Cartel formation and the business cycle

ABSTRACT Several theoretical propositions suggest that changes in economic activity can explain the decision to form a cartel. The majority of the propositions claim that cartel formation is more likely in expansionary and less likely in contractionary phases of the business cycle. The propositions are re-examined theoretically and by using data on detected cartels for the European Union. In both cases, the results cannot confirm that more cartels are formed in any of the business cycle phases and that correlated economic growth rates create higher incentives to collude. Furthermore, it cannot be confirmed that more cartels form shortly after respectively the trough or the peak in the business cycle. The conclusion is that cartel formation is unaffected by changes in the business cycle.


Introduction
One of the major problems in cartel enforcement policy is developing reliable ways to detect them.What works in one case may not apply to the next especially when cartels adopt new ways to keep their activities hidden from competition authorities.Despite the obvious challenges cartel enforcement relies in practice on direct knowledge about the functioning of individual markets at the national level, on the Internal Market and in the European Economic Area.Even with this knowledge, it is not easy to detect a cartel by only relying on data on how markets function or are supposed to function.Instead, the detection of cartels relies in practice heavily on information provided by customers, suppliers and competitors outside of the cartel and even more so, from cartel members who would like to quit but run the risk of punishment if they tell all.Altering the punishment for cheating is something the European Commission has been working on.When competition policy became a common policy for the EU member states cartel members who wanted to quit also faced sanctions from the competition authorities if they told all.This changed with the introduction of a leniency programme in 1996, later revised in 2002 and again in 2006.The final revision now gives immunity from fines when a cartel member informs all about the cartel to the competition authorities. 1This has created a situation under which any potential cartel cheater only faces potential punishments from the remaining cartel members, which after detection is unlikely to be used.
Limited tools are available to caseworkers for identifying real restraints on competition.This is especially a challenge when such restraints are applied by an entire industry at the same time, which is what cartels tend to do.Any parallel adjustments in prices, quantities and strategies look similar to other industry-wide adjustments caused by for example changes in demand or the cost of resources.And better tools would be preferable, especially those that use easily observable changes in economic conditions.This is where several theoretical articles may suggest could be the case.The articles suggest that cartel formation may be linked to specific periods in the business cycle.The main problems are that the articles propose different links to the business cycle and that the propositions have not been empirically tested.Identifying which of the propositions is correct may help competition authorities greatly improve their cartel detection skills.
The first article on the matter is made by Rotemberg and Saloner in 1986. 2 In the article, they show that collusion prices are predominantly counter-cyclical irrespective of whether prices or quantities are used as the strategic parameters of the firms.The articles that followed this article, all focus on prices as the strategic variable, and all find that collusion prices are more likely to be pro-cyclical or at least weakly so. 3 The different results can be explained by the different assumptions used by the authors, and especially their choice of model to study the phenomenon in.When reading the articles it especially becomes clear that the  The article by Bagwell and Staiger points out the limitation of Rotemberg and Salinger's article when it comes to their view on demand fluctuations. 5They explicitly explain that the changes in the business cycle do not imply fluctuations in demand.Instead, the business cycle is a way to describe the level of economic activity over time and although economic activity always fluctuates from a state of low to high activity, it does not mean that demand necessarily fluctuates from low to high.Over a normal business cycle, demand will always be increasing with the economy's growth rate, and given that the economy is always growing, demand will be growing as well.Albeit sometimes faster and sometimes slower.
Harrington and Haltiwanger talk about cyclical demand movements and explicitly highlight the relevance of their findings for macroeconomics, although they essentially look at the same kind of demand fluctuations as Rotemberg and Salinger. 6he "newest" article on the topic is from Fabra.7She places her theoretical contribution in the middle of macroeconomics by talking about cyclical demand as a natural part of the business cycle, but it is unclear why that is so and it is not the aim of her article to explain why.
In other words, we have ambiguity in the theoretical results, although it may just be a matter of understanding the assumptions correctly.This calls for a study of the relationship that can shed more light on the issue and this is the purpose of this article by taking two approaches.The first approach is to re-examine the original theoretical framework, but this time with demand endogenized instead of exogenously given.The second approach is to use existing data on cartel formation to investigate to what extent the data can clarify if cartels are more likely to form during certain periods of the business cycle.The data have been used to study similar problems but in contrast to earlier studies, I do not assume that cartel formation is a count variable or fits a given distribution such as a Poisson distribution or can be predicted by a hidden Markov-series. 8sing this approach I avoid, in the author's opinion, the highly problematic assumption that is used by some authors, that cartel formation has randomness or is some other probabilistic event.Cartel formation cannot be random it is carefully planned and executed just like any other white-collar crime.Cartel formation is a carefully planned executed choice by the involved managers of the involved firms.
Data on cartel activity is not readily available for obvious reasons, but it does exist for sanctioned cartels.Sanctioned cartels are cartels that have been prosecuted, litigated and convicted by the European Commission (EC) (or rather by the Court of Justice of the European Union), the EU's main competition authority.These cases involve some of the biggest firms in the world operating on the EU's internal market.In most cases, these firms have a global reach and not only affect trade in the internal market but in many other parts of the world.Together with data on economic activity in the EU, the cartel data forms the core of the analysis for the empirical part of this article.
Naturally, economic theories are more believable and have increased quality when actual economic data backs up the theory or as Allais explains "when neither the hypothesis nor the implications of a theory can be confronted with the real world that theory is devoid of any scientific interest".9On the other hand, using real economic data to investigate a proposed theoretical connection is not easy.Often empirical testing is so difficult that even though an economic theory can be refuted a couple of times it is simply not enough for it to be dismissed entirely. 10In light of this, it is not the purpose of this study to find definite answers but to highlight what the available data can say about the decision to start a cartel over the business cycle.
The remainder of this article is organized as follows.Section 2 describes the existing theoretical models proposing links between cartel activity and the business cycle.In section 3 the theoretical propositions are re-examined and discussed using the collusion incentive constraint.Section 4 presents various studies that have focussed on the role of cooperation between firms and competition policy during adverse economic events and other related studies.Section 5 presents the data used for the study.Section 6 provides the analysis and main results, and Section 7 concludes.

