Laboratory Evidence on the Effectiveness of Corporate Leniency Programs
Author(s): Jeroen Hinloopen and Adriaan R. Soetevent
Source: The RAND Journal of Economics, Vol. 39, No. 2 (Summer, 2008), pp. 607-616
Published by: Wiley on behalf of RAND Corporation
Stable URL: http://www.jstor.org/stable/25474385 .
Accessed: 21/11/2014 06:57
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
Wiley and RAND Corporation are collaborating with JSTOR to digitize, preserve and extend access to The
RAND Journal of Economics.
http://www.jstor.org
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
RAND JournalofEconomics
Vol. 39,No. 2, Summer2008
pp. 607-616
Laboratory evidence on the effectiveness
of corporate leniency programs
Jeroen Hinloopen*
and
Adriaan R. Soetevent**
The number of cartels detected in theUnited States and inEurope has increased considerably
since the introductionof corporate leniencyprograms inantitrust legislation. It cannot, however,
be ruled out that thisapparent success results inpart from increased cartel activity.We explore
theeffectsof corporate leniencyprograms onpricing and cartel activityby use of an experiment.
Wefind that in the lab (i)fewer cartels are established when a leniencyprogram is inplace, and
(ii) cartels thatdo exist are less successful in chargingprices above thestaticNash equilibrium
price and have lower survival rates.
1. Introduction
Corporate leniency programs have become an increasingly important tool for antitrust
authorities to break cartels. These programs provide fine reductions and/or rewards for reporting
a cartel toantitrustauthorities by one of thecartelmembers. The firstsuch programwas initiated
by theU.S. Department of Justice in 1978. Since then, they have become part of antitrust
legislation intheUnited States, theEuropean Union, and other countries (OECD, 2002; Spagnolo,
forthcoming).
The number of cartels detected has increased considerably since the introductionof leniency
programs, but "in principle this could even be due to an increase in cartel activity" (Spagnolo,
2004). The reason for this is thatalong with the enhanced incentive to cheat on a cartelwhile
simultaneously reporting it, leniency programs can also make cartel activity less costly for
*Universiteit van Amsterdam and Katholieke Universiteit Leuven; j.hinloopen@uva.nl.
**Universiteit van Amsterdam; a.r.soetevent@uva.nl.
The comments of Philip Haile (the editor) and of two anonymous referees substantially improved thequality of the article.
We furthergratefully acknowledge valuable comments fromMichiel Bijlsma, Cory Capps, Stephen Martin, Jose Luis
Moraga-Gonzalez, Theo Offerman, Michael Price, Arno Riedl, Joep Sonnemans, Giancarlo Spagnolo, Charles Thomas,
and seminar participants theUniversity ofAmsterdam (CREED), Copenhagen Business School, Katholieke Universiteit
Leuven, 2005 EARIE meeting in Porto, and the 2006 ASSA meetings and 2006 IOS meeting, both held inBoston.
We thank Jos Theelen for developing the software. Financial support of the Economics Network forCompetition and
Regulation (ENCORE) is gratefully acknowledged. Soetevent's research was supported by a grant from the Program of
Fellowships for Junior Scholars of theMacArthur Research Network on Social Interactions and Economic Inequality.
The usual disclaimer applies.
Copyright ? 2008, RAND. 607
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
608 / THE RAND JOURNALOF ECONOMICS
participants. This occurs when cartel profitsnet of reduced fine payments exceed both defection
and expected cartel profits.When thishappens, toomany cartelmembers are given excessive fine
reductions upon reporting,and cartels are formed and reported continuously. Leniency programs
with thisfeatureare called exploitable (Spagnolo, 2004). So itremains an open question whether
thedramatic increase in leniency applications isproof of theirsuccess infightingcartels ormerely
reflects increased cartel activitydue to theirconditions having become too generous.
