Faculty of Business and Economics
MSc Economics: Industrial Organisation, Regulation and
Competition Policy
To collude or not to collude?
The dilemma faced by a manager who
could see her pockets lightened
By Filippo Bernardoni
Student Number: 10827358
Master Thesis
Supervisors:
prof. dr. J. Hinloopen
dr. A. M. Onderstal
25/01/2016 – AmsterdamTable of Contents:
1 - INTRODUCTION AND MOTIVATION …...3
2 - LITERATURE REVIEW...7
2.1 - BACKGROUNDTO LENIENCY PROGRAMS... 7
2.2 - THEORY OF LENIENCY PROGRAMS... 9
2.2.1 - Ex-Ante Deterrence... 9
2.2.2. - Post-Conviction Behaviour... 11
2.2.3. -Negative Side Effects... 12
2.3 - EXPERIMENTAL EVIDENCE... 12
2.4 - LEGAL FRAMEWORK: A COMPARISON BETWEEN THE EU AND US LEGISLATIONS...13
3 - EXPERIMENTAL PROTOCOL AND PROCEDURE...15
4 - HYPOTHESIS...19
5 - RESULTS AND ANALYSIS...20
6 - CONCLUSIONS...29
7 - BIBLIOGRAPHY...32
ACKNOWLEDGEMENTS...37
APPENDIX A- EXPERIMENTAL INSTRUCTIONS AND DOCUMENTS...38
APPENDIX B – ANALYSIS STATA DO-FILE...44
1 - Introduction and Motivation
Ever since the birth of Antitrust Policy back in 1890 with the passage of the Sherman Act by the US Congress, cartels have been the focus of attention of Antitrust Authorities worldwide. Whether involved in price-fixing, sharing the markets or restraining production, cartels are commonly considered by economists as undertakings reducing economic efficiency.1
Antitrust Authorities have traditionally fought cartels by imposing fines after a successful investigation: the success of this approach is conditional on the ability of the Authorities of proving, through hard evidence, the coordination of anti-competitive actions by the
companies involved in the cartel. Given the hardship of collecting incriminating meaningful hard evidence, since 1993 most Antitrust Authorities2 have adopted Leniency Programs as
main tool for fighting cartels.
Leniency Programs can be generally defined as schemes for discounts3 on fines
imposed on cartel members in exchange of incriminating proofs of collusion. The main idea behind Leniency Programs is therefore rewarding, whether through actual rewards or through discounts of fines, cartelists for their collaboration with the investigative authorities in charge of proving cartel activity. It has always to be juxtaposed with supplementary hard evidence and investigation on the reliability of the delation.
The theory of Leniency Programs4 describes two different and contrary results
stemming from the implementation of such programs: on the one hand, the value of being a member of a cartel is decreased due to the possibility of being reported to Antitrust Authorities by fellow cartelists, and subsequently being convicted and paying a fine5; on
1 As stated in the Section 1 of the Sherman Act, 1890: “Every contract, combination in the form of trust or otherwise, or conspiracy in restain of trade or commerce among the several States, or with foreign nations, is declared to be illegal.”
2 Bryant and Eckard (1991) and Combe and Monnier (2008) respectively found detection probability to be comprised in the 13%-17% and 12.9%-13.3% range. In addition, 1993 was a pivotal year since the US Leniency Program was modified and augmented with the possibility for companies to report after an investigation was launched: this extension proved to be very helpful to both US Antitrust Authorities, the Federal Trade Commission (FTC) and the Department of Justice (DoJ), and was subsequently introduced in other countries. The previous program, allowing companies to provide information only before an investigation had been launched, was not effective: the hardship of proving collusion could not be dampened by the additional information companies possess. As reported by Rey (2003), in 1999 the Department of Justice alone secured $ 1 billion, more than the cumulative fines in the previous 109 years since the entry into force of the Sherman Act.
3 Leniency Programs can greatly vary in many aspects: the magnitude of fine discounts, the number of applicants that can apply for leniency, the timing of application, whether the ringleader can apply or not, specific offences that are excluded from leniency, etc.
4 For an exhaustive argumentation see D'Apremont and Motta (2000)Spagnolo (2000a), Spagnolo (2000b), Motta and Polo (2003), Spagnolo (2004), Buccirossi and Spagnolo (2006), Spagnolo (2006)
5 As discussed by Kovacic et al. (2006), managers tend to keep the incriminating evidence that can be appealed on in court. This fact explains the source of the decreased expected value of cartel membership under a Leniency
the other hand, the value of being a member of a cartel is increased due to the possibility of reporting and therefore paying a reduced fine, if not being completely exempted from it.
Leniency Programs therefore create a double-edged sword hanging over colluding firms: the basic trade-off of increased exposure to legal actions against the possibility of “free exit” from such legal implications.
A theoretical model cannot “a priori” predict which of these two effects would actually prevail in practice, since many firm-specific and industry-specific details may determine the predominance of either of the effects. The practical effects of Leniency Program also vary greatly depending on the details and design of the policy. Therefore the effects of Leniency Programs have been studied and tested in laboratory experiments since the seminal work of Apesteguia et al. (2007).
The experimental approach to the issues and topics of cartel activity is of extreme help: as anticipated before, the theory of cartels is not conclusive and ambiguous, while the use of the laboratory provides a good proxy for testing the willingness of firms to take part in non-binding price communication. It is also superior to field evidence: it is impossible to retrieve information on yet-to-be-disclosed cartels while also frequent changes in
legislation for appraising different features are unfeasible.
The experimental strand of literature has shown partly conflicting results, as conflicting and ambiguous have been the results achieved by the researchers: this variety is due to the many different features and experimental protocols implemented.
Nevertheless, from some recent papers, specifically Bigoni et al. (2012) and Dijkstra et al. (2014), it seems that “[...] subjects achieve remarkable sophistication in the agreements
they make”6,provided opportunities of rich communication and report after an investigation has been launched. Sophisticated agreements allow the creation of stable cartels that are the most detrimental in terms of welfare. The assumptions described above are very concrete, calling for a serious appraisal of the issue of cartel sophistication. Both the US Leniency Program, reformed in 1993, and the EU Leniency Program, modified in 2002, allow the report to the Antitrust Authority after an investigation has started. In addition, as well underlined by Connor (2001) in his paper on the Lysine Cartel, rich communication is pivotal for the development of a stable and profitable cartel.
This means that those cartels that do not dissolve in their initial phase could achieve a noticeable stability even in the presence of Leniency Programs: stability and sophistication
Program.
coupled can lead to the formation of hard core cartels7, whose member are not likely to
report and that, in case of conviction by the Antitrust Authority, can retain high prices through means of tacit collusion (Harrington, 2004). It is clear that these well-established and hard core cartels have negative effects on social welfare and economic efficiency.8
Therefore, in my opinion, it is appropriate to raise the level of sophistication of the tools deployed by Antitrust Authorities.
“Could the extension of fines, and accordingly leniency, to managers improve the effectiveness of Antitrust Policy?”
This topic is of great importance since it represent the main difference between the current US and EU legislation in the field of Antitrust Policy: in addition, as reported by Spratling (1999) and Hammond (2000), since the introduction of managerial liability in the reform of US Leniency in 1993, “one of the major reasons of the success […] is the fear of
imprisonment for corporate officials.”9 Hence, the importance of testing whether such a feature is to be added to the EU Leniency Guidelines or not has to be further studied in order to have a clearer and more precise understanding of the issue.
