The e¤ect of EU antitrust investigations and …nes on a …rm’s valuation Luca Aguzzoniy, Gregor Langusz, Massimo Mottax 31 May 2012
Abstract A typical EU antitrust investigation involves a sequence of events (surprise inspection, Commission Decision, Court judgment) which affect the investigated …rm’s market value. We …rst model these relationships, and then use event study techniques to estimate the impact of these antitrust events on …rms’share prices. A surprise inspection reduces on average a …rm’s share price by 2.89%, an infringement Decision by the European Commission reduces it by 3.57%. The Court judgments do not have a statistically signi…cant e¤ect on the …rm’s valuation. After weighting for capitalization, we …nd that the total effect of antitrust action ranges from -3.03% to -4.55% of a …rm’s market value. Fines account for no more than 8.9% of this loss, and we conjecture that most of the loss is due to the cessation of illegal activities.
We are grateful to the Editor, Saul Lach, three anonymous referees, Francesco Amodio, Ricardo Cabral, Claudio Calcagno, John Connor, Decio Coviello, Tomaso Duso, Gabrielle Fack, John Fingleton, David Genesove, Vivek Ghosal, Paul Grout, Joe Harrington, Andrea Ichino, Kai-Uwe Kühn, Mel Marquis, Walter Mölls, Michele Polo, Patrick Rey, Mark Schankerman, Richard Spady, Helder Vasconcelos, Frank Verboven, and especially Liliane Karlinger and Kirti Mehta for comments and discussions. We are also grateful for comments received during seminars at IO Annual CEPR Workshop, Leuven, EUI Economics Department, Eitan Berglas School of Economics, Tel Aviv, the 2006 Annual EU Competition Law and Policy Workshop at Florence, Carlos III (Madrid), CORE (Louvain La Neuve), Mannheim, Toulouse, LUISS (Rome), NHH (Bergen). Finally, we are grateful to Dimitrios Magos, Emanuele Tarantino, Francesca Bassetti, Giacomo Pansolli and especially Antoine Chapsal for their research assistance. Part of this work was carried out while Langus was visiting the Eitan Berglas School of Economics, Tel Aviv, whose hospitality is gratefully acknowledged. A previous version of this paper was circulated with the title "Have Antitrust Fines in the EU Had Any E¤ects on the Firms?", CEPR Discussion Paper 6176 (2007). y LEAR, Roma.
[email protected] z CRA International, London.
[email protected]. x ICREA-Universitat Pompeu Fabra and Barcelona Graduate School of Economics;
[email protected]. Ramon Trias Fargas 25-27, E-08005, Barcelona (Spain).
I
Introduction
Antitrust laws are fundamental in market economies, as they prevent …rms from distorting competition in a way that is detrimental to economic ef…ciency, and …nes are a crucial tool for the enforcement of antitrust laws. Only if the penalties that …rms incur when found guilty of antitrust infringement are large enough, will the …rms be deterred from engaging in anti-competitive behavior. In the US, managers who have been found guilty of a conspiracy can be given prison sentences, and …rms are subject to …nes and to the payment of treble damages in private actions. In the EU, which is the object of this study, competition law violators are not subject (at EU level) to criminal penalties, and private damages actions are extremely rare, but …rms can in principle be given …nes up to 10% of their previous year’s turnover. Yet, anecdotal evidence suggests that the impact of antitrust investigations and …nes may be weak. Indeed, a large number of …rms (and in fact several …rms from the sample we analyze in this paper) are repeat o¤enders.1 Moreover, infringement decisions by the European Commission and Community Court judgments do not seem to trigger management changes very often. This raises the question of the extent to which …rms are seriously a¤ected by the …nes they receive, or expect to receive.2 In this paper, we carry out (by using event study techniques) an empirical analysis to explore the e¤ect of antitrust investigations on the share prices of …rms which have infringed European competition law (for abuse of dominance, cartels, or other anticompetitive agreements). There are two main novelties in our work. Firstly, this is the …rst work which estimates the impact of European antitrust investigations on o¤ending …rms, and to this purpose we have constructed an original database.3 Bosch and Eckard (1991) carried out a similar exercise for the US, to estimate the e¤ect on the …rm’s stock market price of an indictment for price …xing.4 They …nd that the shares of indicted …rms in their sample on average lose a cumulative 1:08% of their value in the days immediately after the public announcement of the indictment.5 1 Repeat o¤enders in our sample are typically large …rms which have infringed competition law in di¤erent product markets and di¤erent countries. 2 Connor and Bolotova (2006), building on a large number of studies, estimate the mean cartel overcharge at around 29% per year. With such margins over the competitive prices, one may expect antitrust …nes to result in considerable under-deterrence. 3 See Duso et al. (2007) and Duso et al. (2011) for empirical analyses of the e¤ects of EU merger noti…cations and decisions. They also make use of the event study methodology. 4 Bizjak and Coles (1995) carry out another event study analysis on US data relative to private antitrust litigation. They …nd that, on average, defendants lose approximately 0.6 percent of their equity value (and plainti¤s gain less than what defendants lose). See also Detre and Golub (2004) for an analysis on recent US antitrust data. 5 An indictment by the US Department of Justice should be ’news’to the markets, as the indictment is preceded by investigations which are supposed to be secret.
1
Secondly, since we analyze the e¤ect of di¤erent but related events, we propose a simple model of the antitrust procedure which captures the relationships between these antitrust events. The model allows us to predict the sign that each of these events would have on the …rm’s share prices, and to see why each event brings new information to the market. Our estimates suggest that the capitalisation-weighted average total effect of an antitrust action is to decrease the …rm’s stock market value by between 3:03% and 4:55%. Importantly, we also …nd that only a small part of this e¤ect (no more than 8:9%) is due to the …ne. We conjecture that most of the loss is due to the fact that the market anticipates that, after an antitrust action, an anticompetitive practice will cease, leading to lower pro…ts.
I.i
The EU competition law institutional framework, in a nutshell
Since our objective is to estimate the e¤ect of antitrust investigations in the European Union, it is appropriate to brie‡y remind the reader of the main actors in the …eld of EU competition law, and of the main events which occur in a typical investigation. The European Commission is the main competition authority for the enforcement of EU competition law, whose main provisions are contained in articles 101 (anticompetitive agreements) and 102 (abuse of dominant positions) of the Treaty establishing the European Community. Fines can be imposed on …rms which have infringed articles 101 or 102, and they are decided at the discretion of the Commission, whose decisions are however subject to the review of the Community Courts, i.e. the Court of First Instance (CFI - now renamed General Court) and the European Court of Justice (ECJ). Fines can never be higher than 10% of the …rm’s worldwide turnover in the previous year; they should be proportional to the gravity and duration of the infringements; and they cannot consist of criminal penalties. In 1998, the Commission published a Notice containing the Guidelines (i.e. a code of practice) that it would follow in deciding …nes.6;7 However calculated, commentators (and the Commission itself) agree that, until 1979 (with the Pioneer Decision, which is also the …rst Decision in 6
On 28 June 2006, the European Commission revised the Guidelines for setting antitrust …nes. However, the vast majority of our observations refer to cases which were initiated under the old Notice. 7 Since relevant market turnover data are typically not published in the Commission Decisions for con…dentiality reasons, it is not possible to identify whether the base …ne is computed as a percentage of turnover. This should change in the future: the 2006 Guidelines provide that the base …nes may be up to 30% of the company’s annual sales in the market to which the antitrust infringement relates, multiplied by the number of years of participation in the infringement, provided the total is within the limit of 10% of the …rm’s total annual turnover.
2
our sample), the Commission was very lenient when imposing …nes, whereas towards the mid-90’s the Commission started to impose increasing …nes.8
I.ii
How an antitrust investigation proceeds
The European Commission, or more precisely its Directorate General for Competition (DG-COMP), begins its investigation either at its own initiative or on the basis of a complaint from a third party (although, if complaints occur, the Commission has no obligation to start an antitrust procedure). There is typically no announcement that an investigation has started, and no precise time frame for it. If during the preliminary stages the Commission has serious suspicions that there has been an antitrust infringement, it can carry out a surprise inspection, also called Dawn Raid (henceforth Raid ), on the premises of the …rm(s), to gather documentary evidence (which is absolutely crucial for anticompetitive agreement cases, but relevant for abuse cases too).9 This inspection should represent a genuine surprise for the investors. To verify that this is really an unexpected event, we examined past issues of the Financial Times for any news about the (potential) investigation before the inspection took place, and we could not …nd any, for any of the …rms for which we have dates of the Raid.10 A well-established jurisprudence obliges the Commission to take steps to respect the rights of the defendants during the investigation.11 Among these, the Commission has to send a "Statement of Objections" to the …rms under investigation, where it states its allegations regarding the practices of the …rm and asks for the …rm’s response.12 After having analyzed all the evidence and having heard from the parties, the Commission might either take a formal infringement Decision or decide to close the case. In the latter case, there may be a non-infringement decision; an announcement through a press release; or - by far the most common outcome - no public statement at all (hence, there is little or no in8
See for instance Geradin and Henry (2005, p. 20 and ¤.). Pursuant to Regulation 1/2003, the Commission can also conduct surprise inspections at the homes (and private vehicles) of …rms’managers and employees. 10 It is of course possible that investors may nonetheless anticipate that an investigation will take place. This may be the case in particular for some of the international cartel cases which appear in our sample, where a US antitrust case precedes the EU investigation. To deal with this issue we shall report separate estimates for dawn raids when the same cartel had already been investigated in another jurisdiction and when there have been applications for leniency. 11 Indeed, several Commission Decisions have been annulled by the Community Courts on various procedural grounds. 12 We also carried out an empirical analysis of the e¤ects of the Statement of Objections, but as expected - it is largely a procedural step which does not reveal substantial new information to the market - we did not …nd any signi…cant e¤ect of this event on the value of the …rm. We shall report estimates for the Statement of Objections in Section IV.ii below. 9
3
formation about it). Whatever the Commission’s verdict, it may be reached a long time after the Raid (on average, three to four years to an infringement Decision). A relevant feature for our analysis is that the Decision is a collegial act of the whole European Commission, not of DG-COMP, and before taking it several bodies are consulted, such as representatives of national competition authorities and members of other directorates general. Although all the people involved are bound by con…dentiality clauses, leakages about (or speculations on) the content of the Decision and the level of the …nes are common.13 Firms which have been …ned can appeal to the Community Courts, which can rule upon the merits of the Commission Decision, and whose Judgments can annul, uphold or modify the …ne, as well as of course annul or uphold, completely or partly, the overall Decision.14 The decisions taken by the Court are not made public until the moment they are announced, although in some cases there may be signs of the judges’views.15 The paper continues in the following way. Section II presents a stylized model of the antitrust procedure, with the aim of formulating testable predictions. Section III describes our data and explains our estimation procedure. Section IV reports the results of our analysis and discusses their robustness. Section V discusses the economic interpretation of the results. Section VI concludes the paper.
II
A simple model, to obtain testable predictions
Since the antitrust procedure involves di¤erent events which take place successively and are related, we propose a simple model of this procedure. Although very stylized, the model guides our analysis by o¤ering some testable predictions on the e¤ects of the events. Suppose that a …rm has engaged in a certain anticompetitive business practice,16 and that antitrust events (decision to open an investigation, infringement decision, court judgment) all occur with certain probabilities. This may be rationalized as a situation where the outcome of an investiga13
Nevertheless, rumors and speculations are typically concentrated in a period of one month before the date of the Decision. 14 In older cases, the …rms’ appeal was decided by the ECJ. In more recent years, it is the CFI which decides; …rms can also appeal the CFI’s judgment. We do not look at this ’second’judgment, and only consider the …rst judgment, whichever Court takes it. 15 In particular the opinion of the Advocate General often (though not always) anticipates the judgment of the Court. However, Advocates General are only involved in the ECJ’s procedures and not the CFI’s. 16 The practice may concern either abusive behaviour or cartel participation. We consider a …rm’s infringement decision in isolation for simplicity. The model could be extended to deal with cartel decisions, by analysing the incentive constraint for collusion of the …rms involved, but this is beyond the scope of the present paper.
4
tion depends on some factors - such as the discovery of documental evidence and the respect of the procedures - that may be casual. For simplicity, we also ignore the fact that time elapses between one event and the following one, and accordingly we consider neither discount factors nor interest rates. These assumptions are admittedly very crude, but they allow us to emphasize some simple relationships among the antitrust events.
"place Figure 1 approximately here." A simple formalization of the antitrust procedure is given in Figure 1.17 At time 1, Nature determines whether the …rm will be subject to a surprise inspection - event which takes place with probability m - or not. If no Raid is undertaken, though, we assume that the …rm will never be investigated any longer, and it will have (anticompetitive) net present value V M .18 If a Raid takes place at time 1, the Commission will investigate the practice further. With probability 1 p, the Commission will not …nd proof of the infringement and the case will be dropped. The …rm will not be investigated any longer and it will have value V M . With a probability p the Commission will …nd proof of an infringement and at time 2 it will issue an Infringement decision imposing a …ne, F , and ordering the …rm to cease the anti-competitive practice. In this case, though, we assume that the …rm always appeals the infringement decision. (This is largely consistent with what happens in reality - where most Decisions are appealed -, and of course it makes sense in the model because the cost of appealing is taken to be zero for simplicity.) At time 3, the Court upholds the Commission Decision with probability q and annuls the …ne with the remaining probability 1 q. Of course, the Court is free to set any level of the …nes it deems correct, so the …ne should be a continuous variable. To simplify matters, though, we assume that it has a binary choice.19 If the Judgment is in favour of the Commission, the …rm will pay the …ne F and will have competitive pro…ts forever, resulting in a …rm’s net present 17
One may use this simple model to consider the …rm’s choice between violating and complying. It would be su¢ cient to include a time 0 node where the …rm decides whether to violate the law or not in a particular market. If it does not, its (competitive) present discounted value will be V C . The model could then be used, for instance, to identify the optimal …ne necessary for deterrence. 18 A slightly more sophisticated version of the model would be that in each period the Commission could do a surprise inspection from the pool of the …rms which have not been investigated previously, but this would not qualitatively change the results. 19 In our event study analysis, we de…ne as ’annulment’a Court judgment which reduces the …ne to below the 1/2 of the …ne proposed by the Commission, and ’upholding’when either …rms do not appeal (there are a few such cases in our sample) or the Court …ne is above 1/2 of the original one.