To answer thisquestion, empirical evidence on the effectsof leniency programs would be
most welcome. This typeof evidence, however, ishard tocome by because informationon cartels
not yet disclosed is typicallynot available. Moreover, natural experiments do not easily present
themselves, as it is not feasible to adjust legislation too often.lWith a laboratory experiment,
however, one can study theeffectsof leniencyprograms on thewillingness of subjects to engage
in (nonbinding) price discussions and on themarket prices thatresult.
Here we reporton four treatmentsof an experiment that implements such a setup and that
follow up and extend thework ofApesteguia et al. (2007). In each treatment,subjects repeatedly
play a discrete homogeneous-goods Bertrand pricing game ingroups of three.This game has been
introducedbyDufwenberg andGneezy (2000). In each period, subjects have tochoose an integer
in the range 101-110. Subjects who choose the lowest number (price) p receive net earnings of
(p
?
100)/L, with L equal to thenumber of subjects choosing the lowest number. This lowest
number is labeled themarket price. Any price above themarket price yields no revenue. The
Nash equilibrium price of theone-shot game ispN
= 101. The treatmentsdiffer in theoptions of
subjects tohave nonbinding price discussions prior toamarket phase inwhich theactual price is
determined. In comparing behavior andmarket outcomes across treatments,we study theeffects
of corporate leniencyprograms on pricing, cartel activity,and cartel stability.
One notion of cartel activity considers all situationswhere noncompetitive prices (p > 101)
occur as collusive. This notion is consistentwith models of tacit collusion with homogeneous
products. Apesteguia et al. (2007) apply a differentnotion by defining a cartel as the situation
when all groupmembers unanimously decide todiscuss prices. This latternotion ismore in line
with current legal practice thatfirms are unlikely to be found guilty of collusion when there
have not been explicit attempts by thefirms involved to communicate with each other (Motta,
2004). In our analysis, we consider both notions of cartel stability.Our evidence shows a positive
relationship between having price discussions and theoccurrence of noncompetitive prices, and
for this reason qualitative outcomes are similar,whichever notion is adopted.
The remainder of our article proceeds as follows. Section 2 provides a brief overview of
previous studies and an outline of our experimental design. Section 3 discusses the results and
Section 4 concludes.
2. Literature and experimental design
Apesteguia et al. (2007) examine the effects of leniency programs in a one-shot Bertrand
game. In a one-shot game, thepositive effectsof leniency programs are likely tobe overstated,
though,because a negative backlash ofwhistle-blowing for futurecooperation is ruled out. Our
design,which has subjects repeatedly play the same stage game with the same opponents, reduces
thisbias. In this article,we do not, however, attempt tomodel thedynamic strategies thatmay
ensue from this setup.
InApesteguia et al. (2007), thefine reduction is split among all cartelmembers who report.
In contrast,we explicitly take intoaccount theorder inwhich cartelmembers apply for leniency.
In thisway, we open up thepossibility for subjects to "race to report" by giving early applicants
largerfine reductions. Another difference from their setup is thatour communication phase is
shorterandmore structured,in thatsubjects can only communicate about prices and only forone
1
One possibility would be to compare cartel behavior across jurisdictions with different types of corporate leniency
programs enacted. Yet, additional factors that affect individual firmbehavior are likely to differ between jurisdictions,
including the institutional environment, the ruling legal system, and cultural norms.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
HINLOOPEN AND SOETEVENT / 609
minute ratherthan tenminutes. Inpractice, cartelmembers can talkabout anything foras long as
theywant. In thisrespect, theapproach ofApesteguia et al. (2007) comes closer to reality.2Our
results should thereforebe treatedas complementary to those ofApesteguia et al. (2007).
