The appropriate choice of legal standards (LSs) in antitrust enforcement, that is, decision procedures or decision rules that provide the basis for how potentially anticompetitive conduct must be assessed in order to decide whether there is liability or not, has been hotly debated for many years. The debate has gained in intensity in recent years as a result of the concerns expressed by a significant number of academics and policy makers in many countries about the antitrust treatment of major platforms. Relying on recent developments in the literature on the choice of LSs, this article shows that the adoption of presumption-based LSs, that is, LSs that are closer to Per Se than to Effects-Based in developing jurisdictions, and the standards that would be optimal in developed jurisdictions, is consistent with the principles of minimising the costs of decision errors, and also leads to lower enforcement costs. The standards that would be optimal in developed jurisdictions, is consistent with the principles of minimising the costs of decision errors, and also leads to lower enforcement costs.

legal Standards, Effects-Based, Per Se, Rule of Reason

The appropriate choice of legal standards (LSs) in antitrust enforcement, that is, decision procedures or decision rules¹ that provide the basis for how potentially anticompetitive conduct must be assessed in order to decide whether there is liability or not, has been hotly debated for many years. How widely divergent the opinions have been in this debate and how dominant specific points of view have become in terms of their influence on enforcement practice has varied over time and across countries and continents. Broadly speaking, excluding hard-core horizontal agreements, for which there is broad unanimity that their treatment should rely on a strong presumption of illegality,² for most other conducts that come under antitrust scrutiny, specifically, vertical restraints and monopolisation or abuse of dominance, enforcement practices in the US (or North America) differs quite significantly from that in the EU and the EC in particular, as well as from other less mature jurisdictions,³ the latter being at present much closer to the EU than to the US.

There are a number of differences. To start with, the US tends to treat many more practices, even when undertaken by firms with significant market power, as presumptively legal (rather than illegal), that is, on average benign, than the EU that would treat these practices when undertaken by dominant firms as presumptively illegal. Next and related to this, in the US, the dominant opinion is that the primary objective in antitrust enforcement is to limit false convictions rather than false acquittals – a point of view that only recently started to be strongly criticised – and a point of view that is not held in the EC. As an outcome of these views, there is a strong tendency in the US to treat what is considered presumptively legal conducts using Per Se Legality (or Modified Per Se Legality, or Quick Look) LS (see below for a full explanation of these terms), an approach that leads to a high rate of acquittals and has been particularly criticised with respect to enforcement in high-tech digital markets. If, on the other hand, a practice is considered presumptively illegal in the US, then, at least in the last two decades, there has been a strong tendency to rely on the extensive use of economic analysis and evidence in case-specific investigations, that is, to rely much more on the rule of reason (or full effects-based approach). This is not the case in the EU,⁴ which until recently treated the practices considered presumptively illegal by relying more on object-based or on intermediate LSs rather than full effect-based,⁵ and this is certainly not true in other jurisdictions.⁶ However, based on a detailed empirical analysis of the extent and type of economic analysis applied in the assessment of abuse of dominance cases by DGCOMP (and not on the economic analysis utilised in the assessment of the appealed decisions of these cases by the EU courts), Katsoulacos and Makri (2020) show that there has been a systematic and substantial move towards effects-based in DGCOMP decisions over the last two decades.

The debate on the appropriate choice of LSs has gained in intensity in recent years as a result of concerns expressed by a significant number of academics and policy makers in many countries regarding the treatment of the major platforms. Even in the US, an increasing number of commentators argue that current antitrust doctrines, rules and antitrust enforcement “are too limited to protect competition adequately, making it needlessly difficult to stop anticompetitive conduct in digital markets” and growing market power (Baker et al., 2020).⁷ More generally, it is argued that in the US, “as a result of unsound economic theories and unsupported empirical claims about the competition effects of certain practices… antitrust rules constructed by the courts reflect a systematically skewed error cost-balance:⁸ they are too concerned to avoid chilling procompetitive conduct and the high cost of litigation, and too dismissive of the cost of failing to deter harmful conduct.” In addition, they have “encouraged overly cautious enforcement policies and overly demanding proof requirements and have discouraged government enforcers and private plaintiffs from bringing meritorious exclusionary conduct cases.”⁹