The alignment of legislations also constitutes a tightening of the “internal governance problem” faced by cartel and well explained by Spagnolo (2000a,b): these two influential papers present the specific aspects that differentiate cartels, and corporate crime in general, from common crimes committed by individual wrongdoers.
The first of these features is the intrinsic governance problem faced by cartels: in order to successfully raise their profits through the implementation of collusive strategies, cartel members need to cooperate. Notwithstanding, cooperation cannot be enforced through explicit contracts given the unlawful nature of such actions. This impossibility, coupled with individual incentives to cheat on the collusive agreements, determines the intrinsic
instability of cartels as stated by Stigler (1960).
Second is the monitorability problem: internal cohesion and observance of the
agreements must be internally enforced, without any legal tool, by cartelists themselves, or, in other words, self-enforcing. The impossibility of monitoring the actions of others, hence ascertaining whether they stick to the agreements or not, shifts the scope of
7 Dijkstra et al. (2014) 8 Spagnolo (2004)
9 Kovacic et al. (2006) p.3; both Hammond and Spratling are officials of the Antitrust Division of the Department of Justice in the US.
collusive agreements towards a term future, hence the definition of cartel as a
long-term dynamic criminal relationship.10
Third, and partially stemming from the second, by misbehaving together, cartelists end up having information on the unlawful actions committed by their fellows. This last point is the most important when considering the structure and implementation of a Leniency Program: it must be designed in a way that incentivises and facilitates the transmission of such information to the Antitrust Authority.
In my opinion it is thus important to appraise the potential benefits that the extension of fines to the managers that lead firms throughout collusive agreements: it is relevant not only because it is the main legislative difference between the European and American Leniency Programs, but also because it theoretically represents a positive tightening of both the corporate incentives to collude as well as the agency problems faced by a firm that has to implement collusive practices within its managerial hierarchy.
The remainder of the paper is structured as follows: chapter 2 presents the related literature. Chapter 3 describes the experimental protocol and details. Chapter 4 presents the hypotheses of the paper, while Chapter 5 discusses and analyses the data collected in the experimental sessions. Finally, Chapter 6 covers the results of data analysis,
presenting some explanations for the results obtained, while Chapter 7 gives some conclusive remarks and ideas for future research.
2 - Literature Review
The literature covering Leniency Programs and their effects is wide and mainly twofold: it aroused to importance at the beginning of the third millennium with the theoretical
analyses of Spagnolo (2000), Motta and Polo (2003) and Spagnolo (2006), while, from the pioneering work of Apesteguia et al. (2007), the experimental strand of literature has been active in testing the effects that theoretical works bestow to Leniency Programs.
Section 2.1 covers the background to Leniency Programs. Section 2.2 presents the theoretical discussion of Leniency Programs, separating ex-ante deterrence (2.2.1), post-conviction behaviour (2.2.2) and negative side effects (2.2.3). Section 2.3 deals with the results obtained by the experimental strand of literature, while Section 2.4 is a legislative comparison between the Leniency Programs deployed in EU and US.
2.1 - Background to Leniency Programs
In order to fully understand the use and impact of Leniency Programs, the first thing to
be done is analysing the effects that cartels have on the overall welfare of an economy: cartels, defined by Rey (2003) as “a joint profit maximising entities”, are detrimental to economic welfare since they create both allocative and dynamic inefficiency, resulting in “billions of dollars of overcharges to consumers around the world”.11 This means that
production is suboptimal, as a result of price-fixing or restrained production, and that also innovation, whether product innovation or process innovation, is sacrificed on the altar of joint profit maximisation and anti-competitive market practices.
Other than their effects on the market, another important feature of cartels that is pivotal in understanding the use and impact of Leniency Programs is cartels' internal governance problem.12
As defined by Spagnolo, cartels are “long-term dynamic criminal organisation facing an internal governance problem”: this definition allows to better understand why the use of Leniency Programs in Antitrust Policy is a powerful tool for destabilising the internal equilibrium reached by colluding firms.
As sharply noted by Spagnolo, the impossibility of writing enforceable contracts for collusive agreements leaves unsolved, and unsolvable, the governance problem since there is no legal mean for a cartel to prosecute those members that do not stick to the agreement. In addition, even if jointly profitable, any unilateral deviation from the 11 OECD (2001).
agreement is individually profitable in the short run, before punishment mechanisms can be deployed by cheated-upon cartelists. The creation of an internally stable cartel is therefore a process that requires time for cartelists to know and trust each other.
The internal stability of a cartel is directly linked to the cartel ability of imposing
compliance to the agreement dealt upon its members, rather than unilaterally deviate and increasing their own profits at expenses of the fellow cartelists. This condition is satisfied when
E(ΠC) – E(AC) ≥ 0 ; (PC)
and
E(ΠC) – E(AC) ≥ E(Πd) − E(Ad) ; (ICC)13
Both the Participation Constraint (PC) and the Incentive Compatibility Constraint (ICC) must be respectively satisfied for any member for a cartel to be internally and dynamically stable.
Here in this framework the design of Leniency Programs comes under the focus of attention: if properly implemented and set up, Leniency Programs tighten the ICC. They reduce, if not drive to zero14, the expected legal cost of deviation, E(Ad).
The optimal Leniency Program is therefore able of deterring cartel formation “a priori” by invalidating the ICC: the result are fewer short-lived cartels that collapse in their initial stages of life due to the excess number of reports and deviation that the Leniency Program instil through its ex-ante deterrence effect.
Nevertheless, for the implementation of a successful Leniency Program some conditions and requirements need to be in place. As reported by Hammond (2004), the reform of the US Corporate Leniency in 1993 was conducted with three principles in mind: the thr\eat of severe sanctions, a law-enforcing environment and the need for
transparency. Each of these hinges is fundamental in developing a Leniency Program able of deterring cartel formation and ensuring a short and troublesome life to those cartels that
13 Respectively the Participation constraint and the Incentive Compatibility constraint, where c labels collusion and d labels deviation. Π defines profits while A defines the Antitrust cost, including fines, that the firm could bear in that scenario. Both must be satisfied for a cartel to be stable.
14 Actually, even if not implemented in reality mainly for political and ethical arguments concerning the
appropriateness of rewarding wrongdoers, the best Leniency Program reverts the sign of the expected legal cost of Antitrust since it distributes to the reporting firm(s) the fines imposed on the reported firms.
eventually form. The fines imposed should be severe so that they can financially recover the damages created by collusion; the Antitrust Authority should be active in fighting and investigating cartels in order to create a concrete risk of detection; finally, the mechanisms of the Leniency Program should be clear and predictable, so that firms and officials
applying for it can foresee with high certainty the treatments they will receive by the Antitrust Authority.
Now the discussion turns to the theory of Leniency and the effects Leniency Program have on agents in the market and the outcome of the market.
2.2 - Theory of Leniency Programs
The theory of Leniency Programs15 provides mixed evidence about the effectiveness16
of such programs in Antitrust Policy. Nevertheless, it is impossible to “a priori” ascertain which of these two effects prevails17. Despite this impossibility, the theory of Leniency
Programs separates two main aspects that have to be carefully understood and coped with when designing a Leniency Program: Ex-Ante Deterrence and Post-Conviction Behaviour.
