;s;�;�)

@s> 0 and @pgi (�;s;�;�)

@�< 0

(ii) pdeci (�; s; �; �) � pcenti (�; s; �; �) unless both � and s are su¢ ciently large, in which

case pcenti (�; s; �; �) � pdeci (�; s; �; �)

Part (i) of the proposition illustrates the properties shared by both governance structures.

First, the quality of information acquired is naturally increasing in the strength of incentives.

Second, the quality of information is decreasing in the level of operational integration because

increasing � reduces the value of information. Third, the quality of information is increasing

in the degree of incentive con�ict. This result follows because increasing s leads the manager

to put more weight on the (positive) value realized by his division while placing less weight

on the negative externality imposed on the other division, an e¤ect which always outweighs

the associated reduction in the true value of information. It is thus possible to motivate

information acquisition both through the strength of incentives and through the degree

of con�ict between the division managers. Finally, the quality of information is naturally

decreasing in the cost of information. Because I have assumed that the marginal cost of

information is proportional to environmental volatility, the quality of information acquired

is independent of the level of volatility, which allows me to separate the e¤ects of � and

� on the preferred design. Having separated the e¤ects �rst, it is then straightforward to

understand the impact of any co-movements of � and �2:

Part (ii) of the proposition summarizes how the quality of information acquired depends

on the governance structure. It states that, for a given (�; s; �), the quality of primary

19

information acquired is higher under decentralization unless both the level of operational in-

tegration and the degree of incentive con�ict between the division managers are su¢ ciently

high, in which case the opposite holds. This result re�ects the di¤erence in the perceived

value of information under the two governance structures, which is determined by the dif-

ference in the true value of information (discussed above) and the di¤erence in the size of

the negative externality. This negative externality is always higher under decentralization,

so that decentralization is able to generate more information even when the true value of

information is lower. However, when both � and s are su¢ ciently large, the di¤erence in the

true value dominates and centralization is able to generate more information.

5 Choice of Organizational Design

Having derived the expression for the expected pro�ts of the organization as a function of

the design parameters, we can now analyze the organizational design problem. Substituting

the missing components into the net surplus function and utilizing the symmetry between

the divisions, we can write the design problem as

max�

�K (�)� ��

2

3+ �

2

3maxg;s;�

(pgi (�; s; �; �)gi (s; �)� C (�; p

gi (�; s; �; �))�G (�))

�s.t. pgi (�; s; �; �) =

�egi (s;�)�egi (s;�)+� :

We can thus view the design problem as composed of two steps. First, for any given level of

operational integration, choose the governance and compensation structures (g; s; �) to max-

imize organizational performance. This step maximizes the net surplus that the organization

can realize through local responsiveness for any given level of operational integration and

the cost of information and is analyzed in section 5.1. Note that, importantly, this solution

is independent of the level of volatility. Intuitively, given the level of operational integra-

tion, the optimal way to acquire and use information is independent of the overall value of

that information. Second, choose the level of operational integration to maximize overall

performance. Given the solution to the �rst stage, this step simply balances the bene�ts of

operational integration in terms of the improved potential e¢ ciency of the operations with

the cost of integration in terms of the lost local responsiveness under the pro�t-maximizing

compensation and governance structures. This solution is discussed in section 5.2, together

with the link between overall organizational design and environmental volatility. To illustrate

20

the equilibrium, I will solve the model numerically. The bene�ts of operational integration

and the cost of incentives are given by

K (�) = K

�1� e��1

�r�2

��r�3

��and G (�) = B�

(1��) ;

The qualitative results are not a¤ected by the particular functional forms used.20

5.1 Choice of (sg; �g) and g

Consider �rst the choice of compensation structure given the level of operational integration

and the governance structure. The �rst-order conditions characterizing the solution are

�g :�gi (s; �)� �egi (s; �)� @pgi (:)

@�= G� (�)

sg :�gi (s; �)� �egi (s; �)� @pgi (:)

@s+ pgi (:)

@gi (s;�)

@s= 0:

Because the only role of incentive strength (�g) is to motivate information acquisition, the

choice of �g simply equates the marginal bene�t of the information generated with the mar-

ginal cost of incentives. The choice of incentive alignment (sg), on the other hand, balances

two e¤ects. Increasing the degree of incentive alignment improves the use of existing informa-

tion and thus increases the value of information (@gi (s; �) =@s < 0) but this improvement

comes at the cost of reducing the quality of information acquired (@pgi (:) =@s > 0). As to the

impact of the cost of information and the level of operational integration on these choices,

we can make the following observations:

Cost of information: A reduction in the cost of information increases the equilibrium

degree of incentive alignment while having an ambiguous impact on the strength of incen-

tives. Intuitively, a reduction in the cost of information increases the amount of information

acquired, which increases the value of incentive alignment while allowing the organization

to economize on incentive provision. However, because the increase in incentive alignment

increases the value of information while reducing the amount of information acquired by the

manager (changes which increase the value of incentives), the net impact on the strength of

incentives is ambiguous.

Level of operational integration: An increase in the level of operational integration has

a generally ambiguous impact on the compensation structure. First, increasing the level

of operational integration decreases the value of information, thus decreasing the value of

20The particular parameterization used for the bene�t function is A = 4; �1 = 2:5; �2 = �3 = 0:5 and� = 1:2: For the cost of incentives function, B = 0:01:

21

incentives. Second, increasing the level of operational integration changes the agency losses

in the use of information. As discussed in section 4.3, these losses are increasing in the

level of operational integration under centralization while being non-monotone under decen-

tralization. The equilibrium degree of incentive alignment then generally parallels the size

of these agency con�icts, thus being increasing in the degree of incentive alignment under

centralization while being non-monotone under decentralization. The equilibrium strength

of incentives, in turn, balances the decreasing value of information with the variation in the

managerial information acquisition resulting from changes in the equilibrium degree of in-

centive alignment, resulting in a non-monotone relationship between the level of operational

integration and the equilibrium strength of incentives.21

The equilibrium compensation structure (sg; �g) determines for each governance structure

the net value that it can realize through local responsiveness. The choice of governance

structure is then straightforward: choose the governance structure that yields a higher net

value. The dependency of this choice on � and � is summarized in the following proposition:

Proposition 2 Choice of governance structure: Centralization is preferred over decen-tralization if and only if the level of operational integration and the cost of information are

su¢ ciently high

Proof. See Appendix A

This result follows directly from the discussion in section 4.3 on how the two governance

structures di¤er in their ability to generate and use information. When the level of opera-

tional integration is su¢ ciently low, a decentralized structure is able to both generate more

information and make better use of that information for a given compensation structure.

When the compensation structure is endogenous, it is then further able to economize on

incentive provision. For higher levels of operational integration, the choice of governance

structure depends on the cost of information. In particular, as the cost of information goes

down (and as a result, the equilibrium degree of incentive alignment goes up), decentraliza-

tion becomes increasingly attractive because of the di¤erential impact of incentive alignment

on the quality of decision-making and communication.