We can refer to a large number of broad considerations influencing the choice of LSs that have been the subject of an extensive literature. The most important of them are: the desire to minimise decision errors;¹⁰ the desire to minimise implementation / enforcement costs;¹¹ the deterrence effects and the legal uncertainty effects of different LSs;¹² reputational concerns of the competition authorities (CAs); substantive (or liability) standards applied.¹³ The first four considerations are encapsulated in the so-called normative or welfare maximising approach to the choice of LSs (Katsoulacos & Ulph, 2009; 2015; 2016; 2020). Reputational concerns can be important, given that when decisions are reached with different LSs, they encapsulate economic analysis of a different extent and degree of sophistication, and thus are likely to be treated differently by appeal courts, which leads to annulment rates of decisions that differ depending on the LS used – decision annulment negatively affects the reputation of CAs (Avdasheva et al., 2019; Katsoulacos, 2019b). Finally, the adoption of non-welfarist substantive standards¹⁴ leads to optimal LSs closer to Per Se (Katsoulacos, 2019a).

Here, we focus on the consideration that has had the greatest influence on thinking in this area and that has been discussed most extensively and for a longer period than all the others: the desire to minimise the welfare costs of decision errors¹⁵ (see, Easterbrook (1984), Beckner and Salop (1999), Hylton and Salinger (2001), Evans and Padilla (2005), Katoulacos and Uplh (2009), and for a very recent authoritative nontechnical review applied to exclusionary conduct, see Gavil and Salop (2020)).¹⁶ In particular, Katsoulacos and Ulph (2009), supplemented by their 2016 paper followed by Seifert (2020), Katsoulacos and Ulph (2020), and, especially, Katsoulacos and Ulph (2022)¹⁷ provide models that examine all the factors that a CA or a court must take into account and derive simple representations, in terms of conditions expressed by simple formulae, of exactly the way that these factors interact and influence the error-minimising choice of LSs.

Katsoulacos and Ulph (2022) recognise that the task usually facing CAs and courts is to choose to what extent their assessment should rely on additional distinct economic analyses and information gathering investigations that improve our ability to correctly discriminate between genuinely harmful and benign conducts of the same type. They think of the additional assessment tests as lying along a sliding scale or continuum, at the extremes of which are, on the one hand, assessments based purely on presumptions (the strict Per Se that relies only on the characterization of conduct) and, on the other hand, assessments based solely on the findings of all potential case-specific economic analyses and tests that could influence the conduct’s impact (full effects-based, or rule of reason). The idea that “the modes of antitrust analysis represent a continuum, or “sliding scale” with different fact finding requirements for different situations” was initially developed in the Antitrust Law treatise by Areeda and Hovenkamp.¹⁸ This idea’s articulation that best represents the approach of Katsoulacos and Ulph (2022) is that of Jones and Kovacic (2017). As they note, “the general progression in U.S. doctrine has been toward recognition of an analytical continuum whose boundaries are set, respectively, by categorical rules of condemnation (per se illegality) or acquittal (per se legality) and an elaborate, fact-intensive assessment of reasonableness (Rule of Reason). These poles are connected by a range of intermediate tests that seek to combine some of the clarity and economy of bright-line rules with the greater analytical accuracy that a fuller examination of evidence can produce.”¹⁹ In Katsoulacos and Ulph (2022), the continuum, with the range of intermediate tests, is described by a sequence of steps or stages, in each of which additional screens are examined using additional blocks or components of economic analysis, generating additional information, building on the information already gathered in previous steps. The question then is wouldn’t it be better to add another step of economic analysis and hence move to a LS closer to full effects-based. Thus, the objective of each stage of the information gathering and analysis process is to examine whether certain preconditions or screens that are considered necessary for demonstrating liability (welfare harm) are satisfied, such as significant extant market power, potential for exclusion, potential for consumer harm, and potential for efficiencies. Then decision error costs are compared across stages to determine the optimal (error-minimising) LS.