5
value V C . Otherwise, the …rm will not have to pay any …ne and will have expected value V C + (1 )V M , with 2 [0; 1], included between two extreme cases. (i) The Court may annul or drastically reduce the …ne while agreeing with the Commission that there has been a violation, resulting in the …rm having to cease the practice at hand despite the (relatively) favorable judgment. In this case, = 1, with the …rm having a value V C . (ii) The Court may not only annul the …ne, but also disagree with the analysis of the Commission, rejecting its allegations of anticompetitive behavior. In this case, = 0, with the …rm having a value V M . We believe that case (i) corresponds with what would happen in cartel cases where the Commission is unlikely to issue an infringement decision unless it has very strong documental evidence, while case (ii) might …t cases of abuse of dominance or anticompetitive agreements (other than cartels) where economic and legal analysis play a bigger role than factual evidence and where discretion therefore matters more. (In fact, though, the Community Courts are typically in agreement with the Commission’s analyses.) In order to investigate how the occurrence of a certain antitrust event a¤ects the valuation (that is, the net present value) of the …rm, let us …nd …rst the value of the …rm after each particular event. The expected value of a …rm that violates competition law is: VV iolation = mpq V C =V M
F + [1
mpqF
q)( V C + (1
mp] (1
mp V M
V C (q (1
)V M ) + (1
mp)V M
(1) (2)
) + ):
After a Dawn Raid, and before a Decision, it is: VRaid = V M
pqF
p VM
V C (q (1
) + ):
(3)
After an infringement Decision, it is: VDecision = q(V C
F ) + (1
q)
V C + (1
)V M :
(4)
Finally, after a judgment upholding or annulling the Commission Decision, the …rm’s expected value will respectively be: VU pheld = V C
F ; VAnnulled = V C + (1
)V M :
(5)
We can now compute the e¤ect of an event on the expected value of the …rm. First of all, the occurrence of a Dawn Raid will change the …rm’s value as follows:
Raid
=
VRaid VV iolation = VV iolation
p(1 VM
m) qF + V M V C (q (1 mpqF mp (V M V C ) (q (1
6
)+ ) < 0: (6) )+ )
When a Dawn Raid takes place, the market correctly understands that the probability that the …rm may be obliged to stop the lucrative anticompetitive conduct and pay the …ne is now higher than before the Dawn Raid took place, resulting in the …rm’s expected market value to decrease. This leads to: Prediction 1: If the event "Commission undertakes a Dawn Raid" is observed, then we should expect the share price of the …rm to decrease. An infringement decision will change the expected …rm’s value as:
Decision
=
VDecision VRaid = VRaid
(1 p) qF + V M V C (q (1 V M pqF p (V M V C ) (q (1
)+ ) < 0: (7) )+ )
In words, if the Commission issues a negative decision, the …rm’s expected market value will decrease because it is more likely that the …rm will have to stop anticompetitive conduct and it will ultimately have to pay the …ne. We can then state: Prediction 2: If the event "Commission issues an infringement decision" is observed, then we should expect the share price of the …rm to decrease. After a Court’s Judgment which upholds the Decision, the change in the …rm’s value will be:
U pheld
=
VU pheld VDecision = VDecision
q) F + V M V C (1 F ) + (1 q) ( V C + (1
(1 q(V
C
) < 0, (8) )V M )
whereas after a judgment which annuls the Decision, it will be:
Annulled
=
q F + V M V C (1 ) VAnnulled VDecision = > 0. C q(V F ) + (1 q) ( V C + (1 )V M ) VDecision
This results in the following: Prediction 3: If the event "Judgment upholds the …ne" is observed, then we should expect the share price of the …rm to decrease. If the event "Judgment annuls the …ne" is observed, then we should expect the share price of the …rm to increase.
7
(9)
Our stylized description of the antitrust procedure also allows us to perform some comparative static analysis. In particular, we are interested in studying the e¤ects of q and F , since there are clear indications that in the period we consider these variables change value over time. It is easy to see that the probability that the Court upholds the Commission Decision a¤ects the expected changes in the …rm’s value as follows (let us focus on absolute values of the changes): @j
Raid j
@q
> 0;
@j
Decision j
@q
>0
@j
U pheld j
@q
< 0;
@j
Annulled j
@q
> 0;
That is, if the court upheld the Commission Decision with a higher probability one would expect to …nd a stronger e¤ect both of the Raid and of the Decision (since the market would anticipate that the infringement is more likely to be con…rmed in appeal). As for the judgment, a higher probability that a Decision is upheld will imply a lower e¤ect on share returns from upholding (since the market has already discounted it), but a stronger e¤ect from annulling the Decision (which would now ’surprise’ more the market). The severity of the …nes imposed by the Commission a¤ects the expected changes as follows: @j
Raid j
@F
> 0;
@j
Decision j
@F
>0
@j
U pheld j
@F
> 0;
@j
Annulled j
@F
> 0:
That is, a higher …ne would ceteris paribus increase the magnitude of the e¤ects of all antitrust events: the more important the …ne the stronger the impact of the events on the …rms’pro…ts, and thus on their market values. Several commentators have expressed the view that the probability that the Court approved the Commission Decisions has increased after the early enforcement period, re‡ecting the fact that the Commission has dealt with the rights of defence of the …rms more carefully, and that it has imposed …nes in a less arbitrary way (for instance by issuing guidelines), both leading to higher upholding rates by the Courts. Also, it is well established that …nes have increased over time (for instance see the below Figures 2 and 3). Unfortunately, though, the fact that around the year 1998 (when the …rst Guidelines on the imposition of …nes were issued) both q and F increase makes it di¢ cult to evaluate separately the role played by these two variables. However, we can identify the following prediction. Prediction 4: If both the probability that the Court upholds the …ne, q, and the …nes imposed by the Commission, F , increase, then we should expect the events "Raid", "Infringement 8
Decision", and "Judgment annuls the Decision" to result in a stronger e¤ ect on the share price.20
III
Estimation of abnormal returns
In this section, we describe our data and estimation procedure.
III.i
Data
Our data come from Commission Decisions, published in the O¢ cial Journal of the European Communities, and judgments of the Court of First Instance and the European Court of Justice, published in the European Court Reports and other sources. In these Decisions the Commission describes the investigation and often reports the date of the surprise inspection, if it was made. The data refer to all the infringement decisions from January 1969 until November 2009. By looking at these Decisions we could …nd 984 observations relating to a total of 190 infringement cases involving 823 di¤erent …rms.21 From this sample we have retained only the Decisions involving the …rms listed in a stock exchange.22 The objective is indeed to estimate the e¤ect of antitrust investigations on the market value of listed …rms.23 Our …nal sample refers to 91 cases (the …rst of which dates from 1979) involving 180 …rms (some are repeat o¤enders), for a total of 240 single observations in which the Commission …nds an infringement of either article 20
In principle it is also possible to look at how the probability p of an infringement Decision a¤ects the magnitude of the changes in share values (one can check that @j Raid j j > 0; @j Decision < 0). However, the Commission does not systematically release @p @p information about …rms which have been raided, so in our database nearly all the …rms have been raided and have received an infringement decision, which makes it impossible to calculate p. 21 An observation refers to a particular …rm/case combination. For instance, when the same …rm is object of two di¤erent infringement decision it will appear twice in our dataset. 22 Over the period considered, listed …rms represents around one quarter of the entire population of antitrust investigations (when considered at the single …rm level). However, if we look at more recent years our sample is even more representative. For instance, in year 2008, around 50% of the observations refer to listed …rms. The …nancial data are retrieved for the Datastream International database. 23 The results of this paper might not re‡ect the average impact of antitrust investigations on the total value of non listed …rms. Publicly listed …rms are typically subject to a closer scrutiny by the regulators and, more generally, the public. One consequence of this might be that the events which a¤ect the future income streams of such …rms are likely to receive a wider publicity. The public oversight and dispersed ownership might also mean that a larger share of these …rms’value is in the form of “goodwill” and it might be that events which we study a¤ect such goodwill particularly strongly. This might imply, that the e¤ect of the events which we study, are systematically stronger for the listed …rms than is generally the case and that our conclusions should not be extrapolated to the …rms which are not listed on stock exchange.
9
101 or 102 (in one case of both). We also have dates of Court judgments for 99 appeals (39 annulments and 60 upheld observations), as well as exact dates of surprise inspections for 130 infringements.24 Table I presents some descriptive statistics on the selected sample (also broken down by periods), while Table C.I in the Appendix lists all the observations in our sample with the relevant main characteristics, such as the type of antitrust infringement, the dates of the relevant events and the amount of the …nes (they range from 0 to over one billion euro).25 Most of the observations, about half of our sample, are related to Commission Decisions taken in the years between 2003 and 2009 (118 observations and 43 cases). Among our observations we also identify 17 occurrences in which total immunity was granted to …rms. Also, we …nd that the average duration between the Raid and the Decision is fairly constant over time and is about three and a half years. "place Table I approximately here." Overall, the average …ne is 55.3 million euro, it is 4.6 million euro, in the sub-period 1979-1997 and 88.2 million euro in the sub-period 2003-2009, denoting a sharp increase in the amount of …nes. When we look at the average ratio between …nes and …rms’capitalization we …nd that on average this ratio is 0.27%.26 Also for the …ne capitalization ratio we …nd a sharp increase starting in the period 1998-2002 while in the …rst period (19791997) the …ne represented, on average, only the 0.06% in terms of a …rm’s capitalization. In the latest period, 2003-2009, the ratio decreases slightly relatively to the previous period, 1998-2002. To better illustrate the change in severity of the Commission, we also draw Figures 2 and 3, which show the evolution of average …nes and average …ne capitalization ratios over these thirty years. 24 Note that while we know all the Commission Decisions and their dates, surprise inspections do not always take place or sometimes their date is not made public by the Commission (we dropped several dawn raids observations because their dates were not revealed or were not made precise); also, …rms may decide not to appeal. This explains why we have a di¤erent number of observations for the three di¤erent antitrust events. 25 A noteworthy element of the Commission’s …ning policy is the possibility to grant, under its Leniency Programme, reductions in …nes to …rms which cooperate in cartel investigations. A zero …ne is due to the fact that the Commission can grant a 100% …ne reduction to the …rst …rm which reports information allowing the Commission to have su¢ cient evidence to convict …rms involved in a cartel. See Motta (2004) for a textbook analysis of leniency programmes. At date of writing the maximum …ne given by the Commission reached a record 1.06 billion euro. The …ne was given to Intel for violating antitrust rules in the computer chip market. 26 By looking at the …ne capitalisation ratio we can control both for …rm’s size and for the di¤erent value of money over time. Hence, it is a useful metric when comparing di¤erent types of …rms and …rms over time.
10
Figure 2 shows the evolution of the …nes (in absolute terms) over the period of interest. As suggested above, starting from 1998 there is a sharp increase in the amount of …nes. Figure 3 shows the evolution of the average …ne capitalization ratio and it also suggests a sharp change in the amount of …nes starting in 1998.
"place Figure 2 approximately here." "place Figure 3 approximately here."
We can also characterize our sample by looking at the sectors in which investigated …rms operate and at their country of origin. Figure 4 shows the distribution of the sample in the di¤erent six macro sectors we have identi…ed. Most of our observations are related to …rms operating in the Hydrocarbon and Chemicals sectors. About the geographical origin of …rms (we considered the country of listing), Figure 5 shows that the majority of the observations in the sample are related to German (50), Japanese (39), English and French (respectively 29 and 28) …rms, followed by US (19) and Dutch (17) …rms. The other 58 observations in the sample relate to …rms from other 22 countries (both European and extra-European).
"place Figure 4 approximately here." "place Figure 5 approximately here." The …rms in our sample are listed on di¤erent stock exchanges. In accordance with the country distribution seen above, the majority of …rms are listed in Frankfurt and Tokyo, followed by Paris, London and New York. The remaining stock exchanges where the …rms are listed are Amsterdam, Korea, Hong Kong, Singapore, Stockholm, Oslo, Brussels, Copenhagen, Milan, Luxembourg, Taiwan, Malaysia, Athens and Vienna.27
III.ii
Event Study Methodology and Estimation Procedure
The central concept in the event study methodology is the e¢ cient market hypothesis (EMH). Under this hypothesis, the price of the security re‡ects the value to investors of all the relevant available information about the fundamentals of the …rm. Moreover, under the EMH, any news about the fundamentals are immediately re‡ected in the share price. 27
In case of multiple listing we select the stock exchange with higher capitalization.
11
The question that the event study attempts to answer is: what is the value of a change of a particular fundamental? Under the EMH, if we knew the exact time at which the news became available to investors and the security price that would have prevailed in the absence of this news we could compute the value of the change of the fundamental that is re‡ected in the news, as the di¤erence between the counterfactual and the actual price. We use standard event study methodology to estimate the e¤ect of the three above mentioned events in the antitrust investigation on the value of the …rm. Our main references for the event study methodology are Campbell et al. (1997) and MacKinlay (1997).28 To obtain a counterfactual return we use a simple market model of returns:29 R i = i + i Rm + i ; (10) where Ri and Rm are the period- returns on security i and the leading index of the stock exchange where the security is listed, respectively. We compute the returns as ln Pit ln Pit 1 , where Pit is the price of the share on trading day t. "place Figure 6 approximately here." Figure 6 illustrates our approach. We de…ne = 0 as the event date, = T2 to = T3 form the event window and the periods from = T0 through = T1 form the estimation window. Let L1 = T1 T0 + 1 and L2 = T3 T2 + 1. We then run OLS on the equation 10 to estimate the parameters i and i for the …rm i using 101 trading days in the period T0 = 130 to T1 = 30.30 Then we use the estimated model to estimate counterfactual returns in the periods of interest to then construct abnormal returns in the event window as ^?i = Ri?
? ^ i + ^ i Rm
;
(11)
? are L 1 vectors of actual returns on the security i and where Ri? and Rm 2 of the leading index of the stock market where i is listed.
We aggregate individual daily abnormal returns by averaging them over securities (N ) and summing them over the days of the event window to obtain cumulative average abnormal returns (CAAR) for the event. 28
See also Brown and Warner (1980, 1985). A convenient assumption that we will make is that the (N 1) vector of asset returns, Rt , is independently multivariate normally distributed with mean and covariance matrix for all t. Under this assumption, given that the model is correctly speci…ed, the abnormal returns, conditionally on the market return, are jointly normally distributed. This result is the basis of our inference. 30 We have also performed robustness checks by modifying the length of the estimation windows, and checked that the results are not very sensitive to such variations. 29
12
CAAR =
T3 X
=T2
N 1 X ? ^i N i=1
!