Leniency programs differ in the size of fine reduction for reporting a cartel, in the extent
towhich leniency is still an option if the antitrustauthorities have started an investigation into
theparticular cartel before amember reports it,whether it ispublic informationwhich and how
many firms have applied for leniency, and whether fine reductions are available for reporting
cartelmembers who are not the first to report the cartel. For example, in The Netherlands,
complete exemption from finepayments isavailable to cartelmembers who are thefirstto report
a cartel, thatfullycooperate with the ensuing investigation, thatare not the cartel's ringleader,
and provided thatno investigation into the particular cartel is ongoing. In case the antitrust
authorities are already investigating theparticular cartel, thefine reduction is reduced by 0-40
percentage points. For laterreporting,cartelmember fine reductions in the range of 10^0% are
available. As a rule, informationas towhich firmsapplied for leniency isnotmade public during
the investigation.The leniency program we investigate resembles the currentpractice inmany
jurisdictions: thefirstapplicant receives full amnesty, the second is given a 50% fine reduction,
the third to come forward has to pay the fullfine, and applicants remain anonymous.3 As the
experiment ismotivated inpart by practice, we consider amoderate program only as opposed to
a courageous program. In the latter,thefirstapplicant receives amonetary reward in addition to
immunity.
Theory offers a number of predictions we can testwith our design. First, given that the
leniency program we consider is nonexploitable, we expect introductionof theprogram to lead
to a reduction of thenumber of cartels formed (Motta and Polo, 2003). Second, we expect the
leniency program to lead to lower average prices, as cartel activity is generally associated with
higher price levels. Third,with regard to the cartels thatdo formdespite the leniencyprogram,
Spagnolo (2004) points out that leniency programs provide cartelmembers who defect from the
collusive agreement, that is, who charge a lower price than the agreed-upon price, a protection
from-finesmotive as long as the expected reduced fines of the leniency program are below the
expected fine of a defecting agent who does not report.Because of this effect,we expect to
observe thatdefecting cartelmembers always apply for leniency in theperiod of defection.
Experimental design. We implement four treatments.Benchmark is the standard game
where subjects cannot discuss prices. Communication adds thepossibility for subjects todiscuss
prices, at no cost (see Step 1below). Ifgroup members unanimously agree to discuss prices, a
communication screen opens: a cartel is established. InAntitrust, groups thatdiscuss price face
ineach period a probability of 15% of being detected by an antitrustauthority.4Upon detection,
groupmembers have topay a fine equal to 10% of theirrevenue inthatperiod. Leniency resembles
Antitrust, except thatsubjectswho participate ina cartel have thepossibility to reportthecartel in
exchange fora fine reduction (see Step 3 below). In theexperiment, thefine liabilityof individual
cartelmembers automatically carries over to thenext period unless the cartel breaks down due
to detection or a leniency application by one ormore cartelmembers. This is analogous to an
infiniterepose period.
Step 1.Communication phase. Each subject has todecide whether shewants todiscuss prices or
not by pressing the appropriate button; a communication window opens only ifall threegroup
members simultaneously choose to discuss price. The experiment thus precludes the creation
2
Restricting the communication to price discussions precludes subjects bringing the outside world into the
experiment ("If Iwin Iwill buy you a beer").
3
Unlike, for example, Motta and Polo (2003), we do not distinguish between the situation where the antitrust
authorities have started an investigation or not.
4
Strictly speaking, having a discussion about prices isnot illegal. To label the third treatmentAntitrust emphasizes,
however, the introduction of fines. The chosen probability reflects the empirical finding by Bryant and Eckard (1991) that
in a given year, 13%-17% of the existing price-fixing cartels are detected.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
610 / THE RAND JOURNALOF ECONOMICS
of partial cartels. Subjects only learn the joint communication decision in theirgroup, not the
individualwillingness to communicate or thepossible cartel activity inother groups.
Ifa cartel is established, a communication window opens thatallows each cartelmember to
indicate his or her acceptable price range.All groupmembers simultaneously supply aminimum
andmaximum price from thechoice set {101, 102,..., 110}. Ifa subject insistson one particular
price, she can articulate thisby choosing thatprice both as herminimum andmaximum price.
