Cartels Between Life and Death. What Determines Cartel Survival?

CONTENTS

1 introduction

2 determinants of cartel survival
2.1 Fundamentals of Cartel Stability
2.2 Geographical Scope
2.3 Number of Firms
2.4 Ringleader
2.5 Type of Infringement
2.6 Internal Enforcement Methods
2.7 Entry and Exit Dynamics
2.8 Cartel Enforcement
2.9 Macroeconomic Determinants

3 data
3.1 Data Set Description
3.2 Variables

4 methodology and estimation strategy
4.1 Principles of Survival Analysis
4.2 Duration Modeling and Competing Risks
4.3 Time-varying Covariates
4.4 Models

5 results and discussion
5.1 Estimation Results
5.2 Robustness Checks and Biases

6 conclusion

A Appendix - Tables and Figures

LIST OF FIGURES

Figure 1 Time-varying firm number in the power cable cartel

Figure 2 Cartel duration over time - cause-specific (12 months)

Figure 3 Kaplan-Meier survival estimates (12 months)

Figure A1 Cartel duration over time - cause-specific (6 months / standard)

Figure A2 Demand growth and six-month volatility

Figure A3 Bank of England Official Bank Rate (IUMBEDR)

Figure A4 Kaplan-Meier survival estimates (Categorical covariates)

Figure A5 Kaplan-Meier survival estimates (6 months / standard)

Figure A6 Estimated baseline cumulative hazard for natural vs agency death

LIST OF TABLES

Table 1 Descriptive statistic - average over cartels

Table 2 Cartel duration range in years

Table 3 Descriptive statistic - time-varying variables

Table 4 Censored Cox PH model with competing risk (12/6 months)

Table A1 Hypotheses

Table A2 Variables in dataset

Table A3 Cartel sample classification according NACE Rev

Table A4 Stratified Cox PH model with competing risk

Table A5 Censored Exponential PH model with competing risk (12/6 months)

Table A6 [Daily] Censored Cox PH model with competing risk (all types)

LIST OF ABBREVIATIONS

Abbildung in dieser Leseprobe nicht enthalten

1

INTRODUCTION

Over €28 billion fines imposed by the European competition authority (ECA) in 135 decisions between 1990 and 2018,¹ of which the majority took place after the turn of the millennial, underline that the fight against cartels is far from over. The anticompetitive practices of car- tels can lead to increased consumer prices and remain an important point on the agenda of antitrust authorities. As a reaction, the competition policy of the ECA has undergone sev- eral changes in the last decades, most prominently observable by the introduction of the EU corporate leniency program in 1996 and its modification ten years later. Almost all of the current cartels are detected by leniency applications of one of their members because whistle blowing policies allow cartel members to get reductions in fines when they self-report the committed infringements. Before the implementation of these policies, the most crucial threat for internal cartel stability was strategic deviations from the cartel agreement (e.g. price cuts) by one of the cartel members to increase short-term profits. The risk of being undercut on the one hand and the risk of leniency applications on the other hand place today’s cartel firms in a ’between-life-and-death’ situation in which persistence of the cartel is equally likely as its breakdown.

Early theoretical literature related to cartel stability (e.g. Friedman (1971)) investigates the threats of deviations, whereas the introduction of leniency policies paves the way for a variety of new (game-)theoretical research that examines the effects of whistle blowing policies on cartel stability (e.g. Harrington & Chang (2009)). However, cartels are not fully defenseless to both threats and develop several techniques like monitoring or compensation mechanisms to secure their agreements. Some cartel organizations use a well-founded set of techniques and share properties so that they are either never detected or survive significantly longer as other cartels which refrain from these methods. It is at the heart of competition authorities interests to know these organizational and operational properties, that support the duration of cartels, to detect hardcore cartels and evaluates the current level of antitrust law. Beside the theoretical research, a strand of empirical literature examines, by the methods of survival analysis, the effect of various cartel-specific factors on the duration of cartels. Modern survival analysis, also denoted as time-to-event analysis, originates from medical and biostatistic and tracks, for example, the life of a cancer patient and evaluate the effectiveness of a certain medical treatment on the survival time of this patient. Similarly, cartel survival studies like De (2010) or Levenstein & Suslow (2011) examine cartel’s survival by treating the durations of discovered cartels as life cycles. These ex-post, cross-sectional studies assume that the characteristics of a cartel affect its survival and can be metaphorically compared to the influence of patient- specific factors (e.g. genetics, smoking) on the patient’s life expectancy. Cartel duration studies mainly consider internal factors like the number of cartel firms but also include, as proposed by theoretical research (Green & Porter (1984), Rotemberg & Saloner (1986)), external factors like demand growths.