(12)
Under the null hypothesis the event has no e¤ect on the mean returns and we use the test statistic below to draw inference about the cumulative abnormal return p sCAAR N p J= (13) ssCAR 1 + (N 1)r
Where the above J test is the adjusted BMP test developed by Kolari and Pynnönen (2010).31 This test is based on the established standardized test developed in Boehmer et al. (1991). It uses scaled cumulative average abnormal returns sCAAR, an unbiased estimator of the standard deviation of the scaled cumulative abnormal return ssCAR and also accounts for average sample cross-correlation (r; estimated in the estimation period). We adopt this statistical tests since Kolari and Pynnönen (2010) show that it accounts for cross-sectional correlation and is robust to serial correlation.32 We assess the robustness of our main approach adopting a set of alternative speci…cations and methods (see section IV.iii). First (see section (IV.iii.a)) we replicate the analysis substituting the market model with the mean model, where the mean return of the individual security is used as the counterfactual return. In this case the model is simply Ri = i + i . In principle, it is possible that a change in the share price of a very large …rm may cause a change in the relevant stock market index, giving rise to endogeneity problems. Using the mean model rather than the market model avoids this problem.33 We also develop a complete non-parametric event study (see section (IV.iii.b)) following the suggestion of Dombrow et al. (2000). Indeed, the OLS estimation adopted to estimate equation (10) relies on the strong assumption of normally distributed errors i . Dombrow et al. (2000) show that when the underlying distribution of the errors is uncertain, distribution free statistics (like non-parametric tests) are more robust and accordingly should be adopted. Given the concerns on normality of stock returns we do 31
The test is asymptotically distributed as a standard normal (asymptotics with respect to number of securities and lenght of estimation window), for a proof see the online Appendix B in Kolari and Pynnönen (2010). 32 It should also be noted that the simulation results of Kolari and Pynnönen (2010) suggests that the adjusted BMP test, in terms of power, is superior to the commonly used portfolio approach, to account for serial correlation. Also, Kolari and Pynnönen (2010) show that, being a scaled statistics, the adjusted BMP test is robust to serial correlation as the same order autocorrelation is present both at the numerator and denominator. 33 The statistical test used to evaluate the estimates of the mean model is the same J test seen for the market model.
13
implement this approach to check the robustness of our estimates. The nonparametric event study follows the estimation approach known as Theil’s non-parametric regression (…rstly suggested by Theil (1950) and reported in Dombrow et al. (2000), see Appendix A) and the statistical test we adopt is the multi-day rank test developed in Cowan (1992) and based on the rank test developed in Corrado (1989) (see Appendix A).
IV
Results
In this section, we …rst describe our main results, then we report the various robustness checks we have carried out. Table II reports daily average abnormal returns for the di¤erent events and it reports the cumulative average abnormal returns for four di¤erent aggregate event windows. The table, for each estimate, also reports the respective J test, where, consistently with the hypotheses tested (see section II), the tests are all one-sided. In what follows, we discuss the estimates for each event of interest.
"place Table II approximately here."
IV.i
A …rst look at the results
The average abnormal return on the day of the Raid is negative and highly statistically signi…cant, suggesting a :64% drop in the …rm’s share price the day the Dawn Raid is carried out. This implies a very quick relay of the news to investors.34 We also …nd statistically signi…cant e¤ects both before and after the raid. While the latter is expected (typically the markets will continue to adjust to the news in the following days), the former is more surprising, since Dawn Raids are widely considered truly unexpected events (and since we had found no mention of possible investigations in newspapers before the raids took place). Note, though, that the strongest e¤ect does take place on the day of the inspection. If we aggregate the abnormal returns, we …nd highly signi…cant (at 1% signi…cance level) negative cumulative returns for the Dawn Raid. For instance, by choosing the 31-day event window, the raid is estimated to lead, on average, to a 2:89% drop in the investigated …rm’s stock market valuation (signi…cant at 1%). 34 A large number of studies indicate that stock markets react very quickly to unexpected news. Brooks et al. (2003) investigate a sample of 21 fully unexpected negative news events - such as the Exxon-Valdez oil disaster, plant explosions, plane crashes, deaths of executives - and …nd that share prices fall by an average of 1.6% after a mere 15 minutes.
14
In the column for the Commission Decision we have negative abnormal returns for some particular days before and after the event. In the case of the Decision, though, …nding signi…cant estimates before the event is not entirely unexpected, since - as explained above - there are rumors and possible anticipations before the Decision. As for the cumulative average abnormal return, we …nd that for the 31-day window the Decision is estimated to reduce the returns by 3:57%, e¤ect which is statistically signi…cant at the level of 1%. The last two columns in Table II show the e¤ects of the Court judgments.35 We de…ne as “annulled ”all judgments which either annul the …ne or reduce it by more than 50%, and “upheld ” all remaining judgments. About Court judgments, although for some particular days we …nd some statistically signi…cant e¤ects, they do not all carry the same sign. Moreover, when aggregating over several days, we rarely …nd any statistically signi…cant e¤ect, for either annulments or upholding judgments. These are the base results. We now discuss the issue of market expectations and then discuss more thoroughly our estimates, dealing with each of the antitrust events in turn. (IV.i.a)
Market expectations
Our modelling approach, at each event, estimates the market reaction to the unexpected component of the antitrust action. For this reason, there might be concerns that this component does not fully represent the impact of the antitrust action. Nevertheless the following should be noted. The …rst antitrust event we consider, the Raid, is highly unexpected. Starting from this event, as the model makes it clear (see section II), the estimated market reactions already incorporate expectations about the possible outcomes of future events. For instance, the reaction at the Raid stage (the …rst event) already accounts for the possible future developments of the Commission Decision and of the Court judgement. Hence, by looking at all the events in turn we can control for this anticipation e¤ect. Moreover, we do capture the market reactions around the date of the event (due to speculations on expected actions) by adopting a relatively large event window. For instance, for the event Commission Decision, by examining past issues of the Financial Times we found that rumors on the potential infringement decision, and speculation on the magnitude of the …nes may occur but, if they do, they are typically concentrated in a period 35
We run the analyis only on the Judgements of the Court of First Instance. This because the ruling of this court are rarely overturned by the last court of appeal (i.e. the European Court of Justice).
15
of one month before the date of the Decision.36 For the above reasons the unexpected reaction to antitrust events that we assess in this study should be representative of the full impact of the antitrust action. Nevertheless, for some cases, we might still not capture some anticipated e¤ects, for instance if some novel information about some antitrust actions reach the market between the Raid and the Decision, or even before the Raid (and outside our event windows).37 If this is the case our estimates only o¤er a lower bound of the full impact of the antitrust action.38
IV.ii
Analysis and re…nement of the results
(IV.ii.a)
Dawn Raids
Antitrust experts and practitioners would agree that - apart from exceptional cases - "Dawn Raids" do come unexpectedly and surprise commentators and markets. It is di¢ cult to reconcile this fact with our …nding of negative (statistically signi…cant) returns before the Dawn Raid. In case of investigations already under way in other jurisdictions, it is possible that the market knows - or suspects - that the European Commission may also investigate similar infringements in the EU. It is also conceivable that news of an investigation being under way may occur when a …rm participating in a cartel has applied for leniency to the European Commission, and rumors about the application may have …ltered to the market. For these reasons, we have carried out separate estimates of the e¤ects of raids by excluding all cases which had already been investigated in the US (we did not …nd any prior investigations in other jurisdictions), as well as those where leniency applications were made. Table III shows that the estimated e¤ects of the raid do become stronger after excluding all such cases. Nevertheless, we still …nd a statistically signi…cant e¤ect also when using the long 31-day window which includes a period well ahead of the raid. Clearly, our results suggest that the market seems to anticipate that investigations may soon take place.39
"place Table III approximately here." 36
Our main event window, from 20 days before the event to 10 days after, covers 20 working days before the event date to account for possible rumors and speculations taking place one month before the event. 37 As we do not …nd systematic evidence of such cases we conjecture that for most of our cases novel information is only released around the date of the event. 38 The "full" e¤ect of the antitrust action would indeed be higher than what we estimate. 39 We also checked for signi…cant average abnormal returns between 30 and 21 days before the Raid but we did not …nd any. Hence, if there is anticipation this seems to be limited to one month before the Raid.
16
These results show that the surprise inspection has a strong negative e¤ect on the investigated …rm’s valuation, although the precise magnitude of the e¤ect is sensitive to the length of the event window used. The choice of a particular event window is always to some extent arbitrary, but in what follows we shall favour the 31-day window to fully account for market reactions taking place before the event day and to account for the adjustments that follow the event date. We can then state the following result. Prediction 1 seems consistent with the data: If the event "Commission undertakes a Dawn Raid" is observed, then the share price of the raided …rm decreases on average by 2.89%. (IV.ii.b)
Commission Decisions
We have veri…ed the impact of the infringement Decision of the Commission by considering di¤erent event windows. The results reported in Table II indicate that several days ahead of the Decision the market anticipates the event as we register signi…cant negative daily abnormal returns. This is not surprising, since it is well known that there are informational leakages occurring prior to the date of the Decision. Accordingly, our favorite estimate remains the one obtained with the 31-day event window. The cumulative average abnormal return for this event window is -3.57% and is signi…cant at 1%. Prediction 2 seems consistent with the data: If the event "Commission issues an infringement Decision" is observed, then the share price of the …rm decreases on average by 3.57%. (IV.ii.c)
Court judgments
As seen in Table II above, neither judgments annulling the Commission Decision, nor those upholding them, are statistically signi…cant for almost all event windows (although the sign are somewhat as expected, especially when considering the shorter event windows).40 The lack of a signi…cant e¤ect may be attributed to di¤erent reasons. One possibility is that through the hearings and the reports of the witnesses, the public may form a correct expectation of the judge’s view, so that at the time the judgment comes the market has already discounted it. Another possibility is that - as discussed in Section V below - the …ne itself is only part of the loss that a …rm incurs because of the antitrust investigation. For instance, suppose the judgment annuls a cartel decision for procedural reasons: the …rms might have won the case, but it is unlikely that they could continue colluding: despite the 40
This might be the case because shorter event windows are less unbalanced toward the days ahead of the event. Especially for Annulments it seems that around and after the event realizations the "positive" daily abnormal returns outnumber the "negative" ones.
17
annulment, they will still bear the loss in market value due to the ceasing of a lucrative (anticompetitive) activity. We therefore state the following: Prediction 3 does not …nd support from the data: Neither the event "Court annuls the Commission Decision" nor the event "Court upholds the Commission Decision" have any statistically signi…cant e¤ ect on the …rm’s share price. (IV.ii.d)
Variability across …rms
To give a better idea of the variability of the e¤ects across …rms, Figures 7 and 8 show the estimated cumulative abnormal returns for each …rm in the sample (notice that there are several points corresponding to the same date whenever there is a cartel case). Visual inspection shows that overall the Raids and the Decisions negatively a¤ect the share prices, but there is a lot of variability across …rms (and positive abnormal returns are not rare).41 Moreover, it seems that, over time, the support of the distribution of …rms’ cumulative abnormal returns gets larger as we observe increased variability and larger ranges of CARs. The variability across …rms might be explained by many factors like the sectors in which …rms operate, the type of infringement, the size of …rms, the level of …nes.42 "place Figure 7 approximately here." "place Figure 8 approximately here." If we consider the cross-section of CARs we can also look at the quantiles of the empirical distribution to inspect the cross-section variability in the response to the same event. For both events Raid and Decision the median CAR is slightly lower (in absolute terms) than the average CAR. By looking at quantiles we can see that more than 50% of the observation exhibit a 41
This paper studies the changes in …rms’stock prices around common events of interest. The focus is indeed on the average overall e¤ect rather than on the single …rm’s outcome. Individual realizations might well be biased by unrelated events, which occurrence explains individual positive CARs. On the contrary, it is the mean that consistently represents the impact of the event of interest, as individual unrelated phenomena, occurring around the date of the event, should on average cancel out. 42 We are aware that our sample of listed …rms is heterogeneous under di¤erent dimensions, one of which is the size of …rms. Some …rms are large multiproduct and multinational …rms. For this type of …rms, the e¤ect of antitrust interventions related to one particular product and geographic market may well be smaller than for a smaller, single-product …rm operating only in a single domestic market. However, it should be recalled that the Commission can impose …nes up to 10% of the total (world) turnover of a …rm, so that for any given violation a larger multiproduct and multinational …rm might generally be given a larger …ne.
18
negative CAR around the date of the event and the support of the negative part of the distribution is longer than the support of the positive part. These elements support the …nding that on average there is a negative and sizeable reaction of the market to these events. "place Table IV approximately here."
(IV.ii.e)
E¤ects by type of infringement
The overall e¤ects of Raids and Decisions may hide di¤erences across types of infringement. In Table V we show the separate estimates of CAARs by infringement. The data show a very strong e¤ect of both the Raid and the Commission Decisions in abuse of dominance cases,43 even though we should stress that there are only few observations in this sub-sample. The e¤ects of raids and decisions are also strong in cartel cases (not surprisingly, since they account for the vast majority of our observations), whereas the estimates for the remaining cases (anticompetitive agreements other than cartels) are not statistically signi…cant, perhaps re‡ecting the fact that there are very di¤erent types of infringements in this category. "place Table V approximately here."
(IV.ii.f )
E¤ects by sector
The impact of the antitrust events seems also to vary depending on the sector in which investigated …rms operate. Table VI presents the estimated e¤ects for the four sectors in which we …nd statistically signi…cant e¤ects. Indeed, for the two macro sectors Consumer goods and Food/Agriculture/Paper we do not …nd statistically signi…cant e¤ects neither for Raid nor for Decision. The sector that seems to exhibit the higher impact is Machinery and Vehicles followed by Services and Transport, and Construction and Metal. The …rms that operate in the sector Hydrocarbons and Chemicals seem not to be a¤ected by the Raids while at the Decision event they exhibit a drop of around 2.13% in their market value (signi…cant at 5%). "place Table VI approximately here." 43
We do not report estimates for Court judgments because they are not signi…cant for nearly all event windows and types of infringement. Nevertheless, we found a negative (-2.45%) and statistically signi…cant e¤ect (at 10%) only for judgments upholding abuse cases (7 observations) in the 11-day event window (-5,+5).
