The Network Law Review is pleased to present you with a Dynamic Competition Initiative (“DCI”) symposium. Co-sponsored by UC Berkeley, EUI, and Vrije Universiteit Amsterdam’s ALTI, the DCI seeks to develop and advance innovation-based dynamic competition theories, tools, and policy processes adapted to the nature and pace of innovation in the 21st century. The symposium features guest speakers and panelists from DCI’s first annual conference held in April 2023. This contribution is signed by Pierre Régibeau, Chief Competition Economist at the European Commission.
Disclaimer: The opinions expressed in this short paper are mine alone. They do not reflect the position of either DG Competition or the European Commission.
There is little doubt that all of the sectors where Competition Policy is commonly enforced exhibit some dynamic characteristic, be it because of investments, innovation, network effects, switching costs or, more simply, entry and exit. Some of these features have long been incorporated into our standard approaches. However, a large number of academics, policy-makers or simply observers are now insisting not only that dynamic forces are more crucial than ever to our understanding of competition but that the nature of some of those forces has been changing.1See, for example, Dynamic competition, the trendy concept in Brussels and Spain policy debates – Telefónica (telefonica.com), Gérard et al. (2018).
In this paper I review both the validity and some of the implications of such claims. I begin by distinguishing between dynamics and the more pedestrian – though important – issue of the appropriate timeframe for both merger and antitrust analysis (section 2). In section 3, I quickly review the dynamic dimensions, which, in my view, already receive considerable attention in our current practice. Section 4 discusses aspects of industry dynamics that might warrant both further academic work and increased policy scrutiny. Section 5 briefly turns to the standard of proof and the dangers of building complex theories of harms in the absence of sufficient evidence. I also discuss the kind of evidence that an increasing reliance on dynamic theories of harm is likely to require. Section 6 concludes.
2. The Relevant Time Horizon
The issue of the relevant time horizon arises mostly in merger control where the myth that the Commission essentially limits itself to an horizon of two years still appears to have significant purchase. In particular, consultants and politicians often insist that the Commission does not pay sufficient attention to the threat of entry that is supposed to discipline the parties after the merger is complete. A shining example of this debate is the Siemens-Alstom (2019) case, where both France and Germany argued that Chinese entry would prevent the parties from exploiting their increased market power.2See Competition law: blocking of Siemens-Alstom merger provokes Franco-German calls for rules changes | International Bar Association (ibanet.org).
This time of claim is ill-founded, for two reasons. Firstly, as we will see below, the disciplinary role of entry in the context of mergers is greatly exaggerated. Secondly, as a matter of practice, DG Competition does tailor its time horizon to the specifics of the industry and the specifics of the case. For example, the time horizon considered will be longer if the main theory of harm concerns investment or innovation. One also needs to consider a more significant post-merger period if the industry currently operates at full capacity and is expected to do so for a number of years. Contrary to what some parties propose, full capacity does not eliminate the potential negative effects of the merger, it just shifts them, in two main ways. Firstly, today’s decisions still affect prices for the period where today’s orders will finally be realised. Secondly, the very presence of full, lasting, capacity utilisation shifts the attention to the parties’ incentives to invest in additional capacity. As a third example, the presence of significant network effects also makes it important to evaluate the strength of these effects, and how they materialise over time.
Time Horizon and Dynamics
A sector is characterised by significant dynamics if today’s actions affect the choices available to economic agents in the future. This might be because there is some form of hysteresis (i.e. today’s actions have ongoing, irreversible long-term effects) or because of strategic modification of one’s own choices or of the choices available to other economic agents. Network externalities are an example of hysteresis. Commitments (irreversible decisions), such as physical tie-in, are the main source of strategic actions that modify the dynamic evolution of competition in a given sector.
It should then be clear that, while dealing with dynamic theories of harm requires us to also consider a sufficiently long time horizon, a substantial horizon is also called for when the sector is fast changing or experiences temporary rigidities, even if the theory of harm remains essentially static (see the capacity example above).