The intersection of all threeprice ranges becomes the choice set of a second round of price
negotiations. If the intersection is empty, the common choice set isnot adjusted. This updating
process continues until eithera unique price is reached or aminute has passed (atwhich point the
communication window closes).5 Because inour experiment at least 20 roundswere played, we
set the communication limitat one minute. Inmost cases, thisminute was more thanenough to
allow subjects to reach an agreement. InCommunication, Antitrust, and Leniency, thefraction of
price communications thatdid not yield a unique agreed-upon price was 7.3%, 1.0%, and 2.7%,
respectively.All price agreements were nonbinding.
Step 2.Market phase. Each subject chooses her price from the choice set {101, 102,..., 110}.
Subjects learn themarket price, being equal to the lowest of all threeprices submitted.
Step 3. Reporting phase. This step only arises inLeniency. If a cartel exists, subjects have to
decide whether or not to report the cartel. Each cartelmember has to push either the report
button or thenonreport button.6 In this step,no information isgiven about thereportingdecision
of other cartelmembers. The decision to report costs one point, irrespective ofwhether or not
leniency is granted. This cost reflectsadministrative costs, legal fees, and other consultant fees
thatfirms typically incurwhen filing a leniency application. Practically, levyinga small reporting
cost precludes a cartelmember who has been undercutfrom punishing defectors forfree.7
If the cartel is reported,all groupmembers' revenues are reduced by 10% of currentperiod
revenues. Subjects can thus realize negative earnings. For instance, ifprice negotiations resulted
in an agreed-upon price of 107, one member undercuts by quoting a price of 106, the cartel is
subsequently reported, and theundercutting cartelmember did not apply for leniency, itsnet
earnings in thisperiod would be: (106
?
100)
?
106/10
= ?4.6.
If the cartel still exists afterall subjects have made theirreportingdecision, the cartelwill
be detected with 15% probability by theantitrustauthority.Ifdetected, all groupmembers face a
revenue deduction of 10% of currentperiod revenues.8
All relevant informationabout the stage game is thendisplayed: submitted price, market
price, revenues gross of possible revenue deduction, revenue deduction, reporting costs, and net
earnings. The value of participants' totalearnings is also updated and isvisible at all times.
Experimental procedure. The experiment was conducted at the CREED experimental
laboratoryof theUniversity ofAmsterdam. There were 52 groups of threeparticipants each (12
groups inBenchmark, 13 inboth Communication and Antitrust, and 14 inLeniency). Subjects
were drawnfrom a largepool ofundergraduate studentsacross all subject fields. Every participant
was matched with the same other two subjects for a total of at least 20 periods.9 Subjects were
randomly assigned to a treatmentto ensure that therewere no systematic differences in subject
5
Although this communication scheme is not used elsewhere, it is rooted in the related literature. For instance, in
having a communication window of one minute, we are in linewith Cason and Davis (1995), who allow for a two-minute
communication phase.
6
We explicitly introduced these two buttons inorder to force subjects tomake their reporting decision consciously.
7
"Free punishment" is possible inApesteguia et al. (2007), which may lead to an overestimation of the effect of
leniency programs on cartel reporting.
8
Note that this leniency program is nonexploitable. Per-period expected cartel profits are (110
?
100)/3
?
0.15
0.10 110/3
=
167/60. These exceed per-period expected profitswhen applying for leniency: (110
?
100)/3
? 1
?
0.5
0.1 110/3
=
1/2.