19
(IV.ii.g)
Changes over time
Our sample spans over more than 30 years, during which antitrust policy has consolidated and changed. In particular, all parties involved have gained considerable experience in dealing with competition law issues, and markets have become increasingly aware of the role played by antitrust law and the risks faced by …rms. At the same time, also the European Commission’s antitrust policy has evolved. This has been a continuous process, but one might conceivably identify in the 1998 "Guidelines on the imposition of the …nes" a landmark in European competition policy, giving rise to a more transparent (less arbitrary) imposition of …nes as well as con…rming an increased severity in handing them out. Accordingly, we have …rst split our data in two sub-samples, the …rst for events which took place between 1979 and 1997, and the second for events from 1998 onwards. The estimated CAARs for these sub-samples, reported in Table VII, indicate that indeed there is a stronger e¤ect of the Commission Decision in the period 1998-2009, but that is not true for the Raid. For the Commission Decision we do …nd that after 1998 …rms’s valuation is strongly and signi…cantly a¤ected by the event, while before 1998 we do not …nd any signi…cant e¤ect. For the Raid, we do …nd a drop in the …rm’s market value both before and after 1998, but the estimated CAAR is higher in the …rst period.44 (We do not report the breakdown of First Court judgments over time because estimates are not statistically signi…cant, in any period considered.) "place Table VII approximately here." In light of the above results we can state the following: Prediction 4 …nds support from the data only for what concerns Commission Decision: The increase in both q and F which took place in about 1998 leads to the expected increase in the e¤ ects on …rms’ market value of Commission Decisions.45 To gain some further insights on the evolution of the e¤ects of the investigations over time, we have proceeded to a further breakdown of the sample 44
If we exclude from the sample the …rms which have received leniency immunity, and which therefore receive zero …ne, the estimated e¤ect of both raids and Decision increase in the period 1998-2009, but the qualitative results would not change: the e¤ect of the raid would still be stronger in the period 1979-1997. Instead, we have repeated the exercise with the mean model, and we found that the estimated CAARs are higher in the period 1998-2009 for both the raids and the Decision, as predicted by the model. 45 Notice that there is no e¤ect of q and F on judgments, but as we know neither annullments nor upholds seem to have any e¤ect on share prices in any period.
20
into sub-periods. First of all, since commentators (see for instance Geradin and Henry, 2005) argue that the Commission started to apply increasing severity in the …nes in the early-mid 90’s, we have carried out separate estimates for the two sub-periods 1979-1993 and 1994-1997. Next, we know that other notable changes in EU antitrust policy took place in the early 2000’s. Most importantly, in February 2002, the Commission drastically changed its leniency policy, …rst introduced in 1996, but which had not been successful. The 2002 review o¤ered full transparency and certainty of immunity from …nes to the …rm …rst reporting a cartel. Further, it also introduced immunity whenever providing su¢ cient evidence for the Commission to establish an infringement, even when an investigation had already started. These changes turned out to be extremely e¤ective and have led scores of …rms to apply for leniency immunity (obliging the Commission to introduce very recently a settlement procedure to deal with all the cartels unveiled by leniency applications). A change in EU enforcement also took place in 2003, when the O¢ ce of the Chief Economist at the DG-Competition was created, in response of a series of CFI judgments which quashed Commission Decisions in merger cases, and which heavily criticized the economic reasoning of the Commission. The Chief Economist Team has in‡uenced not only merger policy but also antitrust policy in general, leading among other things to a more cautious approach in abuse cases. Accordingly, we have also broken the sub-sample 1998-2009 further, and we have carried out separate estimates for the periods 1998-2002 and 20032009.46
"place Table VIII approximately here."
Table VIII reports the estimates for the above mentioned sub-periods. The results show that the e¤ect of the antitrust events on share prices is much stronger in the periods immediately before and immediately after 1998. In particular the higher e¤ect for Raid is found in the period 1994-1997 (7.41%) while for Commission Decision it is found in the period 1998-2002 (-5.92%). Moreover, it shows that the early years of antitrust enforcement are characterized by weak e¤ects on the market valuation of …rms. This can be attributed to the relative indulgence of the Commission, at the time, and the expectation that Courts may annul the Commission Decisions. On the contrary, it is more di¢ cult to explain the lower e¤ects found after 2002, for Raid. As for the 55 Raid observations relative to the period 2003-2009, we do not …nd a signi…cant e¤ect on …rms’market value. Nevertheless, for 46
We have also performed a robustness check by breaking the sample in 2003 and 2004, rather than in 2002, but the results are qualitatively very similar.
21
Commission Decisions we do …nd a signi…cant e¤ect with the expected sign, although lower in magnitude than the one found in the period 1998-2002. The following argument might help explain what we …nd. On the one hand, a more generous leniency programme might possibly lead to lower expected …nes if the market expects …rms may bene…t from …ne reductions. At the same time, the market might discount less …rms’market value at the Raid stage as it waits to see the outcome of the Commission (and learn which …rms will bene…t from leniency reductions, and to what extent).47 On the other hand, the availability of evidence due to whistleblowers might make it easier to prove infringements, also on appeal, leading to higher probability of upholding the Commission Decisions in Courts. It is also possible that after 2002 the market assigns a higher probability that the Commission will investigate the …rms (one of the e¤ects of leniency programmes is that they make it more likely that cartels break down because one of its members reports to the antitrust agency),48 leading to a lower valuation of the …rm even before any Raid takes place. In conclusion it is di¢ cult to explain the di¤erences, over periods, in the estimated e¤ects as many factors, that could in‡uence market reaction, change over time. Indeed, the aggregate estimates for each smaller subsamples are also in‡uenced by the particular selection of …rms and cases (sectors involved, type of infringements, number of leniency immunity granted), hence without controlling for these factors it is di¢ cult to sensibly compare di¤erent periods. Nevertheless, the conclusion that after 1998 Commission Decisions have a higher impact on …rms’market value seems to be robust. (IV.ii.h)
Market predictions
The fact that Court judgments are never statistically signi…cant make it di¢ cult to evaluate the ultimate e¤ect of antitrust investigations. To gain some insight on the role of judgments, though, we have split the sample into two subsets: that of …rms whose …nes have been eventually upheld, and that of …rms whose …nes have been eventually annulled. Table IX shows that in the former case the e¤ect of Raids and Decisions is much stronger (both statistically and economically), whereas in the latter the e¤ect of Raids is not statistically signi…cant, while that of Decisions is signi…cant at the 5% level but is less strong. "place Table IX approximately here." 47
We replicate this analysis excluding …rms that bene…ted from leniency immunity but the main conclusions remain the same. 48 In terms of our model, the probability m of an investigation would have increased after 2002.
22
It is di¢ cult to interpret this result, but it seems to suggest that the market tends to have some reasonably good expectations about what is likely to be the …nal outcome of the antitrust proceedings. The …nding that the market at the Raid and at the Commission Decision stages predicts reasonably well the probable overall outcome of the investigation is therefore consistent with no signi…cant e¤ect of Court judgments.
IV.iii
Robustness checks
In this section, we carry out some robustness checks of the estimates obtained. In particular, we shall estimate the mean model, the non-parametric model and perform some placebo analysis. (IV.iii.a)
The mean model
The fact that the …rms in our sample are often large companies that enter in the composition of stock market indices, which in turn appear as independent variables in the model of counterfactual returns, may be a source of endogeneity bias in the estimates. As a check of robustness of our estimates, we then estimate the counterfactual model using the mean-model, described at the end of Section III above. Table X reports the cumulative abnormal return estimates, for the 31-day event windows, also used in Table II. The results of the mean model go in the same direction, and have similar magnitude, as the market model. Hence, these results seems to suggest that endogeneity does not a¤ect our main estimates based on the market model. Given that the market model should have greater explanatory power and produce better counterfactual scenarios, in the discussion of the results we favour the market model estimates shown above.49
"place Table X approximately here." (IV.iii.b)
Non-parametric tests
Although the OLS estimation of the market model is the traditional choice in the majority of event studies Dombrow et al. (2000) show that when the normality condition (in the stock return distribution) fails to hold, non-linear estimators may be preferred. Indeed, under non-normality, OLS is only the 49
Apart from estimating the mean model, we have also estimated a restricted sample that excludes those observations which stock market indices contains few …rms (indeed, it is possible that the …rm which receives the …nes has a considerable weight on such an index). The estimated e¤ects appear to be stronger in this sub-sample. For instance, when excluding indices containing thirty …rms or less, and focusing on the 31-day window, the estimated e¤ect of the raid is -4,4% (signi…cant at 1%) and that of the Decision is -4.73% (signi…cant at 1%), whereas for the whole sample (see Table II) was -2.89% and -3.57% respectively for the raid and the decision.
23
best linear unbiased estimator. Therefore, as a robustness check we also perform a complete non-parametric event study implementing Theil’s nonparametric estimator in combination with a non-parametric test statistic (the rank test). Table XI presents the estimates for the non-parametric event studies at the main event window of interest (31-day). The nonparametric estimates con…rm all the main results of the OLS event study for all of the events of interest. In the case of the Raid and the Decision, nonparametric estimates are highly signi…cant (at 1%) with estimates having a comparable magnitude to the OLS estimates, whereas the Judgments are still statistically insigni…cant. Again this evidence support the robustness of our main estimates discussed above. "place Table XI approximately here."
(IV.iii.c)
Placebo analysis
As a last robustness check we also perform a series of placebo tests running event studies around dates on which no relevant news about the antitrust investigation should reach the market. This should check that our estimates are not biased by other possible confounding e¤ects. First of all, for the event Raid, we run a series of three placebo tests: 50 days, 75 days and 100 days before the date of the Raid (see Table XII). For the 31-day event window ( 20; +10), the placebo test "50 days before the Raid" estimates a CAAR of -1.09% not statistically signi…cant. The placebo test "75 days before the Raid" estimates a CAAR of 0.71% again not statistically signi…cant. Finally, the placebo test "100 days before the Raid" …nds a positive CAAR of 2.85% statistically signi…cant at 1%. Nevertheless, this latter estimate is not robust over di¤erent event windows. "place Table XII approximately here." For the Commission Decision we ran two placebo tests (see table XIII), one 100 days before the Decision and one at the date of the Statement of Objection (SOB). Focusing on the 31-day event window, for the …rst placebo test we …nd a CAAR of -0.91% but not statistically signi…cant. For the event SOB we …nd a CAAR of -0.07% again not statistically signi…cant. The …ndings at the SOB stage are especially interesting. Usually when the European Commission issues a SOB an anonymous press statement is also released. It seems indeed that at this stage the market has no means to infer useful news about the investigation process as we do not …nd any market reaction.
24
"place Table XIII approximately here." Overall, the placebo tests are reassuring in that when we apply the same methodology used above on some …ctitious date, as expected, we do not …nd statistically signi…cant and robust e¤ects, whereas at the dates of the Raid and of the Decision we do …nd e¤ects with the expected sign, and which are both economically and statistically signi…cant.
V
Economic interpretation of the results
In this section, we discuss the results obtained, and carry out some additional analysis with the aim of better interpreting them. Section V.i discusses the economic signi…cance of the estimated e¤ects by …rst comparing them with estimated e¤ects of similar events, and then by calculating the overall e¤ect of antitrust investigations, as opposed to considering the single events ’Raid’ and ’Infringement Decision’in isolation. Section V.ii is devoted to the role played by the …nes in explaining the overall loss in market value.
V.i
Economic signi…cance of the estimated e¤ects
To see whether our estimated e¤ects on …rms’ share prices should be considered large or small, it is useful to compare our results with the estimates of similar events. We …rst compare our results with the e¤ects of antitrust events in two other jurisdictions (one published study for US data and our own original investigation on Italian data), and then with other events which share similar features. (V.i.a)
Comparison with other jurisdictions
We are familiar with only one exercise which aimed at estimating the effects of antitrust investigation. Bosch and Eckard (1991) used US data and estimated an average drop in the …rm’s value of 1.08% following the indictment (the closest event to the Dawn Raid). However, the US antitrust system is very di¤erent from the European one, in several respects: company …nes tend to be much lower, there are harsh criminal sentences (upon which probably falls most of the weight of cartel deterrence), and private damages play a vastly bigger role than in the EU. For this reason, we have carried out a similar event study analysis also on the investigations carried out by the Italian Competition Authority (ICA), where Italian antitrust law and enforcement is borrowed from the EU. Also for the Italian case we can distinguish three di¤erent steps: the start of the investigation; the Decision (either an infringement decision or an acquittal decision); and the judgment of the Court of Appeals (that can uphold or 25
annul ICA’s decisions). The results of this event study (the data cover the period 1991-2007, for 154 observations) show that the e¤ect of the ICA’s interventions on the …rms’market value is either not statistically signi…cant or much weaker than what found for the EC investigations.50 (V.i.b)
Comparison with similar events
It may also be useful to compare our results with those obtained in works estimating the e¤ects of events with characteristics similar to those of antitrust events. Gunthorpe (1997) uses event study techniques to investigate the e¤ect of the …rst announcement in the Wall Street Journal that a …rm is involved in some form of illegal behavior, such as racketeering, patent infringements, or fraud (for instance, misleading advertising and securities fraud). She …nds that on the very same day of the announcement, the average abnormal return is -1.32%, and that the cumulative average abnormal return is -2.3%. The magnitude of these e¤ects is similar to that of the Dawn Raids, which are also unexpected events. Since Commission Decisions are not entirely unexpected events, we need to …nd events sharing this feature for the sake of making comparisons. MacKinlay (1997) analyses the e¤ects on share prices of announcements that actual earnings are more than 2.5% less than expected. On the same day as this announcement is publicly made, the …rm’s share drops by -.68%, while the cumulative average abnormal return on the 41-day event window (comparable to the length of the long event window we used for the Decision) is of about -1.26%. The estimated e¤ects of such relatively minor ’bad news’are therefore of a slightly lower order of magnitude than the estimated e¤ects of the news that the European Commission has decided to …ne a …rm for an antitrust infringement. (V.i.c)
Total loss in market value
So far, we have considered the e¤ects of the events in isolation. However, it is natural to ask what is the overall impact of the antitrust investigation, as determined by the e¤ects of both the Raid and the Decision. As a …rst approximation, one could simply take the sum of the estimated e¤ects of these events, namely 2:89% for the Raid and 3:57% for the infringement 50
At the start of the investigation the market does not seem to react as we do not detect any signi…cant abnormal returns. Di¤erently, following an infringement decision the market value of investigated …rms drops by around 0.48% and 0.72%, but this holds only for relatively short event windows (only the 7 days and 3 days event windows have statistically signi…cant estimates). Following an acquittals decisions we …nd that on average there is a positive reaction of the market (the CAARs have a positive sign) but the estimates are not statistically signi…cant. After an Upheld judgment the market value of …rms drops by around 2.82%, but this estimate is statistically signi…cant only when we consider the largest event window. Finally, an annullment of the ICA’s decision does not signi…cantly a¤ect the market value of …rms. See Appendix B.