The proposed links between cartel activity and the business cycle
Rotemberg and Saloner investigate the effect of the business cycle on optimal collusive pricing when demand is subject to shocks. 11Their opinion is that oligopolies may find collusion more difficult when demand is relatively high.Assuming that the price is the strategic variable of firms in a market, the benefits to a deviating firm from a cartel agreement are larger if demand is increasing than otherwise.At the same time, the punishment for cheating the other cartel members is less affected by changes in demand, according to the authors.This means that the incentives to collude change whenever demand changes.Especially if demand changes from low to high.When demand is high, the benefits from deviating from the collusion agreement may exceed the punishment a deviating firm can expect to receive as a parting gift from the former co-conspirators.This leads to competitive outcomes or price wars during boom periods and restored collusion during periods of lower changes in demand.They conclude that collusion is more likely to be counter-cyclical.This implies that cartel formation is more likely during business-cycle contractions.
Haltiwanger and Harrington investigate how cyclical demand movements affect collusive behaviour. 12They find that while the benefits of deviating from the collusive outcome tend to increase in periods of booming demand, firms will find it more difficult to collude during downturns, as the punishment outcome placed on cheaters is lower when demand decreases.They also find that in a market with declining prices collusion has a higher tendency to be pro-cyclical as the payoff from colluding decreases.Moreover, they find that collusive prices tend to be pro-cyclical during booms and counter-cyclical during downturns.They conclude that incentives to deviate from the collusion outcome are greater during a period of economic expansions in the business cycle and it is harder to collude during economic downturns.
Bagwell and Staiger see collusion as a balancing act of short-term temptations to cut prices against the long-term cost of the price war that may come because of such actions. 13They examine sub-game perfect equilibria in collusion games under the assumption that firms adopt symmetric strategies.Their games lead to the most-collusive prices, which are the highest prices sustainable in equilibrium given the one-time benefit from deviating from a collusive agreement and maximal punishment.Maximal punishment, or a grim trigger strategy, means post-cartel breakdown profits of zero.Under these conditions, they can show that most collusive prices are weakly pro-cyclical when growth rates in demand are positively correlated over time.This would mean that cartel formation is more likely during the expansionary phase in the business cycle.Correlation means that the expected profit is larger in the next period compared to today.In that sense, deviating from the cartel means forgoing a future higher profit.Depending on firms' valuation of future profit, or the discounted value of such, the cost of deviating today may be higher than the benefits of doing so.
Fabra investigates the effect of capacity constraints on the sustainability of collusion in markets with cyclic demand.14She finds that collusion is more likely to form during upturns in the business cycle but at the same time also more likely to break down.Capacity constraints imply that punishment profits move pro-cyclical making collusion more likely to be pro-cyclical than counter-cyclical.The periods of increasing demand may lead to lower losses from cheating even if collusion profits are high.Similar to Bagwell and Staiger mentioned above this means that cartel formation is likelier in the expansionary phase of the business cycle.
The articles do not specify exactly when cartel formation is likely to appear in which phase of the business cycle.A business cycle consists of a trough, an expansionary phase from the trough to the peak, the peak itself, and a contractionary phase from the peak to the trough.If cartel formation is linked to the business cycle, it would be reasonable to expect that more cartels are formed as close to the trough or the peak depending on which of the articles you believe in.This ensures that the payoff of the cartel is maximized over time following the progression in the business cycle.
In other words, if Rotemberg and Salinger are right and negatively correlated growth rates create higher incentives to collude, then we would expect to see more cartels forming shortly after the peak in the business cycle.On the other hand, if Haltiwanger and Harrington, Bagwell and Staiger and Fabra are right and positively correlated growth rates cause more cartels to form, then we would expect more cartels to form close to the trough in the business cycle.

