Organizational Dynamics∗

Shingo Ishiguro†
Graduate School of Economics
Osaka University

May 14, 2012

Abstract
In this paper we develop a dynamic general equilibrium model to investi-
gate the co-evolution of production organizations and economic development
over time. We compare two organizational modes: centralized organizational
mode wherein the principal holds the decision making authorities and decen-
tralized organizational mode wherein some of the decision making authorities
are delegated to agents. We present a novel mechanism that accommodates
the dynamic general equilibrium interactions between organizational choice
and market equilibrium conditions, and then show that the dynamic pro-
cesses of organizational changes are diverse and non-monotonic over time in
the process of economic development.

Keywords: Centralized Organization, Decentralized Organization, Incen-

tives, Economic Development
JEL Classification: D82, G30, L22, O11

∗
I am grateful to V. Bhaskar, Eric Chou, and Ryoji Odoi for their helpful comments on an
earlier version of the paper. I began this research during my visit to University College London
and am grateful for the hospitality and support I received there. Of course, all remaining errors
are my own.
†
Correspondence: Shingo Ishiguro, Graduate School of Economics, Osaka University. 1-7
Machikaneyama, Toyonaka, Osaka 560 0043, Japan. E-mail: ishiguro@econ.osaka-u.ac.jp

1
1 Introduction
In this paper we develop a dynamic general equilibrium model to investigate the
co-evolution of production organizations and economic development over time. In
particular we present a novel dynamic mechanism that accounts for diverse and
non-monotonic changes in production organizations in the endogenous process of
economic development. As several economic historians have discussed (see, for
example, Mokyr (1998, 2004, 2010)), technological revolutions and the economic
success of countries are dynamically associated with the creation of new orga-
nizational modes of production. More precisely, our paper is motivated by the
following historical and empirical observations that discover the links between
the changes in organizations and economic development.
First, dynamic processes of organizations are not uniform but diverse across
countries. Several historical pieces of evidence have documented that different
countries, even in a specific region such as Western Europe, experience different
paths of economic development, which are associated with the differences in the
development of production organizations (Chandler (1990), Landes (1969), and
Mokyr (2010)). For example, Landes (1969) argued that in the period of the First
Industrial Revolution, France lagged behind Britain because British firms entered
into partnerships, which allowed them to create innovative and new businesses
by being more open to accepting new partners from the outside, while the French
firms were more closed and based on family-limited ownership and management.
Chandler (1990) also stated that the advent of the Second Industrial Revolution
in the 19th century was accompanied by the emergence and rise of the ’modern
business enterprise’ (MBE), which made large scale production possible. This
gave American firms an advantage over the firms operating in Britain that had
small-scale production and hired less professional salaried managers.
The recent empirical study by Bloom and Van Reenen (2009) discovered the
fact that firm organizations are diverse across countries. They measured the
extent of decentralization in firms in terms of delegating decision making to the
managers at lower levels of the corporate-hierarchy, and found that there are
significant cross-country variations in the decentralization of firms.1 By using
their measure of firm decentralization, Figures 1–3 show that the decentralization
of firms is positively associated with the level of economic development (in terms
of per capita GDP) and the investment in R&D and IT technology.2
1
Bloom and Van Reenen (2009) collected the data on the extent of decentralization in firms
by interviewing managers across countries. They assessed the data using a 5-point scale with
1 denoting the least decentralization and 5 denoting the most, in terms of the extent to which
plant managers are delegated the decision-making authority in several matters (such as hiring,
marketing, and product introduction).
2
The recent paper by Bloom, Sadun and Van Reenen (2010) also found that the delegation
of decision making from firm owners to managers is higher in more-developed countries/regions
such as the U.S., Japan, and Europe than in less-developed countries such as Brazil, China,
and India. They argued that the lack of delegation in the latter countries may be due to the

2
 Figures 1–3 around here.

Second, the dynamic processes of organizational changes are not necessarily

monotonic over time. Decentralized organizations are associated with not only
well-developed phases but also low-development stages. The British Industrial
Revolution in the mid-18th–early 19th century witnessed drastic innovations in
newer technologies such as the steam engine. Such technological changes emerged
with the changes in organizational modes of production (Clark (1994), Landes
(1968), Mokyr (2010)). The predominant organizational form of production be-
fore the British Industrial Revolution was the putting-out system: employers rent
materials to workers who were located in geographically dispersed areas and who
controlled the production process and output. However, the moral hazard prob-
lem on the side of the workers who produced low-quality goods became more
severe as the market expanded. As a consequence, the employers (capitalists)
created a new organizational form called the factory system where workers were
collected on a shop floor and the control of the production process was centralized
in the hands of the capitalists.
The factory system was related to the creation of new technologies in the mid-
18th century England (see, for example, Landes (1968) and Mokyr (2010)).3 With
regard to the extent to which the decision-making authority is delegated to agents,
the British Industrial Revolution can be interpreted as follows: because economic
development proceeded owing to the significant technological improvements, the
organizational mode of production changed from the decentralized mode (putting-
out) to the centralized mode (factory).
However, as is well known from the studies by Chandler (1962, 1977), in
the late 19th century, the organizational mode of firms again changed to the
decentralized mode. During this period, in the U.S., the personal firms with
small-scale production and owner-managers were replaced with MBEs that have
large-scale production and salaried managers who are delegated the decision-
making authority by the firm owners (see Figure 4 for the evidence that the
number of people engaging in managerial jobs has been increasing along with the
growth of the U.S. economy).4 In the early 20th century, technological advances
such as railways resulted in the expansion of the U.S. economy, which led to
dominance of family-owned and family-managed firms there. Acemoglu et al (2007) also showed
related evidence that firms closer to the frontiers of technology are more decentralized.
3
Williamson (1985) claimed that the factory system saved transaction costs, which were
larger under the traditional putting-out system. Here, transaction costs can be interpreted as
inclusive of the cost of workers’ moral hazard (embezzlement) under the putting-out system.
Geraghty (2008) provided empirical evidence that the factory system was complementary to
the technological changes occurring then, such as quality control and investment in machinery
during the British Industrial Revolution.
4
With regard to Japanese firms, Morikawa (2001) provided a related fact. The number
of salaried managers in large Japanese firms significantly increased between 1905 and 1930,
when the Japanese economy entered the modern development stages. Only 5 among the 75
large, leading firms in Japan had two or more salaried managers in 1905. However, this figure

3
the change in the form of the MBEs (see, for example, the cases of GM and Du
Pont) from Unitary form (U-form), wherein the top management coordinates and
monitors functional departments, to Multi-divisional form (M-form), wherein the
top management focuses only on strategic decision making and delegates daily
decision making to the division managers (see Fligstein (1985) and Palmer et al
(1987) for related evidence. Figure 5 confirms the fact that the number of firms
adopting the M-form has been increasing after World War II).5

Figures 4 and 5 around here.

Based on these empirical and historical facts, we develop a dynamic general

equilibrium model wherein we incorporate organizational choice into an over-
lapping generations model. In our model, there are two possible organizational
modes: (1) centralized organization wherein each principal (project owner) holds
the authority over all the decisions (two decisions in our model) that affect the
revenue of her projects (e.g., product specification, marketing decisions, and so
on), and (2) decentralized organization wherein the principal delegates some (one
in our model) of decision-making authorities to agents working on her projects.
The delegation of the decision-making authority has two competing effects. Since
the principal must incur the overload inefficiency when she makes all the decisions
at a time, from the view point of efficient division of labor, the authority to make
some decision should be delegated to the agent. We refer to this effect as the
delegation gain. On the other hand, when the principal delegates a decision to
the agent, she faces the moral hazard problem regarding the decision making by
the agent because then the agent has discretionary power to control the decision.
Thus the principal may need to leave some rent to the agent who is motivated
to make the decision aligned to her interest. In other words, the principal loses
a control power over the decision delegated to the agent. We refer to this effect
as control loss.
The economy-wide technological improvement and the increase in the mar-
ket wage favor the delegation gain more than the control loss. First, when the
technological level increases, the demand for the principal’s products increases,
and consequently, they then need to organize the productions by more efficient
division of labor such that the overload inefficiency is avoided. This makes the
delegation more attractive for the principals. Second and more importantly, the

increased to 29 in 1913 and 113 in 1930 (see Morikawa (2001, pp.49)). These facts show that the
rise of the MBEs was a phenomenon observed in not only in the U.S. but also in other countries
as they entered the modern development era.
5
Related to Chandler’s argument, Williamson (1975) discussed that as the firms grow (the
firm size becomes larger), they face the ’cumulative control loss’ in that it becomes too costly for
the top management to perform all functions of the organization. Beyond the critical limit, the
firms’ organizational mode eventually changes from a centralized mode to a more-decentralized
one.