26
Decision, and …nd that the overall e¤ect of antitrust action would be a 6:46% drop in share prices. This a rather sizeable e¤ect on the investigated …rms, especially if one considers that some of the …rms in our sample are huge conglomerates which have been investigated in markets which represent a very small subset of their business operations. This simple calculation has two shortcomings. First of all, summing the two estimates leads to an overestimation of the total e¤ects of the antitrust investigation. To see why, de…ne the latter as T OT
=
VDecision VV iolation VV iolation
(14)
with VV iolation being the value of the …rm before any antitrust investigation takes place and VDecision its value after the …rm has experienced both a Raid and an infringement Decision. The total e¤ect can be rewritten as: T OT
=
VDecision VRaid VV iolation
+
VRaid VV iolation VV iolation
(15)
where VRaid is the …rm’s value after the Raid. VRaid VRaid Since VDecision < VDecision , it will also be: VV iolation VRaid
T OT
= =
VDecision VV iolation VV iolation Decision
+
<
VDecision VRaid VRaid
+
VRaid VV iolation VV iolation
= (16)
Raid
It follows that the "true" T OT will be lower than b Decision + b Raid = 6:46%. Second, we have so far followed the standard methodology used in event studies, and we have not weighted the estimates of the events for the …rms’ capitalizations. This unweighted average estimate might be very informative if we were to predict the impact of the events for a generic …rm independently of its capitalization, but - by treating the estimated e¤ect of a very small …rm on par with the estimated e¤ect of a very large one - might be biased if we were to understand the overall impact of the antitrust events on the economy as a whole. In what follows, we calculate the overall e¤ects of antitrust actions by weighting for …rms’capitalization. Weighted average of the e¤ects of antitrust events In the sample, we have 130 observations for which we have estimated e¤ects of both the Raid and the Decision and the remaining 110 (out of 240) for which we have only the estimated e¤ects of the Decision.51 By de…ning as CAARidec and CAARiraid the estimated e¤ects of the Decision and the raid respectively (note that for some …rms the latter may not exist), and as CAPi …rm i’s 51
Strictly speaking, we should talk about …rms/observations, as some …rms are repeat o¤enders.
27
capitalization, we can compute the estimated loss in market value for the Decision as 240 X LDecision = (CAARidec CAPi ) (17) i=1
and for the Raid as
LRaid =
130 X
(CAARiraid CAPi )
(18)
j=1
By de…ning total capitalization of a subset of S …rms (with S CAPS =
S X
CAPs
240) as (19)
s=1
we compute the weighted estimate of the market losses/capitalization ratio due to the Decision as
and due to the Raid as
b Dec = LDecision CAP240 b raid = LRaid CAP130
(20)
(21)
By applying this methodology, we obtain that b Dec = 2:12% and that b raid = 1:61%. Note that these estimates are lower than the unweighted average e¤ects of the events that we have reported so far, suggesting that the bigger the …rm (size being proxied by capitalization) the lower the e¤ect of antitrust events, other things being equal. The next step is to apply this principle to compute the total e¤ect of the antitrust events. To do so, we have two possibilities. The …rst one is to consider the whole set of 240 …rms but treat the …rms for which we have no observation about the Raid as if they had an estimated CAAR for the Raid equal to zero. In this way, we approximate the total loss as LT OT;240 =
240 X
(CAARidec
CAPi ) +
i=1
130 X
(CAARiraid CAPi )
(22)
j=1
Accordingly, the estimated total e¤ect of antitrust event would be given by b T OT;240 = LT OT;240 CAP240 28
(23)
By applying this method we obtain that b T OT;240 = 3:03%. Obviously, this is an underestimation (in absolute value) of the total e¤ects, since we expect that, were we able to observe it, on average the estimated CAAR of the Raid would be negative. We can then turn to a second methodology, which is to look only at the subset of 130 …rms for which we have estimations of the e¤ects of both the Raid and the Decision. In this case, the estimated total loss would be LT OT;130 =
130 X
(CAARidec CAPi )
i=1
+
130 X
(CAARiraid CAPi )
(24)
i=1
The estimated total e¤ect of antitrust event would then be given by b T OT;130 = LT OT;130 CAP130
In this way, we would obtain that b T OT;130 =
(25) 4:55%.
We estimate therefore that the total e¤ects of antitrust action upon the …rm’s market valuation are between 3:03% and 4:55%.
V.ii
The role of …nes in explaining the loss in market value
In the US, Bosch and Eckard (1991) estimate that …nes and damages account for only 13% of the total loss of stock market value caused by the …rm’s antitrust indictment. Perhaps the main reason why an antitrust investigation may create a loss in the …rm’s value which goes well beyond the …ne is that the …rm will likely have to put an end to a pro…table activity (be it a cartel, an abusive practice, or any other business practice considered illegal by the antitrust agencies and the courts).52;53 Other possible sources of loss in value, in addition to the direct e¤ect of the …nes, could be: (i) legal and consulting fees for antitrust proceedings; (ii) the …rm may have to give up pro…table projects either because the management is distracted by the antitrust investigations, and/or because, in case of large …nes, the …rm will have lower retained earnings and cash (in imperfect …nancial markets, lower assets will limit the …rm’s ability to obtain credit); and (iii) the …rm may be hurt by the negative publicity following an antitrust investigation. In order to calculate the weight of the …nes in the total loss of valuation in our sample, compute the total value of the …nes for a subset of S …rms (with S 240) as 52
In our model, this e¤ect would be given by the di¤erence between payo¤s V M and
C
V . 53
Furthermore, in some cases, the …rm may also have to comply with (structural or behavioural) remedies which could lower its pro…ts even more.
29
F IN ET OT;S =
S X
F IN Es
(26)
s=1
The overall loss in capitalization due to the …nes would therefore be given by b S = F IN ET OT;S CAPS
(27)
Following what we have done above, we can then use two methods to estimate the loss in capitalization due to the …nes. First, if we consider the whole sample we would have that the loss in capitalization due to the …ne is b 240 = 0:27%, and that the …nes explain only
b 240 b T OT;240
= 8:90% of the overall loss in market value due to antitrust
action. Second, if we consider the sub-sample of …rms for which we observe both Raid and Decision, we obtain that total loss due to …nes is b 130 = 0:28%. b This implies that the …nes explain only b 130 = 6:25% of the overall loss T OT;130
in market value due to antitrust action. We can then conclude that the …nes explain between 6:25% and 8:90% of the loss in market value due to the antitrust events. Also, the fact that Court judgments are found not to be statistically signi…cant may be interpreted as indication that whether the …rms have or not to eventually pay the …nes may matter little compared with the fact that it is unlikely that they can continue to engage in anticompetitive conduct. (V.ii.a)
Regressing abnormal returns over the …nes
To determine whether the magnitude of a negative market reaction at the time of the Raid and Commission Decision depends on the relative magnitude of the …ne imposed on the …rm by the Commission, we regress the abnormal returns on a constant, the ratio of the …ne over the total capitalization of the …rm, and a number of controls.54 The results are summarized in Tables B.III and B.IV (Appendix B) respectively for Raid and Commission Decision. We …nd that the coe¢ cient on the relative size of the …ne is not statistically signi…cant for the Raid, but it is highly signi…cant for the 54 The control variables included a trend, and dummies for the period from 1998 onwards, for the type of infringement (article 102, article 101 non-cartel), for the sector of activity (except hydrocarbon and chemicals), and for the nationality of the …rm (except for EU). We also included the square of the …ne capitalization ratio to control for non linear e¤ects but the associated coe¢ cient was not statistically signi…cant. We also carried the same regressions with the absolute level of the …ne (rather than the …ne/capitalisation ratio) as explanatory variable, and the estimated coe¢ cient was not signi…cant. Details available from the authors upon request.
30
Commission Decision.55 Perhaps this is not entirely surprising, since at the time the Raid is carried out, the market still does not know how large the …ne is going to be. This may be seen as an indication that …nes (normalized by capitalization) do have some impact on the loss of market value due to the antitrust investigation, although they are far from explaining all of it. (V.ii.b)
E¤ects of the investigation on …rms which have received immunity
To investigate further the role played by the …nes, we see whether …rms which have been the object of an infringement Decision but have received no …nes experience systematically di¤erent e¤ects than …rms which did receive a …ne. To do so, we focus on cartels and we split the sample into …rms which have received immunity from …nes within the leniency programme and all other …rms.56 All these …rms - those which have won leniency and their rivals - have been found guilty of collusion, and accordingly we can conjecture that the cessation of illegal pro…ts will occur for all of them, but only the latter will have to pay the …ne. In the extreme case, if the level of the …nes did not matter at all, we should expect a similar impact of investigations independently of the value of the …nes. Table XIV though, shows this is not the case: whereas both the Raid and the Decision signi…cantly decrease the share prices of the …rms which do receive the …nes (column "Cartels excl. Leniency"), they do not a¤ect in any statistically signi…cant way those of …rms which have received immunity within the leniency programme.
"place Table XIV approximately here."
The di¤erent impact on these two groups of …rms may be explained in di¤erent ways. First, …nes may a¤ect the relative ability of …rms to …nance investments and projects. Suppose for instance that …rms were …nancially constrained. If some …rms in the industry had to pay the …nes whereas one of their rivals did not have to, the former may have lower assets and therefore lower access to credits than the latter, which would therefore bene…t from this. Second, it may be not the level of the …nes themselves but rather the fact that one …rm is guaranteed immunity that explains this di¤erential 55
Note that an estimated coe¢ cient of :56% implies that an increase by :1% in the …ne/capitalisation ratio would increase the average estimate of the CAAR by :056%, that is, from 3:57% to 3:63%. 56 We also carried out another estimation by separating the whole sample into …rms which - whatever their infringement - received a very small …ne (that we de…ned as less than :01% of their capitalisation ratio, and compare it with all other …rms. The results are very similar to those reported here for the case of immunity …rms only.
31
impact: while other …rms and their managers have to devote resources and assets to the antitrust investigation, the …rm with immunity may forget about legal issues, and focus on its business. Third, it is in principle possible that there is a reputational e¤ect in the market, with the …rm having ’spilled the beans’receiving a favorable treatment by consumers, who want instead to punish the others. In other words, there may be two e¤ects at play when a cartel investigation is based on evidence gathered within a leniency programme. First, the inability to continue collusion a¤ects negatively all the …rms in the industry. Second, there is a "redistributional" e¤ect whereby (because of credit constraints e¤ects, deviation of managerial resources to less productive activities, or loss of reputation) some market shares move from the …rms which receive the …nes to the …rm which has received immunity. To understand more in depth these di¤erential e¤ects, though, we should analyze more carefully the interaction between …rms in each cartel case, and …nd additional information on them, something which is beyond the scope of the present paper.
VI
Summary and conclusions
We have modelled the e¤ects of successive events in an antitrust investigation on a …rm’s stock market value and we have estimated, by using event study techniques, these e¤ects by making use of an original database on EU antitrust law proceedings. Our main result is that the Dawn Raid (i.e., the surprise inspection of the …rm’s premises carried out by the Commission), which is the …rst piece of information received by market operators indicating that the European Commission intends to investigate an antitrust infringement, has a strong and statistically signi…cant e¤ect on the …rm’s share price: on average, on the same day as the Dawn Raid the …rm’s return is around :64% lower than the counterfactual return provided by the market model; the cumulative average abnormal return due to the Dawn Raid is estimated to be 2:89%. We also …nd that the Commission’s infringement Decision results in a strong and statistically signi…cant cumulative abnormal return of about 3:57%. Instead, we do not …nd evidence of statistically signi…cant e¤ects of the Court’s judgments. We also estimate the weighted (by capitalization) total e¤ects of antitrust action (that is, the e¤ect of both the Raid and the infringement Decision) to be between 3:03% and 4:55% of market value, a rather sizeable effect when considering that some of the …rms in our sample are very large multinational and multiproduct …rms. Finally, we show that only a fraction (at most 8:9%) of this loss in capitalization is due to the …nes. Our conjecture is that most of the loss is due to the cessation of anticompetitive pro…ts, but other factors - for instance
32
the management having to focus on legal rather than on operational issues, legal expenses, or a loss of reputation - may also play a role. In a recent book, Whinston (2008) expresses doubts on the e¤ectiveness of antitrust intervention, referring to some empirical work which suggests that anti-cartel activities may have not led to a price decrease in the markets at hand. We regard our paper as o¤ering instead some (admittedly indirect) evidence on the e¤ectiveness of antitrust intervention. In our sample, composed predominantly by cartels, we feel that most of the drop in the share prices is likely due to the cessation of pro…table cartel activity. In turn, this should imply that investors expect investigated and …ned …rms not to be able to sustain such high prices as in the past. Therefore, although we cannot o¤er direct evidence on this issue, our paper indirectly suggests that antitrust intervention may indeed have an e¤ect on market prices.57
57 In case of abuse of dominance cases, as well as (non-cartel) horizontal or vertical agreements, the e¤ects on prices would depend on whether the antitrust decision has duly taken into account e¢ ciency justi…cations and overall e¤ects on consumer welfare. To the extent that an abuse, say, would have led to higher pro…ts of the violator and higher consumer prices, the antitrust actions would decrease both of them.