2.2.1 - Ex-Ante Deterrence
Ex-Ante Deterrence is the ability of a Leniency Program of discouraging cartel formation through the hardship of the legal punishments prescribed for such anti-competitive law infringements: it is also the main criterion for evaluating the effectiveness of a Leniency Program, since the complete abstention from any collusive agreement whatsoever is the ultimate goal of Antitrust Policy.
The theory of Leniency Programs summarises seven notable effects related to ex-ante deterrence:
Reward to Whistle-Blower Effect: as argued by Spagnolo (2000), the restriction of
lenient treatment only to the first firm applying and paying it the sum of the fines collected from the others is the first best solution in deterring cartels: sufficiently high rewards to parties that spontaneously report may deter all cartels and other illegal business practices at no cost at all, since what is paid as rewards is nothing else than what is collected in
15 For an exhaustive argumentation see D'Apremont and Motta (2000)Spagnolo (2000a), Spagnolo (2000b), Motta and Polo (2003), Spagnolo (2004), Buccirossi and Spagnolo (2006), Spagnolo (2006)
16 Nevertheless, the implementation of Leniency Programs in every major developed economy and their successes balance the trade-off between positive and negative effects of Leniency towards the former, Brenner (2009). 17 The actual prevalence of either of the effects is determined by real-world features, encompassing market size, the
number of firms in a given industry, the industrial history and developments of the specific industry, the overall economic situation, etc.
fines. It is hardly observed in real-world cases because of political and ethical considerations.
Protection from Fines Effect: as described by Spagnolo (2004), despite
constraints on the implementation of a first best solution, there is still room for the implementation of moderate Leniency Programs with positive ex-ante deterrence effect. This effect, also called “Deviator Amnesty Effect” by Harrington (2008), ensures that, within the framework of second best law enforcement policies, the whistle-blower is granted full exemption from fine payments: this feature further tightens the ICC by driving to zero the expected legal cost of deviation, thus allowing an optimal deviation coupled with report, detrimental for cartel stability.
The provision of coupling deviation with a report granting full exemption was already stressed by Ellis and Wilson (2001) and Hinloopen (2003).
Protection from Punishment Effect: within the frame of second best law
enforcement policies by Spagnolo (2004), this effect partially protects deviating firms from the punishment they would have incurred by other cartelists once deviation is detected. Its dynamic is very similar to the Protection from Fines effect: enhanced when report is
restricted to the first applicant only and allowing the combination of deviation and report. Nevertheless, given the vast array of punishment strategies that could be carried out, the actual protection is variable.
Increased Risk Effect: a moderate Leniency Program that provides decreasingly
discounted fines depending on the order of application can enhance the perception of risk associated with the entry and keeping of an anti-competitive agreement. This increased perception of the risk associated with cartel membership can lead to cartel breakdowns before a given level of mutual trust is reached by cartelists. Restricting lenient treatment to the first applicant enhances the ex-ante deterrence and the leniency effectiveness.
Raise Rivals' Cost Effect: as sharply observed by Ellis and Wilson (2001) in their
dynamic oligopoly model à la Bertrand, self-reporting cartel membership can have some significant strategic advantages and can help building a strong market position18:
self-reporting imposes to the other firms high legal expenditures, while also dampening their operations and strategic plans by imposing jail conviction to executives and other
reputational costs. This effect increases in the fine level and the market size, while decreases in market concentration.
18 It has to be noted how Ellis and Wilson (2001) refer to the US Jurisdiction analysing the US Leniency Program: the strategic advantage is therefore strengthened by the fact that executives incur in criminal penalties for leading their company into a collusive agreement. Penalties that encompass convictions are to be served in federal jails.
Rush to Courthouse Effect: as for the Increased Risk effect, the perception of the
risk associated with cartel membership is crucial: as stated in Harrington (2008), in a dynamic model where leniency is restricted to first applicant only, the perception of the risk leads to a literal rush to the courthouse for being the first filing the application. Hence, deterrence increases in the perceived riskiness and decreases in the number of lenient treatments granted: the rush to the courthouse would not happen, or happen to a lower extent, if arriving second would still grant a discount on the fine to be paid.
Reward for Employees Effect: this effect is closely related to the Reward to
Whistle-Blower effect, but differs in the extent of such rewards. While a general reward to the firm can reach a first best solution, rewarding individual employees can leverage on the internal agency problems. As noted by Aubert et al. (2006), “the Antitrust Authority can try take advantage of and even exacerbate these agency problems in order to deter collusion more effectively". This actually increases the risk connected to implement collusion, while also raises the cost of collusion since “colluding firms will have to bribe informed employees in order to secure their fidelity."
2.2.2. - Post-Conviction Behaviour
If Leniency Programs are evaluated through their ultimate goal of Ex-Ante Deterrence, another very important aspect of the topic is Post-Conviction Behaviour: it is relevant to understand the dynamics that drive former cartelists' behaviour both in terms of price evolution and recidivism.
As reported by Sproul (1993)19, there has been an observed average 7% price increase
in 25 industries whose members had been convicted on the grounds of cartel activity. It is therefore critical for the design of an effective Leniency Program to understand how firms react to the imposition of fines and how their behaviour is affected. Such a prominence is highlighted by Harrington (2008), that defines post desistance as a dynamic part of ex-ante deterrence, hence stressing how the deterrence of cartel activity is a goal that has to be achieved by invalidating the ICC for potential cartelists in a dynamic setting that
considers the reactions and behavioural adjustments that conviction elicits.
Relative to the evolution of prices after conviction a possible explanation of the average price rise described by Sproul (1993) is offered by Baliga et al. (2005) and Offerman and Potters (2006) in their discussion of Antitrust fines as sunk cost. They argue how
managers suffer from a sunk cost bias: the payments of high sunk costs leads to an 19 This paper analysed the Post-Conviction Behaviour of industries convicted under the then in force Leniency
increase in prices as well as the undertaken of (over)risky behaviour. Antitrust fines could be considered as sunk cost, since they are incurred once conviction is official and
definitive, and cannot be recovered. This argument is strengthened by the common managerial practice of full-cost pricing rather than the economic efficient variable-cost pricing20.
2.2.3. Negative Side Effects
As previously written, the Theory of Leniency Programs is not exhausting and of
univocal interpretation. Indeed, it provides mixed evidence of the effectiveness of leniency programs. The bad design of Leniency Programs can be extremely harmful: excessive rewards or number of applicants or lack of diligence in checking the applications filed can nullify, if not reverting, the beneficial effects Leniency Programs have in Antitrust Policy. There are therefore some negative side effects of misimplemented Leniency Programs. First, as Spagnolo (2004) describes, Leniency Programs can be “exploitable”, that is “allowing firms to increase the value of collusive agreements by reporting it to the law enforcing agency.”21 An exploitable Leniency Program is therefore of great harm for
consumers in the economy since it completely overrides the beneficial effects of leniency, while actually being beneficial to cartelists.
Second, a badly written Leniency Program can lead to a Cartel Amnesty Effect: all cartel members could simultaneously file the report application and being granted a discount on fine. Notwithstanding, as discussed by Harrington (2008), it seems that both the Deviator Amnesty Effect and the Rush to Courthouse Effect dominate it.