Standard of Proof
The only difficulty implied by longer time horizons is that we are likely to be less sure about what market conditions and behaviour is likely to be over the later part of the period. This should be acknowledge not only by applying the usual discount factors to benefits and harm arising later (using the same factor for both) but by further discounting the later part of the analysis to reflect its lower degree of accuracy.
3. Main Sources of Dynamics
3.1. Investment and Innovation
Investment in productive infrastructure or innovation are two inherently dynamic decisions since they affect the nature and degree of subsequent price/quality competition in the market as well as the ease of entry into the market. The impact of mergers or of potentially anti-competitive behaviour has long been considered in DG Competition’s practice. There has however been increased attention given to innovation theories of harm over the last ten years. The reasons for such increased emphasis are compelling. Firstly, innovation increasingly appears to be the main engine for growth in a large number of sectors. Secondly, with the emergence of new technologies and new business models, we now realise that innovation can take a multiplicity of forms, from straight R&D into new products or new production processes, to new business models, the building of ecosystem or the combination of data, for example. We must therefore invest in learning about innovation’s new guises. Finally, and most importantly, there is an increased realisation that the welfare losses from a decrease or misdirection of innovation efforts are likely to be significantly larger than the direct effect of market power on prices and service quality, which have long captured most of our attention.
Incentives (and ability) to invest are now routinely considered, be it as a basis for a theory of harm, in both mergers (e.g. Jazztel-Orange, 2015, ship-building cases) and antitrust (Czech Network-sharing case), or as part of the remedy assessment (e.g. Novelis-Aleris, 2019). Innovation has also received plenty of attention, starting as early as 2004 with Microsoft, where the effects of the deterioration of interfaces on the investment incentives of rival server OS providers were discussed. A number of recent high-profile cases, have intensified the attention given to the whole innovation process: Dow-Dupont (2017), Bayer-Monsanto (2018), Illumina-Grail (2021), or Nvidia-Arm (2022). A common characteristics of the first three cases Is that the theories of harm concerned not just the incentives to innovate but access to the input necessary to the innovation process. Other cases had already looked at technology markets where protected technical knowledge and know-how are sold. For Example QualComm – NXP looked into NXP’s incentives to stop licensing its key security technology post-merger as an element in protecting QualComm’s own market power in some types of chips’ design.3Note the significantly dynamic character of the theory of harm. Step 1 involves the change in licensing incentives, a second step concerned the innovation of NXP’s rivals, and a third step looked at the feedback effect on QualComm through the effect on the innovation incentives of its own rivals. Indeed, a fourth step involving a decision to bundle was also part of the story. In this sense, Illumina did not introduce a novel theory of harm since it simply argued that, post-merger, Illumina would have reduced incentives to make its must-have technology available to Grail’s innovation rivals. Still, a few features deserve to be noticed. However, Illumina’s technology was an essential input not only in the production process (hence the abuse would be in the technology market, affecting the product market directly) but also in the research process itself (hence the abuse is still in the technology market but, this time, affecting the innovation market). It is therefore a “bridge” between QualComm-NPX and Dow-Dupont and Bayer-Monsanto, where at least one of the theories of harm concerned horizontal overlap – and hence increased market power – in the market for innovation itself.
While we are of course still learning how to better protect the competitive innovation process, this is an angle that the Commission is beginning to have well in hand.4 See Motta-Tarantino or Régibeau-Rockett (2019), for example.
3.2 Dynamic Economies of Scale
Dynamic economies of scale mostly refer to network effects and learning by doing. Dynamic network effects are usually captures as network externalities, whereby a product’s attractiveness increases with the number (or proportion) of its buyers either directly or because greater numbers leads to a greater choice of complementary products. With learning by doing, greater levels of production today lower production costs tomorrow.
While I am not aware of any EU cases where the theory of harm invoked learning by doing, network effects often get a mention, often under the guise of the “risk of market tipping”. This “tipping” argument is most directly relevant for merger control as it can justify opposition to horizontal mergers involving what might, at first sight, look like a very moderate increment in market shares. The general idea is that, as affirm grows larger, its relevant “network” and its associated positive externality makes the company’s product more and more attractive compared to its competitors, until one reaches a point where rivals simply cannot make up for the disadvantage through greater efficiency or lower prices. If a firm already has a large market share then, in theory at least, even a small increment could alter the contestability of the market drastically.
