Leniency Policies and Cartel Success: An Experiment

Cartels are often fought by granting leniency, in the form of forgiveness of penalties, to whistle-blowers. This study employs a laboratory experiment to compare leniency programs that differ with respect to fine size and whether a second whistle-blower may apply for leniency. The results show that leniency does not affect the probability that a cartel forms, but is effective in exposing cartels and thereby inhibiting cartel success. Higher fines are more effective, but allowing leniency to a second whistle-blower is no more effective than granting leniency to only one whistle-blower.


Introduction
Economists and regulators have long been interested in the most effective policies to prevent the formation of cartels.While collusion and the formation of monopolies have generally been prohibited in the United States since the Sherman Antitrust Act of 1890, there have nevertheless been many attempts by companies to form cartels in a variety of industries.One reason that cartelization is an enduring problem is because once a cartel has formed, it can be difficult for antitrust authorities to expose it.Bryant and 1 3 Eckard (1991) estimate the probability of catching a cartel at between 13 and 17% in the US.Combe et al. (2008) estimate the probability of uncovering a cartel in the EU at 12.9-13.3% in the period from 1969 and 2008.In addition, the costs of the investigations necessary to expose cartels can be substantial.For example, the US Department of Justice allocated $188.5 million for antitrust enforcement in 2021.The duration of investigations can also be quite long.For instance, in the EU, cases brought between 2000 and 2011 had an average length of investigation of 50.8 months (Huschelrath et al. 2012).
In order to make it easier to expose cartels, the US introduced a leniency program in 1978.This initial program allowed the first applicant for leniency to receive a partial exemption from the penalties and fines for collusion, including reduced criminal punishment.The hope was that by giving a partial exemption, whistleblowing would be encouraged, leading to the breakup of existing cartels and the deterrence of future cartelization.The incentives were strengthened in 1993, with whistle-blowers receiving full immunity from any government-imposed penalties.As a result, leniency applications have increased by a factor of 10, with convictions and fines skyrocketing as well. 1ollowing the success of the US policy, leniency policies were adopted by other countries.The European Union's leniency policy, instituted in 1996, guarantees the first whistle-blower a penalty exemption, but also grants fractionally reduced fines for second and third whistle blowers.South Korea introduced a leniency policy in 1997, and then in 2005 additionally guaranteed a partial penalty exemption of 50% for the second whistle-blower.Japan adopted a leniency program in 2006 that allows a partial fine exemption additionally for a third applicant.The magnitudes of the penalties for collusion differ by country.For example, the EU and South Korea impose a maximum penalty of 10% of sales revenue for firms that have acted as a cartel, while the US specifies a maximum penalty of 20% of "affected volume." In this paper, using a controlled laboratory experiment, we study the effects of the two key components of leniency policies that vary internationally: the size of the penalty, and how many whistle blowers receive leniency.We consider whether, ceteris paribus, incentivizing a second whistle-blower with leniency and changing the fine size affect the rate of cartel formation, the probability of cartel exposure, the probability that an industry successfully colludes, industry profits, and the fines that the authorities collect.
The choice of experimental design and parameters was guided by a comparison of the American and South Korean leniency policies.Our research question is the following: Is the system of leniency in the US more or less effective at reducing cartels than the system in place in South Korea?The policies differ in two dimensions: First, in the US, leniency is granted to only one (the first) whistle-blower, and in South Korea it is granted to two whistleblowers.Second, the fine size differs in the two countries, and is higher in the US than in South Korea. 2 The two-factor, 1 3 Leniency Policies and Cartel Success: An Experiment two-level design of the experiment considers which of the two dimensions might be the source of any of any differences in outcomes that we observe.An additional control treatment is used to establish whether some of the versions of the leniency policy might not even be better than no leniency policy at all.
Our data show that all of the leniency policies that we include in our study reduce the ability of firms to successfully collude, with the exception of a policy granting leniency to two whistle blowers in conjunction with a relatively low fine.High penalties have the effect of reducing overall industry profits and increasing the fines that are collected by the state.The success of the leniency policies does not result from deterring the formation of cartels, but rather from increasing the rate at which they are reported and punished.
The paper is organized in the following manner: Sect. 2 discusses the prior experimental literature on cartel leniency.Section 3 describes the experimental design.Section 4 presents our hypotheses.Section 5 reports the results and Sect.6 provides a summary of the results and a concluding discussion.