My paper contributes to literature of cartel survival analysis by investigating cartel duration based on an up-to-date record of 123 cartels discovered and punished by the ECA.² The major- ity of data, especially related to internal characteristics, sources from the non-confidential ver- sions of the ECA’s prohibition decisions whereas macroeconomic data stems from the OECD and the Bank of England, respectively. The main goal of this thesis is firstly to replicate the analyses by Hellwig & Hüschelrath (2017, 2018) on cartel level and secondly to extend their analyses by incorporating new covariates related to either internal enforcement methods and demand volatility. The paper at hand is methodologically in line with previous litera- ture since it uses several Cox proportional hazard specifications and evaluates the findings to different natural death definitions. Furthermore, I also provide nonparametric Kaplan- Meier survival estimates of all categorical variables and estimate time-varying covariates by a piecewise-constant exponential PH model.

My results confirm most of the outcomes by Hellwig & Hüschelrath (2017, 2018) on a broader data set. I found slight evidence that more democratic organized cartels have lower probabil- ities of breakdowns compared to cartels which are led by a ringleader. Internal enforcement techniques like monitoring mechanisms support cartel members to enforce their agreements while trade associations simplify communication in young cartels but increase the hazard in adult cartels. The strongest driver of cartel instability is, apart from gradual cartel exits, the availability of the EU corporate leniency program. This finding underlines the effectiveness of leniency and whistle blowing policies. According to the macroeconomic determinants, my analysis points out mixed results for changes in the interest rate but a decrease in demand growth and an increasing demand volatility are associated with a hazard-decreasing effect on cartel duration in all models. Hence, a policy recommendation could be that cartels are formed and endure in markets with worsening and volatile economic conditions.

The structure of my paper is as follows. Chapter 2 presents the determinants of cartel stability by reverting to the theoretical literature on cartel stability and case study evidence. More- over, each section in this chapter outlines a testable hypothesis. Chapter 3 describes the data sets and explains the construction of variables. Methodology and estimation strategy can be found in Chapter 4. Chapter 5 evaluates the hypothesis according to the regression results, discusses the robustness of the findings and points out sources for biases. Finally, I conclude in Chapter 6.

DETERMINANTS OF CARTEL SURVIVAL

This chapter reviews factors which are known for affecting the stability and duration of cartels based on contributions of the theoretical literature and detected real-world cartel cases. All case studies used in this chapter are also present in my cartel data sets. The use of case studies as a tool for analyzing cartel behavior is very common, because they offer valuable insights of the organization and behavior of cartels. Despite the fact that each cartel is unique in its characteristics, it is reasonable to assume that most cartels face similar challenges and risks.

Early case studies by Posner (1970) or Hay & Kelley (1974) evaluate the impact of antitrust policies and examine product and market characteristics which are more susceptible to price fixing agreements. Their sources are several US court records provided by antitrust divisions like the Federal Trade Commission (FTC). More recent papers using the case study approach are for example Grout & Sonderegger (2005) who investigate econometrically the formation of European cartels or Harrington (2006) who considers 20 European cartel case decisions in the period between 2000 and 2004. My case study is most closely related to his paper since he focuses on revealing operational and organizational patterns of detected cartels. While all mentioned papers are selling their case study approach as the main contribution, my case study section serves more as a preliminary part of the paper and is therefore very brief com- pared to those papers. The main purpose of this chapter is not only to review important determinants but rather to draft testable hypotheses by combining results of theoretical re- search and case study evidence.

