ESSAYS IN POLITICAL ECONOMY
by
Sang Hyun Kim

A DISSERTATION
Submitted
to Michigan State University
in partial fulﬁllment of the requirements
for the degree of
Economics - Doctor of Philosophy
2013

ABSTRACT
ESSAYS IN POLITICAL ECONOMY
by
Sang Hyun Kim
This dissertation analyzes various issues in Political Economy. In the ﬁrst chapter, I revisit the issue
that has been popularly debated, namely whether corporations afﬁliated in business groups build
stronger ties to the government. With the focus on the Japanese political economy, this chapter
theoretically and empirically investigates how a ﬁrm’s incentive to build political connections is
affected by its ownership structure. I ﬁrst show that intercorporate stockholding encourages the
equity issuer to build political network if the equity holder exercises non-trivial controlling power
over the equity issuer, but discourages it otherwise. Next, I analyze a large data set of Japanese
corporations publicly traded from 1991 to 2003. The empirical analysis conﬁrms the theoretical
predictions: the number of retired elite bureaucrats in a private ﬁrm’s board of directors, which
is a measure of political connection of the ﬁrm, increases as the share of equity held by nonﬁnancial ﬁrms (friendly shareholders) decreases or as that by ﬁnancial institutions (controlling
shareholders) increases. These ﬁndings suggest that companies afﬁliated to business groups, in
contrast to popular belief, might build relatively weaker political connections.
By reinterpreting Jean-Jacques Rousseau who considered self-government and self-regulation
of people as the fundamental problem of politics, the second chapter explores several fundamental
issues of political economy including how to allocate resources to achieve efﬁcient self-regulation,
how the agency problems in different parts of a society interact each other, and why the governments of poor societies work so poorly. The analysis shows that once a society is sufﬁciently
developed in economic and political spheres, the agency problem on the citizens’ side becomes
negligible, in which case citizens (and researchers) can entirely focus on the issue of government
accountability. When a political community is economically and politically poorly developed,
however, the agency problem on the citizens’ side exacerbates agency problems of other parts of

the society. Therefore, in such a case one should explicitly take the problem of self-regulation of
the citizens as an integrated part of the entire political economic system.
In the last chapter, I develop a spatial voting model in which political elites play an active role
in increasing or decreasing polarization. Key assumptions are: (i) voters respond to changes in
policy positions of parties only if they pay attention to politics; (ii) political elites can disinterest
a speciﬁc group of voters away by making the voters believe that implemented policies will be
less preferable to them. Under intermediate range of parameters, the model generates multiple
equilibria, i.e. political elites can choose whether to polarize their policy platforms or not. Either
when economic inequality sufﬁciently grows or when media tend to mobilize partisans sufﬁciently
more than they do centrists, polarization becomes the unique equilibrium.

ACKNOWLEDGEMENTS

This dissertation would not have been possible without the guidance and the help of several individuals who in one way or another contributed and extended their valuable assistance in the
preparation and completion of this study. My utmost gratitude goes to my main advisor, Jay Pil
Choi for his patience and steadfast encouragement to complete this study. I am truly indebted
and grateful to my other committee members, Christian Ahlin, John D. Wilson and Charles Hadlock, whose encouragement, supervision and support from the preliminary to the concluding level
enabled me to develop the subjects.
I would also like to express my gratitude to my parents and elder brother, Dong-hyun for their
love and support.

iv

TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
CHAPTER 1
1.1
1.2

1.3

1.4

1.5

OWNERSHIP STRUCTURE AND POLITICAL CONNECTION:
THE CASE OF JAPAN . . . . . . . . . . . . . . . . . . . . . . . .
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.2 Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data and Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.2 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.1 Estimation with ﬁxed-effects . . . . . . . . . . . . . . . . . . . . . . . .
1.4.2 Estimation with the instruments . . . . . . . . . . . . . . . . . . . . . .
1.4.3 Delayed responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.4 Subsample analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

CHAPTER 2
ON THE OPTIMAL SOCIAL CONTRACT
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Environment . . . . . . . . . . . . . . . . . . . .
2.2.2 Government . . . . . . . . . . . . . . . . . . . . .
2.2.3 Timing of the events . . . . . . . . . . . . . . . .
2.2.4 Equilibrium . . . . . . . . . . . . . . . . . . . . .
2.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.1 Structure of the government . . . . . . . . . . . .
2.3.2 State capacity and stability . . . . . . . . . . . . .
2.4 Other Issues . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1 Illegitimate government . . . . . . . . . . . . . . .
2.4.2 Poverty trap . . . . . . . . . . . . . . . . . . . . .
2.4.3 Civic virtue . . . . . . . . . . . . . . . . . . . . .
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

1
1
6
6
13
15
15
18
21
21
25
25
26
28

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

30
30
34
34
36
37
38
42
45
48
50
50
52
54
55

CHAPTER 3
VOTER ATTENTION AND POLITICAL POLARIZATION
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Basic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

57
57
60
64
70

v

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vi

71

LIST OF TABLES

Table 1.1

Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Table 1.2

First Stage Regression on the Instruments . . . . . . . . . . . . . . . . . . . . . 21

Table 1.3

FE Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

Table 1.4

IV Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Table 1.5

Delayed Responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

Table 1.6

Subsample Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

vii

LIST OF FIGURES

Figure 1.1

Trends of economy-wide ﬁnancial ﬁrms’ shareholding and non-ﬁnancial ﬁrms’ . 19

viii

CHAPTER 1
OWNERSHIP STRUCTURE AND POLITICAL CONNECTION: THE CASE OF JAPAN

1.1

Introduction

Recent empirical works have shown that business groups are pervasive in modern economies. La
Porta et al. (1999), Claessens et al. (2000) and Faccio and Lang (2002) provide evidence on the
ubiquity of business groups in various countries across the world. The relationship that business
groups have with other parts of society considerably varies depending on historical, political and
societal context. But in most economies where business groups play an important role, critics and
the public have suspected and raised concerns about the possibility that the economic giants, utilizing their vast resources, might build strong connections with politically powerful groups, and
unlevel the playing field. Despite such widespread concern and interest, academic investigation on
political behavior of financially interlinked firms has been scarce until recently. Recent works by
some financial economists have discovered a series of interesting aspects of political economy of
pyramidal business groups in developing countries. They show that most large business groups are
formed with government support, and enjoy close ties to the government for long periods of time.
1

These findings, together with the occasional news about marital ties and personal networks be-

tween political and economic elites, seem to justify the above mentioned concern. Other scholars,
however, have found that in some advanced economies, companies affiliated in business groups
have weaker ties to the government in contrast to the public’s belief. For instance, Colignon and
Usui (2003) and Raj and Yamada (2009) report that group affiliation of a Japanese company is
correlated negatively with the strength of its political connection which is measured by the number
of former high-ranked government officials hired as board members. Once this inconsistency is
taken seriously, a couple of new questions arise: why are the government-business relationships
1

Khanna and Yafeh (2007) and Morck, Wolfenzon and Yeung (2005) provide excellent surveys on this subject.
1

in different countries dissimilar? Why do group afﬁliated ﬁrms build weaker or stronger political connections? What is the underlying mechanism which generates the discrepancy? These
are questions that are difﬁcult to answer in a satisfactory manner, but addressing them properly
is undeniably essential in understanding the political economy of business groups. This paper, as
a ﬁrst step, examines whether intercorporate stock ownership, the deﬁning characteristic of business groups, indeed encourages private ﬁrms to build stronger political networks, and explores the
conditions under which ﬁnancially interlinked corporations have relatively weaker political connections. The focus of this paper is on the Japanese political economy which is an ideal laboratory
for this investigation particularly due to its extensive reciprocal shareholding practice. In the ﬁrst
part of the paper, I develop a simple theoretical framework in which ﬁnancially and politically related ﬁrms simultaneously decide how much resources to spend in building political connections.
Key assumption used throughout the analysis is that the political network built by a company
generates positive externality to other ﬁnancially connected ﬁrms. In other words, when a group
afﬁliated ﬁrm builds political network and spends resources in pursuing political goals, the beneﬁts
are shared among the member ﬁrms. 2 At least since Olson (1965), it has been well understood
that political action imposes certain externalities on other political players and that outcomes of
political games are often socially inefﬁcient due to the absence of internalization mechanism. The
problem considered here departs in a signiﬁcant way from the situations that Olson described. Because intercorporate holding of stocks works as a ﬁnancial contract that enables corporations to
at least partially internalize such externalities, the ﬁrms holding each other’s stocks are expected
to enjoy greater political beneﬁts at lower costs. However, this does not immediately imply that
ﬁnancial ties among corporations encourage them to build stronger connections to the government.
The analysis presented in this paper shows that ﬁnancial ties between corporations can strengthen
or weaken the incentive for political network building, depending on the type of the relationship
between the equity holder and the equity issuer. The intuition can be clearly demonstrated by
considering the simplest case with two ﬁrms, an equity issuer, call it ﬁrm 1, and an equity holder,
2

Justiﬁcations for this assumption are presented in Section 1.2.

2

ﬁrm 2. 3 Suppose ﬁrst that prearranged institutional barriers or conventional corporate governance
practice, which will be discussed at length below, keep ﬁrm 2 from intervening in the other corporation’s decision making process. In this case, as the share held by ﬁrm 2 increases, ﬁrm 1’s
incentive for building political connection will be weakened for the following reason. Whenever
ﬁrm 2 increases its holding of the other ﬁrm’s equity, it will increase its political expenditure as
well, because it can absorb back a larger amount of political beneﬁt through the strengthened ﬁnancial linkage. And, when ﬁrm 2 participates more enthusiastically in government-related activities,
ﬁrm 1 would be more tempted to free-ride on the other’s effort, and consequently will reduce the
resources assigned to political activities. By contrast, if ﬁrm 2 can bypass the institutional barriers
and exercise controlling power over ﬁrm 1’s business, ﬁrm 1 will increase activities that beneﬁt
ﬁrm 2 and reduce those which harm it. And more importantly, such tendency will become stronger
as the share held by ﬁrm 2 grows up. In this situation, the company owned largely by the other
ﬁrm tends to build a stronger connection with the government, than do ﬁrms of which equity is
diffusely held. In the context of Japanese political economy, this theoretical result can be easily
transformed into empirically testable hypotheses. Since 1980’s, economists who study Japanese
companies have consistently reported that ownership of a corporation is often concentrated to a
small number of large shareholders (or block holders), and that such large shareholders are mostly
two distinct types of corporations: trade partners and ﬁnancial institutions. According to previous studies, companies in trade partnership refrain themselves from participating in other ﬁrms’
governance, and furthermore, protect each other from the risk of hostile takeovers by outsiders.
Financial institutions holding both debt and equity, on the other hand, act as outside monitors who
discipline the managers, and often actively intervene in the ﬁrm’s corporate governance. Combining these narratives with the theoretical result of this paper generates the following hypotheses:
non-ﬁnancial ﬁrms’ shareholding would have a negative impact on the ﬁrm’s political connection,
while ﬁnancial institutions’ would have a positive impact. The second half of the paper is devoted
to testing these hypotheses, by analyzing roughly 2,300 corporations publicly traded from 1991
3

In the analysis below, I consider a more general case of cross-shareholding, i.e. ﬁrm 1 can
hold ﬁrm 2’s equity.
3

to 2003. Following the previous studies mentioned above, I employ the number of retired elite
bureaucrats in a private ﬁrm’s board of directors, called amakudari, as the measure of political
connection of the company. The amakudari practice is not merely idiosyncratic reemployment
arranged by individuals. It is systematically (sometimes ofﬁcially and often not) arranged by ministries and government agencies, and the size of such arrangement is non-trivial by any criteria.
In the sample period, nearly 40 percent of the companies have had at least one former bureaucrat as their board member. These ex-bureaucrats are known to provide channels of information
and negotiations between the public and the private sector. The empirical strategy is built on the
economy-wide ﬁnancial deregulation, so called the "ﬁnancial big bang" which was initiated and
in progress during the sample period. Fuelled largely by the reform, the ownership structure of
most Japanese corporations has undergone rapid change. During the sample period, the share held
by ﬁnancial institutions has dropped from 34 percent to 25 percent, and non-ﬁnancial ﬁrms’ share
from 33.5 to 29.8 percent. If the change in ﬁnancial structure is largely exogenous to individual
ﬁrms’ effort for better political connection, as the ﬁnancial deregulation story suggests, then the
empirical correlation between the change in the ownership structure variables and the change in
the number of amakudari would reveal how the degree of the political connection is affected by the
ownership structure. Regression analysis with ﬁxed-effects shows that the number of amakudari
is negatively correlated with the share of equities held by non-ﬁnancial ﬁrms and positively with
ﬁnancial ﬁrms’ shareholdings. Even if the changes of the ownership structure were largely driven
by the exogenous shock, the estimates are likely to be biased, unless a ﬁrm’s political connection
does not affect its ownership structure at all. Thus, I further explore the causal relationship by
exploiting the information of ﬁrms which have never hired any ex-bureaucrat for the entire sample
period. The results of instrumental variable regression also conﬁrm the theoretical predictions. The
effects of changes in the ownership structure are not only statistically but also economically significant: a ten percent point increase of non-ﬁnancial ﬁrms’ share induces ﬁrms to hire about 0.5 less
ex-bureaucrat. The same change in ﬁnancial ﬁrms’ shareholding encourages the ﬁrm to hire 0.7 additional amakudari. This paper offers an intuitive explanation for the puzzling ﬁnding documented

4

in the previous studies. Some scholars, motivated by the observations from developing countries,
have suspected that Japanese business groups, often called keiretsu, and their member ﬁrms have
better connections with the government. For instance, in a historical review on Japanese business
groups, Morck and Nakamura (2005) conjecture that "[s]ince the great keiretsu ﬁrms included the
most attractive amakudari landing spots and were the most enthusiastic about amakudari, these
groups may have enjoyed advantage, in the short term at least, due to their better connections
with government." 4 Thus, the negative correlation between group afﬁliation and the number of
amakudari reported by Colignon and Usui (2003) and Raj and Yamada (2009) appeared at ﬁrst as
a puzzle. Whereas Colignon-Usui and Raj-Yamada analyze only the cross-sectional distribution of
retired bureaucrats, the present paper explores the deeper mechanism by asking a more articulated
question and by adopting more sophisticated empirical strategies. Group afﬁliated ﬁrms might enjoy larger political beneﬁts at lower costs because they can partially internalize political externality
through the ﬁnancial linkages. At the same time, however, this internalization mechanism might
encourage the ﬁnancially interlinked ﬁrms to free-ride on others’ political effort, which might lead
to lower levels of political expenditure by the member ﬁrms. The current work also provides a
framework to understand the cross-country difference. The theory says that ﬁnancially interlinked
ﬁrms tend to assign relatively more resources to political activities if the equity holder (or parent company) exercises signiﬁcant control over the equity issuer ﬁrms (or subsidiaries). Studies
of pyramidal groups convincingly show that this is indeed the case. So, even though it may not
be directly comparable, the evidence found in this paper is consistent with the empirical patterns
observed in developing economies. This work is related to a burgeoning literature on politically
connected ﬁrms. Analyzing data of manufacturing ﬁrms in the U.S., Agrawal and Knoeber (2001)
show that politically experienced directors are more prevalent in ﬁrms where sales to government,
exports and lobbying are greater. Goldman et al. (2008) ﬁnd positive abnormal stock return fol4

See also Okimoto (1989) who shows that group afﬁliated ﬁrms and banks were "generously
subsidized" after the World War II. Similarly, Beason and Weinstein (1996) ﬁnd that in the postwar Japanese economy, keiretsu afﬁliated ﬁrms were granted more favorable industrial subsidies
in mining business.

5

lowing the announcement of the nomination of a politically connected individual to the board. For
the case of Japan, Miwa and Ramseyer (2005) report that more ex-bureaucrats are found in ﬁrms
doing business with the government. Studies exploring politically connected companies in developing countries are more abundant. For example, see Fisman (2001) for Indonesia, Classens et al.
(2008) for Brazil, and Khwaja and Mian (2005) for the case of Pakistan. By analyzing politically
connected ﬁrms in 47 countries, Faccio (2006) ﬁnds that ﬁrm values increase when ofﬁcers or
large shareholders of the ﬁrms are entering politics. The remainder of the paper is organized as
follows. Section 1.2 analyzes political-network-building incentives of proﬁt-seeking ﬁrms under
changing ﬁnancial environments. In Section 1.3, I provide a description of the data, and present
the empirical strategy. Empirical ﬁndings are in Section 1.4. Finally, I discuss a few limitations of
this work and directions for future work in Section 1.5.

1.2 Theory
1.2.1 Model
Consider two ﬁrms (i, j = 1, 2) which hold each other’s equity, and which simultaneously and
independently decide how much effort and resources to spend in building political network. The
political connection of a ﬁrm imposes a certain externality on the other which might be positive
or negative.5 Let πi (li , l j ) denote the total revenue of ﬁrm i as a function of its own political
connection li and the other’s l j . For expositional simplicity, assume
πi (li , l j ) = πi (li + δ l j )
and πi (·) is increasing and strictly concave. Under this simplifying assumption, the sign and the
size of the externality are determined by a single parameter δ .
The political externality prevailing between ﬁnancially interlinked ﬁrms is expected to be positive on average (i.e. they share political beneﬁts) for at least two reasons. In a detailed analysis
5

For political externalities that arise in various context, see Olson (1963).