9
From period 20 onward, the session ended with 20% probability, to attenuate possible final end-round effects.We
analyze only thefirst20 periods.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
HINLOOPEN AND SOETEVENT / 611
FIGURE 1
PATHSAVERAGEMARKET PRICE (LEFTPANEL)AND CUMULATIVEDISTRIBUTION FUNCTIONOF
MARKET PRICES (RIGHTPANEL)
1081 I-*? Communication!
-*~Antitnjst.
e 103. Jr?m
^^$8?'^K~^'yi^^ H* 0-4 -|
?
j
--Benchmark
If
^^
w^^3|Sv
^ ^ - ?Communication
102'
^_
/^_^/<\^*% 3 0.2
-|Nr-Antitrust
101.^ .^
" ~~""^
^ S ^ ? "^ ~- '^,_ ._ 0.1 -I
l~;;:r-Leniency
0.0-I-1-1-1-1-1-1-1-.-1-,
100-1?.?.?,?,?,?,?.?,?,?,?,?,?.?-.?,?,?.?.?, 101 102 103 104 105 106 107 108 109 110
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 1920
per** P'iC*
characteristics across treatments.Participants were paid theircumulative earnings ineuros (with
theconversion rate 1point
=
0.25) plus a 5 show-up fee.Average earningswere 14.40 and
themaximum andminimum paymentswere 41.40 and 6.80, respectively.10The lengthof the
sessions was between one hour and one hour and 40 minutes.
3. Experimental results
Pricing. First we consider the impact of Leniency programs on prices. The leftpanel of
Figure 1 shows the average price path for all four treatments.Prices decrease over time and the
average price forCommunication and Antitrust does not differ significantly fromBenchmark
(p > 0.1), but average prices inLeniency are considerably lower (p
=
0.019). The plot of the
cumulative distribution function (cdf) of prices depicted in the rightpanel of Figure 1 confirms
this finding.The cdf ofmarket prices in the other treatmentsclearly first-orderstochastically
dominates thecdf ofmarket prices in theLeniency treatment(p < 0.01).
These lowerprices can be due to lower cartel activity in theLeniency treatmentonly ifthe
cartels inour experiment aremore successful thannon-cartels incharging prices above the static
Nash equilibrium price. Figure 2 shows that this condition is satisfied despite the nonbinding
nature of the price discussions: forCommunication, Antitrust, as well as Leniency, the cdf of
market prices in cartel groups first-orderstochastically dominates the cdf of those innon-cartel
groups (p < 0.001).
The figurefurthermore shows that the cdf ofmarket prices of cartel groups inLeniency
is first-orderstochastically dominated by the corresponding cdfs ofmarket prices inAntitrust
and Communication (p
= 0.284 and 0.166, respectively). That is, cooperation in cartels thatdo
establish in theLeniency treatmentis less successful than in theother treatments.11
Thus, we have twopotential explanations for theobserved lowermarket prices inLeniency:
first,as shown above, cartels thatexist seem less successful in charging prices above the static
Nash equilibrium price; and second, leniencymay lead to fewer cartels being established. We
investigate the latterpossibility in thenext subsection; in the subsection on cartel stability,we
address thequestion why cartels in theLeniency treatmentare less successful.
Cartel activity. The bottom row of Table 1 shows that introduction of an antitrust
authority about halves the frequency of price discussions, thep value of no treatmenteffect
10
None of the subjects ended up with negative cumulative earnings.
11
For non-cartels, average prices are higher inBenchmark than inCommunication, Antitrust, or Leniency (p
=
0.076,0.350, and 0.036, respectively). This can be ascribed to a selection effect: inBenchmark, subjects have no option to
communicate, so that those subjects who wish to form a cartel inorder to establish higher prices also end up innon-cartel
groups.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
612 / THE RAND JOURNAL OF ECONOMICS
FIGURE 2
CUMULATIVEDISTRIBUTION FUNCTIONOF MARKET PRICES INCARTEL AND NON-CARTEL
GROUPS
1.0
-j^^^^^j
n .~~~^p-~_?, j
, ,,
||
..,.,
JWp^^g?j
jso.4-/ jA?
- ?non-cartels-Benchmark
^
fAjr
? ?non-cartels-Communication
3 0.3-
/??