Once such concerns are legitimate, one should also insist on a note of caution. The first issue is that, in spite of the popularity of the “tipping” story, documented examples of tipping are in fact few and far between. Take the canonical example of the choice between QWERTY and DVORAK keyboard arrangements. In Paul David’s initial account, Dvorak was a superior display that was undone by the first mover advantage of the Qwerty solution.5Paul David, 1985 and Brian W. Arthur, 1983. However, later authors6E.g. Liebowitz and Margolis, 1990. have cast serious doubt on this simple story pointing out that the presumed superiority of Dvorak was not well established and that the costs of switching between display were, at the time, likely too small to justify tipping. Another canonical example is the “format war” between VHS and Beta video cassettes. Here the network effect arises indirectly through the availability of titles for each format. In this case, Ohashi (2004) shows that network effects were indeed a significant factor in the eventual triumph of VHS. Still, he points out that, as in the Querty case, firms’ strategic decisions also played a very significant role.
Even if we turn to more casual observation, it is hard to identify many examples of markets that have tipped significantly. Clearly the market for general internet search has tipped in favour of Google. Indeed, this might be a rare case where both learning by doing and network effects (on the ad front) worked together. Facebook is also a likely candidate among social networks. Here the tipping is unlikely to have resulted from network effects since most users have a limited network of friends/acquaintances. It is more likely the result of a first mover advantage cemented by the high switching costs created by a lack of coordination (how to get 25o people to move together) combined with the fact that there are interlocking friends network (my friend will not move if other networks she is part off do not). There aren’t many more examples.
Overall then, even in a dynamic world, we should temper our enthusiasm for the “tipping” argument and the theories of harm that rely on it. If do look at tipping we should first identify the precise mechanism involved in the alleged tipping before engaging in a serious measurement exercise aimed at evaluating the magnitude of network effects, especially at the margin.
3.3 Sunk Costs, Entry and Exit
While entry and exit can affect the competitive landscape drastically, they are not necessarily part of industry “dynamics” in the sense of this note. There are in fact two types of entry (and exit): entry which will occur anyway (unconditional) in the future and entry that arises only as a response to a merger or to the behaviour of a dominant undertaking (conditional).
Unconditional entry is just part of the scenery. It is an element of both the factual with merger or behaviour and of the counterfactual without them. This does not mean that such entry is irrelevant, especially in the context of merger review as it changes the market structure of reference. For example, the projected entry of one competitor within a reasonable time from would turn a 3 to 2 merger into a de facto 4 to 3 merger, lowering the potential for concern.
Conditional entry is another kettle of fish altogether as it only occurs as a response to a decrease in competition following a merger or an abusive conduct. At first sight the argument is compelling: if the merged entity increases prices and barriers to entry are not too high then there would be entry, bringing prices down. However, as pointed out by Cardonna, Miller and Sheu (2023) and Werden and Froeb (1998), this view fails to account for the interdependency between merger-specific efficiencies, price levels, barriers to entry and actual entry. Assuming Cournot or Bertrand competition, Werden and Froeb conclude that “the entry issue can be collapsed into the efficiency issue: if a presumably profitable merger does not generate sufficient efficiencies, it cannot be expected to induce entry”. Cardonna et al. extend the analysis to include consumer welfare. This creates the following tension. On the one hand, if the merger will trigger entry then the efficiencies must be large in order to make the merger profitable. (They also might need to be even larger to make consumers better off). On the other hand, if efficiencies are high then entry is less profitable and hence less likely. The result is that the circumstances under which conditional entry would indeed discipline the merging parties sufficiently to make consumers better off are very specific, not too likely to occur….and hard to determine empirically.
If we add that opining about conditional entry means guessing about the behaviour of a number of third parties about whom neither the party nor the competition authority have very good information, I can only conclude that entry has received exaggerated attention. It is neither a crucial element of merger dynamics, nor one that we can assess reliably.