6

of the ownership structure of large Japanese corporations, Miwa (1996) documents that the largest
shareholders are mostly corporations, especially ﬁnancial institutions and business partners. When
two ﬁrms are in business partnership, any good news to one ﬁrm which does not drastically alter
the distribution of bargaining power is good to the other as well. For example, suppose a company
assembling components into complete products succeeds in inﬂuencing the government to implement a more favorable policy, and consequently decides to increase its production. The growth in
the demand by the assembler will beneﬁt the component suppliers as well. The beneﬁt potentially
spills over into up and down of the entire supply chain. Of course, political success that improves
the ﬁrm’s general performance will beneﬁt its creditor-shareholders, too.
Information sharing is another good reason to expect positive political externality to prevail
among ﬁnancially connected ﬁrms. Arguably, one of the main goals of building political connection is to get information related to the public policies and regulations. Once a ﬁrm successfully
obtain policy related information, such information can be transmitted to other ﬁrms without any
direct cost (due to non-rivalry). If the managers of the ﬁrm expect losses from information sharing,
then they can freely keep the information inside. Therefore, the externality generated by information sharing is highly likely to be positive in its nature. Studies on keiretsu document that the
group afﬁliated ﬁrms share information through various channels, which allows them to better utilize political connections. Motivated by this observation, I make a key assumption, namely that
a ﬁrm’s political network building effort generates positive externality to other ﬁnancially connected ﬁrms.6 To ensure the existence of stable Nash equilibrium, I further assume the size of the
externality is sufﬁciently small, i.e. 0 < δ ≪ 1.
To introduce the objective of the companies, let us deﬁne the value of ﬁrm i as
Vi (li , l j ) = πi li + δ l j − ci li + q ji π j (l j + δ li ) − c j l j
where q ji denotes the share of equity of ﬁrm j held by ﬁrm i, and ci is the constant marginal cost
6

Of course, we cannot preclude the possibility that some ﬁnancially interlinked ﬁrms suffer
a conﬂict of political interests. However, since my focus is on the average behavior, exceptions
would not change the main message of the paper. At the end of this subsection, I brieﬂy discuss
what theory predicts when negative externality prevails.
7

of the network building. The terms in the ﬁrst square bracket is the proﬁt generated within ﬁrm i,
and the second by ﬁrm j. Because ﬁrm i holds q ji of ﬁrm j’s equity, it is entitled to receive the
corresponding proportion of the cash generated by ﬁrm j.
Maximizing the value Vi would be the objective of ﬁrm i if ﬁrm j does not participate in i’s
decision making process.7 If, however, ﬁrm j is active in the corporate governance of ﬁrm i, its
objective must at least partially reﬂect shareholder j’s interest, and ﬁrm j’s inﬂuence should grow
larger as it holds more equity of ﬁrm i. In sum, ﬁrm i maximizes a weighted average of Vi and V j :
maxVi (li , l j ) + φi j qi jV j (l j , li )
li

(1.1)

where qi j is the share of equity of ﬁrm i held by ﬁrm j. The formula shows that as qi j gets larger,
ﬁrm j is more able to enforce ﬁrm i to make decisions on behalf of ﬁrm j. φi j ≥ 0 is a parameter
introduced to capture pre-arranged implicit contract, conventions and institutional features that are
not explicitly modeled in this simple framework. In particular, if ﬁrm j is a friendly shareholder
(e.g. trading partners) which does not actively participate in ﬁrm i’s decision making, the managers
of ﬁrm i will not take much of ﬁrm j’s interest into account, i.e. they let φi j small . On the other
hand, if j is a controlling shareholder (e.g. ﬁnancial ﬁrms) that has a substantial inﬂuence on
ﬁrm i’s corporate governance, ﬁrm i’s decisions will reﬂect largely ﬁrm j’s interest, which can be
parsimoniously captured by a large φi j .8
It is noteworthy that when either the externality is negligible (δ = 0) or the ﬁrms do not hold the
other ﬁrms’ equity (q12 = q21 = 0), the objective of the ﬁrms boils down to maximizing their own
proﬁt, [πi (li ) − ci li ]. Hence, in the textbook environment where markets are perfectly competitive,
and the ownership of ﬁrms is dispersed among small individual shareholders, the assumption that
ﬁrms maximize (1) is identical to the standard assumption such as proﬁt or shareholder wealth
maximization.9
7

In the context of cross-shareholding, the assumption of ﬁrm-value maximization is previously employed in Farrell and Shapiro (1990) and Clayton and Jorgensen (2005).
8 In the next subsection, I discuss at length the corporate governance practice in Japan with an
emphasis on the difference between friendly and controlling shareholders.
9 According to recent empirical studies on corporate ownership structure, such an environment
8

Because the goal of this section is to derive simple and intuitive predictions, I add restricts on
the ownership structure to focus on the empirically relevant situation. For most pairs of ﬁrms in
reality, share held by another ﬁrm is non-negative but far less than one, so assume in what follows
that q ji ≪ 1 for i, j = 1, 2. I further assume
φi j < 1 − 2qi j q ji /q2j ,
i
which ensures that ﬁrm i weighs the direct beneﬁt given to its own proﬁt more than the external
effect. It is easy to show that under these assumptions, there is unique stable equilibrium of the
game of political network building.
The following proposition describes how ﬁrm i’s equilibrium political expenditure would respond to a change in its ownership structure when ﬁrm j is a friendly shareholder, i.e. when φi j is
small.
Proposition 1 For sufﬁciently small φi j , ﬁrm i lowers its political expenditure in response to an
increase in the share held by ﬁrm j, i.e. ∂ li∗ /∂ qi j < 0.
Proof. Suppose ﬁrst φi j = 0 so that ﬁrm i maximizes its value Vi . In this case, ﬁrm j’s share qi j
does not appear in ﬁrm i’s objective function, which means qi j affects the equilibrium political
expenditure li∗ only through an indirect channel, the change in l j . The ﬁrst-order condition for
program (1) is
πi′ − ci + q ji δ π ′j = 0.
Differentiating this ﬁrst-order condition with respect to q ji and rearranging the terms, one can
derive the following.

′
∂ li (l j |q ji ) −δ π j
=
∂ q ji
SOC

(1.2)

is rather exceptional. For example, La Porta et al. (1999) found that corporate ownership is
concentrated to a small number of shareholders in most countries other than the U.S. and the
U.K.

9

where li (l j |q ji ) is the best response function of ﬁrm i. And, the slope of the best response function
is obtained by differentiating the ﬁrst-ordier condition with respect to l j :
πi′′ + q ji π ′′
∂ li (l j )
j
= −δ ′′
2 π ′′
∂lj
πi + q ji δ j

(1.3)

which is negative, i.e. strategic substitute. Combining (2) and (3) yields
πi′′ + q ji π ′′
∂ li (l j ) ∂ l j (li |qi j )
∂ li∗
−δ πi′
j
=
= −δ ′′
×
∂ qi j
∂lj
∂ qi j
SOC
πi + q ji δ 2 π ′′
j
which is negative given the assumptions. Because the objective function is continuous in φi j , for
φi j in a small neighborhood of zero, ∂ li∗ /∂ qi j is negative.
The logic behind this proposition is quite straightforward. As qi j increases, ﬁrm j’s incentive
to hire former government ofﬁcials also increases because a larger part of the political beneﬁt appropriated by ﬁrm i can be absorbed back through the strengthened ﬁnancial connection. Knowing
this change in ﬁrm j’s incentive, ﬁrm i has an incentive to reduce its political expenditure and freeride on ﬁrm j’s political effort. Consequently, the equilibrium political connection li∗ becomes
lower as ﬁrm i gets more ﬁnancially integrated to ﬁrm j.
A few remarks are worth mentioning. First note that this proposition is obtained by restricting
our attention to the case where qi j is small. If φi j > 0 and ﬁrm j is the dominant shareholder of
ﬁrm i, the above proposition would fail to remain relevant. Second, the above proposition does
not predict that ﬁnancially interlinked or group afﬁliated ﬁrms will enjoy smaller political beneﬁts.
Even if each member ﬁrm builds a weaker connection, they might enjoy greater political beneﬁts
than stand-alone ﬁrms do, by internalizing the externalities. For instance, suppose the positive externality is mainly originated from information sharing among ﬁnancially connected companies, so
the political beneﬁts stand-alone corporations enjoy are solely from its own expenditure. Assume,
for simplicity, φi j = φ ji = 0. In this case, a stand-alone ﬁrm k’s ﬁrst-order condition is
′ S
πk lk − ck = 0,

whereas that of a group afﬁliated ﬁrm is
πi′ liG + δ l G − ci + q ji δ π ′j l G + δ liG = 0
j
j
10

′ S
where the last term in the equation is positive. Hence, πi′ liG + δ l G < πk lk so long as ck is
j
S
much smaller than ci , that is, even if liG may be smaller than lk , the total beneﬁt liG + δ l G must
j
S
be greater than lk .

The ﬁrst proposition highlights the incentives to free-ride on other ﬁnancially connected ﬁrms.
Not surprisingly, this incentive is sufﬁciently mitigated when a shareholder ﬁrm actively participates in the corporate governance of the equity issuer ﬁrm, i.e. when φi j is large.
Proposition 2 For sufﬁciently large φi j , ﬁrm i increases its political expenditure in response to an
increase in the share held by ﬁrm j, i.e. ∂ li∗ /∂ qi j > 0.
Proof. The ﬁrst-order condition for program (1) is
1 + φi j q2j
i

πi′ − ci + q ji + φi j qi j δ π ′j = 0.

As ﬁrm i increases its holding of ﬁrm j’s equity, the best-response li (l j ) will move outward, i.e.
δ π ′j
∂ li (l j |q ji , qi j )
=−
> 0.
∂ q ji
2 π ′′ + q + φ q δ 2 π ′′
1 + φi j qi j i
ji
ij ij
j
However, its size shrinks as φi j increases. Therefore, for a sufﬁciently large φ j , ﬁrm i does not
increase its political spending much in response to an increase in q ji , which implies that the incentive analyzed in the previous proposition is sufﬁciently mitigated. On the contrary, the direct effect
of a change in ownership structure does not shrink. The direct effect of a change in ownership
structure on the best response function of ﬁrm i is
2φi j qi j πi′ − ci + φi j δ π ′j
∂ li (l j |q ji , qi j )
=−
∂ qi j
1 + φi j q2j πi′′ + q ji + φi j qi j δ 2 π ′′
i
j
q +φ q
φi j δ π ′j 1 − 2qi j ji i j 2i j

=−

1+φi j qi j

1 + φi j q2j πi′′ + q ji + φi j qi j δ 2 π ′′
i
j

which is positive since φi j < 1 − 2qi j q ji /q2j by assumption. And, it does not shrink as φ j
i
increases. Therefore, for a sufﬁciently large φ j , the direct effect dominates and the equilibrium
political expenditure increases as the other ﬁrm’s share increases.
11

The intuition is the same with the standard logic of internalization via integration: as a ﬁrm
(or subsidiary) gets ﬁnancially more integrated to another company (or parent company), the subsidiary’s goal will become more aligned with that of the parent ﬁrm, and they will behave more
cooperatively. This effect is particularly strong when ﬁrm j is able and willing to exercise considerable power on the managers of ﬁrm i.
The contrast between the two propositions is striking. The free-ride effect highlighted in Proposition 1 has not been widely recognized in the literature probably because most previous studies
has focused on pyramidal business groups within which there are a controller and the controlled.
Within groups of trade partners which cross-hold each other’s stocks, however, the free-ride effect
may dominate the cooperation effect which is emphasized in Proposition 2. Given that the two
effects push the ﬁrm toward the opposite directions, crucial in empirical analysis is to discern the
friendly and the controlling shareholders. In the next subsection, I selectively survey the literature
on corporate governance of the Japanese ﬁrm so as to identify shareholders which tend to exercise
controlling power and those which do not.
Lastly, I discuss brieﬂy the case where the positive externality assumption is violated. If negative political externality prevails between ﬁnancially interlinked ﬁrms i and j, intercorporate shareholding would have negative impacts on their political connections regardless of whether the shareholder is controlling or friendly. To see why, suppose ﬁrm j is a friendly shareholder. As ﬁrm j
holds more of ﬁrm i’s equity, ﬁrm j will lower its political expenditure, because a larger negative
externality will ﬂow back through the ﬁnancial linkage. In turn, this change leads to a decrease in
ﬁrm i’s political expenditure, since with negative externality, the political network building game
is of strategic complement, i.e. li (l j )/l j > 0. Next, suppose ﬁrm j is a controlling shareholder. As
the share held by ﬁrm j increases, it is more able to force ﬁrm i to lower activities that do harm to
ﬁrm j’. Thus, ﬁrm i will become less enthusiastic in political network building.

12

1.2.2 Prediction
Since at least 1980s’, researchers have reported that some of the stylized facts observed in the
U.S. ﬁnancial market are absent in the Japanese counterpart. Allen and Zhao (2007) describe the
Japanese system simply as "shareholders are not rulers." Economists have identiﬁed a couple of
reasons why the inﬂuence of individual shareholders is particularly limited in the Japanese corporation. First, the boards of directors which typically are dominated by insiders (senior employees)
do not guide and discipline the managers to work for shareholders’ best interests. Instead, both
directors and managers are expected to make decisions for the sake of broader stakeholders, especially creditors and employees.10 Another signiﬁcant difference between the U.S. and Japanese
ﬁnancial markets is that hostile takeovers which are quite common in the U.S. ﬁnancial market
are extremely rare in the Japanese market. The primary reason for this is that cross-shareholdings
were put in place by many Japanese corporations to prevent hostile takeovers. It means that market
discipline which enables the U.S. shareholders to control large corporations plays only a limited
role in the Japanese economy.
Despite all these institutional barriers and implicit contracts, however, one should not jump
to the conclusion that every shareholder is powerless. While friendly shareholders (also called as
antei-kabunushi, meaning stable shareholders), mostly the corporations in business partnership,
stay away from others’ corporate governance unless there is a risk of hostile takeover, ﬁnancial
institutions holding both debt and equity, on the other hand, act as outside monitors with signiﬁcant
inﬂuence, guiding and sometimes replacing the managers (see Aoki and Patrick, 1994). Kaplan
and Minton (1994), Kang and Shivdasani (1995) and Yafeh and Yosha (2003) provide evidence
of active intervention by ﬁnancial institutions. Shleifer and Vishny (1997) note that "their power
comes in part because of a variety of control rights they receive when ﬁrms default or violate
debt covenants (Smith and Warner, 1979) and in part because they typically lend short term, so
10

For more details, see Aoki (1990) and Miwa (1996) for instance. Berglof and Perotti (1994)
assess the issue from more theoretical perspective, but they do not distinguish controlling and
friendly shareholders.

13

borrowers have to come back at regular, short intervals for more funds."11
This observation suggests that when a corporation appears as a shareholder, it is likely to be a
friendly shareholder if it is a non-ﬁnancial ﬁrm, and a controlling one if a ﬁnancial company. Combining this additional information with the theoretical results obtained in the previous subsection,
the following predictions are immediate.
1. As the share of equity held by non-ﬁnancial ﬁrms increases, a measure of political connection
of the ﬁrm decreases.
2. As the share of equity held by ﬁnancial institutions increases, a measure of political connection of the ﬁrm increases.
It is noteworthy that these predictions depart from the hypothesis tested in previous empirical
studies in a few important ways. Previous studies have mainly explored the cross-sectional variation asking whether group afﬁliated ﬁrms have relatively stronger ties to the government. One of
the problems the previous studies suffered is the ambiguity in the deﬁnition of business groups,
namely it is often controversial whether a given ﬁrm is afﬁliated in a business group or not. And
as Miwa and Ramseyer (2002) convincingly show, this ambiguity in group afﬁliation can make
the empirical ﬁndings non-robust. More importantly, even if researchers overcome the problem
and get a robust test result, it is still unclear why or why not business groups have stronger political connections. By contrast, the hypotheses derived here do not suffer the ambiguity problem in
deﬁning group afﬁliations. Moreover, since they are derived from explicit considerations about the
incentives of ﬁnancially interlinked ﬁrms, a test of the hypotheses can provide suggestive answers
to the why-questions, too.
To investigate the causal relationship between a ﬁrm’s ownership structure and its political expenditure, a data set in panel structure and a proper empirical strategy are needed. In the following
section, I introduce the data used in the analysis and my empirical strategy.
11

A related, fundamental question is why countries have different ﬁnancial systems. Perotti and
von Thadden (2006) and Roe (2003) provide political-economic theories explaining how different
corporate governance systems come to exist.
14

1.3 Data and Empirical Strategy
1.3.1 Data
The source of the data is Nikkei: Annual Corporation Reports issued from 1992 to 2004 which
contains summary information about every corporation publicly traded from 1991 to 2003. In each
year, about 2,500 to 2,900 companies have been listed in the market, but non-negligible fraction
of them failed to register the requested information properly. Dropping out the observations with
missing variables, the number of corporations analyzed here is about 2,300 per year. The panel
structure is naturally unbalanced since every year some ﬁrms enter and others exit the ﬁnancial
market. The report provides the list of directors and managers and the name of the former employer of each board member. It also shows a rough picture of ownership structure of each ﬁrm. In
the report, owners of a ﬁrm are categorized into six groups: the government, non-ﬁnancial ﬁrms,
ﬁnancial institutions, securities companies, foreigners, and the rest.12 Each of the shares held
by non-ﬁnancial ﬁrms, ﬁnancial institutions, and the rest accounts for roughly 30% of all stockholdings (so altogether 90%), and the shares held by the government, securities companies, and
foreigners altogether account for 10%.
To explore the government-business relationship, students of Japanese political economy have
utilized data of retired high-ranked bureaucrats hired by private ﬁrms, called amakudari literally
meaning "descent from heaven", as a measure of political connection or a proxy for political expenditure of private companies.13 This measure is expected to serve the purposes well particularly
because the role of bureaucrats in Japanese politics is predominant. They "actually initiate and
draft virtually all important legislation." (Johnson, 1975) A remark made by Sahashi, former viceminister of the Ministry of International Trade and Industry (MITI), shows clearly the extent and
importance of their role: the Diet (Japanese parliament) is merely an "extension of the bureaucracy."
12

Here, the majority of "the rest" are known to be individual investors.
The information on private ﬁrms’ direct expenditure on political activities is not available to
researchers since lobbying activity is not legalized in Japan.
13