The equilibrium to this �rst stage of the design problem is illustrated in �gure 4, which

plots the equilibrium compensation structure under the two governance structures, together

with the preferred governance structure, for two levels of the cost of information.22 The

21The relationships among s; � and � are discussed more extensively in Appendix B22� = 0:13 and � = 0:117:

22

0

0.8

0

0.8

Centralization

Decentralization

level of operational integration, β/(α+β)

level of operational integration, β/(α+β)

0 1

0 1

centdec

cent decdec

firmlevel incentives, 2λg(sg1)

divisionlevel incentives, λg(2sg1)



centralization (cent)decentralization (dec)

Figure 4: Choice of compensation and governance structures as a function of � and �

key point to note is that, re�ecting the di¤erence in the agency costs, the di¤erence in the

equilibrium compensation structure under the two governance structures is largest for inter-

mediate levels of operational integration. In particular, centralization is able to economize

on incentive provision by using interdivisional con�ict instead of incentive strength as the

tool for motivating information acquisition.

This result, in turn, leads to our �rst main conclusion: The relationship between the

level of operational integration and the preferred governance structure can be non-monotone.

When the cost of information is high, then the ability of centralization to economize on

incentive provision is su¢ ciently valuable and it is preferred for all levels of operational

integration beyond some cuto¤ level. As the cost of information is reduced, the value of this

advantage is reduced and decentralization is eventually preferred for all levels of operational

integration. The preference for centralization, however, remains the longest for intermediate

levels of operational integration, where its relative advantage is the largest. As a result, for

intermediate costs of information, decentralization is preferred for both low and high levels

of operational integration while centralization is preferred for the intermediate levels.

23

5.2 Choice of � and the relationship between organizational design

and the environment

Having solved for the equilibrium compensation and governance structures (sg; �g; g) as a

function of the level of operational integration � and the cost of information �; we can then

consider the choice of operational integration. Let

N g (�; �) = maxs;�pgi (:)

gi (s

g; �)� C (�; pgi (:))�G (�)

denote the net value that each governance structure is able to realize through local re-

sponsiveness for any given (�; �) and let

M (�; �) = maxgN g (�; �)

be the upper envelope of the two functions. The optimal choice of operational integration is

then characterized by

K 0 (�) + �2

3dd�M (�; �) = 0;

which simply equates the marginal improvement in the potential e¢ ciency of the opera-

tions with the marginal cost of lost local responsiveness. The properties of this equilibrium

choice are summarized in the following proposition

Proposition 3 The choice of operational integration: The level of operational in-tegration is decreasing in the volatility of the environment while increasing in the cost of

information. Whenever a reduction in operational integration is associated with a change in

governance structure, this reduction is discrete.

Proof. See Appendix A

The intuition behind this result is straightforward. An increase in the volatility of the

environment increases the value of responsiveness while leaving the bene�ts unchanged, thus

leading to a reduction in the level of operational integration. Similarly, an increase in

the cost of information reduces the amount of information acquired in equilibrium, which

decreases the responsiveness of the organization and thus the cost of operational integration.

As mentioned earlier, the only direct link between organizational design and environmen-

tal volatility is through the level of operational integration. However, because the choice of

24

0

1centralization

volatility, θ 2_

0

0.8

level of operationalintegration, β/(α+β)

volatility, θ 2_



centralization decentralization centralization decentralization

Figure 5: Organizational design and environmental volatility (1)

operational integration impacts the rest of the organizational structure, changes in volatility

indirectly impact both the optimal compensation and governance structures. This relation-

ship is illustrated in �gure 5 for a relatively high cost of information. The equilibrium out-

come in this case is broadly consistent with the accepted wisdom regarding the relationship

between organizational design and environmental volatility.23 First, stable environments are

characterized by organizations that pursue a strategy of global e¢ ciency through the com-

bination of tightly integrated operations, centralized decision-making and relatively heavy

use of �rm-wide incentives to manage the interdependencies. As the environment becomes

more volatile, the organization reduces the level of operational integration to increase its re-

sponsiveness, increases the use of division-level and decreases the use of �rm-level incentives

to motivate additional information acquisition and eventually decentralizes decision-making.

As a result, volatile environments are then characterized by organizations that pursue a

strategy of local responsiveness through the combination of loosely integrated operations,

strong divisional incentives and decentralized decision-making.

While the equilibrium outcome in this case is consistent with the accepted wisdom, the

results also shed additional light on the much-discussed relationship between decentralization

and volatility. It is generally argued that an increase in uncertainty increases the preference

for decentralization because decentralization is able to to make better use of local information

and to respond faster to changes in the local conditions, and there is increasing empirical

evidence of a positive association between decentralization and volatility (discussed below).

This logic is re�ned on two fronts. First, in the present setting, while decentralization and

volatility are generally positively associated in equilibrium, volatility is not directly causing

23In particular, the relationship between the environment and the overall organizational design is broadlyconsistent with the typology of multinational corporations (MNCs) developed in Prahalad and Doz (1987)and Bartlett and Ghoshal (1989). See also Porter (1980).

25

0

1centralizationdecentralization

volatility, θ 2_

volatility, θ 2_

0

0.8

level of operationalintegration, β/(α+β)

Centralization

Decentralization



decentralization centralization decentralization

figure 5

Figure 6: Organizational design and environmental volatility (2)

decentralization. Instead, decentralization arises only as a part of the overall adjustment in

organizational design: an increase in volatility decreases the equilibrium level of operational

integration, and it is only because decentralization is a more e¢ cient governance structure for

managing a loosely integrated organization that decision-making authority is delegated to the

division managers. Indeed, because the optimal compensation and governance structures are

independent of the level of operational integration, simply decentralizing decision-making in

response to increased volatility would actually worsen organizational performance. Instead,

all design parameters need to be adjusted together in an internally consistent fashion to

maintain the �t between organization and the environment.

Second, note that the relationship between organizational design and environmental

volatility illustrated in �gure 5 holds only for a su¢ ciently high cost of information. For

intermediate costs, the non-monotone relationship between decentralization and the level

of operational integration translates into a non-monotone relationship between volatility

and decentralization. An alternative equilibrium is illustrated in �gure 6, which plots the

equilibrium organizational structure as a function of volatility for an intermediate cost of

information. Now, for stable environments, the preferred organizational design is a tightly in-

26

tegrated but decentralized structure that makes heavy use of �rm-level incentives to manage

the large agency con�icts caused by the high level of interdependence. As the environment

becomes more volatile, the organization reduces its level of operational integration but tem-

porarily centralizes decision-making to motivate information acquisition while economizing

on incentive provision for intermediate levels of volatility.

5.2.1 Existing empirical evidence

While there is an increasing body of empirical work examining the relationship between the

environment and di¤erent organizational variables, I am unaware of any work that would

take a comprehensive view of the overall design problem, as suggested by the present results.

However, as reviewed by Colombo and Delmasto (2008), some results regarding the co-

movements of di¤erent subsets of parameters are starting to emerge. For the relationship

between decentralization and volatility, there is now an increasing amount of evidence that

�rms that operate in environments that exhibit more volatility and higher informational

asymmetries are more decentralized and that decentralization is generally associated with

pay that is more tied to �rm-level performance (Nagar 2002 and Wulf 2006, for example).24

These papers, however, do not consider the level of operational integration. As a result, the

question of causality remains open. For the role of operational integration, there is evidence

that an increase in interdependencies across operating units is generally associated with a

decrease in delegation of decision-making authority and an increase in the use of �rm-level

performance measures (Bushman et al 1995, Christie et al 2003, Colombo and Delmasto

2004 and Abernethy et al 2004). The level of operational integration is, however, treated

as an exogenous parameter.25 In consequence, while these results are broadly consistent

with the theoretical results presented above, the theoretical framework also suggests that

they present only a partial and potentially misleading picture of the link between the overall

organizational design and the environment.

