Pros and Cons of Presumptions
2020 year's Pros and Cons seminar addressed the issue of Presumptions.
Contributors (top row from left): Margaret Slade, David Bailey Damien Neven. (Middle row from left) Andriani Kalintiri, Griet Jans, Kevin Coates. (Bottom row from left) Theon van Dijk, Ana Sofia Rodrigues and Cani Fernández.
Summary of the seminar
To Presume, or not Presume, that is the question
David Bailey began his presentation by drawing a topical comparison between presumptions and the US presidential election. What they have in common is that they provoke very different reactions in people. For some, presumptions are inappropriate shortcuts, which run a high risk of leading to wrong answers. For others, presumptions are an invaluable way to make competition law administrable and workable in practice.
According to Bailey, it is something true and this might be a fact, a legal conclusion or a procedural position. By way of example, a well-known presumption is the presumption of innocence in criminal law. Bailey noted that presumptions can have different strengths. For example, a conclusive presumption is, really, a rule. These conclusive presumptions are not common in EU law, but in the US they have the per se rule which characterizes US antitrust. Bailey also noted that presumptions do not always point to illegality, for example, a block exemption contains a presumption of lawfulness.
Bailey remarked that in his view it is important for presumptions to be rebuttable in most instances. They also have to be anchored in some sort of fact and this is what distinguishes them from assumptions. Presumptions should also be grounded in and consistent with common sense.
The most popular presumption takes the form of an evidentiary presumption, which is rebuttable. This is when you draw an inference from facts in order to reach a provisional conclusion.
One of the most well-known examples comes from the Akzo 1991 case. In this case, the court attempted to work out what rule or presumption it should apply when a dominant undertaking prices below a measure of cost. The court concluded that if a dominant firm sets its prices below cost, especially if below average cost, it loses money on every product sold. This makes no economic sense unless the dominant firm has as its objective to eliminate competition against it. Bailey mentioned how the court stated this categorically and that it appeared to become a rule. In the French telecoms case, the court added to the rule by stating that where you price below average variable cost it is prima facie abusive behaviour. Bailey remarked that the addition of “prima facie” is important because it recognises that it is possible to rebut what is clearly a presumption, but what may be a strong presumption.
Bailey further mentioned that it may be appropriate to use presumptions where they reflect the lessons learned from experience and where they reflect what ought to be obvious. An example of the latter is the parental liability where it is relatively rare that a subsidiary can act autonomously if a parent owns 100% shares with voting rights.
Bailey commented that if you are not able to make a reliable and consistent inference from facts then you should not be making presumptions.
Secondly, when it is necessary and/or desirable to prove actual or likely harm to the competitive process it is inappropriate to use presumptions. By way of example, in the Microsoft 2004 case Microsoft tied its media player with Windows. The court did not presume that technical tying was harmful but instead went on to examine it. This approach appears to have been the springboard for what has been decided in more recent cases.
Thirdly, Bailey stated that it is inappropriate to use presumptions that are not rebuttable and generally these should be avoided. For example, the treatment of loyalty rebates in Hoffmann-La Roche received some criticism because loyalty rebates are not always bad. This was later clarified in Intel, 2017.
Bailey stated that presumptions are not set in stone, new ones can be created as human thinking and commercial circumstances change. This is why it is good not to have legislative presumptions because these are harder to change. It is instead preferable to have judicial and administrative presumptions.
Bailey concluded by stating that presumptions are important because they make the system of competition law workable and can increase legal certainty while reducing costs. However, they need to be rebuttable and grounded in fact or experience. Bailey recognised that of course presumptions have drawbacks but ultimately that it boils down to what system of competition law we want to have.
The economic and enforcement aspects of presumptions
Damien Neven discusses whether and when it is appropriate to use presumption in decision-making. He explains use of presumption in decision theoretic perspective, where "presumption" is defined as a prior informed by economic theory, empirical evidence, enforcement experience, with respect to the consequence of a conduct that is informed by a limited case-specific information. The alternative to a decision-making based on a prior is to acquire additional information and update the prior, and to decide based on the updated prior.