2.3 - Experimental Evidence
The experimental strand of literature, pioneered by Apestuguia (2007), has focused on testing the magnitude of the theoretical effects described in section 2.2 and the different policy features that can be implemented into a Leniency Program.
Apesteguia (2007) found a small non significant deterrence effect, whereas there also was a positive effect on prices. These results have to be interpreted with caution since he used one-shot Betrand interaction, thus lacking to include into the analysis an important aspect such as the dynamic nature of an ongoing cartel.
Hinloopen and Soetevent (2008) adjusted this mistake by developing a 20-rounds repeated market interaction à la Bertrand with a decreasingly discounted fines in case of
20 Balakrishnan, R., & Sivaramakrishnan, K. (2002).
multiple leniency application; they used a communication protocol allowing for the exchange of minimum-acceptance prices. Their findings are consistent with the theory, presenting a significant difference both on ex-ante deterrence and prices when a Leniency Program is in place. The work of Bigoni et al. (2012), extending the literature by allowing coupling deviation and report, confirms the theory. Notwithstanding, the “protection from fines effect” is only partially confirmed since 30% of the players deviating did not couple it with the report to the Antitrust.
The results relative to post-conviction behaviour are mixed: both Hinloopen and
Soetevent (2008) and Bigoni et al. (2012) confirm higher prices in cartels that reform under Leniency Programs in comparison to those that reform under traditional Antitrust Policy. On the other hand, the desistance effect, hence the degree of recidivism of previously-convicted cartelists, is significant in Bigoni et al. (2012), while it is not in Hinloopen and Soetevent (2008).
2.4 - Legal Framework: a comparison between the EU and US Legislations
As briefly mentioned in the introduction, what differentiates the most the Leniency Programs deployed in the European Union and the US is the different approach to
managerial liability and, accordingly, the different treatment executives receive once their companies are found guilty of collusion. It is therefore worthy to undertake a comparison between the two different legislations in force.22
The US Corporate Leniency Program23, as reformed in 1993, comprises norms and
principles that apply to both corporations and officials and is juxtaposed by the Amnesty Plus Act. Under this program, leniency is granted only to the first firm reporting the collusive agreements it is involved in and the opportunity of reporting is granted both before and after24 an investigation has been launched. The corporate fine amounts to a
maximum of $ 100 millions, while the individual fine for officials amounts to a maximum of $ 1 million. Officials also risk a maximum of 10 years to be served in federal detention centres. In addition, a complete protection from criminal prosecution is granted to all corporate employees, regardless of their position in the corporate hierarchy, if they actively collaborate with the Antitrust Division.
22 This comparison is also motivated by the fact that these are the two oldest and most active Programs, reflecting the importance of the respective economies in the world. In addition, particularly for the US Leniency Program, they also represent the ground-setting legislations on which many other Programs have been drawn on worldwide. 23 US DoJ. (1993).
24 The acceptance of a report after an investigation has been launched is conditional on the disclosure of information that lead to the conviction of cartel: the firm has to add meaningful incriminating evidence whose knowledge is not in the hands of the Department of Justice yet.
Such a program is complemented by the Amnesty Plus program that stimulates the report of all the cartels a firm is involved in and provides an opportunity for softening the measures taken against a firm that do not qualify for leniency in first place. Firms that do not qualify for leniency in a certain case can obtain a significant discount of fine upon the report of another cartel they are involved in, while maintaining the full exemption in the second case. Following the same rationale, firms that are granted leniency in a given case, if discovered active in a second cartel are wiped out of the benefits previously accrued. It is therefore in the best interest of reporting firms to disclose all the collusive agreements they take part in.
Such a Leniency Program is therefore well suited for instilling the Rush to Courthouse effect, while it also produces a significant Protection from Fine effect. In addition, as sharply noted by Festerling (2005), the US Corporate Leniency, through its section on the individual liability of employees and officials, enhances the agency problems and internal conflicts a colluding firm faces, hence resulting in an higher rate of corporate reports.
On the other hand, the EU Leniency Program as enacted by the European Commission in 2002 and subsequently modified in 200625 presents different features. Leaving aside the
absence of any norm or principle governing the individual liability of corporate officials, there are two other differences with the US Corporate Leniency.
The first, is a more generous scheme of discounts that grants partial leniency to more than the mere first reporting firm: the first firm reporting is indeed granted full amnesty, while, conditional on the order of further applications and on the significance of the value added by the information disclosed, also the following reporting firms are granted a
discount on the fine. From 30% to 50% for the second applicant, from 20% to 30% for the third applicant and from 0% to 20% for any subsequent applicant. This approach softens the Rush to Courthouse Effect and, as argued by Spagnolo (2006), it is worse than a
“first-wins-all” approach: it could actually lead to the implementation of a waiting strategy for
colluding firms. Firms, realizing that leniency is partially granted also to following reporters, may decide to wait the first firm to report and then rushing for being the second firm
reporting, hence shifting the Rush to Courthouse effect on the second place available. The second, is the lack of a tool comparable to the Amnesty Plus: European cartelists are therefore not incentivised to disclose all the cartels they are involved in at once: there is no benefit from it. This could potentially lead to less cartels discovered and dismantled.
3 - Experimental Protocol and Procedure
In this section the experimental protocol and procedure are described. First, the
protocol, that has been drawn on the work of previous researchers and adapted in order to accommodate the novelty of modelling managerial liability, is presented; second, the procedure, describing how the experiment has been carried out and the data collected, is introduced.
The core of the experimental protocol is drawn on the work of Hinloopen and Soetevent (2008): hence the experiment is set up as 20 rounds of repeated homogeneous triopoly à la Bertrand with a 80% continuation probability after round 20.26 Price choice is restricted
to p={101, …, 110} assuming constant marginal cost, c=100, and subsequent profits being π=(p-100)/L, where L is the number of players setting the lowest price.
Rounds were structured as follows:
Step 1: Communication Decision
Each subject is asked whether he wants to communicate with the other players.
Following the protocol implemented by Hinloopen and Soetevent (2008), each subject has to express whether he wishes to communicate or not within 15 seconds. If all players decide to communicate the round will advance to step 2, otherwise to step 3. In case there is no unanimous communication decision, players will wait the duration of step 2 (max. 45 seconds), and then proceed to step 3.
A cartel is established if all players agree to communicate.27
Step 2: Communication/Chat
In case of unanimous communication decision, the three firms in CG and TG1 enter a chat: the length of the chat is 45 seconds.28 Alternatively, trios in TG2 enter a process of
reiterated elimination of minimum acceptance prices inspired on Hinloopen and Soetevent (2008): players have to enter both a minimum and a maximum acceptance price in the set p={101, …, 110}, with the possibility of expressing a single price as both minimum and
26 This reflects the aim of minimizing end-of-game effects. In addition, the observations after the 20th round of
interaction are not included into the analysis.
27 An already formed cartel that has not been discovered yet can enter the communication stage normally: such a possibility alters neither the cartel definition nor its subsequent legal circumstances.