Understanding how the demand fluctuations affect incentives
The results of the articles mentioned in the previous section are found using incentive-based models of firm choices.To illustrate how demand fluctuations affect the choices of firms to collude we can use the collusion incentive constraint also used in the articles.The constraint is often used to illustrate how collusion incentives are affected under different market conditions. 15ssume that an industry has n firms.Moreover, assume that the total profit attainable when firms collude is denoted by p c , this will also be the maximum profit possible for any firm if it undercuts the agreement with the other cartel members.In addition, assume that upon cartel break down the firms return to the most competitive state where all receive a profit of p r .Normally this profit could be the purely competitive outcome or it could be less than that if the non-cheating cartel members can agree to punish the cheater.It does not complicate the model by including some kind of punishment for the cheater but it helps towards simplifying the expression in the collusion incentive constraint.The last assumption is that all firms have the same discount factor d, which they use in all periods to discount future profits.
An individual firm chooses to collude if the payoff of doing so is greater than or equal to the payoff it gets from deviating from the cartel and receiving the most competitive outcome payoff for as long as the cartel would have lasted, i.e. for a non-determined infinite period.
This gives the following collusion incentive constraint, which must hold for collusion to be preferable for the firms, (1) The equation tells us that to change the incentives to collude the relative profit between the most-competitive profit outcome, i.e. p r , and the cartel profit, i.e. p c , cannot change in the same way if demand changes.If it did the incentives to deviate (and collude) would not change when demand changes.It is unlikely that demand affects the profits differently given that the profits are subject to the same market conditions.Nevertheless, it is worth examining in more detail.
For the sake of simplicity, assume that the particular market has a q i .Moreover, assume that all firms have the same marginal cost of c and that the payoff in the case of collusion and most-competitive outcome both relates to this demand illustrating that demand in each period for the decision to form a cartel, is made under the same demand conditions.Using an nfirm Cournot model gives us the following symmetric payoff for the collusion outcome of p c = (A − c) 2 4 and a most-competitive outcome payoff that depends on the number of competitors of . 16 When inserting the two payoffs in Equation (1) we get, In other words, demand conditions are not at all relevant for assessing incentives to collude, since changes in demand change collusion profit and most-competitive profit in the same way.
We get the same result if we simply assume that the reversal profit is zero, which is what for example Haltiwanger and Harrington assume when they say that firms after a cartel member cheats apply a grim trigger strategy in all future periods afterwards. 17In both cases, the relevant question would then be why changes in demand have different effects on the incentives to collude if the changes affect the cartel profit and most competitive profit in the same way.This means that we must assume some kind of asymmetry in how demand changes affect the two payoff outcomes to conclude in the same way as the theoretical models in Section 2.
Since the mentioned articles lead to theoretical results, it is important to understand just how they arrive at them.In all four cases, the articles do not incorporate the demand change into the model they use for deriving their results.In that way, demand changes become an exogenous variable in the models.For example, in Bagwell and Staiger it is assumed that the demand changes do not have price effects. 18This is a problematic assumption.It does help to simplify the analysis but it implies that we have either a completely inelastic demand or have Bertrand competition.To assume inelastic demand is, of course, a possibility but it seems unrealistic that there are no price effects on markets with fluctuating demand.Industrial sectors that are very susceptible to changes in the business cycle such as manufacturing are usually also industrial sectors where prices are elastic.Bertrand competition is a theoretical possibility more than an observable fact in oligopolistic markets.
In Bagwell and Staiger the cartel profit is multiplied by the growth in economic activity. 19Haltiwanger, Harrington and Fabra assume that a grim trigger strategy applies if the cartel breaks down. 20This result in a reversal profit of zero, similar to firms facing Bertrand competition, and means that only the collusion profit will change over time.This is exactly what seems to lead to the conclusion that cartel formation is affected by the business cycle.Grim trigger strategies or not it is not a simple thing to dismiss that cartel activity may be linked to the business cycle.
The part of the incentive constraint that specifies the payoff from forming a cartel and from not doing so can tell us how changes in the business cycle affect firm profit.Finding the derivative of both profits 17 See Haltiwanger and Harrington (n 3). 18See Bagwell and Staiger (n 3). 19ibid. 20See Haltiwanger and Harrington (n 3) and See Fabra (n 3).
will tell us how the incentive change when demand changes.Again using the n-firm Cournot model the derivative of the cartel solution is,21 And, the derivative of the reversal profit is, A comparison of the two derivatives reveals that, This means that when the demand is increased, then p c increases more than p r .This means that the benefit of cartelization increases more than the costs and the incentives to collude should be higher.It also implies that whenever demand is decreasing then p m gets closer to p r and makes collusion less likely to happen.This is of course a different way to look at incentives as an absolute difference between the two payoffs.It does imply that the effect of a change in demand on the incentives to collude is, The implication of this is that when the economy exhibits high growth, which typically coincides with the period just after the trough in the business cycle, then both the difference in payoffs, i.e. p c − p r , and d are relatively large.This means that more firms will find cartelization more attractive than otherwise, and the opposite will be the case if A decreases.In other words, if we just look at the effect of demand changes on individual profits, it may be the case that cartel formation is linked to the way demand fluctuations change firms' profits.