24In a closely related paper, Guadalupe and Wulf (2008) show that an increase in product market com-petition has led �rms to eliminate management layers (related to decentralization) and increase both �rm-and division-level pay for division managers.25Of course, industries vary in the bene�ts of operational integration, but �rms in any given industry still

face the choice of how integrated to be. For example, Philips used to be signi�cantly less integrated thanMatsushita (Barlett and Ghoshal, 1989) and Hewlett-Packard used to be less integrated than IBM (Prahaladand Doz, 1987), while changes in market conditions and the production technologies have led all four �rmsto navigate towards an intermediate solution.

27

5.3 Some extensions

Variation in the cost of incentives: In the discussion above, I did not consider varia-tion in the cost of incentives. The reason for this apparent neglect lies in the observation

that variation in the cost of incentives is qualitatively equivalent to variation in the cost of

information. Recall that the primary trade-o¤ in the framework is between motivating the

acquisition of information and then motivating appropriate use of that information. A re-

duction in the cost of information allows for more information acquisition at a given strength

of incentives. A reduction in the cost of incentives allows for the provision of stronger incen-

tives to acquire information.

Bene�ts of operational integration: For brevity, I have not discussed the role of oper-ational integration explicitly. In short, the optimal level of operational integration depends

on both the maximal bene�ts of operational integration and how quickly these bene�ts are

realized. The higher the maximal bene�ts, the higher the equilibrium level of integration

while the quicker the bene�ts are realized (the more concave the function), the lower the

equilibrium level of integration.

Measures of complexity: Complexity of the environment is sometimes o¤ered, togetherwith volatility, as another important dimension when analyzing the link between organi-

zational design and the environment. Cost of information can be seen as one aspect of

environmental complexity but I have not discussed it as such because there are other aspects

to complexity that are not captured by the cost of information and which can have a di¤erent

impact on the organizational design. As an example, another dimension of complexity is the

di¢ culty of transmitting information. First, the recipient of the messages can misunderstand

the content of the messages, where the likelihood of misunderstandings is increasing in the

complexity of the task at hand. Second, information can be veri�able but only at a cost,

and the cost of veri�ability is increasing in the complexity of the environment. Rantakari

(2008c) discusses the impact of both dimensions on the relative performance of the two gov-

ernance structures and shows that an increase in both the likelihood of misunderstandings

and the cost of veri�cation decrease the attractiveness of centralization. In other words, the

impact on the optimal governance structure is exactly opposite to that of cost of information.

For comparative statics, it is then crucial whether a given measure re�ects the di¢ culty of

generating information or the transmission of that information.

28

6 Conclusion

I have examined the �t between the overall organizational design, composed of the level

of operational integration, allocation of decision rights and the compensation structure of

division managers, and the environment, as determined by its volatility and the cost of in-

formation. The results were broadly consistent with the accepted wisdom regarding the �t

between the organizational design and the environment. Volatile environments were char-

acterized by organizations that were loosely integrated, with decentralized decision-making

and the compensation of division managers based primarily on divisional performance, while

stable environments were characterized by organizations that were tightly integrated, with

decision-making generally centralized and compensation structure that made more extensive

use of �rm-wide incentives.

The results also highlighted the importance of �t, not only between the design and the

environment but also among the di¤erent design parameters. This presence of interdependen-

cies provides one explanation why many corporate restructurings have yielded disappointing

results. A successful restructuring requires more than simply redrawing the organizational

chart. Instead, it requires an internally consistent readjustment of all organizational design

parameters together with the strategic orientation of the organization. For example, the

attempt of Brown-Boveri to rationalize its production in the 1970s got thwarted by the fail-

ure to simultaneously adjust managerial compensation and authority structures: "Division

managers in Germany and France were still measured and evaluated on their own short-term

results and had little incentive to help their Italian colleagues. To the contrary, the di¢ -

culties in Italy gave the German managers an ability to turn around their business on their

own and ignore the joint integration plan." (Prahalad and Doz, 1987/1999:206).

Finally, the results shed new light on the much-discussed relationship between delegation

and uncertainty. First, the results showed that the equilibrium likelihood of decentralization

need not be monotonically increasing in the volatility of the environment. Second, even if

decentralization and volatility were generally positively associated in equilibrium, volatility

did not directly cause decentralization. Instead, an increase in volatility led to a reduction

in operational integration, which in turn increased the preference for decentralization. As

a consequence, despite the positive equilibrium association of decentralization and volatility

in the present framework, simply decentralizing decision-making as a response to an increase

in volatility would actually worsen organizational performance. Instead, to maintain the �t

between the overall design and the environment, one needs to account for the interactions

among the design parameters and adjust them together in an internally consistent fashion.

29

References

[1] Abernethy, M.A., J. Bouwens and L. van Lent (2004), "Determinants of Control System

Design in Divisionalized Firms," The Accounting Review, 79(3), pp. 545-570

[2] Aghion, P. and J. Tirole (1997), "Formal and Real Authority in Organizations," Journal

of Political Economy, 105, pp. 1-29

[3] Alonso, R. (2007), "Shared Control and Strategic Communication," working paper,

University of Southern California

[4] Alonso, R., W. Dessein and N. Matouschek (2008), "When Does Coordination Require

Centralization?," American Economic Review, 98(1), pp. 145-79

[5] Athey, S. and J. Roberts (2001), "Organizational Design: Decision Rights and Incentive

Contracts," The American Economic Review, 91(2), pp. 200-205

[6] Bartlett, C.A. (1986), "Building and managing the transnational: the new organiza-

tional challenge." In: Porter, M.E. (Ed.), Competition in Global Industries. Harvard

Business School Press, Boston, MA.