15

In more general term, amakudari practice refers to the reemployment system of elite bureaucrats. It shares some common features with the "revolving door" practice in the United States, but
differs in many ways.14 Most notable discrepancy is in the supply side. Because the bureaucratic
hierarchy is pyramidal, a bureaucrat is pressured to depart the public sector after she becomes a
section director, if she does not continue to rise in the administrative hierarchy. By the time that a
member of her cohort becomes a vice-minister, all but the most successful must leave the bureaucracy to give the vice-minister absolute seniority. So, the supply of ex-bureaucrats has been sizable
and stable. These retired bureaucrats start their second career in national or local politics, in private
and public corporations, or in other institutions in need of their consultation. Those who are hired
in private ﬁrms are known to provide channels of information and negotiations between the public
and the private sector. According to Colignon and Usui (2003), "[a]rranged by the ministry, not the
individual, it in effect provides private corporations with lobbyists" and "ministries with windows
to private corporations."15
Following the previous studies, I utilize the number of ex-bureaucrats in a ﬁrm’s board room as
the measure of its political connection. More speciﬁcally, exploiting the information of the former
employer of board members, I measure the degree of political connection in two ways. First,
narrowly deﬁned amakudari includes only the apparent ones, the listed board members whose
former employer is a ministry or a government agency. The problem of the narrow deﬁnition is
that it might underestimate true political connection, because some retired bureaucrats go to the
private sector in multiple steps: ﬁrst to a public company, then to a private one. This practice is
also quite popular, and is even given a name, yokosuberi meaning "sideslip." So, in addition to
the directors from ministries and government agencies, ones who were previously hired in public
14

See for instance Che (1995) for an economic analysis of the "revolving door" practice.
A few systematic investigations on the effect of hiring amakudari, focusing on the ﬁnancial industry, are available in the literature. Horiuchi and Shimizu (2001) show that those banks
accepting amakudari have reduced capital adequacy levels and increased non-performing loans.
Similarly, Van Rixtel and Hassink (2002) ﬁnd that amakudari appointments have a positive impact
on future proﬁtability and lending to risky industries.
15

16

Table 1.1: Summary Statistics
Narrowly deﬁned amakudari
Broadly deﬁned amakudari
Share held by non-ﬁnancial ﬁrms
Share held by ﬁnancial institutions
Number of directors
Number of employees
Sales

Mean
0.27401
0.38239
31.6676
29.8896
17.4372
2365.75
223116

Std. dev Minimum Maximum
0.74889
0
13
1.02232
0
13
18.6569
0
100
15.8656
0
100
7.32593
4
77
5286.23
6
97474
896680
10.33
2.13e+07

companies are counted as broadly deﬁned amakudari.16 Table 1.1 shows the summary statistics of
the variables that I use in the analysis. The fraction of the ﬁrms which have at least one (broadly
deﬁned) amakudari is 38.19%. This number is signiﬁcantly greater than that in the U.S. where less
than 10% of the ﬁrms have directors with political background.17 Given that the average number
of amakudari is 0.38239 as shown in the table, it means that most ﬁrms hire one or zero retired
bureaucrats. The share held by other corporations also make a sharp contrast with ownership
structure of the typical U.S. ﬁrm. At its highest, about 70% of the entire equity was once held by
other corporations, while most large ﬁrms are diffusely owned in the U.S.
As brieﬂy mentioned in the previous section, the structure and the role of board of directors
of Japanese companies differ quite signiﬁcantly from those of American ﬁrms. In Japan, the distinction between board members and executive directors is rather vague, and the vast majority of
directors are selected from among employees. An employee is elected to be a director in his or her
early ﬁfties, and stays on the board for six to seven years. Unless they resign, these new directors
are promoted four years later to a higher position, such as managing director, executive managing
director, vice-president or president. This is a reason why the number of directors appears much
larger than in the U.S. ﬁrm. See Miwa (1996) and Miwa and Ramseyer (2005) for more detailed
description.
It is well-known that the number of employees and sales, which are used as control variables in
16
17

These deﬁnitions are originally suggested by Rhyu (2008).
See Agrawal amd Knoeber (2001).

17

the regression analysis, follow considerably right-skewed distributions. Because in their original
form, they poorly "explain" the dependent variable, in the analysis I take log on these variables to
make them follow more bell-shaped distributions.

1.3.2 Empirical strategy
In this subsection, I explain the empirical methods used in the analysis, which exploit the exogenous changes in the ownership structure prompted by the economy-wide ﬁnancial reform. In the
late 1980s and the early 1990s, the Japanese economy experienced a series of signiﬁcant collapses
of asset price bubble, and entered into a long period of economic recession, often referred to as
"lost decade." Beginning in the 1990s, the Japanese government introduced short-term stimulation
policies and economy-wide structural reform for economic recovery. Substantive deregulation of
the ﬁnancial sector, known as the "ﬁnancial big bang" program was initiated in 1996, as a major
part of this effort. The aim of the deregulation was to transform a highly regulated and bankoriented ﬁnancial system into a transparent, market-based one. It eliminated regulations related to
banks’ payoff ratio, additional stock issuance, and a large part of the controls on foreign exchange
transactions. Largely fuelled by this institutional reform, the ownership structure of Japanese corporations has changed substantially during the sample period. Figure 1.1 shows the trends of the
shares held by non-ﬁnancial companies and those held by ﬁnancial institutions.
This observation suggests that taking the change of the ownership structure variables as "exogenous" to its political connection is a reasonable starting point. In this subsection, let us just
focus on a single ownership variable denoted by qit for expositional simplicity. The relationship
of our interest is captured by β1 in the following equation:
lit = α1i + β1 qit +Wit γ1 + et + νit

(1.4)

where lit is the political expenditure of ﬁrm i at period t, α1i is the unobserved individual heterogeneity (ﬁrm ﬁxed-effect), Wit is vector of exogenous variables, and et is time trend shared by
all ﬁrms. If the change in the ownership structure variable is largely exogenous to the degree of

18

Figure 1.1: Trends of economy-wide ﬁnancial ﬁrms’ shareholding and non-ﬁnancial ﬁrms’

political connection, i.e. E(νit |α1i , qit ,Wit , et ) is close to zero, a simple least-square estimation
of (4) would be enough to reveal how intercorporate stockholding affects the degree of political
connection. Thus, I ﬁrst regress the number of amakudari on the ownership structure variables and
the other controls including ﬁrm and time ﬁxed-effects.
Even if the changes of qit are largely driven by the exogenous factor, however, it might not
be completely exogenous for the following reason. If political beneﬁts are shared only within
ﬁnancially connected companies, and if the sharing makes the member ﬁrms better off, ﬁrms without ﬁnancial connections have incentives to join the club so long as the cost does not exceed the
beneﬁt. In other words, a strong political connection might attract investments from other corporations which seek for better and more diverse political connections, which means qit may also be
a function of lit :
qit = α2i + β2 lit + Xit γ2 + dt + fIt + εit

(1.5)

where α2i is ﬁrm ﬁxed-effect, Xit is vector of exogenous variables, dt is time trend, and fIt is
19

industry-speciﬁc effect of ﬁnancial reform. I denotes the industry to which ﬁrm i belongs.18 The
error terms εit is assumed to be mean zero and orthogonal to the other variables. If this simultaneity
problem is not properly controlled for, the method suggested above may generate signiﬁcantly
biased estimates unless β2 is close to zero.
Alternative strategy is to estimate (4) with an instrument for qit . If the error term νit is orthogonal to q jt for all j = i, the industry-year average of q jt can be used as an instrument for
qit . Unfortunately, that is not the case. Because ﬁnancially interlinked corporations strategically
decide how much to expend in building political connections, the set of unobserved variables must
include l jt where j is a ﬁrm that can impose political externality on ﬁrm i, i.e.
νit =

∑ δ j l jt + ηit
j∈Pi

where Pi is the set of ﬁrms whose political expenditure affects ﬁrm i’s performance, and ηit is the
unobservable that is orthogonal to l jt for all j. Thus when β2 is not zero, for any ﬁrm j in Pi , q jt is
not orthogonal to νit .
Because Pi is not observable to the researcher, one cannot build an instrument precisely based
on Pi . Instead, I propose an instrument based on a similar idea. Notice that the mean of νit can be
safely assumed to be zero thanks to α1i . It implies ﬁrm j can be included in Pi only if l jt changes
at least once. Therefore, any set of ﬁrms which hire a constant number of amakudari can be used
in constructing an instrument. In the main analysis, I instrument qit with qJ t = #1 ∑ j∈Ji q jt where
i
J
Ji is the set of ﬁrm j such that i) l jt = 0 for all t, and ii) ﬁrm j and i are in the same industry. Note
that the intersection of Ji and Pi is an empty set. Since for j ∈ Ji , q jt is orthogonal to νit , so is its
average qJ t .
i

Validity of the the instruments relies on the assumption that hiring amakudari is the dominant
form of political network building. If hiring ex-bureaucrats is an inferior way to build connections
to the government, this instrument would not help to mute the strategic interaction between li and
l j , and might not be valid. However, no other signiﬁcant political channel between the government
18

Corporations are categorized into 29 industries by the Nikkei Reports, and no ﬁrm has moved
from one industry to another.
20

Table 1.2: First Stage Regression on the Instruments

Industry-year average
of non-ﬁnancial ﬁrms’ shareholding
Industry-year average
of ﬁnancial ﬁrms’ shareholding
All other control variables
Time FE
Firm FE

Non-ﬁn. share
0.27679
(4.87)

Yes
Yes
Yes

Fin. share

0.39204
(8.20)
Yes
Yes
Yes

Note: The industry-year average variables are calculated only with ﬁrms that
have never hired retired bureaucrats. The numbers in parenthesis are
t-statistics based on the standard errors clustered by individual ﬁrms.

and private ﬁrms has been reported in the literature thus far. An alternative assumption that can
support the validity is that total political expenditure of a ﬁrm is proportional to the number of
amakudari in its board room. This is the case if the total political beneﬁt that the ﬁrm enjoys is the
multiplication of its spendings in various political channels (e.g. in Cobb-Douglas form).
The instruments should be highly correlated with the instrumented variables, which can be
satisﬁed if fIt varies considerably across time and industry. Table 1.2 shows that the industry-year
average qJ t is strongly correlated with individual ﬁrms’ ownership structure qit , even after all the
i

other variables are controlled for.

1.4 Empirical Results
1.4.1 Estimation with ﬁxed-effects
As a preliminary analysis, I ﬁrst estimate (4) without instrumenting the ownership variables. Table
1.3 shows the regression coefﬁcients and t-statistics (in parenthesis) in various speciﬁcations. For
panel A, I use the narrow deﬁnition of amakudari, and the broad one for panel B. The total number
of observations (N × T ) is 30,126, and the panel is unbalanced. t-statistics which are calculated
using the standard errors clustered by individual ﬁrms. Firm and time ﬁxed-effects are included in

21

Table 1.3: FE Estimates
A. Narrowly deﬁned amakudari
Shareholding of non-ﬁnancial ﬁrms

(1)
-0.00254
(-2.75)

Share held by ﬁnancial institutions
Number of directors
Log of number of employees
Log of sales

(2)

(3)
(4)
-0.00234 -0.00331
(-2.43)
(-2.96)
0.00147 0.00063 -0.00025
(1.49)
(0.62)
(-0.21)
0.01136
(4.71)
-0.05510
(-1.76)
0.00424
(0.41)

B. Broadly deﬁned amakudari
Shreholding of non-ﬁnancial ﬁrms

(1)
-0.00283
(-2.53)

Share held by ﬁnancial institutions

(2)

0.00278
(2.16)

Number of directors
Log of number of employees
Log of sales

(3)
-0.00219
(-1.95)
0.00199
(1.52)

(4)
-0.00342
(-2.56)
0.00121
(0.79)
0.01571
(5.05)
-0.11131
(-2.47)
-0.01989
(-1.42)

Note: The dependent variable is the number of ex-bureaucrats in the boardroom.
The numbers in parenthesis are t-statistics based on the standard errors clustered by
individual ﬁrms. Firm ﬁxed effect and time ﬁxed effect are included in all regressions.

all regressions.
Even in this simple regression, one can identify some interesting patterns. Notice ﬁrst that
the share held by non-ﬁnancial ﬁrms appears to be negative and statistically very signiﬁcant. The
coefﬁcient for ﬁnancial institutions’ shareholding appear positive in most speciﬁcations, but it is
statistically different from zero only in the second column of panel B. Not surprisingly, the total
number of directors appears to be positively correlated with the number of amakudari. The number
of employees, a measure of ﬁrm size, shows negative correlation with the level of political effort.
No statistically signiﬁcant pattern is found with regard to the sales of a ﬁrm, a measure of the ﬁrm’s
22

performance.
A few implications can be derived from the ﬁndings. First of all, they clearly show that ownership structure of a ﬁrm does matter in determining its political connections, which suggests political behavior of group afﬁliated ﬁrms is indeed different from that of stand-alone companies. But
the observed pattern says group afﬁliated ﬁrms would not necessarily build stronger connections
to the government. This ﬁnding echoes those of Colignon and Usui (2003) and Raj and Yamada
(2009) who report negative correlation between the number of amakudari and business group afﬁliation. However, the results presented here show more than the previously explored cross-sectional
distribution of amakudari. Because the cross-sectional correlation between the ownership structure variables and the number of ex-bureaucrats is already captured by individual ﬁrm ﬁxed-effect,
the remaining effect of the ownership structure variables must come from their changes over time
dimension. Also note that they are largely consistent with the theoretical predictions.

23

Table 1.4: IV Regression
A. Narrowly deﬁned amakudari
Share held by non-ﬁnancial ﬁrms

(1)
-0.05138
(-3.85)

(3)
-0.02799
(-2.87)
0.02467 0.02255
(3.83)
(3.25)

0.79721

0.64812

0.75566

(1)
-0.1228
(-5.45)

(2)

0.07601
(7.83)

(3)
-0.04773
(-3.40)
0.07237
(6.64)

0.76473

0.83603

Share held by ﬁnancial institutions

(2)

Number of directors
Log of number of employees
Log of sales
Fraction of variance due to α1i
B. Broadly deﬁned amakudari
Share held by non-ﬁnancial ﬁrms
Share held by ﬁnancial institutions
Number of directors
Log of number of employees
Log of sales
Fraction of variance due to α1i

0.86793

(4)
-0.03028
(-2.88)
0.01584
(1.96)
0.00912
(4.29)
-0.09117
(-2.15)
-0.03034
(-1.84)
0.73631
(4)
-0.04937
(-3.27)
0.07035
(5.47)
0.00669
(1.93)
-0.35242
(-4.97)
-0.13437
(-4.47)
0.81809

Note: The dependent variable is the number of ex-bureaucrats in boardroom. The
numbers in parenthesis are t-statistics based on the standard errors generated by
bootstrap. Firm and time ﬁxed effects are included in all regressions.

24

1.4.2 Estimation with the instruments
In this subsection, I present the main result obtained by instrumenting the ownership variables.
Table 1.4 reports the estimated coefﬁcients and t-statistics (in parenthesis). The t-statistics are
calculated based on standard errors generated by bootstrap method. As before, ﬁrm and year ﬁxedeffects are included in all regressions. First notice that the patterns found in table 1.3 are repeated
here. The share of equities held by non-ﬁnancial ﬁrms has a negative effect on the ﬁrm’s political
connection, whereas the share held by ﬁnancial ﬁrms has a positive, which conﬁrms the theory.
Also note that statistical signiﬁcance of ﬁnancial ﬁrms’ shareholding is dramatically improved in
this analysis. Both variables are statistically signiﬁcant in all speciﬁcations.
The effects of changes in ownership structure are economically signiﬁcant as well. A ten percent point increase of non-ﬁnancial ﬁrms’ shareholding decreases the number of narrowly deﬁned
amakudari by 0.3, and the number of broadly deﬁned ones by 0.5. The same change in ﬁnancial
ﬁrms’ shareholding induces ﬁrm to hire 0.2 additional narrowly deﬁned amakudari and 0.7 broadly
deﬁned one. Both the number of employees and sales are negatively correlated with the number of
ex-bureaucrats.

1.4.3 Delayed responses
So far I have implicitly assumed that the adjustment of the board composition is immediate. However, one might legitimately suspect that assumption. If it takes time for ﬁrms to adjust their board
composition in response to a change in their ownership structure, lagged variables must be introduced and appear statistically signiﬁcant. Hence, in this subsection I explore how the dependent
variable responds to a change in lagged explanatory variables. Table 1.5 shows the estimates of
regressions with contemporaneous and one-year lagged variables. As before, ﬁrm and year ﬁxedeffects are included in all regression, and the endogenous variables are instrumented. First, panel
A shows that an increase in non-ﬁnancial ﬁrms’ shareholding at period t − 1 reduces the number
of ex-bureaucrats at period t. The size of coefﬁcients of the lagged variable appears larger than
that of the contemporaneous variable in all speciﬁcations, and the statistical signiﬁcance of the
25

Table 1.5: Delayed Responses
A. Non-ﬁnancial ﬁrms’ share
Narrowly deﬁned amakudari
Contemporaneous
-0.05138
-0.01079
(-3.85)
(-0.54)
Lagged
-0.05567 -0.04601
(-3.58)
(-2.00)
B. Financial ﬁrms’ share
Narrowly deﬁned amakudari
Contemporaneous
0.02467
-0.02174
(3.83)
(-1.29)
Lagged
0.02891 0.04685
(4.04)
(2.70)

Broadly deﬁned amakudari
-0.1228
-0.04520
(-5.45)
(-1.42)
-0.12990 -0.08962
(-4.75)
(-2.28)
Broadly deﬁned amakudari
0.07601
-0.01283
(7.83)
(-0.50)
0.07882 0.08934
(7.31)
(3.36)

Note: The dependent variable is the number of amakudari. The numbers in parenthesis
are t-statistics based on the standard errors generated by bootstrap. Firm and time
ﬁxed effects are included in all regressions.

contemporaneous variable is lost when the lagged variable is included.
Similar pattern is found in panel B. An increase in ﬁnancial institutions’ shareholding at period
t − 1 tends to increase the number of ex-bureaucrats at period t. When the lagged variable is
included, the explanatory power of the contemporaneous variable completely disappears. These
ﬁndings are consistent with the expectation that for the composition of boards to fully be adjusted
would take time. It should not be missed that the directions of response of political connection
remain consistent with the theoretical predictions.