28 The initial chat protocol was drawn on Dijkstra et al. (2014): 90 seconds for the first communication, 45 for each subsequent communication happening while the cartel was active and 60 seconds for any communication taking place after a cartel was dismantled, regardless of the causes of such breakdown. The idea has been dropped due to the layers of complexity it was adding to the writing of the zTree files.
maximum. The intersection of the price ranges then becomes the price set for the following round of price post. This process continues until a single price emerges or one minute elapses.
Communication is non-binding and any player reporting his identity, insulting or discussing non relevant topics will be excluded from the experiment.
Step 3: Pricing Decision and 1st Report Chance
Each player sets the price he wishes from the set p={101, …, 110} and decides whether or not to report the communication taking place at step 2 or in any previous round as far as it occurred and was undiscovered. This step will last 30 seconds and participants will be informed that, in case they do not decide a price within the time span, their production will be automatically set to zero, hence their profits would be 0.
Step 4: Market Outcome and 2nd Report Chance
Players learn the market price and their resulting payoffs net of eventual fines triggered by a report at stage 3. A second chance for report is provided. The Antitrust Authority also conducts its investigation automatically after all the reporting decisions have been taken. The probability of being caught, prosecuted and convicted is α=15%.
Step 5: Summary
A final summary of the round comprising price choice, market prices, profits or losses net of eventual fine, cumulative profits accrued is made available.
With respect to this experimental protocol there are some clarifications that have to be stated, in order to fully understand the mechanisms and incentives behind participants' actions.
First, high prices, even if symptom of tacit collusion, cannot “per se” lead to conviction, hence a cartel is established only if all players agree to communicate: this follows the ruling legal framework that requires hard evidence of communication29 in order to lead to
successful conviction of cartelists.
Second, the Leniency Program provides full exemption to the first applicant, a 50% discount for the second, and no discount for the third. In case of simultaneous
applications30, each player will enjoy the expected fine reduction. A single applicant will be
29 It can be safely assumed that communication between players leaves behind hard evidence that can be used for successfully prosecuting cartel's members.
awarded a 100% fine reduction, two applicants will receive 50% fine reduction each, whereas three applicants will receive no reduction. This program applies both to firm-level leniency and managerial-level leniency. Firm-level fine equals 10% of the current round revenues, while the probability of being successfully caught by the Antitrust Authority is α=15%: detection by Antitrust Authority leads to sure conviction and fine.31
There are some main differences with the protocol implemented by Hinloopen and Soetevent (2008): the first, and most important, is the fact the participants are asked to play the role of the top executive implementing collusion within the firm, not the usual one of the firm colluding on a market. In order to stress such a difference I changed the
rewarding scheme for participants linking firm performance to managerial salary. With respect to this point two considerations have to be taken into account: the locus of
decision for such anti-competitive actions usually is at the highest layers of the corporate hierarchy and salaries at such a corporate level are usually threefold, comprising a fixed salary, bonuses conditional on results and stock options32.
Participants received a fixed base salary of 4 points, while the bonus will be equal to firm-level profits, hence
ManagerialSalary = base + π
iThe managerial-level fine is F=3.5. Set at this level, the fine retains the initial idea of punishing liable managers by imposing a fine equal to their bonus: 3.5 points (€ 0.35) are just above the bonus earned in the scenario of “best” collusion, with p=110 and πi= 3.3.
The second relevant difference with the protocol tested by Hinloopen and Soetevent (2008) is the communication protocol: they used a reiterated post of minimum acceptance price, whereas, in light of the results of Harrington et al. (2013) on communication
protocols and of Dijkstra et al. (2014) on the sophistication of agreements reached when rich communication is allowed, chat communication is provided. Nevertheless, given the opportunity of organising several experimental sessions, I also tested a treatment group keeping the reiterated post of minimum acceptance prices as communication protocol.
price: the choice of granting the average reduced fine to all applicants reflects the difficulty of keeping track of the application order in the laboratory. It is also motivated by the need of providing a clear set up to participants in order not to create confusion or a sense of uncertainty.
31 This last feature actually contradicts the Leniency Programs employed by the European Union: under the current legal framework firms can apply to leniency also after an investigation is launched. My decision of discarding this feature is motivated by the experimental necessity of having a simple frame for participants given that the nature of the experiment is already quite complex.
32 Stock options are outside the scope of the experiment. They would increase the complexity of the simulation, with the risk of jeopardising the quality of the observations.
This choice is motivated by the desire of reinforcing the conclusions reached by Harrington et al. (2013) that show how the less structured and freer is the communication protocol, the more effective, stable and forward-looking are the collusive agreements reached.
Three different treatment groups were therefore set up for testing the experimental hypotheses: a Control Group with chat communication and no managerial liability, hence no managerial-level fine and leniency. A first Treatment Group with chat communication and managerial liability and a second Treatment Group with reiterated elimination of minimum acceptance prices as communication protocol and managerial liability. All groups are henceforth labelled as CG, TG1 and TG2 respectively.
The main difference33 between CG, on the one hand, and TG1 and TG2, on the other, is
that Antitrust Policy impacts managerial salaries in a different way: in CG managers incur in a salary reduction only if, being a cartel active, a report happens at step 3, hence before the salary is paid. Its effect on managerial salary is on the variable part of the salary that is driven to zero by the fine imposed on the firm34. If either report at step 4 or an Antitrust
investigation happens, there is no effect for the manager. On the other hand, both in TG1 and TG2, Antitrust Policy has a twofold effect on managerial salaries: it impacts
managerial salaries as in CG, plus the fine directly imposed on managers, F=3.5 points. Therefore, both in TG1 and TG2, managers may see their salary reduced also if a report at step 4 or an Antitrust investigation happens. Therefore:
Salary = b + γπ
i; (CG)
35Salary = b + γπi - βF ;
(TG1 and TG2)36The experimental sessions have been carried out at the CREED laboratory at the Universiteit van Amsterdam the week between 13-17 July 2015: 5 sessions have been carried out for a total of 90 participants37 and an average payment of € 15.9 per participant.
33 The different communication protocol is not the main difference and does not represent the main idea of the thesis. 34 Firm-level profits are decreased by 10% of the revenues, hence always driven to zero. The bonus paid to the
manager is therefore equal to zero.
35 The degree of the impact of firm-level Antitrust Policy on managerial salary is labelled by γ. The impact can be triggered only by a secret report at step 3, otherwise the salary is paid at step 4. γ=0 if all firm simultaneously report or a firm is being reported; γ=0.5 if 2 firms, including the player's, simultaneosly report; γ=1 if no firm reports. 36 The degree of the impact of managerial-level Antitrust Policy on salaries is labelled by β. The impact can be
triggered by report both at step 3 and step 4 and by an investigation by the Antitrust Authority. β=0 if no firm reports and no AA investigation happens; β=0.5 if 2 firms, including the player's, simultaneously report; β=1 if all firms simultaneously report or a firm is being reported or an AA investigation happens.
Each session has been run with zTree.38
Due to some technical mistakes I committed in writing the zTree files I had to interrupt the first session and reschedule the following, so I have 11 observations for the Control Group, 7 for the Treatment Group 1 and 5 for the Treatment Group 2. One observation consists of the data collected from a group of three participants.
