ESSAYS ON CARTELS IN KOREA
By
Yung Shin Jang

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Economics – Doctor of Philosophy
2017

ABSTRACT
ESSAYS ON CARTELS IN KOREA
By
Yung Shin Jang
The first essay “An Empirical Study of Anti-Cartel Enforcement Policies in Korea” investigates
and compares the impacts of the anti-cartel policies on stabilizing cartels and voluntarily revealing
incriminating evidence. The three policies, which were innovated in Korea around 2005, are the
following: i) a more generous and predictable leniency program, ii) expansion of resources, and
iii) a harsher maximum fine ceiling. In order to identify these policy effects, hazard rate models
are used with a sample of 327 cartel cases determined by the Korea competition agency from 2000
to 2010. The results show that the new leniency program and the resource expansion have a significantly positive impact on destabilizing cartels in the short run, while the long run effects for both
policies are ambiguous. Also, the announced harsher fine rule, as well as the actual fines imposed
by the competition authority, decrease the duration of discovered cartels. Pertaining to evidence
disclosing sthe effects of the three policies, a raised fine ceiling shows significantly greater impact
than the new leniency program. However, the expansion of resources fails in disclosing incriminating evidence.

The second essay “An Empirical Study of Cartels with Ringleaders in Korea” explores how
a cartel organization strategically chooses its structure to alleviate cheating and to sustain collusion. It focuses on organizational roles of cartel ringleaders. Using a choice model with a sample
of Korea’s 327 cartel cases from the 2000 to 2010, the research identifies the features of cartel
ringleaders, and then analyzes in what kinds of situations they tend to appear. The findings indicate that cartel ringleaders are more prevalent in the circumstances where a cartel is more difficult
to sustain. The probability of ringleader-involved cartels increases as more firms, greater size
asymmetry, or less product homogeneity exists in a cartel. It is probable that a cartel endogenously develops its organizational structure in response to market circumstances. By choosing a

ringleader-based cartel structure, firms can make incentive-compatible constraints less binding by
reducing distrust among themselves. This finding is confirmed for aggressive ringleaders who play
an active role in cartels (i.e., determining cartel plans) compared to organizational ringleaders who
play an administrative role in cartels (i.e., convening meetings). Last, the findings indicate that
ringleaders are significantly less likely chosen in cartel structures under the circumstances where
other alternative monitoring mechanisms, such as an industry association, exist.

Finally, the third essay “What Governs Cartels?: Evidence from Korea” explores what actually
governs a cartel. Even though multiple firms cooperate to manage a cartel, the cartel frequently
develops additional monitoring mechanisms to sustain it. This study investigates what types of
monitoring mechanisms a cartel has, and whether they effectively stabilize a cartel or not. The
study also compares the impact of different types of monitoring on cartel stability. Expectedly,
mechanisms controlling a cartel significantly decrease the likelihood of cartel death. However,
involvement of an industry association is less effective in stabilizing a cartel than existence of
regular meetings composed of cartel firms. Also, an industry association increases stability less in
cases of “death by anti-trust” cartels than in cases of “natural death” cartels. However, a regular
meeting in a cartel decreases the likelihood of death more in “death by anti-trust” cartel cases than
in “natural death” cases. Last, involvement of a higher level of personnel in each firm with a cartel
stabilizes “natural death” cartels, while it destabilizes “death by anti-trust” cartels.

ACKNOWLEDGMENTS

This dissertation is debt to many people, who have helped me for the last five years at Michigan State University. I might not have been able to complete this journey without their dedication.
First and foremost, I would like to express my sincere appreciation to my committee chair Jay Pil
Choi. Without his valuable guidance and persistent support from both personal and professional
perspectives, I could not have overcome numerous obstacles that I have faced through my research.
I also owe a great feeling of gratitude to the other members of my dissertation committee, Thomas
Jeitschko, Kyoo il Kim, and Aleksandr Yankelevich. They not only patiently helped me nurture
research ideas and develop them better, but also made it possible for me to experience various
opportunities in academia. I am so honored and privileged to have all four of them in my dissertation committee. I would like to thank my fellow graduate students for their meaningful feedback
and cooperation. Also, special thanks to Lori Jean Nichols, Dean Olson, and Margaret Lynch in
the department for their friendly and helpful support should be acknowledged. And finally, last
but by no means least, with a very special gratitude to my family! My parents, Daesoo Jang and
Jungsook Lee, always stand behind and support me. I am so grateful to them for giving me their
endless amounts of love, even when I probably do not deserve it. A most special gratitude goes
out to my lovable wife, Yunhui Choi, and my adorable two kids, Lucy and Michael. When I was
almost frustrated with the grad program, I would have dropped out of the program without their
sacrifice. Without expecting anything in return, they stand right next to me in order to support my
challenge.

iv

TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
CHAPTER 1
An Empirical Study of Anti-Cartel Enforcement Policies in Korea
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 The Collusion Game . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1.1 Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1.2 AA’s Policy Parameters . . . . . . . . . . . . . . . . . . . .
1.3.1.3 Comparative Statics . . . . . . . . . . . . . . . . . . . . . .
1.3.2 Anti-Cartel Policy Innovations in Korea . . . . . . . . . . . . . . . . .
1.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.1 Data and Sample Information . . . . . . . . . . . . . . . . . . . . . . .
1.4.2 Empirical Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Cartel Deterrence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5.2 Empirical Frameworks . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6 Incriminating Evidence Revelation . . . . . . . . . . . . . . . . . . . . . . . .
1.6.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6.2 Empirical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

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

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

1
1
6
9
9
9
10
10
11
14
14
22
24
24
25
27
34
34
35
36
38
40

CHAPTER 2
An Empirical Study of Cartels with Ringleaders in Korea
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Data and Stylized Facts . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1 Empirical Frameworks . . . . . . . . . . . . . . . . . . . . .
2.4.2 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . .
2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

43
43
47
51
59
59
60
67
69

CHAPTER 3
What Governs Cartels?:
3.1 Introduction . . . . . . . . . . . .
3.2 Data and Descriptive Statistics . .
3.3 Empirical Frameworks . . . . . .
3.4 Empirical Results . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

72
72
75
78
80

Evidence from Korea
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
v

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

3.4.1 Involvement of Industry Association
3.4.2 Existence of Regular Meetings . . .
3.4.3 Hierarchical Structure . . . . . . . .
3.4.4 Other Determinants . . . . . . . . .
3.5 Different Roles of an Industry Association .
3.5.1 Roles of Industry Association . . .
3.5.2 Results . . . . . . . . . . . . . . .
3.6 Conclusions . . . . . . . . . . . . . . . . .
REFERENCES . . . . . . . . . . . . . . . . . .

vi

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

80
81
82
83
85
85
85
87
89

LIST OF TABLES

Table 1.1

KFTC’s Leniency Filed Cases and Total Cartel Fine Amount (’01-’10) . . . 13

Table 1.2

Descriptive statistics of detected cartels in Korea (2000 - 2010) . . . . . . . 17

Table 1.3

Impacts on Cartel Stability(Survival model with 3-year short term) . . . . . 28

Table 1.4

Impacts on Cartel Stability(Survival model with 9-monthshort term) . . . . . 31

Table 1.5

Impacts on Revealing Evidence(OLS estimation) . . . . . . . . . . . . . . . 36

Table 2.1

Incidence of (ringleader involved) cartel case 2000-2010 . . . . . . . . . . . 52

Table 2.2

Comparison of cartels with/without ringleader . . . . . . . . . . . . . . . . 53

Table 2.3

Number and size of cartel ringleaders . . . . . . . . . . . . . . . . . . . . . 55

Table 2.4

Incidence of ringleader by types and roles . . . . . . . . . . . . . . . . . . . 56

Table 2.5

Incidence of ringleader types across agreement types . . . . . . . . . . . . . 57

Table 2.6

Incidence of products homogeneity across RL types . . . . . . . . . . . . . 58

Table 2.7

Incidence & roles of industry associations across RL types . . . . . . . . . . 58

Table 2.8

LPM, Probit, and Logit Estimates of Ringleader-based Cartel . . . . . . . . 61

Table 2.9

LPM, Probit, and Logit Estimates of Ringleader-based Cartel . . . . . . . . 64

Table 2.10 Multinomial Logit Estimates with 3 ringleader choices

. . . . . . . . . . . 65

Table 3.1

Descriptive statistics on Korea cartel by death reasons (2000 - 2010) . . . . 76

Table 3.2

Hazard rate estimation results . . . . . . . . . . . . . . . . . . . . . . . . . 80

Table 3.3

Hazard rate estimation results by Roles of Industry Association . . . . . . . 86

vii

LIST OF FIGURES

Figure 1.1 Number of Fined case (2000-2010) . . . . . . . . . . . . . . . . . . . . . . 19
Figure 1.2 Cartel durations (2000-2010) . . . . . . . . . . . . . . . . . . . . . . . . . 21
Figure 1.3 Fine reduction rates (2000-2010) . . . . . . . . . . . . . . . . . . . . . . . 22

viii

CHAPTER 1
An Empirical Study of Anti-Cartel
Enforcement Policies in Korea
1.1

Introduction

There have been debates over how mindful an anti-trust agency should be when intervening in a
market to regulate collusive agreements. However, one of the important fields of anti-trust policies
today, in practice, is to detect cartels in the market. Stigler (1964) questions the sustainability
of collusive agreements because of the enforcement problems such as detection and punishment.
Along this vein, it is not always necessary for governments to intervene in the market to deter collusion because collusion ends in a natural death due to firms’ strong temptation to cheat. However,
in recent studies on cartels- especially since a leniency program was introduced in the US in 1978many economists and policy-makers agree that cartels such as price fixing or market allocation
behaviors should be regulated through anti-trust intervention to enhance market competition and
consumer welfare.1 In reality, despite continuous efforts of anti-trust agencies, firms still form
cartels in various and secretive ways. They have incentives to collude if long-run forgone future
profits responding to the retaliation by other firms are greater than short-run deviation profits, even
though their behaviors can be adjusted for costs and probability of capture. Therefore, many antitrust agencies worldwide have cultivated policy environments by innovating anti-cartel policies
and enhancing detecting capabilities.2 The change of policy parameters has an effect on a firm’s
1 See

Motta and Polo (2003), Spagnolo (2004), Hammond (2004), Harrington and Chang (2009), and
Miller (2009). They show theoretical framework or empirical analysis on anti-trust interventions. According to the FTC (Fair Trade Commission), about 130 jurisdictions have antitrust or consumer protection laws.
(https://www.ftc.gov/policy/international/competition-consumer-protection-authorities-worldwide)
2 Harrington and Chang (2008) argue that when a competition policy is made tougher with a higher detection rate
and a harsher penalty rate, the value to forming a cartel is reduced.

1

incentive compatibility constraints (ICC) regarding colluding versus deviating. In other words,
firms face different long-term benefits from colluding and short-term benefits from deviating as
the anti-cartel policy parameters change.
The most successful innovation against cartels is a leniency program, which originated in 1978
from the US Anti-trust Division of the Department of Justice (DOJ). However, it was not so effective until it was revised in 1993. The main change was that the eligibility of granting leniency is
wider and the requirements have become more predictable. A full automatic fine reduction to a first
informant and a criminal amnesty for individuals are provided when reported before an investigation has been launched.3 After the success of the US DOJ in 1993, many competition authorities
in other countries began to introduce the program.4 Anti-trust agencies also increase maximum
fines and resources such as budget and workforce for detecting and deterring cartels. Motta and
Polo (2003) argue that if resources are sufficient to deter collusion, utilizing full fines is preferred
over a leniency program for maximizing social welfare. However, they also argue that a leniency
program may be optimal due to limited resources.
It is difficult to discover whether or not these anti-cartel policies are effective as expected, since
we observe only discovered cartels, not a potential population of cartels. For example, even though
anti-trust agencies have been discovering more cartels since the changes in the anti-trust policy, it
might result from the success of the policy (where detection rates increase) or its failure (where
formation rates increase). To specify, innovation of the leniency program can increase cartel formations because a reduction of expected fines is granted if firms report when or before cartels
are detected. In contrast, the program can lead cartels to collapse by increasing the probability
of detection. À priori, it is difficult to conclude which impact dominates. If the current detection
capability of the anti-trust agency is high enough, enhancing its capability may have less effect
3 There are three major changes. First, amnesty is automatic if there is no pre-existing investigation. Second,
amnesty is also available even if cooperation begins after an investigation is underway. Third, if a corporation qualifies
for automatic amnesty, then all directors, officers, and employees who come forward with the corporation and agree
to cooperate also receive automatic amnesty. Refer to “The modern leniency program after ten years” presented by
James M. Griffin in 2003. (https://www.justice.gov/atr/speech/modern-leniency-program-after-ten-years-summaryoverview-antitrust-divisions-criminal)
4 The European Commission sets up an LP in 1996, Korea in 1997, UK in 2000, France and Germany in 2001,
Japan in 2005.

2

on the ICC that a firm faces. However, if the detection capability of the anti-trust agency is low
enough, the policy change can disrupt cartels by enhancing the agency’s detection capability even
a little. Also, it is not simple to determine which policy between increasing resources and a leniency program contributes more to disrupting cartels. In addition, it is conventional wisdom that
increasing penalties may make firms more likely to deviate by raising the expected cost of collusion, holding other things constant. However, if the current fine level imposed on cartels is too
low, and still below the current maximum fine ceiling, increasing the ceiling may not necessarily
change a firm’s incentive. Thus, evaluating the efficacy of anti-cartel policies is an empirical issue.
Around 2005, the Korea Fair Trade Commission (KFTC) innovated anti-cartel policies toward
more aggressive enforcements. These policies can be categorized into three groups: i) revising a
leniency program more generously and predictably, ii) enlarging a cartel investigation organization with more staff and budgets, and iii) raising the maximum fine ceiling against cartel behavior.
These changes of the policy environment with time and case-variations can be a favorable opportunity for a natural experiment, where the impact of anti-cartel policies on cartels can be empirically
estimated. The main objective of these policy changes is to destabilize and deter cartels by modifying firms’ incentives. Additionally, when the KFTC investigates and prosecutes the violation,
these policies can contribute to enhancing the conviction probability by making firms voluntarily
provide more hard evidence necessary for establishing the wrong-doing. This article mainly attempts to obtain adequate answers to the following questions: i) Do the three anti-cartel policies
have sufficient impact on disrupting cartels in the market before cartels are detected? ii) Do these
policies make cartel members voluntarily disclose their incriminating evidence after cartels are
detected?
In the current paper, the impact of both the new leniency program and empowered resources on
cartel stability are explored and compared, using survival analysis. Also, an increased fine ceiling
against cartels is investigated, which is compared to the effect of an actual fine amount on cartel
behaviors to derive more abundant conclusions. Second, the effect of the policies on revealing
incriminating evidence is also tested using OLS. Previous studies mainly focus on the effect of a

3

leniency program. Even though anti-cartel policies such as the leniency program usually tend to
be planned and enforced with the change of fine rule and organizational resources of the anti-trust
agencies, those studies do not distinguish the impact of these policies. For example, Brenner (2009)
tests Harrington and Chang’s (2009) hypotheses with EC (European Commission) cartel data from
1990 and 2003 without considering the change of “Guidelines on the method of setting fines”,
which became effective on January 14, 1998. Also, Choi and Hahn (2014) examine the impact
of the corporate leniency program using Korea data from 1981 and 2012. However, neither the
restructuring of the cartel investigation organization in December 2005, nor the revised guideline
which doubled the fine ceiling against cartels is explicitly considered. It may be because it is
apparently difficult to divide the policy effects. However, without isolating the individual effects,
there may exist the possibility that the impact of the leniency programs is overestimated if other
policies modified with the leniency program also have disruptive effects on cartels. Exceptionally,
Zhou (2012) isolates the effect of the leniency program in the EU from that of other institutional
changes by restricting the sample to the cartel cases that dissolved before the major change in the
EU’s anti-cartel regime.5 However, the study does not compare the effect of the leniency program
and other institutional change. Also, the revision of “Guidelines on the method of setting fines” in
2006 has not been examined.6
The present paper extensively tests the main hypotheses that Harrington and Chang (2009)
propose regarding anti-cartel policies. This study might be one of the first empirical papers that
separately compare impacts of a leniency program and increased resources on cartel sustainability.
The results show that both the newly revised leniency program, which grants an automatic full fine
reduction to a first reporter, and the increased resources with a fourfold increase in investigation
workforce have significantly positive effects on destabilizing cartels in the short run. “Marginal
cartels,” rather than stronger and longer duration cartels, are more likely to dissolve, and so we
5 The White Paper, announced in April 2, 2008, “suggests specific policy options and measures that would help
giving all victims of EU antitrust infringements access to effective redress mechanisms so that they can be fully
compensated for the harm they suffered.” See http://ec.europa.eu/competition/antitrust/actionsdamages/index.html for
details.
6 The detailed EC’s guideline can be available in the website, http://ec.europa.eu/competition/antitrust/legislation/fines.html

4

observe the longer durations of discovered cartels as Harrington and Chang (2009) predict. These
results are consistent with Motta and Polo (2003), who argue that, when a competition authority has
limited resources, a leniency program can be optimal in a second-best perspective.7 However, in
the long run, both policies show ambiguous results. A possible interpretation is that the increased
cartel durations in the short run due to marginal cartel break-down seem to be almost offset with the
reduced cartel duration due to the deterrence effects in the long run. The results cannot significantly
confirm, for the sample period, which effect is dominant between the deterrence effects (decreasing
cartel durations) and the short run effects (increasing cartel durations). Another possibility is that a
cartel may endogenously develop its alternative collusive governance to facilitate collusion through
learning-by-doing effects or some trustworthy tools such as industry associations or ringleaders in
the long run. If so, an antitrust policy desisting current collusive agreements may fail in deterring
future collusion in the long run. The third policy that raises the fine rate ceiling, as well as the
actual fine imposed by the KFTC shows positively significant effects on disrupting cartels. This
result implies that when a firm considers the expected cost for collusive behavior, not only the
history of penalty, but also the announced fine rule affects the behavior.
For the incriminating evidence effects, the revised leniency program and the harsher maximum
fine ceiling have significantly positive effects on revealing information. Leniency program effects
are smaller. This result is possibly because a leniency program applies to only leniency applicants,
while a general fine guideline applies to all cartel members. However, the increased resources do
not show any significantly positive effects on evidence revealing.
The remainder of the paper is organized as follows. The next section reviews the previous
literature on the impact of anti-cartel policies. Section 3 explains institutional backgrounds and
offers a simple model that shows how an anti-trust agency changes an incentive compatibility
constraint that a firm may face. In Section 4, the data set is described. In Section 5, effects
of the anti-cartel policies on cartel stability are explored before of those policies on revealing
incriminating evidence are presented in Section 6. Finally, Section 7 concludes the paper.
7 They argue that when a competition authority has sufficient resources, using a full fine is the best for maximizing
social welfare rather than implementing a leniency program.

5

1.2

Literature Review

From the perspective of firms, the antitrust agency’s policy enters through policy parameters such
as a fine rate, the probability of detection and prosecution, monitoring resources, etc. (Aubert et
al, 2006; Harrington and Chang, 2009; Motta and Polo, 2003). Theoretically, a leniency program
has ambiguous impact on the sustainability of cartels, as it has two opposite effects. A leniency
program can make enforcement more effective by ex-post destabilizing cartels (Harrington, 2008;
Spagnolo, 2004), while it might also induce collusion by ex-ante decreasing the expected cost
of violation of anti-trust law (Chen and Harrington, 2007; Ellis and Wilson, 2003; Spagnolo,
2000). Motta and Polo (2003) further argue that a leniency program is supposed to be the second
best in detecting cartels. In other words, if the investigation resources available to an anti-trust
agency are sufficient to prevent collusion using full fines, a leniency program had better not be
used. However, a leniency program may be socially desirable because investigation by an antitrust agency requires scarce resources, which may be saved by making cartel participants reveal
any incriminating evidence (Kaplow and Shavell, 1944). These concerns should be empirically
tested in order to resolve them adequately.
However, as many previous empirical papers on cartels indicate, studying the efficacy of anticartel policies on cartel stability has a few inherent limitations for the following reasons. Since we
cannot observe the potential population of cartels in the market, there can be a sample selection
bias (Brenner, 2009; Harrington and Chang, 2009; Levenstein and Suslow, 2006; Miller, 2009).
We only perceive cartels discovered by anti-trust agencies, but do not exactly know how many
undiscovered cartels exist in the market. For instance, even though the number of detected cartels
following policy changes increases, we cannot argue that the new anti-cartel policy is efficacious.
It is because the number of detected cartels can decrease with more effective policy than before.
If the new policy is semi-successful, such that the cartel formation rate decreases but the detection
rate remains unchanged, the number of detected cartels can decline. Thus, adequate proxies and
conditions to analyze the effectiveness of the policies are needed. Miller (2009) and Harrington and

6

Chang (2009) independently developed theoretical frameworks with which to study the efficacy of
anti-cartel policies.8 Both of their frameworks argue that the short-term and long-term efficacy
of the policy should be considered separately. Miller (2009) shows that if a leniency program is
effective to deter cartels, the number of detected cartels will increase in the short run, and then
decrease to below the pre-leniency level in the long run. On the other part, Harrington and Chang
(2009) show that an efficacious leniency program should increase cartel durations in the short run
while the long run effect is ambiguous. The intuition is somewhat simple. Assume that before
a leniency program is introduced or revised more leniently, there were marginal cartels, where
firms were indifferent between colluding and competing because long-term profits from colluding
and short-term profits from deviating were equal. Now, if a new or innovative leniency program
is introduced and has destabilizing effects on current cartels, then the marginal firms’ deviating
benefits exceed colluding benefits, which leads the marginal cartels to dissolve. These short-run
effects make the average durations of discovered cartels longer. However, in the long run, even
strongly sustainable cartels can collapse because of the forceful deterring effects of the leniency
program, which shorten the cartel duration. Thus, the total effect depends on the magnitude of
both short-run and long-run impacts. However, unlike a leniency program, the policy that increases
detecting probability shortens the duration of discovered cartels as the industry goes from short-run
to long-run.
Miller (2009) also empirically predicts the impact of the new leniency program on the US cartel cases. The introduction of the new leniency program was implemented on August 10, 1993.
Under the old leniency program, the DOJ approved leniency status only on a discretionary basis and for cartel members who had reported their violation before an investigation started. He
finds that cartel detection and deterrence capabilities are enhanced under the new leniency pro8 Miller (2009) develops a theoretical model that provides empirical predictions and moment conditions to analyze
the effects of the new leniency program revised in 1993 in the US. He assumes that the formation rate, the detection
rate, and the dissolution rate of cartels are constant and are determined outside of the model, while Harrington and
Chang (2009) assume that those rates are endogenized in the model. Also, Miller (2009) focuses on the frequency
of cartel, whereas Harrington and Chang (2009) argue that the frequency of cartel cannot be representative proxy in
evaluating the success of some policies such as a leniency program and detecting capability, because they can increase
or decrease the number of discovered cartels.