4
rise in the market wage affects each principal’s incentive problem. Since the mar-
ket wage represents the reservation payoff which each agent could obtain if he
rejected working on the principal’s project, the increase in the market wage re-
duces the rent of the agent that is defined as the surplus the agent can obtain in
excess of his reservation payoff. It becomes cheaper for the principal to capture
the delegation gain when the agent’s rent is small. The increase in the market
wage thus makes the decentralized mode more profitable for the principals.
By using such trade-off of delegation, we present a novel mechanism wherein
the choice of an organizational mode has a dynamic feedback effect on the de-
termination of the market wage and the economic development process. Since
delegation of decision making results in more efficient division of labor, decen-
tralized organization can enhance the profitability of the goods produced by the
principals, which in turn fosters the economic development. Then, this progress
increases the aggregate demand for the goods, and hence the total demand for
labor, which in turn actually raises the market wage. The increase in the mar-
ket wage makes the decentralized organizational mode more profitable for the
principals, resulting in the co-evolution of the organizational mode toward more
decentralization and economic development.6
The main contribution of this paper is to show that there exist equilibrium
paths which exhibit both diverse and non-monotonic changes in the organiza-
tions over time as the economy endogenously develops, and which confirm the
aforementioned empirical and historical evidence. More precisely, we show that
there exist multiple equilibrium paths for similar economies even given the same
initial condition. Some equilibrium paths converge to a steady state with a higher
technological level and more decentralized organizations while others converge to
a steady state with a lower technological and less decentralized organizations.
Unlike the existing literature which stresses the role of increasing returns to
scale, we show that the dynamic interactions between endogenous organizational
choices and market equilibrium conditions lead to the existence of multiple equi-
librium paths: On the one hand, when the principals expect a high market wage (a
high reservation payoff of agents), the rent of the agents becomes low so that the
delegation gain dominates the control loss which thus makes the delegation more
favorable for the principals. The delegation of decision making authority to agent
then exploits more efficient division of labor, which results in a large production
and hence a large labor demand. Thus the market wage actually increases and
the original principal’s expectation about high market wage is confirmed. On the
other hand, when the principals expect a low market wage, the delegation gain
is dominated by the control loss, which thus favors centralized organization with
less efficient division of labor. Then the aggregate production and hence labor
6
Our result is thus consistent with the finding of Claessens, Djankov, Fan and Lang (2000)
that the East Asian corporations differ from each other in terms of the degree of separation
between ownership and control. They find that ’the concentration of control generally diminishes
with the level of a country’s economic development’. (pp.82)

5
demand decreases, resulting in low market wage. Thus the principal’s original ex-
pectation about low market wage is confirmed as well. Such dynamic interaction
between organizational choice and equilibrium market condition causes multiple
development paths of similar economies.
We also show that there exists a non-monotonic equilibrium path wherein the
organization of production changes over time non-monotonically between cen-
tralized and decentralized modes. In early stages of development the principals
choose decentralized organization but they change the organizational mode to
centralized one as the economy develops in an intermediate range. However, as
the economy enters well developed stages, the principals choose decentralized or-
ganization again. This non-monotonic process reflects the dynamic changes of
the trade-off between the delegation gain and control loss: Since the principals
need less effort from agents in early stages of development, the control loss which
would be caused by delegation is small so that they choose decentralized organi-
zation. In the intermediate range of development levels, the principals need high
effort from agents by leaving some rent to them under delegation. However, the
control loss is larger than the delegation gain when the development level is not
so high enough that the principals can gain much from efficient division of labor.
This favors centralized organization. Eventually, as the economy develops well,
the delegation gain becomes larger than the control loss, resulting in the choice
of decentralized organization again.
In addition to these results, we also show that relatively backward economies
with worse initial conditions can catch up and overtake more advanced economies
with better initial conditions because the former adopt the more-efficient pro-
duction modes earlier. This result is useful in understanding why some countries
which were initially lagging behind outperformed those which were initially ahead,
and how such reversal of economic development is related to the differences in
the evolution process of organizational modes across countries (Chandler (1990)
and Landes (1969)).7

Related Literature
Some papers recently attempted to address the macroeconomic implications of
organizational choices. Thesmar and Thoenig (2000) considered organizational
choice in an endogenous growth model. They however simplified the choice of
alternative organizations as if the firms chose two different technologies, one of
which is assumed to incur a fixed operation cost but be more productive than the
other. In contrast, we investigate more explicitly the conflicts within the inter-
nal organizations of firms by considering the trade-off of delegation. Marin and
Verdier (2008) extended the model of Aghion and Tirole (1997) in the general
equilibrium framework and investigated how the allocation of the decision mak-
7
For more evidence, Gershenkron (1962) discussed why Germany caught up with Britain in
the period of the Second Industrial Revolution. Another example is the rapid growth of some
Asian countries such as Korea and Japan after World War II.

6
ing authority in internal organizations is affected by market competition. Their
model is however static; on the other hand, our main interest is in the dynamics
of organizations. Marin and Verdier (2006) tested their theory using the data
for Germany and Austria. They then found that the larger country (Germany)
has bigger firms with more decentralized structures. This finding is consistent
with our theory that more developed countries with larger markets favor more
decentralized organizations, although their focus is on international trade and
market competition which are different from ours. Garicano and Rossi-Hansberg
(2012) investigated the impacts of information and communication technology
on the economic growth process by looking at their effects on organization and
innovation. However, they did not consider the trade-off of delegation. More
importantly, we provide several new insights not present in their paper. The dy-
namic processes through which firm organizations become decentralized are not
uniform but diverse. Because of this diversity, relatively backward economies may
overtake initially advanced ones in the process of development and technological
progress.
Acemoglu, Aghion and Zilibotti (2002, 2006) examined the contracting choice
of intermediate inputs in terms of vertical integration or outsourcing in a growth
model. However, their main interest was the effect of the distance of the economy
from the world technological frontier on the contracting choice, which is not our
concern here. Our main interest is to show that the dynamic processes of the
manner in which firm organizations become decentralized over time are diverse in
the process of economic development. Acemoglu and Zilibotti (1996) investigated
the agency contracts in a dynamic equilibrium model, although their main interest
was the role of asymmetric information in the process of economic dynamics which
differs from ours. Legros, Newmand and Proto (2012) revisited Smith’s classical
thesis about the division of labor but their focus was on the monitoring of workers
in organizations.

The remaining sections are organized as follows. In Section 2, we set up the

basic model of an overlapping generations economy with organizational choice.
In Section 3, we derive the organizational equilibrium, by taking as given the
market prices. In Section 4, we introduce the market equilibrium conditions and
derive the full equilibrium dynamics of the economy. Section 5 concludes the
paper. The Appendix contains all proofs.

2 Model
2.1 Overlapping Generations Economy
We consider an overlapping generations economy with discrete time indexed by
t = 0, 1, 2, .... There is a unique final good which is produced by a continuum of
intermediate goods indexed by [0, 1] and technologies which we will explain below.

7
The final good is used for both consumption and investment. In each period two
types of individuals with measure one each are newly born and each individual
lives for two periods (young and old). One type of individual is an intermediate
good producer (principal ) and the other type is an agent. Each individual is
concerned only with his/her consumption level when old.8 Each young agent
born in period t is endowed with one unit of labor which is inelastically supplied
to the perfectly competitive labor market and earns the market wage wt . Since
young agents are concerned with their consumption level when old, they will save
all their wage income wt and lend it to the competitive credit market. We use ρt to
denote the interest rate of the perfectly competitive credit market. Each old agent
is also endowed with one unit of labor, which is used either for the production
of an intermediate good or supplied to the competitive labor market; this will
be discussed in detail below. Each principal holds a continuum of intermediate
goods and enters the production process of these intermediate goods only when
old.9
In period t, each final good firm produces the final good in the perfectly
competitive sector using a technology At and intermediate goods (xit ) as follows:
∫ 1
Yt = BAt1−α xαit di, (1)
0

where At denotes the technological level used in the production, xit ≥ 0 denotes
the input level of intermediate good i, and α ∈ (0, 1) and B > 0 are parameter
values. We take the final good as numèraire.
In the perfectly competitive R&D sector, there are many competitive R&D
firms, each of which develops a technological level At using capital Kt and labor
Lt as follows:
At = φKt1−β Lβt , (2)
where φ > 0 denotes the R&D productivity. Here, to simplify exposition, we
rule out any uncertainty which might affect the innovation of a technology, even
though it is easy to introduce an idiosyncratic shock into the model.10 The R&D
firms invest in capital Kt one period in advance (i.e., in period t − 1), and capital
fully depreciates after use within one period. Each R&D firm sells its developed
technology At to the final good firms in the perfectly competitive market for

8
This assumption is not essential to our analysis. Our results hold even when we allow young
individuals to consume.
9
As such, a principal does nothing when young. Although we can allow the principal to work
in the perfectly completive labor market when young, we do not consider such a possibility to
simplify exposition.
10
For example, it is possible to consider a more general form At = θt Kt1−β Lβt , where θt ∈ [0, 1]
denotes an idiosyncratic shock which is independently distributed across the R&D firms and is
not serially correlated. We can also allow the R&D productivity φ > 0 to depend on the past
technological level (knowledge). We will comment on this in Section 5.

8
technologies. We use Qt > 0 to denote the price per technology level in the
technology market.
After purchasing a technology At from the R&D sector, each final good firm
chooses
∫ 1 the demand for intermediate goods (xit ) in period t to maximize its profit
Yt − 0 qit xit di by taking the prices of intermediate goods qit ≥ 0 as given. This
gives the following first-order condition:

qit = BαAt1−α xα−1

it , (3)

which gives the demand function of intermediate good i.

Each principal owns one unit mass of intermediate goods. Each principal
first enters the development process of each intermediate good by hiring one old
agent. When the development process of an intermediate good is successful, the
principal can produce the intermediate good by using the final good. When
the development process of an intermediate good is not successful, the principal
obtains nothing from that intermediate good.11
Given the above demand function (3), the principal who succeeded in devel-
oping intermediate good i chooses its production level xit ≥ 0 to maximize its
profit qit xit − xit which gives the following first order condition:

Bα2 A1−α
t xα−1
it = 1. (4)

Equivalently, the optimal price which maximizes the profit is set by qit = 1/α for
all i ∈ [0, 1]. Then, the maximum profit of each successfully developed interme-
diate good before subtracting the wage of an old agent is given by

πt ≡ Bαα/(1−α) (1 − α)At . (5)

In what follows, we will normalize Bαα/(1−α) (1 − α) = 1 so that πt = At .