4 - Hypothesis
The general hypothesis at the core of my work is the positive effect on welfare,
stemming from the abatement of cartel activity, that the inclusion of managerial liability into Antitrust Policy generates. As discussed in Section 2.1, the introduction of managerial liability tightens both the Participation Constraint and the Incentive Compatibility Constraint by lowering E(ΠC) and raising E(AC). Firms have to compensate the risk beared by their
employees in charge of implementing collusive agreements, hence decreasing E(ΠC). At
the same time E(AC) is higher due to the fines imposed to employees implementing
collusive practices within the firm.
Despite the simplicity of such an hypothesis, the complexities related to cartel activity and the dynamic nature of collusion lead me to test more and narrower hypotheses in order to cope with the sophistication of cartels and policies deployed to deter their formation and success.
Hypothesis 1: “Market prices are higher in CG than in TG1, and they are higher in TG1
than in TG2”
Hypothesis 2: “More cartels are formed in CG than in TG1, while more in TG1 than in
TG2”
Hypothesis 3: “Cartels have a longer duration in CG than in TG1, and in TG1 than in
TG2”
Hypothesis 4: “Post-conviction prices are higher in CG than TG1, and also higher in
TG1 than in TG2 ”
groups: I valued this risk as more harmful than allowing participants to learn the rules of their group in details. 38 Fischerbacher (2007)
Hypothesis 5: “Post-conviction prices are higher in cartels that collapsed due to Antitrust
Investigation rather than in cartels collapsed due to report”
Hypothesis 6: “Price deviation are less severe and persistent in CG and TG1 than in
TG2” and “players should couple price deviation with report to the Antitrust Authority in order to enjoy the protection from fines effect”
5 - Results and Analysis
In this section the analytical results of the tests conducted for checking the hypotheses are presented along with the methodology applied to hypothesis testing. The discussion follows the sequence of hypotheses as presented in the previous paragraph.
Result 1: “Market prices are not statistically different in CG and TG1; market prices are
statistically higher in CG and TG1 than in TG2.”
The main and more visible implication of cartels is an increase in price above the competitive or oligopolistic equilibrium resulting in a loss of consumer surplus and more profits to colluding firms. By tightening the Incentive Compatibility Constraint (ICC) through the introduction of managerial liability in Antitrust Policy, lower prices should be observed in TG1 and TG2 due to such policy enhancement.
Figure 1 – Price Evolution – per treatment
Despite this premise, it already seems at first glance that such an hypothesis is only 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 100 102 104 106 108 110
Price Evolution - per treatment
Control Group Treatment Group 1 Treatment Group 2 Period P ri ce
partially confirmed: as Figure 1 depicts, the evolution of average market price is almost collinear39 between CG and TG1, while in TG2 it is clearly closer to a competitive market
behaviour.
First glance impression is confirmed by the results of Mann-Whitney U tests, that cannot reject (p-value= 0.7513) the null hypothesis that average market prices in CG and TG1 come from the same distribution, while rejecting it at any significance level for what concerns the relationship between average price distribution between CG and TG2 and between TG1 and TG2 (p-value=0.00 in both cases).
Therefore, this first part of analysis suggests two different results: first, it appears that the introduction of managerial liability does not have any effect40 on the general pricing
behaviour of firms; second, the conclusions about communication protocol expressed by Harrington et al. (2013) seem confirmed by the fact that the price distribution of TG2 is dominated by both CG and TG1, hence the more free and less structured the
communication is, the more effective and long-lasting collusive agreements are stroked.
Figure 2 - Cumulative distribution function for market prices per treatment
Result 2: “The number of cartels formed in CG and TG1 is statistically indifferent; the
number of cartels formed in CG and TG1 is higher than the one in TG2.”
39 Actually, the average of market prices in all periods is higher in TG1 than in CG, even if differing by a mere 0.1, being respectively 105.8 and 105.7.
40 It actually seems that the effect has the opposite direction to what expected since prices tend to be higher, even if insignificantly higher, in TG1 than in CG.
100 101 102 103 104 105 106 107 108 109 110 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Cumulative Distribution Function - Market Prices
CG TG1 TG2 Price P e rc e n ta g e
The rationale is the same of the previous hypothesis: by tightening the ICC, the expected value of being a cartel member decreases, hence affecting firms' individual willingness to enter collusive agreements and, accordingly, the number of cartels formed. This hypothesis allows to test whether managerial liability enhances the “ex-ante”
deterrence power of Leniency Programs.
An initial and primary analysis of cartel formation, as summarised in Table 1, highlights some interesting aspects: it appears that, as happened for Hypothesis 1, TG1 presents figures and behaviours more collusive than CG, hence contradicting the experimental hypothesis. Both the average number of cartels formed per group and the average number of near-unanimity41 decisions is higher in TG1 than in CG. It should conversely be noted
that the number of groups that formed a cartel at least once is higher in CG than in TG1.
Table 1 – Summary Statistics for Cartel Formation and Willingness to Communicate42
Figure 3 – Number of Active Cartels per period
Nevertheless, such differences do not a have statistical significance: as represented in Figure 4, the Mann-Whitney U test performed on the cumulative distribution of collusive
41 By near-unanimity it is meant those rounds where 2 members out the 3 in each group expressed the willingness to communicate, resulting in cartel formation. Such indicator is helpful given the fact that cartels can be formed by a fraction of the active firms in a market.
42 All the figures in Exhibit 2 are expressed per group.
Control Treatment 1 Treatment 2 Number of Cartels 1.6 1.7 0.6 Round of Last Formation 12 4 8 Number of Near-Unanimity 6.2 7.7 10.8 Number of Rounds of Perfect Collusion 8.8 9.3 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 2 4 6 8
Number of Active Cartels
Control Group Treatment Group 1 Treatment Group 2 Periods N U m b e r o f C a rt e ls
decision43, as well as the Kolmogorov-Smirnov test, cannot reject the null hypothesis that
collusive decisions in both CG and TG1 come from the same distribution. For the Mann-Whitney U test p-value=0.8214, for the Kolmogorov-Smirnov test p-value=0.898. On the other hand, both test fully reject, at any significance level, the hypothesis that the collusive decisions in TG2 are drawn from either of the other two groups distributions: all the p-values are equal or drawing near 0.
Figure 4 – Cumulative distribution functions of Collusion per treatment
Figure 5 - Cumulative Distribution of Joint Collusive Decision
The same results emerge from the analysis of near-unanimity decisions: the cumulative
43 Collusion is defined so after that the all the players decided to enter price discussion and it lasts until either one of the players reports the agreements or the Antitrust Authority launches an investigation. The variable Collusion is a dummy variable allowing also for the Kolmogorov-Smirnov test to be performed.
0 1
0.9 0.93 0.96 0.99
Cumulative Distribution of Collusive Decisions
CG TG1 TG2 Collusion P e rc e n ta g e 0 1 2 3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Cumulative Distribution of Willingness to Communicate
CG TG1 TG2
Joint Collusive Decision
P e rc e n ta g e
distribution functions of the sum of communication decision44 of CG and TG1 are almost
collinear and their difference is statistical insignificant (p-value=0.5), while they both clearly dominate, as evident in Figure 5, the distribution function of TG2 (p-value=0.00 when tested against either CG or TG1).
Result 3: “Cartel lifespan is statistically equal in CG and TG1; cartel lifespan is longer in
CG and TG1 than in TG2.”