Previous Related Literature
Previous research has studied the effect of leniency policies on cartel activity and has verified that they do have an effect.Motta and Polo (2003) show that leniency increases the chance that a cartel is exposed, but also note that the cartel formation rate may actually increase due to the drop in the expected cost of cartelization.Leslie (2006) finds that allowing exemptions for cartel ringleaders may promote the destabilization of cartels.Miller (2009) verifies empirically that leniency programs lead to greater levels of cartel revelation and deterrence, using cartel data in the US from 1985 to 2005.However, Brenner (2009), using data from the EU during the period of 1990 to 2003, shows that it is possible for leniency policies actually to lead to cartel stabilization.Harrington (2008) studies the effect of fine reductions and finds that a full fine exemption is more effective for catching cartels than a partial fine exemption; but the increase in the number of full fine-exempt firms may make it difficult to destabilize cartels.Zhou and Gartner (2012) confirm that larger fine reductions give rise to the faster breaking-up of cartels.
A number of experiments have investigated the effect of leniency policies.See Marvao and Spagnolo (2014) for a survey.The initial studies considered environments with Bertrand price competition.Apesteguia et al. (2007) study the effect of leniency on cartel formation in a one-shot Bertrand game with homogeneous goods.There are four treatments in their experiment.In the Ideal treatment, no communication among firms is permitted, and there is no opportunity for whistle-blowing.Standard allows communication among firms, but there is no fine reduction from reporting a cartel.Leniency awards a fine reduction to firms that report their cartel.If only one firm reports, it is fully exempt from paying the fine (which is 10% of revenue).If there are k > 1 whistle-blowers, each receives a 1/k reduction in its fine.
Bonus transfers the fine from those cartel members that are fined to whistle-blowers as a reward. 3The results show that the Leniency treatment has the lowest rate of cartelization among the treatments, and also leads to lower prices than the Standard or Bonus conditions.Hinloopen and Soetevent (2008b) consider behavior in a repeated Bertrand game, extending the setting of Apesteguia et al. (2007) to repeated interaction.Their design has four treatments: Under Benchmark, no communication is permitted.Under Communication, firms can communicate in every period.In Antitrust, a 15% probability of detection is introduced; and in Leniency, firms in the cartel can blow the whistle and obtain an exemption from their fine.There is a full fine exemption for the first applicant and a 50% fine reduction for the second applicant, as in two of our treatments.Hinloopen and Soetevent (2008b) find that Leniency reduces cartel formation and destabilizes existing cartels, but fails to reduce cartel recidivism from the level observed in Antitrust.
Bigoni et al. ( 2012) compare leniency policies in a repeated Bertrand competition with two firms selling differentiated products.After deciding whether or not to communicate with the competitor, firms choose their prices.The treatments include a L-Faire condition, in which firms are free to collude, and a Fine treatment in which there is a probability of detection and penalty.In the Leniency treatment, if one firm reports the cartel, it receives full leniency, while if both report, each receives leniency equal to 50% of the fine.In the Reward treatment, if there is only one firm reporting, it receives a reward that is equal to the fine that is paid by the other firm.There are two chances to blow the whistle: before and also after setting prices.The authors find, as in the prior studies, that Leniency does reduce cartel incidence compared to the Fine treatment.Fine, in turn, leads to lower cartelization than L-Faire.Exempting the ringleader, defined as the first firm to start communication, from leniency, and well as increasing fines in the absence of leniency, have no additional deterrent power.Bigoni et al. (2015) study the effects of leniency, fine size, and detection probability.Their design allows some interaction effects to be studied that we cannot include, such as the interaction between the fine size and the probability of detection; and the interaction between leniency policy and the probability of detection.However, like the studies mentioned above, they use a Bertrand pricing paradigm, which is quite different from the Cournot setup that we employ.It differs both in underlying theoretical structure (the Bertrand game is one of strategic complements while the Cournot game is one of strategic substitutes), and in its propensity for collusive behavior (Bertrand competition leads to more collusion than does Cournot competition).In the Bigoni et al. (2015) design, there are two whistle-blowing opportunities: one before prices are set, and one afterward.They find that both higher fines and leniency help deter cartels, and fines tend to be more effective under leniency.They note that low fines can be counterproductive to the deterrence of cartels, since they may be used as punishments to stabilize rather than to deter cartels.