Existing studies on cartels and the business cycle
Not many studies have been made on cartel activity and the business cycle.Except for a working paper by Garcia et al., 22 all the studies mentioned in this section look at specific events in the business cycle and their relevance to competition policy.The events are the financial crisis of 2008-2009 and the Covid19-pandemic of 2020.The working paper of Garcia et al. focuses on the relationship between cartel lifecycles and business cycles and to what extent collusion is pro-or countercyclical.The authors use a Poisson regression model on data collected from sanctioned cartels identified by the EC in published cartel cases.The results show that collusion appears to be pro-cyclical and counter-cyclical concerning their demise.Using a Poisson regression seems attractive as the number of cartels that starts in any given period is a count variable.On the other hand, the restricting assumptions of variable association and mean and variation being of the same size is problematic.This may especially be so as the data does not have such characteristics, see Sections 5 and 6.
In 2009, the OECD wrote a discussion paper on the role of competition policy during the unfolding financial crisis. 23In the paper, it is discussed whether competition has enhancing effects on systemic financial crises.Although no clear conclusion is made the tone of the paper and the inclusion of several policy recommendations suggest that competition is not preferable during severe economic crises.Among the policy recommendations, is different exit strategies that could be used by governments to revert to a more competitive environment in the financial sector when the crisis has been resolved.The paper addresses how to implement different types of competitive constraints to prevent bank runs, moral hazards and excessive risktaking of financial institutions, and to eliminate contagion.The paper concludes that the best way to suspend competition rules for a period is by allowing governments to intervene in the management of financial firms and disregard anticompetitive measures including coordination made by private firms to prevent the aforementioned problems.Although this is an example from the financial crisis the principles outlined in the OECD paper of suspending competition during a severe economic crisis or downturn became a practical solution to the economic adverse situation.
In 2020, the OECD published the report "Competition policy in times of crisis" in which the role of competition policy was discussed in light of the outbreak of the Covid19-pandemic.The conclusion is the same as for the financial crisis, that competition rules should be suspended or at least be unusually lenient until the crisis is over. 24n both cases, the more lenient approach to competition policy during severe adverse events in the economy may explain why some cartels are started or even have lasted longer than expected.