4. “New” Sources of Dynamics
We now move to newer, or neglected sources of dynamics. We begin with an old insight.
4.1 Entry Paths
Long ago, Caves and Porter (1977), pointed out that our canonical model of entry, where one or more single-product incumbents face potential entry into their market, does not fully capture the complexity of entry – and hence entry-deterrence –. Their main point is that entrant do not just sit on the sidelines waiting for their time. They come from somewhere. In other words, entrant are often not start ups but companies that are already well established in adjacent markets. They can then leverage their resources and know-how from this established market into the new one, either as a source of efficiency or through tactics such as tying.
This raises the question of what are “adjacent markets”. Caves and Porter suggest that the main link comes from common skills and resources. This is very much in line with Rumelt’s7See, for example, R.P. Rumelt (2005) view as a collection of hard to imitate resources that are deployed across various markets through a strategy of diversification. To this, one might add that adjacent markets could also be those that best enable strategies such as tying or strategic discounting.
The implication of this work for dynamic competition analysis is that, guided by our traditional market definition methodology, which relies mostly on demand-side substitution,8While we do routinely look at « supply-side substitution, our analysis tends to focus on immediate entry from closely related market without a deep look at resources or more complex entry paths. we might often be blind to the possibility of entry from different quarters. Moreover, as Caves and Porter notice, entry into market A might not come directly from market B but from D to C to B and finally to A. This has two consequences. Firstly, we might underestimate the likelihood that negative effects from a merger or anticompetitive behaviour might be mitigated by entry. This, however, does not worry me much. As we have seen, it is hard for conditional entry to exert sufficient restraint on merging parties to make the transaction both profitable and welfare improving. This severe drawback would also apply to indirect entry. Moreover, indirect entry is likely to take time, decreasing its relevance for merger analysis.
On the other hand, the relevance of entry path also suggests more complex entry-deterrence strategies where the dominant firm in A does not intervene in that market or in a substitute market but in a market that is on the natural “route” for another established firm to eventually make into the dominant’s firm main market. Under our current approach we would most likely miss such strategies. While such concerns might at first blush seem far-fetched, I would suggest that they arise quite naturally in a world where large multi-product firms tend to build ecosystems that eventually compete with each other.
4.2 Eco-systems Dynamics
The development of eco-systems raise a number of important issues for competition policy enforcement. For example how do we assess competition in a world where open eco-systems compete with (parts of) closed eco-systems as well as with stand-alone companies? What is the role for interoperability and data access in such a world?
From the point of view of dynamics, though, eco-systems only create a few of additional difficulties. The first one is very similar to the implications of entry paths. Eco-systems can be built in many different orders, starting in search to get into mobile phones, advertising, health services and so on, starting as a market place to branch out into cloud and health or moving to phones to app stores to producing one’s own apps. Does the precise order of construction of these systems matter, i.e. does it affect the eventual equilibrium configuration of the sector(s)? Can entry or rivalry be reduced by blocking access to some of the more popular system components? More concretely, should we move from a simple counterfactual where acquisition by an eco-system is simply compared to either the pre-acquisition state or to some prediction as to the stand-alone future of the un-acquired target (including as a possible seed for a rival eco-system)? Or should we extent our analysis to allow for the fact that the target might eventually be acquired by another player and that this acquisition might significantly affect the balance of power between eco-systems?
The second difficulty is how to implement meaningful access/interoperability policies? Should access only be considered for the “core” of an eco-system? The cores? Also, can we anticipate the possible dynamic reorganisation of an eco-system to respond to mandatory access to some of its parts? Competition authorities have begun to venture into this very complex terrain, with sometimes diverging conclusions and mixed success (see for example the EU and CMA investigations of the Microsoft-Activision Acquisition)
4.3. Behavioural Biases
Over the last thirty years, economists have increasingly accepted that their workhorse model of rational choice is not always appropriate to guide some policy choices. A large number departure from rational behaviour (behavioural “biases”) have been documented. Some, like patterns of behaviour in how individuals handle risk, appear to be prevalent in most economic environments. Others are more specific to particular circumstances.