3
Leniency Policies and Cartel Success: An Experiment Chowdhury and Wandschneider (2018) study conditions with: (i) a low cartel detection probability and high fine size; and (ii) a high detection probability and low fine size.Each combination is in a setting with and without leniency.Detection probability and fine size are not varied independently while holding the other variable constant.Their setup is also a repeated Bertrand setting with a random ending rule.They find that under leniency, there are fewer cartels under low detection rates and high penalties, but observe the reverse pattern in the absence of leniency.Andres et al. (2021) observe the contrasting finding that leniency does not have an effect on cartelization.Their experimental design has the distinctive feature that a human experimental participant is placed in the role of a regulator who can award leniency at his/her own discretion after reviewing the content of the communication between firms.Their setup does not include a voting stage where firms choose whether or not to join a cartel, and the authors argue that not including this stage makes cartel formation more difficult.Their design also includes open communication, which they argue builds trust and reduces whistle-blowing.Bodnar et al. (2023) are concerned with the effect of the ability to sue colluding firms for damages on the effectiveness of a leniency policy.In their leniency system, the first whistle blower receives a full fine exemption, and the second receives a 50% reduction.Damages are set at 60% of the difference between the cartel and the Nash equilibrium revenues, summed over the life of the cartel.The damages are won and awarded with probability 0.95.All cartel members must share the costs equally.They find that private damages significantly reduce the likelihood of cartel formation.Damages also reduce the number of applications for leniency and lower prices.The study also compares structured and free-form communication, and finds that cartel stability and leniency applications are lower under chat communication, though the overall amount of collusion is higher under more restricted, structured communication.Hamaguchi et al. (2009) study how the effectiveness of leniency policies is affected by the number of firms in the industry, and by the percentage of the fine that is exempted to firms that report the cartel.They also compare a policy of granting leniency only to the first reporter versus to all reporters.Their experiment is framed as a prisoner's dilemma with two possible actions that are labeled with the abstract terms A and B. They observe that leniency does reduce the incidence of cartels, but that there is no difference between granting full or partial leniency, or between giving leniency to one or all reporters.They also find that giving a reward to reporters reduces the incidence of cartels more than merely granting exemptions from fines.
Hinloopen and Soetevent (2008a) investigate two-firm oligopolies in a setting with multiple equilibria.Like Hamaguchi et al. (2009), they use a two-action prisoner's dilemma, rather than a price setting framing.There are four treatments in the experiment: Under the Benchmark treatment, there is no penalty for collusion, Under Antitrust, there is a probability of detection of 40%.In the Exploitable treatment, there is full leniency for a firm who is the only one to report, and 90% leniency for the two firms if they both report.In the Non-Exploitable treatment, full leniency is given to the first applicant and 50% to the second, as in some of our treatments.The results show that in the Exploitable condition, firms learn to increase their earnings by colluding and then reporting on their cartel.In the Non-exploitable treatment, many pairs of firms are able to collude by taking turns monopolizing the market in successive rounds of the interaction.Feltovich and Hamaguchi (2018) are concerned with the relative power of the direct effect of whistle-blowing, which would serve to deter cartel formation, and the indirect effect of leniency in reducing the cost of exiting the collusive agreement, which might make such agreements more likely.In their experiment, being caught in collusion by the competition authority is very costly and means that the firms are unable to charge high prices in any later periods.They found that leniency significantly lowered prices and reduced cartel stability.They conclude that leniency is a good anti-collusion policy.Clemens and Rau (2019) investigate the behavior of leniency policies that exclude ringleaders from possible leniency.The idea is that doing so would discourage cartel formation.In their experiment, firms can choose to collude or not, and firms who do not collude receive the payoffs that they would as Cournot players.Firms engage in a finitely repeated game.Their design includes a baseline treatment, called AA with No-leniency, and another called LEN, in which firms sequentially have the opportunity to report the cartel.Two other treatments, RD2 and RD4, make either two or four ringleaders ineligible for leniency.The results show that discriminatory leniency policies are not effective in reducing the incidence of cartels; but the results also confirm that non-discriminatory leniency is effective in doing so.
In summary, previous experimental research on the effect of leniency programs on cartels generally find that leniency programs do reduce the likelihood that cartels are successful.Granting leniency to one or to multiple reporting firms does not seem to make a difference insofar as it has been directly compared.There is evidence that higher fines reduce cartel formation-whether or not a leniency policy is in place.Leniency policies reduce successful cartel formation both in paradigms in which prices are set from a relatively extensive menu of prices and those that are framed as two-or three-action social dilemmas.