The data on cartels and the business cycle from 1991 to 2012
The cartel data used in this study covers cartels identified by the European Commission in the period from 1991 to 2012.The information about the cartels is gathered from cartel cases published by the European Commission's General Directorate for Competition.
The legal definition of a cartel differs from the economic definition on several important points.In economics, collusion takes place when firms coordinate their prices or quantities, whereas in a legal context collusion covers all types of cooperation between firms that legally fulfil the criteria for being a restrictive agreement according to the EU-Treaty article 101 (1).This implies all imaginable types of cooperation between firms on all different matters including those that are under normal circumstances would be legal such as joint ventures on technological improvements or R&D. 25 The vast majority of the firms in the dataset are firms that have agreed on geographically exclusive areas of interest, dividing customers, or set quantity restraints on the supply of their products, and only a smaller number of the cartels involve price coordination.
The data consists of 108 cartel cases, covering 94 different products, and 631 firms of which 72 are repeat offenders.The oldest cartel started in 1969 and the newest in 2011.The data covers the biggest firms in the EU and includes firms which originate outside of the EU but are included since they have operated at the time of the cartel agreement within the EU's internal market.The companies match one or more of the following conditions: having an operating revenue of more than 10 million EUR, having total assets of more than 20 million EUR, or having employees of more than 150.A clear majority of firms originate from Germany followed by France, and the UK a clear indication of where the largest firms in the EU come 24 See Competition Policy in Times of Crisis: Supplement to Competition Policy in Eastern Europe and Central Asia (OECD-GVH Regional Centre for Competition in Budapest (Hungary) July 2020). 25For more on this see COUNCIL REGULATION (EC) No 139/2004 of 20 January 2004 on the control of concentrations between undertakings (the EC Merger Regulation).Official Journal of the European Union, L 24/1.
from.Due to limited access to data for firms with employees of less than 250 persons, not all firms in the dataset are included in Table 1.Any investigation on the link between the decision to start a cartel and the business cycle must first examine what the available data can tell us about the industries to which the cartels belong.If the majority of cartels start in industries that are less sensitive to changes in the business cycle, then the data would not be suitable to investigate how changes in the business cycle affect the decision to form a cartel.
The majority of cartels in the dataset come from industries that are very sensitive to business cycle changes, such as manufacturing, transportation, financial services and wholesale.Using the NACE rev. 2 to categorize the cartel members' production reveals that a clear majority of 98.1% of the cartels come from highly business cycle-sensitive industries.Other industries are typically perceived as less susceptible to changes in the business cycle.Agriculture is the only category in the dataset that fits this description.The category only consists of 1.9% of the total number of firms, see Table 2.
The period the final dataset covers are from 1991 to 2012.No meaningful measurement of EU GDP exists before 1991 and 2012 is the last published cartel case publicly available.The period is interesting from a competition policy point of view as it covers some important competition policy changes with relevance for cartel enforcement.In 1996, the so-called leniency programmes were introduced.They allowed firms in a cartel to get a fine reduction when helping the EC with the cartel case.In 2002 and again in 2006, the programme was revised making it possible for cooperating firms not to be fined. 26The objective was to increase the build-in instability of all cartels and the programme has in many ways been the most important instrument for decreasing the duration of cartels and limiting their numbers.The same period is also characterized by increased competition in the Internal market by creating better opportunities for trade between the EU member states and between the EU members and external partners.With the conclusion of the Maastricht Treaty in 1991, the foundation was laid for the Euro and border-free travel.In 1999, the first euro was issued and in 2007, the Lisbon Treaty was concluded further strengthening the integration of the member states' markets.In addition, it became easier during this period to cooperate by merging.In 2004, the EU created a new common merger regime with common rules on what a merger is and how it should be treated. 27he other part of the data used in this study is for the EU business cycle.The data was used to identify expansionary and contractionary phases.Respectively defined as economic growth rates above the natural trend or the average growth rate for the period in the economy and contractions as phases with lower growth than the natural trend.The data set has four periods of expansion and four periods of contractions of different lengths.
It was also possible to identify two severe adverse economic events during the period, defined as periods with negative growth rates.The first event occurred in 1993 and the other in 2009.The negative growth in 1993 was the result of a too-tight monetary policy in an attempt to control rapidly rising inflation.The negative shock in 2009 is the financial crisis.Another well-known crisis that took place in the period was the 2001 to 2002 dot-com bubble.It does not appear in the data as a severe shock to the economy in the same way as the aforementioned ones.The main reason for this is that despite the repercussions to the economy it was not an economy-wide crisis affecting many sectors at the same time like for example the financial crisis of 2009.Instead, the impact was on a relatively smaller part of the economy.The economic growth dropped but never became negative.The dot-com crisis is visible as a lower-than-usual growth rate for 2003.Over the entire period, the average duration of the cartels that start and the number of cartels detected is decreasing, see Table 3.
The trend in the number of cartels detected is decreasing over time and no new cartels were detected at the end of the data period.The last year is included in the dataset as it is the last reporting year for cartel activity available on the EC's website.The number of starting cartels drops in 2005 and afterwards.This is likely a consequence of the announcement of the EC's intention to change the leniency programme, see Figure 1.

