Do Sanctions or Moral Costs Prevent the Formation of Cartel Agreements?

2.1 The Experiment

2.2 Game Structure and Matching Procedure

Experimental studies have been carried out to better understand collusive mechanisms and the effect of competition policies.^[5] Previous researchers have mainly studied the effect of communication, price announcements and leniency on the formation and stability of cartels and price levels.^[6]

Most experiments on cartels have been conducted with Bertrand games with the possibility to communicate before the price choice, and have therefore been focused on collusion via the level of price generated after non-bidding agreements.^[7] Recently, Engel and Zhurakhovska (2014) and Haucap et al. (2024) have used a standard symmetric two choices, one-shot, prisoner’s dilemma. Some studies, however, limit the ability to communicate.^[8] Moreover, Hamaguchi et al. (2009) ^[9] do not rely on a Bertrand game. In their study, participants are not given a choice on whether or not they want to participate in a cartel, but they are offered possibilities to denounce it or not. This framework allows them to focus on the determinant of individual decisions to blow the whistle.

Our experiment also relies on a limited market environment. In our setting, participants must decide whether or not to form a cartel. Our work does not aim at documenting the level of collusion (measured via the price increase) but to focus on the effect of sanctions schemes and moral costs on the individual decision to form a cartel. We study variations of sanctions schemes and moral costs that cannot be cleanly introduced in a market game, as the returns from a cartel agreement have to be held constant in the experiment, which is the case in our design. Moreover, to focus on the individual determinants of a cartel decision, in our setting, participants cannot directly communicate before choosing to collude or not.^[10]

Matching procedures in the above-cited articles rely on fixed matching. This design best captures the features of actual markets, in which firms interact with each other repeatedly in the same market.^[11] But in some articles, the fixed matching is altered in the course of the experiment. Chowdhury and Crede (2020) allocate participants who interacted with the same two other participants throughout the experiment, apart from a treatment in which participants at some pre-announced point in time are rematched into new groups. In Bigoni et al. (2015) there is a rematching procedure^[12] (at the end of each period, participants were rematched with the same competitor with a probability of 85 %).

To match and contribute to the preexisting literature, our experiment relies on fixed matching.

3.1 The Influence of Sanctions Schemes on Collusion and Cartels

The literature on economics of crime has highlighted the role of sanctions and detection to deter crime. Following the pioneering work of Gary Becker (1968), the economics of crime framework allows us to analyze a firm’s decision to engage in a cartel: a firm has an incentive to form a cartel when the illegal gain resulting from this practice exceeds its expected cost, which depends on both the sanction (fine) and the probability of detection. Therefore, crime deterrence relies on fines and detection (see e.g. Connor and Lande 2012, for a summary of this literature). Also, the corporate governance approach highlights the existence of a possible divergence of interests between shareholders and managers, which argues for penal sanctions (prison or disqualification). A large body of literature has therefore addressed managerial and policy issues, from the optimal level of fines to the appropriate form of the sanction between fine and imprisonment, which raises the individual versus corporate liability issue (see Buccirossi and Spagnolo 2008; Polinsky and Shavell 2000). Concerning leniency, the theoretical literature suggests that such an approach enhances cartel instability and could deter cartel formation (see Cooter and Garoupa 2014 for an overview, and Spagnolo 2008 for specific applications^[13]). Finally, other individual factors (risk appetite or aversion, psychological aspects or ethical behavior) may influence the individual’s choice of whether or not to engage in cartel practices (see Combe and Monnier 2020; Stucke 2011).

As regards experimental studies on cartel deterrence, Apesteguia et al. (2007) test a fine equal to 10 % of the revenue, with positive probability, and study the effect of leniency and bonus (in which firms are rewarded with a share of the fines paid by other firms). In the presence of leniency, they find that prices are lower but that the rate of cartel formation remains high. On the other hand, Hinloopen and Soetevent (2008) find that leniency reduces the rate of cartel formation.^[14] In Hamaguchi et al. (2009), players who reported the information can receive a partial exemption, full exemption or reward. The reward of whistleblowers has been found to help deter collusive practices. Bigoni et al. (2012, 2015) demonstrate that fines have a significant deterrent effect. Their results also suggest a strong cartel deterrence potential for well-run leniency and reward schemes.^[15] Chowdhury et al. (2018) conduct an experiment involving two combinations of fines and detection probabilities. They find that in the absence of leniency, the probability of detection and fines are substitutable. Last, concerning the influence of private claims and penal sanctions, Bodnar et al. (2023) run an experiment to study if private damage claims against cartels may have negative effects on leniency (as whistleblowers have no or only restricted protection against private third-party damage claims). This may stabilize cartels. In their experimental setting, firms choose whether to join a cartel, may apply for leniency afterwards, and then potentially face private damages. They find that the implementation of private damage claims reduces cartel formation but makes cartels more stable. Fonseca et al. (2022) compare legal regimes in which firms are fined with legal regimes in which managers are prosecuted. Their results suggest that there is less collusion in legal regimes in which managers are prosecuted.