However, it is important to note that this chapter in particular and my analysis in general make no claim to include all relevant aspects which may or may not have an effect on car- tel survival. For instance, the well-founded strand of theoretical literature regarding product heterogeneity and, more importantly, asymmetries between cartel member firms (e.g market shares, capacities) are not considered. There are mainly two problems which make such an analysis a challenging task. First, a lack of information like sales data, capacity utilization or major customer allocation makes it difficult to get a clear picture of the underlying power relations within a cartel. Second, the depth of information in the Directorate-General for Competition (DG COMP) prohibition decisions is quite heterogeneous.³ Some prohibition decisions, especially those from cartels which are detected because of a leniency application, provide a rich set of information, whereas others suffer from detailed firm-specific informa- tion. Furthermore, I did not find applicable case studies for all of the following motivated determinants.

2.1 Fundamentals of Cartel Stability

Before I begin to outline the determinants, which are tested later in this paper, I will address some introducing words to the terminology used in this branch of research and highlight a few seminal theoretical contributions to the literature of cartel stability.

The paper at hand is related to cartel survival. A cartel survives until a certain point in time if it remained stable in all previous periods.⁴ Hence, I denote the persistence over a series of consecutive periods by cartel survival, whereas a cartel is said to be stable if it survives in a certain period. Formally, the cartel Γ survives until the beginning of period t if it has not experienced a breakdown in all t ✁ 1 previous periods. Therefore, the determinants of periodical cartel stability are directly linked to determinants of overall cartel survival.⁵

Instabilities can arise from several different threats which can be grouped loosely into internal and external threats.⁶ First and foremost, a cartel can break down internally because of its or- ganizational and operational characteristics. Organizational characteristics refer to the overall structure of the cartel. Examples are the number of active firms in the cartel, the hierarchy and distribution of power or the number of cartel entries and exits. Operational characteristics in- corporate the acting of the cartel within its market but also any form of communication among cartel members. Examples are the type of infringement used by the cartel, the geographical scope of the cartel activity, enforcement related mechanisms like monitoring, retaliation and compensation.

Until the development of leniency and whistle blowing policies by cartel authorities, schol- ars view the deviation bye one of the cartel members from the collusive agreement as the most crucial threat for cartel stability. Thus, early literature related to cartel stability treats organizational and operational aspects, comprehensively.

One of the earliest contribution to the modern literature of collusive agreements was made by Stigler (1964). In his paper ’A Theory of Oligopoly’ he postulates that prices chosen by members of a collusive agreement underly several factors like the elasticity of demand, entries of other firms and customers in the market, the industry concentration and the possibility to monitor and enforce ’secret price-cuts’ from members of the collusive arrangement.

If (self-) enforcement mechanisms are weak, the collusive price exceeds the competitive price only slightly because of the permanent fear of being undercut (Stigler 1964, p. 46). Collusion members are assumed to pool their sales information and use these information to detect price cuts in his buyers-sellers model. The results show that incentives to deviate from the collusive price increase with the number of rival firms, the probability of repeated purchases external and exogenous threat, but the breakdown occurs internally because of lacking communication among the cartel members and/or a poor mutual cartel strategy made by the buyers and more importantly by a rising number of entries from buyers into the market (Stigler 1964, p. 51). The latter point is equivalent to an increase in demand. To sum up, Stigler (1964) regards cartels as organizational entities, that suffers from deviation and breakdowns in some periods of time, but their members are able to detect these deviations by pooling sales information and reverting to collusion afterwards.