2.2 Organizational Modes

We suppose that the intermediate goods held by the principals in the economy are
matched with old agents in a one-to-one manner in each period. To simplify this
argument, we assume that the matching process involves no frictions so that every
old agent certainly matches one intermediate good held by some old principal in
each period.12
The development of an intermediate good requires one old agent to be hired;
the outcomes of this development process are stochastic. An intermediate good
is successfully developed with probability ht ∈ (0, 1) such that the principal
11
Since each principal owns a large number of intermediate goods (one unit mass of the
intermediate goods), she can fully diversify the risk of developing these intermediate goods.
12
Our results do not change even when we introduce matching friction in that some old agents
fail to match the principals. In that case, the old agents who do not match the principals go to
the competitive labor market and earn the market wage wt .

9
who holds that intermediate good can enter the production process, while the
development of the intermediate good is failed with probability 1−ht such that the
principal obtains nothing. The success probability ht depends on two decisions
(actions) at ∈ {0, 1} and bt ∈ {0, 1}. That is, we denote by ht = h(at , bt ) the
success probability of an intermediate good in period t. Here we suppose that
each principal can handle both of these actions at a time whereas each old agent
can perform only the action of at due to his limited expertise. We will also denote
by at = ∅ null decision: the principal chooses nothing over the decision at which
is delegated to the matched agent. We assume that actions at and bt are not
observable and hence not verifiable.
Each old principal incurs her disutility of choosing actions as G(at , bt ). Each
old agent incurs his disutility of choosing action at as C(at ) when he is delegated
to make decision at ∈ {0, 1} from the principal.
We then assume the following:

Assumption 1. (i) G ≡ G(1, 1) > g ≡ G(0, 1) = G(1, 0) > G(0, 0) = 0 and

G(∅, b) = 0 for any b ∈ {0, 1}, (ii) c ≡ C(1) > C(0) = 0.

Assumption 1 (i) says that the principal incurs larger disutility of choosing
high actions than low actions when she makes both decisions at a time. Assump-
tion 1 (i) also says that the principal’s disutility becomes larger when she handles
both actions than when she makes only one decision. In particular, when a prin-
cipal delegates action of at to a matched old agent, she incurs a low disutility
G(∅, b) = 0 whereas she will bear a large disutility when she makes both deci-
sions. This assumption reflects the “overload” inefficiency which is caused when
the principal handles both decisions of at and bt by herself at a time because her
expertise and attention are limited. Assumption 1 (ii) says that the marginal
disutility of choosing high action is positive for each old agent who is delegated
to make decision at ∈ {0, 1}.
We also make the following assumption on the success probability h(at , bt ):

Assumption 2. h(1, 1) > h(1, 0) > h(0, 1) > h(0, 0) = 0.

Assumption 2 says that high action increases the success probability of de-
veloping an intermediate good. The requirement that h(1, 0) > h(0, 1) is made
to ensure that there is some benefit to the principal from choosing action at by
herself instead of delegating it to agent. Without this assumption the principals
never choose the centralized control of the decision making, as we will see more
details below.
We suppose that the outcomes of development of each intermediate good are
verifiable. Thus each principal offers to a matched old agent a contract {vtS , vtF }
which specifies a wage payment vti , i = S, F , contingent on the realized outcome,
i = S, F , where i = S (i = F ) means the successful (failed) development of the

10
intermediate good for which the agent works. The old agent then decides whether
or not to accept such a contract. Here, we impose the limited liability constraint
that wages must be non-negative, vti ≥ 0, i = S, F .13 If the old agent rejects it,
he can go to the competitive labor market to earn the market wage wt . Thus,
the market wage wt can be viewed as the reservation payoff of each old agent.
There are two possible organizational modes each old principal can choose.
One is the centralized mode (C-mode) wherein an old principal holds the de-
cision making authorities for both action choices at and bt by herself over an
intermediate good. Although a matched old agent makes neither decision at or
bt in the C-mode, he is necessary for developing a project (for example, agent
has indispensable skill for project development). Thus the principal still needs to
hire one old matched agent in the C-mode. The other is the decentralized mode
(D-mode) wherein the principal delegates the decision making authority of action
choice at ∈ {0, 1} to a matched old agent whereas she keeps the decision over bt
by herself.
The trade-off between the C-mode and the D-mode is as follows. Due to As-
sumption 1, in the C-mode the principal incurs the overload inefficiency when she
chooses these actions at = bt = 1 by herself at a time while she can keep control
powers over both decisions. On the other hand, in the D-mode the principal can
avoid such overload inefficiency by delegating action choice of at to a matched old
agent while she loses the control power over such decision because action choice
by the agent is not observable. Thus, in the D-mode the principal may need to
motivate the agent to work hard (choose high action at = 1) by leaving a rent to
him. Such rent costs the principal who is thus reluctant to delegate the action
choice at to the agent.
The timing of the game in each period is as follows.

1. Intermediate goods held by old principals are matched with old agents in a
one-to-one manner.

2. Each old principal decides whether or not to delegate the authority to make
the decision at ∈ {0, 1} to a matched old agent for each intermediate good.

3. Then, the principal offers a wage contract {vtS , vtF } to the matched old
agent. If the agent accepts it, the game goes to the next stage. Otherwise,
the agent goes to the competitive labor market to earn the market wage
wt .

4. The actions bt ∈ {0, 1} and at ∈ {0, 1} are simultaneously chosen by the

principal and delegated agent.
13
We assume that the interest income each old agent earns ρt wt−1 in period t is not verifiable
and hence the principal cannot extract this income from the matched old agent. This assumption
is made only to simplify the analysis about the optimal contract design although the main results
are not substantially changed even when we assume that the interest income is verifiable (see
Section 5).

11
 5. The outcome of each intermediate good is realized. Then, the wages are
paid according to the contract, and the principal chooses the production
levels of successfully developed intermediate goods.

3 Organizational Equilibrium
Before analyzing the full equilibrium of the model, we will conduct a partial
equilibrium analysis to determine the equilibrium organization, the C-mode or
the D-mode, chosen by each old principal, given the market wage wt and aggregate
technological level At of the economy. We call this organizational equilibrium.
First, we consider the C-mode in which the principal holds the authority to
choose both actions at ∈ {0, 1} and bt ∈ {0, 1} by herself. Since the old agent
chooses nothing for decision at but he is indispensable to develop an intermedi-
ate good, the principal will simply pay the reservation payoff, the wage in the
competitive labor market, wt to the old agent in order to satisfy his individual ra-
tionality constraint. Put differently, the principal offers a contract vtS = vtF = wt
to a matched old agent.
Thus an old principal obtains the following expected payoff from an interme-
diate good in the C-mode in period t:

t ≡ h(at , bt )At − wt − G(at , bt ).

ΠC (6)

Here the principal succeeds in developing an intermediate good and obtains the
revenue At with probability h(at , bt ), pays the wage equal to the market wage
wt to the matched old agent and incurs the disutility G(at , bt ) of choosing both
actions.
Then we can readily show that in the C-mode the principal never chooses
at = bt = 0 due to Assumption 2 (h(0, 0) = 0). Since the principal also never
chooses at = 0 and bt = 1 in the C-mode due to Assumption 1 (i) (G(1, 0) =
G(0, 1) = g) and Assumption 2 (h(1, 0) > h(0, 1)) as well, we can verify that
{ C1
Π (At , wt ) ≡ h(1, 1)At − wt − G if At ≥ (G − g)/(h(1, 1) − h(1, 0)),
Π (At , wt ) ≡
C
ΠC0 (At , wt ) ≡ h(1, 0)At − wt − g otherwise

Next, we consider the D-mode in which the principal delegates the decision
of at ∈ {0, 1} to a matched old agent. In this case the principal offers a contract
{vtS , vtF } to the matched old agent, which induces him to choose an appropriate
action at . In either case of implementing high or low action at ∈ {0, 1} from the
agent, the principal always chooses high action bt = 1 for the other decision bt
which she makes by herself due to Assumption 1 (i), G(∅, bt ) = 0. This is because
the principal can release herself from the overload for handling both decisions by
herself in contrast to the C-mode.
When the principal implements low action at = 0 from the old agent, she will
simply pay vtS = vtF = wt . Then the principal obtains the expected payoff as

12
follows:
ΠD0 (At , wt ) ≡ h(0, 1)At − wt . (7)
On the other hand, when the principal implements high action at = 1 from the
old agent, the contract must satisfy the incentive compatibility (IC) constraint

h(1, 1)vtS − c ≥ h(0, 1)vtF (IC)

which is simplified to vtS − vtF ≥ c/∆h where ∆h ≡ h(1, 1) − h(0, 1) > 0. Also,
the contract must satisfy the individual rationality (IR) constraint

h(1, 1)vtS + (1 − h(1, 1))vtF − c ≥ wt (IR)

as well as the limited liability (LL) constraint

vtS ≥ 0, vtF ≥ 0. (LL)

Here recall that each old agent could earn wt in the competitive labor market if
he rejected a contract offer {vtS , vtF } and worked in the competitive labor market.
Then we can readily show the following lemma (proof is simple and hence
omitted).