Our experiment therefore contributes to the literature in the following way. First, it focuses on individual sanctions as participants’ earnings are directly impacted by their collusive behavior in all our sanctions schemes. Moreover, it includes a greater variability of fines and detection probabilities, which allows us to cleanly assess the role of these determinants. Besides leniency, the experiment adds two sanctions schemes which have not been evaluated in the laboratory for cartels: forced compliance^[16] and exclusion (in the form of temporary disqualification). Evidence on the effect of exclusion with reintegration –which resembles our implementation – has been collected in other type of cooperative games.^[17]

Hence, our experimental setting aims at testing if alternative and/or stronger sanctions lower the propensity to choose the cartel. In line with the pre-existing experimental literature, we hypothesize that the effects of the two additional sanction schemes – forced compliance and exclusion – are associated with a lower propensity to choose the cartel choice.

3.2 The Role of Moral Costs in Cartel Decisions and Formation

Social responsibility in a markets framework has been studied in the literature. Bartling et al. (2015) find a persistent preference among many consumers and firms for avoiding negative social impact in the market, reflected both in the composition of product types and in a price premium for socially responsible products. Previous research has also shown that market framing may increase the likelihood of unethical behavior.^[18] In an experimental framework, Kirchler et al. (2016) ^[19] study how different interventions influence the degree of moral behavior when subjects make decisions that can generate negative externalities on third parties. Sutter et al. (2020) ^[20] further explain that market framing increases the acceptability of unethical behavior and that it is associated with a dilution of responsibility.

Moreover, behavioral economists have studied the conditions under which participants engage in reducing other participants’ earnings if it comes at the expense of their own earnings. The “money burning” experiment of Zizzo and Oswald (2001) shows that participants are willing to pay to reduce other participants’ earnings (see also the joy-of-destruction game introduced by Abbink and Sadrieh 2009, and its related literature^[21]).

In the case of cartel decisions, Stucke (2011) stresses that the fact that damages inflicted on third parties are not always visible to the managers involved in the cartel may undermine their sense of responsibility and their moral cost. Concerning moral costs, cartel settings have the particularity to place cartelists in a situation in which they have to arbitrage between their own profit and fairness towards third parties who suffer from the cartel.^[22] Last, Engel and Zhurakhovska (2014) study collusion in oligopoly inflicting harm on the opposite market side. In their experiment, they have a Baseline with just two active players, and three treatments with an additional passive outsider who is negatively affected by insiders choosing a cooperative move. In their experiment, harm on outsiders significantly reduces the conditional cooperation of insiders. They explain this result by guilt aversion (which is greater if the active insiders increase their own payoff at the expense of the outsider). Haucap et al. (2024) also experimentally analyze cooperation in a prisoner’s dilemma game, with a passive third party that may be harmed when active players mutually cooperate. Applying a within-subjects setting, they compare cooperation under anonymity and social information. Results show that the presence of a negative externality particularly affects guilt-averse women, who cooperate less often independently of the degree of information they receive.

Previous studies show that the reduction in earnings imposed on a participant may contribute to reducing the propensity to choose the cartel option, but the loyalty-fairness dilemma could mitigate this effect. This leads to the following hypothesis:

3.3 The Role of Individual Traits in Cartel Decisions and Formation

The literature on business ethics examines the role of individuals’ traits as a factor of corporate fraud.^[23] Zahra et al. (2005) developed a framework in which individual variables influence the likelihood of corporate fraud. Moreover, risk aversion is also likely to play a role in the choice of individuals. Tan and Yim (2014) show that being more risk-averse decreases the propensity for tax evasion. Bernhardt and Rastad (2016) find that cartelists with high risk aversion collude less. Other studies focus on gender and find that women have a greater aversion to unethical behavior than men (Robinson et al. 2000), that their reasoning is more contextual than men, which includes ethical matters (Van Staveren 2014), and that bribery is more likely to occur when the principal owner is male (Ramdani and Van Witteloostuijn 2012). Experimental studies find that women can be less corrupt than men (Rivas 2013).