The first, ’strategic’ considerations of firms in collusive agreements made by Stigler (1964) experienced an upswing with the development of game theory. The seminal work of this branch of literature is Friedman (1971) who proved that a collusive agreement can always be established in an infinitely repeated non-cooperative game, which is also denoted as supergame, when firms’ discount factors are sufficiently close to unity. Friedman (1971) applied this reasoning to a Cournot oligopoly model with firms earning the collusive profit in all periods of the game as long as no deviation occurred and returning to the non-cooperative Cournot solution when a firm defects from the agreement by producing a larger quantity. In particular, let G be an infinite repetition of the stage game g and denote periods by t ✏ {0, 1, 2, ...✉. A cartel Γ consists of n identical firms indexed by i. All firms are involved in quantity competition, produce a homogeneous good, share current profits equally inside the cartel and discount future profits at the same discount factor δ.⁷ Cartel profits and total non-cooperation profits (i.e. Cournot profits) are denoted as πΓ and πc, respectively, while deviation profits are denoted by πd.⁸ Firms can decide in each period t to stay in the cartel or to deviate to a larger one-time profit, but being exposed to punishment in all following periods. A firm i will choose the former strategy if,

Abbildung in dieser Leseprobe nicht enthalten

A cartel agreement can be stabilized in terms of profits if (1) is fulfilled for all n members of the agreement.

Once the game theory has created the methodological foundation for modern collusion theory, researchers started to extend the implicit collusion models by adding further elements. In the 1980s a strand of literature arose which incorporated business cycles, demand growth and demand volatility into this class of models. In their path-breaking paper Green & Porter (1984) introduce uncertainty regarding the cause of observed price drops in the model framework of Friedman (1971). Firms are assumed to sell a homogeneous product by choosing quantities and behave as a monopolist if prices are high and return to Cournot competition if prices are low (Green & Porter 1984, p. 89). A price drop can occur for two reasons, first, a cartel member secretly undercuts the cartel price by overproduction. Second, prices can fall because of a random demand shock. The crucial assumption is that firms are uncertain whether the experienced drop in prices is caused by the first or the latter event. If the observed price falls below the cartel’s threshold price p¯, the cartel moves back to Cournot competition with larger quantities, lower prices and profits for all members. From a consumer perspective, this pattern looks like a price war, however the cartel is still active. This punishment phase lasts only for T periods since the collusion members are aware of their uncertainty and the potential randomness of the price drop. This finding suggests that cartels are able to survive phases of low demand and contrasts the infinite punishment phase of classic grim trigger strategies as stated in Friedman (1971).

While Green & Porter (1984) identified periods of low demand as driving force of price wars, Rotemberg & Saloner (1986) investigate the effects of high demand phases on the stability of collusion by using the model framework of Friedman (1971), too. They apply demand shocks to an implicit collusion model in which firms are competing in prices. Demand shocks are assumed to be independently and identical distributed (i.i.d.), suggesting that there is no correlation in demand between two consecutive periods. They assume that, due to higher prices in high demand phases, price cuts by cartel members are more profitable compared to price cuts in low demand phases (Rotemberg & Saloner 1986, p. 390). More precisely, when firms compete in prices, incentives to undercut increase with prices because also the one-time deviation profit increases in prices, whereas losses imposed by the future punishment in a low demand state (i.e. the return to Bertrand competition) are independent from the previous price level. Rotemberg & Saloner (1986) point out that full collusion, that is collusion in both states of demand, can be only sustained when firms are very patient. Thus, the stability of a cartel is affected negatively in periods of high demand and average levels of discount factors. A cartel can overcome this threat by setting a price below the monopoly price to diminish the incentives to deviate (Rotemberg & Saloner 1986, p. 393).

Further literature on cartel stability discusses internal characteristics like asymmetries in ca- pacities, products differentiation and external characteristics like the interaction of cartels with firms outside the cartel (fringe).⁹ Since these determinants do not play a role in the follow- ing empirical analysis, an extensive discussion is omitted. The most serious, external risk for cartels, namely leniency programs, are treated separately in section 2.8.