?a?non-cartels-Antitrust
/// ??non-cartels-Leniency
d/y
-o?cartels- Communication
01_ Kf-^-cartels-Antitrust
O ~Q~ cartels -
Leniency
0.0-I-,-,-,-,-,-,-,-,-,-,
101 102 103 104 105 106 107 108 109 110
price
versus the one-sided alternative being 0.028. Introduction of a leniency program provides a
furtherreduction of cartel activity of about 50% (p
=
0.017). As shown in the first row of
Table 1, the alternative notion of cartel activity, that treats all situationswith noncompetitive
market prices as collusive, displays about the same picture. The reduction in noncompetitive
prices when an antitrustauthority is introduced is no longer significant,however (p
=
0.210).
To answer thequestion why fewer cartels come intoexistence in theAntitrust and Leniency
treatment,we firstconsider the development of the fraction of individual subject decisions in
favor of forming a cartel for all treatmentswhere communication is an option. The leftpanel
of Figure 3 shows that inAntitrust, the reduction in cartel activity is related to a significant
drop in the percentage of decisions in favor of price discussions relative to Communication
(p
=
0.025). In contrast, thenumber of subject decisions infavorof price discussions inAntitrust
and Leniency are about equal (p > 0.10). This cannot explain thedrop incartel activitybetween
these two treatments.
Recall thatfora cartel tobe formed,all threegroupmembers unanimously have to say "yes"
to startprice discussions. The rightpanel of Figure 3 shows thatLeniency not only has the lowest
fraction of unanimous decisions but also the highest fraction of "near unanimity" situations
where price discussions are blocked because one groupmember refuses to communicate.
This in turnraises thequestion whether theexperimental differences between Antitrust and
Leniency lead subjects in the lattertreatmenttobe more persistent in theirchoice not to collude.
TABLE 1 Average Fraction of Noncompetitive Prices and Cartels
Formed
Benchmark Communication Antitrust Leniency
Noncompetitive prices 0.47 0.60 0.50 0.16
O> 101)(0.36) (0.31) (0.29) (0.14)
Cartels
-
0.47 0.27 0.13
(-) (0.30) (0.20) (0.13)
Standard errorswithin parentheses.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
HINLOOPEN AND SOETEVENT / 613
FIGURE 3
PER-PERIOD AVERAGE FRACTIONOF SUBJECTS WHO WISH TO FORMA CARTEL (LEFTPANEL)
AND A HISTOGRAMOF THE NUMBEROF GROUP MEMBERS AGREEING TO FORMA CARTEL (RIGHT
PANEL)
1.0
-j
0.9- >^ q Communication
0 8 - ^'''^"^ ^
^^IIK^k P<^ ~*"*N-,
0S
antitrust
A
mS*\ ^OJ'*"^*^*^'^ ^^?~* &
0SsLeniency
- ?Communication
iiSHill /IfiS
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 200123
petiOGfnunterofsubjects
Figure 4 addresses thisquestion bymapping thecdf of thenumber of times each subject consents
to discussing prices. Whereas inCommunication and Antitrust all subjects consent to discuss
prices at least once, in theLeniency treatmentabout 10% of the subjects neverwants todiscuss
prices. Their persistence successfully blocks some of thecartels, although thedifferencebetween
thecdfs of Leniency and Antitrust isnot significant (p
=
0.402).
Cartel stability. In this section,we limitattention to the cartels thatdo form in order to
explain why cartels inLeniency are less successful in establishing highmarket prices. The cdfs
of the agreed-upon prices in the communication phase of thedifferent treatments(Figure 5) do
not reveal a significantdifference of subjects' behavior in theprice discussions (p
=
0.230): in
all treatments,67%-83% of the subjects agrees to charging themonopoly price of 110.