The Basic Setup
The structure of the experiment is based on that of Clemens and Rau (2019).Subjects are assigned to groups of four that represent four firms in a market with identical products to sell.Each group has no contact with or information about any other groups in their session.Thus, each group's activity can be considered an independent observation.Sessions consist of ten periods, and group assignments are fixed for the ten periods.Participants are made aware that their session payoffs are determined by their profits up until a randomly selected final period, plus a participation fee of $5.
At the beginning of each period, group members can talk to each other in a chat box for one minute.All messages are seen by all members of the group, but not by any members of other groups.After the discussion period ends, each individual decides whether to join a cartel or not.In the experiment, this is referred to as 1 3 Leniency Policies and Cartel Success: An Experiment choosing whether or not to "join the market agreement."4Firms make their decision before knowing how others decided.If two or more subjects agree to join a cartel, then a cartel is formed, with those who agreed to collude as the members.There is an exogenous probability of 0.15 that the cartel is discovered and that all cartel members are fined.
The parameters are the same as in Clemens and Rau (2019), and are based on the following underlying structure: there are four identical firms that produce a homogeneous good.Market demand is given by the inverse demand function P = 100 − Q, where Q is the quantity that is produced by the industry.All firms have a constant marginal cost of 60.The monopoly quantity and price are 20 and 80, respectively.If all four firms join a cartel and share the profits equally, each firm would produce 5, and receive revenue of 5 × 80 = 400, making a profit of 5 × (80 − 60) = 100.If a fine of 10% of revenue is imposed on the cartel, the fine is 40 for each firm, and each firm would receive 60 as its net payoff.In the Cournot equilibrium, each firm produces a quantity of 8, and the resulting price is 68.This results in profits of 8 × (68 − 60) = 64.Cartels with two or three members result in different profit vectors.
Table 1 shows the net payoffs for each player, depending on how many firms join the cartel and whether or not it is exposed.The table indicates the final period payoffs in the benchmark No-leniency (No-len) treatment.It shows that partial cartels of two or three firms are less profitable than is the Cournot equilibrium for the cartel participants, while a full cartel of four members is the most profitable arrangement for the industry. 5e induce time discounting with the following procedure: All groups play exactly 10 periods, but the payoffs are equivalent to those that would exist in an indefinitely repeated game with a 0.1 probability of termination in each period.After the 10 periods of the session are completed, we generate a sequence of random numbers that determine the probability that a given period that has already been played would continue to count.For example, a random number is drawn for period 1 from a uniform distribution on [0, 1].If the number drawn is in the interval (0.1, 1], periods 1 and 2 both count.Then, a random number is drawn for period 2. If this number is in [0, 0.1], only periods 1 and 2 count.If the draw is > 0.1, period 3 also counts; and so on.If a number in (0.1, 1] is drawn in period 10, and the game is thus slated to continue to count, then each individual is paid an additional amount equal to the amount that they have earned in the 10 periods that have already been played.This additional payment is equal to their expected additional payment were the game to continue under the same 10% probability of ending after each period (the expected number of future periods would be 10) and assuming the same average-per-period payoffs as in the 10 periods that have already been played.Therefore, in this case, participants' total earnings for the session equal the show-up fee of $5, plus double the amount of money that they earned in periods 1 to 10.
Though there is discounting of the future at a constant discount factor of δ = 0.9, there is a unique subgame perfect equilibrium.This is because period 10 is certain to be the last period, and thus the game is finitely repeated.The only Nash equilibrium in period 10 is the Cournot equilibrium.Thus, the only subgame perfect equilibrium in the finitely repeated 10-period game is to play the Cournot equilibrium in each period regardless of past history.In any subgame where a cartel forms, all players blow the whistle on the cartel. 1 3 Leniency Policies and Cartel Success: An Experiment

Treatments
Our design consists of a total of five treatments: The differences among the five treatments are summarized in Table 2.The benchmark treatment is called No-len.In this treatment, there are no leniency policies in effect, but instead there is an exogenous 15% probability that a cartel is discovered and fined.If exposed, the cartel is fined 10% of revenue.The High1 treatment has two key features of the current US leniency system: (1) The treatment has a relatively high fine and (2) the first applicant for leniency is given a 100% fine exemption, with no other whistle-blowers' receiving any reduction in their fines.Under the High2 treatment, a second applicant receives a 50% reduction in her fine, but the condition is otherwise identical to High1.Under Low1, only the first applicant receives a 100% fine reduction, but the fine size is only 5% of revenue.Finally, the Low2 treatment has two features of the leniency program of South Korea: (1) The fine size is relatively low; and (2) there is a 50% fine exemption for the second applicant, in addition to 100% forgiveness for the first reporter.In all treatments other than No-len, the game has a second decision stage after the choice of whether or not to join a cartel.In this second stage, the subjects who have joined the cartel choose whether or not to blow the whistle on the cartel.Blowing the whistle is described to the participants as "reporting the market agreement."Firms who do not join a cartel skip this stage.If only one or none of the participants chose to join a market agreement, then this stage is also skipped, as there are no cartels that can be reported.
Under the High1 and Low1 treatments, if there is one whistle-blower, she receives full leniency and pays no fine.Thus, she receives the cartel payoff for the period.The remaining cartel members pay the fine: 10% of revenue in High, and 5% of revenue in Low.If there is more than one whistle-blower, only one among them (chosen randomly) receives full leniency.In High2 and Low2, two whistle-blowers, randomly chosen if there are more than two, receive leniency.One receives full leniency and pays no fine, and the second pays only 50% of his allotted fine.If there is only one whistle-blower, she receives 100% leniency.Each firm's earnings are equal to its profits in the market minus any fines levied for participation in a cartel.After each period, each participant receives some information about activity in the period, consisting of how many firms agreed to join the cartel, how many applied for leniency, and one's own earnings.