To evaluate the usefulness of this approach, one can, for example, consider its application for comparing whether, when assessing tying arrangements, a Modified Per Se Illegality LS, under which, in order to reach a decision, we rely on certain contextualization tests and the existence of significant market power, is preferable (in terms of decision errors) to strict Per Se Illegality, under which there is no pre-requirement of extant market power. Also whether a Disadvantaging Rivals (truncated effects-based) LS is preferable to Modified Per Se Illegality – where, under the former, significant market power is not enough for illegality, it is also required to demonstrate that rivals are likely to be excluded (in a broad sense) from the market by their conduct; or whether a full effects-based is preferable to the Disadvantaging Rivals LS. As Evans and Padilla (2005) point out, first, Strict Per Se and then, later, Modified Per Se Illegality were the standards favoured for tying by both the US and EU jurisdictions until about the end of the 1990s,²⁰ and since then it was decided to move to LSs closer to effects-based.

To give another example, the approach can be used to clarify and make precise why it makes sense to recommend that antitrust laws should be updated in order “to recognise that under some circumstances conduct that creates a risk of substantial harm should be unlawful even if the harm cannot be shown to be more likely than not.”²¹

Here, we rely on these recent developments in the literature on the choice of LSs to show that, generally, error-minimising LSs for reaching liability decisions in antitrust enforcement will be closer to Per Se than to effects-based in developing jurisdiction/countries (as we defined these countries above²²) than they will be in developed countries / jurisdictions.²³ To do this, below we provide more details about how various factors influence the error minimising choice of LSs.

As indicated above, a CA can decide that a conduct violates the competition law by undertaking one or more investigations, in each of which it successively examines a screen or precondition for identifying harm. Assuming here that the objective is to identify whether the conduct is harmful to consumer welfare, the CA can consider that this objective was satisfied, depending on the type of conduct investigated, in a number of ways, differing in terms of whether some or all screens are examined and which ones in particular. Specifically, to reach a liability decision, the CA’s investigations can cover one or more of the following stages, each of which is associated with the examination of a specific screen or precondition:

Stage 0: Initial characterisation of the conduct. This includes a detailed examination of all the relevant features of the conduct with a focus on those features that, according to case law and established economic theory, are considered most likely to influence the effects of the conduct. This conduct examination is often accompanied by a description of some basic market magnitudes, such as the level of sales, which are an input to stage 2 and can be also considered as part of that stage. We can refer to this as the “conduct characterisation screen.”

Stage 1: Detailed contextualization of the market(s)²⁴ and, most importantly, establishing that there is Significant Market Power (SMP, or Dominance). We can refer to this as the “market contextualization and SMP screen.”

Stage 2: Establishing that there is potential for significant exclusionary impact or, more generally, a competition lessening effect (by enhancing ability to exercise market power). This can be manifested through the exit of a rival or rivals or through the marginalisation of rivals (so that they cannot exploit economies of scale and/or network effects) or through the exclusion of potential entrants. We can refer to this as the “enhanced ability to exercise market power screen.”

Stage 3: Establishing that there is a potential loss for consumer welfare before accounting for efficiencies. Salop (2017) provides an extensive discussion of how, for many practices usually considered under Abuse of Dominance (AoD), exclusionary potential may or may not be associated with harm to consumer welfare.²⁵ Ideally, consumer welfare should be evaluated in terms of effects on prices, on output, on consumer choice (product variety), on quality, and on innovation.²⁶ We can refer to this as the “potential consumer welfare loss due to anticompetitive effects screen.”

Stage 4: Establishing lack of potential for significant efficiencies that can benefit consumers, specifically, establishing that efficiencies are not sufficiently significant to outweigh the anticompetitive effect of the conduct. We can refer to this as the “efficiencies and balancing screen.”

Depending on the screens examined, we can distinguish the following legal standards.