As regards cartels, Abate and Brunelle (2022) find that there is a correlation between the maintenance of informal networks based on masculine values and the permanence of cartel practices, thereby demonstrating that gender imbalance constitutes a risk factor for cartel practices. Haucap and Heldman (2023), who investigate the sociology of cartels, obtain that only two of the 156 individuals involved in these 15 cartels were female, suggesting that gender also plays a role for cartel formation. The experimental study by Hamaguchi et al. (2009) also concludes that women are more likely than men to end a cartel by denouncing them through a leniency program. Haucap et al. (2024) experimentally analyze gender differences in cooperation in a prisoner’s dilemma game, with a passive third party that may be harmed when active players mutually cooperate. Applying a within-subjects setting, we compare cooperation under anonymity and social information, as personal characteristics are commonly known in real-life relations. Results show that the presence of a negative externality particularly affects guilt-averse women. No gender difference is found absent negative externalities.

Our experiments contribute to these literatures by documenting the role of two individual traits, i.e. risk attitude and gender. In line with the above-mentioned literature, we hypothesize that:

4.1 The Effect of Sanctions Schemes, Fines and Detection Probability

Sanction schemes have strong and persistent effects on cartel choice (option B). Table 2 and Figure 1 report the effectiveness of the four sanction schemes implemented in the experiment. The monetary sanction scheme and the leniency sanction scheme are not very effective in deterring cartel choice, as 61.96 % and 65.45 % of choices are Choices B, in the monetary and the leniency scheme respectively. Compliance is more effective as this proportion drops to 51.54 %. Exclusion is the most effective sanction scheme as this proportion drops to 42.30 %. The proportion of cartel formation in each sanction scheme is coherent with these behavioral results: it is lowest in the exclusion sanction scheme (27.82 %), followed by the compliance sanction scheme (39.67 %) and by the monetary sanction scheme and the leniency sanction scheme (47.31 % and 50 %, respectively). These differences are statistically significant (see Appendix Table A1, Panel a.).

Figure 1:

Means of cartel choices and cartel agreements by sanctions scheme.

Note that in the compliance and the exclusion sanction schemes, the decrease in cartel choices and in cartel agreements can in part be explained by a mechanical effect. As for compliance, since cartel choice is removed in the period following a detection, participants have fewer instances in which they can choose to collude. As for exclusion, any choice is removed. In the exclusion sanction scheme, cartel choice was selected 45.12 % of the time when it was available. The resulting cartel agreement rate is 29.68 %. The decrease in cartel choices and in cartel agreements remains statistically different from the monetary sanction scheme (with a p-value = 0.0000 in both cases). In the compliance sanction scheme, the cartel choice was selected 59.12 % of the time when this choice was available. The cartel agreement rate is then 45.57 %. The decrease in cartel choices remains statistically different from the monetary scheme (with p-value = 0.0075). However, in this case the decrease in cartel agreements is not statistically different from the monetary sanction scheme. The effectiveness of the compliance sanction scheme is thus more a matter of ex post monitoring than ex ante deterrence, unlike exclusion which is efficient ex ante. The leniency sanction scheme is ineffective in deterring cartels. However, it is found to be effective in terms of detection; under leniency, the rate of detection is 89.58 %.

Result 1 (sanction schemes). Compliance and exclusion are very effective at deterring cartel choices and agreements. Monetary and leniency sanction schemes are less effective at deterring cartel choices and agreements. The leniency sanction scheme leads to an increase in the rate of detection.