1.4.4 Subsample analysis
In this subsection, I check the robustness of the main result by focusing on two subsamples. It
has been pointed out by many researchers that there are two forces sustaining amakudari practice:
push and pull factors.19 Push factors refer to the incentives for the government to send former
bureaucrats to private ﬁrms mainly for regulatory purposes. On the other hand, pull factors are
the incentives for private ﬁrms to recruit retired bureaucrats for the purpose of network building.
19

See for example Aoki (1988) and Colignon and Usui (2003).
26

Table 1.6: Subsample Analysis
A. Highly Regulated Industries Excluded
Narrowly deﬁned amakudari
Broadly deﬁned amakudari
Non-ﬁnancial ﬁrms
-0.05665
-0.03121
-0.14196
-0.07167
(-4.20)
(-3.31)
(-5.70)
(-4.54)
Financial institutions
0.03212 0.02716
0.08645 0.07508
(4.36)
(3.54)
(7.59)
(5.76)
# of observation
27473
27473
B. Horizontal Keiretsu (Share held by non-ﬁnancial ﬁrms < 30%)
Narrowly deﬁned amakudari
Broadly deﬁned amakudari
Non-ﬁnancial ﬁrms
-0.05138
-0.10544
-0.47524
-0.14467
(-3.85)
(-2.00)
(-2.90)
(-1.94)
Financial institutions
0.02787 0.02771
0.09007 0.08984
(3.30)
(2.76)
(6.54)
(5.61)
# of observation
16119
16119
Note: The dependent variable is the number of amakudari. The numbers in parenthesis
are t-statistics based on the standard errors generated by bootstrap. Firm and time
ﬁxed effects are included in all regressions.

Since the focus of the analysis has been on the pull factors, the result must become clearer if we
exclude from the sample the industries where push factors are presumably very strong: Finance,
Air transportation, Communications, and Electronic power and gas industries. These industries are
regarded as psuedo-public sectors, and the government thoroughly monitors and regulates them.
Consequently, the average number of amakudari in these industries is almost twice as large as the
economy average. So in this subsample, the absolute value of the coefﬁcients are likely to appear
smaller than before. Panel A of Table 1.6 shows the same pattern observed in the Table 1.3, 1.4
and 1.5.
Next, I try to take into account heterogeneity of business groups. Aoki (1990) argues that there
are two types of "business groups", one of which is "ﬁnancial keiretsu," and the other is "capital
keiretsu". Financial keiretsu groups are characterized by loose cross-shareholdings and identiﬁed
by Presidents’ Clubs whose main function is information sharing. On the other hand, capital
keiretsu groups are characterized by strong vertical relationships where a dominant parent company
holds the majority of stock. Morck and Nakamura (2005) categorize modern business groups into
27

"horizontal and vertical keiretsu" each of which can be loosely matched to ﬁnancial and capital
keiretsu, respectively. According to Morck and Nakamura, the key feature of vertical keiretsu is
that a dominant non-ﬁnancial ﬁrm exercise controlling power over its subsidiaries, which differs
from the intercorporate ownership discussed in Section 2. So, once vertical keiretsu groups are
excluded from the sample, the main result is expected to be a bit clearer.
Panel B of Table 1.6 shows the results from the sample of ﬁrms less than 30% of which stocks
are owned by non-ﬁnancial corporations. As expected, the main result turns out to be robust in this
analysis as well. Other than the fact that, as expected the numbers are a bit inﬂated, the results are
very similar to the ones in Table 1.4.

1.5 Conclusion
Analyzing the reemployment system of retired bureaucrats in Japan, this paper shows that shareholding by non-ﬁnancial ﬁrms (friendly shareholders) has a negative impact on proﬁt-motivated
ﬁrms’ political participation, and that by ﬁnancial institutions (controlling shareholders) a positive
impact. It provides a framework to understand the puzzling ﬁnding reported by Colignon and Usui
(2003) and Raj and Yamada (2009), that business group afﬁliated corporations have weaker connections to the government. I argue that it can be explained by the fact that when ﬁnancially interlinked ﬁrms share political beneﬁts with each other, the incentives to free-ride on others’ political
effort might be signiﬁcantly high, so the member ﬁrms end up with weaker political connections.
By focusing on a simple decision problem faced by a corporation, whether to hire directors
with a bureaucratic background, this paper derives a broad implication for corporate governance
under high degree of intercorporate stock ownership. For many years, researchers have tried to
understand the operation of different ﬁnancial systems. For the Japanese case, the major role
that ﬁnancial institutions play in corporate governance has long been cited as the key mechanism
solving agency problem in the Japanese ﬁnancial market. The empirical speciﬁcation employed in
this paper is formulated to indirectly test these claims, and my ﬁndings support the "main bank"
narratives.
28

There are a few limitations in the analysis, which invite future works on this topic. I analyzed
ﬁrms’ political activities focusing on a speciﬁc political channel, namely retired bureaucrats hired
by private ﬁrms. If there are other less observable political channels, the present analysis might be
showing only an imperfect and biased picture of the entire political-economic system. I believe that
studies utilize other information and data would add valuable insights to the literature. For example,
to see if business group afﬁliated ﬁrms are actually better treated by the government, one may
want to check whether implemented policies have indeed been in favor of group afﬁliated ﬁrms
by directly analyzing the government expenditure. Further work taking into account heterogeneity
of business groups is also called for. If business groups are heterogenous, as argued by Aoki
(1990) and Morck and Nakamura (2005), one can expect that their political behavior might also
be heterogenous. Such heterogeneity is likely to turn out even greater if a researcher compares
business groups in different countries. Granovetter (2005) and Khanna and Yafeh (2007) have
suggested frames to categorize various types of business groups.
Lack of welfare implications is another limitation of this study. The theoretical model suggests that the ﬁrms cross-holding each other’s equity would make more efﬁcient use of each unit of
political connection. However, until the welfare impact of the reemployment system is fully understood, it would hardly be possible to properly evaluate the political consequences of intercorporate
stock ownership.

29

CHAPTER 2
ON THE OPTIMAL SOCIAL CONTRACT

2.1

Introduction

It has been well noted in the literature of development and democracy that the structure and quality
of governments in poor countries are markedly different from those of rich countries. Evidence
shows that on the contrary to the widespread belief, poor democracies do not perform better than
non-democracies with comparable income level, and in many dimension do worse. Also, the policy discrepancy between poor and rich democracies often appears greater than that between poor
democracies and non-democracies.1 If these patterns appear puzzling in the currently dominant
framework of political economy studies, it might be because the framework has been proposed and
developed in already developed countries such as the United States and western European countries, so lack perspectives of less developed societies. Thus, to better understand the relationship
between political system and economic development, an alternative conceptual framework may
have to be adopted, desirably one that contains the perspective of poor economies.
This paper proposes such a framework, being motivated by the following two facts: (i) now rich
countries in Europe and the other parts of the globe were poor countries some 200 or 300 years ago;
(ii) scholarly works directly and indirectly reflect the most prominent problems at the time when
they were written. Provided that problems that poor countries face are largely assimilar across
time and region, this observation suggests that the classics in political theory, which were written
200 or 300 years ago, would help us develop an alternative framework with developing countries’
perspective. As a preliminary attempt, this article proposes a modern economic interpretation
for one of the classics in political theory, On the Social Contract by Jean-Jacques Rousseau. In
particular, adopting Rousseau’s key ideas on Sovereign, government and subjects, I characterize
1

See for example Keefer (2007).

30

the optimal social contract, and examine its properties using modern contract theory.
Principal-agent framework has been applied to political problems in numerous studies, but in
most occasions, the relationship among political bodies has been assumed to be linear and unidirectional. A typical study in political science considers citizens as the principal and government
as the agent. In public ﬁnance, on the other hand, government mostly appears as the principal,
whereas citizens play the role of agents. In the social contract à la Rousseau, individuals appear
twice in completely different positions, once in the position of the ultimate principal (Sovereign)
and then in that of the ultimate agent (subjects).
"What then is government? An intermediate body set up between the subjects
and the Sovereign, to secure their mutual correspondence, charged with the execution of the laws and the maintenance of liberty, both civil and political." Jean-Jacque
Rousseau, On the Social Contract, Book III, Chapter 1.
Thus, the relationship between the citizens and the government looks like a "hierarchy", a chain
of principal-agent relationships, but differs from usual hierarchies in that the top and the bottom
of the chain are the same people. In this sense, the relationship is circular and bi-directional
in Rousseau’s framework. Following his predecessors, most notably Thomas Hobbes, Rousseau
considered the problem of self-regulation of people as the fundamental problem of political theory.
Because the current main stream political economy has exclusively focused on the one side of the
agency problems, namely government accountability, the construction of this circular framework
would add valuable insights to the literature of democracy and development.2
Explicit consideration of this bi-directional agency problem helps highlight a few important
aspects of citizen-government relationship, particularly in relation with economic and political
2

This framework can be alternatively motivated by James Madison who said "[i]f men were
angels, no government would be necessary. If angels were to govern men, neither external nor
internal controls on government would be necessary. In framing a government which is to be
administered by men over men, the great difﬁculty lies in this: you must ﬁrst enable the government
to control the governed; and in the next place oblige it to control itself." Alexander Hamilton, John
Jay and James Madison, The Federalist Papers, No. 51.

31

fundamentals. In this article, I show that richer societies tend to require their government to spend
more resources in monitoring the executive (i.e. checks-and-balances), that they can provide a disproportionately greater amount of public goods, and that when the members’ prospective income
level is greater, the polities are more resilient to negative economic shocks. Well-institutionalized
political system has similar positive effects on the performance of polities by reducing the cost of
disciplining misbehaving politicians. The bottom-line is that as the value generated by the social
contract becomes greater, the self-regulation problem (i.e. the agency problem on the citizens’
side) becomes less severe. And, the government can be constructed to deliver more desirable outcomes when the society is required to spend less resources in mitigating individual citizens’ agency
problem. I also show that the imperfect self-regulation can generate a poverty trap in which the citizens refuse to invest into public good project because the marginal utility of private consumption
is too high, which in turn results in low public good provision, low productivity and low level of
income.
The predictions of the model are largely consistent with the empirical patterns documented
in the literature. Keefer (2007) summarizes the ﬁndings as "poorer countries make signiﬁcantly
different choices along these policy dimensions than richer countries; theses are not easily explained by regime type. However, ... the policy choices of poor democracies differ little from
those of poor non-democracies!" In Section 4, I discuss the possibility that non-democracies might
be able to perform better than democracies by punishing misbehavior more severely. La Porta et
al. (1999), in similar vein, ﬁnd that richer countries show higher public sector efﬁciency, better
public good provision, larger government, and higher level of political freedom. Przeworski et al.
(2000) provide a comprehensive assessment on the dynamic relationship between political regime
and economic development. They do not ﬁnd any evidence that higher economic wealth causes
transition to democracy, but do ﬁnd that economic prosperity stabilizes the political regime.
A few previous studies have explicitly considered the bi-directional agency problem. In a
model in which a self-serving politician provides public goods and citizens can partially evade
taxes, Acemoglu (2005) show that both too weak and too strong states might impede economic

32

development. The underlying logic is very similar to that of hold-up problem that a seller, who
are requested to build buyer-speciﬁc product, has: if one party has too much (bargaining) power,
the counterpart has little incentive to invest in projects which enlarge the total surplus. Whereas he
analyzes the relationship between the "power" of government and its performance, I focus on how
economic development and free, fair and regular elections affect the performance of government.
Acemoglu et al. (2010) studies dynamic Mirrlees taxation under the assumption that political
elites have self-serving motivation. They allow the citizens to discipline politicians by the threat of
replacement. The agency problem on the citizens’ side in their paper is originated from asymmetric
information with regard to the type of agents. In my model, the source of the agency problem is
imperfect enforceability of the social contract. These studies do not endogenize the structure of
government. Acemoglu et al. (2011) is very closely related to the current study in terms of model
in the sense that they assume the government’s ability to collect taxes depends upon the size of the
government (bureaucracy) and that they allow the ruler(s) to decide the size.
To my best knowledge, none of the previous studies has investigated the optimal social contract
problem, namely how to optimally assign resources to minimize the overall costs generated by the
bi-directional agency problem, the problem that I explore in this work. Another notable feature
of this study is rich comparative statics with respect to empirically observable characteristics. The
above mentioned studies provide a very clear picture of a speciﬁc mechanism (for example how
inefﬁcient states can emerge and persist via the corruption between economic elites and bureaucrats), while show little attempts of comparative statics. In contrast, staying at a more abstract level,
I draw broad boundaries for feasible, incentive-compatible political associations, and examine how
the boundaries respond to changes of economic fundamentals and political institutionalization.
In a very different set up, Lagunoff (2001) consider the self-regulation problem of citizens. In
his model, the citizens have heterogenous preference, and democratically, so along the preference
of the median voter, establish the law which everybody has to obey. Because the law enforcement
is subject to errors, the median voter has an incentive to establish more tolerant legal system even
when she has no intrinsic preference for "civil liberty." Unlike the studies mentioned above and

33

the current study, Lagunoff does not explicitly consider the incentives of the government and the
bureaucrats. Lagunoff’s work is rather complementary to the other papers in the sense that it
provides a plausible reason for citizens to choose tolerance over most severe punishment.
This work is also related to the studies of self-enforcing political regimes. Przeworski (2005)
and Fearon (2011) investigate why and when fair and regular elections are preferred by political
elites, and how a democratic regime can be sustained as an equilibrium. Key idea is that since the
political elites cannot credibly make a commitment to give up their power, a democratic regime
can be sustained only when giving up power is more beneﬁcial for themselves.3 Because the
social contract is the contract that provides the basis of law and order, a body politic can be built
and persist only when "civil freedom" is preferred over "natural freedom" by its members. In this
sense, the social contract must be self-enforcing.
The rest of the paper is organized as follows. In the next section, I lay down the basic set up and
the equilibrium concept. Section 3 provides the main analysis of the model. In that, I characterize
the optimal social contract, and conduct a few comparative statics so as to examine the structure,
quality and stability of the social contract. I formally and informally discuss other related issues
in Section 4. Finally, I conclude in Section 5, brieﬂy discussing the usefulness of the framework
developed in this paper.

2.2 Model
2.2.1 Environment
Consider a society with a continuum of inﬁnitely lived individuals i ∈ [0, 1], and t = 1, 2, ...∞. In
each period, a person is endowed with one or zero unit of stochastic "taxable" income:


 1
with prob. yt
yit =
 0 with prob. 1 − y

t
3

Similar idea is examined by Acemoglu (2003).

34

The aggregate level of taxable income is stochastic and independently and identically distributed,
iid

i.e. yt = yit di ∼ F(yt ). The mean of yt is µ, and its distribution is common knowledge. Because
yit is observable only to person i, yt is not directly observable. The person with yit = 1 decides
whether to privately consume it or to submit it for public good production. Denote xit ∈ {0, 1} the
consumption decision: xit = 0 if invest to the public good production, and 1 if consume it privately.
The public good production succeeds with probability p. More precisely, from z unit of the
input (collected private good), γz unit of public good is produced with probability p, and zero unit
with probability 1 − p. Once the public good is produced, everybody in the polity equally enjoy
its beneﬁt. Assume the "tax revenue" zt = yt − xt = (yit − xit ) di is observable, but neither yt
nor xt is. The amount of input for public good production, denoted by zt ≤ zt , is not observable
to the citizens, but when the project succeeds, γzt is publicly observable. In other words, zt is
not observable when the public project is launched, but becomes a public knowledge after the
project generates its outcome. In general, zt is strictly smaller than zt due to the politician’s private
appropriation, legitimate or illegitimate. Such an appropriation must be legitimate in equilibrium,
and is denoted as wt = zt − zt , the wage for the politician.
An individual’s instantaneous utility in each period is u(xit + xit ) + γzt if the project succeeds,
and u(xit + xit ) otherwise, where xit is the baseline income (or the wealth of the citizen) and u(·)
is increasing and strictly concave. To focus on the bi-directional agency problem, I abstract from
heterogeneity of citizens, i.e. xit = xt for all i. For a moment, I assume xt = x, but relax this
assumption in later sections.
In most part of the paper, I use the public ﬁnance terminology such as taxable income, tax
revenue and so on. However, the model can also be understood in more Hobbesian manner. For
instance, yit can be interpreted as an opportunity to steal others’ belongings. In this interpretation,
the choice xit is whether to respect the law and order, and the public good is the secure property
rights.