1. Strict Per Se (SPS) LS is the LS under which the liability decision relies purely on the initial characterisation of the conduct (in stage 0) and the presumption that generates an impact on well-being.
2. Modified Per Se LS (MPS LS). Under this, a liability decision relies only on the information from stages 0 and 1 and the presumption that generates an impact on well-being.
3. Truncated Effects Based I LS (TEB I LS). Under this, a liability decision relies on the information from stages 0, 1 and 2 and the presumption that generates an impact on well-being. The US Quick Look LS can be considered as an intermediate LS between MPS and TEB I with a “quick look” on the efficiency defence. ²⁷
4. Truncated Effects Based II LS (TEB II LS). Under this, a liability decision relies on the information from stages 0, 1, 2 and 3 and the presumption that generates a final impact on well-being.
5. Full Effects Based (or rule of reason) LS (FEB LS). Under this, a liability decision relies on the information from all assessment stages 0–4.

We note that all LSs (1–4) are presumption-based LSs, in the sense that they all rely on some presumption about the outcome of the subsequent assessment(s), if one or more subsequent assessments were made. Only in case (5) the liability decision relies on case-specific information from all assessment steps (0–4). So, the distinguishing characteristic of this LS is that there is no reliance on presumptions when the liability decision is made.

Clearly, for all presumption-based LSs there can be either a presumption of illegality (that is, the assumption that the conduct is on average harmful) or a presumption of legality (that is, the assumption that the conduct is on average benign). To clarify, consider stage 0: in this stage the LS is that of Strict (or, for simplicity, let’s omit the word “strict”) Per Se Illegality if just on the basis of the information collected in this stage the conduct is considered presumptively illegal; or the LS is that of Per Se Legality if in this stage the conduct is considered presumptively legal. To determine this, in stage 0, following the CA’s characterisation of the conduct as being, by virtue of its specific formal features, of a particular type,²⁸ the CA can draw on knowledge of other cases involving this type of conducts, on relevant economic theory and evidence, and the information collected from the complainants and the firm(s) involved in the specific case in order to come to a view that in fraction of such cases²⁹ are genuinely harmful to consumer welfare³⁰, with (average) harm H > 0, while the remaining fractions are genuinely benign, with (average) benefit B > 0. Given this, if the average harm across all cases is ³¹, the conduct is considered presumptively illegal (PI) if , and is considered presumptively legal (PL) if . Clearly, knowledge about the values of these parameters need not be very precise³² in the sense that what a CA actually needs to determine is just whether on average the conduct can be presumed harmful or benign. This essentially involves agencies or courts “creating presumptions³³ through experience, to guide their factual investigations and decision making” (Breckner & Salop, 1999; Gavil & Salop, 2020). CAs or courts have “initial information on the likelihood and magnitude of benefits and harms… (representing) preliminary presumptions for the entire class of similar (conducts) before gathering additional case-specific information.”

At present in many jurisdictions (including those of North America and the EU) only very rarely will a general conduct type identified in stage 0 be characterized as Presumptively Illegal without any additional contextualization of the circumstances under which the specific conduct is undertaken, the exception being that of horizontal hard-core cartels in the US.³⁴ Indeed, this is not the case for abuse of dominance practices and most vertical restraints, and the general conduct types examined under these enforcement categories are currently characterized as PL. Of course, liability decisions on such conducts are never or very rarely taken using a (Strict) Per Se LS: at least some case specific investigations are first undertaken. At a minimum, this is done in order to contextualise market conditions and to establish whether there is significant extant market power (step 1 of the investigative steps defined above). Having undertaken this step, if it is determined that the firms involved have SMP (or are dominant), the question then becomes whether the general type of conduct, when undertaken by dominant firms, is PI (on average harmful) or PL (on average benign) and what is the strength of this presumption. If in this stage the conduct is considered presumptively illegal and no further assessment is made, the LS is that of Modified Per Se Illegality; if in this stage the conduct is considered presumptively legal and no further assessment is made, the LS is that of Modified Per Se Legality. It can easily be seen that if the conduct is presumptively illegal in stage 1, the conduct will be presumptively illegal, and even more so in subsequent stages if the screens in these stages are satisfied.

Using the notation introduced above, in stage 1, the conduct will be presumptively illegal if average harm and presumptively legal if where is the probability that the conduct, when undertaken by dominant firms, is genuinely harmful.³⁵ Thus, in stage 1, the conduct will be presumptively illegal if H is large relative to B and/or is quite large, and it will be presumptively legal if H is small relative to B and/or is quite small. It is important to note that if the conduct is presumptively legal and the Modified Per Se legality LS is chosen, all conducts will be permitted and the cost of decision errors will be equal to the costs of false acquittals, H. If the conduct is presumptively illegal and the Modified Per Se Illegality LS is chosen, all conducts will be banned and the cost of decision errors will be equal to the costs of false convictions ³⁶. Thus, false convictions are large relative to false acquittals if H is small relative to B and/or is quite small.