2.2 Geographical Scope

The first determinant which possibly affects the duration of cartels is the geographical scope of the cartel. For example the amino acid cartel was a worldwide operating cartel whose mem- bers were Eurolysine SA (Europe), Kyowa Hakko Europe GmbH (Europe), Sewon Europe GmbH (Europe), Cheil Jedang Corporation (Asia) and Archer Daniels Midland Ingredients Ltd. (US) (Case COMP/36.545). In contrast, some cartels like the Luxembourg brewery cartel acted only on a small, geographically limited scope (Case COMP/37.800). This nationwide cartel consists of four Luxembourgian breweries. Hellwig & Hüschelrath (2017, p. 9) point out that worldwide cartels are more vulnerable for instabilities because of difficulties in coordinat- ing their worldwide business strategies and heterogeneities in their cultural dimension which may result in different business policies or incompatible temporal corporate targets. I found two further explanations why worldwide cartels may have a shorter life than regional cartels. First, worldwide cartels are, due to their multinational operating, more exposed to exogenous policy shocks as regional cartels. If, for instance, the US government decides to protect the domestic chemical industry by raising the tariffs, the price-fixing and market sharing agree- ments of the amino acid cartel will be affected negatively. In this scenario it is possible that the US firm Archer Daniels Midland (ADM) can not maintain its fixed market share because of a decline in demand. ADM could overcome this situation by lowering the price but this will undermine the cartel’s price-fixing agreement. Therefore, the cartel members have to discuss the tariff policy change and balance its impact on the infringement. Second, the du- ration of multinational cartels can be diminished because they are under the radar of more than one competition authority. The Luxembourg brewery cartel is under observation of the Autorité de la Concurrence and the ECA, whereas the amino acid cartel can be detected by several country-specific competition authorities and supranational authorities like the ECA.¹⁰ Furthermore, the members of the amino acid cartel are able to send their applications for leniency to three different competition authorities (ECA, FTC (US), FTC (Korea)).¹¹ Leniency applicants of multinational cartels have possibly a broader set of leniency programs to choose from which may decrease the stability of the cartel.

H1: Worldwide cartels have a shorter life expectancy than regional cartels.

2.3 Number of Firms

The number of firms within a cartel is considered as a crucial determinant for the duration of cartels. Almost all survival analyses include this variable because of its ease of measure- ment and its ascribed effect on cartel stability. A cartel with more members covers a larger market share and is faced with less competitors. Therefore, larger cartels might be more able to use their market power to establish their infringements in the industries and are more ro- bust against aggressive market entries compared to small cartels which can be attacked more easily by a competitive fringe.¹² Nevertheless, theoretical literature considers an increasing number of firms in a collusive agreement as destabilizing factor. The coordination and moni- toring of the agreement simplifies with a lower number of cartel members and the risk for an internal conflict is diminished (Motta 2004). Furthermore, a rather trivial result is that when the number of firms in the cartel rises also the number of firms rises which eventually will apply for leniency or undercut the agreement in a future period. But not only the number of possible deviants increases in absolute terms also the actual incentives to deviate increase because deviation profits grow with the number of cartel members. Hence, the robustness against aggressive entrants and a competitive fringe, reached by a larger cartel, is borne by an increased probability of deviations.

The theoretical prediction that a rising number of firms makes collusion more difficult is confirmed by experimental literature, too. For instance, Huck, Normann & Oechssler (2004) investigate firm number effects in an experimentally-tested Cournot oligopoly game. They found that, as the number of players increase, the outcomes get more competitive. More precisely, when four or five firms are active in the market all outcomes could be either assigned to be Nash or to be ’rivalistic’ (Huck et al. 2004, pp. 442-3). A collusive outcome could only be observed in the duopoly market. Consequently, we would expect that cartels with more members have shorter life expectancies.

H2: The larger the number of firms in the cartel, the higher is the probability of a cartel breakdown.