Lemma 1. Suppose that an old principal implements high action at = 1 from a

matched old agent in the D-mode. Then the optimal contract becomes (i) vtS =
c/∆h and vtF = 0 if R ≡ h(0, 1)c/∆h ≥ wt and (ii) any (vtS , vtF ) which satisfies
(IC), (IC) and (IR) as equality if R < wt .

When the market wage wt is so low that R > wt , the old agent obtains a
positive rent R − wt > 0 because (LL) and (IC) binds while (IR) is slack. On the
other hand, when wt ≥ R, (IR) becomes binding so that the old agent obtains
no rent at all.
From Lemma 1, we see that the principal obtains the following expected payoff
in the D-mode when she implements high action at = 1 from a matched old agent:

ΠD1 (At , wt ) ≡ h(1, 1)At − max{wt + c, ch(1, 1)/∆h}. (8)

Thus the maximal expected payoff of an old principal in the D-mode is given by

ΠD (At , wt ) ≡ max{ΠD1 (At , wt ), ΠD0 (At , wt )}. (9)

By defining
à ≡ c/∆h, (10)
and
w∗ (A) ≡ −∆hA + ch(1, 1)/∆h, (11)
we can show the following lemma.

13
Lemma 2. The optimal contract in the D-mode is characterized as follows: (i)
the principal implements high action at = 1 from a matched old agent when

(At , wt ) ∈ D1 ≡ {(At , wt ) ∈ ℜ2+ | At ≥ Ã and wt ≥ w∗ (At )}

(ii) the principal implements low action at = 0 from a matched old agent when

(At , wt ) ∈ D0 ≡ ℜ2+ \ D1 .

respectively (see Figure 6).

In the D-mode each principal induces a matched old agent to choose high
action at = 1 when the technological level At and market wage wt are large.
When At is large, the principal needs to implement high action from the old
agent in order to respond to the increase in her profit πt = At . Also, and more
important thing, the rise of the market wage wt increases the reservation payoff
of each old agent, which makes the rent R−wt which is needed to implement high
action becomes smaller. Thus, when the market wage wt is large, the principal
tends to implement high action from a matched old agent. Note here that, when
(IR) is binding (wt > R) and hence the agent’s rent disappears, the market wage
does not affect the principal’s choice of which action should be implemented from
the agent.
Next, we will investigate the principal’s optimal organizational choice. Whether
the C-mode or the D-mode is profitable for an old principal depends on both the
market wage wt and the aggregate technological level At of the economy. Taking
the difference between the principal’s payoffs in these organizations ΠD (At , wt )
and ΠC (At , wt ), we will show that the D-mode dominates the C-mode when the
market wage wt and/or the aggregate technological level At are high.
To this end, we define
g
A≡ . (12)
h(1, 0) − h(0, 1)

This implies that ΠC0 (A, wt ) = ΠD0 (A, wt ) when at = 1 and bt = 0 in the C-
mode while at = 0 and bt = 1 in the D-mode. Thus the principal prefers the
C-mode with at = 1 and bt = 0 to the D-mode with at = 0 and bt = 1 if and only
if At ≥ A because of Assumption 2.
Then there are two cases, Ã > A and A > Ã, to determine which ΠC (At , wt )
or ΠD (At , wt ) becomes larger than the other. Althoug we can deal with both
cases similarly, we will pay our attention to the case that à > A in the main
text, by relegating the case of A > Ã in the Appendix.

Assumption 3. Ã > A.

14
 Note that Assumption 3 implies c > g: old agent chooses high action at = 1
at larger disutility than the principal does when she makes only the decision of
bt = 1.
In what follows we also make the following assumption.

Assumption 4. G > ch(1, 1)/∆h.

Assumption 4 means that it becomes too costly for the principal to choose
high actions for both decisions at = bt = 1 in the C-mode as compared to the
D-mode. This reflects the fact that the principal incurs a large overload loss
when she handles multiple decisions by herself at a time. We merely impose
Assumption 4 in order to keep the analysis simple, although we can slightly
modify our characterization result of equilibrium paths even when we remove
Assumption 4 as long as G is larger than c.14
By defining

w(A) ≡ −[h(1, 1) − h(1, 0)]A + ch(1, 1)/∆h − g, (13)

and
c−g
 ≡ >0 (14)
h(1, 1) − h(1, 0)
(due to Assumption 3), we can show the following result.

Proposition 1. Suppose that Assumption 1-4 hold. Then the organizational

equilibrium in period t (given technological level At and market wage wt ) is de-
termined as follows: each old principal chooses

(i) D-mode with at = 0 and bt = 1 if At ∈ D0∗ ≡ {At | 0 ≤ At < A},

(ii) D-mode with at = bt = 1 if (At , wt ) ∈ D1∗ ≡ {(At , wt ) ∈ ℜ2+ | At ≥ Â and wt ≥ w(At )},

(iii) C-mode with at = 1 and bt = 0 if (At , wt ) ∈ C ∗ ≡ ℜ2+ \ D0∗ ∪ D1∗

respectively (see Figure 7).

In Figure 7, we depict Proposition 1. The region labeled D0∗ is the set of

(At , wt ) for which the principals optimally choose the D-mode with low action
at = 0 implemented from the matched old agents. The region labeled C ∗ is the
set of (At , wt ) for which the C-mode becomes optimal for the principals where
at = 1 and bt = 0. The region labeled D1∗ is the set of (At , wt ) for which the
principals optimally choose the D-mode with high action at = 1 implemented
from the matched old agents. As can be seen from the figure, the equilibrium
14
If G < c, the principal never chooses the D-mode because it is cheaper for the principal to
choose high actions for both decisions by herself rather than delegating a decision of at to the
agent.

15
organization moves from the D-mode with low action to the C-mode and then
to the D-mode with high action as the technological level At increases over time.
More interesting implication of Proposition 1 is that a higher market wage wt
is more likely to favor the D-mode as compared to the C-mode, given the same
aggregate technological level At > Â. This will be crucial for deriving multiple
equilibrium paths as we will see below.
The intuition behind Proposition 1 is explained as follows. When the tech-
nological level of the economy At is so low that At < A, each principal does not
need to implement high action at = 1 from a matched old agent in the D-mode.
Then, the principal prefers the D-mode to the C-mode because the overload cost
incurred under the latter, measured by g, is larger than lower chance of develop-
ing an intermediate good, measured by h(0, 1), under the former. On the other
hand, when the technological level At is so large that At ≥ Â, the principal
wants to implement high action at = 1 from a matched old agent in the D-mode.
Then two competing effects of delegation arise: first, by delegating a decision
at ∈ {0, 1} to a matched old agent, the principal obtains the delegation gain
h(1, 1)At − h(1, 0)At > 0 as the difference between the expected revenues from
the D-mode and the C-mode. Here note that the principal can always choose
high action bt = 1 for the non-delegated decision in the D-mode while she never
chooses high actions for both decisions in the C-mode due to Assumption 4. On
the other hand, by delegating a decision at ∈ {0, 1} to an agent, the principal
incurs the control loss because she cannot directly control the delegated decision
and hence may have to leave a positive rent to the agent who has the discre-
tionary power over the delegated decision. We define such control loss as the
differences between the expected wages of the agent in the D-mode and the C-
mode, max{wt + c, ch(1, 1)/∆h} − (wt + g). Since c > g holds by Assumption 3,
the control loss is positive. When the market wage wt is large, the rent of the
agent becomes smaller and hence the control loss becomes lower than the delega-
tion gain. Thus, the principal prefers the D-mode to the C-mode for large market
wages wt ≥ w(At ). Finally, in the intermediate regions of the technological levels
(A < At < Â), the delegation gain is dominated by the control loss so that the
C-mode becomes optimal.
We refer to case (i), (ii) and (iii) of Proposition 1 as the low D-mode equilib-
rium, the high D-mode equilibrium and the C-mode equilibrium, respectively. On
the other hand, on the boundary of the two different regions shown in Proposition
1, each principal is indifferent between the two different organizational modes.
For example, consider a point (wt , At ) on the curve wt = w(At ) with At > Â
(see a point M in Figure 7). Then, each principal is indifferent between choosing
the C-mode and the D-mode with implementation of high action at = 1. In such
a case, some positive fraction among the principals choose the D-mode and the
others choose the C-mode. We call this the mixed-mode equilibrium wherein the
principals who choose different organizational modes co-exist in the same period.
Let ztD ∈ [0, 1] and ztC ∈ [0, 1] denote the fraction of the principals who choose

16
the D-mode with high action at = 1 and those who choose the C-mode in period
t, respectively. Thus, 1−ztD −ztC denotes the fraction of the principals who choose
the D-mode with low action at = 0 in period t. Proposition 1 then determines the
equilibrium fractions of the principals who choose different organizational modes,
ztD ∈ [0, 1] and ztC ∈ [0, 1], depending on the aggregate technological level At and
market wage wt . For example, when (wt , At ) lies in the region D1∗ in Figure 7,
all principals choose the D-mode with high action at = 1. Thus we have ztD = 1
and ztC = 0. Further, when (wt , At ) ∈ C ∗ , the principals choose the C-mode,
and hence ztC = 1 and ztD = 0. We will thus use z D (At , wt ) and z C (At , wt )
to denote the equilibrium fractions, by making their dependency on At and wt
explicit. With regard to the mixed-mode equilibrium, we will see later how these
fractions ztD and ztC are determined after we derive the labor market equilibrium
condition.

4 Market Equilibrium Conditions

We now turn to the derivation of the market equilibrium conditions and combine
them with the organizational equilibria obtained in the previous section.