[7] Bartlett, C.A. and S. Ghoshal (1989), Managing Across Borders: The Transnational

Solution, Harvard Business School Press, Boston, MA

[8] Bolton, P. and M. Dewatripont (1994), "The Firm as a Communication Network,"

Quarterly Journal of Economics, 109, pp. 809-839

[9] Brickley, J.A., C.W. Smith and J.L. Zimmerman (2003), Designing Organizations to

Create Value: From Structure to Strategy, McGraw-Hill, New York, NY

[10] Bushman, R., R. Indjejikian and A. Smith (1995), "Aggregate Performance Measures

in Business Unit Manager Compensation: The Role of Intra�rm Interdependencies,"

Journal of Accounting Research, 33 (Supplement), pp.101-128

[11] Chandler, A. (1962), Strategy and Structure: Chapters in the History of the American

Enterprise, The MIT Press, Cambridge, MA

[12] Chandler, A. (1977), The Visible Hand: The Managerial Revolution in American Busi-

ness, The Belknap Press of Harvard University Press, Cambridge, MA and London,

England

30

[13] Christie, A.A., M.P. Joye and R.L. Watts (2003), "Decentralization of the Firm: Theory

and Evidence," Journal of Corporate Finance, 9, 3-36

[14] Colombo, M.G. and M. DelMastro (2004), "Delegation of Authority in Business Orga-

nizations: An Empirical Test," The Journal of Industrial Economics, 52(1), pp.53-80

[15] Colombo, M.G. and M. DelMastro (2008), The Economics of Organizational Design,

Palgrave Macmillan, New York

[16] Crawford, V. and J. Sobel (1982), "Strategic Information Transmission," Econometrica,

50, pp. 1431-1452

[17] Dessein, W. (2002), "Authority and Communication in Organizations," Review of Eco-

nomic Studies, 69, pp. 811-838

[18] Dessein, W., L. Garicano and R. Gertner (2007), "Organizing for Synergies," CEPR

Discussion Paper No. 6019

[19] Dessein, W. and T. Santos (2006), "Adaptive Organizations," Journal of Political Econ-

omy, 114, pp. 956-995

[20] Friebel, G. and M. Raith (2007), "Resource Allocation and Firm Scope," Mimeo, Uni-

versity of Tolouse and Simon School of Business, University of Rochester

[21] Galbraith, J.R. (1973), Designing Complex Organizations, Addison-Wesley Publishing

Company, Reading, MA

[22] Galbraith, J.R. (1977), Organization Design, Addison-Wesley Publishing Company,

Reading, MA

[23] Garicano, L. (2000), "Hierarchies and the Organization of Knowledge in Production,"

Journal of Political Economy, 108(5), pp. 874-904

[24] Guadalupe, M. and J. Wulf (2008), "The Flattening Firm and Product Market Com-

petition: The E¤ect of Trade Liberalization," working paper, Columbia University and

Harvard University

[25] Harris, M. and A. Raviv (2002), "Organization Design," Management Science, 48(7),

pp. 852-865

[26] Harris, M. and A. Raviv (2005), �Allocation of Decision-Making Authority,�Review of

Finance, 9(3), pp. 353-383

31

[27] Harris, M. and A. Raviv (2007),"A Theory of Board Control and Size," forthcoming,

Review of Financial Studies

[28] Hart, O. and J. Moore (2005), "On the Design of Hierarchies: Coordination versus

Specialization," Journal of Political Economy, 113(4), pp. 675-702

[29] Harzing, A-W. (2000), �An Empirical Analysis and Extension of the Bartlett and

Ghoshal Typology of Multinational Companies,�Journal of International Business Stud-

ies, 31(1), pp.101-120

[30] Holmstrom, B. (1999), �The Firm as a Subeconomy,�Journal of Law, Economics and

Organization, 15(1), pp. 74-102

[31] Holmstrom, B. and P. Milgrom (1991), �Multitask Principal-Agent Analyses: Incen-

tive Contracts, Asset Ownership, and Job Design,� Journal of Law, Economics and

Organization, 7, pp. 24-51

[32] Holmstrom, B. and P. Milgrom (1994), �The Firm as an Incentive System,�The Amer-

ican Economic Review, 84(4), pp.972-991

[33] Milgrom, P. and J. Roberts (1990), �The Economics of Modern Manufacturing: Tech-

nology, Strategy and Organization,�The American Economic Review, 80(3), pp.511-528

[34] Lawrence, P. and J. Lorsch (1967), Organization and the Environment, Harvard Business

School Press, Boston, MA

[35] Marshak J. and R. Radner (1972), Economic Theory of Teams, Yale University Press,

New Haven and London

[36] Melumad, N. and T. Shibano (1991), "Communication in Settings with No Transfers,"

The RAND Journal of Economics, 22, pp.173-198

[37] Milgrom, P. and J. Roberts (1995), �Complementarities and �t: Strategy, structure,

and organizational change in manufacturing,� Journal of Accounting and Economics,

19, pp.179-208

[38] Mintzberg, H. (1979), The Structuring of Organizations, Prentice Hall, Englewood

Cli¤s, NJ

[39] Nadler, D.A. and M.L. Tushman (1997), Competing By Design: The Power of Organi-

zational Architecture, Oxford University Press, New York, NY

32

[40] Nagar, V. (2002), "Delegation and Incentive Compensation," The Accounting Review,

77(2), pp.379-395

[41] Porter, M.E. (1980), Competitive Strategy, The Free Press, New York, NY

[42] Porter, M.E. (1986), "Changing Patterns of International Competition," California

Management Review, 28(2), pp.9-40

[43] Porter, M.E. (1996), �What is Strategy?�Harvard Business Review, 74(6), pp. 61-78

[44] Prahalad, C. and Y. Doz (1987), The Multinational Mission: Balancing Local Demands

and Global Vision, Free Press, New York, NY

[45] Qian, Y., G. Roland and C. Xu (2006), "Coordination and Experimentation in M-Form

and U-Form Organizations," Journal of Political Economy, 114(2), pp.366-402

[46] Rantakari, H (2008a), �Governing Adaptation,�Review of Economic Studies, Vol. 75(4),

pp. 1257-1285

[47] Rantakari, H. (2008b), �Uncertainty, Delegation and Incentives,�working paper, Uni-

versity of Southern California

[48] Rantakari, H. (2008c), "A supplement to "Governing Adaptation" and "Organizational

Design and Environmental Volatility," working paper, University of Southern California

[49] Roberts, J. (2004), The Modern Firm: Organizational Design for Performance and

Growth, Oxford University Press, New York, NY

[50] Simon, H.A. (1947), Administrative Behavior, The Free Press, New York, NY

[51] Stein, J. (1989), "Cheap Talk and the Fed: A Theory of Imprecise Policy Announce-

ments," The American Economic Review, 79(1), pp. 32-42

[52] Thompson, J.D. (1967), Organizations in Action, McGraw-Hill, New York, NY

[53] Woodward, J. (1965), Industrial Organization: Theory and Practice, Oxford University

Press, London, UK

[54] Wulf, J. (2006), "Authority, Risk, and Performance Incentives: Evidence from Division

Manager Positions Inside Firms," forthcoming, Journal of Industrial Economics

[55] Zabojnik, J. (2002), "Centralized and Decentralized Decision Making in Organizations,"

Journal of Labor Economics, University of Chicago Press, 20(1), pp. 1-22

33

A Proofs and derivations

A.1 Expected pro�ts

The derivation of the expected pro�ts is detailed in Rantakari (2008a), so I will only outline

the impact that imperfect primary information has on the solution.

Decision-making: Let m and n denote the decision-makers controlling decisions i and

j respectively. Let m0s objective function be of the form smi�i+smj�j and let Em denote be-

liefs conditional on information available to m: Then, we can write the �rst-order conditions

for the two decisions as

dmi = a1Em�i + a2Emdj and dnj = b1En�j + b2Endi;

so that the equilibrium decisions are given by (after repeated substitution)

dmi =a1(1�b2a2)Em�i+a2b1EmEn�j+a2b2a1EnEm�i

1�b2a2 ;

where a1 and a2 (b1 and b2) are the relative weights placed on adaptation and coordina-

tion by m (n). Note that the equilibrium responsiveness of the agent to changes in beliefs

is independent of the accuracy of beliefs, so that the only di¤erence to the full-information

case is that all information is discounted by qi; the accuracy of primary information.