2.4 Ringleader

Beside the number of active firms in the cartel also the distribution of power among these firms can be a source for instabilities. Non-democratic cartels, that are cartels with a ringleader, may serve as proxies for asymmetries (market shares, revenues) within a cartel. The following ex- ample shows that cartels with an unequal distribution of power are probably more vulnerable to internal conflicts than less dictatorial cartels. The citric acid cartel (1991 - 1995) fixed prices and quantities of citric acid products and consisted of five firms. ADM and F. Hoffmann-La Roche AG were the ringleader and decided about the fixed quotas, whereas Jungbunzlauer AG and Cerestar Bioproducts B.V played merely a passive role and had only minor participa- tions in cartel’s decisions (Case COMP/36.604, pp. 57-8). Especially, Jungbunzlauer AG tried, due to fear to be pushed out of the market, to attack the ringleaders by deviating from the agreements and priced below the target prices. Lastly, the citric acid cartel could not solve its internal conflicts and broke down in 1995. This example shows that an asymmetric dis- tribution of power may result in a lower duration compared to more democratic organized cartels.

H3: Cartels with ringleaders are more likely to experience a breakdown.

The cartels considered in this paper are involved in different types of infringements. I dis- tinguish among combinations of five types which are price fixing, market sharing, quantity fixing, information exchange and bid-rigging. At a first glance, a cartel with a more complex infringement scheme, that is a combination of multiple types, is more robust and has a longer, expected duration compared to a cartel which rely only on a single infringement. This hypoth- esis follows from the idea that complex cartels are more structured organizations with more efficient monitoring techniques and less incentives to deviate from the agreement. But not only the complexity of the infringement drives the stability of an agreement, even the actual type may influence the hazard for a breakdown. According to Hellwig & Hüschelrath (2017, p. 9), cartels which use solely price-fixing agreements are more likely to be unstable because deviations are easy to conduct and more difficult to detect. This approach is also confirmed by Stigler (1964, p. 46) who underlines the strengthening effect on collusion of market sharing agreements.

H4: Cartels which use solely price-fixing agreements are expected to have a shorter duration.

2.6 Internal Enforcement Methods

The previous three sections emphasize that the stability of a cartel can be weakened by secret deviations from the agreement. Unlike legal business contracts, cartel agreements can not be enforced in courts due to their illegal nature. Therefore, many cartels develop techniques like monitoring and compensation mechanisms to control the effectiveness of the agreement and balance unexpected variations in sales and market shares. Trade associations are mainly used for two reasons. First, they serve as a mutual forum to monitor agreements and second, cartel members are able to cover their meetings under the veil of trade associations. All three methods have in common that they are expected to have a deterring effect on deviations.

Monitoring

Most cartel monitoring mechanisms concern either the reporting of prices or sales. In general, the monitoring of other firms’ prices is often more feasible as the monitoring of other firms’ sales, because the opponents’ charged prices are either directly observable in public price lists or indirectly observable by customer consultations. A close price monitoring mechanism is, especially for cartels involved solely in price fixing, of particular interest to adjust prices to a mutual level and to detect price cuts as soon as possible. While the costs for price monitoring are rather small, the monitoring of other firms’ sales figures is sometimes so complicated that many cartels trust in truthful, self-reporting statistics. A cartel with an extensive monitoring scheme is the copper plumbing cartel (Case COMP/38.069). This cartel was formed in 1988, lasted for 12 years and consisted of eight main copper plumbing suppliers which fixed prices and quantities of the EEA market. The market price of copper plumbings depends strongly on the copper price, a conversion margin and is negotiated with each customer on an individual basis (Case COMP/38.069, pp. 14-5). The starting point for the negotiation was a price list which was fixed by the cartel member. Variations in the cooper prices made a frequent adjustment of prices necessary and certainly increased the stability of the infringement. Apart from the exchange of price information, the cartel monitored sales figures through import statistics and reported them to a central controlling entity within the cartel. Overall, the close and extensive monitoring in the copper plumbing cartel is probably a reason for its above- average duration.