The Sessions
Each treatment is in effect for 10 groups.As indicted earlier, each group has four members.Thus, in total, there are 50 groups and 200 participants in the study.The experiments are implemented via Zoom using Qualtrics.We employ an experimental currency (ECU) to denominate earnings, with each 200 ECUs exchangeable for 1 US dollar at the end of the session.The show-up fee is $5 dollars, and participants earn $12 on average.The subjects are recruited from the subject pool that is maintained by the Economic Science Laboratory (ESL) at the University of Arizona, in Tucson, Arizona, USA.
A session takes on average 45 min.Each session has two or three groups who participate simultaneously.Subjects join a Zoom meeting room at the scheduled starting time of the session.They then receive a link to the experiment and are told that they would earn a show-up fee of $5 and additional money depending on their decisions during the experiment.They read the instructions and are then randomly assigned to groups of four.They then proceed through the experiment.

Hypotheses
We rely on previously obtained experimental results to formulate hypotheses.The available previous work on cartel leniency is in near-complete agreement that leniency reduces the incidence of cartels.The variable that we use as a measure of the performance of an antitrust regime is the rate of successful cartel formation.This is defined as the percentage of periods in which a cartel is formed and not exposed: either through antitrust enforcement or whistle-blowing.We view the objective of antitrust policy to be to minimize this percentage of these "successful" cartels.We first hypothesize that the result obtained in prior studies would also be observed here, and that the rate of successful cartel formation would be reduced by leniency.This is tested by comparing the rate of successful cartel formation in the No-Len treatment with those in the High1 and High2 treatments, which have the same fine size in place: Hypothesis 1 The leniency treatments High1 and High2 have fewer successful cartels than No-Len.
Our treatments vary as to whether a second whistle-blower can receive partial leniency or not.As we discussed in Sect.2, the available evidence is mixed with regard to whether granting leniency to one or to multiple whistle-blowers is more effective in preventing cartels.The evidence is also inconclusive as to whether granting full or partial leniency makes cartels less likely.Moreover, there is no direct previous comparison between awarding full leniency to the first party to report the cartel and awarding full leniency to the first and partial leniency to a second whistleblower as well.Thus, in the absence of prior evidence, we hypothesize that: Hypothesis 2 A leniency program that allows a second applicant a partial fine exemption leads to a similar likelihood of successful cartel formation as does a program that has no second-applicant possibility.Thus, the rate of successful cartel formation is not different between High1 and High2, and not different between Low1 and Low2.

3
Leniency Policies and Cartel Success: An Experiment Bigoni et al. ( 2015) observe that, under Bertrand competition, stronger penalties are more likely to deter cartels in the absence of leniency.They also find that the deterrent power of high penalties is magnified under leniency, since the benefits of leniency to the whistle blower increase as the fine avoided becomes larger.We thus hypothesize that higher fines lead to fewer successful cartels: Hypothesis 3 Under a leniency program, the rate of successful cartel formation is higher under Low1 than in High1, and higher under Low2 than in High2.

Leniency and Cartel Formation
We begin our reporting of the data by considering the frequency with which cartels form and how the likelihood of their formation is affected by leniency policy.Table 3 indicates the cartel formation rate and the percentage of periods in which a cartel is formed; a cartel occurs if two or more individuals agree to join a cartel.For each treatment, the cartel formation rate is the total number of periods in which a cartel was formed, divided by 100 (the total number of periods played under each treatment: 10 groups times 10 periods per group in each treatment.We define a Full Cartel as a cartel that includes all four firms and a Partial Cartel as a cartel with two or three members.The term All Cartels encompasses both full and partial cartels.Table 3a shows the percentage of possible instances in which a cartel-either full or Fig. 1 Cartel formation rate over the 10 periods, each treatment partial-is formed in each of the five treatments.Figure 1a illustrates how the cartel formation rate changes over time, by tracking the percentage of groups that form cartels in each period. The table shows that the cartel formation rates under High2 (84%), High1 (83%), and Low1 (81%) are similar to that under No-len (85%).Low2 (95%), however has an even higher cartel formation rate than No-len.However, Mann-Whitney U tests for pairwise treatment differences between No-len and each of the other treatments are not significant.This means that the various leniency policies do not change the likelihood of cartel formation from that in No-len. 6Indeed Figure 1 confirms that there are no obvious differences among treatments.
The data for full cartels only are shown in Table 3b and Fig. 1b.The full cartel formation rate in a treatment is the number of periods in which full cartels form, divided by the total number of periods.Full cartel formation rates are highest under No-len (49%), followed by Low2 (45%), and then in turn by Low1 (33%), High1 (32%), and High2 (27%).Figure 1b shows some tendency for the incidence of full cartels to decrease in the later periods of a session.MW tests, which are conducted between pairs of treatments, do not show any significant differences between treatments in the percentage of instances that a full cartel forms.Thus, none of our leniency policies has an effect on the incidence of cartel formation relative to a regime of No-leniency.