7

gram by showing that the number of cartel discoveries increases in the short run and then falls
below the pre-leniency level. He uses Poisson regression with finally identified 40 DOJ cases from
January 1, 1985 to March 15, 2005. Brenner (2009) explores an empirical evaluation of the initially introduced EC’s leniency program in 1996 with 61 cartel cases between 1990 and 2003 to
test Harrington and Chang’s (2009) hypotheses. He finds that the duration of discovered cartels
insignificantly increased since introducing the initial leniency program. However, he finds that
this program had a significantly positive effect on reducing the investigation and prosecution costs.
Zhou (2012) develops an empirical framework of Cox hazard model and uses 144 cartel cases between 1985 and 2012 to estimate the impact of newly revised EC’s leniency program in 2002.9 He
finds the destabilizing efficacy of the new leniency program of the EC.10 Choi and Hahn (2014)
show the empirical evidence of the effectiveness of the new leniency program revised in 2005 on
cartel stability with the 619 cases decided by the KFTC from 1981 to 2012. Also, Oindrila De
(2010) and Feinberg et al. (2016) conduct duration analyses to extensively study the determinants
of a cartel including leniency programs in the EU and Korea, respectively. Both studies find the
the programs are effective, even though the leniency programs are not the primary focus. However, as suggested above, these previous empirical papers do not separately consider the effects
of other anti-cartel policies that might have affected cartel stability, even though those policies
were changed and enforced with the leniency program at about the same time, with some interval.
To identify and compare the individual policy effects on cartel firms’ behaviors, it is desirable to
separately evaluate those policies.

9 Like the US leniency policy prior to August 1993, the EC’s policy did not provide automatic and complete
amnesty from monetary fines. However, in February 2002, the program was revised to provide more generosity and
less discretion of the authority. A first leniency applicant became given full fine reduction without uncertainty when
a firm reports a cartel before the EC opens an investigation. This key change is very similar with the policy the US
revised in 1993.
10 He also finds the effectiveness of the US DOJ leniency program on cartel stability, using 294 DOJ cases from
1985 through 2005.

8

1.3

Background

1.3.1

The Collusion Game

1.3.1.1

Timing

The setting of this model is based on Aubert et al. (2006). Two firms play an infinitely repeated
game. Both firms have the same discount factor δ ∈ (0, 1) and maximize the expected discounted
sum of their profits. An Anti-trust Agency (AA) can impose fines on colluding firms and mandate
corrective orders to prohibit them from colluding.11 The timing is as follows:
1. At t = 0, the game begins with an AA setting the three policy parameters, which are the
amount of full fine (F), the degree of leniency (L), and the amount of resources to detect a cartel
(R).
2. At t = 1, when an AA launches an investigation, firms decide whether to sustain collusion or
to deviate for applying for leniency.
- i) If a firm deviates, it obtains a deviation profit, π D , and pays a reduced fine, f . Also, it has
to comply with AA’s corrective order to cease collusion, which thereafter makes the firm obtain
a competing profit, π C , by continuing to compete.12 However, the other firm gets π when a firm
deviates, and thereafter obtains π C .
- ii) If a firm colludes, it continuously enjoys a collusion profit, π M , and faces the expected
fine, ρF, where ρ ∈ (0, 1) is a conviction rate.
We assume that firms gain from collusion, but each firm benefits at the expense of the other
from deviation, i.e., from competing when the other colludes. This implies the inequalities, π ≤
π C < π M < π D and π + π D < 2π M by efficiency effect. We moreover assume that a full fine is not
11 An

anti-trust agency attempts to maximize consumer surplus by imposing fines on colluding firms. Moreover,
the agency spends budgets and workforce to detect cartels. However, for simplicity, government’s constraint is not
considered here.
12 It can be plausible if a firm who has a penalty history might face a substantially heavy fine and indictment.
However, it may not be empirically plausible because a firm can re-negotiate its agreement, rather than reverting to
π C . Also, according to Green and Porter (1984), on-and-off cartels or periodic price wars can be an equilibrium
without defecting from collusive agreements.

9

enough to deter collusion if it is imposed with probability ρ, implying that ρF < π M − π C < F.
The reduced fine is assumed to be lower than the expected fine from the AA’s random investigation,
meaning that f < ρF.
1.3.1.2

AA’s Policy Parameters

The AA exogenously sets F, L, and R to deter collusion by revising anti-cartel policies. Moreover,
we assume the following:
i) The reduced fine, f , is decreasing in the degree of leniency, L, i.e., f (L) < 0.
ii) The detection rate, ρ, is increasing in resources, R, i.e., ρ (R) > 0.13
1.3.1.3

Comparative Statics

The collusion is sustainable only if
14

δ ≥ δ+ =

π D − π M + ρ(R)F − f (L)
π D − π C − f (L)

(1.1)

If anti-cartel policies are made tougher, implying a higher value for a detection rate and a
penalty, the value to forming a collusive agreement is reduced due to the increase of the expected
fine. Also, when a leniency program grants larger leniency for a reporter, the deviating payoff
increases. Therefore, the following comparative statics can be drawn from the model.
i) AA can reduce a discounted fine, f , by enhancing the extent of leniency, L. This policy can
be implemented by introducing a leniency program or increasing the generosity and predictability
of the old program. Then, the threshold, δ + , will increase, implying that firms are more likely to
deviate from collusion, i.e., ∂ δ +/∂ L = (∂ δ +/∂ f ) (∂ f/∂ L) > 0.15
ii) AA can also improve a detection capability, ρ, by increasing budgets and workforce to
detect cartels implying expanding resources R. Then, the threshold, δ + , will increase, implying
13 For

simplicity, it is assumed that a detection probability depends on resources an anti-trust agency has. However,
it might be possible that changes in the degree of leniency endogenously affect the probability of detection.
14 The collusion is sustainable only if a long-term benefit from colluding is greater than a short-term benefit from
deviating. π M − ρ(R)F + δ π M − ρ(R)F + δ 2 π M − ρ(R)F + · · · ≥ π D − f (L) + δ π C + δ 2 π C + · · ·
15 ∂ δ +/∂ f

=

−[π D −π C − f (L)]+[π D −π M +ρ(R)F− f (L)]
[π D π C − f (L)]2

=

π C +ρ(R)−π M
[π D π C − f (L)]2

10

<0

that collusion is more likely to be destabilized, i.e., ∂ δ +/∂ R = (∂ δ +/∂ ρ ) (∂ ρ/∂ R) > 0.
iii) AA can set a higher ceiling for maximum fine, F, which increase the threshold, δ + , meaning that cartels might become unstable, i.e., ∂ δ +/∂ F > 0.

1.3.2

Anti-Cartel Policy Innovations in Korea

Since the enactment of the anti-trust law in Korea, the Monopoly Regulation and Fair Trade Act
(MRFTA) in 1980, and the KFTC the following year, anti-cartel policies have experienced many
variations. The KFTC had introduced the initial leniency program in 1997, but the KFTC seemed
to have difficulties in detecting secretive cartels until 2005, when it was revised. Before 2005, the
KFTC did not have a publicly formal guideline on how to enforce a leniency program. Instead, the
program had been operated based on related provisions of the MRFTA, which wrote only the basic
principles and the maximum fine reduction rate for a leniency applicant. Therefore, it would be
at the KFTC’s discretion whether to grant leniency to a reporter or not and, if granted, how much
of a discount in the fine would be provided. Also, the size of the organization for investigating a
cartel was small, such that only one team with seven staff members oversaw cartel investigations.
However, the KFTC, around 2005, experienced the three major innovative breakthroughs on anticartel policies toward tougher enforcements. The policies can be divided into three categories: a
more generous and predictable leniency program, a deeper cartel investigation organization with
more staff, and an increased maximum fine rate ceiling against cartels.16
At first, the new leniency program, on April 1, 2005, was overhauled to ensure leniency applicants a clear understanding of the program’s procedures. Before then, the first leniency applicant
was eligible for reduction of a fine by 75 percent or more only in the pre-investigation stage, while
the reduction rate was reduced to less than 50 percent if the first applicant cooperated with the
KFTC after an investigation was launched. More importantly, the KFTC had a wide range of
discretion in granting leniency and deciding on the amount of discounted fines. Even worse, a leniency applicant did not know how much fine reduction would be granted until the case was finally
16 The

related information on these three anti-cartel policies can be obtained from the news release of the KFTC.
(http://eng.ftc.go.kr)

11

determined. Under the new program, as soon as a firm applies for leniency, the KFTC has to notify
the firm of the eligibility of the leniency application. The discretion of the KFTC is minimized
in judging the eligibility. Even before a leniency applicant confesses, the applicant can predict an
automatic full leniency only if the applicant firm is eligible for a first reporter. In addition, this
automatic full leniency becomes granted even after an investigation is already opened. The range
of fine reductions also becomes widened and steeper. Under the new program, a first applicant gets
a 100 percent fine reduction for sure, while the second one receives a 30 percent reduction. Overall, the predictability and transparency of the leniency program improve compared to the previous
regime by making new guidelines on a leniency program that specifies standards and conditions as
to procedures and eligibility for leniency application and approval.17
The KFTC had underwent organizational restructuring on December 19, 2005.18 While many
structural changes were made, the most remarkable variation was the birth of the "Cartel Bureau."
It grew from a division-size with only seven staff members to a bureau-size with 31 staff members composed of three divisions: the Cartel Policy Division, the Manufacturing Industry Cartel
Division, and the Service Industry Cartel Division.19 This restructuring was designed to allow
the KFTC to concentrate on cartel detecting tasks with more resources. If the KFTC’s detecting
capability had been very low, this resource expansion possibly increased the KFTC’s enforcement
capability in detecting cartels, which could result in changing cartel firms’ IC constraints.
Finally, in December 2004, the announcement that the cartel fine ceiling rates would be doubled
was made by revising the MRFTA.20 To enhance cartel deterrence, the fine ceiling was raised to
ten percent from five percent of the amount of the related sales. This increase may serve as a
disincentive for a cartel if cartel firms regard it as raising the expected costs related to collusive
17 The

guideline is titled “Public notification on implementation of leniency program including corrective
measures against voluntary confessors, etc. of unfair cartel activities.” For more details, refer to the link,
http://ftc.go.kr/data/hwp/Commssionnews(2005may).doc
18 The previous hierarchical and vertical structure was replaced by a more horizontal and function-oriented structure. Investigation Bureau, Anti-trust Bureau, and Competition Bureau were combined altogether to create Headquarter for Competition Law & Policy Enforcement and Cartel Bureau. Headquarter for Competition Law & Policy
Enforcement is in charge of correcting abuse of market dominance, anti-competitive M&As, and other unfair business
practices, while Cartel Bureau detects cartels.
19 A bureau of the KFTC is made up of 3-5 divisions, where one division has about 6-10 staff members.
20 This new fine rule was implemented in April 2005.

12

Table 1.1 KFTC’s Leniency Filed Cases and Total Cartel Fine Amount (’01-’10)
Year
’01
’02
’03
’04
’05
’06
’07
’08
’09
’10

Total

LP* filed case (percent† )
0 (0)
1 (7.1)
1 (9.1)
0 (0)
6 (26.1)
6 (22.2)
10 (41.7)
20 (46.5)
13 (61.9)
17 (68.0)
74 (31.6)

Fine amount (billion KRW‡ )
27.7
53.1
109.8
29.2
249.3
110.5
307.0
197.5
52.9
585.8
1,722.8

Source: 2011 KFTC Annual Report
*Leniency Program
†[Number of cases with the LP plus a fine/ Number of cases with a fine] x 100
‡ one billion Korean Won can approximately be conversed into one million US dollars

behaviors. As displayed in the simple model above, the increase in expected fines can tighten the
ICC and raise the threshold for colluding by decreasing the long-term benefits from colluding, other
things being equal. That is, the increase in the maximum fine ceiling can cause cartel members to
change their cartel behaviors if they accept this change as the higher expected cost for colluding,
which makes cartels unstable. However, it is also possible that cartel firms do not change their
behaviors if the changed fine ceiling does not affect their expected costs.
These policy changes observably seemed to be more effective in detecting cartels than before.
It is easily noticed in Table 1.1 that both the fined/leniency filed cases and the total amount of
fines dramatically increased since 2005. According to the KFTC, out of the 163 total fined cases
from 2005 to 2010, 72 cases (44 percent) were leniency-filed. The percentage has been on the rise
almost every year. Also, the annual cartel fines imposed by the KFTC soared from roughly 30
million to 250 million US dollars in 2005 when the policies changed. This trend is very similar
to that of the US, where the revised Corporate Leniency Program resulted in a surge in amnesty
applications. Under the old policy, roughly one amnesty application per year is obtained, while
under the new policy, the application rate has jumped to more than one per month.21
21 “The modern leniency program after ten years” was presented by James M. Griffin in 2003.
(https://www.justice.gov/atr/speech/modern-leniency-program-after-ten-years-summary-overview-antitrust-divisions-

13

1.4
1.4.1

Data
Data and Sample Information

This paper covers the data set of finalized 327 cartel cases filed for violations of Article 19 (1) of
the MRFTA between January 1, 2000 and December 31, 2010.22 At first, 396 cases are collected
from public decision reports released on the KFTC’s website. However, out of them, 69 cases are
excluded because they are not original cases. They are only related to the cases on recalculating
fines or for prosecuting alleged conspirators. If one or more cartel members are dissatisfied with
the KFTC’s decision, they can appeal to a higher court. If the higher court reverses the KFTC’s
decision and orders it to recalculate overly-imposed fines, the KFTC should redeliberate its original
decision. Moreover, the KFTC is supposed to decide whether or not to prosecute cartel members
when the Prosecutor’s Office requests that the KFTC accuse them.23 The number of cases is
consistent with the number of decision reports.24
After 327 decision reports are closely read, the relevant variables for this research are chosen
and drawn based on the previous theoretical predictions and empirical analyses. Even though this
unique data set has some limitations, it has a few advantages over previous empirical studies on
the cartel cases of the EU and the US because it includes a larger number of case samples and
provides richer, more extensive, and detailed information on cartel characteristics. The data set
covers the number and names of cartel members, whether or not they are large companies,25 the
criminal)
22 Some English-version documents of the cases filed during the periods are available for downloading from the
Commission website, www.eng.ftc.go.kr.
23 Korea has the ‘Criminal Complaint Exclusive System’ under anti-trust law, implying that the Korea Prosecutor’s
Office alone may not bring a criminal antitrust lawsuit under the MRFTA. Instead of independent prosecuting, the
office should request the KFTC consider the decision whether to prosecute or not. Then, the KFTC makes a decision
after deliberating anti-competition effects of the case in the related markets. However, the KFTC now has little
discretion on prosecuting since the MRFTA was amended in 2014. However, even though this institution is weakened,
the Prosecutor’s Office still should depend on the KFTC, when prosecuting.
24 For example, if the KFTC divides one cartel case into two sub-cases depending on related product markets, two
decision reports are produced and this cartel is formally counted as two cases.
25 In Korea, there is a unique ‘Large Business Group Policy’ to prevent too much concentration of economic power.
That is, ‘Institution of designating large business group’ applies to conglomerates whose affiliates’ combined assets
amount to about 5 billion US dollars or more. The minimum asset requirement doubled up to about 10 billion US

14

type of cartel,26 whether or not it is fined, the related industry, the affected geographic and product
market, the start- and end-date of cartels,27 the involvement of industry associations, the amount
of fines imposed pre- and post-stage where leniency scheme is applied, fine rates, fine reduction
rates, etc.
Even though it is still difficult to know the exact identities of leniency applicants due to the
strict confidentiality of the KFTC, much information on the leniency program is inferred through
a matching process, where the information of anonymous firms and final fines charged to the
cartel firms are paired to identify the relevant firms. The cartel duration, which is one of the most
important variables in testing cartel stability, completely depends on the KFTC’s decision. The
duration of each firm in a cartel can be different due to the dates of entry and exit. So, the value of
each firm’s duration is averaged. Because the KFTC does not operate a plea bargaining process, it
might be relatively cleaner, compared to the DOJ data. The cartel duration information may have
some inherent measurement issues because it is completely dependent on the KFTC’s decision.28
However, due to the limitationof data, a cartel duration is used in this study as a proxy for cartel
success as many previous cartel literature uses the duration. The number of cartel firms implies
maximum value of the variable in each cartel case. That is, even though there are some changes
in the number of firms during the cartel period due to entry and exit, the number of firms includes
all the firms involved in the case. Even though there are eight types of violations regulated by
the MRFTA, most agreement types fall into either price fixing or market sharing. Previous papers
dollars in September 2016. If a conglomerate falls into this category, and is annually designated as ‘Large Business
Group’ by the KFTC, it faces very extensive regulation such as restrictions on stock holding and finance, stricter
disclosure obligations in a stock market, differentiated tax burdens, limitations on entry to specific small industry, etc.
26 There are eight types of violation regulated by the MRFTA. They are price fixing, restriction of the transaction
term, production quantity restriction, market allocation, restriction of product specification, establishment of a joint
company, interfering with business and a bid rigging.
27 The start- and end-date of a cartel are identified, based on the KFTC’s decision reports. In case where those
dates does not explicitly defined in the reports, the first meeting date of any cartel firms is used for the start-date, and
the end-date is either the date when the case is finally decided by the KFTC, or the date when firms self-correct the
violation, whichever is earlier.
28 As Levenstein and Suslow (2006) point out, the cartel duration based on dates of formal dissolution, which are
written in cartel contracts, might overestimate the true economic duration of cartels. In contrast, cartel durations
determined by a anti-trust agency also can be mostly conservative estimates supported by the hard-core evidence
produced in the court. For example, the KFTC can guess prior starting date that is before the one in document, but the
KFTC cannot confirm the starting date due to lack of hard evidence. Also, in the context of Green and Porter (1984),
it is hardly distinct an apparent on-and-off cartel from some episodes of repeated cartels.

15

argue that cartel duration can be affected by industry concentration and the market share of the
collusion (Dick, 1966; Ellis and Wilson, 2003; Hay & Kelley, 1974; Selten, 1973). However,
that information is not included in the data set for the following reasons: There are too much
missing information on industry concentration and market shares in the publicly opened decision
reports. And, they might have endogeneity problem because they result from cartel durations. This
missing information may be covered to some extent by industry effects and a large firm dummy
to some extent.29 The amounts of cartel fines are included in two ways, which are fines pre and
post-leniency scheme. The fine amounts of the two types are the same for a non-applicant, while
the fines can be different for a leniency applicant. That is, for an applicant, fines after applying a
leniency scheme are smaller than those before the leniency scheme, if the applicant is eligible for
granting leniency. Fine rates are different, depending on the severity of violations. If the severity is
more substantially significant rather than minor, higher fine rates are applied within the maximum
fine ceiling. When the KFTC calculates the final fine amount before a leniency scheme, the KFTC
can adjust the fine amount depending on other factors. For example, compliance with the authority
decreases fines, while recidivism increases them.30 That is, even though firms are not eligible
for leniency, all firms are granted fine reduction if they cooperate in revealing more evidence.
In final stage, leniency applications are considered, where firms’ names cannot be identified for
confidentiality.
Table 1.2 compares descriptive statistics for the cartel cases identified. Note that the number of
cases, the cartel duration, and annual fines sharply increased, even though considering the slightly
longer period covered after the policies changed. In particular, the number of total cartel cases
(cases without bid-rigging) increased by 66 percent (31 percent). The bid-rigging cartels relatively
increased more than the non-bidding cases. The KFTC detected fined cases about three times more
than before from 51 to 156. The proportion of fined cases increased from 41.5 to 76.5 percent
29 The one-digit industry dummy is used, which is based on the Korea Standard Industry Classification. The large
business group dummy is based on the list of the large business group when the case is determined.
30 The KFTC, in order to calculate cartel fines, firstly defines the sales amount of the cartel-related products. Second, it obtains the basic fine amount through multiplying the cartel-related sales by fine rates, which are determined
depending on the severity or importance of the violation. Then, it adjusts the basic fine in accordance with a firm’s
cooperativeness and history of recidivism. Finally, a leniency scheme is applied under confidentiality and anonymity.