Hypothesis 3 tests whether the inclusion of managerial liability into Antitrust Policy constitutes a good tool for hindering the internal coordination of cartel members, hence exacerbating the internal governance problem of cartels as described by Spagnolo (2000 a,b).
In order to understand whether such an effect exerts its influence on the lives of cartels, Kaplan-Meier survival estimates are performed: this analysis allows the prediction of the share of observations for which a given event has occurred or not. In this case the event is represented by cartel breakdown.
Figure 6 – Kaplan-Meier Survival Estimates ranked by Breakdown Cause.
The analysis has been conducted separating the observations both by treatment and by breakdown cause. The results of survival forecasts by treatment result in a
statistically-44 Sum of Communication Decision is the variable used for identifying the joint willingness to communicate: it is the sum of the individual communication dummies of group members. A value of 3 means a cartel has been created, while lower values stand for the number of members who wished, unsuccessfully, to communicate.
0. 00 0. 25 0. 50 0. 75 1. 00 0 5 10 15 20 analysis time failure = 0 failure = 1 failure = 2 failure = 3
insignificant almost collinear overlapping curves45, while, as shown in Figure 6, when
analysed by breakdown cause46 the survival analysis provides more useful insights.
It appears that the cartels broken down by internal report coupled with price deviations are those who most rapidly see the internal coordination melting down and reverting to competitive strategies: this is a powerful insight, since it clearly establishes a positive relationship between price deviations and both cartel duration and recidivism, hence suggesting that positive results for welfare could be achieved by policies that pose clear incentives on coupling price deviation and report.
Result 4: “Post-conviction prices are statistically indifferent in CG and TG1;
post-conviction prices are higher in CG and TG1 than in TG2.”
Now another important vein of the literature on cartels is tested: as argued by Harrington (2008), ex-post desistance is a dynamic part of ex-ante deterrence, since it prevents the reformation of previously broken-down cartels. With respect to this feature the introduction of managerial liability may create a very strong two-fold obstacle: on the one hand it raises the E(AC), hence tightening the ICC; while, on the other hand, it may also
result in a behavioural obstacle to recidivism, since top executives will have to renegotiate collusive agreements with the same companies whose whistleblower has concluded with personal fines levied on their own pockets.
45 Kaplan-Meier analysis is affected by the extremely small number (3) of cartels formed in TG2, hence resulting in an extremely superficial and almost meaningless forecasts; on the other hand, the analysis by break-down cause allows for more meaningful, though statistically insignificant, forecasts.
46 In Figure 6, “failure” labels the breakdown causes: 0 is no breakdown, 1 is Report1, 2 is Report 2 and 3 is AA Investigation.
Figure 7 – Cumulative distribution functions of post-conviction price
A superficial analysis already suggests how post-conviction prices seem to be
unaffected by managerial liability, while reinforcing the findings related to communication protocols: the average post-conviction prices for CG and TG1 are respectively p=107.15 and p=107.21, while for TG2 p=101.55. This finding is also supported by the results of the Mann-Whitney U test on the distribution of post-conviction prices. The null hypothesis of distributions of post-conviction prices drawn from the same unknown distribution between CG and TG1 cannot be rejected at 1% (p-value=0.0148), while it is rejected when testing the relationship between CG and TG2, and between TG1 and TG2, with both
p-values=0.00. Such relationships are portrayed in Figure 7, where the cumulative
distribution functions of post-conviction prices are depicted: both the inexact collinearity between the CDFs of CG and TG, and the domination of the CDF of TG2 by those of CG and TG1 are highlighted.
Result 5: “Post-conviction prices are higher in cartels that collapsed due to Antitrust
Investigation, than in cartels that collapsed due to report.”
As sharply defined by Spagnolo (2004 a,b), cartels are long-term dynamic criminal
relationships that need to form trust among members in order to stick them to the respect
of the collusive agreement. Henceforth, it is interesting to test which type of cartel breakdown has the largest effect in terms price abatement.
101 102 103 104 105 106 107 108 109 110 0 0.2 0.4 0.6 0.8 1
Cumulative Dsitribution of Post-Conviction Prices
CG TG1 TG2 Price P e rc e n ta g e
The assumption that a breakdown resulting from internal turmoil in the cartel leads prices to faster approach the competitive equilibrium stems from the aforementioned need of reciprocal trust: the uncertainty deriving from internal enforcement issues has a stronger disruptive power on cartelists' trust than an outside investigation.47
Figure 8 reports the average post-conviction prices by breakdown cause. Two effects are clearly identifiable from the graph: first, the conjectures of Harrington et al. (2013) on communication protocols find another piece of evidence. The average prices in post-conviction rounds for the reiterated elimination of minimum.acceptance prices protocol are clearly dominated by those of chat protocol.48 Second, from the comparison of
post-conviction prices ranked by breakdown cause it appears that Report 1 and Report 2 have a strong direct effect on prices, reducing them quickly to values close to the competitive equilibrium. On the other hand, breakdown resulting from an antitrust investigation seem to have little effect: prices drop by a small fraction and tend to be stable afterwards. This fact suggests, that, even if Antitrust Authorities must increase the effectiveness and efficiency of their investigations, the ultimate driver of post-conviction prices remains the
trustworthiness shown by fellow cartelists.
Figure 8 – Post-Conviction Prices by Breakdown Cause (6 rounds time-span49)
47 Actually, an outside investigation that sees no cartelist whistle blowing even after its own launch, may simply prove to cartelists the trustworthiness of their fellows.
48 The comparison can be done only on AA breakdowns since they are the only observed in TG2.
49 The time span is 6 rounds since such a time span allows for the collection of post-conviction prices of all-cartels formed in the three treatments. In addition, cartels that underwent a process of quick form-dissolution cycles where not included as long as their final dissolution came.
1 2 3 4 5 6 100 102 104 106 108 110
Post-Conviction Prices by Breakdown cause
Report 1 - (general) Report 2 - (general) AA - (general) AA - (chat protocol) AA – (posts protocol)
Number of Post-Conviction rounds
P
ri
Result 6: “Price deviations in CG are as severe and persistent as in TG1; price
deviations in CG and TG1 are less severe and persistent than in TG2. In addition, players do not fully couple price deviation and report.”
As discussed in the literature review, the usual way of getting out of a collusive
agreements expects a price deviation for undercutting the collusive price and capturing all the market.50 Nevertheless, due to the enforcement of Leniency Programs, deviators must
be aware of the “punishment effect” of fines and, accordingly, a price deviation is expect to come coupled with the report to the Antitrust Authority. The “protection from fines” effect is expected to be enhanced by the introduction of managerial liability.
As Figure 9 summarises, it appears that the frequency of price deviations is definitely affected by the nature of the communication protocol, hence further strengthening the already strong evidence supporting the discussion of Harrington et al. (2013), as
highlighted by the staggering frequency of price deviations in TG2. Another aspects that glimpses at first sight is the double frequency of price deviation in TG1 than in CG, 31.5% and 15.3% respectively, hence completely subverting the predicted result. Despite this figure, the results of the Mann-Whitnry U test conducted on the distribution of market prices during cartel activity, cannot reject (p-value=0.3348) the null hypothesis of samples drawn from the same population. The null hypothesis testing the relationship between CG and TG2, and between TG1 and TG2 are rejected at any significance level (both
p-values=0.00).