33
References Bizjak, J. M. and J. L. Coles (1995). The E¤ect of Private Antitrust Litigation on the Stock-Market Valuation of the Firm. American Economic Review 85 (3), 436–461. Boehmer, E., J. Masumeci, and A. Poulsen (1991). Event study methodology under conditions of event induced variance. Journal of Financial ½ Economics 30 (2), 253U272. Bosch, J.-C. and W. Eckard, Jr. (1991, May). The Pro…tability of Price Fixing: Evidence From Stock Market Reaction to Federal Indictments. The review of Economics and Statistics 73 (2), 309–317. Brooks, R. M., A. Patel, and T. Su (2003). How The Equity Market Responds to Unanticipated Events. Journal of Business 76. Brown, S. J. and J. B. Warner (1980). Measuring Security Price Performance. Journal of Financial Economics 8, 205–258. Brown, S. J. and J. B. Warner (1985). Using Daily Stock Returns: The Case of Event Studies. Journal of Financial Economics 14, 3–31. Campbell, J. Y., A. W. Lo, and C. A. MacKinlay (1997). The Econometrics of Financial Markets. Princeton University Press. Connor, J. M. and Y. Bolotova (2006, November). Cartel overcharges: Survey and meta-analysis. International Journal of Industrial Organization 24 (6), 1109–1137. Corrado, C. J. (1989, August). A nonparametric test for abnormal securityprice performance in event studies. Journal of Financial Economics 23 (2), 385–395. Cowan, A. R. (1992). Nonparametric event study tests. Review of Quantitative Finance and Accounting 2, 343–358. 10.1007/BF00939016. Detre, J. and A. Golub (2004, February). A reexamination of the profitability of price …xing using stock price movement: Has new antitrust legislation been a more e¤ective deterrent of price …xing? Sta¤ Paper 04-03 - Department of Agricultural Economics - Purdue University. Dombrow, J., M. Rodriguez, and C. F. Sirmans (2000, June). A complete nonparametric event study approach. Review of Quantitative Finance and Accounting 14 (4), 361–80. Duso, T., K. Gugler, and B. Yurtoglu (2011). How E¤ective is European Merger Control. European Economic Review 55 (7), 980–1006. 34
Duso, T., D. J. Neven, and L.-H. Röller (2007). The political economy of european merger control: evidence using stock market data. Journal of Law and Economics 50 (3), 455–489. Geradin, D. and D. Henry (2005). The ec …ning policy for violations of competition law: An empirical review of the commission decisional practice and the community courts’judgments. European Competition Journal 1, 401. Gunthorpe, D. L. (1997). Business Ethics: A Quantitative Analysis of the Impact of Unethical Behavior by Publicly Traded Corporations. Journal of Business Ethics 16. Kolari, J. and S. Pynnönen (2010). Event study testing with cross-sectional correlation of abnormal returns. Review of Financial Studies 23 (11), 39– 96. MacKinlay, C. A. (1997). Event Studies in Economics and Finance. Journal of Economic Literature 35 (13-39). Motta, M. (2004). Competition Policy, Theory and Practice. Cambridge: Cambridge University Press. Nicolau, J. L. (2001, March). Parametric and nonparametric approaches to event studies: An application to a hotel’s market value. (2001-08). Saleh, W. (2007). Investors reaction to dividend announcements: parametric versus nonparametric approach. Applied Financial Economics Letters 3 (3), 169–179. Theil, H. (1950). A rank invariant method of linear and polynomial regression analysis. I, II, and III Nederl. Akad. Wektensch. Proc. 53 (4), 386–392, 521–525 and 1897–1912. Whinston, M. D. (2008, October). Lectures on Antitrust Economics, Volume 1 of MIT Press Books. The MIT Press.
35
Appendix A
Non Parametric Event Study
I.i
Theil’s method
OLS estimation of the market model is the traditional choice in the majority of event studies. However Dombrow et al. (2000) show that when the normality condition fails to hold other non-linear estimators may be preferred. The same authors argue for the adoption of robust statistics when the underlying distribution of the errors is uncertain. Accordingly, they propose to use a nonparametric estimator, …rstly suggested by Theil (1950), for its high e¢ ciency and ease of computation and implementation.58 In contrast to OLS the Theil’s estimator does not need any distributional assumptions and can be implemented as follows:59 for each case in the sample: 1. Sort the L1 (L1 = T1 order of Rmt ;
T0 + 1) data pairs of (Rmt ; Rit ) in ascending
2. Separate the data into two groups based on the median;60 3. Calculate the slope parameters and choose the median value, T heil L i;(j;j+ 21 )
T heil ; L i;(j;j+ 21 ) T heil : i
=
Ri;j+ L1
in 28, for all the
L1 2
pair
Ri;j
2
Rm(j+ L1 )
Rmj
(28)
2
T heil 4. Use the estimated d to estimate the L1 parameters: i
:
d T heil = R it it
d T heil R i
mt
d T heil as the median of the L T heil ; 5. Choose d 1 it i
Then, similarly to equation 11, for each day and …rm we proceed to estimate the nonparametric abnormal returns as in equation 29. heil ^?T = Ri? i
d T heil T heil + d Rmt i i
(29)
Notice that given the median based nature of this estimator the undue in‡uence of outliers is removed. Both the OLS and Theil’s estimators are 58
For an event study that uses the Theil’s estimator see Nicolau (2001) and Saleh (2007). The step procedure follows closely the methodology outlined in Dombrow et al. (2000). 60 In case of an odd numbered interval we drop the median observation. 59
36
easy and fast to compute and implement. However the latter one does not need any distributional assumptions on the error term. Moreover, Dombrow, Rodriguez and Sirmans …nd that Theil’s nonparametric estimation has relatively greater power, than OLS, to detect abnormal performance in presence of non normally distributed errors and o¤ers comparable results to OLS under normality.
I.ii
Non Parametric Test
The nonparametric test we use is known as the rank test and was outlined by Corrado (1989). The test is developed as follow. First, for each observation, we compute abnormal returns for all the days considered both in the estimation and event window. Then for every observation i we convert all the daily abnormal returns into their rank, within the distribution of abnormal returns of that case. d it ) Kit = rank(AR
(30)
Higher values of rank K denote an higher abnormal return. This transformation turns the distribution of the abnormal returns into a uniform distribution of the possible ranks. Under the null hypothesis of zero abnormal returns the expected rank is just one plus half the number of days considered (if we run the analysis for 250 days the expected rank is 125,5). Then the below two tests, depending on the level of aggregation, are computed:
1 N
N X
(Kt;i
( T +1 2 ))
i=1 N P _testDAAR =v t u T " N u X X t1 1 Kt;i T N t=1
i=1
( T +1 2 )
a
#2 ~N (0; 1)
(31)
The test in 31 refers to daily average abnormal return estimates and T represents the sum of days both in the event and estimation window (i.e. T = T1 T0 + T3 T2 + 2): When we aggregate the daily average abnormal return to construct the CAR measure we use the test proposed by Cowan (1992). This test extend the original test proposed by Corrado (1989) to multi day event window assuming that the daily return ranks within the window are independent.
37
1 T3 T2 +1 1
N P _testCAAR = (T3 T2 +1) 2
T3 X T2 +1
"
1 N
N X
(Kd;i
i=1 d=1 v u T " N u X X t1 1 Kt;i T N t=1
38
i=1
#
( T +1 2 ))
( T +1 2 )
#2
a
~N (0; 1)
(32)
B
Tables
Event W.
Start (154 obs.) CAAR J
Infring. (139 obs.) CAAR J
Acquitt. (15 obs.) CAAR J
(-20;+10) (-5;+5) (-1;+5) (-1;+1)
0:34 0:14 0:22 0:03
0:65 0:48 0:72 0:48
2:97 1:38 0:35 0:51
0:34 0:28 0:56 0:12
0:61 0:79 1:50 1:53
0:98 0:80 0:26 0:58
Abnormal Returns as percentagess; One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table B.I: The impact of the investigations of the Italian Competition Authority on …rms’market value
Event W.
Upheld (49 obs.) CAAR J
CAAR
(-20;+10) (-5;+5) (-1;+5) (-1;+1)
2:82 0:89 0:77 0:12
0:50 0:56 0:56 0:48
1:40 0:78 0:85 0:21
Annuls (27 obs.) J 0:25 0:50 0:64 0:84
Abnormal Returns as percentages; One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table B.II: The impact of the judgment of the Italian Court of Appeal (Consiglio di Stato) on …rms’market value
39
Coe¤. Constant Fine/Cap
t
3:42 0:03
3:41 0:32
Coe¤.
t
:01 1:77
0:08 0:63
Controls Leniency Article Post98 Sector Nationality
YES NO NO NO NO
YES YES YES YES YES
obs R2
130 0.02
130 0.03
Dep. variable
31-day CAR at Raid
Two-sided test, signi…cance levels *** 1% ** 5% * 10%
Table B.III: Regression of abnormal returns for Raid on …ne/cap ratio
Coe¤. Constant Fine/Cap
t
2:40 0:59
3:07 7:39
Coe¤.
t
0:99 0:55
0:39 6:98
Controls Leniency Article Post98 Sector Nationality
YES NO NO NO NO
YES YES YES YES YES
obs R2
240 0.19
240 0.20
.
Dep. variable
31-day CAR at Decision
Two-sided test, signi…cance levels *** 1% ** 5% * 10%
Table B.IV: Regression of abnormal returns for Decision on …ne/cap ratio
40
C
The sample
41
42
Date Dec.
14/12/79 25/11/80 17/12/81 23/11/84 23/11/84 23/11/84 14/12/85 18/12/85 18/12/85 23/4/86 23/4/86 23/4/86 23/4/86 23/4/86 10/7/87 5/12/88 21/12/88 21/12/88 21/12/88 21/12/88 21/12/88 21/12/88 13/12/89 19/12/90 19/12/90
Art.
NC 101 NC 101 NC 101 101 101 101 102 NC 101 NC 101 101 101 101 101 101 NC 101 102 101 101 101 101 101 101 NC 101 101 101 21/8/86 21/11/83 21/11/83 21/11/83 21/11/83 21/11/83 21/11/83
13/10/83 13/10/83 13/10/83 13/10/83 13/10/83
9/12/80 9/12/80 9/12/80
Date Raid Pioneer Johnson & Johnson Siemens Solvay Degussa Air Liquide AKZO Siemens Fanuc Solvay Imperial Chemical Hoechst BASF Shell Beiersdorf BPB Industries Bayer Imperial Chemical Dow Chemical BASF Hoechst Shell Bayer Imperial Chemical Solvay
Firm
0.300 0.200 0.039 3.000 3.000 0.500 10.000 1.000 1.000 2.500 10.000 9.000 2.500 9.000 0.010 0.150 2.500 3.500 2.250 5.500 1.000 0.850 0.500 7.000 7.000
Fine Com.
2.500 9.000 9.000 2.125 8.100 0.010 0.150 0.000 0.000 0.000 0.000 0.000 0.000 0.500 0.000 0.000
7.500
0.039
0.200
Fine 1st judg. 0.040% 0.005% 0.001% 0.457% 0.349% 0.026% 4.548% 0.007% 0.013% 0.170% 0.113% 0.128% 0.035% 0.025% 0.001% 0.009% 0.029% 0.036% 0.017% 0.076% 0.013% 0.002% 0.006% 0.085% 0.362%
Fine/Cap. ratio.
Table C.I: Sample of …rms and antitrust cases
13:87% 1:44% 8:52% 10:46% 6:38% 1:99% 2:90%
4:86% 0:51% 0:14% 2:80% 13:11%
5:69% 11:96%
CAR Raid 0:42% 23:08% 5:95% 0:48% 5:02% 9:04% 6:89% 8:85% 9:87% 5:34% 8:31% 8:65% 6:53% 6:59% 3:05% 9:95% 0:10% 3:65% 0:06% 0:02% 3:13% 0:08% 4:65% 3:55% 10:15%
CAR Dec.
43
Date Dec.
1/4/92 15/7/92 5/6/91 13/7/94 27/7/94 27/7/94 27/7/94 27/7/94 27/7/94 27/7/94 30/11/94 30/11/94 30/11/94 30/11/94 30/11/94 30/11/94 30/11/94
30/11/94 21/12/94
21/12/94
21/12/94
Art.
101 & 102 NC 101 NC 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101
101 101
101
101
23/4/91
19/9/89
Date Raid Nedlloyd Herlitz Toshiba SCA Holding Imperial Chemical BASF Solvay Norsk Hydro Hoechst Shell Dyckerho¤ Heidelberger Ciments Francais Lafarge Titan Cement Buzzi Unicem CementirCementeries del Tirreto Italcementi Orient Overseas Container Line Nippon Yusen Kabushiki Kaisha Neptune Orient Lines
Firm
0.010
0.010
33.580 0.010
0.026 0.040 2.000 2.200 2.500 1.500 3.500 0.750 1.500 0.850 13.284 15.652 25.768 23.900 5.625 3.652 8.248
Fine Com.
0.000
0.000
26.789 0.000
1.500 0.850 8.043 7.056 13.570 15.276 0.000 0.000 7.471
2.200 1.550 1.500
0.040
Fine 1st judg.
0.001%
0.000%
2.794% 0.004%
0.005% 0.008% 0.014% 0.265% 0.033% 0.015% 0.152% 0.025% 0.015% 0.001% 1.445% 0.692% 4.221% 0.470% 2.259% 1.310% 4.489%
Fine/Cap. ratio.
2:86%
0:70%
CAR Raid
0:79%
5:72%
8:83% 3:49%
13:85% 15:27% 2:60% 13:78% 4:04% 2:13% 2:45% 7:24% 4:91% 1:86% 6:60% 6:65% 3:71% 6:57% 6:57% 6:77% 11:67%
CAR Dec.
44
Date Dec.
21/12/94 21/12/94
12/7/95 10/1/96 21/1/98 21/1/98 28/1/98 16/9/98 16/9/98 16/9/98 16/9/98 16/9/98 16/9/98 16/9/98 14/10/98 9/12/98 14/7/99 8/12/99 8/12/99 8/12/99 16/5/00 16/5/00 16/5/00 16/5/00 16/5/00
Art.
101 101
NC 101 NC 101 101 101 NC 101 101 101 101 101 101 101 101 101 101 102 101 101 101 101 101 101 101 101 27/5/94 5/7/94 12/6/97 1/12/94 1/12/94 1/12/94
23/10/95
26/6/91
Date Raid Mitsui OSK Lines Kawasaki Kisen Kaisha BASF Bayer Acerinox Thyssenkrupp Volkswagen Hyundai Merchant Hanjin Shipping P & O Nedlloyd A.P. Moller-Maersk Orient Overseas Nippon Yusen Neptune Orient Tate & Lyle Minoan Lines British Airways Nippon Steel Sumitomo Metal Vallourec Evergreen Malaysia Shipping Hanjin Shipping P & O Nedlloyd A.P. Moller Maersk
Firm
2.700 3.000 3.530 8.100 102.000 18.560 20.630 41.260 27.500 20.630 20.630 13.750 7.000 3.260 6.800 13.500 13.500 8.100 0.368 0.134 0.620 1.240 0.836
0.010 0.010
Fine Com.