16

Table 1.2 Descriptive statistics of detected cartels in Korea (2000 - 2010)
2000 - 2004

2005 - 2010

Annual fines* (million KRW)

123
94
51
3
97
6.59
(4.70)
82,097

Fines per case* (million KRW)

8,377

204
123
156
70
166
5.86
(4.42)
269,976
[175,300]†
10,728
[6,966]†
32.58
(34.36)
16.42
(10.96)
4.36
(1.97)
3.74
(6.6)

Number of total cases
Number of cases (without bid-rigging)
Number of fined cases
Number of leniency filed cases
Number of price fixing cases
Average number of firms

Average cartel duration (month)‡
Average investigation duration (month)
Average fine rate (percent)
Average fine reduction rate for compliance (percent)**

13.95
(19.50)
5.11
(3.09)
2.75
(8.63)
0.11
(0.59)

Remark: Standard deviations in parentheses
*Fines are averaged and expressed in million KRW, where 1,000 KRW can be approximately conversed into 1 US
dollars.
† The average fine amounts after the leniency scheme is applied are expressed in the bracket.
‡ The average cartel duration is obtained except for bid-rigging cartels.
** Prior to the application of leniency scheme, the KFTC discounts fines for firms who cooperate into its investigation
through providing incriminating evidence.

during the same periods. It seems that the KFTC discovered more serious cartel cases after the
anti-cartel policies were revised. The new anti-cartel policies probably resulted in the discovery
of underground cartels discovered that might not have been discovered otherwise. Most of all, the
number of leniency filed cases soared from 3 to 70 when comparing cases before and after 2005.31
The average annual fine amount from 2000 to 2004 was only approximately 82 million dollars.
31 These

figures might be little different from the exact number from the KFTC. The KFTC does not disclose the
individual information pertaining to a leniency program for each case. However, when compared to the aggregated
figures that were announced in 2011 annual report(http://www.ftc.go.kr/policy/whitebook), it is not much different.

17

But, it increased more than threefold to 270 million US dollars since 2005, whereas the averaged
fines per case increased only slightly. Therefore, it may be possible to think the increase in total
fine amount is mainly attributed to the expansion of those cartel cases that were discovered.
Excluding 110 bid-rigging cartels, the average cartel duration is obtained because bid-rigging
cartels tend to be a one-time event, rather than continuous or repeated like normal price-fixing or
market allocation agreements. The average cartel duration increased more than two times from
14 to 33 months during the same periods. It seems to be apparently consistent with Harrington
and Chang’s (2009) prediction that cartel durations, if an anti-cartel policy is in effect, are reduced
in the short run due to breaking down of marginal cartels. However, more rigorous analyses are
needed to draw confident conclusions. One of the interesting points pertaining to cartel duration is
that the duration is very short, compared to that of the US or EU data. According to Levenstein and
Suslow (2006), the cartel duration in prior research ranges from 3.7 to 10 years. The current data set
divides a geographic market into three categories; local, entire nation, and international markets.
As the scope of markets becomes larger, a cartel organization may require higher monitoring cost,
which can reduce cartel stability. However, it shows the contradictory results. The average cartel
duration for local, entire nation, and international markets are 13 months, 32 months, and 67
months, respectively. There possibly exists a monitoring mechanism to facilitate collusion in a
larger geographic market. Or, multi-market contact can improve firms’ ability to sustain higher
prices by combining the incentive compatibility constraints (Bernheim and Winston, 1990).
The increasing rate of investigation durations is even larger than that of cartel durations, which
is about a threefold increase. It contradicts Brenner’s (2009) findings, where investigation duration, used as a proxy for revealing information, decreased since the EC leniency program was
introduced. On the one hand, the KFTC’s anti-cartel policies apparently seem to fail at inducing
cartel firms to reveal information. However, on the other hand, investigation duration may not be
representative of revealing information. I would lean toward the possibility of an less proper proxy.
The investigation periods can increase in the amount of revealed evidence. For example, if cartel
members do not disclose any evidence, an anti-trust agency has few sources to verify cartel firms’

18

Figure 1.1: Number of Fined case (2000-2010)

Remark: When one cartel case is documented with multiple cases for some administrative convenience reasons, only
the first case is counted to avoid a nuisance.

violations. In contrast, if more complex and detailed evidence is disclosed, an anti-trust agency
has to spend more time in investigating and finalizing the cartel case. The lengthier investigation
duration could result from longer cartel duration with more abundant and complicated incriminating evidence. Rather than the investigation duration, a fine reduction rate in return for cooperative
compliance would be more appropriate for a proxy for revealing information. The KFTC adjusts
basic fines depending on how cooperative cartel firms are. This fine reduction rate due to revealing
information soared since anti-cartel policies were revised around 2005. It seemed that the policy
changes led cartel firms to disclose more incriminating evidence when they are detected.
Figure 1.1 shows that the number of fined cartel cases increased since 2005 in Korea.32 The
trend is partially in line with Miller’s (2009) findings. Using DOJ’s data set, he shows theoretically
32 The

number of total cases including non-fined ones also shows increasing trend, but it is more clear in the fined

cases.

19

and empirically that the number of discovered cartels rises in the short run and drops below a
pre-leniency level if a leniency program is effective. However, the trend in Korea does not show
that the number of discovered cartels drops below pre-leniency levels in the long run. In 2008
and 2010, the number of fined cartel cases discovered soared than in previous years. In those
years, there was a nuisance issue. Some cartel cases, which were discovered in an industry at the
same time, were documented as multiple decision reports for some administrative and convenience
reasons. For example, the chemical cartel case which falls into 2008 was split into seven decision
reports depending on product types such as toluene, ethylene, xylene, etc. Also, the bid-rigging
cartel case for constructing apartment complexes in 2010 was also documented in seven decision
reports, based on the districts of construction. If these cartel cases over the sample periods are
regarded as only one case to avoid this sort of noise, the trend becomes more consistent with
Miller’s (2009) findings, even though it seems to require a larger sample period in order to confirm
long-run decreasing trends to below the pre-policy change level.
The cartel duration of the discovered cartels in this paper is one of the most important dependent
variables in analyzing cartel stability, despite some of its limitations, as stated above. Regarding
this duration, Figure 1.2 demonstrates a similar trend in Figure 1.1. The raw data offers some preliminary insight. The average duration of cartels peaked in 2007, and then there was a decreasing
trend. Harrington and Chang (2009) argue that the duration of discovered cartels increases in the
short run because if an anti-cartel policy is effective, marginal cartels among population cartels begin to break down. However, in the long run, even strong cartels can collapse so that the duration
of discovered cartels either rises or drops. Note that in order to determine the year that a cartel case
falls into, a decision date is used. That is, the cartel is categorized based on when it was decided,
regardless of when the cartel actually started. If a cartel’s duration is categorized based on the
cartel start date, a different trend can be displayed. Choi and Hahn (2014) analyzed the effects of
Korea leniency programs on cartel deterrence, using basically Zhou’s (2012) methodology. They
found that if a cartel case is grouped by a cartel start date, the average duration of discovered cartels
that started on and after 2005 is shorter than one that started before 2005.

20

Figure 1.2: Cartel durations (2000-2010)

Remark: Bid-rigging cartels are excluded

Figure 1.3 displays the time trend of the average fine reduction rate due to complying with
investigation, which is used as a proxy for revealing evidence. Before 2005, firms were not willing
to disclose incriminating evidence to the KFTC. One possible reason is that cartel firms have less
incentive to voluntarily reveal hard evidence on their side. It might be because that the KFTC
had very low prosecution capability to prove a firm’s violation even after a cartel behavior had
been detected.33 However, the fine reduction rate increased since 2005. If anti-cartel policies are
effective, the KFTC can be provided with more incriminating evidence, implying a higher fine
reduction rate due to complying with the KFTC. The descriptive results show the possibility of the
effectiveness of the KFTC’s anti-cartel policy innovations around 2005. However, we cannot argue
that the changes in anti-cartel policies around 2005 have any significant effects on destabilizing a
33 There

were less fined cases before 2005. During those periods, about 60 percent of total cases were finalized
without imposing fines, which is far higher proportion comparing to about 25 percent from 2005 to 2010.

21

Figure 1.3: Fine reduction rates (2000-2010)

Remark: The fine reduction rates is related to how well cartel members comply with anti-trust agency’s investigation.
More cooperations result in higher fine reduction rates, where the rates are ranged from 0 to 20 percent.

cartel and saving investigation costs with these descriptive statistics. A more rigorous analysis will
be conducted later in this paper.

1.4.2

Empirical Issues

Note that when we compare the impact of anti-cartel policies on cartel stability, we should check
whether there have been some significant changes in cartel organization before and after policy
changes to avoid mistaken interpretations. If they are largely different under each regimes, we can
have another selection bias and should be cautious when interpreting the impact of the changed
policies. Because there may exist structural changes in forming cartels other than duration or
number of cartels, we cannot conclude whether the policies make cartels unstable or not. In the

22

current paper, the comparison test is conducted to alleviate this issue by checking the differences
in the characteristics of the discovered cartels. The number of cartel members, the geographic
scope of cartel markets, and the type of cartels are compared between pre- and post-2005 when
anti-cartel policies were revised. The differences of the every mean are not statistically significant
at conventional levels of significance. The t-values of the mean comparison test with the three
variables are 1.406, 1.501, and 0.553 with df =324, respectively. These results show that there are
no distinguishable changes in the structural characteristics of cartels before and after 2005 when
anti-cartel policies were innovated. With this finding, we have fewer problems in estimating the
effectiveness of the anti-cartel policy changes.
Also, some measurement issues are addressed in empirically analyzing anti-cartel policies. The
duration determined by an anti-trust agency is mostly the length of a single event. The duration data
used in this study, like most previous cross-section duration analyses, can be regarded as a legal
definition rather than an economic definition. As mentioned above, it is possible that the duration
supported by incriminating evidence can either overestimate or underestimate the true value of an
actual duration. In addition, as Miller (2009) points out, the cartel duration information reported by
the DOJ may be distorted due to the negotiation of a plea bargaining process. However, the KFTC
does not operate a plea bargaining institution unlike the DOJ. So, the relatively less contaminated
data set as to cartel durations can be obtained from the publicly opened decision reports even
though this issue cannot be completely eliminated. Therefore, in spite of a possibility of selection
bias that can be unavoidable in empirical research on cartels, the KFTC’s duration data is used in
the current study. The theoretical framework of Harrington and Chang (2009) is applied using the
KFTC’s public data set.

23

1.5
1.5.1

Cartel Deterrence
Hypotheses

Brenner’s (2009) methodology is adopted and modified to test the hypotheses of Harrington and
Chang’s (2009) model. According to their theoretical framework, the increased detection rate
may raise the duration of the population of discovered cartels because “marginal cartels” start to
collapse first, and thus the surviving cartels tend to be stronger and longer in the short run. First, the
new leniency program that grants an automatic full leniency to a first confessor and the enlarged
resources with a larger investigation organization and more staff are considered. If the two anticartel policies have been effective in enhancing the KFTC’s detection and conviction capabilities,
then the duration of discovered cartels increases in the short run. Hence, this section first focuses
on analyzing the short-run predictions of the two anti-cartel policies.34 In addition, Motta and
Polo’s (2003) argument is tested by comparing the impact of the two anti-trust policies. They
conclude that even though using a full fine to deter cartels should be preferred when the resources
are enough, a leniency program may be an optimal enforcement for the second best when the
anti-trust agency’s resources are not sufficient.
In addition, the impact of the aggressive new fine rule on cartel deterrence is investigated. It is
conventional wisdom that higher penalties destabilize current cartels and lead firms to form fewer
cartels because this change can tighten the incentive compatibility constraints a firm faces, given
other factors fixed. However, as Harrington and Chang (2009) point out, if an intensified penalty
is perceived by a cartel firm, and so can increase its incentive to deviate from cartels or not to form
a cartel, cartels are more likely destabilized. However, in spite of a policy change, if a firm feels
indifferent and has the same expected cost as before, this policy may be ineffective in disrupting
34 The long run effects of the two anti-cartel policies are different.

The effects of a leniency program, in the long run,
are ambiguous because this program also has a pro-collusive impact as well as a destablizing one. That is, the average
duration of discovered cartels can increase or decrease. However, an enlarged resources in investigation manpower can
decrease the duration of discovered cartels if it is effective in improving an anti-trust agency’s detection and conviction
capability. For more details, refer to Harrington and Chang (2009)

24

cartels. This may happen if a maximum fine ceiling is raised but proves to fail in affecting firms’
expected penalties.35 To test this prediction, the effect of the new fine rule - doubling the fine
rate ceiling - on cartel stability is explored. The result is also compared to the effect of actual
fine amount imposed in the previous year, because firms may build their expectations based on the
past.36 Therefore, the following three hypotheses are proposed.
[Hypothesis 1] The new leniency program with certain full immunity to a first reporter increases the duration of discovered cartels in the short run.
[Hypothesis 2] More cartel detection resources with upsized organizations increases the duration of discovered cartels in the short run.
[Hypothesis 3] The increase in a maximum fine rate decreases the duration of discovered
cartels.

1.5.2

Empirical Frameworks

While Cox’s semi-parametric hazard model is used in many previous analyses of cartel durations
(Choi and Hahn, 2016; De, 2010; Zhou, 2012; Zimmerman and Connor, 2005), Brenner (2009)
uses a Weibull specification, where the shape of the hazard is parametrically estimated.37 An
accelerated failure-time (AFT) parametric survival model with Weibull distribution rather than a
proportional hazard (PH) model is mainly applied in the current paper, but the Cox model is also
employed to investigate the robustness of the results.38 In this paper, unlike the previous empirical
35 Based

on the fact that the averaged fine rate during the sample periods is only 3.88 percent, increasing maximum
fine rates ceiling from five percent to ten percent might not affect cartel firms’ behaviors.
36 Here, the time-dependent effect of the new fine rule is not investigated. It would be more informative to compare
the general effects of the maximum fine ceiling to ones of the actual fines imposed by the KFTC rather than comparing
time-dependent effects.
37 Cox regression is considered a ‘semi-parametric’ procedure since the baseline hazard function, h (t) does not
0
have to be specifically assumed, while a fully proportional parametric hazards model should assume the baseline
hazard function. Hazard rates can be parameterized according to the survival times whose distributions are mainly
exponential, Weibull and Gompertz. The advantages of Cox hazard model in analyzing durations have been widely
recognized. (Meyer, 1990)
38 Zhou (2012) points out that the Weibull specification used by Brenner (2009) is likely to be plagued by unobserved heterogeneity that might cause biased estimates as Meyer (1990) indicates. He argues that Cox approach may
have advantages in controlling for unobserved heterogeneity. In Cox model, the hazard function for cartel is hi (t; Xi ) =
h0 (t) × exp(Xi β ) where t is the elapsed time since the start of a cartel, and Xi is a vector of observed explanatory variables. The parameter vector β is the vector of coefficients, measuring the influence of observed characteristics. The
term exp(Xi β ) shifts the baseline hazard function h0 (t), and a positive coefficient indicates that the observed charac-

25

estimation where only a leniency program is considered, the impacts of other important anti-cartel
policies are also separately tested. This paper may be one of the first empirical papers that directly
compare the impact of a leniency program and enlarged resources of an anti-trust agency. Previous
studies do not separately identify the effect of a leniency program from the effect of other anticartel polices, even though each policy affects cartel stability. Those effects should be individually
evaluated when a leniency program is planned and enforced at a similar time with the change of fine
rules and organizational resources. If not, the effect of a leniency program can be overestimated.
The following model is estimated to test Hypothesis 1 to 3:

ln(DCi ) = β1 + β2 LPDi + β3 LP STin + β4 RDi + β5 R STin + β6 NFRi + Xi β + εi

(1.2)

where for a particular cartel case i, DCi is the average duration of the cartel. LPDi is the duration
of the cartel that lasted since the new leniency scheme had been introduced in April 1, 2005.
LP STin equals one if the cartel was discovered within the first n periods since the new leniency
program had existed, and equals zero otherwise. In this variable, n periods can be three years or
nine months. At first, Brenner’s (2009) short-term definition, three years is used. However, the
same estimation is also conducted except for setting the short-term nine months in order to avoid
overlapping periods of the two policies’ short-terms and to isolate the impact of the new leniency
program from that of the enhanced resources. RD is the duration of the cartel that lasted since the
KFTC’s Cartel Bureau with more resources was initiated on December 19, 2005. R STin is also a
dummy variable that takes the value of one if the cartel was discovered during the first n period
since the time that the cartel detecting resources were upsized. NFRi is the new fine rule dummy,
which is one if case i was subject to the new fine rule that doubled the fine ceiling. Xi is the row
vector variables affecting the duration of cartels, including the following variables: the number
of cartel members (NFi ), the large business group dummy (LBGi ), the price fixing cartel dummy
(PFi ), the annual total cartel fines imposed by the KFTC one year before the observed cartel starts
teristics increase dissolution hazard rates and reduce cartel durations. And the ratio of the hazard rates does not depend
on time but only depends on Xi and β like the following: h(t,x1 )/h(t,x2 ) = h0 (t)×exp(β x1 )/h0 (t)×exp(β x2 ) = exp{β (x1 − x2 )}
(Zhou, 2012). When controlling for other variables, the coefficients shift the hazard rates of cartels at time t.

26

(Fi ), the geographic market dummy (Geoi ), and the industry dummy (INDi ).39 εi is an error term
following an extreme value distribution. We call a β the parameter to be estimated. According
to Hypotheses 1, 2, and 3, the positive β3 and β5 , and the negative β6 are expected, respectively.
These coefficients capture the differences between the durations of cartels that failed before the
related anti-cartel polices that were introduced and those under the policies after controlling for
other factors. Therefore, the positive coefficients increase the durations of discovered cartels while
the negative ones decrease the durations.40

1.5.3

Results

In this subsection, the regression analyses for the Equation 1.2 are performed in two steps. First,
a survival analysis is conducted after setting the short-term three years as Brenner (2009) does.
The impact of anti-cartel policies on cartel stability is estimated in both Weibull and Cox specifications, so that Harrington and Chang’s (2009) hypotheses are tested. In the second step, the same
regression is re-run, except that here a short term is nine months to avoid the overlapping effects
between the new leniency program and the expanded cartel organization.
In Table 1.3, the first step - with a three-year short-term - regresses four specifications of survival analyses, which consist of two short models without Xi and two full models with Xi . The
short and the long equations are regressed using both the parametric hazard model with a Weibull
specification and the semi-parametric Cox model. Note that in an AFT parametric model with the
Weibull distribution [(1) and (3)], a positive coefficient means increasing cartel durations, which
says a covariate extends the time until failure. However, the interpretation of the coefficients, in
the semi-parametric Cox model [(2) and (4)], should be the opposite. In these models, a positive
coefficient increases the hazard ratio, therefore decreasing durations, the time until failure.
39 The large business group dummy equals one if at least one of cartel members falls under designated large business

group in the year when the case is finalized. The price fixing cartel dummy equals one if the type of the cartel is
price fixing, and equals zero otherwise such as market allocation, restriction of product quantity, etc. The geographic
market dummy is categorized into local, entire nation, and global. The industry dummy is based on the Korea Standard
Industry Classification.
40 Note that in Cox model, positive coefficients mean increasing hazard ratios, while negative ones decrease the
ratios, after controlling for other factors.

27

Table 1.3 Impacts on Cartel Stability(Survival model with 3-year short term)
(1)
Wei(S)3y
-.001
(.001)
1.171***
(.239)
.002
(.002)
-.039
(.262)
-.317
(.323)

(2)
Cox(S)3y
.001
(.001)
-1.189***
(.249)
-.002
(.002)
.126
(.259)
.261
(.309)

(3)
(4)
3y
Variables
Wei(F)
Cox(F)3y
LPD
-.0003
-.001
(.001)
(0.002)
LP-ST3yrs
.487**
-.862**
(.215)
(.304)
RD
.001
-.001
(.001)
(.002)
R-ST3yrs
.14
-.016
(.206)
(.288)
NFR
-.336
.555*
(.224)
(.309)
F
-.331*** .536***
(.042)
(.07)
NF
.012
-.014
(.013)
(.017)
LBG
.064
-.053
(.156)
(.21)
PF
-.338*
.404*
(.173)
(.232)
Geo
.705*** -.966***
(.13)
(.182)
IND
No
No
Yes
Yes
Cons.
6.084***
8.506***
(.098)
(.526)
2
Chi
50.13
47.67
175.59
181.08
n
211
211
210
210
Remark: Standard errors in parentheses. Significance levels: *<10 percent, **<5 percent, ***<1 percent.