4.1 Technology Market

In the competitive technology market, each final good firm buys a technology
At at per unit price Qt to maximize its profit Yt − Qt At . This then yields the
following first-order condition:
∫ 1
−α
B(1 − α)At xαit di = Qt . (15)
0

By combining the choice of xit (see (3)) and the optimal price qi,t = 1/α for all
i ∈ [0, 1] with this, we can show that

Qt = B(1 − α)α2α/(1−α) ht , (16)

where
ht ≡ ztD h(1, 1) + ztC h(1, 0) + (1 − ztD − ztC )h(0, 1)
denotes the average number of the principals holding successfully developed in-
termediate goods. Here, recall that ztD = z D (At , wt ) ∈ [0, 1] and ztC = z C (At , wt )
denote the fractions of those principals who choose the D-mode with high action
and the C-mode, respectively. A principal choosing the D-mode with high action
at = 1 can succeed in developing an intermediate good with probability h(1, 1),
while a principal choosing the C-mode can successfully develop an intermediate
good with probability h(1, 0). A principal choosing the D-mode with low ac-
tion at = 0 succeeds in developing an intermediate good with probability h(0, 1).

17
Thus, ht is the average of the fractions of the principals who succeed in devel-
oping intermediate goods across different organizational modes. In what follows,
we will use α̃ ≡ B(1 − α)α2α/(1−α) .
Given Qt = α̃ht , each competitive R&D firm chooses capital Kt and labor Lt
to maximize its profit Qt At − wt Lt − ρt Kt , where At = φKt1−β Lβt . This gives the
following first-order conditions:

wt = φβKt1−β Ltβ−1 α̃ht (17)

and
ρt = φ(1 − β)Kt−β Lβt α̃ht . (18)

4.2 Labor Market

Using the fact that labor supply comes from one unit mass of young agents,
the labor market clears if Lt = 1. From this and (17), we can determine the
equilibrium wage in the labor market as follows:

wt = φα̃βht Kt1−β (19)

which can be written as

wt = w(At |ztD , ztC ) ≡ γht At , (LME)

where At = φLβt Kt1−β = φKt1−β is the aggregate technological level of the econ-
omy in period t, and we denote γ ≡ β α̃. We define the equilibrium market
wages in different organizational equilibria as wD (At ) ≡ w(At |1, 0), wC (At ) ≡
w(At |1, 0), and wd (At ) ≡ w(At |0, 1). Here, wD (At ) > wC (At ) > wd (At ) holds
due to Assumption 2 (h(1, 1) > h(1, 0) > h(0, 1)).
In a mixed-mode equilibrium, ztD ∈ [0, 1] and ztC ∈ [0, 1] are determined
as follows. If At > Â and wt = w(At ), then each principal is indifferent be-
tween the C-mode and the D-mode with high action at = 1 . As such, we set
ztD + ztC = 1, and determine ztD ∈ (0, 1) such that it satisfies (LME) for wage
w(At ): w(At ) = α̃β[ztD h(1, 1) + (1 − ztD )h(1, 0)]At . Further, when At = A, each
principal is indifferent between the C-mode and the D-mode with low action
at = 0. Then, ztD = 0 holds and ztC ∈ [0, 1] can be any value for which (LME) de-
termines an equilibrium market wage wt = α̃β[ztC h(1, 0) + (1 − ztC )h(0, 1)]A.
In such a case, the equilibrium market wage wt can be any value satisfying
wt ∈ [α̃βh(1, 0)A, α̃βh(0, 1)A].

4.3 Equilibrium Dynamics

In the credit market, the total demand for capital Kt+1 must be equal to the
total credit supply from the savings of young agents (wage income wt ):

wt = Kt+1 . (CME)

18
Since Kt+1 = (At+1 /φ)1/(1−β) and wt is given by (LME), (CME) can be rewritten
as
w(At |ztD , ztC ) = λ(At+1 ; φ) (20)
where λ(A; φ) ≡ (A/φ)1/(1−β) . Thus λ is increasing and convex function.
Then, from (LME) and Lemma 2, the following dynamical system

w(At |ztD , ztC ) = λ(At+1 ; φ)

ztD = z D (At , w(At |ztD , ztC )),
ztC = z C (At , w(At |ztD , ztC ))

determines an equilibrium path {At , ztD , ztC }∞ t=0 , given an initial condition A0 =
1−β
φK0 . A steady state {A, z , z } of the economy is then defined as the solution
D C

to λ(A; φ) = w(A|z D , z C ), z D = z D (A, w(A|z D , z C )), and z C = z C (A, w(A|z D , z C )).

We define the high D-mode steady state value of A, denoted by AD , as wD (AD ) =
λ(AD ; φ), and the low D-mode steady state value of A, denoted by Ad , as
wd (Ad ) = λ(Ad ; φ). Further, we define the C-mode steady state value of A,
denoted by AC , as wC (AC ) = λ(AC ; φ). Under our assumptions, we can verify
that the steady state solutions defined above—AD , Ad , and AC —are all uniquely
determined.
The explicit solutions for these steady states are given by

AD ≡ φ1/β [α̃βh(1, 1)]β/(1−β) , AC ≡ φ1/β [α̃βh(1, 0)]β/(1−β) , Ad ≡ φ1/β [α̃βh(0, 1)]β/(1−β)

so that AD > AC > Ad due to Assumption 2.

For a mixed-mode steady state, we define A such that λ(A; φ) = w(A|z D , z C )
for some z D ∈ (0, 1) and z C ∈ (0, 1). For example, when each principal is
indifferent between the C-mode and the D-mode with high action at = 1 in
a steady state A′′ , there arise two cases. First, when (IR) is binding in the D-
mode, we have A′′ = Â. Second, when (IR) is not binding in the D-mode, we have
w(A′′ ) = w(A′′ |z D , 1 − z D ) in the steady state. As such, this steady-state value of
A′′ is determined by λ(A′′ ; φ) = w(A′′ ) and the equilibrium fraction z D ∈ (0, 1)
is found by solving λ(A′′ ; φ) = w(A′′ |z D , 1 − z D ). Finally, when each principal
is indifferent between the C-mode and the D-mode with low action at = 0 in a
steady state A′ , A′ = A must hold in such a steady state.
We need the condition for which the profit of an old principal is non-negative
in any period t. Since the principal can obtain the profit max{ΠC (At , wt ), ΠD (At , wt )}
by her optimal organizational choice, it suffices to ensure that ΠD (At , wt ) ≥ 0
in any period t, which holds if h(0, 1)At − wt ≥ 0. This can be rewritten as
h(0, 1)At ≥ α̃βht At due to (LME). Using Assumption 2, the sufficient condition
for this is that α̃ and β are small enough as in the following.

Assumption 5. h(0, 1) ≥ α̃βh(1, 1).

19
 In what follows, we will maintain Assumption 5, and deal with the character-
ization of equilibrium paths.
First, we show that, if the productivity parameter φ is so large, then the
economy eventually converges to the high D-mode steady state AD . To see this,
we define A∗ as wC (A∗ ) = w(A∗ ) where such A∗ uniquely exists and is given by

ch(1, 1)/∆h − g
A∗ ≡ > 0. (21)
(γ − 1)h(1, 0) + h(1, 1)

Here both the denominator and nominator are positive because h(1, 1) > h(1, 0),
and Assumption 3 implies that c > g and hence ch(1, 1)/∆h > g.
Since AC and Ad are increasing in the productivity parameter φ and both
A∗ and  are independent of φ, we have AC ≥ max{Â, A∗ } and Ad ≥ A when
φ is large. In particular we define φ as the lowest value of φ which satisfies
AC ≥ max{Â, A∗ } and Ad ≥ A. Then AC ≥ max{Â, A∗ } and Ad ≥ A for all
φ ∈ (φ, ∞).
Then, we can show the following result (see Figure 8).

Proposition 2. Suppose that the productivity parameter φ is so large that φ ∈

(φ, ∞). Then, the equilibrium paths have the feature that the economy eventually
converges to the high D-mode steady state (AD , z D = 1, z C = 0).

Proposition 2 implies that as the economy develops with technological progress,

the organization of firms change to obtain the more advantages available from de-
centralized forms—such as MBEs and M-form modes—delegating more decision-
making authority to the hired agents, which confirms the finding of Chandler
and Williamson. Moreover, the positive association between per capita GDP
and the more decentralized form of organizations is consistent with the empiri-
cally observed phenomenon (see Figure 1). Furthermore, our result confirms the
empirical finding that organizations become more decentralized in more techno-
logically advanced economies which invest more in technologies such as IT and
R&D (see Figures 2 and 3).15
The intuition behind this result is explained as follows. In the early stages of
development both the market wage wt and technological level At are relatively
low. In particular, suppose that A0 ∈ (A, Â). 16 Then, the delegation gain is
dominated by the control loss because a small market wage implies that a large
rent must be left to the matched old agents in the D-mode (R−wt becomes larger),
and a small technological level implies a small demand for intermediate goods.
Both these effects need less motivation to elicit high action from the old agents.
Thus, the principal prefers the C-mode to the D-mode, resulting in the C-mode
equilibrium. However, as technological progress fosters economic development
15
See also Acemoglu et al (2007) for related evidence.
16
The case of A0 < A can be similarly handled.