Results on the choice to start a cartel and the business cycle
The cartel data were used to make two types of investigations both to examine if positive or negative correlated growth rates create higher 27 ibid.
incentives to collude and if more cartels form shortly after respectively the trough or the peak in the business cycle.The first investigation looked at the prevalence of cartels formed during contractions and Notes: "E" is for an expansion and "C" is for a contraction.Cartels start is the number of cartels that starts that particular year.Source: The author.expansions in the business cycle.In the second investigation, the dataset was divided into three parts in line with the number of full cycles from trough to peak in the dataset.
Cartels start in every period and all years except the last year in the dataset.This confirms that cartelization can happen irrespective of the state of the business cycle.The dataset has 22 years in total during which 7 years were contractions and 15 years were expansions.Of the 88 cartels, 21 started during years of contraction and 67 started during an expansion.This gives an over-representation of cartels that start during expansions and an under-representation of cartels starting during contractions.Compared to the natural probability of a cartel starting in any of the two periods with the same probability there is an 8.0 percentage point difference between the observed occurrence of starting cartels and the natural probability.Applying a statistical test for similar proportions it can be confirmed that the proportions are significantly different to the natural occurrence.It is therefore much more likely that a cartel starts during an expansionary period than during a downturn.This supports the propositions of Haltiwanger and Harrington, Bagwell and Staiger, and Fabra, see Table 4.
After identifying the number of full business cycles in the dataset the data was split into three smaller subsets.The purpose was to examine to what extent cartels are started early or late in the business cycle.If the results show that firms start early in the business cycle it may suggest that starting a cartel is primarily pro-cyclical.The number of cartels starting in each of the three periods can be seen in Table 5.
To be sure that the regressions only looked at expansionary periods observations after the peak was excluded.This limited the data and it was deemed that using the same method for looking at tendencies in contractionary periods would be pointless given the thin data material.The data for the three regressions can be seen in Figure 2. The next step was to calculate the three regression lines using the split dataset for data covering the trough to the first coming peak and examine the resulting coefficients and constants statistically.The regression line for the expansionary period from the trough to the peak covering the period from 1993 to 2001 produced a positive but statistically insignificant value for the coefficient (0.814 > 0.05).The constant value was statistically significant (0.009 < 0.05).This confirms the result found in Section 4 that changes in demand do not influence the incentives to start a cartel.The next regression covers the period from 2003 to 2006 and includes the year the revised leniency   programme was introduced in the EU.The purpose of the leniency programme was to destabilize cartel formation further by undermining the incentives to collude.The programme had the intended effect seen by the relatively large decrease in the number of starting cartels in comparison to before the programme change.The regression line, in this case, has a negative but statistically insignificant coefficient of −0.545, with a p-value of 0.455.The constant is also statistically insignificant.
The last regression was for the period 2009 to 2011 and produced a negative insignificant coefficient close to the 0.05 confidence level.The constant was only significant at the 0.05 level.The results are summarized in Table 6.
Given the results of the three regressions, a strong association between periods and the number of starting cartels does not seem to exist.One should be careful when making assertions on the available data since the observation points rapidly decreased in the two last periods.The likely explanation is the progressive change in competition policy and how it dramatically changed the incentives to form cartels.