The estimated results for the Equation 1.2 are suggested in Table 1.3. First, the coefficients
of main interest are β3 (the coefficient of the the leniency program dummy, LP-ST 3yrs ) and β5 (the
coefficient of the resource expansion, R-ST 3yrs ). They are the short-term effects of two anti-cartel
policies on cartel stability. Most importantly, the new leniency program, from Specifications (1)
and (3), has a positive effect on dissolving cartels as Hypothesis 1 predicts. The new leniency program causes the duration of discovered cartels to increase in the short run because it breaks down
the “marginal cartels.” The program moves the fragile cartels from a population of sustainable cartels to a population of unstable ones. Both the short and the full model specifications estimate the

28

statistically significant coefficient at a conventional level of significance. These findings are not
consistent with Brenner (2009) and O. De (2010). In those studies, the EC’s leniency program
did not have a significant impact on disrupting cartels. However, our study is in accord with Zhou
(2012) and Choi and Hahn (2014), in which the new leniency programs of the EC and Korea significantly affected cartel stability in the short run, respectively. For robustness, Cox regressions are
also employed in Specifications (2) and (4). The relevant coefficients from the Cox model show
similar results to Weibull specifications. However, concerning Hypothesis 2, no significant estimate is obtained both in the short and the full model. Even worse, they show a mixture of opposite
signs. Thus, we cannot argue that the upsized workforce for detecting cartels is successful. Relating to Hypothesis 3, we cannot strongly argue that the aggressively changed penalty scheme (the
new fine rule dummy, NFR) does not increase the expected cost that cartel participants face. All
of the coefficients of interest on Specifications (1) through (4) are insignificant at a conventional
level of significance, even though the signs seem to reduce cartel durations. That is, increasing
a fine ceiling from five to ten percent of the related sales amount does not significantly affects
cartel firms’ incentives to deviate. Rather than the fine ceiling, the actually imposed fines seem
to be more effective to deter cartels. The coefficients of the variable Fi , the annual total cartel
fines imposed by the KFTC one year before firms join cartels, are negatively associated with cartel
durations at a conventional level of significance.
However, as mentioned above, the first step with a three-year short term may have serious overlapping period issues. In other words, the estimated effects of the new leniency program on cartel
stability might be from the new leniency program or the expanded cartel organization. Because
there is a nine-month time gap between the new leniency program and the KFTC’s organizational
restructuring, the three-year short term may be too lengthy to isolate two different policies. It
might be difficult to separate the impact of two different anti-cartel policies, because the policy
of upsizing resources were operated in the following nine months since the innovation of the new
leniency program, which is within the first three years after the leniency policy changed. Thus, in
order to address this issue, two actions are taken. First, the short term is shortened to nine months

29

in the second step regressions, meaning that n equals nine months in Equation 1.2. Also, either
of two cartel spans spent under two policies, which are variables LPD and RD, is excluded in the
full specifications. By doing so, two benefits are obtained. The impacts of the two anti-cartel policies can be separately estimated. That is, technically, there might be fewer overlapping problems
between the two anti-cartel policies. On top of that, the long-run effects of the policies can be estimated. If the detecting capability is enhanced through upsizing resources, more time spent under
the policy should have a negative effect on cartel durations in the long run. However, regarding a
leniency program, the long-run effects are ambiguous (Harrington and Chang, 2009).

41

The estimation results for the second step are given in Table 1.4. The new leniency program
significantly increases the duration of discovered cartels like the first step. Here, however, unlike
the results shown in Table 1.3, the signs of β3 (the coefficient of variable R-ST9m ) under Specifications (1) through (4) - the effects of the expanded resources on desisting cartels - also become
significant. This finding significantly supports Hypothesis 2. After isolating the two anti-cartel
policies, the results display not only the new leniency program, but also increasing investigation
resources with more staff is also successful in disrupting collusion in the short run. These findings are compatible with Harrington and Chang (2009), who argue that both enhancing detection
probability and a leniency program increase cartel durations in the short run. Interestingly, when
comparing the effect of the new leniency program of the KFTC to the results of Choi and Hahn
(2014), the value of related estimate is smaller in the current paper.42 It leaves the possibility that
the previous results overestimate the impact of the leniency program because two anti-cartel policies are not isolated. The results pertaining to the impact of the two anti-cartel policies are also
confirm Motta and Polo’s (2003) conclusion. They argue that when an anti-trust agency has insufficient resources, it is desirable to use a leniency program with other policy tools for disrupting
collusion. However, it is difficult to argue which policy is more effective in the short run. Compar41 It is because marginal cartels will be dissolved or not be formed in the short run, causing an increase in discovered
cartel durations. However, in the long run, even strongly sustainable cartels are finally destabilized, decreasing the
duration of the detected cartel, if anti-cartel policies are effective.
42 According to Choi and Hahn (2014), the coefficients, related to the short-run effects of the new leniency program,
range from -24.008 to -1.203. They employ Cox model.

30

Table 1.4 Impacts on Cartel Stability(Survival model with 9-monthshort term)

Variables
LPD
LP-ST9m

(1)

(2)

(3)

(4)

Wei_LP(F)9m

Cox_LP(F)9m

Wei_R(F)9m

Cox_R(F)9m

.001***
(.0001)
.365**
(.203)

-.001***
(.0002)
-.761**
(.308)

.484**
-.918***
(.205)
(.309)
RD
.001***
-.002***
(.0002)
(.0003)
R-ST9m
.504***
-.665**
.599***
-.793***
(.196)
(.265)
(.198)
(.266)
NFR
-.439**
.71**
-.406*
.638**
(.213)
(.293)
(.21)
(.285)
F
-.335***
.526***
-.329***
.511***
(.042)
(.069)
(.042)
(.068)
NF
.007
-.007
.004
-.004
(.013)
(.016)
(.013)
(.016)
LBG
-.068
.124
-.106
.171
(.154)
(.206)
.154
(.205)
PF
-.328**
.387*
-.319
.379*
(.167)
(.223)
(.168)
(.223)
Geo
.798***
-1.072***
.822***
-1.092***
(.12)
(.172)
(.12)
(.171)
IND
Yes
Yes
Yes
Yes
Cons.
8.405***
8.33***
(.532)
(.536)
2
Chi
175.44
179.49
172.77
176.07
n
216
216
216
216
Remark: Standard errors in parentheses. Significance levels: *<10 percent, **<5 percent, ***<1 percent.

ing the magnitudes of the coefficients of LP-ST9m and R-ST9m , in Specifications (1) and (3), upsizing
resources increases cartel durations more than the new leniency program does, while Specifications
(2) and (4) display the opposite results.43 Even though the two policies are significantly effective
in destabilizing cartels in the short run, it is not evident whether the revised leniency program is
superior to the increased resources or not.
According to Brenner (2009) and O. De (2010), cartel life under an anti-cartel policy can be
43 For

example, when hazard ratios are compared in Specification (1) AFT model, the ratio of the LP-ST9m (0.619)
is larger than the one of the R-ST9m (0.515), implying that expanding resources increases the duration of discovered
cartels slightly more than the new leniency program does. However, in Specification (2) Cox model, the hazard ratio
of the LP-ST9m ( 0.467) is less than the one of the R-ST9m (0.514).

31

interpreted as a long run effect of the policy. If a policy is effective, more time spent under the
anti-cartel policy should have a negative effect on the cartel duration. Here, the coefficients of the
variables, LPD and RD can be considered as the long-run effects of the two anti-cartel policies.
The results are not much different from zero, even though they are statistically significant at the
conventional level of significance. One possible reason is that the short-term effects, which increase discovered cartel durations due to the demise of “marginal cartel,” can be offset with the
long-term effects that decrease the cartel durations due to the effectiveness of the anti-cartel policies. Also, there is a possibility that cartel governance may endogenously respond to the external
policy changes. That is, a cartel may develop its alternative collusive agreements and facilitate
sustaining the IC constraint through either learning-by-doing effects or some trustworthy tools,
such as industry associations or ringleaders. If so, an anti-trust policy desisting current collusive
agreements may fail in deterring future collusion in the long run. Last, but not least, it is possible
that a five-year period after the policy changes is too short a time to evaluate long-term effects.
More sample periods might be needed to test it.
In addition, as Hypothesis 3 predicts, the harsher fine rule increasing a fine ceiling as well as
the actual fine amount imposed by the KFTC shows the positively significant impact on disrupting
cartels. These findings are also consistent with the predictions of Harrington and Chang (2009),
that harsher penalties lead industries to the cartelized state for a shorter period. It seems that cartel
firms perceived increased costs for collusive agreements due to the revised fine rules, even though
the average fine rate before 2005 is only 2.42 percent, which is still below the old regime, five
percent of related sales. In fact, the average fine rate increased to 4.55 percent under the revised
fine rules with a ten percent fine rate ceiling. Therefore, it can be understood that cartel firms
consider their expected penalties based on both the actual fines imposed by an anti-trust agency in
the past and the announced guideline for future penalty.
In addition, the coefficients of the number of cartel members, NF, fail to show the conventional
wisdom that more cartel members destabilize collusion. As Levenstein and Suslow (2006) pointed
out, it might be because the number of firms are endogenously determined when they form cartels.

32

The coefficients of the Large Business Group dummy, LBG, are not significant. It implies that
even though one of the cartel firms belongs to a conglomerate, called Chaebol in Korean, it is
not evident whether a firm joining a cartel stabilizes collusion. On the contrary, it even reduces
the duration of discovered cartels, even though the estimates are not significant. The results are
consistent with Robert et al. (2016), where they attribute the results to the overly-broad definition
of the Large Business Group. Or, as Motchenkova and van der Lann (2005) argue, reputation
concerns are more important for large firms so that the large firms have shorter cartel duration.
The price fixing dummy, PF, significantly reduces the duration of discovered cartels. This result is
in accordance with the findings of O. De (2010). Price fixing cartels are more difficult to monitor,
and they are exposed to the risk of entry and cheating than are market sharing cartels. A larger
scope of geographic market seems to increase cartel durations, even though the cartel needs more
monitoring costs under the circumstances. One possible reason is that there exists a monitoring
mechanism to save the costs. Or, multi-market contact can improve firms’ abilities to sustain higher
prices by combining the incentive compatibility constraints. (Bernheim and Winston, 1990). If so,
international cartels and domestic cartels that cover a whole country have longer durations of the
discovered cartels comparing local markets.

33

1.6
1.6.1

Incriminating Evidence Revelation
Hypotheses

This section explores another policy effect of cartel members voluntarily revealing evidence. Another important goal of anti-cartel policies is obtaining hard evidence to succeed in proving a
violation and convicting wrong-doers when a cartel is detected. It does not matter whether an
investigation is initiated by a leniency program or not. The change of external policy environments affecting firms’ incentives can make it more valuable to reveal their information related to
the involved cartel case. For example, introducing a leniency program can lead firms to store more
evidence and submitting it for leniency benefits when detected. Improving detection capabilities or
increasing penalties, absent a leniency program, may decrease the extent of revealing information
because any evidence reported to the authority causes higher cost by helping the authority substantiate its claim. However, on the other hand, if cartel firms think a prosecution or a conviction
is highly unavoidable due to the evidence from a leniency applicant, and an anti-trust agency provides a non-applicant with benefits of additional fine reductions in return for complying with the
authority, they are likely to provide more evidence even under a higher penalty regime. The extent
of revealed information depends on how an anti-cartel policy is designed and enforced. The effect
of anti-cartel policies on revealing evidence is also an empirical issue.
Brenner (2009) finds that the EC’s initial leniency program in 1996 was so successful that cartel
members voluntarily reveal more incriminating evidence. Due to the problem of non-observability
of revealing evidence, Brenner used the total amount of fines per case and investigation periods
as proxies for the revelation. He finds that after introducing the leniency program, the amount
of fines per case increases, but the investigation periods decrease due to the efficiency effects of
the leniency program. However, two proxies might be less representative for revealing evidence.
Regarding amount of fines, it mainly depends more on market definitions, seriousness of violations,
fine rates, etc., rather than revealed evidence. That is, even though cartel members do not cooperate

34

in investigation with an anti-trust agency - if the relevant market size is large, the extent of the
violation is sufficiently serious, or the fine rate in the case is high - the amount of fines can be
large, with less evidence. Also, the investigation periods can increase or decrease at the level of
revealed evidence. For example, if cartel members do not disclose any evidence, an anti-trust
agency has few sources to substantiate a cartel firm’s violations. In contrast, if more complicated
but detailed evidence is disclosed, an anti-trust agency has to spend more time investigating and
finalizing the cartel case, even though it is easier to establish the infringements.
So, here, a fine reduction rate proportional to the extent of investigation cooperation is used
as a proxy for revealing incriminating evidence instead of the amount of fines and investigation
periods. That proxy might be more reasonable in testing the direct impact of anti-cartel policies on
revealing evidence. Hence, the following three hypotheses will be tested.
[Hypothesis 4] The new leniency program with certain full immunity to a first reporter raises
the fine reduction rate for revealing evidence given detection of cartels.
[Hypothesis 5] More resources with upsized organizations under the new leniency program
raise the fine reduction rate for revealing evidence given detection of cartels.
[Hypothesis 6] The increase in a maximum cartel fine rate under the new leniency program
raises the fine reduction rate for revealing evidence given detection of cartels.

1.6.2

Empirical Framework

Here, the OLS regression model based on Brenner (2009) is applied with two modifications. First,
as a proxy for revealing evidence, fine reduction rates due to cooperating with investigation are
used instead of fine amounts and investigation periods. Second, the impact of the other two anticartel policies on disclosing evidence is also considered. In order to test the effect of the policies
on disclosing incriminating evidence, the following equation is estimated.

FRi = α1 + α2 NLPi + α3 ERi + α4 NFRi + α5 F i + α6 lnDCi + α7 NFi + α8 INDi + υi

35

(1.3)

Table 1.5 Impacts on Revealing Evidence(OLS estimation)
Variables
NLP
ER

NFR

(1)
3.932***
(.966)
-.24
(1.026)
4.507***
(1.039)

F
NF

lnDC
IND
Constant
R2
Observations

No
.751
(.613)
.216
204

(2)
2.135**
(1.061)
-.156
(.938)
3.594***
(1.067)
.717**
(.316)
.108
(.087)
.25
(.226)
Yes
-9.28
(3.895)
.231
192

Remark: Dependent variable is fine reduction rates for cooperating in the KFTC’s investigation. Standard errors in
parentheses. Significance levels: *<10 percent, **<5 percent, ***<1 percent.

where FRi is the average fine reduction rates per case in return for cooperating with an investigation, NLPi is a dummy for the new leniency program, which equals one if the new leniency
program is applied to the case i, ERi is a dummy for enhanced resources, which equals one if an
enlarged cartel organization is subject to the case, and υi is a normally distributed error term. Note
that DCi is natural log transformed. The positive α2 , α3 , and α4 support Hypotheses 4, 5, and 6.

1.6.3

Results

In Table 1.5, the estimation results of Equation 1.3 are provided. Only policy variables are included in Specification (1). Then, to compare with the effect of the new fine rule, the annual total
cartel fines imposed by the KFTC one year before the observed cartel begins are added in Specification (2). Also, the number of firms involved, cartel durations, and industry effects are included,
following Brenner (2009).
As Hypothesis 4 predicts, the positive estimates are obtained at a significance level. The results

36

display that the new leniency program revised in 2005 increases the average fine reduction rates
due to revealing evidence by roughly two percent points in the full model. Considering the average
fine reduction rate for evidence revealing is 0.34 percent from 2000 to 2004, the revised leniency
program had a substantial impact on revealing evidence by cartel members. Also, the estimate
results regarding the new fine rule confirm Hypothesis 6. The relevant coefficients range from 3.6
to 4.5, which are larger than those of the new leniency program, implying that the harsher fine
rule induced firms to disclose more incriminating evidence than the new leniency program did. It
may be because the new leniency program affects only leniency applicants, while the new fine rule
affects all cartel members involved in the case. When detected by the KFTC, there might be “raceto-the-court” effects to obtain eligibility for leniency. If a leniency applicant provides the KFTC
with sufficient incriminating evidence, it may be more reasonable that even non-applicants reveal
more evidence to benefit fine reductions in return for cooperating in the investigation. There also
exists another possibility. Therefore, the results should be interpreted cautiously because it may be
that the KFTC has become so generous that it has reduced more fines under the harsher fine rule.
However, testing Hypothesis 5 proves to be a failure, implying that expanding a cartel investigation organization seems to be unsuccessful in getting cartels to voluntarily disclose incriminating
evidence. Perhaps, from the perspective of cartel members, there is almost no difference between a
small cartel division and a large cartel bureau. When detected, only one or at most two staff usually
takes charge of a cartel case, irrespective of the size of the cartel organization. Thus, after a cartel
has already been detected, a larger organization has an impact whether or not cartel members will
reveal information.

37

1.7

Conclusions

This research investigates how a variety of anti-cartel policies affect firms’ cartel activities and
reporting behaviors. The results should be cautiously interpreted because the findings are based on
cartels that were discovered by the competition agency. This cartel analysis cannot avoid inherent
“selection bias” issues. Also, given that cartel durations are also determined by the competition
agency, there is a possibility of “measurement error.” Nonetheless, this study may be meaningful
and humbly contribute to empirical research on cartel behaviors and anti-cartel policy effects.
Many previous empirical papers regarding anti-cartel policies have mainly concentrated on the
impact of a leniency program on cartel deterrence or desistance, even though a leniency program
is introduced or revised with changes of other anti-cartel policies. The results of the current study
show the possibility that some previous results on the impact of a leniency program might have
been overestimated, because the program is not isolated from other anti-cartel policies. This paper
explores the impact of the three important changes in anti-cartel policies of the KFTC around
2005 on cartel stability and evidence disclosure. The three policies are the following: i) a more
generous and transparent leniency program, which gives automatic full immunity to a first reporter,
ii) expanding resources with upsized Cartel Bureau, and iii) a harsher fine rule with a doubled
maximum fine ceiling.
The empirical estimation results support Harrington and Chang’s (2009) theoretical predictions
that a harsher enforcement or policy can disrupt cartels by making them expect the shorter range of
a firm’s profit realization. The results can be categorized into two parts; impact on cartel stability
and impact on revealing evidence. First, all of the three policies have positive impacts on destabilizing cartels. More specifically, the leniency program and resource expansion destabilize cartels
in the short run through increasing the detecting capabilities of the anti-trust agency or reducing
cartel formations in the market. As a result, the duration of discovered cartels increases due to
breakdown of “marginal cartels.” However, the long-term effects of those policies are ambiguous.
Also, harsher penalties disrupt cartels, even though they do not directly affect the detection capa-

38

bility of anti-trust agencies. When determining collusive behaviors, firms consider the future fine
amount based on announced fine rules, as well as actual fines imposed by the KFTC in the past.
Second, the three policies each have more or less a different impact on how much of incriminating
evidence a cartel firm discloses. The doubled maximum fine policy has a larger effect on revealing
evidence than the revised leniency program does. However, we cannot argue that the expanded
organization for detecting cartels has an impact on disclosing evidence. The implication of these
results is that when policy-makers fine-tune anti-cartel policies, they should understand the impact
of different policy instruments on regulating cartels. They first should specify the goal to achieve
and match the effective policies with it in order to achieve better outcomes.