20
(At increases), both market wage and technological level increase together so
that the demand for intermediate goods increases over time. Then, the trade-off
between the delegation gain and the control loss is changed in favor of the former
because the rent R − wt of the old agent in the D-mode becomes smaller, and the
larger demand for intermediate goods requires more effort from the old agents.
Thus, the principals prefer the D-mode to the C-mode in well-developed stages.
Furthermore, the dynamic general equilibrium interactions between technological
progress and organizational choice reinforce this process. The change from the
C-mode to the D-mode can exploit more efficient division of labor, which in turn
raises the aggregate technological level and market wage over time. This actually
confirms the optimal choice of the D-mode by the principals.
Our result also helps to understand why some relatively backward economies
which were initially lagging behind succeed in catching up and sometimes over-
taking those which were initially ahead.17 In Figure 9, we depict two different
economies which have different initial conditions A′0 and A′′0 where A′′0 > A′0 . In
the early stages of development, the economy with better initial condition A′′0 is
ahead the economy with the worse initial condition A′0 . However, as technological
progress fosters economic development, the relatively backward economy adopts
the D-mode earlier than the initially advanced economy. The reason why the
relatively backward economy can overtake the initially advanced economy is that
the former adopts the D-mode with high action at = 1 in more early stages of
development than the latter. Hence what timing the economy switches to more
decentralized organization becomes crucial for determining the success and failure
of the economy.

4.4 Diverse Development of Organizations

Next, we turn to more interesting case where the R&D productivity φ is in the
intermediate range. Then, we will show that there exist multiple equilibrium
paths which converge to different steady states—the C-mode steady state and
the high D-mode steady state—as shown in Figure 10. To see this more formally,
we will make the following parametric condition: Â < AC < A∗ . The condition
of  < A∗ is equivalent to w(A∗ ) < R, i.e.,

ch(1, 1)/∆h − g
γh(1, 0) < ch(1, 1)/∆h − c
(γ − 1)h(1, 0) + h(1, 1)

which is satisfied for all γ ∈ (0, γ) for some γ. The condition that  < AC < A∗
is then satisfied for some intermediate values of φ, given γ ∈ (0, γ), because both
 and A∗ are independent of φ and AC is increasing in φ with AC = 0 at φ = 0
and limφ→∞ AC → +∞. Then we can find a unique φ∗ > 0 such that AC = Â
17
Relevant examples might be the growth of Germany and the U.S. relative to Britain in the
late 19th century, the rise of some East Asian countries such as Japan and Korea after the World
War II, and so on. See also Chandler (1990) and Gershenkron (1962).

21
 ∗ ∗
holds at φ∗ . We also define φ such that AC = max{Â, A∗ } holds at φ where
∗
max{Â, A∗ } = A∗ for γ ∈ (0, γ). Thus we have  < AC < A∗ for all φ ∈ (φ∗ , φ ),
given γ ∈ (0, γ).
Then, we can show the following result (see Figure 10).

Proposition 3. Suppose that the R&D productivity φ is intermediate such that

∗
φ ∈ (φ∗ , φ ) given γ ∈ (0, γ). Then, there exist multiple equilibrium paths which
converge to different steady states—the C-mode steady state and the high D-mode
steady state—even when the economy starts with the same initial condition A0 ∈
(A, AC ).

In Figure 10 we depict two different equilibrium paths both of which starts

with the same initial condition A0 ∈ (A, AC ). One equilibrium path has the
C-mode in any period and converges to the C-mode steady state AC . The other
equilibrium path has the C-mode before it reaches the C-mode steady state. Just
before reaching the C-mode steady state, the organizational equilibrium changes
from the C-mode equilibrium to the high D-mode equilibrium. Such “jump”
of equilibrium path is possible because both the C-mode and the high D-mode
equilibria become organizational equilibria given the same technology level At .
After that, the economy converges to the high D-mode stead state.
Thus, even when economies have the same initial condition A0 , it is possi-
ble that they experience different development processes with different patterns
of evolution of organizational modes. Some economies may be stuck in a low-
development steady state AC with the C-mode while others develop well and
eventually are in the high D-mode steady state. This result confirms the recent
empirical finding that there are significant differences in the decentralization of
firms across countries and that the degrees of decentralization are positively as-
sociated with the development and technological levels of these countries (Bloom
and Van Reenen (2009), Bloom, Sadun and Van Reenen (2010)).
Multiple equilibrium paths are derived from a novel dynamic mechanism
which has not been addressed in the literature which has emphasized the roles of
increasing returns to scale to obtain multiple equilibria. The key to our result is
not increasing returns to scale but the dynamic general equilibrium interactions
between the market wage (reservation payoff of old agents) and organizational
choices; this is given as follows. Anticipating a high market wage wt , the rent left
to the old agent in the D-mode, R − wt , becomes lower which favors the D-mode.
Then, since the D-mode can gain from efficient division of labor, the success prob-
ability of intermediate goods increases and the aggregate technological level of
the economy goes up. This in turn raises the capital and labor used in the R&D
sector which actually results in an increase in the market wage. This reinforces
the process that the principals tend to choose the D-mode more often. On the
other hand, anticipating the low market wage wt , the rent of the old agent in

22
the D-mode becomes larger, and then the principals tend to choose the C-mode.
Since the C-mode suffers from the overload inefficiency causing the inefficient di-
vision of labor, the success probability of intermediate goods decreases such that
the capital and labor used in the R&D sector decrease as well, confirming the
low market wage.
It should be emphasized here that multiple equilibrium paths never arise when
the incentive problem on the action choice of old agents is absent in the D-mode.
In such a first-best case, there are no rents left to the old agents in the D-mode,
and hence the above mechanism based on self-fulfilled expectations does not work
anymore. In this case, the equilibrium path is unique and converges to a unique
steady state.

4.5 Non-Monotonic Changes of Organizations

Next, we will show the equilibrium path which displays the feature that equi-
librium organizations start with the decentralized mode but change to the cen-
tralized mode and then to a decentralized mode again over time as the economy
develops well.
We here suppose that φ ≥ φ and that A0 < A where recall the definition of
φ: φ is the lowest value of φ which satisfies AC ≥ max{A∗ , Â} and Ad ≥ A.
Then we can show the following result (see Figure 11).

Proposition 4. Suppose that φ ≥ φ and A0 < A. Then, there exists an equilib-

rium path in which the equilibrium organization non-monotonically changes over
time as the economy develops well (At increases over time) in the following man-
ner: from the D-mode with low action to the C-mode and then to the D-mode
with high action.

In Figure 11 we depict an equilibrium path which displays the non-monotonic

feature shown in Proposition 4. When At < A, the low D-mode equilibrium
becomes an organizational equilibrium because the principal does not need to
elicit high action at = 1 from a macthed old agent when the economy is in low
development phase. As the economy enters the intermediate range of development
(At ∈ (A, max{Â, A∗ })), the principal needs to implement high action at = 1 from
a matched old agent under delegation but she must then leave a positive rent to
him for doing so. This is the control loss caused by the delegation. When the
technology level At is not so high enough, the control loss becomes larger than the
delegation gain. Thus the principals choose the C-mode in the intermediate range
of technology levels At . Eventually, as the economy enters well developed stages
with At ≥ max{Â, A∗ }, the delegation gain becomes larger than the control loss
so that the high D-mode equilibrium emerges as an organizational equilibrium.
This pattern of the equilibrium path captures the aforementioned histori-
cal observation that decentralized production organization modes such as the

23
putting-out system preceded the rise of more centralized production organiza-
tion modes such as the factory system, but as technological progress proceeded
further in the 20th century, the production organization mode became more de-
centralized with the emergence of MBEs and the M-form, which Chandler and
Williamson have extensively discussed.

5 Conclusions
In this paper, we have investigated how organizational modes of production evolve
over time in the process of economic development with technological progress.
First we have shown that the principals of organizations tend to move from a
centralized mode to a decentralized mode when the economy develops well. This
is because delegating the decision making authority to the hired agents reduce the
overload cost incurred by the principals and the trade off between the delegation
gain and the control loss changes in favor of the former as economic development
proceeds further. Following this, we have characterized equilibrium paths and
have shown that the equilibrium patterns of organizational change are not unique
but rather diverse over time: some economies adopt more decentralized mode
in the early stages of development while others are stuck in more centralized
mode for longer periods. Third, equilibrium organizational modes change from
a decentralized mode to a centralized mode and then to a decentralized mode
again over time along with development process. We have also discussed that
these theoretical predictions are consistent with several empirical and historical
observations.
We conclude the paper by discussing several extensions of the model. First,
as we have mentioned, we have thus far assumed that the interest income each
old agent can earn ρt wt−1 is not contractible. Thus, the payment to the matched
old agent must be non-negative owing to the limited liability (LL) constraint.
However, if the interest income is verifiable, the principal can make a contract
{vtS , vtF } which satisfies vti ≥ −ρt wt−1 for i = S, F . Thus, the LL constraint
becomes less stringent, and hence, the D-mode becomes more preferable to the
principals than the C-mode, as compared to the case where the interest income
is not contractible. Further, since ρt wt−1 = φ(1 − β)Kt1−β ht , the interest income
increases over time as the economy develops well through capital accumulation,
and hence the limited liability constraint is more likely to be satisfied along with
the process of economic development. Thus, the parameter range for which the D-
mode with high action becomes an equilibrium can be broader when the interest
income is verifiable.
Second, as empathized in the theory of endogenous growth, we can allow
technological progress to depend on the past accumulated technological level (ac-
cumulated knowledge) of the economy. Specifically, suppose that the current
technological level At in period t depends on the average technological level ac-

24
cumulated up to the previous period t − 1, At−1 : At = φ(Kt1−β Lβt )f (At−1 ) where
the function f captures the effect of the past technology At−1 on the current tech-
nological level. Then, we can show that the equilibrium dynamics of the basic
model are modified as At+1 = φf (At )(α̃βht At )1−β . However, as long as the effect
of the past technology is not so strong in the sense that limA→∞ f (A)/Aβ < 1 is
satisfied, we can verify that the equilibrium path {At }∞ t=0 converges to a steady
state A, and has features similar to the results characterized in Propositions 2–4.