Communication: Communication is modeled as one round of simultaneous cheap talk.The equilibrium takes a partition structure, where the cuto¤s of the partition are deter-

mined by the sender�s indi¤erence condition

Ei�sii�i

�tMi ; di(:;m

L); dj(:;mL)�+ sij�j

�tj; di(:;m

L); dj(:;mL)�j:;mL

�= Ei

�sii�i

�tMi ; di(:;m

H); dj(:;mH)�+ sij�j

�tj; di(:;m

H); dj(:;mH)�j:;mH

�:

That is, given that the realized signal tMi falls on the boundary of two intervals of the parti-

tion, then the division manager needs to be indi¤erent between saying that the realized state

belongs to the lower interval (mL ! ti 2 (tLi ; tMi ]) or the higher interval (mH ! ti 2 (tMi ; tHi ]).The solution is equivalent to the full information-case (ti = �i). The reason for this result

follows from the fact that because both the sender and the receiver discount information

at the rate qi; the accuracy of information cancels out in determining the relative incentive

con�ict between the sender and the receiver.

34

Expected pro�ts: Substituting the equilibrium decisions into the pro�t function, we can

write the decision-dependent component (in loss terms) as

kib21((1�ri)a22+ria21)(1�b2a2)2

�(EnEm�i)

2 + (EmEn�j)2�

+ ki ((1� ri) a22 + ria21) (Em�i � EnEm�i)2 + kirib

21 (En�j � EmEn�j)

2

+ ki (1� ri)V arm�i;

where ki = (�+ �) and ri = �= (�+ �) ; with (a1; a2; b1; b2) as given by the �rst-order con-

ditions for the equilibrium decisions. Because the communication equilibrium is unchanged

as a result of inaccurate primary information, the �rst two lines are identical to the full-

information case, with the exception that all components are scaled by pi = q2i : This result

follows because the �rst two lines re�ect only the di¤erences in the posterior beliefs held by

the sender and the receiver. The only component that depends directly on the quality of

primary information is V arm�i:Whenm = i; so that manager i decides di (decentralization),

then

E (�i � Ei�i)2 = qiE (�i � qi�i)2 + (1� qi)E (�i � qixi)2 ;

where xi is a random draw from��; �

�. As a result,

E (�i � Ei�i)2 = (1� q2i )�2i

3= (1� pi) �

2i

3:

When m = P (centralization), we have that

E (�i � Em�i)2 = E (�i � qiEmti)2 = E (�i � qiti + qiti � qiEmti)2 :

Now, E (�i � qiti + qiti � qiEmti)2 = E (�i � qiti)2+q2iE (ti � Emti)2 because in equilibrium,

the headquarters�beliefs must be unbiased so that E (ti � Emti) = 0: Above we already

showed that

E (�i � qiti)2 = (1� pi) �2i

3;

so we have that

E (�i � Em�i)2 = piE (�i � Em�i)2 + (1� pi) �2i

3;

with the �rst component re�ecting the inaccuracy of information transmission and the second

35

component giving the additional loss due to inaccurate primary information, as in the case

of decentralization. The expected pro�t can then be written as if information was perfect

but distributed on U��qi�i; qi�i

�and a common extra component re�ecting the inaccuracy

of primary information:

E (�gi (s; �)) =�K (�)� ��

2

3

�+ pgi [��

gi (s; �)� �

gii (s; �)V ('

gi (s; �))]

�2

3

� pgj��gi (s; �) + �

gij (s; �)V

�'gj (s; �)

��2

3;

where V ('gi (s; �)) =1

4+3'gi (s;�); re�ecting the accuracy of information transmission. Pooling

coe¢ cients, we can then write the value of information as

gi (s; �) = �� gi (s; �)� �

gii (s; �)V ('

gi (s; �))� �

gi (s; �)� �

gji (s; �)V

�'gj (s; �)

�:

Under centralization, the coe¢ cients are

�centi = ��(�+4�)

; �centii = �� centi and �centij = ��centi ;

while under decentralization, the coe¢ cients are:

�deci =��(�+s2�)(s�+2�)2

; �decii =��(�+s2�)(s�+�)2

� �deci and �decij = ��

s�s�+�

�2� �deci :

A.2 Information acquisition

Recall that the solution to the manager�s information acquisition problem is given by

pi =�egi (s;�)�egi (s;�)+� :

Then, it is immediate that @pi@�

> 0 since egi (s; �) > 0 and that @pi@�

< 0: To establish@pi@�and @pi

@s; we need to examine their impact on the perceived value of information. While

taking the derivatives is analytically cumbersome, they are easily veri�ed numerically, which

is done in �gure 7.26 Intuitively, an increase in � reduces the true value of information and,

while it also typically increases the negative externality, the �rst e¤ect dominates. Similarly,

while an increase in s reduces the true value of information, the division manager puts more

26While the solution under centralization is generally tractable, the solution under decentralization isunfortunately extremely cumbersome.

36

0.60.8

1

00.5

10

0.60.8

1

00.5

10

0.60.8

1

00.5

1

0

0.60.8

1

00.5

1

0

r sr s

r s r s

dΦcent(s,β)/ds dΦdec(s,β)/ds

dΦcent(s,β)/dβ dΦdec(s,β)/dβ

~ ~

~ ~

Figure 7: Partial derivatives of perceived value of information

Φdec(s,β)Φcent(s,β) Φdec(s,β)Φcent(s,β)∼ ∼

0

0

1

0

1

0.5

r s

1

0

1

0.5

r s

Figure 8: di¤erences in the true and perceived value of information

weight on the value realized by his division and less on the negative externality. This latter

e¤ect dominates, so that the perceived value of information is increasing in s:

The comparison of the quality of information acquired under the two governance struc-

tures follows directly from the di¤erence in perceived value of information, which is given in

�gure 8. The �rst determinant of the perceived value of information is the true value of infor-

mation, which is greater under centralization only when both � and s are su¢ ciently large,

as discussed in section 4.4. The second determinant is the size of the negative externality

imposed on the other division. Re�ecting the bias in equilibrium decisions, this externality

is always larger under decentralization (�decj (s; �) � �centj (s; �)). The di¤erence is largest for

intermediate � and high s (when the decisional bias is largest), while converging to zero when

� ! 0; � ! 1 and s ! 1=2: The net e¤ect is such that the perceived value of information

37

and so the quality of information acquired is almost always higher under decentralization for

given (�; s; �), except when both � and s are very large so that the externality converges to

zero and the di¤erence in the true value of information dominates.

A.3 Choice of g

The proof for the choice of governance structure follows a replication argument and relates

back to di¤erences in the true value of information, gi (s; �) : Recall that if (�; s; �; pi) are

the same across the governance structures, then decentralization is preferred over centraliza-

tion i¤ deci (s; �) � centi (s; �) : In section 4.4, we saw that decentralization makes better

use of a given amount of primary information whenever the level of operational integration

is su¢ ciently low, independent of the amount of pro�t-sharing, and also whenever the de-

gree of pro�t-sharing is su¢ ciently high, independent of the level of operational integration.