Figure 9 - Deviations Frequency and Magnitude (frequency in brackets)
Finally, the “protection from fines” effect described in Spagnolo (2004) is not fully 50 This is the case in a Bertrand competition scenario, where price is the only determinant of demand and profits; in a
Cournot competition scenario deviation takes the form of an increase in the quantity produced in order to reduce the price and compensating the loss due to lower prices with the extra profits gained on the additional units sold.
# Deviations Av. Price Spread 0
2 4 6 8
Deviations Frequency and Magnitude
CG (frequency 31.6%) TG1 (frequency 15.4%) TG2 (frequency 100%)
observed: price deviations and reports to the Antitrust Authority before prices are observed on the market have a frequency of 27.3%, hence exposing the deviator to the use of fines as punishment for its infringement of the collusive agreement. The “fine punishment effect” is conversely observed in 75% of the cases, where players who suffered from a price deviation punish the deviator by reporting the collusive agreement to the investigative body.
6 - Conclusions
There are two main conclusions that can be inferred from the experimental evidence of the introduction of managerial liability in Antitrust Policy: the rejection of the hypothesis of welfare improvement due to the introduction of managerial liability and the support to the results of Harrington et al. (2013) on communication protocols.
The first is the rejection of the general hypothesis of welfare improvement from the introduction of managerial liability in Antitrust Policy. None of the tests comparing the CG with TG1 has rejected the null hypothesis of equal distribution of the data generated in the two different experimental sections. Further, the effect observed, despite being statistically insignificant, points towards welfare worsening spurred from the extension of Antitrust Policy, as prices, whether post-conviction or not, are higher in TG1. This result is striking and has to be dealt with caution: there may be two possible explanations behind it. The first, is simply the noise in the specific data collected in this experiment.
The second is more interesting and relates to the improvement of the experimental protocol. The protocol implemented does not fully replicate the incentives and mechanisms that lie behind the implementation of collusive agreements: the managerial salary has been modelled in a simple way, while also the scheme of managerial fine has been shaped using the most straightforward and plain idea.
Therefore, there are two potential areas where the experimental protocol can be refined. The first concerns the model of managerial salaries. The implementation of relative
performance salaries may improve the quality of observations: modelled this way the variable part of the managerial salary would be sensible to others' participants salaries. An idea would be to have different variable parts of salary depending whether there is an ongoing cartel or not: in case of cartel, the variable part would be negatively affected by the spread in firm-level profits. In case of competition, the variable part would be positively affected by the spread in firm-level profits.
Competitive Salary = Fixed Part + θ (пi) + Γ (пi – пj)
Collusive Salary = Fixed Part + θ (пi) + Γ (пi + пj)
With this scheme, participants have a twofold variable part of their salary that captures two important aspects: θ (пi) replicates the attempt of maximising firm-level profit, while Γ
(пi +/- пj) values the performance relative to the other market participants, also
incorporating the state of the market51.
Also the structure of the managerial fine may be modified in a way that helps improving the experimental protocol. I decided to set it equal to 3.5 points since 3.3 were the
maximum collusive variable profit. The benchmark idea behind was to punish managers by taking away the bonuses obtained during collusion: modelled this way it takes away only the expected bonus paid in the last round of collusion. In order to enhance the
experimental protocol a different structure that considers also past bonuses may represent an improvement. Specifically, extending the timespan of fines tightens the ICC even more: rather than imposing a fixed fine, it could be better to charge a fine equal to a fraction of all the bonuses collected in previous rounds, hence giving the managerial fine also a
retroactive scope. Otherwise, applying a reversed rationale, it could be forbidden to pay bonuses to managers liable of collusion for a certain amount of rounds after conviction. Additionally, simulating the jail conviction prescribed by the US individual Leniency Program, players could be taken out of the market for a given number of rounds, while being substituted by an “ad interim” player. Both these solutions increase the risk of
51 The state of the market depends on the joint communication decision of participants: when a cartel is active it is safe to assume ownership would value firm-level profit homogeneity, in order to foster cartel stability and maximising industry profits; while, when there is no ongoing cartel, it is safe to assume ownership would value firm-level profit heterogeneity. This modelling of managerial profits suffers the risk of being circular: such a concern is legitimate, but it has to be noted that firms implementing collusion in real life have to compensate their managers for their exposure in the collusive agreement and implementation, and they have to do so in a way that do not tempt manager to deviate from the agreement. Hence the very structure of the managerial salary may determine whether collusion is actually implemented or not.
sticking to collusive agreements for managers, with the twofold effect of increasing the E(AC), while decreasing E(ΠC).
The second conclusion is the complete support to the conjectures of Harrington et al. (2013) provided by the results of the experiment: all the parts of the analysis clearly highlight the fact that chat communication protocol allows the development of more
sophisticated and comprehensive collusive agreements than the “reiterated elimination of
minimum acceptance prices” communication protocol. This explains the fact that members
of CG and TG1 had performances more in line with collusive behaviours than those of TG2. Such a result is very important in light of the different communication protocols used by researchers in different papers: it points out that, in order to properly replicate the coordination devices employed by real-world cartelists, chat communication protocol has to be preferred over others, given the freedom it provides to participants in discussing their agreements.52
Finally, it can be safely said that no effect has been noted by the introduction of
managerial liability in Antitrust Policy, while the conjectures of Harrington et al. (2013) on communication protocols have been sustained by the experimental evidence collected: the freer and less structured the communication protocol is, the more sophisticated and
comprehensive collusive agreements are struck.
52 Such “freedom” is easy to observe when reading the transcripts of the chat conversations occurred: they greatly vary in the degree of sophistication of the deals struck, hence leaving the participants free of dealing with the other according to their own spontaneous reaction, without being forced into more schematic negotiations.
7 - Bibliography
Apesteguia, J., Dufwenberg, M., & Selten, R. (2007). “Blowing the whistle.” Economic
Theory, 31(1), 143-166.
Aubert, G., & Kornprobst, P. (2006).”Mathematical problems in image processing: partial differential equations and the calculus of variations”(Vol. 147). Springer Science &
Business Media.
Balakrishnan, R., & Sivaramakrishnan, K. (2002). “A critical overview of the use of full-cost data for planning and pricing.” Journal of Management Accounting Research,14(1), 3-31.
Baliga, S., Besanko, D., Al-Najjar, N., (2005). “The sunk cost bias and managerial
pricing practices." Meeting Papers 851 (2006), Society for Economic Dynamics.
Bigoni, M., Fridolfsson, S. O., Le Coq, C., & Spagnolo, G. (2012). “Fines, leniency, and rewards in antitrust.” The RAND Journal of Economics,43(2), 368-390.
Brenner, S. (2009). “An empirical study of the European corporate leniency program.”
International Journal of Industrial Organization,27(6), 639-645.
Brisset, K., & Thomas, L. (2004). “Leniency program: A new tool in competition policy to deter cartel activity in procurement auctions.” European Journal of Law and
Economics,17(1), 5-19.
Bryant, P. G., & Eckard, E. W. (1991). “Price fixing: the probability of getting caught.”
The Review of Economics and Statistics, 531-536.
Buccirossi, P., & Spagnolo, G. (2006). “Leniency policies and illegal transactions. Journal of Public Economics,” 90(6), 1281-1297.