35

2.2.2 Government
Following Rousseau, I introduce a government as an intermediary agent which helps solve the selfregulation problem more efﬁciently. Speciﬁcally speaking, because individuals cannot monitor
each other or punish misbehavior of others, they delegate the tasks to the government. Two roles
given to the government in this model is to monitor the individuals and to produce the public good.
The head of the executive can divert the collected resources for personal use if she is willing to take
the risk of being replaced by the citizens. However, because the true investment zt is not observable
and the success of the project is uncertain, the government’s misbehavior is hard to detect, which
aggravates the moral hazard problem. So if possible, the citizens would want to force politicians
to build a separate agency which monitors the use of the public fund.
I assume the government consists of two agencies: Internal Monitoring Agency (IMA) and
External Monitoring Agency (EMA). The role of EMA is, like a IRS in the U.S., to monitor
the citizens whether to pay taxes when they have to. Speciﬁcally, this government agency can
randomly monitor φ (nE ) fraction of the citizens where nE is the number of civil servants working
in EMA. And, IMA check the works of the executive. Since neither the organization of bureaucracy
nor the mechanism of checks-and-balances is the focus of this paper, I simply assume that if more
resources are invested in the monitoring activity, misbehavior is more likely to be detected and
punished. More precisely, when the politician diverts the public resource, such misbehavior is
detected, and the fund is impounded with probability ψ(nI ) where nI is the number of civil servants
working in IMA. It is assumed that even if she gets caught, zero public good is provided in that
period, since the opportunity to produce the public good is lost. Both φ (·) and ψ(·) are increasing
and concave. Let nP = 1−nI −nE be the number of citizens working in the private sector. I assume
that the citizens who work in the public sector lose the opportunity to get the additional income
(i.e. yit = 0 if i works in the public sector), but the baseline income x is preserved.
The above description of the government does not need to be understood literally. Because
for the citizens working in the public sector, yit = 0, the more are hired in the public sector, the
less resources the society has for the public good production. It implies that nE + nI is a real
36

cost from the society’s perspective, which would have been avoidable if the information were
complete. Therefore, the number of government employees can, and probably should, be more
broadly interpreted as all the resources spent so as to solve or mitigate the agency problems.

2.2.3 Timing of the events
In each period, the people as subjects (the agents in the principal-agent relationship) decides
whether to submit the taxable income when it is given, and as the Sovereign (the principal) whether
to replace the politician. The politician in the ofﬁce maximizes the discounted utility by choosing
how much to divert the public fund. Speciﬁcally, the events take place in the following order.
1. The politician in the ofﬁce announces the composition of the government (nI , nE ).
2. The public sector is ﬁlled in. The individuals hired in the public sector must give up the
opportunity obtain the taxable income.
3. EMA secretly and randomly chooses citizens who will be investigated.
4. yit is realized and observed only by i. Individual i chooses xit which is observable only to i,
unless investigated by EMA.
5. zt is collected and publicly observed. The politician secretly decide how much to divert the
public fund in addition to the given wage.
6. The outcome of the public project is realized.
7. Observing the outcome, the citizens collectively decide whether to replace the head of the
executive.
8. IMA and EMA inspect their monitoring objects, and pre-determined sanctions are imposed
if any misbehavior is detected.
9. All agents consume what is given to them.

37

Before proceeding further, a few comments may clarify the setup. First, in any equilibrium
where the citizens working in the private sector submit their taxable income for the public good
production (i.e. xit = 0 for all i and t), the individuals are indifferent between working in the private
and in the public sector. Thus, conditional on that the public investment takes place, the number
of the government employees and the composition of it are completely determined by the head of
the executive. Second, the assumption that EMA inspects randomly chosen citizens is innocuous
because the citizens are ex-ante homogenous. EMA can use a different inspection strategy by
making an artiﬁcial distinction among the citizens. Analyzing equilibria based on such a strategy
is beyond the scope of this paper.4 Third, note that since the number of the citizens is inﬁnite, there
exists the coordination problem in deciding whether to replace the politician. However, the citizens
who share the common goal of disciplining the politicians can easily sidestep the problem by
conditioning their action on the observable outcome. So in what follows, I focus on the equilibria
where the citizens jointly make a decision whether to replace the ofﬁce holder. Lastly, the role of
IMA is to investigate the person who was in charge after the project fails in order to reduce the
incentive to divert the public fund.

2.2.4 Equilibrium
Citizen i as an ultimate agent maximizes her discounted utility
∞

max ∑ Et β t−1 [u(x + xit ) + Zt − Rt − sit ]
xit

t=1

subject to
xit ≤ yit for all t
4

Also, one may raise a question on the way EMA is working. Instead of randomly monitoring every citizen, it can focus on the ones who do not pay taxes. One justiﬁcation could be
that the government agency might not able to adopt this apparently more efﬁcient strategy due to
administrative lags, i.e. EMA does not know who did not pay taxes when they monitor.
If we relax the assumption that EMA randomly samples over every citizen, we will observe
more efﬁcient workings of the government. But the main comparative results of the paper would
still hold.

38

where sit is the sanction imposed on the citizen, and


 γzt if the project succeeds
Zt =

 0
otherwise
Rt =





if replace the politician

R

 0


.

if keep the politician

The replacement cost R is zero if the polity is well institutionalized (i.e. free, fair and regular
elections are held), and λ > 0 if poorly institutionalized (e.g. the cost of revolution). Throughout
the paper, I maintain the assumption that the fundamental political institution (free, fair and regular
election), therefore R, is exogenously given. For the analysis, it is very useful to rescale the utility
for the citizens as follows.
∞

max ∑ Et β t−1 xit +
t=1

Zt − Rt − sit
.
u(x + 1) − u(x)

(2.1)

And deﬁne
γ
u(x + 1) − u(x)
R
R =
u(x + 1) − u(x)
γ =

Note that γ is the "marginal rate of substitution" between the private and the public goods, and
is increasing as the baseline income x grows larger. Similarly, the "effective" cost of revolution
λ / [u(x + 1) − u(x)] becomes larger as x increases.5
Before discussing the punishment schemes, let us consider anarchy, as a benchmark, where nobody invests to the public good production nor works in the public sector. In anarchy equilibrium,
Zt − Rt − sit = 0 for all i and t. From (1), the value of anarchy can be easily calculated as
VA =

µ
.
1−β

Throughout the paper, I make an assumption that the government cannot make any citizens’ utility
lower than VA . The basic idea is that for a social contract to be legitimate, the participation of its
5

Przeworski (2005) formally argues that the higher cost of social conﬂicts in richer society
can be a source of the observed relationship between the economic development and democracy.
39

members must be based on their free-will, and they must be able to freely opt out. If that is the case,
the citizens cannot be worse off than the utility that they get in the anarchy. This assumption might
appear restrictive and unrealistic at ﬁrst glance, particularly because in reality, no polity grants
opt-out option to its members. But in fact, it does not impose much restriction. So long as the
disutility from the sanction is ﬁnite (and independent of key parameters such as x, γ, λ and etc.),
all the qualitative results of this paper does not change.6 Section 4 provides a brief discussion of
the situation in which the legitimacy constraint is violated, i.e. the government is allowed to make
the citizens’ utility lower than that in anarchy. I also assume that it is costless for the government
to punish individual citizens.
On the other hand, the politician maximizes
∞

max ∑ Et β t−1 Ot [w (zt ) + (1 − ψ)Tt ]
Tt t=1

(2.2)

where w(zt ) = zt − z (zt ) is the wage for the politician as a function of the tax revenue zt ,
Tt ≤ zt − w(zt ) is the misuse of the public fund (T for theft), and


 1 if being in the ofﬁce at t
Ot =
 0

otherwise

As in most other repeated games, multiple social contracts can be sustained as subgame perfect

equilibrium. Especially, for any parameter values, there always exists the anarchy equilibrium
which is the unique equilibrium of the static game. This implies that (i) inefﬁcient political system
can persist over a long period of time and (ii) the structure and quality of governments are not
completely shaped by the underlying economic environment. The latter has been a signiﬁcant
obstacle when exploring the relationship between economic environment and political system.
In the following sections, I consider Markov Perfect Equilibrium in which the strategies are
mappings from the current state variables to actions. In particular, I focus on the optimal social
6

One may ask why the government cannot make the misbehaving citizens inﬁnitely miserable. Kantian principle or the arguments relying on human rights can be provided as an answer.
More practically, human imperfection can also be a reason for such restriction. That is, when law
enforcements are subject to errors, citizens collectively decide to tolerate such misbehavior. See
Lagunoff (2001) for a discussion of related issues.
40

contract which can be interpreted in two ways. First, it is an upper bound for efﬁcient public good
provision. Thus, the analysis provides a suggestive answer to the question why some countries
showing poor performance could not do better. Second, it can be understood as the "social demand" for a certain form of government. Just as Marshallian demand for a private good is derived
from consumers’ maximization problem, the "social demand" for a better political system would
be derived as the solution of the social contract problem. This latter interpretation provides a perspective for understanding the empirical relationship between economic development and political
system. Thus, in what follows I interpret the optimal social contract mainly as the social demand,
but with caution.
The optimal social contract is the strategies of the citizens and the politician(s) such that
• Given the state variables (yit , nI , nE , zt , Zt ), the citizens maximize (1), and decide to replace
the politician if
(i) the government composition (nI , nE ) does not maximize the citizens’ utility
(ii) the politician fails to produce the expected amount of the public good, i.e. Zt < γz(zt ).
• Given the state variables (Ot , zt ), the politician chooses Tt to maximize (2).
• The individuals working in the private sector ﬁnd it optimal to obey the law (xit = 0). The
government jobs are ﬁlled in as planned.
• The government maximally punish the citizens who are detected as keeping taxable income
for personal consumption, subject to the legitimacy constraint.
To focus on "working" social contracts, I restrict our attention to a limited range of parameter
values. Speciﬁcally, assume β p is large enough for the politician(s) not to ﬁnd that nI = 0 serves
her best interest. Also assume γ p > 1 so that the public good project is worthwhile to invest into.

41

2.3 Analysis
In this section, I ﬁrst characterize the optimal social contract, and then conduct comparative statics
with respect to the wealth level x and the political institution R. Let us ﬁrst consider the problem
of a citizen as an agent. When keeping the contract, the expected value of the social contract is:
∞

VC =

∑ Et β t−1
t=1

γ pE [z(zt )] − (1 − p)R
Zt − Rt
=
u(x + 1) − u(x)
1−β

(2.3)

For the social contract to be sustained as an equilibrium, the individuals must ﬁnd it optimal to
obey the rule. So, the incentive compatibility constraint for individual i with yit = 1 is
γ pE [z(zt )|yit = 1] − (1 − p)R + βVC
≥ 1 + γ pE [z(zt )|yit = 1] − (1 − p)R + β [φVA + (1 − φ )VC ]
where γ pE [z(zt )|yit = 1] is the expectation for the public good provision conditional on that the citizen is given the taxable income. This is in general different from γ pE [z(zt )] because E [yt |yit = 1]
is greater than the unconditional expectation E [yt ] = µ. The LHS of the inequality is the expected utility for the citizen when she pays the tax (i.e. xit = 0), while the RHS is when she,
instead, uses the endowment for personal pleasure (i.e. xit = 1). The public beneﬁt and cost
γ pE [z(zt )|yit = 1] − (1 − p)R are given to the citizen regardless of what she does. If she evades
the tax, she immediately enjoys the beneﬁt (1 on the RHS), and such misbehavior is detected with
probability φ , in which case the continuation value drops to VA , because the government would
maximally punish the citizen. The IC constraint can be simpliﬁed into
1
≤ VC −VA .
β φ (nE )

(2.4)

For the politician, the value of being in the ofﬁce is
VO = E [w(zt )] + β pVO =

E [w(zt )]
1−β p

where w(z) = z − z (z) is the wage schedule for the politician. Utilizing this notation, the incentive
compatibility constraint for the politician can be written as
w(zt ) + β pVO ≥ w(zt ) + (1 − ψ)z(zt ).
42

The RHS is the expected value when the politician appropriates the entire public fund for private
use, the value with "maximal theft." Given zt publicly observed, accordingly, the politician is
supposed to invest z(zt ) to the public project, otherwise she would be removed from the ofﬁce at
the end of the period. Put it differently, even if she diverts the public fund just a little bit, she is
surely replaced by the citizens. Thus, it is optimal for her to divert the entire fund (i.e. Tt = z(zt )),
once she decides to use the fund for private beneﬁt. However, the politician can enjoy the fruits of
the "theft" only when IMA fails to detect it, thus with probability (1 − ψ). The IC constraint for
the politician can be simpliﬁed into
z(zt ) ≤ min zt ,

β pVO
1 − ψ(nI )

= min zt ,

β pE [zt − z (zt )]
[1 − ψ(nI )] (1 − β p)

(2.5)

There are many social contracts that satisfy the two incentive compatibility constraints. Here, I
only consider the optimal social contract which maximizes the expected utility for the participants
of the contract. This contract must be "self-enforcing": the participants must ﬁnd it optimal to keep
the contract. Thus, the optimal social contract solves
max VC

nI ,nE ,z

subject to the IC constraints (4) and (5).
To ﬁnd the solution of the program, ﬁrst note that constraint (4) must be binding. Otherwise,
one can slightly reduce nE , keeping (4) held, and increase the number of citizens working in the
private sector. This modiﬁcation increases the tax revenue, and also the expected amount of public
good provided. Similarly, (5) must be binding, too, otherwise by decreasing w, one can increase
the public good provision z at least for some zt .
From (3), it is apparent that maximizing VC is equivalent with maximizing E [z(zt )]. From (3)
and (4), one can derive the following formula:
E [z(zt )] =

1
γp

1−β
+ (1 − p)R + µ
βφ

43

(2.6)

Using (5), (6) and the fact that in equilibrium zt = nP yt , one can rewrite E [z(zt )] as
E [z(zt )] = E [min {nP yt , Γ}]
=
where
Γ=

Γ/nP
0

nP yt dF(yt ) + 1 − F

Γ
nP

Γ

(2.7)

β p nP µ − γ1p 1−β + (1 − p)R + µ
βφ
(1 − ψ) (1 − β p)

Note that (7) is strictly increasing in Γ, i.e. VC is a monotone transformation of Γ. Therefore, an
alternative representation of the optimization problem is:
max Γ

nE ,nI

subject to
1
γp

1−β
+ (1 − p)R + µ
βφ

=

Γ/nP
0

nP yt dF(yt ) + 1 − F

Γ
nP

Γ.

The equality constraint is obtained by combining (6), the IC for the citizens and (7), the IC for
the politician. Under the assumption that the program has a proper interior solution, the following
proposition characterizes the optimal social contract.
Proposition 3 Suppose there exists an interior solution for the social contract problem. Then, the
composition of the government (nE , nI ) is characterized by
1−ψ
1 1−β
+ (1 − p)R + µ
′ µ = nP µ − γ p
ψ
βφ
ϒ/nP
ϒ
1 1−β
+ (1 − p)R + µ =
nP yt dF + 1 − F
ϒ
γp βφ
nP
0

(2.8)
(2.9)

βp
where ϒ = 1−β p µ′ . And, the optimal compensation scheme for the politician is
ψ


1−β
 β p nP µ − 1
+ (1 − p)R + µ 
γ p β φ (nE )
z(zt ) = min zt ,


[1 − ψ(nI )] (1 − β p)

In the following subsection, I investigate how the structure of the government responds to

changes in the society’s wealth and the degree of political institutionalization. Modifying the basic
model slightly, in the next subsection I analyze the quality and the stability of the social contract.
44

2.3.1 Structure of the government
To understand the results more intuitively, let us introduce a measure of checks-and-balances.
CB = nI /nE
Greater nI /nE ratio means the polity spends greater amount of resources in disciplining the person(s) in power than in monitoring the citizens. When CB is low, too much resources have to be
given to the political elites in the form of "wages", otherwise they would divert the entire public
fund. Thus, a government with higher CB can be thought of as a better device for self-government.
High CB could also be counted as a desirable feature for political systems in that the government
is more accountable, and more civil rights and liberty are given to its members.7
The following two propositions state how the structure of the government responds to changes
in the society’s wealth and political institutionalization. The basic message is that the optimal
social contract yields more "democratic" government structure when the total value generated by
the contract is larger. The total value of the social contract is larger when the society’s baseline
income is greater, and when the cost of replacing the incompetent politician is smaller.
Proposition 4 For a polity with well-institutionalized system (i.e. R = 0) the social demand for
higher CB becomes stronger as the society’s wealth increases.
Proof. Assuming R = 0, differentiating (8) and (9) with respect to γ gives
−

1−β φ′
(1 − ψ)ψ ′′ µ ∂ nI
+ µ−
∂γ
γ pβ φ 2
(ψ ′ )2
−

∂ nE
∂γ

(1 − F)ψ ′′ β pµ ∂ nI 1 − β φ ′ ∂ nE
+
γ pβ φ 2 ∂ γ
(ψ ′ )2 1 − β p ∂ γ

=

1
γ2 p

1
= − 2
γ p

1−β
+µ
βφ
1−β
+µ
βφ

Note ﬁrst that the coefﬁcients of ∂ nI /∂ γ in the both equations are positive, and the RHS of the
ﬁrst equation is positive while that of the second one is negative. Adding up these two equations,
7

Although it is not incorporated in this model, one can think of the possibility that higher
nE results in more suppressive states, as frequently observed in authoritarian political systems. I
discuss this issue in the next section.