Now, let the particular (s; �) equal�scent; �cent

�, the optimal centralized solution for the par-

ticular environment and for now assume that � and pi are the same across the governance

structures. Then, if either s or � is su¢ ciently low, decentralization is able to replicate the

performance of centralization by choosing the same parameters. Further, by re-optimizing

its choice it can do strictly better. To complete the proof, recall from section A.2 that

deci (s; �) � centi (s; �) ! pdeci (�; :) > pcenti (�; :) ; so that more information is acquired

under decentralization for a given level of incentives, further supporting the advantage of

decentralization. In short, if deci�scent; �cent

�� centi

�scent; �cent

�; decentralization will

always be preferred over centralization.

The proof in the other direction (centralization is still sometimes preferred) follows sim-

ilarly. Let the particular (s; �) be now�sdec; �dec

�: For (s; �) both su¢ ciently large, we

know that deci (s; �) < centi (s; �) : Also, for (s; �) both su¢ ciently large, we know that

pcenti > pdeci ; implying that the centralized (but suboptimal) structure both generates more

information and makes better use of that information than an otherwise identical (and op-

timally chosen) decentralized structure, which completes the proof.

Of course, the exact location of the boundary in the (s; �)-space depends on the di¤er-

ences in the equilibrium quality of information and the strength of incentives that a¤ect

the quantitative trade-o¤s involved. We can conclude that the boundary lies somewhere

between�sdec; �dec

�-pairs where pcenti = pdeci (upper bound) and

�scent; �cent

�-pairs where

deci (s; �) = centi (s; �) (lower bound)).

38

A.4 Choice of �

Recall that that the �rst-order condition characterizing the choice of operational integration

was given by

K 0 (�) + �2

3dd�M (�; �) = 0;

where M (�; �) = maxgN g (�; �) and N g (�; �) = max

s;�pgi (:)

gi (s

g; �) � C (�; pgi (:)) � G (�) :Consider �rst the net pro�t realized from local responsiveness under each governance struc-

ture. Then, we have that dd�N g (�; �) = @

@�N g (�; �) by the envelope theorem, and

@@�N g (�; �) = pgi (:)

@gi (sg ;�)

@�< 0, @2

@�2N g (�; �) = pgi (:)

@2gi (sg ;�)

@�2> 0 and

@2

@�@�N g (�; �) =

@pgi (:)

@�

@gi (sg ;�)

@�> 0:

The net pro�t from local responsiveness under both governance structures is thus a de-

creasing and convex function of the level of integration, and the marginal cost is decreasing

in the cost of information. The result that @�

@�< 0 and @�

@�< 0 is then immediate as

long as the governance structure does not change as a result of the change in the envi-

ronment. To see why the potential change in governance structure does not change the

result, note that M (�; �) is then the upper envelope of two net pro�t functions. Be-

cause each N g (�; �) is decreasing and convex and they are equal at � = 0 and � ! 1,the points at which N cent (�0; �0) = Ndec (�0; �0) are characterized by the property that

lim�!�0+

��dM(�;�)d�

�� < lim�!�0�

��dM(�;�)d�

�� ; independent of the order of governance structures. As

a result, the pairs (�0; �0) for which N cent (�0; �0) = Ndec (�0; �0) provide a local minimum

of the optimization problem and a change in the governance structure is always associated

with a discrete change in the level of operational integration in the direction of the original

change.

39

B Interactions among (�; s; �)

In this subsection, I examine the interactions among operational integration, incentive align-

ment and incentive strength in a little more detail. We can write the objective function of

the principal as

maxg

�max�;s;�

�K(�)� ��

2

3+ pgi

��; s; �; �

�gi (s; �)

�2

3� C

��; pgi

��; s; �; �

��

2

3G (�)

��s.t. pgi (�; s; �; �) =

�egi (s;�)�egi (s;�)+� :

Now, taking the �rst-order condition, say, with respect to �; we get�gi (s; �)

�2

3� Cpi

��; pgi (:)

�� @pgi (:)

@�= �

2

3G0 (�) ;

and noting further that since the division manager�s information acquisition choice satis-

�es �egi (s; �) �23 = Cpi ��; pgi (:)� ; we can write this solution as�g :

�gi (s; �)� �egi (s; �)� @pgi (:)

@�= G0 (�) ;

and similarly for the other two design parameters, for which we get

sg :�gi (s; �)� �egi (s; �)� @pgi (:)

@s+ pgi (:)

@gi (s;�)

@s= 0

�g : K 0 (�) +h�gi (s; �)� �egi (s; �)� @pgi (:)

@�+ pgi (:)

@gi (s;�)

@�

i�2

3= 0:

To examine the interdependencies across the design parameters, I will consider the pair-

wise relationships among the three. I do this by analyzing the cross-partials of the pro�t

function, but let us manipulate the �rst-order condition a little bit more to build up the

intuition. Recall that gi (s; �) = ��i (s; �)��j (s; �) ; while egi (s; �) = s (�� i (s; �))�(1� s)�j (s; �) : Thus, we can write

gi (s; �)� �egi (s; �) = (1� �s)gi (s; �)� � (2s� 1)�gj (s; �) ;which measures how far the quality of information acquired is from the pro�t-maximizing

level (given �; s). This gap depends on two aspects. First, how large a margin the manager

internalizes of the true value of information, as given by �s; minus the extra incentives pro-

vided by the incentive con�ict (2s� 1) through the negative externality. I will use�gi (s; �)to denote this valuation gap.

40

To examine the pairwise dependencies, let us begin with incentive strength and alignment.

We have that @2�@s@�

Q 0; so that incentive strength and incentive alignment can generally beboth pairwise complements and substitutes. To see this, we have that

@2�@�@s

=@�gi (s;�)

@s<0

@pgi (:)

@�>0

+�gi (s; �)>0

@2pgi (:)

@�@sR0

:

We already know that �gi (s; �) > 0 as long as G0(�) > 0 and that @pgi (:)

@�> 0: The

impact that an increase in s has on the bene�ts of increasing � is two-fold. First, an in-

crease in s increases the amount of information acquired and thus reduces the valuation gap

�gi (s; �) ; which in turn reduces the value of the strength of incentives. Second, increasing

the degree of incentive con�ict also alters the responsiveness of the agent to incentives, which

is composed of two competing e¤ects. First, an increase in s increases the perceived value of

information, making the agent more responsive to incentives. However, because this change

induces further information acquisition, it moves the agent to a more convex part of his cost

function and so reduces his responsiveness to incentives. The two e¤ects work in the opposite

directions. Since

pgi (�; s; �; �) =�egi (s;�)�egi (s;�)+� ;

we have that

@pgi@�=

�egi (s;�)(�egi (s;�)+�)2 and

@2pgi@�@x

= �

��egi (s;�)�(�egi (s;�)+�)3

�@egi (s;�)

@x:

In the case of s; @egi (s;�)@s

> 0 so that the sign of @2pgi@�@x

equals the sign of � � egi (s; �)�:When �egi (s;�) � � (pgi � 0:5); the coe¢ cient is negative and the partial derivative is then un-ambiguously negative, so that incentive strength and incentive alignment are complements.