A competition authority can make decision based on information from experience, or gather new information that is specific for a particular conduct and make a decision. Hence, whether to use presumption or not is optimal stopping rule for a competition authority to investigate or not. The goal of a competition authority is to achieve consumer welfare. The metric to make decision is probability whether a conduct is anti-competitive or not. The authority undertakes investigation if and only if the default case based on the initial information is likely to be overturned, which is more likely when the investigation is sufficiently precise.
Neven presents that there are two strands in the literature; Bayesian framework and the choice between alternative decision rules, i.e. per se vs rule of reason, or per se vs effects. Neven's model is closely related to the Bayesian framework to enforcement decisions; Beckner and Salop (1999) and Salop (2017). They consider a framework in which an investigation proceeds by allowing parties (plaintiffs and defendants) to bring evidence in turn to modify the prior into a new posterior distribution, which leads to a process of "burden shifting" in which each party attempt to shift the prior in a direction that is favorable to its case. Neven's work attempt to develop a normative analysis in a framework, in which even if the standard of proof does not change as the evidence accumulates, the "incremental standard" changes at each iteration.
Neven's model shows in which circumstances an authority wants to do investigation rather than to make immediate decision based on prior. If the expected consumer welfare before the acquisition of the information is negative, an investigation is only attractive when a good news leads to a clearance, which occurs when it is sufficiently likely and precise. If the expected consumer welfare before the acquisition of the information is positive, it is still better off doing investigation.
The standard for a clearance and the standard for a prohibition before the investigation cannot both be identical to the standard that applies respectively for a clearance and for a prohibition after the investigation. If it were the case, no investigation would ever be undertaken.
If the conduct is no longer prohibited after obtaining good news the overall errors will fall. In this case, there is a type II error after the investigation when good news has been received. Hence, a prohibition without investigation will only be dominated by an investigation in terms of expected errors if in the case of good news the conduct is allowed. Similarly, an investigation will dominate a clearance according to the standard if and only if in case of bad news the practice is prohibited.
When maximizing consumer welfare, it is optimal to undertake an investigation if and only if the default case is likely to be overturned. This is more likely when investigation is sufficiently precise.
When mapping Neven's model with the case law, capability applies to restrictions by object under 101. Analysis of plausibility requires evaluating the practice in its legal and economic context. e.g. Intel. Capability needs to be interpreted as an incremental standard of proof.
Merger control is different, where there is no presumption to start with. Unlike what happens with 101 and 102, cases cannot be prohibited before searching for additional information. The merger regulation establishes a high threshold for a clearance in phase I.
The use of presumptions in the application of Article 102 TFEU
Cani Fernández, President of the Spanish National Commission for Markets and Competition, presented a paper on the topic of the use of presumptions in the application of Article 102 TFEU.
Fernández began by explaining the general approach regarding the burden of proof in EU Competition Law. She emphasized that it is up to the Commission, or the National Competition Authority, to prove that what they have found actually constitutes an infringement.
Presumptions are, according to Fernández, very useful to minimize type I and II errors, but have been overused and relied on too much, to the point where they in some instances have been almost non-rebuttable. The presumption of innocence is important, and presumptions have to be rebuttable under proportionate conditions. Regarding the burden of proof, Fernández emphasized the difference between burden of proof and the standard of proof in this context. If the standard of proof is stricter, then the objective burden of proof is stricter.
Fernández continued by listing several examples of presumptions used in EU Competition Law, e.g. the liability of the parent of the wholly owned subsidiary, or the participation of a firm in an anticompetitive conduct if it attended an anticompetitive meeting. Restrictions by object has been seen as "illegal per se", and has shifted the focus of the court from the proof of infringement, to determining liability and duration of the infringement. The ECJ judgement in Cartes Bancaires has, according to Fernández, ended an expansive trend in the use of presumptions in EU Competition Law. Since then, restrictions by object are a presumption that must be applied in a restrictive way.
Two cases were then discussed to show the evolution of the use of presumptions in this context, from Hoffman-La Roche, to Intel. Regarding Hoffman-La Roche, Fernández highlighted the presumption that volume rebates are considered legal, and the presumption that exclusivity or fidelity rebates are considered illegal. Fernández then used the Intel case as a basis for discussing how the presumption that exclusivity rebates are illegal can be rebutted. Overall, Fernández is of the view that it would be positive if presumptions under Article 102 TFEU continue to be rebuttable, and showed a number of cases as examples of this in practice.