39

REFERENCES

40

REFERENCES

Aubert, C., P. Rey, and W. E. Kovacic (2006). The Impact of Leniency and Whistle-Blowing
Programs on Cartels. International Journal of Industrial Organization, 24, 1241–66
Bernheim, M.D. and D. Whinston (2006). Multimarket Contact and Collusive Behavior. Rand
Journal of Economics, 21 (1990), 1–26
Brenner, S. (2009). An Empirical Study of the European Corporate Leniency Program. International Journal of Industrial Organization, 27(6), 639–45
Chen, J. and J.E. Harrington (2007). The Impact of the Corporate Leniency Program on Cartel
Formation and the Cartel Price Path. The Political Economy of Antitrust, ed. by Vivek Ghosal and
Johan Stennek, Oxford: Elsevier 59–80
Choi, Y.J. & Hahn, K.S. (2014). How does a Corporate Leniency Program Affect Cartel Stability?
Empirical Evidence from Korea. Journal of Competition Law and Economics, 10(4), 883–907
Cox. D.R. (1972). Regression Models and Life-Tables (with Discussions). Journal of Royal
Statistical Society Series, B(34), 187–220
Ellis, C.J. and W.W. Wilson (2003). Cartels, Price Fixing, and Corporate Leniency: What Doesn’t
Kill Us Makes Us Stronger. University of Oregon, Manuscript http://darkwing.uoregon.edu/wwilson
De, Oindrila (2010). Analysis of cartel duration: Evidence from EC prosecuted cartels. International Journal of the Economics of Business, 17(1), 33–65
European Commission Commission (1996). Notice on the non-imposition or reduction of fines
in cartel cases. Official Journal of the European Commission C, 207, 4–6
European Commission (1998). Guidelines on the method of setting fines. Official Journal of the
European Commission C, 9
European Commission (2002). Commission notice on immunity from fines and reduction of fines
in cartel cases. Official Journal of the European Commission C, 45, 3–5
Green, E.J. and R.H. Porter (1984). Noncooperative Collusion under Imperfect Price Information. Econometrica, 52(1), 87–100
Hammond, Scott D. (2004). Cornerstones of an Effective Leniency Policy. Speech International
Competition Network Workshop on Leniency Programs, Sydney, Australia,
Harrington, J.E. and M.H. Chang (2009). Modeling the birth and death of cartels with an
application to evaluating antitrust policy. Journal of the European Economic Association, 7(6),
1400–35
Kaplow L. and S. Shavell (1994). ptimal law enforcement with self-reporting of behavior. Journal
of Political Economy, 102, 583–606
41

Korea Fair Trade Commission (2014). Commission notice on immunity from fines and reduction
of fines in cartel cases. Statistical Yearbook 2014, http://eng.ftc.go.kr/bbs.do
Levenstein, M. C. and V. Y. Suslow (2006). What Determines Cartel Success? Journal of Economic Literature, XLIV, 43–95
Meyer, B.D. (1990). Unemployment Insurance and Unemployment Spells. Econometrica, 58,
757–82
Miller, N. H. (2009). Strategic Leniency and Cartel Enforcement. American Economic Review,
89, 750–68
Motchenkova, E. and R. van der Lann (2005). Strictness of leniency programs and cartels of
asymmetric firms. Discussion paper,
Motta, M. and M. Polo (2003). Leniency Programs and Cartel Prosecution. International Journal
of Industrial Organization, 21, 347–79
Robert M. Feinberg et al. (2016). The Determinants of Cartel Duration in Korea. Review of
Industrial Organization, 48, 433–448
Spagnolo, G. (2000). Optimal Leniency Programs. Fondazione Eni Enrico Mattei working paper,
42
Spagnolo, G. (2004). Divide et impera: Optimal deterrence mechanisms against cartels (and
organized crime). Discussion paper, 4840 Mimeo
Stigler, George (1964). A Theory of Oligopoly. Journal of Political Economy, 72(1), 44–61
Zhou, Jun (2012). Evaluating Leniency and Modeling Cartel Durations: Time-Varing Policy
Impacts and Sample Selection. Bonn University,
Zimmerman, J.E. and J.M. Connor (2005). Determinants of Cartel Duration: A Cross-Sectional
Study of Modern Private International Cartels. Working paper http://ssrn.com/abstract=1158577

42

CHAPTER 2
An Empirical Study of Cartels with
Ringleaders in Korea
2.1

Introduction

A large amount of prior theoretical literature on oligopoly focuses on price leadership (e.g., Stackelberg leader). However, previous economics literature on cartels is less focused on cartel ringleaders despite their significant role in forming a cartel. Price leadership is an economic term related
to price parallelism by using an instrument such as a dominant power or tacit collusion.1 The term
“cartel ringleader” is relatively close to a relevant legal term defined as a “leader,” an “instigator,”
or an “originator” of the formal cartel by anti-trust authorities. More specifically, price leadership
is different from a ringleader of an explicit formal cartel for the following reasons. First, a cartel
ringleader restricts competition by conspiring and instigating explicit collusion, whereas parallel
pricing incurred by a dominant firm or a barometric firm results from a limited market competition. Second, a cartel ringleader is based on an explicit agreement in a real-world cartel detected
by an anti-trust agency, whereas collusive leadership based on tacit collusion uses public preannouncements as a coordinating device.2 Third, a cartel ringleader has been studied mainly in the
context of a corporate leniency program due to the trans-Atlantic policy difference,3 whereas price
1 Since

Stigler (1947) and Markham (1952), three types of price leadership has been studied: dominant firm price
leadership, barometric firm price leadership, and collusive price leadership.
2 Despite the Tobacco Case, where parallel pricing is illegal regardless of being implemented by an agreement or
not, it is generally recognized that tacit collusion in itself is difficult for an anti-trust agency to prosecute. Without
explicit evidence of agreement, illegal firms are not easily convicted, even though there apparently exists parallel
pricing among firms. Refer to American Tobacco Co. v. United States 328 U.S. 781 (1946) and Brooke Group Ltd. v.
Brown & Williamson Tobacco Corp., 509 U.S. 209 (1993)
3 The leniency program offers firms involved in a cartel—which self-report and hand over evidence—either total
immunity from fines or a reduction of fines. Even though both regimes charge harsher penalty to a cartel ringleader,
they treat a cartel ringleader differently. The US DOJ basically does not grant leniency to a cartel ringleader while

43

leadership has been studied to explain firms’ behaviors in an oligopoly market. Fourth, the role of
a cartel ringleader is related to a market allocation cartel as well as a price fixing cartel, whereas
price leadership is mainly intended to parallel pricing. However, from the viewpoint of public
policy, it is commonly important to understand what kinds of market environments determine the
likelihood of those instrumental devices, which supports collusion or affects market structures.
This research focuses on finding adequate answers to the empirical puzzle about cartel organization by using the concept of a cartel ringleader. Conventional wisdom asserts that collusion is
more sustainable with fewer firms and less asymmetries in the industry. However, we can easily
observe cartels with relatively many firms and their considerable size asymmetries. There are a few
possible reasons for this discrepancy between theoretical predictions and empirical results. Among
these reasons is sample selection bias. Most empirical studies on cartels use policy-dependent data,
which are determined by an anti-trust agency. The stronger cartel with few and symmetrical features is too secretive to be discovered by an anti-trust agency. Strong enforcement of anti-cartel
policies drives those cartels underground. Another reason may be that theory and empirics use different types of collusion; theoretical literature usually focuses on tacit collusion, whereas empirical
analysis uses a set of explicit cartel data.
Toward addressing this puzzle, this paper inquires deeper into the second possibility. Explicit
collusive agreements can support a cartel by reducing distrust among firms. Also, a cartel can
develop an organizational mechanism for monitoring and enforcing the agreement to offset the
destabilizing forces resulting from more firms and greater asymmetries. If this mechanism is effective, we can observe cartels under certain environments where it would be difficult to sustain a
the EC symmetrically treats a cartel ringleader with other regular firms. In the US, in order to apply for amnesty or
leniency, the leniency policy requires that “the corporation did not coerce another party to participate in the illegal
activity and clearly was not the leader in, or originator of, the activity” (US Department of Justice, Antitrust Division,
Corporate Leniency Policy at A.6). The EU had adopted the discriminating regime following the US, but adopted a
non-discriminating leniency program in 2006 and Canada changed in 2010. The Korea Fair Trade Commission revised
its leniency program from the US-style to the EC-style regime in 2005. In a unique situation, China’s State Administration for Industry and Commerce (SAIC) has a leniency policy with a discriminating regime (applying to “organizers
of monopoly agreements”) for non-price monopoly agreements while the National Development and Reform Commission (NDRC) has a non-discriminating regime in its leniency policy related to price-monopoly agreements. See Ye
(2014). For more details for US and EU policy on leniency programs, refer to “Corporate Leniency Policy” of the DOJ
(https://www.justice.gov/atr/corporate-leniency-policy) and “Commission Notice on Immunity from fines and reduction of fines in cartel cases” (2006, http://eur-lex.europa.eu/legal-content/EN/ALL/?uri=CELEX:52006XC1208(04)

44

cartel otherwise. Levenstein and Suslow (2006) suggest that organizational responses such as an
industry association can help firms avoid the “cartel problems.”4 That is, when cartel firms confront difficult situations where the incentive compatibility constraint becomes dissatisfied for some
reason, the cartel strategically responds to the environment to alleviate the problem of cheating.
This research focuses on the role of a ringleader as an organizational response in a cartel and finds
relevant policy implications. A cartel ringleader can be one of the important factors in determining
a cartel structure if it plays as an organizational solution to overcome cartel problems. The results
of this study find that a cartel ringleader is more prevalent under circumstances where it is easier
for cartel firms to deviate from collusion—with more firms, greater asymmetries and less product
homogeneity.
This paper extracts some stylized facts about cartel ringleaders in Korean cartel cases, using a
data set of 327 finalized cartel cases filed for violations of Article 19 (1) of the MRFTA (Monopoly
Regulation and Fair Trade Act) between January 1, 2000 and December 31, 2010.5 Here, a “cartel ringleader” is defined as a firm that instigates or leads a cartel based on its role stated in the
case documents of the Korea Fair Trade Commission (KFTC). A cartel ringleader is basically
categorized into two types: an organizational ringleader that instigates collusive behaviors by approaching other firms or convening a meeting for collusive agreements; or an aggressive ringleader
that plays an active role in determining collusive plans regarding price, market allocation, bidding,
etc. This research contributes to the literature by empirically showing that a cartel strategically
chooses its structure in response to market characteristics. This is one of the first papers about
cartel ringleaders using the 327 Korea cartel samples. The results confirm the argument of Levenstein and Suslow (2006) that an alternative monitoring mechanism can be an organizational
solution to alleviate the issue of cheating. This study also confirms the findings of Davies and De
(2013), which analyzes the effects of market characteristics on the incidence of a cartel ringleader,
using the 89 European Commission’s (EC) cartel samples. The abundant organized facts about
4 Osborne (1976) argues that enforcing a cartel has inherent difficulties called “cartel problems” in locating firms’
contract surface, choosing an agreeable focal point on it, and monitoring possible deviation.
5 Some major documents filed during the periods are available for download from the Commission website,
www.eng.ftc.go.kr.

45

a ringleader include how frequently they exist across cartel types (e.g. price-fixing, bid-rigging,
etc.), the roles of ringleaders (i.e., organizational, aggressive, or both), the frequency of alternative
monitoring mechanisms (i.e., industry association), and types of punishment/compensation mechanisms (e.g., side payments, bidding rotation, etc.). Three main estimations are conducted: a linear
probability model (LPM) and two index models, a probit and a logit forms.
Three significant results are found. This study confirms that ringleaders are more likely to be
identified in cartel structures with more cartel firms, greater asymmetries in size measured by the
normalized Herfindahl-Hirschman Index (HHI), and less homogeneity of cartel products.6 Using
a multinomial logit shows that the choice of aggressive ringleaders is pronounced, compared to
cases of organizing-only ringleaders or no ringleaders. Last, this study finds ringleaders are less
likely chosen in cartel structures when other alternative monitoring mechanisms such as an industry
association are involved in cartels. More specifically, when industry associations already play
administrative roles in cartels, an organizing-only ringleader is preferred less than an aggressive
ringleader. The cartel with involvement of the industry association tends to delegate an aggressive
role to a ringleader in order to additionally enforce or monitor the cartel. The remainder of the
paper is organized as follows. The next section reviews the previous literature. Section 3 describes
the data set and distills the descriptive statistics and stylized facts about ringleaders in Korean cartel
cases. In Section 4, empirical results are displayed, which is followed by a conclusion.

6 HHI is an index measuring the degree of competition or concentration in the industry,

which is defined as the sum
of the squares of the market shares of the firms within the industry. However, the normalized HHI is used because it is
more suitable for symmetry or equality, while HHI is more suitable for concentration. The formula of the normalized
HHI is H*=[H-(1/N)]/[1-(1/N)] if N>1, and where H is HHI and H*=1 if N=1, where N is the number of firms in the
industry, and market shares are calculated by using cartel shares.

46

2.2

Literature Review

Levenstein and Suslow (2006) point out that a cartel should continuously develop a beneficial
structure to all cartel firms in order to sustain their collusive agreement through dealing with three
key problems: coordination, cheating, and entry. If an incentive compatibility constraint is dissatisfied, like Stigler (1964) indicates, a cartel becomes exposed to the incentive to cheat. Many
studies have examined the market characteristics that affect the incentive compatibility structure of
a cartel member. These studies generally argue that more firms in collusion can decrease stability,
other factors being equal (Tirole, 1988).7 Intuitively, a coordination problem leads to more costly
haggling, and makes it more difficult to maintain or enforce a cartel, as there are more firms in
the industry. Asymmetries in the characteristics of firms also affect the firms’ choices of whether
to collude or not. The impact of asymmetry on cartel stability varies depending on model specification.8 However, the theoretical literature has usually recognized that asymmetry resulting from
multiple sources tends to destabilize collusion (Ivaldi et al, 2003).9 Firm-level asymmetry can
lead to asymmetric market shares, which make it more difficult for cartel participants to choose an
agreeable focal point in the contract surface due to differential preferences.
However, contrary to theoretical expectation, much empirical research based on real-world formal cartel data displays more or less contradictory results (Posner, 1970; Dick, 1996a; Ganslandt
et al, 2012).10 A few possible reasons may account for these divergences, among them, sample
selection bias which is unavoidable in empirical research on cartels. Only cartel data detected and
7 The

threshold discount rate of expected future profits necessary to sustain collusion increases as the number
of firms in the industry increases (Tirole 1988). Ralph M. Bradburd and MEad A. Over Jr. (1982) argue that the
organizational costs of both cartel formation and cartel maintenance are expected to increase with the number of firms
in the industry.
8 For example, product homogeneity may have two conflicting effects on cartel stability; the homogeneity makes
firms coordinate more easily while product differentiation facilitates collusion by decreasing the benefits to deviating.
9 Motta (2004) argues that collusion is easier to sustain when firms in the market have a more similar distribution
of assets because IC constraints become more relaxed. According to Ivaldi et al. (2003), the diversity of cost structures
can make collusion unstable for two reasons: it exacerbates a coordination problem due to difficulties in finding an
agreeable focal point; a low-cost firm can gain more from undercutting, and it is less threatened by a high-cost firm’s
potential retaliation. Compte et al. (2002) argue that asymmetries in capacity can make collusion more difficult to
sustain when the aggregate capacity is limited.
10 Posner (1970) and Dick (1996a) find cartel duration increases with the number of participants. Ganslandt et al.
(2012) discuss how a certain amount of asymmetry within cartels may actually facilitate collusion.

47

prosecuted by an anti-trust agency can be used for research with the implication of non-random
sampling. It is difficult to detect more strongly sustainable collusion with fewer firms and/or symmetrical features. Another reason may be that most theoretical literature on collusion assumes tacit
collusion, which is possible without formal agreement. However, in the real world, an anti-trust
agency detects and prosecutes a formal cartel only when sufficiently established evidence of an explicit agreement has been provided. This explicit agreement can make cartel firms’ IC constraints
less binding even in the market structure with many firms and considerable asymmetries among
them.
Monitoring mechanisms can also play a similar role in organizing cartels. Levenstein and
Suslow (2006) state that the most successful cartels develop organizational mechanisms (e.g., industry associations) that accommodate fluctuations in the external environment. They assert that
firms probably make such a significant investment in building a more complicated collusive organization and in developing organizational skills in an effort to avoid the issue of cheating that
undermines cartels. The endogenously changing structure of cartels may have an effect on market equilibrium. For example, an industry association can alleviate inherently unavoidable “cartel
problems” suggested by Osborne (1976) in an un-concentrated industry. An endogenously chosen
cartel structure possibly allows marginal firms to survive and to lower market concentration. If so,
we observe that collusion is more sustainable even in the less concentrated industry that would,
otherwise, be highly concentrated. Using the EC cartel data set, Davies and De (2013) focus on
the role of a cartel ringleader to empirically show this organizational response of a cartel structure.
They find a cartel ringleader instigates explicit collusive agreements or plays an important role in
setting up the agreements. Ringleaders are more prevalent in situations where market features are
difficult to sustain collusion,—i.e., more cartel firms and greater asymmetries in firm sizes.
As discussed above, previous theoretical research mainly pays attentions to a price leadership,
rather than a cartel ringleader. Following Stigler (1947) and Markham (1952), the previous literature examined three types of price leadership. The first two, a most efficient dominant firm type
and a representative barometric firm in the industry, commonly lead the price by moving first, and

48

then other firms in the industry set their price and quantity by paralleling the price determined by
the leader. However, both of these cases are, especially when an anti-trust agency deals with a
cartel case, legally less meaningful because the resulting equilibria are consequences of restrictive competition rather than adopting a strategic conspiracy to maximize their joint profit. That
is, the two types of the price leadership are not intended to avoid price competition among rivals,
but are instead an inevitable consequence of the industry’s structure. Thus, effective corrective
orders/remedial measures, which can be taken by an anti-trust agency to cease these behaviors, are
limited. These types of price leadership can be corrected through reforming the industry structure,
rather than correcting an individual firm’s behavior.
Different from these two types, collusive price leadership comes from tacit adoption of a price
leader to avoid price competition.11 Markham (1952) argues that collusive price leadership assuming overt agreements with the implication of public pre-announcement of price is more likely
under certain market features, such as fewer firms, restricted entry, product homogeneity, price
inelasticity of demand, and similar cost structures among firms in an industry. However, considering a structural response in organizing collusion, empirical results based on real-world cartels
exhibit findings that contradict these predictions. Rotemberg and Saloner (1990) argue that collusive price leadership facilitates collusion under asymmetric information; and if information is
sufficiently asymmetric, the less informed firm prefers to follow the better-informed firm, so that
the leader can emerge endogenously. Mouraviev and Rey (2011) show when firms face asymmetric
cost structures or produce differentiated products, price leadership can enhance the profitability of
collusion. In case of asymmetric cost, the less efficient firm acts as a leader because the incentivecompatible constraint of an inefficient firm is binding without collusion. Ishibashi (2008) shows
that although setting a price early in a homogeneous good capacity-constrained setting is disadvantageous per se, a large firm has the incentive to move first, in order to demonstrate its commitment
not to deviate. However, as already discussed above, these price leadership types are different from
11 Rotemberg and Saloner (1990) define “collusive price leadership” as a situation where “one of the firms announces a price change in advance of the date at which the new price will take effect, and the new price and date are
swiftly matched by the other firms in the industry.”

49

an explicit formal cartel with ringleaders.
Some papers have explicitly dealt with cartel ringleaders using EC data. Reviewing EC cartel
data, Ganslandt et al (2012) find some case evidence of ringleaders in cartels with size-asymmetric
firms. They build a theoretical model that predicts that a certain amount of asymmetry within
cartels may facilitate collusion due to indivisible fixed costs of collusion borne by a leader. Grout
and Sonderegger (2005) observed considerable heterogeneity in the market shares held by cartel
members from 2002 to 2007, and they identified a ringleader in 10 out of 43 EU cartel cases.
Davies and De (2013) used 89 European cartels between 1990 and 2008 to empirically estimate
the frequency of ringleaders under the features of more firms and greater asymmetries in a cartel.
They found a ringleader plays a critical role in reducing “cartel problems” under situations of many
firms exhibiting substantial size asymmetries.

50

2.3

Data and Stylized Facts

This paper finds some stylized facts about ringleaders in Korean cartel cases using the data set of
327 cartel cases finalized between January 1, 2000 and December 31, 2010 filed for violations of
Article 19 (1) of the MRFTA (Monopoly Regulation and Fair Trade Act). All of the case documents
were read closely to collect relevant information. The detailed information largely covers marketand firm characteristics. Observation i represents each cartel case.
It is natural to ask how a ringleader cartel and a no ringleader cartel can be differentiated
because any cartel is triggered by a certain form of involvement of at least one firm or its employer.
Here, no ringleader cartel can be regarded as an alternative to a default where all cartel members
share the legal responsibilities. In this default case, all cartel members should be fined equally,
which implies that they are fined in proportional to their sales. In ringleader cases, all members
of the cartel structure, willingly or sometimes reluctantly, agree to delegate the role of leader to a
ringleader(s) who can be singled out.
The existence of a ringleader can be drawn by carefully reading the section of the “detailed
process of agreement and coordination” or “calculating the amount of fines” in the case documents. Following the guidelines for calculating fines, the KFTC usually charges ringleaders more
severe penalties.12 However, not all cartel cases formally define a ringleader in the documents.
Therefore, a “ringleader” is defined in this paper as any firm in a cartel case that plays an important role in organizing the cartel. In order to define a cartel ringleader, this paper follows the
EC’s rule that distinguishes two broad types of behaviors in a cartel’s decisions—“instigation”
and “leadership.”13 While not all the anti-trust agencies always formally define the term, “cartel
ringleader,” this standard is widely used to categorize various behaviors organizing cartels. Note
12 An actual fine amount imposed by the KFTC is determined following several steps.

First, a basic fine is calculated
based on sales amounts, and then the basic fine is adjusted according to certain factors. A final fine is determined after
leniency discount is applied if needed. Basic Fine = Related sales of the cartel product x Fine rates (0 - 10%) where
fine rates are determined by KFTC according to the seriousness and importance of violation. Adjusted Fine = Basic
Fine x (100% - Fine reduction rate + Fine addition rate) where a fine can be reduced if firms cooperate in investigation
(0 - 20%) or involvement is limited and it can be increased if firms are ringleaders or repeat violations. For more
details, refer to http://eng.ftc.go.kr/bbs.do
13 European Commission (2006), Case T-15/02 BASF v Commission Summary of Judgment, March 15, 2006.