The Effect of Leniency Programs on Cartel Exposure and Cartel Success
We have seen that the leniency policies do not significantly affect the likelihood that a cartel forms.We now consider whether the leniency policies expose more of the cartels that do form: Table 4 presents the data on the Cartel Exposure Rate.A cartel is exposed when it is either detected by an antitrust regulator or it is revealed by a whistle-blowing action on the part of a cartel member.The cartel exposure rate is defined as the number of cartels that are exposed divided by the total number of cartels that are formed.
Panel (a) in the table shows that each of the four leniency treatments has a much higher cartel exposure rate than that under No-len.The cartel exposure rate is greatest under High1 (64%), followed by Low1 (60%), High2 (52%), Low2 (43%), and finally 6 In this paper, all comparisons between No-len and High1, as well as between No-len and High2, are one-sided, since we have a hypothesis (1) with regard to the sign of the differences between these treatment pairs.Furthermore, all comparisons between Low2 and High2, and between Low1 and High1 are also one-sided since we also have a hypothesis (3) about differences between these two treatments.Though the hypotheses refer to the cartel success rate -the primary measure of policy effectivenesswe also use one-sided tests for the cartel formation rate, cartel exposure rate, industry profit, and fine revenue, since these are all measures that are related to the cartel success rate.Thus, we test one-sided hypotheses that High1 and High2 lead to lower cartel formation rates, higher exposure rates, lower industry profit, and greater fine revenue than does No-Len.High1 exhibits the same differences relative to Low1 and High2 the same relationships with Low2.All other p-values are based on two-sided tests.In all tests, each group's activity over the 10 periods that they played is taken as one observation, so that we have ten observations under each treatment.For example, in testing whether the cartel formation rate differs between two treatments, we have ten observations in each treatment, where each observation is the percentage of periods in which a group has formed a cartel in the ten periods that the group interacted.

3
Leniency Policies and Cartel Success: An Experiment No-len (12%).MW test results (with each group as the unit of observation) indicate that each of the four leniency programs reveals significantly more cartels than does No-len (No-len vs. High1, p = 0.001; No-len vs. High2, p = 0.001; No-len vs. Low1, p < 0.001; and No-len vs. Low2, p = 0.007).However, there are no statistically significant MW test results between any pair among the four leniency treatments. 7 Table 4b and Fig. 2b include the data for full cartels only.All leniency treatments expose many more full cartels than does No-len, Full cartels are exposed at the highest rate under High1 (91%), followed by Low1 (64%), High2 (59%), and Low2 (58%).Under No-len, only 14% of full cartels are exposed.
MW test results show that High1 and Low2 break more full cartels than does Nolen (No-len vs. High1, p = 0.003; No-len vs. Low2, p = 0.032); but other treatments are Fig. 2 Cartel exposure rate over the 10 periods, each treatment 7 For this test, the cartel exposure rate is calculated as: (the number of cartels that a group makes that are exposed)/(the number of cartels that a group forms).Thus, for groups that do not form cartels, the variable is not defined, and groups for which this is the case are not included in the test.This particularly affects the comparisons of the exposure rate of full cartels, which form relatively infrequently.
not statistically different from No-len.There are no statistically significant differences among the remaining treatment pairs.Thus, some leniency policies expose more cartels than does No-leniency-with High1 the most effective among the policies.A more complete measure of the performance of a cartel mitigation policy is the Cartel Success Rate.This is calculated as the probability that both a cartel forms and is not exposed.In such a situation, the cartel can be said to be successful since it was able to avoid detection.The rate of cartel success is thus a variable that a regulator seeks to minimize since it is a measure of undetected cartels.Table 5 and Fig. 3 show the rate of cartel success in the different treatments, for all cartels as well as for full cartels only.The hypothesis presented in Sect. 4 were formulated in terms of this variable.
The data in the table show that the cartel success rate is higher under No-len (75%) than under any of the leniency policies.The lowest success rate occurs under High1 (30%), followed by Low1 (32%), High2 (40%), and Low2 (54%).As can be seen in Table 9 of Appendix A, Hypothesis 1 is supported in that the differences between No-len and the two High fine conditions are significant (No-len vs. High1 p = 0.001, No-len vs. High2 p = 0.004).It is also the case that there are significantly High1 and High2 also lead to significantly fewer successful full cartels than does No-len.This supports Hypothesis 1.There are no significant differences among the leniency policies in the full cartel success rate.All four policies yield a full cartel success rate of between 3 and 19%, as compared to 42% without leniency.