However, for � su¢ ciently small (relative to the cost of information so that pgi is low), the

coe¢ cient can be positive and incentive strength and incentive alignment can be substitutes.

Numerical examples suggest that this can occur but � needs to be quite low relative to the

cost of information �.

Second, I will argue that @2�@�@�

- 0; so that the level of operational integration and the

strength of incentives are generally substitutes. Taking the cross-partial, we have

@2�@�@�

=@�gi (s;�)

@�-0

@pgi (:)

@�>0

+�gi (s; �)>0

@2pgi (:)

@�@�

Q0

:

From earlier, we have that @pgi (:)

@�> 0 and that �gi (s; �) > 0: For the e¤ects of �; we

have �rst that @�gi (s;�)

@�- 0: To see this, recall that �gi (s; �) = (1� �s)gi (s; �) �

41

� (2s� 1)�gj (s; �) : As discussed, gi (s; �) is decreasing in � because of the reduced adap-

tiveness of the organization. Second, �gj (s; �) is generally increasing in � because of the

increasing dependency of the divisions, further reducing the valuation gap. The only ex-

ception occurs under decentralization when both � and s are su¢ ciently large so that an

increase in � actually improves decision-making and thus lowers �decj (s; �). But in this re-

gion, decentralization would never arise as the preferred governance structure (increasing

operational integration further would actually reduce agency losses). Further, when � is

su¢ ciently small, the e¤ect on true value always dominates.

The e¤ect on the sensitivity of the manager is again ambiguous and from above, exactly

the opposite to the degree of incentive con�ict, with the sign of @2pgi@�@�

equal to the sign ofegi (s; �)�� (since @egi (s;�)@�< 0). Thus, the sign of the cross-partial is (almost) unambiguous

when �egi (s; �) � � (so that pgi � 0:5). Numerical estimations suggest that this sign remainsnegative under centralization while being positive under decentralization only in regions of

high s; � and �; where it is dominated by centralization (reasons for which were discussed

with the choice of governance structure).

Finally, the relationship between operational integration and strength of incentives is

characterized by

@2�@s@�

=@�gi (s;�)

@�-0

@pgi (:)

@s>0

+�gi (s; �)>0

@2pgi (:)

@s@�

R0

+@pgi (:)

@�<0

@gi (s;�)

@s<0

+ pgi (:)>0

@2gi (s;�)

@s@�-0

- 0:

Let us begin by summarizing what we already know. From above, @�gi (s;�)

@�- 0 because of

the reduced true value of information and increased externality. In other words, an increase

in � reduces the value of motivating information acquisition and thus supports a decrease in

s (an increase in incentive alignment). Second, the resulting change in the perceived value

of information alters the sensitivity of the agent to incentives to acquire information. Since

@pgi@s=

��

(�egi (s;�)+�)2�

@egi (s;�)@s

;

we have that

@2pgi@s@�

= ��

@2 eg

i(s;�)

@s@� (�egi (s;�)+�)�2� @ egi (s;�)@s

@ egi(s;�)

@�

(�egi (s;�)+�)3!:

We know that @egi (s;�)@s

> 0 and @egi (s;�)@�

< 0; but @2 egi (s;�)@s@�

< 0; as shown in �gure 9. As

a result, we cannot unambiguously sign @2pgi@s@�

, but we can observe that it becomes negative

when � is su¢ ciently large. The third e¤ect is the change in the quality of information ac-

quired. An increase in � decreases the amount of information acquired�@pgi (:)

@�< 0�; which

42

0.60.8

1

00.5

1

0

0.60.8

1

00.5

1

0

0.60.8

1

00.5

1

0

0.60.8

1

00.5

1

0

d2Φcent(s,β)/dsdβ d2Φdec(s,β)/dsdβ

d2Φcent(s,β)/dsdβ d2Φdec(s,β)/dsdβ~ ~

r s r s

r s r s

Figure 9: cross-partials for true and perceived value of information

makes the fact that incentive con�ict reduces the value of information�@gi (s;�)

@s< 0�less

costly. This e¤ect works in the opposite direction, reducing the value of incentive alignment.

Finally, a change in � directly alters how much the value of information is impacted by the

degree of incentive con�ict, which in turn impacts the current value of information. For this

e¤ect, we have that @2gi (s;�)

@s@�- 0; as shown in �gure 9: In other words, an increase in oper-

ational integration makes the true value of information to deteriorate faster with increases

in s because of increasing agency con�icts. Again, the exception occurs for decentralization

when operational integration actually reduces the size of the agency con�ict, a region where

centralization would be preferred.

For the overall e¤ect, numerical estimations suggest that the sign is always negative

under centralization, while it can be positive under decentralization but again only when

centralization would generally be preferred. Intuitively, this should be the case because the

�rst-order e¤ect of increased operational integration is to decrease the value of information,

increase the free incentives for information acquisition generated by the negative externality

and increase the value of incentive alignment because of the increase in agency con�icts.

However, it is also possible that for di¤erent cost functions, the drop in the quality of

information acquired caused by an increase in � is so large that the organization must

actually increase s to restore incentives for information acquisition.

43

C Cost of incentives

The analysis presented took a reduced-form approach to the cost of providing incentives, with

the further assumption that the cost of incentive provision is proportional to the volatility

of the environment and that the governance structures shared the same cost of providing

incentives. In this appendix, I �rst relax the proportionality assumption and then discuss

some of the potential microfoundations for these costs.

C.1 An alternative cost formulation

The analysis assumed that both the cost of information and the cost of incentives were

proportional to volatility. This assumption allowed me to separate the impact that variations

in volatility, cost of information and cost of incentives have on the organizational design.

Having separated the e¤ects, it is then straightforward to consider any co-movement of the

parameters. For example, an increase in volatility that leads a more than proportional

increase in the cost of information has an ambiguous impact on organizational design: an

increase in volatility supports a reduction in operational integration while an increase in the

cost of information supports an increase in operational integration.

To give a particular illustration of such co-movements, consider that, instead of be-

ing proportional to volatility, the cost of information is independent of volatility, given by

G (�) = B�1�� : Now, as volatility increases, the relative cost of incentives decreases (value of

incentives goes up while the cost stays constant). Therefore, relative to the results presented

in the analysis, there is an additional upward trend to the strength of incentives. And be-

cause of the stronger trend in the quality of primary information, the downward trend for

the level of operational integration is stronger. An example of this is given in �gure 10. Now,

when the environment is very stable, information is not very valuable while incentives are

costly. Therefore, the organization provides only weak incentives. And because the equi-

librium quality of information is then low, no �rm-wide incentives are used. As volatility

increases, the equilibrium level of operational integration decreases and the relative cost of

incentives goes down. As a result, the organization increases the strength of incentives sig-

ni�cantly and eventually the quality of information acquired becomes su¢ ciently high that

�rm-wide incentives are also introduced to make better use of the information generated. Fi-

nally, as volatility increases further, the equilibrium level of operational integration becomes

su¢ ciently low so that the use of �rm-wide incentives again diminishes.