51

Table 2.1 Incidence of (ringleader involved) cartel case 2000-2010
00 01 02 03 04 05 06 07 08 09 10 Total
Total 30 22 34 14 23 30 31 28 39 38 38 327
RL∗ 6 3 5 7 8 8 12 9 17 12 19 106
*”RL” means the number of cartel cases with at least one ringleader.

that a ringleader, as discussed above, is not limited to price fixing cartels, unlike price leadership, but related to all cartel types. Davies and De (2013) also applied this approach to define
ringleaders. First, an “instigator” is involved in the initial birth and/or subsequent enlargement of
the cartel. In this study, an alternative term, an “organizational/organizing ringleader” is used for
those behaviors. This type of ringleader plays an relatively moderate or mild role in organizing
cartels by approaching potential members or convening a cartel meeting in the beginning of cartel
coordination. Second, “leadership” implies an active or leading role in determining/suggesting a
specific cartel plan or in designing a penalty/compensation mechanism. Likewise, an alternative
term, an “aggressive ringleader” is used in the current research. The roles of these types, for the
most part, are associated with a price/sales plan or bidding plan such as complementary- or rotating
bidding. Sometimes, these ringleaders set up a punishment mechanism (e.g., monetary penalty)
against the violator of the agreement, or a compensation mechanism (e.g., side-payment) for the
firm that sacrifices its individual share of the joint profits in the cartel agreement.
As shown in Table 2.1, a single or multiple ringleader(s) are identified in 106 out of the 327
cases, which accounts for about 32% of cases. This frequency of the existence of a ringleader
apparently seems to be greater than that found in previous studies, where 10 out of 43 (23%)
ringleader cases in Ganslandt (2012) and 19 out of the 89 (21%) cases in Davies and De (2013)
are ringleader-related in EC cartel cases. However, as Davies and De (2013) pointed out, it is
probable that the actual incidence of ringleaders in the EU cases would be higher than their finding.
Anti-trust agencies such as the EC officially identify a ringleader only when it feels sufficiently
confident to name a ringleader explicitly with established evidence. Table 2.1 shows a sharply
increasing trend of ringleader cartels since 2005. It might be because that the KFTC introduced
the new leniency program around that time, which gave more benefits to ringleaders than prior to

52

the policy change. The new policy dummy will be included in the rigorous analysis to control for
the impact of the policy modification.
Table 2.2 Comparison of cartels with/without ringleader

Total cases
Price fixing cases
Fined cases
Leniency filed cases
Incidence of industry association
Average number of firms
Average cartel share of the largest firm
Average range of market share
Average HHI
Average normalized HHI
Average fines∗ (million KRW)
Average fine rate (percent)
Average cartel duration (month)†

RL cartels

Non-RL cartels

106
87 [82.1%]
82 [77.3%]
38 [35.9%]
9 [8.5%]
6.89

221
176 [79.6%]
125 [56.6%]
35 [15.8%]
26 [11.8%]
5.78

(4.71)

(4.41)

50.61

50.3

(22.22)

(20.2)

44.12

38.31

(21.89)

(20.6)

.39

.39

(.22)

(.19)

.23

.18

(.21)

(.17)

10,116

10,175

(25,272)

(25,469)

3.84

4.27

(2.37)

(4.39)

15.87

17.54

(23.39)

(22.68)

Remark: Standard deviations in parentheses, and the proportion of certain types of cartel cases to total
ringleader cartels or total non-ringleader cartels are expressed in the bracket.
*Fines before the leniency scheme is applied are averaged and expressed in million KRW, where 1,000
KRW can be approximately conversed into 1 US dollars.
† The average cartel duration is obtained execpt for bid-rigging cartels.

Table 2.2 exhibits descriptive statistics that compare ringleader-related cartels with non-ringleader
cartels. The number of participating firms in a cartel is at least two, and the maximum number of
firms is 25. The average number of cartel firms is about 6.14. The longest cartel duration in
the sample set is 110 months. The average one is about 17 months. Pertaining to cartel types
53

and whether fined or not, there is not much difference regardless of the existence of a ringleader.
However, a ringleader-based cartel is more likely to apply for leniency. About 36% of ringleader
cartels is leniency filed, whereas only 16% is for non-ringleader-involved cartels. Note that 77
cases conditional on being ringleader-involved are subject to post-2005, when the new leniency
program allowed a ringleader to apply for leniency benefits, such as fine reduction and immunity.
Even though this new anti-cartel policy might have had a positive impact on destabilizing cartels,
it seems to have been highly associated with increasing ringleader involvement with cartel formation. Interestingly, an industry association is slightly less involved with a cartel in the presence of
a ringleader. It implies that an industry association and a ringleader can be substitutive monitoring
mechanisms when a cartel is formed or enforced. Also, ringleader-based cartels have on average more firms and greater asymmetries in a size. Note that a range of cartel share14 between the
largest and smallest firm, and normalized Herfindahl-Hirschman Index (HHI)15 are used as proxies
for asymmetry. Higher values imply greater asymmetries in firm sizes for both variables.
Table 2.3 displays how many ringleaders are involved in a cartel and how large they are.
The average number of cartel firms is about 6.14, which is a little less than the average in 7.25
of Hay and Kelley’s (1974) study. On average, 1.68 cartel ringleaders are involved in cartels. Out
of the 63 cartel cases with only two cartel firms, 47 cases (75%) are not involved with a cartel
ringleader. It seems that a cartel structure with fewer firms is less likely to develop a special
monitoring mechanism in order to sustain collusion. Rigorous analysis is needed to confirm this
observation. It is easy to see more than one ringleader. Among the 106 ringleader-involved cases,
multiple ringleaders appear in 41 cases. About 40% of ringleader-involved samples shows that
14 Here,

a cartel is assumed to be all inclusive, even though it is still possible that some minor firms do not join
the cartel. Cartel shares, the amount of sales related with goods and services in the cartel case, is used to calculate
market shares of the related firms for two reasons. The first reason is inaccessibility of market share information. Even
though some decision documents of the KFTC include market shares of the cartel firms in the industry, most public
documents do not. Also, it is reasonable that if not all inclusive, the cartel is more likely to collapse.
15 HHI is an index measuring the degree of competition or concentration in the industry, which is defined as the sum
of the squares of the market shares of the firms within the industry. However, the normalized HHI is used because it is
more suitable for symmetry or equality, while HHI is more suitable for concentration. The formula of the normalized
HHI is H*=[H-(1/N)]/[1-(1/N)] if N>1, and where H is HHI and H*=1 if N=1, where N is the number of firms in the
industry.

54

Table 2.3 Number and size of cartel ringleaders
6.14

Average number of firm

(4.53)

Average number of RL

1.68
(1.09)

Average cartel share ratio of 2nd largest to largest firm

55.09
(26.87)

Average cartel share of the largest RL

41.57
(27.99)

Multiple RL cases
Largest firm RL cases∗
2nd largest firm RL cases∗
Smallest firm RL cases

41
57 [53.77%]
14 [13.21%]
9 [8.49%]

Remark: Standard deviations in parentheses.
*In the case of multiple ringleaders, the largest ringleader among them is counted. The proportions of
ringleader-related cases are expressed in the bracket where total number of ringleader-related cases is 106.

multiple ringleaders play important roles in cartel structure.16 In three cartel cases with 25 firms,
which is the largest in the data set, only one case is ringleader-related, with three ringleaders.
However, an industry association is involved with the cartels in other two cases. It is probable that
an industry association plays an alternative role in organizing cartels instead of a cartel ringleader.
The cartel share of the second largest firm is on average 55% of that of the largest firm, which
implies the existence of size asymmetry. In Ganslandt et al. (2012), market share of the second
largest firm is about 70%. Average cartel share of the largest ringleader in the cartel is about 42%.
Not surprisingly, cases where the largest firm in the cartel is the ringleader accounts for about 54%
of ringleader-involved cases. This finding is in a line with Ishibashi’s (2008) finding, where a firm
with larger capacity is suitable for the role of a leader. In contrast, nine cartel cases are related to the
smallest ringleader. It is possible that the incentive-compatible constraint of small—presumably
the least efficient—firms binds without the explicitly collusive agreement (Mouraviev and Rey,
2011).
Table 2.4 summarizes ringleaders’ roles across their types. As discussed above, the generally
recognized rule is used to categorize a ringleader. Two broad types of behaviors are distinguished in
16 The

number of multiple ringleaders in the current research range from two to five. In Davies and De (2013), the
maximum is seven.

55

Table 2.4 Incidence of ringleader by types and roles
Types
Organizational RL (Organizing-only∗ )

Roles
approaching potential firms
convening initial meeting

Aggressive RL†

Frequencies

72 (28)
46 (11)
26 (17)
78†
30
8
42
35
2
8
3
3
5

determining price plan‡
determining sales plan
determining bidding plan
- complementary bidding
- rotating bidding
- subcontracting
- side-payment
penalty scheme
set-up joint firm
* Many organizational ringleaders are also involved in aggressive roles. Out of the 72 cases, only 28 cases
are subjected to organizing-only roles.
† Many ringleaders engage in more than one role. For example, out of the eight ‘determining sales plan’
cases, four cases are also related to ‘determining price plan.’ Four cases in ‘subcontracting’ are also associated with ‘complementary bidding.’ It also happens in ‘penalty’ and ‘set-up joint firm.’
‡ This role is only associated with non-bid-rigging cartels.

the KFTC’s decision documents. An organizational ringleader plays a role in organizing a cartel;
however, its role is limited to a weakly initiating role such as approaching potential members
or convening cartel meetings in the beginning of cartel coordination. In contrast, an aggressive
ringleader plays a more active or leading role in the cartel by having a larger influence on the cartel
by outlining a specific cartel plan or a penalty/compensation mechanism. In the 72 cartel cases in
the sample set, single or multiple ringleaders are involved with organizational roles. However, 44
cases out of the total are also associated with aggressive roles. Thus, only 28 of the cases are subject
to organizing-only. Second, 78 cartel cases show the existence of an aggressive ringleader(s).
This type displays a variety of activities. Most of them are related to leading roles in designing
specific cartel plans. Note that many ringleaders engage in more than one role. For example, four
‘determining sales plan’ cases out of eight are also related to ’determining price plan.’ Four cases
in ‘subcontracting’ are also associated with ‘complementary bidding.’ It also happens in ‘penalty’
and ‘set-up joint firm’ cases. Only three cases of penalty scheme—which are mostly monetary

56

Table 2.5 Incidence of ringleader types across agreement types
RL involved
Types

No RL

Organ. only

Aggres.

Subtotal

Total

Not bid-rigging
160
22
35
57
217(66.4%)
∗
127
20
31
51
178(54.4%)
- Price-fixing
- Others†
33
2
4
6
39
Bid-rigging
61
6
43
49
110 (33.6%)
∗
49
4
32
36
85(26.0%)
- Price-fixing
- Others†
12
2
11
13
25
Total
221 (67.6%)
28 (8.6%)
78 (23.8%) 106 (32.4%)
327
* The typology is based on the application of Article 19 (1) of the MRFTA, which implies that when firms’
activities are related to price-fixing and market allocation together, this case is counted as ‘price-fixing’ case.
† ‘Others’ include non-price-fixing cartels such as market allocation, limit or control production or investment, etc.

penalty—are found. However, it is still possible that non-monetary penalty schemes such as a
price war are implicitly agreed upon among cartel firms, and the related evidence is not discovered
by the KFTC.
Table 2.5 reports the incidence of ringleader types across agreement types. A bid-rigging
cartel has different features than a non-bid-rigging cartel, in that the latter usually has longer cartel
duration, while the former is implemented only for one time bidding.17 As we can predict, pricefixing cartels are the most prevalent one in the real world. Out of the 327 cartel cases, 263, or 80%,
are cases of price-fixing.18 Note that following the KFTC’s method to classify violation types,
when firms’ activities are related to both price-fixing and market allocation, this case is counted
as a ‘price-fixing’ case. In bid-rigging cartels, aggressive ringleaders are more likely to exist than
organizing-only ringleaders do. In contrast, in cases neither price-fixing nor bid-rigging cartels,
where market allocation types constitute the majority, a ringleader is less likely to appear. This
situation raises the possibility that a market allocation cartel is less costly than a price-fixing or a
bid-rigging cartel in the context of monitoring cheating. This finding is in line with Davies and
17 Bid-rigging

conspirators can have a long period of a collusive agreement to share multiple one-time projects.
However, bid-rigging still has different features from non-bidding cartels from the perspective of impacts on the
market.
18 Some price-fixing cartels are combined with market sharing behaviors. Harrington (2006) shows many cartels
are related to price-fixing or bid-rigging which can be combined with a market allocation scheme.

57

Table 2.6 Incidence of products homogeneity across RL types
Homogeneous products

criteriona ∗
criterionb †

No RL

Ringleader involved
Organ.-only Aggres. Subtotal

134
87

10
9

48
27

58
36

Total

192
123

* Goods and services are considered homogenous if they are intermediate ones (Jacquemin et al., 1981).
† Based on the KFTC’s identification strategy, goods and services are considered homogeneous if the KFTC
identifies the related goods and services homogeneous ones, the proportion of raw material in cost structure
is substantially higher, or only price rather than quality or brand is important factor to decide demand.

De (2013). However, as stated before, it should be noted that many aggressive ringleader cases are
also instigated by organizing a process by at least one ringleader.
Table 2.6 displays how product homogeneity differs across ringleader types. Here, two different criteria for classifying homogeneous/heterogeneous products are applied. First, following
Jacquemin et al. (1981), criteriona classifies final goods and services as heterogeneous, while
classifying intermediate ones as homogenous. Second, criterionb is based on the KFTC’s identification strategy. Goods and services are considered homogeneous if the KFTC classifies the related
goods and services as homogeneous. However, in cases the proportion of raw materials in cost
structure is substantially higher, or only price rather than quality or brand is important factor to determine demand, the relevant products are considered homogeneous. Interestingly, in both criteria,
a ringleader is not involved in about 70% of homogeneous product cartel cases.
Table 2.7 Incidence & roles of industry associations across RL types
Monitoring mechanisms
Industry association
- administrative role
- aggressive role

No RL

Ringleader involved
Organ.-only Aggres. Subtotal

68
32
36

2
1
1

10
8
2

12
9
3

Total

80
41
39

Table 2.7 shows how often industry associations as alternative monitoring mechanisms appear
in cartel structures across ringleader types. In 80 out of the 327 cartel cases, industry associations
play important roles in organizing or controlling cartels. Only 12 out of these 80 cases (15%)
are ringleader-involved. It is easily observable that an organizing-only ringleader is less likely to
appear when an industry association is pre-involved with the cartel.
58

2.4
2.4.1

Empirical Results
Empirical Frameworks

First, the linear probability model (LPM) is first estimated using OLS, where the binary response
y is specified as

pr(RLi ) = β1 + β2 N i + β3 Asymi + β4 Homoi + β5 Assoi + Zi β + εi

(2.1)

where pr (RLi ) is the probability that cartel case i has a ringleader(s), measured using a binary
variable with one denoting existence of a ringleader. Variable N is an maximum number of cartel
firms over its duration. Variable Asym is the proxy for asymmetry in cartel size, which is measured
by the normalized HHI.19 Variable Homo is a dummy for products homogeneity where it equals
one, if products and services related to cartel cases are considered to be homogeneous. Variable
Asso is a dummy indicating the presence of an industry association, which operates as an alternative monitoring mechanism. Z includes the variables post-2005 and IND. Post-2005 is a dummy
variable with one, if the case is subject to the new leniency program revised in 2005. The new
program allowed a ringleader to be eligible for leniency benefits. IND is an industry fixed effect.20
The error term, εi follows a normal distribution. Positive β2 and β3 , and the negative β4 and β5 ,
are expected.
Second, the equation will be estimated with index models of probit and logit forms to consider changing the marginal effects of the explanatory variables on the probability of ringleader
existence.

pr(RLi ) = f (N i , Asymi , Homoi , Assoi , Zi )
19 Unlike

(2.2)

Davies and De’s (2013) study on EC data where they use a ‘range’ of firm size in a cartel, a different
proxy for market asymmetry in size is used here. However, for robustness, a ‘range’—the difference of market shares
between largest and smallest firm in a cartel—is also used. The results are almost similar.
20 The industry dummy is based on the Korea Standard Industry Classification (KSIC).

59

where the function f is the CDF of the standard normal distribution for the probit model while
it is the CDF of standard logistic distribution for the logit model. The positive coefficients for N
and Asym, and the negative coefficients for Homo and Asso are predicted.
Also, the equation will be re-estimated as a multinomial logit to test how the effects of those
cartel determinants on ringleader incidences differ across ringleader types. The cartel chooses
one of the three alternatives, no ringleader, an organizational only ringleader, or an aggressive
ringleader:

pr(ki ) = pr {U (k) > U (l)}

(2.3)

for all other l=k where U (k) = f (N i , Asymi , Homoi , Assoi , Zi ).

2.4.2

Empirical Results

In Table 2.8, full samples are included for analysis, making no distinctions across cartel types or
ringleader types. Specifications, (1), (3), and (5) include only main explanatory variables, while
(2), (4), and (6) contain additional anti-cartel policy change effects and industry effects. The results
are significantly consistent. Both variables N and Asym are positively significant at conventional
levels. This finding implies that a cartel structure is more likely to choose ringleader-based structure in a situation where a cartel problem is prevalent. Conventional theory predicts that features of
“more firms” and “greater asymmetry” make collusion unstable by leading cartel firms to deviate
from collusion, all other things being equal. However, the empirical results show that it might be
possible for cartel firms to overcome or at least alleviate these situations by choosing a ringleader
structure. This sort of ringleader-based structure can make the IC constraint less binding by increasing trust or decreasing coordination cost among firms.
First, Specification (2) shows that one unit increase in a participating firm enhances the probability of ringleader-based cartel structure by 0.028. As the number of firms increases, a cartel
chooses ringleader-based structure to sustain collusion. In order to consider the changing marginal
effects, probit and logit models are also estimated. By using the rough rule of thumb, we can divide
60

Table 2.8 LPM, Probit, and Logit Estimates of Ringleader-based Cartel
(Full samples)
Dependent Variable: Ringleader involved
LPM (OLS)
short
Variables
(1)
(2)long
Number of firms
.03***
.028***
Asymmetry
Homogeneity
Industry Association

Cons.
Obs.

LR

chi2

Correct Predictions(%)

ll
Peudo R2†

Logit (MLE)

(4)long

(5)short

(6)long

.09***

.095***

.15***

.16***

(.007)

(.007)

(.021)

(.024)

(.036)

(.04)

.35***

.34***

1.015***

1.129***

1.66***

1.924***

(.148)

(.147)

(.431)
-.271*
(.163)
-1.065***
(.222)

(.468)
-.317*
(.189)
-1.008***
(.243)
.888***
(.216)

(.709)
-.45*
(.27)
-1.87***
(.412)

(.786)
-.568*
(.322)
-1.718***
(.433)
1.49***
(.374)

-.089

-.085

(.055)
-.337***
(.066)

(.059)
-.276***
(.067)
.27***
(.068)

No
.243***

Yes
.134

No
-.73

Yes
-1.138

No
-1.2

Yes
-1.868

(.069)

(.102)

(.204)

(.339)

(.338)

(.58)

290

289

-

-

290
38.9

287
68.76

290
39.51

287
68.89

.124

.214

69.7

73.5

70.0

73.2

-169.22
.103

-152.96
.184

-168.92
.105

-152.9
.184

Post-2005
Industry effect

Probit (MLE)

(3)short

Remark: Standard errors in parentheses. Significance levels: *<10%, **<5%, ***<1%.
†The usual R-squared is reported for LPM.

the probit estimates by 2.5 and the logit estimates by 4, so that these estimates can be compared
to the LPM estimates. The scaled estimate on the variable N is 0.038 for probit and 0.04 for logit.
The values of the conditional marginal effects at means for the variable N are 0.034 for both probit
and logit.21 Observably, there is not much difference in the scaled estimates. Interestingly, when
the squared term of the variable N is added in the OLS regression, the probability of ringleader
incidence peaks at N=16, and then decreases.22 A possible reason for this finding is that too many
firms in a cartel—even in the case of an explicit collusion—exposes the cartel to being detected by
an anti-trust agency authority. That is, as the number of firms in a cartel increases, that cartel, to
some extent, can alleviate issues of cheating by choosing a ringleader structure. However, above
21 The

results are not displayed.
results with the squared term of the variable N are not displayed here. However, the estimation results are
qualitatively not much different from those shown in Table 2.7.
22 The

61

the critical point, the high risk of detection or heavier penalty to a ringleader increase the expected
cost of the collusive behavior. Thus, the probability of ringleader incidence begins to decrease.
The positively significant estimates are obtained for the proxy for asymmetry, Asym. Here,
the normalized HHI is used.23 As the asymmetry in size increases, the odds of the ringleaderbased cartel structure increases. A cartel needs to have a ringleader to respond to the difficulties of
coordination problems. In other words, the cartel shows willingness to choose a ringleader-based
structure with more asymmetries in the industry in order to maximize joint profit, instead of no
ringleader. All three models consistently show the same directions. These results are consistent
with the findings of Davies and De (2013), where the EU data set is used.
Product homogeneity could be an important factor in analyzing ringleader presence in cartels.
According to theoretical predictions, it is generally recognized that product homogeneity facilitates
collusion.24 Therefore, it can be argued that a cartel with a homogeneous product, compared to
the case with heterogeneous products, would be unwillingly to delegate the ringleader’s role to any
firm. However, the estimates are significant at only 10% in both index models, even though the
directions of the coefficients are all consistently negative.25
The coefficient for the industry association dummy Asso is regarded as the impact of an alternative monitoring mechanism on ringleader incidence in a cartel. As expected, the results consistently
show significantly negative directions for the three models. The existence of a ringleader in a cartel
is less likely if an alternative monitoring mechanisms is involved. This result confirms that a cartel
uses a ringleader as an organizational response. Being a ringleader can increase profits by coordinating collusions, while it is also risky and costly because many anti-trust agencies impose heavier
fines on ringleaders detected for collusion. Thus, if other mechanism plays a ringleader-like role
in a cartel, a firm would be less willing to take the role of a ringleader in the cartel. Compared to a
case without involvement of an industry association, when an industry association is involved in a
23 When an alternative proxy for asymmetry—the range of cartel share between the largest and the smallest
firms—is used, the results are similar at the conventional level of significance.
24 Product homogeneity feature also has a destabilizing effect on a cartel. Homogeneity makes it possible for a firm
to capture larger market shares when undercutting.
25 Here, the criterion is applied. When criterion is used, the estimate results are qualitatively similar. The results
a
b
are not displayed here.