Profits and Fines
We now consider the overall payoffs to firms: these are increased by the successful formation of full cartels, reduced by the fines that are paid when the cartel is exposed, and increased by leniency awarded to reporters.The average profits, fines paid, and resulting payoffs to the industry, summed over the 10 periods that are played, by treatment, are shown in Table 6.In the table, the industry's payoff is defined as its profit minus the fines it pays.These payoffs to firms differ considerably among treatments.The highest industry payoffs are obtained in Low2, with No-len second highest.These treatments are followed by High1, Low1, and High2.
MW tests show that there are no significant pairwise differences in industry profit among No-len, Low1, and Low2 (No-len vs. Low1, p = 0.131; No-len vs. Low2 p = 0.880).However, High1 and High2 do lead to significantly lower firm payoffs than does No-len.(No-len vs. High1, p = 0.010; No-len vs. High2, p = 0.021).Low2 leads to greater industry payoffs than High1 (p = 0.008) or High2 (p = 0.004).Firms earn greater profit under Low1 than High1 (p = 0.048).Thus, a leniency policy reduces firms' payoffs relative to a No-leniency regime when the fine is sufficiently high.High fines reduce industry profit.Granting leniency for two whistle-blowers increases industry profit when the fine level is low.
Table 6 also indicates the average revenue from fines over the ten periods in each treatment.The fine revenue is greatest in High1 (533), followed by High2 (346), Low1 (219), Low2 (200),.As can be seen in Table 10 in Appendix A, MW tests reject the hypothesis that fines are the same under No-len and High1 (p = 0.001), as well as between No-len and High2 (p = 0.012).Thus, both High1 and High2 lead to greater fines than does No-len: this indicates that despite the forgiveness of fines through leniency, the revenue to the authority is greater under leniency.Whether leniency is granted to one or two whistle blowers has no impact on fine revenue: Revenue under High1 is not statistically different from revenue under High2 (p = 0.112); Low1 and Low2 also do not generate different revenue from each other (p = 0.762).Both High conditions generate greater fines than does Low2 at p < 0.05, and High1 leads to more revenue than does Low1 (p = 0.008).

Discussion
In this paper, in the context of a controlled laboratory experiment, we compare different cartel whistle-blower leniency programs.We vary the size of the fine for being caught participating in a cartel, and we vary whether one or two whistle-blowers can receive leniency from penalties.We advanced three hypotheses: (i) that leniency would reduce the likelihood that firms would successfully collude; (ii) that it would make no difference whether leniency was offered to one or to two whistle-blowers; and (iii) that the larger is the fine size, the fewer are cartels that will be successful.The first hypothesis was strongly supported, and the other two received qualified support.
In our experiment, we observe that: (1) Leniency policies do not reduce cartel formation; (2) however, they tend to expose more cartels, and thus do reduce the probability that firms successfully collude; (3) higher fines reduce firm profits and increase the fine revenue to the state; and (4) there is no consistent difference between the effects of awarding leniency to one or to two whistle-blowers.
Our results show that cartel formation rates are not significantly affected by leniency policies.However, leniency policies do increase cartel exposure rates through whistle-blowing.The cartel success rate-which is an overall measure of effectiveness in reducing cartels-refers to the likelihood that a cartel is both formed and unexposed.This is a measure that is unobservable in the field, but is straightforward to measure in laboratory experiments.In the experiment, the data from High1 and High2 show that both leniency policies reduce full cartel success rates, reduce firms' payoffs, and increase the state's fine revenue.Hypothesis 1 is supported in the data.
Why does a leniency policy not reduce the incidence of cartels?There appear to be offsetting effects of leniency on the rate of cartel formation: leniency reduces the incentive to form a cartel, since others may report on the collusive agreement, which reduces a potential cartel member's payoffs compared to a policy of no leniency.However, it makes a strategy of joining a cartel and subsequently blowing the whistle more attractive as well.It appears that in our data, these effects offset each other, and the result is that a leniency policy does not, on balance, deter attempts to form cartels. Rather than deterring cartels, a leniency policy merely causes more cartels to be exposed, and that is why they are effective in reducing cartel success.
The cartel success rates-which depend directly on how much whistle blowing occurs-are not different when leniency is granted to one or to two whistle blowers when the penalties are high.While cartels enjoy a higher rate of success with two whistle-blowers when fines are low, there is no difference for the full cartelswhich are the ones that are profitable.Hypothesis 2 is therefore mostly supported.It appears that a marginal incentive of a 50% reduction in the second whistle-blower's fine is not a strong enough incentive to alter collusive behavior.It is a small difference in the overall incentive to report on the cartel since it applies only if a firm one is the second whistle-blower: This is an event that is not very likely, and the fine reduction is only partial.Phase 2 If at least two companies have chosen to join the market agreement, an agreement is made and the round moves on to phase 2. Only those who have entered the market agreement participate in phase 2. In phase 2, each company that has entered the market agreement can choose whether or not to report the agreement.Based on whether or not you and others have reported the agreement, you may lose some of the earnings that you had at the end of phase 1.
• If you choose to report the agreement and you are the first to report, then you do not lose any earnings.• If you choose to report the agreement but are not the first to report, then you will lose some of your phase 1 earnings.The amount you lose can be calculated using the table below.• If you choose not to report the agreement but one of the other companies in the agreement reports, then you will lose some of your phase 1 earnings.The amount you lose can be calculated using the table below.• If no companies in the agreement report, then there is an 85% chance that none of them will lose any earnings.However, there is a 15% chance that the agreement is discovered by a market monitor and then all of them including you lose some of your phase 1 earnings.The amount you lose can be calculated using the table below.
Those companies that have not entered the market agreement cannot report the agreement, and cannot lose any of their phase 1 earnings.If fewer than two companies join the market agreement, there is no phase 2 for the round.