Apparently, if no differences in agreed-upon prices are observed but market prices in
Leniency are considerably lower,cartelmembers inLeniency defectmore often and/orundercut
FIGURE 4
CUMULATIVEDISTRIBUTION FUNCTIONOF THE NUMBEROF PERIODS INDIVIDUALSUBJECTS
WANT TO COLLUDE
? ?Communication
JT
0.9 --^Antitrust
/
0.8 - ~~Q~~ Leniency
gj
|
0.6 -
JjJ
3 no .
r-y^a^^mi ~.*JQT
0.1
Cf
jf M-+--+**
0.0 Imi Mi-1-1-?-1-1-1-1-r-1-?-1-1-1-1-1-1-r-1-1
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
number of times subjects say "yes"
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
614 / THE RAND JOURNALOF ECONOMICS
FIGURE 5
CUMULATIVEDISTRIBUTION FUNCTIONOF THE AGREED-UPON PRICES BY CARTEL MEMBERS
? ? Communicati on
j
0.9
.-^Antitrust
/
0 8
. ~-o~~ Leniency I
? 0.7 -
I
1 0.6 -
II
I 0.4 - 11
3 0.3
-
jt^B
0.0 -I?D r^nZj^'^'r^Cj.\?\Ji^~i^-1? -?-1-1-1->
101 102 103 104 105 106 107 108 109 110
price
TABLE 2 Causes of Cartel Breakdowns
Fraction of Cartels
Defection Detection Reporting
Communication 0.72
--
Antitrust 0.75 0.17
Leniency 0.97 0.03 0.78
TABLE 3 Fraction of Leniency Applications in Leniency among
Winning and Nonwinning Defectors and Compliers
Defect Comply
Win Not Win Win Not Win
Report 0.51 0.29 0.330.37
1st0.67 0.67 1.00 0.56
2nd0.25 0.22 0.00 0.33
3rd0.08 0.11 0.00 0.11
Observations 47 31324
more rigourously on the agreed-upon prices. Table 2 shows that this is indeed the case. The
agreed-upon price isundercut inLeniency by one ormore cartelmembers in97% of the cases,
as compared to about 75% in theother treatments.12The table also shows that in about 80% of
all cases, defection from theagreed-upon price is followed by a leniency application fromone or
more cartelmembers in the same period.
On basis of theprotection-from-fineseffect,we expect thatwinning defectors will always
be among theapplicants for leniency.Table 3 only partly supports thisexpectation: although the
12
The other 3% of cartels is detected by the antitrust authority so no cartel inLeniency survives formore than one
period.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
HINLOOPEN AND SOETEVENT / 615
FIGURE 6
KAPLAN-MEIERSURVIVALESTIMATES BY TREATMENT
8
in
d_
1
""
(DO. ?_
CM I
M--J_
l_ ^ ~h--^S-1-1-1-r?
0 5 10 15 20
duration Inperiods
?
Benchmark
?-??
Communication
" -"""" "
Antitrust
-..? ?-
Leniency
percentage of leniency applications in thepool ofwinning defectors, 51%, is considerably higher
than in theother groups (p
=
0.029), only about half of thewinning defectors file for leniency.
All this should affect theduration of cartels. To estimate the cartel survival rate,however,
the above definition of cartels is ambiguous, as it is unclear whether sequential periods with
price discussions and lower-than-agreedmarket prices constitute the same cartel or not. Figure 6
thereforedisplays theKaplan-Meier cartel survival estimates,with all prices above 101 treatedas
collusive. The estimated survivor functionsare not differentforBenchmark, Communication, and
Antitrust (p
=
0.37). But the log-ranktestclearly rejects equality across treatmentswhen Leniency
is included (p
=
0.00). That is, theprobability thatthemarket price exceeds 101 forany number
of consecutive periods ismuch lower inLeniency than inany other treatment.An alternative test
thatdefines cartel duration as thenumber of periods between the firstsuccessful price discussions
and the firstperiod inwhich market price is lower than themost recent agreed-upon price leads
to similar results. Equality of the survival functions can in this case not be rejected due to the
stricterdefinition of cartel breakdowns. This shortens considerably the recorded cartel lifetime,
therebyeliminatingmost of all variation across treatments.