2.700 0.000 3.136 4.032 90.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 5.600 3.260 6.800 10.935 10.935 8.100 0.000 0.000 0.000 0.000 0.000
0.000 0.000
Fine 1st judg.
0.025% 0.020% 0.219% 0.132% 0.574% 8.462% 65.889% 14.284% 0.501% 13.082% 0.641% 4.494% 0.354% 0.697% 0.099% 0.082% 0.422% 2.232% 0.020% 0.004% 0.274% 0.227% 0.008%
0.000% 0.001%
Fine/Cap. ratio.
8:14% 8:47% 16:81% 12:00%
10:43%
2:21%
5:79%
CAR Raid
1:67% 5:92% 4:02% 6:12% 7:23% 6:37% 15:44% 31:88% 10:18% 14:44% 3:06% 11:60% 8:95% 8:24% 16:88% 6:66% 4:17% 4:46% 1:71% 0:62% 11:66% 14:44% 3:60%
8:99% 6:73%
CAR Dec.
45
Date Dec.
16/5/00 16/5/00 16/5/00 16/5/00 16/5/00
7/6/00 7/6/00 7/6/00 7/6/00 20/9/00 20/6/01 29/6/01 18/7/01 18/7/01 18/7/01 18/7/01 18/7/01 18/7/01 18/7/01 25/7/01 10/10/01 21/11/01 21/11/01 21/11/01
Art.
101 101 101 101 101
101 101 101 101 NC 101 102 NC 101 101 101 101 101 101 101 101 102 NC 101 101 101 101 11/12/96
15/6/00 5/6/97 5/6/97 5/6/97 5/6/97 5/6/97 5/6/97
11/6/97 11/6/97 11/6/97 11/6/97 11/12/96
Date Raid Orient Overseas Nippon Yusen Neptune Orient Mitsui OSK Kawasaki Kisen Kaisha Daesang Kyowa Hakko Ajinomoto Archer Daniels Opel Nederland Michelin Volkswagen SAS Carbide Graphite SEC Nippon Carbon Tokai Carbon Showa Denko SGL Carbon Deutsche Post DaimlerChrysler Eisai Solvay Daiichi Pharmaceutical
Firm
8.900 13.200 28.300 47.300 43.000 19.760 30.960 39.375 10.300 12.200 12.200 24.500 17.400 80.200 0.001 71.825 13.230 9.100 23.400
0.134 0.620 0.368 0.620 0.620
Fine Com.
18.000
9.800
6.480 6.138 6.274 12.276 10.440 69.114
43.875 35.475 19.760 0.000
7.128 13.200
0.000 0.000 0.000 0.000 0.000
Fine 1st judg.
5.751% 0.303% 0.338% 0.760% 13.213% 0.396% 0.145% 4.771% 126.536% 13.587% 6.620% 4.873% 0.714% 8.649% 0.000% 0.185% 0.154% 0.179% 0.255%
0.046% 0.011% 0.017% 0.022% 0.059%
Fine/Cap. ratio.
0:48% 0:07% 7:66% 6:07% 10:77% 27:36% 15:38%
41:27% 5:05% 2:34% 21:10% 0:86%
CAR Raid
16:50% 7:05% 0:29% 5:04% 7:28% 6:72% 8:05% 11:23% 79:75% 1:36% 38:93% 5:59% 53:42% 8:62% 5:40% 7:54% 3:74% 3:06% 7:22%
2:49% 3:23% 23:19% 0:96% 1:89%
CAR Dec.
46
Date Dec.
21/11/01 21/11/01 21/11/01 21/11/01 21/11/01 5/12/01 5/12/01 5/12/01 5/12/01 11/12/01
11/12/01 11/12/01 11/6/02
2/7/02 2/7/02 2/7/02 24/7/02 24/7/02 24/7/02 24/7/02 30/10/02 30/10/02 30/10/02
Art.
101 101 101 101 101 NC 101 101 101 101 101
101 101 101
101 101 101 101 101 101 101 101 101 NC 101
16/6/99 16/6/99 16/6/99 11/12/97 11/12/97 11/12/97 11/12/97
16/2/99 16/2/99
16/2/99
13/7/99
Date Raid Merck Takeda Chemical Aventis BASF Ho¤man La Roche Danone Archer Daniels Bayer Ho¤man La Roche Bayerische Hypound Vereinsbank Dresdner Bank Commerzbank Erste Bank der oesterreichischen Sparkassen Nippon Soda Degussa Aventis Air Liquide Air Products BOC Linde Sotheby Christie Nintendo
Firm
9.000 118.125 0.000 3.640 2.730 1.170 12.600 20.400 0.000 149.128
28.000 28.000 37.690
9.240 37.060 5.040 296.160 462.000 44.043 39.690 14.220 63.500 28.000
Fine Com.
119.243
12.600
91.125
0.000 0.000 37.690
0.000
43.225 30.690
236.854
Fine 1st judg.
2.837% 1.584% 0.000% 0.028% 0.027% 0.018% 0.228% 5.831% 0.000% 1.085%
0.123% 0.272% 0.985%
0.171% 0.080% 0.008% 1.129% 0.637% 0.228% 0.344% 0.053% 0.088% 0.117%
Fine/Cap. ratio.
3:13% 4:04% 4:66% 1:09% 2:69% 2:86% 0:53%
11:71% 4:26%
9:67%
8:23%
CAR Raid
3:74% 7:49% 4:14% 6:31% 5:83% 6:13% 10:80% 1:52% 6:28% 17:37%
2:65% 8:22% 8:42%
6:77% 5:49% 9:38% 1:81% 3:67% 0:33% 9:43% 2:23% 17:76% 5:52%
CAR Dec.
47
Date Dec.
27/11/02 27/11/02 27/11/02 27/11/02 21/5/03 16/7/03 1/10/03 3/12/03 3/12/03 10/12/03 10/12/03 10/12/03 16/12/03 16/12/03 24/3/04 26/5/04 3/9/04 3/9/04 3/9/04 29/9/04 29/9/04 9/12/04 9/12/04 9/12/04 15/6/05 5/10/05
Art.
101 101 101 101 102 NC 101 101 101 101 101 101 101 101 101 102 NC 101 101 101 101 101 101 101 101 101 102 NC 101 9/2/00 21/9/99
22/3/01 22/3/01 22/3/01 25/1/00 25/1/00
22/3/03 22/3/03
25/11/98 25/11/98 15/1/01 15/1/01
Date Raid BPB Lafarge Aventis Merck Deutsche Telekom Yamaha Hoechst SGL Carbon Carbone Lorraine Degussa Ato…na AKZO Outokumpu KME Microsoft Topps KME Outokumpu Halcor Danone Heineken BASF AKZO UCB AstraZeneca Peugeot
Firm
138.600 249.600 2.850 0.000 12.600 2.560 99.000 23.640 43.050 16.730 43.470 0.000 18.130 18.990 497.196 1.590 32.750 36.140 9.160 1.500 1.000 34.970 20.990 10.380 60.000 49.500
Fine Com.
35.024 20.990 1.870 52.500
32.750 36.140 8.247
18.130 18.990 497.196
74.250 23.640 43.050 16.730
12.600
118.800 249.600
Fine 1st judg. 6.712% 2.462% 0.006% 0.000% 0.025% 0.095% 0.474% 7.694% 12.914% 0.314% 0.048% 0.000% 0.994% 8.247% 0.230% 0.551% 21.246% 1.524% 5.276% 0.009% 0.008% 0.124% 0.233% 0.183% 0.110% 0.354%
Fine/Cap. ratio.
11:79% 5:68%
5:04% 20:57% 4:15% 0:10% 18:16%
0:11% 14:73%
29:17% 10:30% 14:47% 25:10%
CAR Raid 3:42% 4:15% 16:85% 3:25% 6:77% 15:48% 4:33% 26:88% 18:41% 8:66% 5:00% 6:21% 5:94% 7:80% 0:49% 7:66% 6:49% 5:09% 1:94% 6:08% 0:85% 5:60% 1:34% 8:79% 10:65% 7:03%
CAR Dec.
48
Date Dec.
20/10/05 30/11/05 30/11/05 30/11/05 30/11/05 30/11/05 30/11/05 21/12/05 21/12/05 3/5/06 3/5/06 3/5/06 3/5/06 3/5/06 31/5/06 13/9/06 13/9/06 13/9/06 13/9/06 13/9/06 20/9/06 20/9/06 20/9/06 8/11/06 29/11/06 29/11/06
Art.
101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 27/3/03 27/3/03
18/4/02 26/6/02 26/6/02 26/6/02 26/6/02 26/6/02 26/6/02 26/9/02 26/9/02 25/3/03 25/3/03 25/3/03 25/3/03 25/3/03 25/3/03 1/10/02 1/10/02 1/10/02 1/10/02 1/10/02 22/3/01 22/3/01 22/3/01
Date Raid Dimon Low & Bonar British Polythene Kendrion Sachsa Verpackung Trioplast UPM-Kymmene Repsol Bayer Edison FMC Kemira Solvay AKZO Total BAM BP Heijmans Shell Total Mueller IMI Aalberts Arcelor Shell Unipetrol
Firm
10.000 12.240 0.000 32.900 13.200 17.850 56.550 3.380 58.880 58.125 25.000 33.000 167.062 25.200 219.131 13.500 0.000 17.100 108.000 20.250 0.000 48.300 100.800 10.000 160.875 17.550
Fine Com.
10.000
25.200
3.380
17.850
Fine 1st judg. 4.296% 6.217% 0.000% 21.755% 10.916% 1.443% 0.666% 0.011% 0.228% 0.850% 1.247% 1.890% 2.106% 0.192% 0.174% 0.782% 0.000% 1.889% 0.064% 0.017% 0.000% 1.941% 6.999% 0.032% 0.093% 1.212%
Fine/Cap. ratio.
4:04% 6:59%
7:89% 7:09% 5:99% 14:03% 0:07% 11:17% 9:69% 5:54% 4:42% 6:59% 13:57% 11:15% 9:01%
8:74% 22:52% 13:05% 12:59% 8:35% 21:26% 11:84% 4:58% 2:31%
CAR Raid 23:02% 10:55% 6:71% 31:57% 11:43% 2:72% 0:11% 7:32% 0:43% 1:51% 3:56% 6:39% 8:25% 2:91% 1:22% 19:96% 7:72% 16:84% 12:94% 8:10% 27:92% 4:34% 14:71% 16:04% 0:36% 11:21%
CAR Dec.
49
Date Dec.
29/11/06 29/11/06 29/11/06 30/11/06 20/12/06 24/1/07 24/1/07 24/1/07 24/1/07 24/1/07 24/1/07 24/1/07 24/1/07 24/1/07 21/2/07 21/2/07 21/2/07 21/2/07 18/4/07 18/4/07 4/7/07 3/10/07 3/10/07 3/10/07 3/10/07 20/11/07
Art.
101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 101 102 101 101 101 101 101 13/3/07
11/5/04 11/5/04 11/5/04 11/5/04 11/5/04 11/5/04 11/5/04 11/5/04 11/5/04 28/1/04 28/1/04 28/1/04 28/1/04 22/3/00 22/3/00
27/3/03 27/3/03 27/3/03
Date Raid Bayer Dow Eni Trade-Stomil ThyssenKrupp Alstom Fuji Electric Mitsubishi Electric Schneider Siemens ABB Areva Hitachi Toshiba KONE Otis Schindler ThyssenKrupp Heineken InBev Telefónica BP Cepsa Galp Repsol Sony
Firm
0.000 64.575 272.250 3.800 3.168 65.025 3.750 118.575 8.100 396.563 0.000 53.550 51.750 90.900 142.120 224.933 143.748 479.670 219.275 0.000 151.875 0.000 83.850 8.663 80.496 47.190
Fine Com.
3.168
Fine 1st judg. 0.000% 0.224% 0.275% 2.319% 0.018% 0.491% 0.128% 0.792% 0.038% 0.540% 0.000% 0.247% 0.291% 0.550% 2.205% 0.439% 2.273% 2.381% 1.142% 0.000% 0.192% 0.000% 0.439% 0.098% 0.262% 0.144%
Fine/Cap. ratio.
2:86%
41:58% 3:22% 4:73% 2:75% 1:87% 8:46% 9:68% 0:46% 1:60% 1:66% 6:47% 0:65% 5:22% 14:12%
18:94% 1:23%
CAR Raid 1:45% 1:29% 4:36% 24:27% 17:46% 15:21% 19:96% 4:91% 8:50% 9:09% 4:21% 12:11% 10:94% 8:86% 7:79% 1:79% 6:07% 5:45% 1:38% 6:06% 5:23% 5:50% 1:16% 2:53% 11:28% 19:07%
CAR Dec.
50
Date Dec.
20/11/07 28/11/07 28/11/07 28/11/07 5/12/07 5/12/07 5/12/07 5/12/07 5/12/07 5/12/07 23/1/08 23/1/08 27/2/08 11/6/08 11/6/08 25/6/08
1/10/08 1/10/08 1/10/08 1/10/08 1/10/08 1/10/08 1/10/08 1/10/08 15/10/08
Art.
101 101 101 101 101 101 101 101 101 101 101 101 102 101 101 101
101 101 101 101 101 101 101 101 101
28/4/05 28/4/05 28/4/05 28/4/05 28/4/05 28/4/05 28/4/05 28/4/05 2/6/05
13/3/07 22/2/05 22/2/05 22/2/05 27/3/03 27/3/03 27/3/03 27/3/03 9/7/03 27/3/03 27/3/03 27/3/03
Date Raid Fuji Asahi Saint-Gobain Pilkington Tosoh DuPont Bayer Dow Denka ENI Bayer Zeon Microsoft Arkema Uralita Industrial Quimica de Mexico ExxonMobil Shell Total Sasol Repsol ENI MOL RWE Chiquita
Firm
83.588 0.000 128.163 318.200 19.800 29.120 23.700 37.440 0.000
13.200 65.000 133.900 140.000 4.800 59.250 0.000 4.425 47.000 132.160 28.870 5.360 889.000 59.020 9.900 1.670
Fine Com.
Fine 1st judg.
0.029% 0.000% 0.131% 1.771% 0.082% 0.040% 0.342% 0.101% 0.000%
3.443% 0.582% 0.528% 5.741% 0.247% 0.203% 0.000% 0.016% 2.894% 0.132% 0.069% 0.561% 0.519% 2.486% 0.864% 5.299%
Fine/Cap. ratio.