It seems fair to say that competition authorities are now willing to consider well-established biases in consumer behaviour when they are likely to affect the conclusion of a case. In particular, policy-makers have received the message that, quite often, behavioural biases introduce more “rigidities” into the economic system than rational behaviour might predict. In turn, such rigidities are likely to protect entrenched incumbents and might be useful when implementing potentially anti-competitive strategies like tying or exclusive dealing.
Switching and search costs are a case in point. In the presence of such frictions, competition is inherently dynamics, as today’s competition affects after-market outcomes and the expectation of after-market outcomes affect today’s competition. This is not the place to discuss the merits of after-market theories of harm, or of theories of harms based on contractual clauses that affect consumer search. My point here is simply that the prevalence of behavioural biases significantly increases the attention that we should give to such theories of harm. Moreover, consumers; limitation in processing and obtaining information can also be used strategically to protect or enhance a dominant position.9On this count, see the discussion of “non-steering” clauses by Apple Store in both the Apple-Epics US litigation and the ongoing Apple-Spotify EU case.
5. Methodologies, Data and Standard of Proof
Suppose that we were determined to investigate dynamic theories of harm more systematically. How would we proceed? One possibility is to turn to the increasing academic literature on dynamic competition. At the theoretical level alternating moves have been studied by Maskin and Tirole (1988 a and b.), Besanko and Doraszelski (2004) specify a model of capacity dynamics leading to various patterns of market shares. However these models would be hard to apply to actual industries as the informational demands would be prohibitive. Soe authors have therefore strived to develop theoretical models of dynamic competition which are computationally tractable and might be calibrated or even estimated. For example, Doraszelski and Satterthwaite (2010) account for investment, entry and exit, Aguirrebaria and Ho (2012) estimate a dynamic model of network competition that explains hub adoption in the airline industry and might therefore be helpful in assessing airline merger.10See Aguirrebaria et al., 2021, for a recent survey. There have also been interesting extensions of more reduced form models, such as Steen and Salvanes (1999)’s version of Bresnahan model allowing for habit formation and other dynamics.
While some of these models might provide useful insights in specific cases, they are still far from providing us with a “dynamic” toolbox in which we can search for the right benchmark, as we do to examine more static theories of harm. In most cases, for now then, our ambition should be more modest and steer away from explicit modelling, calibration or estimation.
Dynamic theories of harm imply more than one step in the reasoning. We are not just saying that “the merger will lead to higher prices”, we are, for example saying that “the merger reinforces the parties’ market power in market A (step 1) which gives it the ability and incentives to tie A with B (step 2), weakening competitors in B, and making it possible to pursue foreclosure strategies based on B (step 3)”. It might therefore be useful to rely on simple “game trees” decomposing our dynamic theories of harm into a number of well-identified steps.
Let us, for example, look at a theory of harm considered (but rejected) by both the CMA and the EU in the Microsoft-Activision case. To recall, Activision is a large game developer and sellers. Its best-known product is Call of Duty (COD), which is a leader in the sub-category of “shooting” games. The dynamic theorem that I evoke starts with the belief that, while Activision itself had resisted giving access to COG for streaming, Microsoft would put significant energy in developing this type of market. The second step is the prediction that the Cloud will become important to streaming in particular and to gaming overall. The third step is that, Microsoft might make it difficult to reach this new “streaming universe” through non Microsoft OS/browsers, to protect its current Windows-based dominance. Once these steps have been explained, one should then think of the kind of evidence that could confirm or infirm the validity of each step.
Clearly, to validate such theories of harm, we need to validate each of the steps involves independently to a satisfactory standard of evidence. This is not the end of the road, however. Since the theory of harm can only hold if every step holds one must also step back and think of the joint probability that all steps are simultaneously justified. For example, if we assign a probability of 80% to the veracity of each of the three steps in the example above, the probability that the theory of harm is supported as a whole is only 51.2%.
Of course, one might argue that, in a dynamic world, it is worst testing theories of harm even if they have an overall degree of certainty that is lower than for our traditional, well-rehearsed static theories of harm, but we should do this knowingly, fully aware of the greater inherent fragility of any dynamic approach.