4. Concluding remarks
The study explores the effects of moderate corporate leniency programs on pricing and
cartel activity.We find that introduction of a leniency program leads to lower prices for two
reasons. First,puttinga leniencyprogram inplace leads toa reduction incartel activity.A notable
difference from theother treatmentsis that in theLeniency treatment,about 10% of the subjects
refuses to discuss prices throughout the experiment. Second, those cartels thatare established
are less successful incharging prices above the staticNash equilibrium price. This isnot caused
by differences inbehavior in the communication phase but bymore frequent and more severe
undercutting of theagreed-upon price in themarket phase. This behavior in turn is related to the
shorter cartel lifetime in theLeniency treatment; theprobability that themarket price exceeds
the competitive level for any number of consecutive periods is significantly lower inLeniency.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions
616 / THE RAND JOURNALOF ECONOMICS
The data furthermoreindicate thatdefectors use the leniencyprogram to receive protection from
fines.
As noted by Spagnolo (forthcoming), "more empirical and experimental evidence would
be extremely welcome on all aspects of leniency and whistleblower programs." This article
responds to thiswish, but thereare stilla number of importanttheoretical results thatremain tobe
scrutinized. For example, although the leniency program we considered here complies with the
currentpractice inmost jurisdictions, Spagnolo (2004) stresses that themost effective leniency
programs are those thatgive a (more than) fullfine reduction to the first applicant and none
to subsequent ones. Likewise, experiments that incorporate endogenous detection probabilities
would also be most welcome (e.g., Harrington, forthcoming;Chen andHarrington, 2007).
References
Apesteguia, J.,Dufwenberg, M., and Selten, R. "Blowing theWhistle." Economic Theory, Vol. 31 (2007), pp. 143-166.
Bryant, P.G. and Eckard, E.W. "Price Fixing: The Probability ofGetting Caught." Review ofEconomics and Statistics,
Vol. 73 (1991), pp. 531-536.
Cason, T.N. and Davis, D.D. "Price Communications inaMulti-Market Context: An Experimental Investigation." Review
of Industrial Organization, Vol. 10 (1995), pp. 769-787.
Chen, J.and Harrington, J.E. "The Impact of the Corporate Leniency Program on Cartel Formation and the Cartel
Path." InV Ghosal and J.Stennek, eds., The Political Economy ofAntitrust. New York: North-Holland, 2007.
Dufwenberg, M. and Gneezy, U. "Price Competition andMarket Concentration: An Experimental Study." International
Journal of Industrial Organization, Vol. 18 (2000), pp. 7-22.
Harrington, J.E., Jr. "Optimal Corporate Leniency Programs." Journal of Industrial Economics, forthcoming.
Mason, C.F. and Phillips, O.R. "In Support of Trigger Strategies: Experimental Evidence from Two-Person Noncoop
erative Games." Journal ofEconomics andManagement Strategy, Vol. 11 (2002), pp. 685-716.
Motta, M. Competition Policy. Cambridge, UK: Cambridge University Press, 2004.
-and Polo, M. "Leniency Programs and Cartel Prosecution." International Journal of Industrial Organization,
Vol. 21 (2003), pp. 347-380.
OECD. Fighting Hard-Core Cartels; Harm, Effective Sanctions and Leniency Programs. Paris: OECD, 2002.
Spagnolo, G. "Divide et Impera: Optimal Leniency Programs." Working Paper no. 4840, CEPR, 2004.
-. "Leniency and Whistleblowers inAntitrust." In P. Buccirossi, ed., The Handbook of Antitrust Economics.
Cambridge, Mass.: MIT Press, 2008, pp. 259-304.
?RAND 2008.
This content downloaded from 147.251.185.140 on Fri, 21 Nov 2014 06:57:35 AM
All use subject to JSTOR Terms and Conditions