11:18% 1:93% 0:68% 7:09% 3:67% 1:37% 10:11% 1:31% 2:69%
18:94% 17:68%
4:52% 3:43% 5:17% 0:92% 13:20% 0:44% 18:94% 1:23% 16:77%
CAR Raid
3:33% 1:39% 13:67% 19:29% 10:30% 21:12% 9:25% 11:67% 25:83%
5:66% 2:36% 0:97% 21:87% 36:61% 5:88% 2:80% 2:72% 26:90% 4:09% 0:30% 9:90% 15:93% 11:32% 3:19% 4:15%
CAR Dec.
51
15/10/08 12/11/08 12/11/08 12/11/08 28/1/09 28/1/09 28/1/09 28/1/09 28/1/09 13/5/09 8/7/09 22/7/09 7/10/09 7/10/09 7/10/09 7/10/09 7/10/09 7/10/09 7/10/09
101 101 101 101 101 101 101 101 101 102 101 101 101 101 101 101 101 101 101 16/5/06
2/6/05 22/2/05 22/2/05 22/2/05 2/5/07 2/5/07 2/5/07 2/5/07 2/5/07
Date Raid Del Monte Asashi Saint-Gobain Pilkington Bridgestone Yokohama Dunlop Trelleborg Parker ITR Intel GDF Suez AKZO Alstom Fuji Electrics Siemens ABB Areva Hitachi Toshiba
Firm
14.700 113.500 896.000 370.000 58.500 0.000 18.000 24.500 25.610 1060.000 553.000 0.000 9.735 1.734 0.000 33.750 6.765 2.460 13.200
Fine Com.
Fine 1st judg.
Note Art.: article infringed (NC 101-non cartel 101) Date Dec. date of EC decision Date Raid: date of surprise inspection Fine Com.: …ne in‡icted by the European Commission Fine 1st judg.: …ne set by the Court of …rst instance, when applicable
Date Dec.
Art.
1.988% 2.239% 8.326% 21.218% 0.703% 0.000% 0.834% 6.387% 0.526% 1.660% 1.005% 0.000% 0.066% 0.183% 0.000% 0.104% 0.048% 0.032% 0.086%
Fine/Cap. ratio.
5:54%
12:67% 3:43% 5:17% 0:92% 5:90% 10:82% 0:69% 6:44% 3:45%
CAR Raid 12:92% 18:74% 21:79% 51:57% 4:94% 30:76% 36:66% 8:06% 8:87% 15:42% 3:10% 8:39% 3:83% 4:43% 0:27% 2:69% 10:72% 1:62% 6:50%
CAR Dec.
52
Date Dec.
Date Raid
Firm
Fine Com.
Fine 1st judg.
Fine/Cap. ratio.
CAR Raid
Fine/Cap ratio: ratio between EC …ne and …rms’market capitalization CAR Raid: …rms’individual Cumulative Abnormal Return at Raid(-20:+10) CAR Decision: …rms’individual Cumulative Abnormal Return at Decision (-20:+10)
Art.
CAR Dec.
D
Figures to be inserted in the text
Figure 1: Description of antitrust procedure
53
Figure 2: Yearly average EU antitrust …nes (1979-2009), in mil.euro
54
Figure 3: Yearly average …ne capitalisation ratio, (1979-2009)
Figure 4: Sample distribution by …rms’sector
55
Figure 5: Sample distribution by …rms’country of origin
Figure 6: Timeline: Estimation and event windows
56
Figure 7: Cumulative abnormal returns by dates of dawn raids (points represent …rm’s estimated CARs); 31-day event window (-20;+10)
Figure 8: Cumulative abnormal returns by dates of Decisions (points represent …rm’s estimated CARs); 31-day event window (-20;+10)
57
E
Tables to be inserted in the text Commission’s Decision
1979-1997
1998-2002
2003-2009
1979-2009
Cases Observations Firms involved Leniency immunities Raids Avg. duration between Raid and Decision (in years) Avg. …ne (in millions of euros) Avg. …ne/cap ratio
21 50 36 0 17 3.7
27 72 64 3 33 3.6
43 118 93 14 80 3.5
91 240 180(*) 17 130 3.5
4.6 0.06%
36.7 0.33%
88.2 0.28%
55.3 0.27%
First Judgement
1979-1997
1998-2002
2003-2009
1979-2009
Upheld decisions Annulled decisions Avg. duration between Decision and First Judgment (in years) Avg. …ne (in millions of euros)
21 16 4.9
26 22 3.6
13 1 3.5
60 39 4.1
3.2
20.1
52.1
18.3
(*) The sum does not amount to 180 since there are repeated o¤enders from a period to another.
Table I: Description of the sample
58
Days to event t= t= t= t= t= t= t= t= t= t=
Raid (130 obs.) AAR J
Com.Dec. (240 obs.) AAR J
Upheld (60 obs.) AAR J
Annulled (39 obs.) AAR J
-20 -19 -18 -17 -16 -15 -14 -13 -12 -11
0:21 0:13 0:18 0:00 0:06 0:04 0:12 0:12 0:13 0:26
1:29 0:76 0:07 0:15 0:41 1:08 0:58 0:36 0:15 1:16
0:38 0:15 0:07 0:18 0:10 0:01 0:10 0:14 0:17 0:19
2:59 1:39 0:33 1:26 0:54 0:18 0:26 0:10 1:86 1:23
0:12 0:11 0:01 0:42 0:11 0:28 0:34 0:27 0:11 0:01
0:39 0:08 0:25 2:01 0:34 0:32 1:06 1:02 0:62 0:72
0:04 0:55 0:27 0:30 0:72 0:29 0:21 0:10 0:21 0:39
0:85 2:00 0:82 0:74 0:67 0:73 0:42 0:96 0:85 0:04
t= -10 t= -9 t= -8 t= -7 t= -6 t= -5 t= -4 t= -3 t= -2 t= -1 t= 0
0:12 0:34 0:37 0:51 0:28 0:05 0:13 0:20 0:01 0:17 0:64
0:94 1:65 1:59 0:90 1:10 0:50 0:08 0:18 0:15 0:24 2:08
0:03 0:27 0:31 0:03 0:06 0:17 0:09 0:00 0:09 0:04 0:19
0:43 1:78 2:07 0:09 0:48 0:70 0:22 0:39 0:99 0:17 1:02
0:10 0:15 0:13 0:32 0:37 0:16 0:09 0:33 0:09 0:29 0:17
1:06 0:18 0:09 0:07 1:99 0:75 0:29 1:48 0:48 1:00 0:59
0:49 0:10 0:78 0:41 0:64 0:27 0:23 0:38 0:18 0:32 0:39
2:14 0:38 1:41 1:11 0:60 0:37 0:33 1:04 0:81 0:00 0:90
t= 1 t= 2 t= 3 t= 4 t= 5 t= 6 t= 7 t= 8 t= 9 t= 10
0:43 0:31 0:15 0:26 0:04 0:14 0:18 0:21 0:09 0:00
1:45 1:48 0:62 0:32 0:30 0:96 1:25 0:68 0:40 0:43
0:14 0:00 0:05 0:10 0:05 0:17 0:16 0:05 0:10 0:42
0:19 0:51 1:00 0:51 0:35 1:76 0:61 0:39 1:06 2:07
0:37 0:09 0:22 0:10 0:19 0:26 0:01 0:11 0:28 0:39
0:95 0:57 0:62 0:19 0:08 0:86 0:07 0:00 0:92 1:89
0:70 0:56 0:03 0:09 0:84 0:16 0:13 0:39 0:29 0:52
1:56 1:30 0:29 0:04 1:90 0:64 0:12 1:41 1:61 1:04
Event Window
Raid (130 obs.) CAAR J
(-20;+10) (-5;+5) (-1;+5) (-1;+1)
2:89 1:17 1:14 0:90
2:50 1:53 2:34 2:18
Com.Dec. (240 obs.) CAAR J 3:57 0:93 0:56 0:37
3:73 1:56 0:96 0:59
Upheld (60 obs.) CAAR J
Annulled (39 obs.) CAAR J
0:77 0:25 0:25 0:09
1:35 1:77 1:25 1:41
0:11 0:05 0:01 0:26
Top panel: Average abnormal returns; Bottom panel: Cumulative average abnormal returns Abnormal Returns as percentages; One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table II: Base results - Daily and cumulative average abnormal returns 59
0:36 1:08 0:93 1:45
Event
Raid (130 obs.)
Raid (100 obs.)
Full sample
Excl. Leniency and US
Window
CAAR
(-20;+10) (-5;+5) (-1;+5) (-1;+1)
2:89 1:17 1:14 0:90
J 2:50 1:53 2:34 2:18
CAAR 2:96 0:84 1:23 1:05
J 2:86 1:21 2:37 1:92
Abnormal Returns as percentages One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table III: Raid estimates - Full sample and excluding cases previously investigated in the US and leniency cases
Quantile 90% 80% 75% 70% 60% Median 40% 30% 25% 20% 10%
CAR as % Raid Decision 8:55 4:70 2:86 2:05 0:33 1:76 4:55 7:80 9:51 10:84 13:93
8:85 5:26 3:75 2:72 0:02 2:89 5:65 6:94 8:23 9:52 16:53
Table IV: Distribution of CARs, Decision and Raid; 31-day event window (-20;+10)
60
Infringement Raid
Cartel CAAR J 2:72
2:03
Abuse CAAR J 11:27
4:53
(121 obs.)
Decision
3:73
Art. 101 not Cartel CAAR J 2:02
(3 obs.)
3:97
7:66
(209 obs.)
1:25 (6 obs.)
2:56
0:00
(12 obs.)
0:88 (20 obs.)
Abnormal Returns as percentages One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table V: Estimated CAARs for Raid and Decision, by type of infringement, 31-day event window (-20;+10)
Hydrocarbon Chemicals Raid (-20;+10)
CAAR 1:22
J 0:61
(obs. 65)
Decision (-20;+10)
CAAR 2:13
Machinery Vehicles CAAR 4:38
Construction Metal
J 2:07
(obs. 30)
J 1:84
(obs. 108)
CAAR 9:16
CAAR 8:98
J 4:09
(obs. 15)
J 1:96
(obs. 42)
CAAR 2:44
J 1:69
(obs. 28)
Services Transports CAAR 5:36
(obs. 8)
CAAR 3:47
Table VI: Estimated CAARs for Raid and Decision, by sectors, 31-day event window (-20;+10)
Raid
1979-1997 CAAR J
1998-2009 CAAR J
5:15
1:95
2:83 (38 obs.)
Decision
0:98
1:41 (92 obs.)
0:41 (50 obs.)
4:25
3:72 (190 obs.)
Abnormal Returns as percentages One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table VII: Estimated CAARs for Raid and Decision, by period (2 periods), 31-day event window (-20;+10)
61
J 2:19
(obs. 39)
Abnormal Returns as percentages One-sided test, signi…cance levels *** 1% ** 5% * 10%
Periods
J 2:94
Periods Raid
1979-1993 CAAR J
1994-1997 CAAR J
1998-2002 CAAR J
2003-2009 CAAR J
2:37
7:41
4:85
0:00
1:46 (17 obs.)
Decision
1:49
0:53
2:43 (21 obs.)
0:33
(28 obs.)
0:00
2:47 (37 obs.)
5:92
(22 obs.)
3:84 (72 obs.)
0:31 (55 obs.)
3:23
2:22 (118 obs.)
Abnormal Returns as percentages; One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table VIII: Estimated CAARs for Raid and Decision, by period (4 periods), 31-day event window (-20;+10)
Appeal Outcome Raid
Fines upheld CAAR J
Fines annulled CAAR J
7:37
2:74
3:51 (39 obs.)
Decision
5:16
0:33 (9 obs.)
3:27
3:07
(71 obs.)
1:69 (41 obs.)
Abnormal Returns as percentages One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table IX: Estimated CAARs for raids and decisions or …rms whose …ne has eventually been annulled (resp. upheld); 31-day event window (-20;+10)
Raid (130 obs.) CAAR J 2:44
1:98
Com.Dec. (240 obs.) CAAR J 3:78
2:83
Upheld (60 obs.) CAAR J
Annulled (39 obs.) CAAR J
1:13
0:16
0:57
0:36
Abnormal Returns as percentages; One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table X: Robustness check estimates: Mean Model; 31-day event window (-20;+10)
62
Raid (130 obs.) CAAR test 2:00
3:44
Com.Dec. (240 obs.) CAAR test 2:50
5:14
Upheld (60 obs.) CAAR test
Annulled (39 obs.) CAAR test
0:99
2:72
0:84
0:83
Abnormal Returns as percentages; One-sided test, signi…cance levels *** 1% ** 5% * 10% The test is the multi-day rank test (Cowan, 1992)
Table XI: Robustness check estimates: Non-Parametric model; 31-day event window (-20;+10)
Days before raid Event Window (-20;+10) (-5;+5) (-1;+5) (-1;+1)
50 days (130 obs.) CAAR J 1:09 0:03 0:04 0:02
75 days (130 obs.) CAAR J
0:95 0:40 0:77 0:69
0:71 0:70 0:64 0:36
0:69 1:20 1:29 1:76
100 days (130 obs.) CAAR J 2:85 0:62 0:58 0:56
2:57 0:88 0:93 1:78
Abnormal Returns as percentages; One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table XII: Robustness check estimates: Placebo event before Raid
Event Window (-20;+10) (-5;+5) (-1;+5) (-1;+1)
100 days before Dec (237 obs.) CAAR J 0:91 0:37 0:02 0:03
0:85 0:87 0:33 0:33
SOB (225 obs.) CAAR J 0:07 0:33 0:14 0:07
0:05 0:54 0:34 0:59
Abnormal Returns as percentages; One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table XIII: Robustness check estimates: Placebo event before Decision
63
Selection
Cartels CAAR J
Firms w. leniency CAAR J
Firms no Leniency CAAR J
Raid
2:72 2:03 (121 obs.) 3:73 3:97 (209 obs.)
3:76
3:37 2:58 (110 obs.) 4:02 4:19 (192 obs.)
Decision
0:59 (11 obs.) 0:43 0:38 (17 obs.)
Abnormal Returns as percentages One-sided test, signi…cance levels *** 1% ** 5% * 10%
Table XIV: Estimated CAARs for raids and decisions, for cartel with (resp.without) leniency; 31-day event window (-20;+10)
64