A final complication stems from the scarcity of appropriate data. One way to support dynamic theories of harm is to, at least, verify the past evolution of the market. For this, we need longer series on prices, outputs, capacities, innovation and entry than we usually have at our disposal. More extensive academic work on the nature and magnitude of network effects and switching costs as well as more systematic studies of entry paths would also be helpful.
Aguirregabiria, Victor and Chun-YU Ho, 2012, “A Dynamic Oligopoly Game of the US Airline Industry: Estimation and Policy Experiments”, Journal of Econometrics, 168:1, pp. 156 – 173.
Agguirregabirria, Victor, Allan Collard-Wexler and Stephen P. Ryan, 2021, “Dynamic Games in Empirical Industrial Organization”, in Kate Ho, Ali Hortaçsu and Alessandro Lizzeri, Eds, “Handbook of Industrial Organization, Vol. 4”, Elsevier.
Arthur, Brian, W., 1983, “On Competing Technologies and Historically Small Events: The Dynamics of Choice under Increasing Returns, Working Paper, International Institute for Applied Systems Analysis, Stanford University.
Besanko, D. and Doraszelski, 2004, “Capacity Dynamics and Endogenous Asymmetries in Firm Size”, Rand Journal of Economics, 35:1, pp. 23 – 49.
Cardonna, Peter, Nathan Miller and Gloria Sheu, 2023, “Merger, Entry and Consumer Welfare”, SSRN 3537135.
Caves, Richard, E. and Michael E. Porter, 1977, “From Entry Barriers to Mobility Barriers: Conjectural Decisions and Contrived Deterrence to New Competition”, Quarterly Journal of Economics, 91:2, pp. 241 – 261.
David, Paul, 1985, “CLIO and the Economics of QWERTY”, American Economic Review, 75:2, Papers $ Proceedings, pp. 332 – 337.
Doraszelski and M. Satterthwaite, 2010, “Computable Markov-perfect Industry Dynamics”, Rand Journal of Economics, 41:2, pp. 215 – 243.
Gérard, Damien, Eric Morgan De Riveryand Bernd Meyring, 2018, « Dynamic markets, dynamic competition and dynamic enforcement : the impact of the digital revolutionand globalisation on competition law enforcement in Europe”, Bruylant, Bruxelles.
Liebowitz, S.F. and S.E. Margolis, 1990, “The Fable of the Keys”, The Journal of Law & Economics, 33:1.
Maskin, E. and J. Tirole, 1988 a., “A Theory of Dynamic Oligopoly I: Overview and Quantity Competition with Large Fixed Costs”, Econometrica, 56:3, pp. 549 – 569.
Maskin, E. and Tirole, J., 1988 b., “A Theory of Dynamic Oligopoly II: Price Competition”, Econometrica, 56:3, pp. 571 – 599.
Motta, Massimo, and Emanuel Tarantino, 2021, “The effects of horizontal mergers, when firms compete in prices and investments”, International Journal of Industrial Organization, 78.
Ohashi, Hiroshi, 204, “The Role of Network Effects in the US VCR market, 1978 – 1986, Journal of Economics and Management Strategy, 12:4, pp. 447 – 494.
Régibeau, Pierre and Katharine Rockett, 2019, “Mergers and Innovation”, Antitrust Bulletin, 64:1.
Rumelt, Richard, P., 2005, “Theory, Strategy and Entrepreneurship”, Handbook of Entrepreneurship Research, Chapter 2, Springer.
Steen, F. and Salvanes, 1999, “Testing for Market Power Using a Dynamic Oligopoly Model”, International Journal of Industrial Organization, 17:2, pp. 147 – 177.
Werden, Gregory, J. and Luc M. Froeb, 1998, “The Entry-Inducing Effects of Horizontal Mergers: An Exploratory Analysis”, Journal of Industrial Economics, 46:4, pp. 525 – 543.
|Citation: Pierre Régibeau, “Competition Policy Enforcement in a Dynamic World”, Network Law Review, Summer 2023.