45

we have
−

ψ ′′ µ
(ψ ′ )2

1−ψ +

(1 − F)β p ∂ nI
∂n
+ µ E = 0,
1−β p
∂γ
∂γ

which shows that nI and nE move to the opposite directions in response to a change of γ. In the
rest of the proof, I show that ∂ nI /∂ γ > 0 and ∂ nE /∂ γ < 0. First, notice that the coefﬁcicent of
∂ nE /∂ γ in the ﬁrst equation is
µ−

βp
∂Γ
1−β φ′
=−
γ pβ φ 2
(1 − ψ) (1 − β p) ∂ nE

If ∂ Γ/∂ nE is negative, that means too much resources are spent in monitoring the citizens. So
in the social optimum, ∂ Γ/∂ nE must be positive. That means the coefﬁcient of ∂ nE /∂ γ must
be negative. If ∂ nI /∂ γ < 0 and ∂ nE /∂ γ > 0, the LHS of the ﬁrst equation cannot be positive. Thus, ∂ nI /∂ γ > 0 and ∂ nE /∂ γ < 0, which in turn implies ∂ (nI /nE )/∂ γ > 0. Because
γ = γ/ [u(x + 1) − u(x)] increases as x increases, ∂ (nI /nE )/∂ x > 0, which completes the proof.
In the optimal social contract, the resources spent in monitoring the citizens nE decreases as
the wealth x increases for the following reason. As the level of baseline income is enhanced, the
marginal utility of private consumption becomes smaller, which means the agency problem on the
individuals’ side becomes less signiﬁcant, thus less amount of resources are required to incentivize
the citizens. By reducing the resources spent in the private sector monitoring, the utility given to
the citizens can be improved either by generating more taxable endowment (i.e. increasing nP ) or
by monitoring the politician more tightly (i.e. increasing nI ). At the optimum, it turns out that
the citizens want to discipline the politician and to reduce the share given to her by enhancing the
probability of detecting potential misuse of the public fund.
Note, however, that this proposition does not mean that poor countries prefer a government
structure with lower CB. Let us for a moment take ψ as an exogenously given parameter while
keeping nE as endogenous. It is easy to show that the optimal level of nE decreases as ψ increases. In other words, as the government becomes more accountable, the agency problem on the
citizens’ side is also mitigated. As mentioned at the beginning of this section, VC is a monotone

46

transformation of E [z(zt )] whereas
E [z(zt )] =

1−β
1
+ (1 − p)R + µ .
γ p β φ (nE )

This clearly shows that lower nE indicates greater utility for the citizens. So, one can claim that
in general, the citizens prefer more accountable government regardless of the level of their wealth.
What the proposition really means is that richer societies are willing to spend greater amount of
time, resources and efforts in monitoring and disciplining the political elites.
Whereas the above proposition describes the relationship between the economic fundamental
and the structure of government, the following shows how the fundamental institution of democracy (free, fair and regular elections) shapes other derivative political institutions.
Proposition 5 Institutionalization of free, fair and regular elections promotes to build a government with higher CB.
Proof. This proposition can be proved by conducting comparative statics with respect to R. The
proof is almost identical with the above proof, thus omitted.
As the cost of replacing the politician goes down, the expected utility from keeping the contract
becomes larger. It means IC constraint for the citizens is loosened, so the cost of monitoring the
private sector nE can be reduced. The citizens can and will reassign the resources saved in EMA
into IMA, i.e. nI goes up. Since the social welfare can be improved by reducing nE and increasing
nI , the citizens are likely to demand more accountably structured government.
The discussion thus far clearly shows that the agency problems on the two sides aggravate each
other. When a country is economically less developed, the citizens are more seriously tempted to
violate the law and order, if by sacriﬁcing the public good they can become privately a little bit
better off. To incentivize the citizens, more resources should be spent in keeping the law and order,
and consequently relatively less resources are spent in disciplining the political elites. This implies
the government provides public goods in a less efﬁcient manner, and less efﬁcient government
reduces the citizens’ willingness to keep the law and order. In this way, the agency problems

47

becomes disproportionately more severe in less developed countries, and thus their governments
are likely to perform poorly.

2.3.2 State capacity and stability
So far I have assumed that the amount of taxable endowment yit is given as 1 or nil. The basic
model can be easily generalized into the case where yit is h with probability yt /h, and 0 with the
complementary probability. As before, deﬁne the marginal rate of substitution as
γ=

γh
u(x + h) − u(x)

and the effective cost of replacement as
R=

Rh
.
u(x + h) − u(x)

Then, the optimal social contract is chracterized by (8) and (9) with a few modiﬁcations: γ and R
are substituted by γ and R, respectively, and µ is scaled up (or down) into hµ. It is apparent from
these deﬁnitions and the proofs of the propositions that Proposition 2 and 3 stay true insofar as the
optimal social contract problem has the interior solution.
To investigate the relationship between the size of government spending and the economic
development, suppose that at period t = 0, the citizens jointly decide whether to construct a polity
which produces h unit of taxable endowment per period at the initial setup cost of H(h) where H
is an increasing function. The individuals construct the polity if and only if
VC − H > VA .
If the initial construction cost H is high, then the polity cannot exist unless the value generated
by the social constract (VC −VA ) is large enough. Recall that according to the above propositions,
when the society is economically developed, and the fundamental political institution is well established, the value of the social contract (VC − VA ) is large. In such a case, it can successfully
provide a greater amount of public goods, i.e. it can sustain higher h. For simplicity, let us assume
for a moment that the society has a well institutionalized political system, i.e. R = 0.
48

Proposition 6 If a society with wealth level x1 is able to sustain a certain level of government
spending h, so is a society with greater wealth x2 (> x1 ).
Proof. By envelope theorem, the expected value generated by the social contract
VC =

γ pE [z(zt )]
1−β

is increasing in γ. The proposition is immediate from the observation that the marginal rate of
substitution γ is increasing in x.
This proposition is closely related to an empirical regularity often referred to as Wagner’s law.
The law says the development of an industrial economy will be accompanied by an increased share
of public expenditure in gross national product. A potential source of the observed pattern is the
agency costs of the citizens and the political elites. As an economy grows, the both agency costs
decreases as shown in Proposition 2, which in turn allows the economy to sustain a disproportionately large public sector.
The same logic sheds light on the stability of political regimes, too. Now, let us assume the
baseline income xt is subject to a negative shock. Speciﬁcally, suppose xt drops by ∆ at period s,
is recovered to x at the next period and on, i.e. xs = x − ∆, and xs+1 = xs+2 = ... = x. The polity is
dismantled if the citizens refuse to comply with the rule when the shock hits the economy. When
the private consumption falls down to x − ∆, the citizens would be more likely to keep the social
contract when the continuation value of the social contract is greater.
Proposition 7 The polity becomes more stable as the prospective wealth level x is greater, or as
the political system gets more institutionalized.
Proof. The IC constraint for the citizens with yis = 1 is
γ pE [z(zs )|yis = 1] − (1 − p)R + βVC
≥

u(x − ∆ + h) − u(x − ∆)
+ γ pE [z(zs )|yis = 1] − (1 − p)R + β [φVA + (1 − φ )VC ]
u(x + h) − u(x)

49

which can be simpliﬁed into
1 u(x − ∆ + h) − u(x − ∆)
≤ VC −VA .
βφ
u(x + h) − u(x)
As shown in the proof of Proposition 4, VC becomes larger as x increases whereas by the assumption, where as the LHS of the inequality (weakly) decreases because u(·) is concave. It means the
IC constraint is easier to be satisﬁed when the society’s wealth is greater.
Przeworski (2005) and others have argued that economic development enhances the stability of
political regimes. The previously suggested argument focuses on the incentive of political losers
to conform the democratic rules. Because revolution destroys economic assets and institutions,
it becomes more costly as the society is more economically developed. The above proposition
provides a different perspective for the same phenomenon. In this model, a polity becomes more
resilient as the economic wealth in the future is expected to be larger, because the citizens’ incentive
to behave more selﬁshly in hard times is weakened.
The results derived so far are summarized as the following. When the expected value generated
by the optimal social contract is larger, (i) the government tends to be more accountable, (ii) the
society can provide a greater amount of public goods, and (iii) the polity becomes more resilient
to negative economic shocks. And, the value of the social contract is larger when the society’s
baseline income is greater, and when the cost of replacing the misbehaving politician is smaller.
Although the basic model highlights some of the important underlying mechanisms, there still exist
notable gaps between the basic model and the reality. In the next section, I try to narrow the gap
by addressing a few issues including endogenous change of the wealth and direct monitoring by
the citizens.

2.4 Other Issues
2.4.1 Illegitimate government
The analysis thus far shows that the legitimacy constraint, namely the government cannot punish
more harshly than the Nature does in anarchy, increases the burden of self-government by increas50

ing the agency costs. From this, it is rather apparent that if the government is allowed to punish its
citizens more severely, the overall efﬁciency can be improved. Does this mean that governments
that watch over citizens and suppress their liberty would perform better? If the government works
are not subject to errors, it seems that it would perform better at least in theory. However, it is not
necessarily the case if the government does not try to maximize the citizens’ welfare. If the government maximizes a certain objective function other than VC , for example the beneﬁt of political
elites, it is very probable that the citizens are worse off. Even though it is likely that the government equips itself with some self-regulating device, i.e. nI > 0, the level of it would be less than
the socially optimal level.8 The severe agency problem on the government side would exacerbate
that on the citizens side, and in order to incentivize the citizens, the society should invest a great
amount of resources in monitoring the citizens (i.e. high nE ).
In reality, more often than not, oppressive governments tend to serve the interests of political
and economic elites over those of the public. Such government are imposed upon rather than
constructed by the public. This observation may suggest that if a polity is constructed by the
public, they would not want to have oppressive state apparatus. If so, the legitimacy constraint
adopted in this paper would be not only of theoretical interest, but also empirically relevant in
working democracies.
Extending model, one can think of the possibility that higher nE results in more suppressive
states, as frequently observed in authoritarian political systems. In such a situation, the results of
this paper would be strengthened because the government with more suppressive state apparatus
would be able to increase the cost of revolution, λ . As shown above, when the cost of replacing
politicians becomes larger, the total value generated by the social contract becomes smaller, which
in turn exacerbate the agency problem of the citizens.
8

McGuire and Olson (1996) convincingly argues that even if rulers have only self-serving
incentives, it is better for them to abstain from fully extracting surplus, otherwise the ruled will
refute to produce surplus at all.

51

2.4.2 Poverty trap
For clear exposition, I have assumed that the wealth level is exogenously given. However, there
is ample evidence that economic prosperity heavily relies on the efﬁciency of public sector and
the quality of public infrastructure. In this subsection, I address this issue by assuming that the
productivity of the economy depends upon the level of public good provided to the citizens.9
For analytical simplicity, let us assume that the politician is perfectly monitored and disciplined
(ψ = 1), every citizen is given one unit of taxable endowment in each period (yit = 1 for all i
and t), the probability of success of the public project is one (p = 1), and the probability that a
misbehaving citizen gets caught and penalized φ is exogenously given and time-invariant. For
detected tax evasion, a punishment is given once, and the size of sanction is ﬁnite (S < ∞). The
production technology of private goods is given as
α
yit = zt kit

and the productivity zt is determined by the public investment in the previous period:
zt+1 =

(yit − xit ) di + z = 1 −

xit di + z

where kit is capital stock at period t, α is smaller than 1, and z is the minimum productivity.
Citizen i solves the following optimization problem:
V (kit , zt ) = max {u(xit + xit ) − φ Sxit + βV (kit+1 , zt+1 )}
x ,k
it it+1

subject to
xit + kit+1 ≤ yit + (1 − δ )kit
xit ∈ {0, 1}
where δ is the depreciation rate of capital. I assume that the initial productivity z1 is z, and the
citizens are ex-ante identical, i.e. ki1 = k1 for all i. It is obvious that if the expected cost of
9

Many previous studies including Acemoglu (2005) have adopted a similar assumption. To
my best knowledge, the poverty trap generated by the agency problems has not been analyzed
before.
52

disobeying the rule φ S is large enough, everybody will choose to invest to the public project,
regardless of the level of capital. To focus on non-trivial case, let us assume φ S is not too large.
More speciﬁcally, I assume the following.
α
α
lim u(zk1 + 1) − u(zk1 ) > φ S

k1 →0

Under the assumptions made above, the following proposition is immediate.
Proposition 8 For sufﬁciently small z, there exists k such that for k1 ≤ k, nobody invests in the
public project in any time, i.e. xit = 1 for all i and t.
Proof. The value for citizen i when xit = 0 is
u zktα + (1 − δ )kt − kt+1 (kt ) + βV kt+1 , zt+1
and that when xit = 1 is
u zktα + (1 − δ )kt − kt+1 (kt ) + 1 − φ S + βV kt+1 , zt+1 .
where kt+1 (·) and kt+1 (·) are the solutions for the optimization problem, given xit . As kt → 0, so
do kt+1 and kt+1 . Thus, as kt → 0, V kt+1 , zt+1 −V kt+1 , zt+1

→ 0. Therefore, according

to the assumption, refusing public investment (xit = 1) is preferred for a very small kt . Next, let us
denote the steady state level of capital by k∗ when xit = 1 for all i:
k∗ = arg max {u [zk∗ + (1 − δ )k∗ − k + 1] − φ S + βV (k, z)} .
k

It is easy to show that as z goes to zero, k∗ also converges to zero. It implies for a sufﬁciently small
z, there exists a steady state level of capital k∗ with which the citizens prefer not to invest in the
public project.
This proposition shows the intuition developed in the previous section can be applied to a more
dynamic situation. When the level of income (wealth) is endogenously determined, the following
negative-feedback mechanism might hinder a country to be economically and socially developed.
If a society is given a low level of initial capital, its production capacity is low, so is its income
53

(wealth) level. As we have seen in the previous section, the agency problem is particularly serious
when the level of income is low. Consequently, the government’s ability to provide public goods
such as social infrastructure, public education and health care would be seriously limited. This
bounds the economy’s expected productivity in the next period. Since the expected return from
private investment is low, they invest little, which results in a low level of capital stock in the next
period. Thus again, low level of income would be given to the citizens.
It is worth mentioning that the result would hold without the simplifying assumptions. If the
agency problems are endogenized as in the analysis of the previous section, the agency cost of a
low income country is greater than that of a high income country. Thus, endogenizing the structure
of the government would do nothing but reinforce the feedback mechanism.

2.4.3 Civic virtue
Numerous studies have argued that "civic virtue" or "social capital" plays a important role in workings of democracy. There are many ways for civic virtue to enhance the quality of self-government.
In this subsection, I demonstrate one way that civic virtue can help reduce the agency cost. Suppose there are informal social networks, so any citizens can be observed by at least one fellow
citizen with probability q. If it is a social norm that misbehavior is reported to the government,
the probability of detection is now q + φ − qφ which is greater than φ . Accordingly, the agency
cost incurred by the citizens can be reduced (lower nE ). As the previous analysis has shown, this
allows the society to spend more resources into disciplining the political elites and to construct a
more democratic government (increase nI ).
Note that the social norm could be ignoring what she observes, instead of reporting it. In such
case, the probability of detection remains as φ even when q is positive. Thus, civic virtue should
be sustained as an equilibrium of repeated interactions among the citizens, and can be broken or
reinforced by historical events.10 The ignoring equilibrium would be more likely to emerge when
10

See Kandori (1995) for a game-theoretic model of social norm, and Putnam et al. (2001) for
a study of social capital and civic virtue.

54

reporting to the government incurs a cost to the reporter.

2.5 Conclusion
This paper, using a simple theoretical framework, explores several fundamental issues of political
economy including how to allocate resources to achieve efﬁcient self-government, why the governments of poor economies work so poorly, and how the agency problems in different parts of a
society interact each other. A novel feature of the current work is that it adopts the frameworks
of the classical writers, particularly that of Jean-Jacques Rousseau. The perspective of the social
contract encourages us to consider bi-directional agency problem of citizens and the government
as a non-separable one. While the agency problem on each side is thoroughly examined in separate
literatures, the joint analysis is rarely attempted.
The analysis of this paper shows that the two agency problems aggravate each other: poor
quality of government lowers the citizens’ willingness to follow the law and order, and at the same
time when more resources are spent in monitoring and disciplining the citizens, less would be
spent in improving the government accountability. It is also shown that the severity of the problem
depends upon the level of economic development and that of political institutionalization of the
society. When the economic performance of the society depends on the amount of public goods
invested in the previous period, the agency problem might generate a poverty trap.
Although Rousseau’s works have been constantly referred to in normative discussions, his
legacy has seldom appeared in positive analyses. So, one may ask why his legacy has been only
partly appreciated. I do not intend to provide here a comprehensive genealogy of the ideas, but
just would like to point out the results of this article do suggest a plausible answer to the whyquestion. It says once a society is sufﬁciently developed in economic and political spheres, the
agency problem on the citizens’ side does not outstand any more, in which case, researchers can
safely ignore it and focus on the problem on the other side, namely the problem of government
accountability. The analysis suggests, however, this strategy might not be valid in analyzing the
political economy of developing countries. When considering developing countries, one should
55

explicitly take the problem of self-regulation of the citizens as an integrated part of the entire
political economic system.
Of course, more scholarly efforts are required for better understanding of the self-regulation
issue. In particular, two important issues which are abstracted from this paper invite further investigations. First, wealth inequality generates a different set of political problems which might
interact with the agency problem which has been the focus of this article. Second, the corruption
of bureaucracy is also very notable feature in developing countries, and it is likely to exacerbate
the agency problem on the other sides of the society.