62

cartel, the probability of a ringleader-based cartel decreases by 0.276 in Specification (2).
Also, as predicted, the generously revised leniency program in 2005 increases the probability
of a ringleader cartel by 0.335 in Specification (2). Because the KFTC grants ringleaders leniency
benefits such as fine reduction or immunity in return for submitting incriminating evidence, the cost
of being a ringleader may decrease. However, it should be noted that this policy change can also
destabilize a cartel because of the higher detection rate by anti-trust agencies. When we restrict
the analysis to ringleader frequency, this policy modification empirically increases the probability
of the ringleader-involved cartels. The overall percent correctly predicted is moderately high.
Davies and De (2013) argue that the likelihood of a ringleader varies with the nature of an
agreement. They discuss that the need for a ringleader is less pronounced where collusion can
be merely forborne by members. (i.e., the agreement is self-enforced.) In such situations, each
firm respects its rivals’ previous market shares, where cheating is easier to detect and there is no
necessity to agree on a common price. Davies and De (2013) finds that a ringleader does not exist
in the market allocation cartels of their EC samples.
Table 2.9 shows the results with a sub-sample set of having only price-fixing or bid-rigging
cartel cases. It is possible that price-fixing and bid-rigging cartels, compared to others, such as
market allocation cartels, are unstable. Among the 39 cartels in the current sample, which were
neither price-fixing nor bid-rigging cartels, only three cases were leniency-filed.26 If that is the
case, the magnitudes of the main estimates with those partial samples would be greater than those
of the coefficients obtained by using a full sample set.
Here, Specifications (1), (3), and (5) are associated with price-fixing cartel samples, and (2),
(4), and (6), with bid-rigging cartels. The estimation results are mixed, with conflicting consequences depending on agreement types. In the case of bid-rigging cartels, compared to the results
from full sample regressions in Table 2.8, the value of the estimates as to Number of firms is greater
than the relevant coefficient of full sample results. Considering a bid-rigging cartel is often related
to only a one-time project, there is also a larger incentive to deviate as the number of firms increases
26 Out

of 288 price-fixing or bid-rigging cartels, the 70 cases were leniency-filed.

63

Table 2.9 LPM, Probit, and Logit Estimates of Ringleader-based Cartel
(Price-fixing or bid-rigging samples only)
Dependent Variable: Ringleader involved
LPM (OLS)

Probit (MLE)

Logit(MLE)

(1) pricing

(2)bidding

(3) pricing

(4)bidding

(5) pricing

(6)bidding

.024***

.032***

.089***

.135***

.147***

.222***

(.007)

(.012)

(.027)

(.047)

(.046)

(.08)

.347**

.231

1.184**

.688

1.934**

1.17

(.164)

(.228)

(.533)

(.727)

(.89)

(1.251)

-.095

-.202**

-.297

-.704**

-.509

-1.212*

(.074)

(.101)

(.238)

(.359)

(.402)

(.628)

-.296***

-.47***

-1.11***

-2.072***

-1.861***

-3.499***

(.076)

(.164)

(.286)

(.689)

(.512)

(1.243)

.22***

.387***

.781***

1.417***

1.262***

2.458***

(.072)

(.123)

(.234)

(.461)

(.401)

(.912)

Industry effect

Yes

Yes

Yes

Yes

Yes

Yes

Cons.

.149

-.001

-1.131**

-1.772**

-1.848

-3.255

(.104)

(.202)

(.348)

(.726)

(.598)

(1.477)

238

105

231

231

-

-

54.67

Correct Predictions(%)

-

-

55.5
71.9

101
40.33
80.2

72.7

101
40.24
79.2

ll
Peudo R2†

-

-

-123.67

-48.72

-124.03

-49.763

.221

.346

.183

.289

.181

.288

Variables
Number of firms
Asymmetry
Homogeneity
Industry Association
Post-2005

Obs.

LR

chi2

Remark: Standard errors in parentheses. Significance levels: *<10%, **<5%, ***<1%.
†The usual R-squared is reported for LPM.

in a cartel. In those situations where cartel problems are prevalent, the necessities of ringleaders
grow to avoid the break-down of a cartel. However, this is not the case with other coefficients
on Asymmetry, Homogeneity and Industry Association in both sub-samples. The magnitudes of
relevant coefficients are not significantly greater than those of the full samples, or on the contrary,
they display exactly opposite consequences. These seemingly contradictory results possibly can
be attributed to the measurement issue, which results from ambiguous typology when the KFTC
classifies cartel types.
27 For

27

example, out of 263 price-fixing cartels classified by the KFTC’s application, 62 cases are also combined

64

Table 2.10 Multinomial Logit Estimates with 3 ringleader choices
Dependent Var: Ringleader type

Base: No ringleader

Variables
Number of firms
Asymmetry
Homogeneity
Administrative role of Industry Association
Active role of Industry Association
Post-2005
Industry effect

Cons.
Obs.

Organizing-only

Aggressive RL

.093†

.183***

(.061)

(.045)

.833

2.379***

(1.243)

(.884)

.513

-.468

(.75)

(.406)

-2.262**

-1.17**

(1.105)

(.532)

-2.131*

-2.33**

(1.128)

(.83)

.717

1.546***

(.651)

(.395)

Yes

Yes

-18.234

-2.302

(1448.503)

(.602)

289

103.92
70.9
-194.46

LR chi2
Correct Predictions(%)

ll

Peudo R2

.211

Remark: Standard errors in parentheses. Significance levels: *<10%, **<5%, ***<1%.
†The p-value is .128.

Table 2.10 shows what kinds of ringleader types can be chosen by a cartel when it confronts an
unfavorable situation, where a deviation could possibly be attractive. The multinomial logit model
is estimated to distinguish aggressive ringleaders from organizing-only ringleaders, as opposed to
no ringleader cartels. Compared to Specification (6)long in Table 2.8, the magnitudes of the first two
main coefficients—related to variables Number of firm and Asymmetry—for aggressive ringleader
types are greater, while those for organizing-only ringleader types are less. This finding implies
that a cartel structure prefers to choose an aggressive ringleader, rather than an organizing-only
with other types such as a market allocation cartel.

65

ringleader in order to maximize the joint profit as those adverse environments become aggravated.
As the number of firms increases, an aggressive ringleader is preferred to an organizing-only
ringleader. However, the estimation results for an organizing-only ringleader are insignificant at
conventional level, where the p-value is 0.128. There are only 28 cases of an organizing-only
ringleader. It is probable that organizational ringleader cartels, due to the fragility features, can
either develop to more stabilized cartels with an aggressive ringleader, or collapse in the middle
of collusion without detection. Actually, the average cartel duration in the current data set is 12.4
months for organizing-only cartels, but 17.1 months for aggressive cartels.
Here, the role of an industry associations is classified into two types in order to determine how
the role of an industry association affects the presence of a cartel ringleader. When an industry
association is involved in a cartel, its role falls into either an administrative, such as providing a
meeting place, distributing determined cartel plans to firms, etc., or an active role, such as presiding
over cartel meetings, suggesting cartel plans, etc. The empirical results consistently confirm that an
industry association plays a substitutive role for a cartel ringleader. When an industry association is
involved in a cartel, the probability of a ringleader-based cartel decreases irrespective of the role of
the industry associations. That is, no ringleader in a cartel is preferred when an industry association
is already involved with a cartel. More specifically, when an industry association already plays an
administrative role in a cartel, an organizing-only ringleader is less likely to be chosen, compared to
an aggressive ringleader. This finding implies that if an industry association plays an administrative
role in a cartel and the cartel needs a ringleader, it delegates an aggressive role to a ringleader in
order to additionally enforce or monitor the cartel. This result is the same when the role of industry
associations is active. That is, when an industry association already plays an active role in a cartel,
an aggressive ringleader is less likely to be chosen. A ringleader does not have to take the risk of
harsher penalty by an anti-trust agency. These estimates are significantly negative at a conventional
level. The Small and Hsiao (SH) test for the independence of irrelevant alternatives (IIA) confirms
that the assumption can be accepted, while the cross equation test fails.

66

2.5

Conclusions

Many previous empirical papers on cartels focus on the overall determinants of cartels. It is generally acknowledged that unfavorable market features against collusion destabilize cartels. However,
these factors can be endogenously determined following the organizational responses of cartel
firms. This endogeneity can be a potential reason for the divergence between conventional wisdom
and empirical estimation results. This research delves deeper into a ringleader’s role in a cartel in
order to find an adequate answer to the empirical puzzle. It finds that a cartel structure develops
more efficient and enforcing mechanism as an organizational response. It identifies the features of
ringleaders’ behaviors and their strategic responses across their roles and agreement types.
The current paper humbly contributes to the literature by concentrating on a ringleader’s role
in organizing a cartel structure using Korea’s cartel cases. It confirms previous findings by using
richer samples and adding meaningful variables that can cause cartel problems. Also, it distills
more extensive and stylized facts about a ringleader. The empirical results show consistent and
significant evidence on a ringleader’s significant role in overcoming or at least alleviating cartel
problems. A cartel willingly or sometimes reluctantly delegates its primary role into a ringleader
under unfriendly environments against the cartel. That is, a ringleader is more likely to instigate
a cartel or even to actively participate in determining cartel plans as the number of firms or the
asymmetry in size grows, but product homogeneity becomes less. These sorts of behaviors, in
spite of the seemingly mixed results, are more pronounced in bid-rigging cartels, where it is more
difficult to coordinate and to agree on a focal point for collusion. Moreover, firms prefer aggressive ringleader choices rather than organizing-only ringleaders to maximize their joint profits, as
opposed to no ringleader structure with more firms and greater asymmetries.
Last, but not least, this paper finds that an industry association is almost perfectly substitutive
for a ringleader in collusion. When an industry association is already involved in a cartel, no
ringleader structure is preferred. And, when an industry association plays an administrative role in
organizing a cartel, but if the cartel still needs a ringleader, an aggressive ringleader is preferred in

67

order to strengthen the management of the cartel.
These findings should be interpreted cautiously because there is the potential issue of sample
selection, which is unavoidable in an empirical study of cartels. We only observe cartels detected
and finalized by anti-trust agencies. More secretive and stronger cartels tend to hide underground,
undetected. The evidence also confirms that the leniency program granting the leniency benefits
to a ringleader increases the incidence of a ringleader in a cartel. Note that this effect is different
from the impact of the policy on cartel stability.
This study can be extended by empirically estimating the impact of cartel ringleaders or industry associations on cartel stability. Existence of those monitoring mechanisms could increase
or decrease cartel stability. Thus, it can be an empirical issue. With regard to public policy, even
though this research has some limitations, the findings can aid in fine-tuning or enforcing anti-trust
policies. For example, an anti-trust agency should be focus more on the role of an industry association because it frequently takes a ringleader-like role in a cartel. And, when the agency treats
a ringleader in penalizing or granting a leniency in order to desist or deter collusion, it could be
more effective to differentiate them by their types and roles. Also, when the agency determines
a merger, it should be noted that apparently pro-competitive features post-merger, such as more
firms or asymmetries can support collusion by making a stronger ringleader more prevailing.

68

REFERENCES

69

REFERENCES

Athey, S., and Bagwell, K. (2008). Collusion with persistent cost shocks. Econometrica, 76 (3),
493–540
Davies and De (2013). Ringleaders in larger number asymmetric cartels. The Economic Journal,
123
Dick, Andrew R. (1996a). When are cartels stable contracts? Journal of Law and Economics,
39(1), 241–83
European Commission (1996). Notice on the non-imposition or reduction of fines in cartel cases.
Official Journal of the European Commission C, 207, 4–6
European Commission (2002). Commission notice on immunity from fines and reduction of fines
in cartel cases. Official Journal of the European Commission C, 45, 3–5
Ganslandt, M. Persson, L. and Vasconcelos, H. (2012). Endogenous mergers and collusion in
asymmetric market structures. Economica, 79(316), 766–91
Grout, P and Sonderegger, S. (2005). Predicting cartels. Economic Discussion Paper: Office of
Fair Trading, 773
Ganslandt, M. Persson, L. and Vasconcelos, H. (2012). Endogenous mergers and collusion in
asymmetric market structures. Economica, 79(316), 766–91
Harrington, J.E., Jr. (2006). How do cartels operate? Foundations and Trends in Microeconomics, 2(1), 1–105
Ishibashi, L. (2008). Collusive price leadership with capacity constraints. International Journal
of Industrial Organization, 26 (3), 347–79
Ivaldi, M. et al. (2003). The economics of tacit collusion. Final report for DG Competition,
European Commisssion,
Jacquemin, A et al. (1981). A dynamic analysis of export cartels: the Japanese case. The
Economic Journal, 91 (363), 685–96
Korea Fair Trade Commission (2014). Statistical Yearbook. http://eng.ftc.go.kr/bbs.do
Levenstein, M.C. ad Suslow, V. Y. (2006). What determines cartel success? Journal of Economic
Literature, 44(1), 43–95
Markham, Jesse W. (1952). Competition in the Rayon Industry. Cambridge: Harvard University
Press,
Motta, M. and M. Polo (2003). Leniency Programs and Cartel Prosecution. International Journal
of Industrial Organization, 21, 347–79
70

Mouraviev, I. and Rey, P. (2011). Collusion and leadership. International Journal of Industrial
Organization, 29(4), 721–41
Osborne, Dale K. (1976). Cartel problems. American Economic Review, 66(5), 837–44
Posner, Richard A. (1970). A statistical study of antitrust enforcement. Journal of Law and
Economics, 13(2), 365–419
Rotemberg, J.J. and Saloner, G. (1990). Collusive price leadership. Journal of Industrial Economics, 39(1), 93–111
Stigler, George (1964). A theory of oligopoly. Journal of Political Economics, 72 (1), 44–61
Tirole, Jean (1988). The theory of industrial organization. Cambridge: MIT Press

71

CHAPTER 3
What Governs Cartels?: Evidence from Korea
3.1

Introduction

According to traditional theories, a cartel is inherently unsustainable due to the temptation of
cutting a collusive price among firms, unless firms are sufficiently patient. However, as Green and
Porter (1984) indicate, imperfect information can facilitate collusion under certain circumstances.
They discuss that a price war is an important mechanism to maintain on-and-off collusion. Many
empirical studies show that in reality, some cartels collapse without intervention of an anti-trust
agency, while others are detected by the agency for various reasons. However, a strong cartel can
survive in the industry without collapsing and being detected.
Irrespective of causes of death, however, it is observed that many cartels frequently operate a
special monitoring mechanism in order to survive. Cartel governance can be strategically chosen
based on an agreement among cartel members. Firms put more effort and funds into a cartel in
order to maintain collusion by reducing uncertainty and preventing firms from cheating. Levenstein
and Suslow (2006) argue that an industry association can be used as an important organization to
avoid reducing profits and undermining trust among cartel firms. For example, assume an industry
association collects price or sales information of each firm and helps cartel firms share it. Then, it
is easier for a cartel firm to distinguish “demand shock” from “cheating by rivals” when the firm
observes less sales or profits. Thus, a firm is less likely to mistakenly punish a rival firm even
when it suffers from a negative demand shock. Thus, it is important to understand what cartel
structure looks like, how it operates in a cartel, and how it affects cartel stability, as well as gaining
an understanding of the behavior of each individual cartel firm.
What actually operates a cartel inside it? There might exist various types of the monitoring
72

mechanism in a cartel. This study focuses on three types of cartel governance: an industry association, a body of regular meetings, and a hierarchical structure in a cartel. It is generally accepted
that these sorts of monitoring mechanisms support cartel stability by reducing distrust among firms
and facilitating communication. However, Levenstein and Suslow (2010) find that the impact of
an industry association on cartel stability can be different depending on the causes of cartel death.
This paper confirms that developing a monitoring mechanism significantly increases cartel stability. However, different types of cartel governance have different degrees of impact on stability.
When a cartel operates a body of regular meetings in which cartel firms participate, the likelihood
of cartel death is smaller than when an industry association is involved in a cartel. Interestingly, the
impact of various types of governance on cartel stability can go in opposite directions depending
on the causes of cartel death. Specifically, involvement of an industry association with a cartel increases cartel stability more in cases of “natural death” cartels than in cases of “death by anti-trust”
cartels. This finding is consistent with Levenstein and Suslow (2010).1
This paper also investigates the impact of other monitoring mechanism, a body of regular
meetings, which is not explored in the previous literature. This mechanism displays opposite
results from an industry association. A body of regular meetings in a cartel decreases the likelihood
of death more in “anti-trust death” cartel cases than in “natural death” cases. It is possible that both
types of cartel governance respond differently when detected, even though they commonly play a
role of facilitator in a cartel. Another finding of the paper is that a higher hierarchical structure
stabilizes “natural death” cartels while it destabilizes “anti-trust death” cartels even though the
point estimates are insignificant at a conventional level. Also, with regard to the role of industry
association in cartel stability, this paper finds that an industry association with administrative roles
decreases the stability of “anti-trust death” cartels, while one with an agreesive role does not.
However, the results are insignificant.
Survival models are used to identify the impact on cartel stability. The data set includes 327
cartel cases finalized between January 1, 2000 and December 31, 2010 by the Korea Fair Trade
1 They find that involvement of an industry association increases the stability of “natural death” cartel while it
decreases the stability of “anti-trust death” cartel.

73

Commission (KFTC). The remainder of the paper is organized as follows. The next section describes the data. In Section 3, empirical frameworks are explained. Section 4 displays the impact
of cartel governance on cartel stability. Section 5 explores how the role of an industry association
affect cartel stability and Section 6 concludes the paper.

74

3.2

Data and Descriptive Statistics

The sample set covers finalized 327 cartel cases filed between January 1, 2000 and December 31,
2010 for violations of Article 19 (1) of the MRFTA.2 Following Levenstein and Suslow (2010), the
data set is divided into two sub-samples across reasons of cartel death: “death by anti-trust” and
“natural death” in order to distinguish the effects of cartel governance on cartel stability. “Natural
death” might be caused by cheating, applying for leniency to an anti-trust agency, new entry, etc.
“Death by anti-trust” may be initiated by the investigation of an anti-trust agency that obtains a
lead from other sources such as consumer complaints, reports from up- or down-stream markets, a
press release, requests from a legislative branch, etc. The previous literature identify the cause of
cartel breakup based on a specific reason in each case document.3 However, here two sub-samples
are separated not by specific reasons of collapsing, but by the time-line of each case. In detail, if a
cartel ends before the KFTC begins its investigation, it would fall under the category of a “natural
death” cartel. In contrast, if the investigation start date of a case comes before the collapse date of
the cartel, this is included in the sub-group of “death by anti-trust.” The magnitudes of the relevant
coefficients in both sub-samples are compared with each other.
As shown in Table 3.1, the proportion of “anti-trust death” in Korea is about 34%, which is
lower by 26% than Levenstein and Suslow’s (2010) findings.4 There might be the possibility
of weaker detection or conviction capability by the KFTC. However, the proportion of leniencyfiled cases is very close, 22.3% in the current research and 20.98% in previous research. Also,
the average cartel duration is statistically longer in natural death cartels than in anti-trust death
cartels, as Levenstein and Suslow (2010) find. Also, natural death cartels, on average, have a
smaller number of firms and higher concentration ratio than anti-trust death cartels. It seems to be
consistent with comparison of the cartel duration. Natural death cartels are more likely stable.
2 Some

English-version documents filed during the periods are available for download from the Commission website, www.eng.ftc.go.kr.
3 According to Levenstein and Suslow (2010), if a cartel case dies due to amnesty application, cheating, or growing
fringe, it falls into the category of “natural death.” On the contrary, it dissolves due to follow-on investigation, customer
complaint, or whistle-blowers, it goes into the group of “anti-trust death.”
4 There are 49 anti-trust death out of total 81 cases.

75

Table 3.1 Descriptive statistics on Korea cartel by death reasons (2000 - 2010)

Number of total cases [leniency-filed]
Average cartel duration (month)*
Average number of firms
Concentration ratio†
Product Homogeneity‡
Industry Association
Regular Meeting
Hierarchy (frequency)
- executive or higher
- general manager
- working manager or lower
- unknown

Natural Death

Death by Anti-trust

216 [73]
29.48
(36.57)
5.80
(4.52)
4,195
(2,030)
.40
(.49)
.23
(.42)
.10
(.30)

111
19.21
(20.51)
6.80
(4.50)
3,258
(1,779)
.33
(.47)
.27
(.45)
.07
(.26)

126
28
19
43

68
20
6
17

Remark: Standard deviations in parentheses
*Cartel duration is averaged over no-bidding cases.
† Concentration ratio is measured by the method calculating HHI (Herfindahl-Hirschman index). But, here the cartel
share of each firm is used instead of its market share.
‡ When a cartel product is an intermediate rather than a final one, it is considered homogeneous.