3
Leniency Policies and Cartel Success: An Experiment

How to Calculate Your Earnings for Each Round
You can use the following table to help you make your decisions.The rows in the table correspond to the number of companies in your market that have chosen to enter the market agreement.The first column shows the number of companies that have entered the market agreement.Column 2 shows the phase one payoff if you enter the market agreement and how it depends on how many other companies have also entered the agreement.Column 3 contains your final earnings for the round if no company reports the agreement or if you are the first to report the agreement.In these cases, your final earnings for the round are equal to your phase 1 earnings.In the fourth column, you can see what you earn if another company is the first to report or if the agreement is discovered by a market monitor.Finally, column 5 shows your earnings if you do not enter the agreement.

How Many Rounds Count Toward Your 'Final Earnings'
In the experiment, you will participate in 10 rounds.However, they may not all count toward your earnings.Imagine that a 10-sided die, with each number 1 to 10 on exactly one side, is rolled after each round.If it comes up 1, the round that has just finished becomes the last round that counts toward your earnings.If the die comes up 2, 3, , 10, the next round also counts.Therefore, there is always a 90% chance that the next round counts, no matter what round you are currently playing.This means that there is a 100% chance that round 1 will count, a 90% chance that round 2 will count, a 0.9 × 0.9 = 81% chance that round 3 will count, a (0.9) 3 = 73% chance that round three will count, etc.
You do not see these die rolls, and you will not know which round is the last one that counts until the experiment ends.If the die does not come up 1 for 10 rounds, then you will receive the earnings from all 10 rounds you play, as well as an additional amount equal to your earnings for the 10 rounds.So in that case, you get 2 times your earnings over the 10 rounds.There is a 35% chance of this occurring.
Let's wait for other players by clicking "Next" and start the game! 1 3 first or only whistle-blower, this firm also obtains 59.If another firm blows the whistle first or the cartel is uncovered by a regulator's monitoring, then the firms in the cartel each get 25.The other situations and the other tables are read and interpreted similarly.See Figs. 4,5,6,7,and 8.
Fig. 4 The payoff table for the No-len treatment Fig. 5 The payoff table for the High1 treatment Fig. 6 The payoff table for the High2 treatment

Table 1
Period payoffs of cartel members and non-members based on size of cartel Payoffs are denominated in terms of experimental currency.200 ECU = 1 US dollar.If all of the four firms join a cartel, the revenue of each firm is 400.If the cartel is exposed, each cartel member is fined 10% of its revenue, 40.Thus, each firm's payoff becomes 100 − 40 = 60.Similarly, 10% of revenue is considered as the fine level for cases in which two or three firms form a cartel

Table 2
Differences

Table 3
Cartel formation rate, by treatment *Cartel formation rate = Number of Cartels formed/100 periods (= 10 periods × 10 groups per treatment) **Full cartel formation rate = Number of full cartels formed/100 periods

Table 4
Cartel exposure rate *Cartel exposure rate = exposed cartels/cartels formed **Full cartel exposure rate = exposed full cartels/full cartels formed

Table 5
Cartel success rate, by treatment *Cartel success rate = Number of unexposed cartels divided by 100 (= 10 periods × 10 groups per treatment) **Full cartel success rate = Number of unexposed full cartels divided by 100 Cartel success rate over the ten periods, all treatments

Table 6
Average industry profits, fine, and payoff, each treatment

Table 7 p
Values from Mann-Whitney U tests of treatment differences in cartel formation rate

Table 8 p
Values of Mann-Whitney U tests of pairwise treatment differences in cartel exposure rate

Table 9 p
Values of Mann-Whitney U tests of pairwise treatment differences in cartel success rate

Table 10 p
Values of Mann-Whitney U tests of pairwise treatment differences in industry payoffs and fines paid If three companies join the market agreement and one company does not: Those who joined each receive 59 ECU.Those who do not join receive 178 ECU.• If two companies join the market agreement and two companies do not: Those who joined receive 50 ECU.Those who do not join receive 100 ECU.• If one company tries to join the market agreement and three companies do not: All four companies receive 64 ECU • If no companies try to join the market agreement: All four companies receive 64 ECU. •