Fines impact behaviors in a consistent fashion as higher fines are associated with lower propensities in both output measures. Table 3 reports the results. The deterrent effect of fines is greater when they are increased from 10 to 15 (15 being the maximum fine level) than when they increase from 5 to 10. These differences in choosing the cartel option and in cartel agreements (which arise when the two Type 1 participants choose to collude) across fines are statistically significant (see Table A1, panel a.). In the monetary sanction scheme, with a low fine (of 5), 74 % of the participants choose the cartel option and the proportion of cartel agreements amounts to 61 %. With a high fine (of 15), these proportions are 44 % and 29 % respectively. These rates are lowest with exclusion and with a high fine (respectively 31 % and 17 %). Overall, by multiplying the amount of the fine by three from 5 to 15, the proportion of cartel agreements is reduced by half. The deterrence effect is highest when the fine exceeds the overprofit of the cartel (which is equal to 10). Cartel overprofit is defined as the cartel profit (20) minus the competitive profit (10). This represents the difference in profits between the cartel and the competitive strategy. These results argue in favor of severe fines. The effect of a fine of 15 is noticeable in the case of the leniency sanction scheme as the reduction in cartel formation (cartel agreements) is comparable to the one in the monetary sanction scheme (Figure 2).

Result 2 (fines). Higher fines are associated with lower propensities to choose the cartel choice and with lower cartel agreements in all sanction schemes, especially when it exceeds cartel overprofit.

Figure 2:

Means of cartel choices and cartel agreements by sanction type and fine.

Probabilities of detection have the expected effect in that higher probabilities of detection are associated with a lower propensity to choose the cartel option and a lower proportion of cartel agreements. Table 4, Figure 3 and Panel b. in Table A1 report the results.

Result 3 (probability of detection). High probabilities of detection are associated with lower cartel choices and lower cartel agreements.

Figure 3:

Proportions of cartel choice and cartel agreements by sanction and probability of detection.

Table 5 reports a logistic regression of the likelihood of choosing the cartel choice by moral costs, which shows the robustness of the effects of sanction schemes, fines and the probabilities of detection across moral cost conditions, controlling for gender and risk-aversion. These results attest that participants are impacted by these sanctions in a systematic fashion.

Another way to investigate the impact of sanctions is to study the combination of fines and probabilities of detection, as this combination allows us to measure the sensitivity of participants to the cartel expected overprofit. We define the cartel expected overprofit as the expectation of getting a higher profit than the competitive one. This cartel “overprofit” is therefore defined as the cartel expected profit, minus the competitive profit (this latter being equal to 10).

Table 6 reports the cartel expected overprofit, computed using the following method:

1. In the Monetary sanction scheme, the cartel expected overprofit is equal to (20 – fine * probability of detection) – 10.

2. In the Compliance sanction scheme, the cartel overprofit is equal to the expected overprofit in the monetary sanction scheme minus an opportunity cost corresponding to the fact that in case of detection, Player 1 will not be given the option to choose the cartel option in the following round. This opportunity cost is equal to the probability of detection times the cartel expected overprofit in the following round.

3. In the Exclusion sanction scheme, the cartel overprofit is equal to the expected overprofit in the monetary sanction scheme minus an opportunity cost corresponding to the fact that in case of detection, player 1 will not be given the option to play. This opportunity cost is equal to the probability of detection*(normal competitive profit (i.e. 10) plus cartel expected overprofit in the following round).

4. In the Leniency sanction scheme, in the absence of denunciation, the cartel expected overprofit is equal to the cartel expected overprofit in the baseline. In the case of denunciation, both players have ex ante the same chance to obtain immunity as it is based on the speed of denunciation. To avoid biasing the computation, the probability that a Type 1 participant will denounce the cartel agreement is set at 50 %.

The rationale behind the use of this overprofit is that if Type 1 participants are sensitive to the cartel agreement expected overprofit, they should take Decision B (collusion) when the cartel agreement expected overprofit is greater than zero. Figure 4 reports the propensity to take the cartel choice by sanction type and by Type 2 participant loss. It shows that this propensity to take the cartel choice increases as a cartel agreement’s overprofit increases. The results show that Player 1 participants are more likely to take the cartel choice when the expected overprofit is high.

Result 4 (expected overprofit). Type 1 participants’ propensity to choose the cartel strategy increases as the cartel expected overprofit increases.

Figure 4:

Cartel choice by cartel expected overprofit.