In this data set, cartel products are classified into two categories, intermediate and final. Also,
following Jacquemin, A. et al. (1981) intermediate goods and services are considered homogeneous, while final ones are not. When a cartel agreed upon intermediate goods and services
rather than final ones, there can be two conflicting effects on cartel stability. Considering price
information as to intermediate products is relatively private, the cartel is likely more unstable.5
Also, buyers of the intermediate products tend to have bigger gaining power, or are more organized/concentrated than unorganized many average consumers who purchase final goods and ser5 Refer

to Harrington and Skrzypacz (2007) for discusssions of private prices and collusion.

76

vices.6 This feature also can hurt cartel stability. In contrast, if intermediate products are relatively
more homogenous than final ones, cartel firms can find an agreeable focal point more easily than
when a cartel product is a final product. This supports cartel stability. It seems that the latter effects
apparently dominate the former effects.
Moving onto the main predictors of the current paper, there are three variables that influence on
cartel activity: involvement of industry association in a cartel, existence of regular meeting among
cartel firms, and a hierarchical structure of personnel involved with the cartel. These variables
play important roles in forming/managing a cartel inside either the cartel itself or even each cartel
firm. There is no significant difference between two sub-samples as to the frequency of an industry
association or a regular meeting. Also, note that a top level position such as an executive or a CEO
is more likely to be involved in cartels in both sub-samples. More rigorous analyses are needed to
draw significant results controlling for other covariates.

6 Dick

(1996) finds that Webb-Pomerene cartels with more concentrated consuming markets are less stable.

77

3.3

Empirical Frameworks

Survival analysis is employed to estimate the impact of organizational characteristics inside a cartel
on cartel stability. First, the Cox semi-parametric hazard model is used following much previous
literature (De, 2010; Zhou, 2012; Choi and Hahn, 2016). Also, a parametric survival model with
Weibull distribution is applied in order to investigate the robustness of the results. These models
specify the probability of cartel collapse as a function of the main features that affect organizing
and maintaining a cartel inside it. In Cox regression, a hazard rate for a cartel i at time t, hi (t; x) is
basically the relative ratio of the probability density function f (x) to the survival function S(x). It
refers to the instantaneous probability with which a cartel breaks down, conditional on its survival
to time t-1. The hazard function can be expressed as follows.

hi (t; X) = h0 (t) × exp(Xi β )

(3.1)

where t is the elapsed time since the formation of a cartel, X is a vector of observed explanatory
variables that influence cartel durations, and β is a vector of parameters to be estimated via a maximum likelihood approach. These parameters measure the impact of the observed characteristics.
The hazard function consists of two terms: the first term, h0 (t), is the baseline hazard that depends only on time and is common to all subjects in the population; the second term, exp(Xi β ),
depends only on the explanatory variables rather than on time. This term shifts the baseline hazard. For the explanatory variables, cartel level information is collected from the sample data set.
The main predictors are the variables as to cartel governance that influence operating a collusive
agreement inside either a cartel itself or cartel firms. They include the dummy for the involvement
of an industry association in a cartel, the dummy for the existence of regular meetings in a cartel,
and the highest hierarchical position of those involved in a cartel. Other determinants are also included, such as the number of cartel firms, the concentration ratio measured by the cartel shares of
the firms, the dummy for the innovative anti-cartel policy change in Korea in 2005,7 and industry
7 The

KFTC innovated anti-cartel policies around 2005. Most of all, it developed an old leniency program in more

78

effects. When the ratio of hazard rates is calculated, the baseline hazard rates are canceled out,
implying that the ratio does not depend on time, but only depends on covariates.
For example, if the main determinant of cartel stability, say x1 , increases by one, while holding
other covariates fixed, the relative hazard rate ratio will be the following.
hi (t, ; x1 + 1, x2 ) h0 (t) × exp {β1 (x1 + 1) + β2 x2 }
=
= exp(β1 )
hi (t; x1 , x2 )
h0 (t) × (β1 x1 + β2 x2 )

(3.2)

where the positive coefficient - an estimated hazard rate ratio greater than one - implies that
the observed variable increases a dissolution hazard rate and reduces cartel duration controlling for
other variables, while a negative coefficient - an estimated hazard rate ratio less than one -reflects
that the relevant characteristic increases cartel stability, given other variables are fixed.

transparent and predictable ways following the US DOJ in 1993 and the EC in 2002.

79

Table 3.2 Hazard rate estimation results
Cox Models
Variables

Natural Death Death by Anti-trust
(3)
(4)
Hazard Ratio
Hazard Ratio
(z-statistic)
(z-statistic)
Asso
.644**
.918
(-1.98)
(-.29)
Reg_Mtng
.431***
.349**
(-3.27)
(-2.20)
.911
1.232
Hierarchy
(-.75)
(1.04)
1.001
.987
NF
(.02)
(-.33)
1.001
1.001
Concent
(.37)
(.44)
Intermed_Prodct
.937
4.407***
(-.30)
(3.79)
.540***
.366***
post_2005
(-2.96)
(-3.35)
IND
Yes
Yes
Cons.
.031***
.001***
(-6.15)
(-8.26)
Obs.
154
77
154
77
Chi2
38.03
41.80
35.49
44.86
Remark: Z-statistics in parentheses. Significance levels: *<10 percent, **<5 percent, ***<1 percent.

3.4

Natural Death
(1)
Hazard Ratio
(z-statistic)
.637**
(-2.00)
.405***
(-3.48)
.899
(-.85)
1.004
(.12)
1.001
(.50)
.954
(-.21)
.510***
(-3.17)
Yes

Parametric Models

Death by Anti-trust
(2)
Hazard Ratio
(z-statistic)
.849
(-.55)
.319**
(-2.30)
1.23
(1.02)
.989
(-.28)
1.001
(.44)
4.376***
(3.53)
.360***
(-3.29)
Yes

Empirical Results

In this section, the regression analyses for the equation 3.1 are performed. Table 3.2 displays the
estimation results. The regression turns first to the Cox model [Specifications (1) and (2) in Table
3.2] with two sub-samples, “natural death cartels” and “death by anti-trust cartels.” Then, parametric models assuming Weibull distribution [Specifications (3) and (4) in Table 3.2] are estimated.

3.4.1

Involvement of Industry Association

As expected, involvement of an industry association (Asso) in a cartel increases cartel stability
overall for all specifications, irrespective of the reasons of cartel death. In detail, for cartels that

80

collapse without being detected by an anti-trust agency (natural death), an industry association
involved in the cartel significantly decreases the likelihood of the cartel death. [Specification (1)]
Involvement of an industry association is also more likely to obstruct the breakdown of the cartel
by an anti-trust agency (death by anti-trust). [Specification (2)] However, the magnitudes of the
hazard ratio as to the industry association variable are different.
The hazard ratio for the sub-sample of “death by anti-trust” is greater than the magnitude for
“natural death.” It implies that the impact of involvement of an industry association on cartel
stability is less in the case of “death by anti-trust” than in the case of “natural death.” The result
is similarly consistent with the findings of Levenstein and Suslow (2010). Assuming the same
base line hazard, the difference of the coefficient on Asso between “natural death” and “death
by anti-trust” is marginally significant at 10% significance level.8 Even though involvement of
an industry association facilitates a cartel, some other effect can reduce cartel stability for “antitrust death” cartels. It might be possible that operation of an industry association leaves a trail
of incriminating evidence, which supports prosecution by anti-trust agency.9 As a result, a cartel
in which an industry association is involved becomes unstable when it is deteced by an anti-trust
agency. However, the point estimate is insignificant.

3.4.2

Existence of Regular Meetings

The estimation results as to the variable, existence of regular meetings in a cartel (Reg_Mtng)
are all significant at a conventional level. Expectedly, the existence of a regular meeting in a cartel
also stabilizes a cartel by facilitating better communication and sharing updated information on the
cartel. The relevant hazard rate ratios are less than one for all specifications. Also, the stabilizing
effects of a regular meeting are greater than those of an industry association for both sub-samples.
However, the sizes of the magnitudes in “natural death” and “death by anti-trust” are completely
opposite, when compared with those of the previous variable, Asso. That is, if a cartel has a regular
8 Using

robust standard errors, p-value is 0.096 with logarithm value of cartel durations.
the involvement of a trade association
helps to stabilize the cartel, providing a mechanism for communication and mediation; but it also increases the cartel’s
visibility and the written record of its conspiratorial activities, making it more vulnerable to prosecution.”
9 Levenstein and Suslow (2010) write that “Cartels appear to face a tradeoff:

81

meeting among cartel firms, the meeting increases cartel stability more when the death is caused
by an anti-trust agency’s intervention than in the case of “natural death.”
What makes a regular meeting yield different results than an industry association? It is probably
because an industry association and a body of regular meetings behave differently during collusion
and respond differently to an anti-trust agency’s investigation. A body of regular meetings usually
consists of cartel firms, while an industry association is a legally independent entity of cartel firms.
When an anti-trust agency discovers a cartel, it tends to focus more on cartel firms rather than an
industry association and charges them a heavier penalty. When detected, cartel members in a body
of regular meetings are more careful in concealing hard evidence as to their illegal behaviors in
order to avoid heavy penalties. Those behaviors can happen during collusion, but they are more
pronounced when the cartel is detected by an anti-trust agency. However, in the case of a cartel
in which an industry association is involved, an industry association invests less effort and care
into hiding incriminating evidence, unless it is actively involved with the cartel. Thus, a cartel
with a body of regular meetings, when an investigation is initiated by an anti-trust agency, likely
has lower probability of being convicted, compared to a cartel in which an industry association is
involved - that is, the former seems to be likely stronger and more stable than the latter. However,
the difference test between two sub-samples are not significant at a conventional level.

3.4.3

Hierarchical Structure

A cartel is composed of firms in the industry. However, a collusive decision is made by personnel
in the firms. Interestingly, a hierarchical structure shows different impact on cartel stability across
the reasons of cartel death, even though the results are insignificant. It is possible that a cartel
becomes stable, as higher levels of positions are involved in the cartel. As expected, when higher
positions are involved in cartels, cartel stability increases in “natural death” cartels. [The coefficient
is less than one in Specification (1).] It is likely that a cartel agreed on by a CEO or executive of
firms is stronger and more stable than one by a working manager or one in a lower position. In
contrast, stability decreases in “death by anti-trust” cartels. [The coefficient is greater than one in

82

Specification (2).] Probably, there is a conflicting effect on cartel stability. A higher level of people
is more likely to cooperate with the investigation of the anti-trust agency in order to avoid personal
liability when a cartel is detected by the agency.

3.4.4

Other Determinants

Even though it is not a main predictor, the variable, intermediate products (Intermed_Prodct), also
shows conflicting results in “natural death” and “death by anti-trust” sub-samples. As noted before,
when a cartel agrees upon intermediate goods and services rather than final goods, two conflicting
effects on cartel stability exist. This feature, on the one hand, decreases cartel stability due to
private price information, or greater bargaining power of more organized/concentrated buyer than
unorganized average consumers who purchase final goods and services. However, in contrast,
this can support cartel stability if intermediate products are relatively more homogenous than final
ones.10 In that case, cartel firms can find an agreeable focal point more easily than when a cartel
product is a final product. The results indicate that the latter slightly dominates the former for the
“natural death” cartels [Specification (1)], while the former strongly dominates the latter for the
“death by anti-trust” caratels [Specification (2)] It is likely that when a cartel with an intermediate
product is detected, it can face organized complaints or lawsuits on the buyers’ side.11 And an
investigating agency can collect stronger evidence from concentrated buyers. These conditions
can hinder cartel stability more when the cartel with intermediate products is detected by an antitrust agency.
However, no significant evidence regarding the negative relationship between the number of
firms (NF) and cartel stability is found. A greater concentration ratio (Concent) measured by cartel
shares does not increase cartel stability, either. Unlike conventional theoretical predictions, many
previous empirical results do not confirm these. Endogeneity issues may be the reason behind this.
The cartels discovered post 2005 - when anti-cartel policies were innovated - have longer cartel
10 Homogeneity

feature also can destabilize cartel because a firm can capture greater market share when undercut-

ting.
11 Unlike the USA, South Korea does not have a treble damage system in civil suits.

A class action lawsuit following
indictment against illegal cartels is not institutionalized as well as a civil lawsuit filed by consumers is not prevalent.

83

durations for both sub-samples. According to Harrington and Chang (2009) and Brenner (2009), a
longer duration following policy change means destabilizing effects on collusion in the short run,
because breakdown of the “marginal cartels” increases the average duration of population cartels.
Expectedly, policy change has a greater impact on “death by anti-trust” cartels.

84

3.5
3.5.1

Different Roles of an Industry Association
Roles of Industry Association

The same regression is conducted except for the distinguishing roles of an industry association.
Here, the roles of industry associations are grouped into two categories, administrative and active.
In cases of administrative roles, an industry association is used as a simple instrument in order to
support collusive decisions made by cartel firms. Here, an industry association provides a meeting
place or distributes information to cartel firms. However, if an industry association actively has
cartel firms form to join a collusive agreement, or determines a specific condition of the agreement,
this case falls into an active role. An industry association might respond differently depending on
its role. As a result, the impact of an industry associations on cartel stability also can vary.

3.5.2

Results

Table 1.3 shows the results. When an industry association with administrative roles is involved
in a “natural death” cartel, its involvement increases cartel stability. However, if the cartel dies
due to an anti-trust agency’s intervention, its involvement hurts cartel stability, and even decreases
cartel durations. Because the penalty related to the cartel case is focused on cartel firms, the
industry association is more likely to comply with an anti-trust agency’s investigation. Or, it at
least is less likely to try to conceal incriminating evidence about the cartel. Generally, an industry
association playing an administrative role is less likely to be heavily punished by an anti-trust
agency unless it forced firms to join a cartel even when a cartel has been detected. This makes
the cartel unstable. Note that the hazard ratio of the variable, Admin_Asso is greater than one
even though it is insignificant. However, assuming the same base line hazard, the difference of the
coefficient on Admin_Asso between “natural death” and “death by anti-trust” is significant only at
10% significance level.12
12 Using

robust standard errors, p-value is 0.078.

85

Table 3.3 Hazard rate estimation results by Roles of Industry Association
Cox Models
Variables

Natural Death
(1)
Hazard Ratio
(z-statistic)
.560**
(-1.96)
.731
(-1.07)
.411***
(-3.41)
.891
(-.92)
1.004
(.13)
1.001
(.40)
.943
(-.26)
.534***
(-2.82)
Yes

Parametric Models

Death by Anti-trust
(2)
Hazard Ratio
(z-statistic)
1.331
(.69)
.604
(-1.28)
.319**
(-2.36)
1.247
(1.07)
.982
(-.44)
1.001
(.25)
4.968***
(3.73)
.354***
(-3.31)
Yes

Natural Death Death by Anti-trust
(3)
(4)
Hazard Ratio
Hazard Ratio
(z-statistic)
(z-statistic)
Admin_Asso
.594*
1.469
(-1.79)
(.96)
Active_Asso
.702
.625
(-1.22)
(-1.19)
.437***
.336**
Reg_Mtng
(-3.19)
(-2.27)
.905
1.244
Hierarchy
(-.80)
(1.08)
1.001
.979
NF
(.02)
(-.50)
Concent
1.001
1.001
(.30)
(.22)
.929
5.118***
Intermed_Prodct
(-.34)
(4.01)
post_2005
.556***
.358***
(-2.67)
(-3.37)
Yes
Yes
IND
Cons.
.031***
.001***
(-6.15)
(-8.24)
Obs.
154
77
154
77
2
38.53
43.86
35.69
47.38
Chi
Remark: Z-statistics in parentheses. in parentheses. Significance levels: *<10 percent, **<5 percent, ***<1 percent.

However, if an industry association plays an active role in a cartel detected by an anti-trust
agency, the industry association - as well as firms - possibly confront a high chance of being
charged or penalized for the wrongdoing. This behavior is likely to lead the industry association to
non-cooperation or hiding cartel evidence especially when detected by an anti-trust agency. This
type of industry association acts like a body of regular meetings before. The hazard ratio related
to the variable, Active_Asso decreases in Specification (2) than in (1). However, the result of the
difference test is insignificant at a conventional level. Other outcomes are very similar to those in
Table 3.2.

86

3.6

Conclusions

This research explores what takes place inside a cartel. A cartel is composed of multiple cartel
firms. If it fails in aligning the incentives of each individual firm to collusive profits, the cartel is
vulnerable to collapse. To maintain collusive profits and avoid undermining uncertainty, the cartel
develops organizational governance within. And, even though a cartel is an agreement among
firms, the cartel or the firms are run by personnel who work for them. The current study focuses on
what actually operates a cartel. More specifically, it studies the impact of three cartel governances
on cartel stability: involvement of an industry association in a cartel, existence of a regular meeting
in a cartel, and hierarchical positions of those involved in a cartel. To identify these effects, survival
analyses are employed for the data set of 327 cartel cases between 2000 and 2010 in Korea.
While the results are marginally significant, involvement of the industry association increases
cartel stability, irrespective of the causes of cartel death. This finding implies that an industry association helps cartel firms communicate with each other and reduce uncertainty/distrust. However,
when the samples are divided into two sub-samples based on causes of death, the increasing effect
in stability is greater for “natural death” cartels than “death by anti-trust” cartels. This result confirms Levenstein and Suslow’s (2010) finding in their data set of international cartels finalized by
the US DOJ and the EC. It is possibe that operation of an industry association in a cartel enhances
the risk of detecting “anti-trust death” cartels due to the trail of incriminating evidence. This effect
undermines the stabilizing effect in those cartels.
Next, existence of a regular meeting in a cartel significantly increases cartel stability for both
sub-samples. However, here the sizes of the magnitudes are opposite to those of the previous
variable, Asso. Possibly, cartel firms in a body of regular meetings put more effort into and are more
careful about concealing incriminating hard evidence than an industry association does. These
behaviors can be pronounced for “death by anti-trust” cartels. Thus, this stronger cartel, when
detected by an anti-trust agency, likely has lower probability of failure.
Third, involvement of a higher level of personnel in each firm with a cartel can increase or

87

decrease cartel stability depending on causes of cartel death. For “natural death” cartels, top executives or CEOs possibly increase cartel stability. In contrast, they can decrease stability for “death
by anti-trust” cartels. The KFTC, like what many competition authorities in other jurisdictions
do, charges harsher penalty and personal liability when top-ranking positions are involved with
a cartel. Thus, this progressive penalty mechanism possibly makes a cartel detected by anti-trust
agency less stable as higher-ranking positions are involved with a cartel.
Last, even though it is not a main predictor, an intermediate product show opposite effects
on cartel stability depending on types of cartel death. The characteristic of intermediate goods
and services, a priori, can stabilize or destabilize a cartel. While stabilizing effects dominate
destabilizing ones for “natural death” cartels, the latter dominates the former for “death by antitrust” cartels.
This research confirms the finding of the prior literature, which argues a cartel endogenously
develops organizational governance to avoid collapsing. This study contributes to the literature by
not only with its methodology, but also its focus. It uses a variety of variables related to structural
features inside a cartel, while it focuses on hierarchical characteristics in a cartel. This might be one
of the first research studies on cartel governance using rich cartel samples. It can be replicated by
using cartel case samples of other anti-trust agencies. Also it can be extended to analyze the impact
of demographical characteristics - such as age, gender, and career of people who are involved in a
cartel - on cartel stability. This study is further extended to an in-depth analysis of an individual
cartel case.

88

REFERENCES

89

REFERENCES

Brenner, S. (2009). An Empirical Study of the European Corporate Leniency Program. International Journal of Industrial Organization, 27(6), 639–45
Choi, Y.J. & Hahn, K.S. (2014). How does a Corporate Leniency Program Affect Cartel Stability?
Empirical Evidence from Korea. Journal of Competition Law and Economics, 10(4), 883–907
Cox. D.R. (1972). Regression Models and Life-Tables (with Discussions). Journal of Royal
Statistical Society Series, B(34), 187–220
De, Oindrila (2010). Analysis of cartel duration: Evidence from EC prosecuted cartels. International Journal of the Economics of Business, 17(1), 33–65
Dick, Andrew R. (1996). When are cartels stable contracts? Journal of Law and Economics, 39,
241–83
Green, E.J. and R.H. Porter (1984). Noncooperative Collusion under Imperfect Price Information. Econometrica, 52(1), 87–100
Harrington, J.E. and M.H. Chang (2009). Modeling the birth and death of cartels with an
application to evaluating antitrust policy Journal of the European Economic Association, 7(6),
1400–35
Harrington, J.E. and Andrzey Skrzypacz (2007). Collusion under monitoring of sales. Rand
Journal of Economics, 38, 314–31
Jacquemin et al. (1981). A dynamic analysis of export cartels: The Japanese case. Economic
Journal, 91(363), 685–96
Levenstein, M. C. and V. Y. Suslow (2006). What Determines Cartel Success?Journal of Economic Literature, XLIV, 43–95
Levenstein, M. C. and V. Y. Suslow (2010). Breaking up is hard to do: Determinants of cartel
duration. Ross School of Business working paper,
Stigler, George (1964). A Theory of Oligopoly. Journal of Political Economy, 72(1), 44–61
Zhou, Jun (2012). Evaluating Leniency and Modeling Cartel Durations: Time-Varing Policy
Impacts and Sample Selection. Bonn University,
Zimmerman, J.E. and J.M. Connor (2005). Determinants of Cartel Duration: A Cross-Sectional
Study of Modern Private International Cartels. Working paper , http://ssrn.com/abstract=1158577

90