6 Appendix A: Proofs
6.1 Proof of Lemma 1
Note first that, when implementing high action at = 1 from a matched old agent,
(IR) is binding if and only if wt ≥ R. When wt ≥ R (hence (IR) is binding),
the principal prefers implementation of high action at = 1 to low action at = 0 if
and only if At ≥ Ã due to the definition of Ã. On the other hand, when wt < R,
(IR) is not binding for implementing high action. Then, the principal decides to
implement high action rather than low action if and only if wt ≥ w∗ (At ) due to
the definition of w∗ (A). Since w∗ is decreasing and w∗ (Ã) = R, we can derive
the set D1 as depicted in Figure 6 in which the principal prefers implementing
high action to low action from a matched old agent. Q.E.D.

6.2 Proof of Proposition 1

First, we show that the principal never chooses at = bt = 1 in the C-mode, in
which case she obtains the expected payoff ΠC (At , wt ) = h(1, 1)At − wt − G. If
the principal chose the D-mode with at = bt = 1, she would obtain ΠD (At , wt ) =
h(1, 1)At − max{wt + c, ch(1, 1)/∆h}, which is greater than ΠC (At , wt ) due to
Assumption 4: when wt + c > ch(1, 1)/∆h, ΠD (At , wt ) = h(1, 1)At − wt − c >
ΠC (At , wt ) = h(1, 1)At − wt − G because of G > ch(1, 1)/∆h > c. When wt + c ≤
ch(1, 1)/∆h, ΠD (At , wt ) = h(1, 1)At −ch(1, 1)/∆h > h(1, 1)At −G ≥ ΠC (At , wt ).
Since the principal can always choose the D-mode with at = bt = 1, she can ensure
at least h(1, 1)At − max{wt + c, ch(1, 1)/∆h} and hence she never chooses the
C-mode with at = bt = 1. Thus Assumption 2 implies that at = 1 and bt = 0
hold in the C-mode which gives ΠC (At , wt ) = h(1, 0)At − wt − g.
Next, we characterize the organizational equilibria.
When At < A, we have à > A > At so that à > At and hence the principal
prefers implementation of low action to high action in the D-mode. Then, when
At ≤ A, we have ΠD (At , wt ) = h(0, 1)At − wt which is greater than ΠC (At , wt ) =
h(1, 0)At − wt − g.
On the other hand, when A < At < Â, we consider two separate cases: First,
when A < At < Ã, ΠD (At , wt ) = h(0, 1)At − wt again which is however less
than ΠC (At , wt ) because of At > A. Second, when à < At < Â, ΠD (At , wt ) =

25
h(1, 1)At − wt − c if wt ≥ R. In such case ΠC (At , wt ) > ΠD (At , wt ) due to
At < Â. On the other hand, if à < At <  but wt < R, then ΠD (At , wt ) =
h(1, 1)At − ch(1, 1)/∆h which is also less than ΠC (At , wt ) because à < At < Â
and wt < R together imply w(At ) > w(Â) = R > wt so that w(At ) > wt
and hence ΠC (At , wt ) > ΠD (At , wt ). Finally, suppose that At > Â. Then, if
wt > w(At ), we have ΠD (At , wt ) = h(1, 1)At − wt − c > ΠC (At , wt ) when wt ≥ R
and ΠD (At , wt ) = h(1, 1)At − ch(1, 1)/∆h > ΠC (At , wt ) when R > wt ≥ w(At )
respectively. On the other hand, when wt < w(At ) (and hence wt < R) for
At ≥ A, ΠD (At , wt ) = h(1, 1)At − ch(1, 1)/∆h < ΠC (At , wt ). Q.E.D.

6.3 Proof of Proposition 2

Suppose that γ ∈ (0, γ) where γ is defined in the main text such that A∗ > Â for
all γ ∈ (0, γ). Suppose also that φ > φ, which implies that AC > max{Â, A∗ }
and Ad > A. Note then that wC (A) > λ(A; φ) for all (A, wC (A)) ∈ C ∗ and
that wd (A) > λ(A; φ) for all (A, wd (A)) ∈ D0∗ . Thus the economy must leave the
regions D0∗ and C ∗ at some finite period in any equilibrium path.
Now we show that, once the economy enters the region D1∗ in which the
high D-mode equilibrium becomes an organizational equilibrium, it never leaves
such region D1∗ . To show this, define A′′ such that wD (A′′ ) = w(A′′ ) (such A′′
exists and is unique). Note here that A′′ < A∗ . Then we also define A′ such
that wD (A′′ ) = λ(A′ ; φ). Since AC > A∗ holds by assumption and λ(A; φ) cuts
wC (A) from below at A = AC , we have wC (A∗ ) > λ(A∗ ; φ). Then, wD (A′′ ) =
w(A′′ ) = λ(A′′ ; φ) and wC (A∗ ) = w(A∗ ) imply that wC (A∗ ) = w(A∗ ) < w(A′′ ) =
wD (A′′ ) = λ(A′ ; φ) and hence wC (A∗ ) < λ(A′ ; φ). Thus, λ(A∗ ; φ) < wC (A∗ ) <
λ(A′ ; φ) which shows that A′ > A∗ because λ is increasing in A.
Now suppose that (AT , wT ) ∈ D1∗ holds at some period T . Thus wT =
wD (AT ). Then we have AT ≥ A′′ and wD (AT ) ≥ wD (A′′ ). Thus AT +1 is de-
termined by the condition wD (AT ) = λ(AT +1 ; φ). By definitions of A′′ and A′ ,
we then have AT +1 > A′ . Since A′ > A∗ , we must have AT +1 > A∗ so that
wC (AT +1 ) > w(AT +1 ). Thus, (AT +1 , wC (AT +1 )) ∈ D1∗ which shows that the
C-mode equilibrium with (AT +1 , wC (AT +1 )) cannot be an organizational equi-
librium. Also, since A∗ > A, we have AT +1 > A which implies that the low
D-mode equilibrium cannot be an organizational equilibrium at period T + 1.
Hence, wT +1 = wD (AT +1 ) must hold for such AT +1 so that the high D-mode
equilibrium becomes only organizational equilibrium at period T + 1. Repeating
this process, we establish that, once the economy enters the region D1∗ in which
the high D-mode equilibrium becomes an organizational equilibrium, it never
leaves such region forever. Once the economy lies in the region D1∗ , it eventually
converges to the high D-mode steady state (AD , z D = 1, z C = 0). Q.E.D.

26
6.4 Proof of Proposition 3
∗
Suppose that φ ∈ (φ∗ , φ ), which implies that  < AC < A∗ . Note that, since
 > A, we have A < AC < A∗ as well (see Figure 10). Suppose also that
A0 ∈ (A, AC ).
Then, we can construct the following different equilibrium paths (Figure 10):
First, we take a path of At such that it starts with A0 ∈ (A, Â) and has the C-
mode equilibrium in any period, which eventually converges to the C-mode steady
state AC . Since At ∈ (A, A∗ ) and wC (At ) < w(At ) for all At ∈ (Â, A∗ ) in any
period along this path, the C-mode can be actually organizational equilibrium
until At approaches AC . The second equilibrium path is the following: The
economy starts from the same initial value A0 ∈ (A, Â) and At increases over
time according to wC (At ) = λ(At+1 ; φ) until it reaches the C-mode steady state
AC . Before reaching AC , At jumps up and increases according to wD (At ) =
λ(At+1 ; φ). Such jump is possible because wD (AC ) > w(AC ) > wC (AC ) so that
both the C-mode and the D-mode with high action at = 1 can be organizational
equilibria, given the same technology level At which is close to AC . After then,
the economy converges to the high D-mode steady state AD . Therefore we have
established that there exist multiple equilibrium paths even for the same initial
condition A0 ∈ (A, Â). Q.E.D.

6.5 Proof of Proposition 4

Suppose that the initial condition A0 < A holds. Suppose also that φ > φ, which
implies that A < Ad and max{Â, A∗ } < AC .
Then we can construct the following equilibrium path (see Figure 11): Since
A0 < A, the economy starts with the D-mode with low action at = 0. Since
wd (A) > λ(A; φ) for all A ∈ (A0 , A) due to Ad > A, an organizational equilibrium
can be the D-mode with low action at = 0 in any period from t = 0 until the
economy reaches A. After that, the economy switches to the C-mode equilibrium
until it leaves the region C ∗ . This is possible because AC > max{Â, A∗ } and
hence the path {At } which satisfies wC (At ) = λ(At+1 ; φ) lies in the region C ∗
for all At ∈ (A, max{Â, A∗ }), for which the C-mode becomes an organizational
equilibrium. After the economy passes max{Â, A∗ }, the D-mode with high action
at = 1 becomes an organizational equilibrium in any period. Then the economy
converges to the high D-mode steady state AD . Q.E.D.