4.2 Moral costs

In addition to studying sanction schemes, we can analyze the impact of moral cost. Table 7 and Figure 5 document the proportion of cartel choice by sanction scheme and moral cost. Overall, the results on the effectiveness of sanction schemes are robust to the introduction of varying levels of moral cost: exclusion is the most effective sanction scheme, followed by the compliance sanction scheme, regardless of the level of loss imposed on Type 2 participants.

Figure 5:

Cartel choice by sanction scheme and moral costs.

The effect of moral costs is statistically significant but small in magnitude, as any loss, i.e. a loss of 3, 7 or 10, imposed on Type 2 participants as opposed to No loss, causes a reduction in the propensity to choose the cartel option of about 2 %. Indeed, the propensity to choose the collusive strategy in any loss condition (loss of 3, 7 or 10) is statistically lower than in the No loss (loss of 0) condition (p-value = 0.0234), indicating that participants take, at least to some extent, moral cost into account. When considering pairwise comparisons, the propensity to choose the cartel option in the Low loss (loss of 3) is statistically lower than in the No loss (loss of 0) condition (p-value = 0.0179). This propensity is also lower in the Full loss (loss of 10) condition compared to the No loss (loss of 0) condition but only at the 10 % level (p-value = 0.065). This leads to Result 5:

Result 5 (moral cost). Moral cost reduces significantly Type 1 participants’ propensity to choose the cartel option but in a magnitude that is small (it causes a reduction in the propensity to choose B of about 2 %).

Table 8 reports the effect of moral cost controlling for fines, probabilities, individual characteristics, and the repetition of the decision by sanction scheme. It documents that, ceteris paribus, the impact of loss remains significant only in the Baseline condition and when there is a full loss imposed on Type 2 participants. This confirms that the effect of moral cost is, relative to the effect of sanctions, small. Note that our setting analyzes moral cost in a between-subject design so we cannot document whether there is a specific relationship between risk-aversion and moral cost.

Result 6 (moral cost and sanction schemes). Ceteris paribus and by sanction scheme, moral cost impacts in a statistically significant fashion the propensity to choose the cartel option in the baseline and when the loss imposed on Type B participants is high.

In addition to the non-parametric tests, it allows us to document that the effect of moral cost is significant only when in the baseline and when the loss is total for Type 2 participants. This suggests that moral cost impacts decisions when the sanction scheme is simple, and when the loss imposed on Type 2 participants is high. Outside these conditions, sanction schemes are the only tools that reduce the propensity to choose the cartel option and that prevent the formation of cartel agreements.

The relationship between the propensity to choose the cartel option and cartel expected overprofit – as reported in Figure 4 – is robust to the varying levels of moral cost, as shown in Figure 6. Figure 6 further shows that dispersion around the cartel agreement expected overprofit is mainly driven by compliance and exclusion, which means that there is higher heterogeneity in behavior when sanctions are strict.

Figure 6:

Cartel choice by cartel expected overprofit, sanction scheme, and moral cost.

This leads to Result 7:

Result 7 (influence of sanctions scheme on the cartel overprofit/loss trade-off). Strict sanction schemes, i.e. compliance and exclusion, lead to more heterogeneous behavior than looser sanctions, i.e. baseline and leniency. The impact of the cartel expected overprofit is robust to the introduction of different moral costs.

4.3 Individual Characteristics

Risk attitudes can affect the decision to form a cartel as it is associated with a lower probability of breaking the law. In terms of risk preference, Figure 7 reports that higher risk aversion is associated with a lower propensity to choose the cartel decision. A logistic regression of the impact of the number of safe choices in the Holt and Laury (2002) task leads to a small (−0.009, with a constant of 0.55 and standard error of 0.001) but statistically significant (p-value = 0.000) decrease in the likelihood to choose collusion. This suggests that risk aversion plays a role in the choice of engaging in such cartel agreement, but that the magnitude of this effect is limited.

Figure 7:

Mean of cartel decisions over the number of safe choices in a Holt and Laury (2002).

Result 8 (risk aversion). Risk aversion reduces the propensity to choose the cartel agreement.

The studying of the impact of gender shows that, when using all data, the propensity to choose B is higher among men than women (p-value = 0.000). Men choose the cartel decision 56.72 % of the time, while women choose it 53.76 % of the time. When comparing the individual propensities of men and women participants, the statistical significance disappears in all conditions except the baseline. The results are reported in Table 9.

Result 9 (gender). Men are found to have a higher propensity to choose the cartel option.