56

CHAPTER 3
VOTER ATTENTION AND POLITICAL POLARIZATION

3.1 Introduction
The nature of the political polarization in the United States has been a subject of intense debates
in a recent few years. Most scholars agree on the fact that the polarization at the elite level has
increased over the last three decades (see e.g. Poole and Rosenthal 1997, Stonecash et al. 2003,
and Theriault 2008). In contrast, there is much less agreement on the ideological landscape at the
mass level. A group of scholars argues that there has been substantial changes in voters’ political
preferences (Abramowitz 2011, and Hetherington 2001), while others counter-argue that those
changes are exaggerated or just a myth (Fiorina et al., 2006).1 The former tends to support the idea
that the polarization is a primarily bottom-up phenomenon, while from the latter’s point of view,
the polarization is best characterized as top-down or elite-driven. Whereas the evidence that has
been submitted is not conclusive, the existing theories of political polarization seem to support the
bottom-up side: in theory, ever since Downs (1957), political parties compete to win as many votes
as possible, thus their policy positions are tightly bound to the fundamental preference of voters.
In this theoretical framework, the top-down side scholars are requested to explain why politicians
who seek for winning elections ever want to change their platforms when there is no major change
in the fundamentals.
This paper provides a spatial voting model in which elite-driven political polarization can
emerge as an equilibrium. Key assumptions are: (i) voters are responsive to changes in policy
positions of parties only if they pay attention to politics; (ii) political elites can disinterest away
some voters by making the voters believe that implemented policies will be less preferable to them.
Under these assumptions, political parties do not have incentive to converge to median voters if
1

See also Abramowitz and Saunders (2008) and Fiorina and Abrams (2008) for the follow-

ups.
57

the median voters believe that the parties’ platforms will diverge away from the center, so do not
even pay attention to politics. By the same logic, the traditional median-voter equilibrium can also
emerge: when voters expect the parties converge to the center, and consequently the median voters
remain attentive and responsive, the political parties are forced to compete for the votes of the
median voters. In this manner, the policy platforms can be polarized even when there is no change
in the fundamentals.2
Although the model provides a logical description of elite-driven polarization, it does not take
a side in the above mentioned debate. Examining the conditions for the median-voter equilibrium
to exist, I show that the centrifugal force grows larger either when economic inequality grows or
when media tend to mobilize partisans more than they do centrists. If the economic inequality and
media slant become considerably severe as many scholars and commentators argue that be the case
in the U.S., platform polarization becomes the unique equilibrium.3 Thus, this paper position is
that it is not theoretically decidable whether the nature of the polarization at hand is top-down (i.e.
without any change in the parameters) or bottom-up (i.e. caused by a change in the fundamentals).
It is widely recognized that in a large electorate rational voters have little incentive to gather
information or to be attentive when it is costly. Downs (1957) pointed out this problem together
with the problem of voters’ turnout decision, namely rational voters who only consider the outcome
of the election do not have incentives to turn out to the voting booth when the number of voters is
sufﬁciently large. The latter problem has been thouroughly explored by economists and political
scientists, but the former has been much less popular in academic discussion.4 The present paper
considers the two problems together: for the turnout problem, on one hand, adopting a widely used
2

The nature of the polarization demonstrated in this paper is better-termed as "belief-driven".
However, so long as such beliefs are formed by the political elites whose symbolic actions are
visible to many citizens, we can also call it as "elite-driven", too.
3 Bartels (2007) and McCarty et al. (2008) argue that the current polarization has been mainly
driven by the change in the structure of the economy, whereas Campante and Hojman (2010) and
Prior (2007) emphasize the role of media in polarization of opinions
4 In the literature, voters’ incentive to acquire information has been investigated exclusively
in committee setting. See Gerling et al. (2005) for a survey of the incentive for information
acquisition in committee setting. See Merlo (2006) for a survey on voters’ turnout decision.

58

assumption which is that voting as a civic duty generates intrinsic utility by itself. For the attention
problem, on the other, I assume that voters rationally evaluate the probability of their turnout,
which depends on the expected utility of voting, which in turn is affected by the policy positions
chosen by the political parties. Because attention is a limited cognitive resource, the voters decide
to pay no attention to politics if they expect they will not show up in the voting booth after all.
This assumption on the link of the two closely related problems helps highlight the role of voters’
belief and politicians’ symbolic actions in elections.
The model is a descendent of the spatial competition models of Downs (1957) and Hotelling
(1929). Given that competing political parties in reality do not always take the same policy position, the famous median-voter result has been challenged by many theorists.5 A recent paper
by Glaeser et al. (2005) suggests a simple and realistic model of political polarization.6 In their
model, voters with different party afﬁliation have accesses to different information sources. Due
to this discrepancy in observability, the political parties are better able to mobilize their own supporters, and in equilibrium, the parties compromise extensive margin (trying to cover as various
ideological position as possible) with intensive margin (trying to mobilize its core supporters as
much as possible), i.e. divergence from the center. The current paper is closely related to their
work in the sense that the main source of platform polarization is voters’ limited, selective exposure to political information. However, the attention decision considered in this paper is absent in
their model, and consequently my model generates multiple equilibria which is rather new in the
spatial competition literature.
This paper is also related to the growing literature of media and political competition.7 Especially, some recent works shed lights on the relations between political polarization and media
enviornment. Bernhardt et al. (2008) show that polarization can induce media censoring, which result in inefﬁcient collective decision making. Campante and Hojman (2010) argue that enhacement
in media variety might cause political polarization by changing the landscape of public opinion.
5

See Grofman (2004) and Roemer (2001) for surveys on platform polarization.
Virag (2008) generalizes the model suggested by Glaeser et al.
7 For a review, see Pratt and Stromberg (2011).
6

59

More encompassing picture is provided by Chan and Suen (2008) who consider political competition together with news market competition. In contrast to what has been observed in recent years,
their model predicts that when more information is available in the market, parties’ positions tend
to become less polarized.
The remainder of paper is organized as follows. In the next section, I illustrate the main mechanism of the model with a few simplifying assumptions. Then, I present the extended model
in which I allow voters’ cognitive ability to be heterogenous and the candidates’ valance to be
stochastic. I discuss the limitations of this study and provide directions for future works in section
3.4.

3.2 Basic Model
Consider an election where two competing parties, L and R, choose their positions in a singledimension policy space [−1, 1] to maximize their votes.8 The election rule is simple majority,
and when tied, the winner is determined randomly. There are inﬁnitely many voters who can
be partitioned into three groups, g ∈ {l, c, r}, left-wings, centrists, and right-wings. The citizens
obtain intrinsic utility from voting itself, but also get disutility when the policy platform she votes
for differs from her ideal point. The utility of abstention is normalized to zero. Speciﬁcally, the net
utility of voting for a party whose platform is at d ∈ [−1, 1] is
ui (d) = B − |d − xi |
where B is the psychological beneﬁt of fulﬁlling a civic duty, xi is the voter i’s ideal point in the
ideology space, which can take one of three values:



 −1 if i ∈ l


xi =
0 if i ∈ c




 1 if i ∈ r

8 One can derive the qualitatively same result with the assumption of "winning probability max-

imization", but the calculation under the "vote maximization" is a bit more straightforward.
60

Denote the sizes of the groups of voters by αl , αc and αr . To focus on symmetric equilibria, let us
assume αl = αr , and normalize αl + αc + αr = 1. In this section, the cost of voting zV and the cost
i
of being attentive zI are assumed to be the same for all voters, i.e. zV = zV and zI = zI
i
i
i
The game consists of three stages: in the ﬁrst stage, given the prior beliefs on the policy platforms d L and d R , the voters decide whether to pay attention (or to gather information) to the politics
or not. Next, observing the size of attentive voters in each group, the parties simultaneously decide
their positions dL and dR . In the last stage, the voters decide whether to turnout and for whom to
vote.
To keep the model simple, I assume an attentive voter can observe only one of the actual
platforms chosen in the second stage: if a voter in group g decides to be attentive to politics,
she observes dL with probability δg and dR with probability 1 − δg (but not both) where the exogenously given probabilities δl > 1/2 and δr < 1/2, i.e. the left-wing voters observe and recognize the left-wing party’s behavior with higher probability than they observe the right-wing
party, and vice versa.9 Hence, the set of the information held by a voter in the third stage is
I = {(d L , dR ), (dL , d R ), (d L , d R )}. To make the environment symmetric, I suppose δl = 1 − δr , and
δc = 1/2. In the next section, I introduce stochastic quality (often called valance) of the candidates,
in presence of which the voters have the incentives to pay attention and to gather information to
make a better choice (or to minimize the regret of voting for a wrong candidate). In this section, as
a short cut I simply assume that the beneﬁt of voting is realized only if the voter is not in complete
ignorance, i.e. only if the voter pay attention to politics. Thus in equilibrium, only the attentive
and interested voters actually show up at the voting booth.
An equilibrium is a pair of belief (d L , d R ), the parties’ actual platforms (dL , dR ), and the voters’
strategy such that
(i) given the prior beliefs, the voters decide to be attentive to politics if and only if the cost of
9

This assumption is very similar to the one suggested by Glaeser et al. (2005), and can be
substituted by a more realistic assumption, e.g. some fraction of voters observe both dL and dR ,
and others observe only one of those. Key is that a non-trivial fraction of voters can observe only
one of the actual platforms.
61

being attentive is smaller than the beneﬁt of it;
(ii) given the sizes of the attentive voters, the parties select their platforms to maximize their
own votes;
(iii) the voters rationally decide whether to turnout and for whom to vote to maximize their net
utility;
(iv) the prior beliefs must be correct, i.e. (d L , d R ) = (dL , dR ).
To focus on symmetric equilibria, let us restrict our attention to the case with d L ∈ [−1, 0] and
d R ∈ [0, 1]. It is rather obvious that if the beneﬁt of voting B is large enough, everybody has an
incentive to turn out, which means that everybody would decide to be attentive in the ﬁrst stage.
With such a large B, therefore, the political parties will always have incentive to move toward
the center as the median-voter theorem predicts.10 On the other hand, if B is very small, nobody
would pay attention to politics, and consequently nobody would turn out. To make the illustration
non-trivial, I assume the beneﬁt B is in the intermediate range:
zI + zV < B < zI + zV + 1

(A1)

This condition says that if the ideological difference between the voter and the party |d − xi | is
zero, the voter always decides to be attentive and turns out, whereas if the ideological difference is
expected to be greater than or equal to one, she abstains.
To characterize the polarization equilibrium, suppose ﬁrst (d L , d R ) = (−1, 1). Then, the maximized expected utility for the centrists whose ideal point xi = 0 is
max {ui (−1), ui (1)} = B − 1,
which is smaller than the total cost zI + zV . Thus, the centrists decide to be inattentive to politics.
On the other hand, the expected utility for the left-wing voters whose ideal point is xi = −1 is
max {ui (−1), ui (1)} = ui (−1) = B,
10

More precisely, with sufﬁciently large B, there does not exist an pure strategy equilibrium
with polarized platforms. The median-voter equilibrium may not exist either.
62

and similarly that for the right-wing voters is also B which is greater than zI + zV by assumption.
Given the distribution of the attentive voters, the parties do not deviate from the presumed positions. First, the parties do not have the incentives to move to the center because the inattentive
centrists would not respond to such a change in platforms. Next, party L does not have an incentive to crowd into party R’s platform since such a move reduce the total votes for party L. This is
because when the left-wing party abandons its supporters, δl fraction of the left-wing voters would
observe the change in the position, and consequently abstain or cast their votes randomly. For
simplicity, let us assume B < 2, so that they abstain. Hence, the left-wing party loses αl δl votes of
the left-wings and in return additionally gain αr δr /2 votes from the right-wing voters. Recall that
αl = αr and δl > δr , which means the total votes for party L decline. Taking any position in (−1, 1)
is clearly a dominated strategy, because it reduces the support of the left-wings while cannot attract
any vote from the right-wings. By the same logic, party R would not deviate from its presumed
position. Therefore, the equilibrium pair of policy platforms is (dL , dR ) = (−1, 1) = (d L , d R ).
In a similar way, we can also characterize the median-voter equilibrium. Suppose the prior beliefs are given as (d L , d R ) = (0, 0). Now, the voters in group c decide to pay attention, whereas the
left-wings and the right-wings become disinterested. Because the attentive voters are concentrated
at the center, any deviation from the center clearly reduces the total votes. Thus, the equilibrium
positions are (dL , dR ) = (0, 0) = (d L , d R ).
Proposition 9 Under assumption A1, there exist multiple equilibria, in one of which the policy
positions of the parties are polarized, whereas in the other the parties converge to the median
voters.
This simple illustration highlights the role of beliefs which are most likely to be formed by
political parties’ positions in the past and their symbolic actions. By constructing such beliefs,
politicians in effect can "choose" whom to be pivotal and whom to be negligible in the election.
As in previous voting models with a large electorate, the voters cannot solve the collective action
problem, thus remain passive in selecting equilibrium.

63

3.3 Extension
In this section, I employ non-degenerated distributions for the cost of voting zV and that of ini
formation gathering zI . Also, I introduce an uncertainty as in probabilistic voting models; the
i
(relative) quality or valance of candidates is unknown to the parties and the voters in the ﬁrst two
stages, and it is realized in the last stage before the voters make decisions.11 Key departure from
the standard probabilistic voting model is that the candidates’ relative valance is observed only to
the voters who pay attention to the politics. As smooth distributions are introduced, the outcome
of the model becomes less stark, but the main idea remains the same.
Speciﬁcally, I assume zV follows uniform distribution of [0, 1/ψ], and zI follows uniform of
i
i
[0, 1/φ ]. Each cost is independent of each other and of the distribution of ideology. I assume that
the cost of voting is unknown in the ﬁrst stage, and is realized at the beginning of the third stage.
The relative valance of the candidate of party R is denoted by y, is drawn from uniform distribution
of [−1/2θ , 1/2θ ], and is observed by attentive voters in the third stage. All the distributions are
common knowledge. Unlike in the previous section, a voter can go to the polling booth without
knowing the actual position (dL , dR ) and the true value of y. But she suffers disutility when she
realizes that she made a "wrong" choice, the choice that differs from the one she would have made
if she knew y. The disutility of regret is denoted by λ .
The utility for voter i when voting for party L is
ui dL = B − dL − xi −

y
2

where dL ∈ dL , d L is the perceived position of party L, and similarly the utility when voting for
party R is
y
ui dR = B − dR − xi + .
2
In this enriched environment, equilibria similar to the ones analyzed in the previous section can
emerge depending on the parameter range. Here, I focus on the situation where at least some voters
11

For early papers which develop and analyze probabilistic voting models, see Hinich et al.
(1972) and Hinich (1977) among others.
64

in all groups decide to be attentive in the ﬁrst stage. The relevant parameter range is
1
1
− 2θ , B − 1 −
2
2ψ
1
1
− 2θ , B −
min λ
2
2ψ

min λ

> 0,
<

(A2)

1
.
φ

The ﬁrst inequality says that the voter with zero attention cost decide to be attentive even if the
ideological difference between herself and the party is as great as one (B − 1 − 1/2ψ is the net
expected utility and λ (1/2 − 2θ ) is the expected disutility of regret for the extremists). The second
inequality, on the other hand, states that the voters with the highest attention cost 1/φ pay no
attention even if the ideological difference is zero. I assume θ < 1/4, so that even the left-wing
voters cast their votes to the right-wing party for some large y, and the right-wings do for the leftwing party when y is very small. I further assume that the disutility of regret λ is sufﬁciently large,
so the regret is irrelevant when making the attention decision, speciﬁcally, suppose λ (1/2 − 2θ ) >
B − 1/2ψ.
In an equilibrium, inattentive voters might show up in the voting booth. However, because
they cast their votes based on the prior beliefs (d L , d R ), the politicians take their votes for granted
when choosing policy position. Hence, here I focus on the attentive voters. And, as before I focus
on symmetric equilibria, thus assume dL ∈ [−1, 0] and dR ∈ [0, 1]. Let us consider the left-wing
party’s position decision, and impose symmetry for dR . Given d R = dR , the expected votes from
the centrists who respond to a change of dL is
y
≥ zV
i
2
y
= αc δc εc Pr y + |dL | ≤ d R Pr B − |dL | − ≥ zV | y + |dL | ≤ d R
i
2

L
πc dL |d R = αc δc εc Pr y + |dL | ≤ d R and B − |dL | −

(3.1)

where αc is the measure of the voters in group c, δc is the fraction of the attentive centrists who
observe party L’s behavior, and εc denotes the fraction of attentive voters in group c, which is a
function of the prior beliefs. Voter i turns out, and votes for the left-wing party if y + |dL | ≤ d R
and B − |dL | − y/2 ≥ zV . Using the distribution assumptions made above, (1) can be simpliﬁed into
i
αc δc εc

1
+ θ dL + d R
2

d + dR
1
ψ B + dL − L
+
4
8θ
65

.

(3.2)

The marginal expected gain of moving toward the center is
L
∂ πc dL |d R
1
d + dR
3 3
+ θ dL + d R + θ B + dL − L
+
= αc δc εc ψ
∂ dL
8 4
4
8θ

.

(3.3)

,

(3.4)

Similarly, the expected votes from the responsive voters in group r is
L
πr dL |d R = αr δr εr

1
+ θ dL − d R
2

d − dR
1
ψ B − 1 + dL − L
+
4
8θ

of which derivative is
L
∂ πr dL |d R
d − dR
3 3
1
= αr δr εr ψ
+ θ dL − d R + θ B − 1 + dL − L
+
∂ dL
8 4
4
8θ

.

(3.5)

Lastly, the expected votes from the left-wings who are responsive to dL is given by
1
πlL dL |d R = αl δl εl
+ θ d R − dL
2

d − dL
1
ψ B − 1 − dL − R
+
4
8θ

,

(3.6)

and the marginal loss of moving toward the center is
∂ πlL dL |d R
1
3 3
d − dL
= −αl δl εl ψ
+ θ d R − dL + θ B − 1 − dL − R
+
∂ dL
8 4
4
8θ

.