7 Appendix B: Case of A > Ã

In this Appendix we will discuss the case where A > Ã although we have so far
assumed the case that à > A in the main text.
Suppose that A > Ã. Then the regions of (At , wt ) which determine orga-
nizational equilibria can be depicted in Figure A1. The difference between the

27
current and previous cases is that the region for which the C-mode becomes op-
timal, denoted by region C ∗ , becomes narrower. However, as shown in Figure
A1, we can derive the similar feature of equilibrium path to the previous case
that, as the economy develops over time (At increases over time), equilibrium
organizations change from the D-mode with low action to the C-mode and then
to the D-mode with high action over time. Also, we can show that there exist
multiple equilibrium paths even for the same initial condition as in the previous
case. This is because there exist multiple organizational equilibria given the same
technological level At > A: Given a At > A, if the principals expect a high wage
wt such that (wt , At ) ∈ D1∗ , they will choose the D-mode with high action. Then,
the equilibrium market wage becomes actually high and lies in the region D1∗ .
On the other hand, if they expect a low market wage wt such that (wt , At ) ∈ C ∗ ,
then they will choose the C-mode, which in turn implies that the equilibrium
market wage becomes actually low and lies in the region C ∗ .

References
[1] Acemoglu, D., Aghion, P. and F. Zilliboti (2006), “Distance to Frontier,
Selection, and Economic Growth,” Journal of the European Economic As-
sociation, 4, 37–74.
[2] Acemoglu, D., Aghion, P., Lelarge, C., Van Reenen, J. and F. Zilibotti
(2007), “Technology, Information, and the Decentralization of the Firm,”
Quarterly Journal of Economics, November, 1759–1799.
[3] Aghion, P. and J. Tirole (1997), “Real and Formal Authority in Organiza-
tions,” Journal of Political Economy, 105, 1–29.
[4] Bloom, N. and J. Van Reenen (2009), “The Organization of Firms Across
Countries,” mimeo.
[5] Boolm, N, Sadun, R., and J. Van Reenen (2010), “Recent Advances in the
Empirics of Organizational Economics,” mimeo.
[6] Chandler, Jr. A. D. (1962), Strategy & Structure: Chapters in the History
of the Industrial Enterprise, MIT Press, Cambridge MA.
[7] Chandler, Jr. A. D. (1977), The Visible Hand: The Managerial Revolution
in American Business, Cambridge, Mass.: Belknap Press.
[8] Chandler, Jr. A. D. (1990), Scale and Scope: The Dynamics of Industrial
Capitalism, Harvard University Press.
[9] Claessens, S., Djankov, S. and L. H.P. Lang (2000), “The Separation between
Ownership and Control in East Asian Corporations,” Journal of Financial
Economics, 58, 81–112.

28
[10] Clark, G. (1994), “Factory Discipline”, Journal of Economic History, 54,
128-163.

[11] Fligstein, N. (1985), “The Spread of the Multidivisional Form Among Large
Firms, 1919–1979” American Sociological Review, 50, 377–391.

[12] Garicano, L. and E. Rossi-Hansberg (2012), “Organizing Growth,” Journal

of Economic Theory, 147, 623-656.

[13] Geraghty, T. (2007), “The Factory System in the British Industrial Revo-
lution: A Complementarity Thesis”, European Economic Review, 51, 1329-
1350.

[14] Gershenkron, A. (1962), Economic Backwardness in Historical Perspective:

A Book of Essays, Cambridge, Mass.

[15] Historical Statistics of the United States: Earliest Times to the Present,
Millennial Edition, Vol. 2, Part B: Work and Welfare. Cambridge University
Press.

[16] Landes, D. S. (1969), The Unbound Prometheus: Technological Change and

Industrial Development in Western Europe from 1750 to the Present, Cam-
bridge University Press.

[17] Legros, P., Newman, A. F. and E. Proto (2012), “Smithian Growth through
Creative Organization,” mimeo.

[18] Marin, D. and T. Verdier (2006), “Corporate Hierarchies and the Size of
Nations: Theory and Evidence,” mimeo.

[19] Marin, D. and T. Verdier (2008), “Market Power Inside the Firm and the
Market,” Journal of European Economic Association, 6, 752–788.

[20] Mokyr, J. (2010), The Enlightened Economy: An Economic History of

Britain 1700-1850, Yale University Press.

[21] Mokyr, J. (2004), The Gifts of Athena: Historical Origins of The Knowledge
Economy, Princeton University Press.

[22] Mokyr, J. (1998), “The New Economic History and the Industrial Revo-
lution”, 1-127, in Joel Mokyr, ed., The British Industrial Revolution: An
Economic Perspective, Second Edition, Westview Press.

[23] Mokyr, J. (1992), Lever of Riches: Technological Creativity and Economic

Progress, Oxford University Press.

[24] Morikawa, H. (2001), A History of Top Management in Japan: Managerial

Enterprises and Family Enterprises, Oxford University Press.

29
[25] Palmer, D., R. Friedland, P. D. Jennings and M. E. Powers (1987), “The
Economics and Politics of Structure: The Multidivisional Form and Large
US Corporations,” Administrative Science Quarterly, 32, 25–48.

[26] Thesmar, D. and M. Thoenig (2000), “Creative Destruction and Organiza-

tional Choice,” Quarterly Journal of Economics, 115, 1201–1237.

[27] Williamson, O. E. (1975), Markets and Hierarchies, Analysis and Antitrust

Implications: A Study in the Economics of Internal Organization, New York,
Free Press.

[28] Williamson, O. E. (1985), The Economic Institutions of Capitalism, Free

Press.

30
 3.5
Sweden
Average Score of Decentralization
3
Germany UK US
2.5 India Portugal
Poland France
2 Italy
Greece Japan
China
1.5
1
0.5
0
0 10000 20000 30000 40000 50000
per
Capita GDP

Figure 1: The vertical axis represents the average score of decentralization from Bloom and Van Reenen (2009).
This is simple average of all scores measuring how much plant managers are delegated decision making about
hiring, marketing, and introduction of new products from firm owners.
For each of these decision making, the scores of decentralization are ranged from 1 (least decentralization)
to 5 (most decentralization). The horizontal axis represents per capita GDP (measured in US dollars)
from IMF, World Economic Outlook Database, April 2006. We took GDP in 2006 to match the data
of Bloom and Van Reenen who conducted the survey research in 2006.

3.5
UK
3
US Sweden
Poland
Portugal
Average Score of
Decentralization

2.5
IndiaItaly Germany
2
Greece China France Japan
1.5
1
0.5
0
0 1 2 3 4 5
R & D Expenditure Share (% of GDP)

Figure 2: The date about R & D expenditure share (% of GDP) is from

IMF, World Economic Outlook Database, April 2006.
 3.5
UK Sweden
Average Score of Decentralization 3
US
Italy Germany
2.5 India
Portugal
Poland China
2 France
Greece Japan
1.5

0.5

0
4 4.5 5 5.5 6 6.5 7 7.5 8
IT Expenditure Share (% of GDP)

Figure 3: The date about IT expenditure share (% of GDP) is from

IMF World Economic Outlook Database, April 2006
0.12 20000
18000
0.1
16000
14000
0.08
12000
0.06 10000 Ratio of US Managers
8000 per capita GDP of US
0.04
6000
4000
0.02
2000
0 0
1860
1870
1880
1900
1910
1920
1930
1940
1950
1960
1970
1980

Figure 4: Ratio of US managers who work at managerial jobs relative to total

number of occupations in US. These jobs include department heads of stores, conductors in railroad,
floor men and managers in stores, managers and superintends in building, and so on.
Source: The data about the number of managers is from
Historical Statistics of US:Earliest Times to the Present, Millennial Edition.
The date about per capita GDP of US is 1990 International Geary-Khamis dollars
from Maddison (2008).

120

No MDF
100
Adopt MDF
80
MDF t

60

40

20

0
1919‐29 1929‐39 1939‐49 1949‐59 1959‐69 1969‐79

Figure 5: Source: Fligstein (1985, p.386)

No MDF: percentage of the firms that do not adopt MDF (MultiDivisional Form)
Adopt MDF: percentage of the firms that adopt MDF during decade
MDF t: percentage of the firms that have MDF at beginning of decade
wt

R D1

D0
wt = w∗ (At )

0 Ã At

Figure 6: Optimal Contracts in D-mode

D0 : it becomes optimal to implement low action at = 0.

D1 : it becomes optimal to implement high action at = 1.
wt

R
D1∗

M wt = w(At )
D0∗ C∗

wt = w∗ (At )

0 A Ã Â At

Figure 7: Organziational Equilibria

D0∗ : low D-mode equilibrium with low action at = 0.

D1∗ : high D-mode equilibrium with high action at = 1.

C ∗ : C-mode equilibrium M : mixed mode equilibrium.

wt , At+1
λ(At+1 ; φ)

wD (At )
D1∗
D0∗ C∗

wC (At )

w(At )

C AD At
0 A A0 Â A∗ A

Figure 8: Convergence to the high D-mode steady state AD .

wt , At+1
λ(At+1 ; φ)

wD (At )

C∗ D1∗

wC (At )

w(At )

A′0 A′′0 At
0

Figure 9: Overtaking
 wt , At+1
λ(At+1 ; φ)

wD (At )

D0∗ C∗ D1∗

wC (At )

w(At )

∗
0 A A0 Â AC A AD At

Figure 10: Multiple Equilibrium Paths

 wt , At+1

λ(At+1 ; φ)
wD (At )
D0∗ D1∗
C∗

wC (At )

wd (At )

w(At )

0 A0 A AC AD At

Figure 11: Non-Monotonic Changes of Organizations

 wt

D1∗

wt = w∗ (At )

D0∗
wt = w(At )
C∗

0 Ã A
At

Figure A1: Organizational Equilibria (when A > Ã)