More generally, a number of other parameters will also influence the relative value of false convictions and false acquittals. Katsoulacos and Ulph (2022) get an exact characterization of all these factors and, thus, can determine under what conditions the Easterbrook (1984) hypothesis is valid, which led to what Hovenkamp (2021) calls “an anti-enforcement bias in antitrust,” namely, that expected error costs from false convictions are higher than from false acquittals. Specifically, the smaller the probability that a screen is satisfied, which in stage 1 means the smaller the prevalence of dominant firms (or, more correctly, the higher the probability of markets’ contestability), the higher the relative value of false convictions. Also, when we can identify that harmful conduct is indeed harmful with a high degree of accuracy (that depends on the probability of identifying correctly if the screen is satisfied and on the probability of identifying that, given this, the conduct is harmful when it is), but cannot identify benign conduct with a high degree of accuracy, the higher will be the relative value of false convictions.

The perception about the value of these parameters explains why, as mentioned above, there are very significant differences in the answer to the question of whether a specific conduct should be considered presumptively illegal or legal and what is the relative value of false convictions and false acquittals across different jurisdictions. Hovenkamp (2021) criticizes, in particular, the Easterbrook (1984) assumption that B is likely to be larger than H, but we see from the list of factors just mentioned that even if this were true, there is no obvious reason to expect that the decision error costs of false convictions are higher³⁹ than those of false acquittals.

Of course, in a jurisdiction in which the dominant economic ideology places greater trust in the markets’ ability to self-correct, which tends to significantly lower the value of H (the dominant US model in the last 40 years) and puts great weight on the incentive effects of false convictions³⁷ (raising the value of B), it is much more likely to characterize a conduct as PL and to consider false convictions more costly than false acquittals than in a jurisdiction that does not place as much trust in markets’ ability to self-correct, de-emphasizes incentive effects and places trust in the governments’ ability to improve outcomes through intervention (the EU model). In the latter, it is much more likely to characterize a conduct as PI and to consider false convictions less costly than false acquittals. This is, of course, a very important consideration when explaining the different enforcement approaches in the US and the EU mentioned above. Indeed, Anu Bradford et al. (2019) attribute to this difference in economic ideology the emerging “global dominance of european competition law over american antitrust law.”

Needless to say, in developing countries, the ability of markets to self-correct will usually be even more limited than in developed economies, as entry barriers and other market failures will be higher, and the CA decisions are unlikely to have significant adverse incentive effects.³⁸ The probability that the conduct undertaken by a dominant firm is genuinely harmful, which, as we have seen, is a very significant factor in determining whether the conduct is presumptively illegal and the relative size of error costs, is likely to be much higher in developing countries. Thus, in the latter case, the presumption is much more likely to be that of illegality and false acquittals are more costly than false convictions. To illustrate, consider a potentially anticompetitive conduct, such as some predatory pricing or rebate schemes, which on their own would not be able to limit entry into markets but can do so (and will be used to do so) when some other entry barriers or market failures are present. In this case, countries/jurisdictions with low other entry barriers or market failures might not even engage actively in enforcement against such practices, while in jurisdictions with many other entry barriers, the value of enforcement will be very high. Further, in the former case, if CAs do remain involved in enforcing competition law for such conducts, they should use an effects-based assessment, but (the administratively less costly) Per Se LSs should be used in jurisdictions in which other barriers are relatively high (prevalence of contestability is low), since in the latter case the presumption of illegality of these conducts – i.e. presumption that they create harm – is much higher, with cost of false convictions more likely to be lower than the cost of false acquittals.

The following analysis of the factors that determine whether an additional assessment stage (and, hence, a movement from Per Se towards effects-based) will lower decision errors also shows that in developing countries this movement is less likely to be justified than in developed countries. There are six factors that need to be taken into account. In the discussion below we assume that the conduct type is presumptively illegal in stage 1, i.e., when conduct is undertaken by firms with significant market power.