(3.7)

The total expected votes as a function of party L’s position is the sum of (2), (4), and (6). Party L
takes the extreme position if given party R’s position, the marginal loss from moving toward the
center is greater than the marginal gain.
In a symmetric equilibrium, d L = −d R , and the attention decisions are made based on these
prior beliefs. Because with large enough λ a voter decide to pay attention if the net utility of voting
is expected to be greater than the cost of being attentive, the fraction of attentive voters in group c
can be written as
εc d L , d R = Pr E max uc d L , uc d R

−

1
≥ zI
i
2ψ

1
1
−
≥ zI
i
8θ 2ψ
1
1
−
= φ B + dL +
,
8θ 2ψ

= Pr B + d L +

that of group r is
1
εr d L , d R = φ B − 1 +
+ 2θ d L
2

d
1
+ L
8θ
2
66

+

1
− 2θ d L
2

d
1
− L
8θ
2

−

1
,
2ψ

and with the symmetric beliefs, the fraction of attentive voters among the left-wing voters εl is the
same with εr .
To characterize the polarization equilibrium, ﬁrst as in the previous section, suppose the voters
expect that (d L , d R ) = (−1, 1). Then, εg fraction of voters in group g decide to pay attention to
politics where
εc (−1, 1) = φ B − 1 +

1
1
−
8θ 2ψ

and
εr (−1, 1) = εl (−1, 1) = φ B − 1 +

1
1
.
+ 2θ −
8θ
2ψ

As in the baisc model analyzed in the previous section, when the voters expect the policy platforms
to be polarized, the voters at the extreme ideological positions pay more attention to politics (εr =
εl > εc ).12 In the second stage, party L sets its platform as dL = −1 if the marginal expected gain
of moving toward the center is negative:
L
L
∂ πlL dL |d R = 1
∂ πc dL |d R = 1
∂ πr dL |d R = 1
+
+
<0
∂ dL
∂ dL
∂ dL

From (3), (5) and (7), it is apparent that the marginal expected gain is greatest when dL = 0, so a
sufﬁcient condition for the polarization equilibrium to exist is
αc δc εc

1 1
1 3
1 1
+ θ + Bθ + αr δr εr
− θ + Bθ < αl δl εl
− θ + Bθ .
2 2
2 2
2 2

(3.8)

The median-voter equilibrium can be characterized in a similar way. Suppose the beliefs are
given as (d L , d R ) = (0, 0). Then, εc , εr and εl are given by
εc (0, 0) = φ B +

1
1
−
8θ 2ψ

,

and
εr (0, 0) = εl (0, 0) = φ B − 1 +
12

1
1
−
.
8θ 2ψ

In the basic model, a group of voters are homogenously attentive or inattentive, i.e. εg ∈

{0, 1}.

67

The fraction of attentive voters among the centrists is now greater than those among the left-wings
and right-wings. The left-wing party has an incentive to set its position at the center if the marginal
expected gain is positive:
L
L
∂ πlL dL |d R = 0
∂ πc dL |d R = 0
∂ πr dL |d R = 0
+
+
>0
∂ dL
∂ dL
∂ dL

which is minimized at dL = −1. Therefore, a sufﬁcient condition for the median-voter equilibrium
to exist is
αc δc εc

1 3
1
1 1
− θ + Bθ + αr δr εr
− 2θ + Bθ > αl δl εl
+ θ + Bθ .
2 2
2
2 2

(3.9)

The analysis so far is summarized in the following proposition.
Proposition 10 Under assumption A2, the polarization equilibrium exists if condition (8) holds.
The median-voter equilibrium exists if condition (9) is satisﬁed.
Note that as in the basic model, multiple equilibria exist for intermediate parameter range. To
see it clearly, suppose the left-wing voters observe only the left-wing party, and the right-wing
voters to the right-wing party, i.e. δr = 0 and δl = 1. And, as assumed from the beginning, the
centrists observe either party with the same probability, δc = 1/2. Then, the equilibrium conditions
(8) and (9) can be rewritten as
1
1 1
1 1
αc εc (−1, 1) ×
+ θ + Bθ < αl εl (−1, 1) ×
− θ + Bθ
2
2 2
2 2
and
1
1 3
1 1
− θ + Bθ > αl εl (0, 0) ×
+ θ + Bθ .
αc εc (0, 0) ×
2
2 2
2 2
Because εl (−1, 1) > εc (−1, 1) and εl (0, 0) < εc (0, 0), for a set of ideological distributions (αl , αc , αr ),
both inequalities are satisﬁed, which means both the polarization and the median-voter equilibrium
exist. The intuition is the same as before: when the voters expect the platforms are concentrated
at the center, the centrists become more attentive to the politics, and the parties come to serve the
attentive voters. When the policy positions are expected to be polarized, the voters at the extreme
positions become more responsive, and the parties optimally decide to mobilize the extremists.
68

To consider the effect of the fundamentals, namely economic inequality and media slant, let
us assume the voters’ attention is affected by the news coverage of media and their attitude in
delivering messages. Speciﬁcally, assume left-wing media deliver news in a way that suits the
left-wing voters’ taste, and right-wing media serve the right-wing voters. Consequently, as media
slant becomes severe, the extremists become more attentive, while the centrists tend to lose their
interests:
∂ εl (s) ∂ εr (s)
∂ εc (s)
,
> 0, and
≤0
∂s
∂s
∂s
where s is the economy-wide media slant.
The ideological distribution would be mainly formed by underlying economic interests. So
when the economic inequality increases the number of the centrists declines, and that of the extremists increases, i.e. as αc goes down, αl = αr = (1 − αc )/2 goes up. Such changes strengthen
the centrifugal force, and let the median voters be a small and unattractive group. In such a case,
the median-voter equilibrium cannot exist any more.
Proposition 11 For sufﬁciently large αl and s, the platforms diverge from the center.
Proof. For sufﬁciently large αl and s, the marginal gain of moving toward the center is always
negative as clearly shown in (3), (5) and (7).
Bartels (2007) and McCarty et al. (2008) argue that the increased economic inequality over
past thirty years might encourage the polarization in political arena. Campante and Hojman (2010)
and Prior (2007), on the other hand, provide some evidence of the effect of changes in media
environment. According to Prior, the media environment in 1950-70’s in the U.S. was best characterized by dominance of broadcast TV which is a low-choice medium. The environment started
changing around mid 70’s when cable TV spreading out all over the country. Cable TV, satellite
TV and the Internet are typical examples of high-choice media. Under this changed environment,
the voters can freely ﬁlter and edit the messages delivered to them, so make the media experience
more entertaining and enjoyable. These narratives are consistent with the above proposition.

69

3.4 Conclusion
In this paper, I develop a simple model in which political parties’ policy positions are polarized or
concentrated at the center depending on which group of voters are expected to actively participate
in politics. When a voter expects that at least one of the chosen platforms is close enough to
her ideal position, she tends to be more attentive to politics than when none of the parties serves
her taste. Thus, by forming beliefs about the policy positions, the politicians play a major role
in determining whom to be pivotal in the election. When the fundamentals such as economic
inequality and media environment allow multiple equilibria, elites’ role in political polarization
is essential. On the other hand, the model also predicts that when the underlying distribution of
voters’ ideology is polarized enough, the parties would be forced to be polarized.
This study has a couple of limitations which invites future studies. First, I assumed many
distributional assumptions to derive analytical conditions. Checking robustness of the main result
in terms of distribution could be a subject of a future work. Second, the prior beliefs has been
assumed to be given from outside of the model. A more comprehensive picture would appear when
the formation of the beliefs are properly considered. Lastly, the mechanism that the extremists
become more enthusiastic in the new media environment is only informally and brieﬂy discussed.13
An interesting future work would be to analyze the interaction between media market and political
competition when voters have limited attention.

13

For an example of modeling such a mechanism, see Mullainathan and Shleifer (2005).
70

BIBLIOGRAPHY

71

BIBLIOGRAPHY

Abramowitz, A. (2011). The disappearing center: engaged citizens, polarization, and American
democracy. Yale University Press, USA.
Abramowitz, A. and Saunders, K. (2008). Is polarization a myth? Journal of Politics, 70:542–555.
Acemoglu, D. (2003). Why not a political Coase theorem? Social conﬂict, commitment and
politics. Journal of Comparative Economics, 31:620–652.
Acemoglu, D. (2005). Politics and Economics in weak and strong states. Journal of Monetary
Economics, 52:1199–1226.
Acemoglu, Daron, D. T. and Vindigni, A. (2011). Emergence and Persistence of Inefﬁcient States.
Journal of European Economic Association, 9:177–208.
Acemoglu, Daron, M. G. and Tsyvinski, A. (2010). Dynamic Mirrless Taxation Under Political
Economy Contraints. Review of Economic Studies, 77:841–881.
Agrawal, A. a. C. R. K. (2001). Do some outside directors play a political role? Journal of Law
and Economics, 44:179–198.
Allen, F. and Zhao, M. (2007). The corporate governance model of Japan: Shareholders are not
rulers. Beijing University Business Review, 12:21–30.
Aoki, M. (1990). Information, Incentives, and Bargaining in the Japaneses Economy. Cambridge
University Press, New York.
Aoki, M. and Patrick, H. (1994). The Japanese Main Bank System: Its Relevancy for Developing
and Transforming Economies. Oxford University Press, New York.
Bartels, L. M. (2007). Unequal Democracy: The Political Economy of the New Gilded Age. Princeton University Press, Princeton.
Beason, R. and Weinstein, D. E. (1996). Growth, economies of scale, and targeting in Japan
(1955-1990). Review of Economics and Statistics, 78:286–295.
Berglof, E. and Perotti, E. (1994). The governance structure of the Japanese ﬁnancial keiretsu.
Journal of Financial Economics, 36:259–284.
Bernhardt, Dan, S. K. and Polborn, M. (2008). Political polarization and the electoral effects of
media bias. Journal of Public Economics, 92:1092–1104.
Campante, F. R. and Hojman, D. A. (2010). Media and polarization. working paper.
Chan, J. and Suen, W. (2008). A spatial theory of news consumption and electoral competition.
Review of Economic Studies, 75:699–728.

72

Che, Y. (1995). Revolving doors and the optimal tolerance for agency collusion. RAND Journal
of Economics, 26:378–397.
Claessens, Stijn, E. F. and Laeven, L. (2008). Political connections and perferential access to
ﬁnance: The role of campaign contribution. Journal of Financial Economics, 88:554–580.
Claessens, Stijn, S. D. and Lang, L. H. P. (2000). The separation of ownership and control in east
Asian corporations. Journal of Financial Economics, 58:81–112.
Clayton, M. J. and Jorgensen, B. N. (2005). Optimal Cross Holding with Externalities and Strategic
Interactions. Journal of Business, 78:1505–1522.
Colignon, R. A. and Usui, C. (2003). Amakudari: The hidden fabric of Japan’s Economy. Cornell
University Press, New York.
Degryse, Hans, M. K. and Ongena, S. (2009). Microeconometrics of banking: Methods, applications and results. Oxford University Press, New York.
Downs, A. (1957). An Economic Theory of Democracy. Harper and Row, USA.
Faccio, M. (2006). Politically Connected Firms. American Economic Review, 96:369–386.
Faccio, M. and Lang, L. H. P. (2003). The ultimate ownership of western European corporations.
Journal of Financial Economics, 65:365–395.
Farrell, J. and Shapiro, C. (1990). Asset ownership and market structure in oligopoly. RAND
Journal of Economics, 21:275–292.
Fearon, J. (2011). Self-enforcing democracy. Quarterly Journal of Economics, 126:1661–1708.
Fiorina, M. and Abrams, S. (2008). Political polarization in the American public. Annual Review
of Political Science, 11:563–588.
Fiorina, Moriss, S. A. and Pope, J. (2006). Culture war? The myth of a polarized America. Pearson
Longman, USA.
Fisman, R. (2001). Estimating the Value of Political Connections. American Economic Review,
91:1095–1102.
Gerling, Kerstin, H. G. A. K. and Shulte, E. (2005). Information acquisition and decision making
in committees: A survey. European Journal of Political Economy, 21:563–597.
Glaeser, Edward L., G. A. M. P. and Shapiro, J. M. (2005). Strategic Extremism: Why Republicans
and Democrats Divide on Religious Values. Quarterly Journal of Economics, 120:1238–1330.
Goldman, Etian, J. R. and So, J. (2008). Do politically connected boards affect ﬁrm value? Review
of Financial Studies, 22:2331–2360.
Granovetter, M. (2005). Business groups and social organization. The Handbook of Economic
Sociology, ed N. Smelser and R. Swedberg, Princeton University Press:429–450.

73

Grofman, B. (2004). Downs and Two-party Convergence. Anual Review, 7:25–46.
Hinich, M. J., J. O. L. and Ordeshook, P. (1972). Nonvoting and the existence of equilibrium under
majority vote. Journal of Economic Theory, 44:144–153.
Hinich, M. J. (1977). Equilibrium in spatial voting: The median voter result is an artifact. Journal
of Economic Theory, 16:208–219.
Horiuchi, A. and Shimizu, K. (2001). Did Amakudari undermine the effectiveness of regulator
monitoring in Japan? Journal of Banking and Finance, 25:573–596.
Hoshi, Takeo, A. K. and Scharfstein, D. (1991). Corporate structure, liquidity and investment:
Evidence from Japanese industrial groups. Quarterly Journal of Economics, 106:33–60.
Hotelling, H. (1929). Stability in Competition. Economic Journal, 39:41–57.
Johnson, C. (1975). Japan: Who governs? An essay on ofﬁcial bureaucracy. Journal of Japanese
Studies, 2:1–28.
Kandori, M. (1992). Social Norms and Community Enforcement. Review of Economic Studies,
59:63–80.
Kang, J.-K. and Shivdasani, A. (1995). Firm performance, corporate governance, and top executive
turnover in Japan. Journal of Financial Economics, 38:29–58.
Kaplan, S. and Minton, B. (1994). Appointments of outsiders to Japanese boards: Determinants
and implications for managers. Journal of Financial Economics, 36:225–257.
Keefer, P. (2007). The Poor Performance of Poor Democracies. The Oxford Handbook of Comparative Politics, ed. Carles Boix and Susan C. Stokes, Oxford University Press, USA.
Khanna, T. and Yafeh, Y. (2007). Business Groups in Emerging Markets: Paragons or Parasites?
Journal of Economic Literature, 45:331–372.
Khwaja, A. L. and Mian, A. (2005). Do lenders favor politically connected ﬁrms? Rent provision
in an emerging ﬁnancial market. Quarterly Journal of Economics, 120:1371–1411.
La Porta, Rafael, F. L.-d.-S. and Shleifer, A. (1999). Corporate ownership around the world.
Journal of Finance, 54:471–517.
La Porta R., F. Lopez-de-Silanes, A. S. and Vishny, R. (1999). The quality of government. Journal
of Law, Economics and Organization, 15:222–279.
Lagunoff, R. D. (2001). A Theory of Constitutional Standards and Civil Liberty. Review of
Economic Studies, 68:109–132.
Merlo, A. (2006). Whither Political Economy? Theories, Facts and Issues. Advances in Economics
and Econometrics, Theory and Applications: Ninth World Congress of the Econometric Society,
R. Blundell, W. Newey and T. Persson (eds.), vol. 1, Cambridge University Press, Cambridge.

74

Miwa, Y. (1996). Firms and industrial organization in Japan. New York University Press, New
York.
Miwa, Y. and Ramseyer, J. M. (2002). The Fable of the Keiretsu. Journal of Economics and
Management Strategy, 11:169–224.
Miwa, Y. and Ramseyer, J. M. (2005). Who appoints them, what do they do? Evidence on outside
directors from Japan. Journal of Economics and Management Strategy, 14:299–337.
Morck, Randall, A. S. and Vishny, R. (1988). Management ownership and market valuation: An
empirical analysis. Journal of Financial Economics, 20:293–315.
Morck, Randall, D. W. and Yeung, B. (2005). Corporate Governance, Economic Entrenchment,
and Growth. Journal of Economic Literature, 43:655–720.
Morck, R. and Nakamura, M. (2005). A Frog in a Well Knows Nothing of the Ocean: A History of
Corporate Ownership in Japan. A History of Corporate Governance around the World: Family
Business Groups to Professional Managers, NBER:367–466.
Morck, R. and Yeung, B. (2004). Family Control and the Rent-seeking Society. Enterpreneurship:
Theory and Practice, 28:391–409.
Mullainathan, S. and Shleifer, A. (2005). The market for news. American Economic Review,
95:1031–1053.
Okimoto, D. (1989). Between MITI and the market. Standford University Press, Standford.
Olson, M. (1965). The Logic of Collective Action: Public Goods and the Thoery of Groups. Havard
University Press, Cambridge.
Perotti, E. and von Thadden, E.-L. (2006). The political economy of corporate control and labor
rents. Journal of Political Economy, 114:145–174.
Persson, Torsten, G. R. and Tabellini, G. (1997). Separation of powers and political accountability.
Quarterly Journal of Economics, 112:1163–1202.
Persson, Torsten, G. R. and Tabellini, G. (2000). Comparative Politics and Public Finance. Journal
of Political Economy, 108:1121–1161.
Poole, Keith T., H. R. (1997). Congress: A Political-Economic History of Roll-Call Voting. Oxford
University Press, New York.
Prat, A. and Stromberg, D. (2011). The political economy of mass media. working paper.
Prior, M. (2007). Post-Broadcast Democracy. Cambridge University Press, New York.
Przeworski, A. (2005). Democracy as an Equilibrium. Public Choice, 123:253–273.
Przeworski, Adam, M. E. A.-J. A. C. F. L. (2000). Democracy and Development, Political Institutions and Well-Being in the World, 1950-1990. Cambridge University Press, New York.

75

Putnam, Robert D., R. L. and Nanetti, R. Y. (1993). Making Democracy Work. Civic Traditions in
Modem Italy. Princeton University Press, Princeton.
Raj, M. and Yamada, T. (2009). Business and Government Nexus: Retired Bureaucrats in Corporate Boardrooms. working paper.
Roe, M. J. (2003). Political determinants of corporate governance: Political context, corporate
impact. Oxford University Press, New York.
Roemer, J. (2001). Political Competition: Theory and Application. Havard University Press,
Cambridge.
Scheussler, A. A. (2000). Expressive Voting. Rationality and Society, 12:87–119.
Shleifer, A. and Vishny, R. (1986). Large shareholders and corporate control. Journal of Political
Economy, 94:461–488.
Shleifer, A. and Vishny, R. (1997). A survey of corporate governance. Journal of Finance, 52:737–
783.
Theriault, S. (2008). Party polarization in congress. Cambridge University Press, New York.
Van Rixtel, A. and Hassink, W. (2002). Monitoring the Monitors: Are Old Boys Networks Being
Used to Monitor Japanese Private Banks? Journal of Japanese and International Economies,
16:1–30.
Virag, G. (2008). Playing for your own audience: Extremism in two-party elections. Journal of
Public Economic Theory, 10:891–922.
Yafeh, Y. and Yosha, O. (2003). Large shareholders and banks: Who monitors and how? Economic
Journal, 113:128–146.

76