Conclusion
The results can confirm that more cartels form during the expansionary phase in the business cycle, but the results cannot confirm that positive or negative correlated growth rates create higher incentives to collude or that more cartels form shortly after the trough or the peak in the business cycle.Therefore, this study is unable to confirm that a relationship between cartel formation and the business cycle exists in line with the theoretical propositions.Consistent with the results it is likelier that there is no relationship at all.
The regression lines cannot confirm that a linear relationship exists between the starting point of cartels and the number of cartels that start.This said the regressions after 2002 produce negative but statistical insignificant coefficients, although with far fewer data points available, which alone may explain the different results to the first regression.
choice of how to incorporate demand fluctuations in the core theoretical model is important for the results and the conclusions.The same can be said about how the authors interpret demand fluctuations and what counter-and pro-cyclical means.The article by Rotemberg and Salinger takes great care in not linking demand fluctuations to the macroeconomic view of the business cycle.4Instead, they talk about demand fluctuation as market demand increases or decreases over time.The other authors describe demand fluctuations as being linked to the changes in the business cycle.

Figure 1 .
Figure 1.The trend in starting cartels in the period 1991 to 2012.Source: The author.
T" stands for trough, and the plus and number indicate the years after the trough."P" stands for peak and the plus number is the years after the peak.

Figure 2 .
Figure 2. Trends in cartel start from trough to peak for three periods.Notes: 'T' stands for trough, and the plus and number indicate the years after the trough.'P' stands for peak and the plus number is the years after the peak.Source: The author.

1
See Commission Notice on Immunity from Fines and Reduction of Fines in Cartel Cases.Official Journal of the European Union, 2006/C 298/11.
2 See J Rotemberg and G Saloner, 'A Supergame-Theoretic Model of Business Cycle and Price Wars during Booms' (1986) 76 American Economic Review 390. 3 See J Haltiwanger and JE Harrington, Jr., 'The Impact of Cyclical Demand Movements on Collusive Behaviour' (1991) 22(1) The RAND Journal of Economics 89.K Bagwell and RW Staiger, 'Collusion Over the Business Cycle' (1997) 28(1) RAND Journal of Economics 82.N Fabra, 'Collusion with Capacity Constraints Over the Business Cycle' (2006) 24 International Journal of Industrial Organization 69.

Table 1 .
Origin of firms and occurrence in sample data and the total number of firms in the economy within the category.

Table 2 .
The cartel data characteristics; broad category code and name and share of total cartel data.

Table 3 .
The business cycle, starting cartels, average duration and real GDP growth.

Table 4 .
Observed cartel starts, natural occurrences and differences.

Table 5 .
The number of cartels starting during three cycles.

Table 6 .
Statistical results on the association between the number of cartel starts and three business cycles.