H5a: Cartels which monitor prices or sales have a higher life expectancy

Compensation Mechanism

If all cartel members truthfully report their sales figures and one or more firms monitor an unexpected drop in their sales, this drop occurs either because of exogenous variations in demand or due to a violation of the cartel agreement by one of its members. In the first case a cartel with an active compensation mechanism can balance these variations for example through a ’buy-back’ scheme, whereas in the latter case the deviant has to be detected and punished by a retaliation mechanism.¹³ An essential prerequisite for both methods is that cartel members monitor their sales and report them to each other.

Harrington & Skrzypacz (2011) examine the influence of truthful or untruthful reporting of firms’ private sales data and interfirm transfer payments on the stability of collusive agree- ments. They use an infinitely repeated multistage game with imperfect monitoring in which each firm (1) chooses a price, (2) observes its sales privately, (3) decides about the value of reported sales and reports it to all other firms and (4) makes a non-private transfer payment, according to the reported sales which is split equally among the other firms (Harrington & Skrzypacz 2011, p. 6). Similar to Friedman (1971), it is further assumed that firms return to the competitive equilibrium when one of the firms do not report its sales or makes an incorrect transfer payment (Harrington & Skrzypacz 2011, p. 6). The results show that collusive agree- ments can be facilitated through a truthful reporting of sales figures and transfer payments, even when firms can not perfectly monitor the origin of the variation in sales. The temptation of underreporting can be reduced since a report of too low sales is followed by a return to the static Nash equilibrium (Harrington & Skrzypacz 2011, p. 18).

Several cartels in the data set at hand combine monitoring mechanisms with compensation schemes. All of these cartels have in common that they are producer of chemicals and involved in complex infringements of either price fixing and market sharing or price fixing and quantity fixing. An exemplary cartel for the first type is the choline chlorid cartel which increased prices to generate higher margins of a low-margin product and allocated market shares at a global and European level (Case COMP/37.533, p. 21). The members of the cartel met every six month to monitor the effectiveness of the agreement and agreed to compensate if a market share exceeded the predetermined market share allocation. The citric acid cartel, an example for the latter type, tried to stabilize its agreement by using a so-called ’buy-back’ method that obliged overproducing members to buy products from members which experienced a loss in sales due to the overproduction in the next period (Case COMP/36.604, p. 30).

[...]

¹ http://ec.europa.eu/competition/cartels/statistics/statistics.pdf [accessed: 20.09.2018].

² I consider prohibition decisions until June 2018.

³ The number of pages in ECA’s prohibition decisions varies from 10 to 377 with mean 75 and st.d. 64.

⁴ The term stable should be distinguished to the terms internally stable and externally stable which are often used in papers concerning incentive constraints of cartel formation

⁵ This assumption is a simplification since it is possible that cartels are stable only in certain phases, break down and are continued in later phases. Cartel duration studies count typically a recurrence of an earlier cartel as two separate cartels.

⁶ Both groups cannot be separated clearly. Interdependencies are present. Demand fluctuation are for example an

⁷ The assumption of quantity competition can be changed easily to price competition assuming that the cartel Γ charges the monopoly price in a collusion state and revert to Bertrand competition in a punishment state.

⁸ The notation is adjusted to the cartel context and differs slightly from the original paper by Friedman (1971) and newer surveys like e.g. Feuerstein (2005).

⁹ See for collusion of capacity-constrained firms e.g. Compte, Jenny & Rey (2002), for effects of product differentia- tion on cartel stability e.g. Ross (1992) and for competition between cartel and fringe e.g. Shaffer (1995).

¹⁰ Autorité de la Concurrence is the French CA which covers also the Luxembourgian market.

¹¹ In this particular case ADM cooperated with the US authority.

¹² The fringe comprises of all firms in the cartel market which do not belong to the cartel. 2.5 Type of Infringement

¹³ Since only a few cartels in my data set use retaliation techniques, I focus on compensation mechanisms.