44

volatility volatility

centralizationdecentralization

(ii) strength and composition of incentives(i) degree of operational integration

0

1

0

0.8centralization decentralization centralization decentralization

Figure 10: An example of the link between organizational design and environmental volatility

C.2 Some microfoundations for the cost of incentives

The simple justi�cation for using a reduced-form approach to the cost of incentives was

that it is generally accepted that incentive provision is costly. Further, the general logic of

the results is not dependent on the particular source of these costs, while complicating the

interpretation of the results. In particular, the relationship between cost of incentives and

volatility becomes context-speci�c. Also, the tractability of the analysis is also compromised.

However, to bridge some of the gap between the assumed reduced-form cost of incentives

and more classical (and microfoundational) approaches, I will discuss some potential sources

of costs here.

Absence of a budget breaker: The simplest possible reason for the inability of theorganization to achieve the �rst-best outcome of � = 2; s = 1=2 is the absence of a bud-

get breaker. The implication of this assumption is that while the organization is able to

divide the existing pro�t stream among the active participants, it is unable to leverage that

pro�t stream to provide the full margin to all participants, e¤ectively setting �max = 1: The

simple justi�cation for this assumption follows from the observation that the moment the

pro�t stream is leveraged, the person responsible for the �nal payments has an incentive to

sabotage the outcome and so avoid making the required payments. The shortcoming of this

approach is that it arti�cially keeps the strength of incentives �xed at � = 1; thus preventing

us from analyzing the interaction between the strength and composition of incentives and

how the strength of incentives can vary across organizational structures and environments.

45

Headquarters as a strategic actor: A simple "�x" to the above shortcoming is to as-sume that, in addition to the absence of a budget breaker, the headquarters also needs to be

motivated to behave appropriately. In particular, one could assume that while information

acquisition, by its local nature, must be undertaken by the division managers, the headquar-

ters is able to engage in e¤ort provision that improves the pro�tability of the divisions. Then,

given that the pro�t streams cannot be leveraged, any share of the pro�ts that is used to

motivate the division managers is a share that cannot be used to motivate the headquarters.

The cost of incentives is then equal to the reduction in pro�ts resulting from the reduction

in the e¤ort level of the headquarters. For example, if the headquarters was able to improve

the pro�ts of division i by �ei by exerting e¤ort ei at a personal cost 12e2i ; then the cost of

incentives to the division managers would be equal to G��; ��= �2�2.

To see this, note that if share �i is used to motivate the division managers, (1� �i) isretained by the headquarters, leading to an e¤ort choice of ei = � (1� �i) and a net surplusof �

2(1��i)(1+�i)2

; while the �rst-best e¤ort level (�i = 0) would produce a net surplus of �2

2:

The di¤erence, which then matches the cost of incentives, is �2

2� �2(1��2i )

2=

�2�2i2: In this

case, because the value of headquarters�e¤ort is independent of volatility, so is the cost of in-

centives (as in the example above). However, it can also be natural to assume that the value

of e¤ort is increasing in volatility. For example, if the marginal value of e¤ort was ��; then

the cost of incentives would be proportional to �2as assumed in the analysis. Further, de-

pending on the shape of bene�t and cost functions, one could generate any shape of G��; ��:

Limited liability and ex ante budget constraints: Another commonly used sourceof incentive costs is the presence of limited liability, where the realized wage of the agent

cannot fall below zero, combined with the presence of an ex ante budget constraint (or moral

hazard on part of the principal), which makes it impossible for the agent to post a perfor-

mance bond ex ante to relax this constraint. In this case, while the total net surplus would

be maximized by setting � = 2; s = 1=2; the principal is generally unwilling to do this be-

cause it would require leaving the agent with unnecessarily high rents. Instead, the principal

distorts the contract and destroys some of the total net surplus while increasing her share of

the surplus. Unfortunately, performing the analysis in detail under this set of assumptions

is infeasible because it would lead to a non-linear compensation structure and thus destroy

the tractability of both the decision-making and communication equilibria. Further, how

stringent the constraint would be depends on the expected level of pro�ts, which are not

pinned down by the model.

What limited liability really boils down to, however, is simply that the principal cannot

46

expropriate the full surplus and thus has incentives to set suboptimal incentives.27 Suppose

that the principal provides a share � of pro�ts to incentivize the agents but is able to extract

only (1� � (�)) of that surplus in terms of ex ante transfers for one reason or another. Then,the objective function of the principal will be to maximize E ((1� � (�)�) (�i + �j)) : Theapparent di¤erence to the formulation used in the paper is that the "cost of incentives" is now

structure-speci�c. This di¤erence is, however, only quantitative. First, note that it is only

the strength of incentives that impacts the share of the surplus received by the principal. As

a result, (g; s; �) are still chosen to maximize the total surplus conditional on the strength of

incentives, �: The only di¤erence is now that a structure that performs better faces a higher

cost of incentives. The organization then optimally reduces the strength of incentives to

economize on these costs. But this reduction in the strength of incentives cannot overturn

the ranking because of this higher cost, because if the organization came to underperform

an alternative structure as a result, then it would face a lower cost of providing incentives.

Thus, while the di¤erences in the strength of incentives would be lower in equilibrium than

under a corresponding setting but a structure-independent cost of incentives, the qualitative

logic and trade-o¤s involved are not changed. The additional shortcoming of this approach is

that, in the present model, there is no natural pro�t benchmark and the expected pro�ts are

monotonically decreasing in the volatility of the environment. As a result, the e¤ective cost

of incentives would be actually decreasing in volatility, further increasing the upward-trend

in the strength of incentives relative to the formulations discussed below.

Risk-aversion: The most traditional approach to the agency cost of incentives is throughthe risk-aversion of the agents. Much like with limited liability, explicitly introducing risk-

aversion to the model makes the analysis intractable, as it changes decision-making (im-

perfect information makes the division managers more cautious when responding to infor-

mation), communication (risk-aversion should improve communication because the division

managers become less adaptive to their own conditions) and information acquisition (risk-

averse division managers will acquire more information to reduce uncertainty). There is,

however, nothing in the nature of risk-aversion that would impact one governance structure

qualitatively di¤erently from the other. Further, as � ! 1; the two solutions continue tobe equivalent because the equilibrium decisions continue to converge (perfect coordination).

Finally, using risk-aversion in the present setting is unattractive because it would need

to ignore the matching of people with di¤erent risk preferences to di¤erent environments.

Indeed, using the assumption of �xed risk preferences would lead to the counterintuitive

result that when the environment becomes su¢ ciently volatile, the strength of incentives

27Ignoring the induced non-linearity of compensation even with linear incentive contracts.

47

goes to zero because the cost of incentives becomes simply too high.28

In summary, while there are potentially interesting interactions that the presence of

limited liability, ex ante budget constraints or risk-aversion could introduce to the analysis,

there is nothing in the logic of either approach that would appear to have a �rst-order e¤ect

on the qualitative trade-o¤s analyzed and the conclusions reached in the present paper.

28While the productive role of uncertainty can lead to a non-monotone relationship between risk andincentives, as discussed, for example, in Rantakari (2008b), for su¢ ciently high levels of uncertainty, therisk-compensation aspect generally comes to dominate.

48

Organizational Design and Environmental Volatility -...

Documents

Transcript of Organizational Design and Environmental Volatility -...