Fernández concluded by stating that it is important that presumptions are rebuttable in a proportionate fashion. Otherwise, the right of defense would be devoid of substance, the principle of innocence would be ignored, and competition enforcement would risk harm to society, in the form of welfare loss stemming from an increase in type I errors.
Kevin Coates provided comments to Fernández. Coates was widely in agreement with Fernández, and expanded on some of the points Fernández made during her presentation.
On the topic of the burden and the standard of proof, Coates emphasized that while the standard of proof may not actually vary, the evidence required might vary depending on the assessment in each case. Particularly, the amount of evidence required might vary depending on what exactly someone is trying to prove. Direct evidence of consumer harm is typically hard, and sometimes impossible, to get, especially in merger cases.
Regarding the Intel case, Coates noted two presumptions. First, that fidelity rebates are harmful, and secondly, that they are not harmful if they pass the "As Efficient Competitor"-test (AEC). Coates expressed concern regarding the latter, and did not agree with the notion that any rebate that passes the AEC-test is not harmful.
Coates continued by discussing how presumptions can be rebutted, and argued that if we are better at narrowly tailoring presumptions, they will be harder to rebut. The Intel case did not shed light on the level of rebuttal needed to rebut the aforementioned presumptions.
On the topic of efficiencies, and why we do not see much efficiency defense, Coates put forth two possible explanations. Either the authority could be unreceptive to such arguments, which would discourage companies from employing costly economists to formulate this type of defense, or efficiencies simply do not exist largely.
In conclusion, Coates reminded us that presumptions are essential, that the better the presumptions are, the harder they are going to be to rebut, and that rebuttals should go "both ways".
Presumptions in vertical mergers
The last topic of the day concerned vertical mergers (VM) and presumptions. Margaret Slade, Professor of Economics at the Vancouver School of Economics, held a presentation with the title "Presumptions in Vertical Mergers: The Role of Evidence". The start of the presentation revisited the historical assumption that VM are efficient, and how the view of VMs over the time has changed from that. Which is also shown in VM guidelines. However VM guidelines rarely contain any presumptions rather they contain likelihoods.
A presumption according to Slade says that if we can observe A we can infer B. Therefor it can work as a guidance for everyone involved and in the case of rebuttable presumptions move the burden of evidence. According to an OECD Background paper (2017), presumptions should be based on experience, economic theory and common sense. But as Slade states:
"The problem with vertical mergers are that we have little experience, there is almost no litigations. Economic theory is ambiguous and I don't know that people have a lot of common sense about a complex transaction like AT&T and Time Warner."
Therefore the presentation moves forward to discuss which conditions according to Slade, any presumptions for VM should satisfy. The first condition is that there should be a body of empirical evidence that shows that A actually leads to an anticompetitive outcome B. However the evidence shouldn't contain the sort of industry which normally doesn't raise concern. Further, Slade states, ex ante merger simulations shouldn't be used as empirical evidence as these are too simple and sensitive to assumptions, and often yield inaccurate predictions. Similarly hypothetical VM simulations should not be used, as the results from these are often ambiguous or not statistically significant or robust to alternative plausible parameter values.
Instead Slade suggests, detailed retrospective case studies concerning both single mergers and multiple mergers should be used as empirical evidence for the first condition. These studies ask the question what would have happened absent of the merger? Data from both pre and post-merger from affected and unaffected markets are used to compare the outcome between affected markets and unaffected markets. Another relevant evidence is stock market events.
The second condition presumptions should satisfy according to Slade is that the problem addressed by a VM presumption should be an inherently vertical problem, and not covered by horizontal merger policy. Thus horizontal problems in either the upstream or downstream market should be treated as just horizontal problems and not vertical ones.
From there Slade moves on to review the existing relevant empirical evidence. The presented evidence are from both different industries, uses different variables of interest and method and show both positive, negative and neutral findings concerning the effects of VM. However, Slade stresses that the classification of these finds can be disputed and that there are too few studies to present any strong evidence. Thus, Slade concludes that the current empirical evidence is too small to form any presumptions based on them. Therefor she finished the presentation urging for more retrospective VM studies and simple VM screens.