1. The probability that the conduct for which a screen (and previous screens) is correctly identified as satisfied is genuinely not harmful. This is reduced with additional assessments and this lowers the costs of false convictions. It is impossible to say whether the reduction in the probability in developing countries is likely to be larger or smaller than in developed countries.⁴⁰
2. The probability that a screen is satisfied given that previous screens are satisfied. This probability becomes lower for additional screens, and this again lowers the costs of false convictions once additional screens are assessed. However, now the reduction in this probability is likely to be smaller or much smaller in developing countries since the smaller the percentage size of the reduction, the greater the probability that in the next screen assessment the screen will be satisfied, and this probability will be greater in developing countries. Thus, to be concrete, the fact that the probability that in the additional screen the conduct has an exclusionary impact is higher in developing countries implies that the percentage reduction of this probability relative to the probability of significant market power (the previous screen) is smaller in these countries, which tends to make the reduction in the costs of false convictions smaller.
3. Then, there are four factors (probabilities) that essentially determine the improvement in the discriminatory power of the assessment after an extra screen is examined. Specifically, the discriminatory power of the assessment depends on the following four probabilities. First, there is a probability that a harmful conduct is correctly identified as harmful and a probability that a benign conduct is correctly identified as benign once a screen is examined. It is expected that the latter two probabilities increase as more assessments (screens) are made (examined). An increase in the former probability lowers the cost of false acquittals, while an increase in the latter lowers the costs of false convictions. There are a significant number of factors that tend to make the level and the increase in these probabilities, if additional assessments are made, much smaller in developing than in developed countries. These factors include limited experience, a “short” case-law history on which to rely, limited skills and resources, and limited data in terms of availability and quality. This lack in the discriminatory power of the assessments is one of the most important factors making LSs closer to Per Se than Effects-Based error minimising choice in developing countries. Second, there are probabilities that a screen is correctly identified as holding when it holds and is correctly identified as not holding when it does not hold. It is not obvious how these probabilities will change for additional screens that may be examined. One important consideration is that additional screens are more likely to require a more sophisticated but also more ambiguous in its predictions economic analysis, and this leads to an increase in the difficulty of correctly identifying the screen. To give an example, it is likely to be less difficult to correctly identify exclusionary impact than a reduction in consumer welfare – thus, the probability of correctly identifying the second screen will be lower than the probability of correctly identifying the first. A decrease in the first probability (correctly identifying a screen as holding when it holds) increases decision error costs from false acquittals, and the decrease in this probability is likely to be more pronounced in developing countries, making the increase in these costs even larger. While the decrease of this probability tends, as a first effect, to reduce the costs from false convictions, the latter will tend to increase due to a decrease in the second probability (correctly identifying a screen as not holding when it does not hold), and this increase is likely to outweigh the first effect. Again, the fact that the decrease in the second probability is likely to be more pronounced in developing countries makes the increase in costs from false convictions in these countries even larger.

In short, with the exception of one factor, the effects of which are ambiguous, all the other factors mentioned above imply a reduced need to apply Effects-Based in developing countries than in developed countries, for reasons of minimizing errors.

The discussion above suggests that the principles of finding decision errors can be used to provide a framework for analysing the choice of legal standards in competition law enforcement in developing countries/jurisdictions and to show that the errorminimising choices in such jurisdictions are more likely to be closer to Per Se than to effects-based than in developed countries/jurisdictions.⁴¹

Over the years, I have benefitted enormously from discussions on the general issues covered in this paper with Svetlana Avdasheva, Svetlana Golovanova, Frederic Jenny, Bill Kovacic, Pierre Regibeau, Patrick Rey, Thomas Ross, Jacob Seifert and David Ulph. Of course, all responsibility for errors, omissions and ambiguities lies with me. I would like to thank two reviewers for their comments, and for their assistance in conducting research I am grateful to Vasiliki Bageri, Eleni Metsiou and Galateia Makri, who contributed in the context of the ELIDEK project “Optimal Design of Competition Policy Enforcement.”