Digital network
VoxEU Column Competition Policy

‘Ecosystem’ theories of harm in digital mergers: New insights from network economics, part 2

Regulators have recently been experimenting with ‘ecosystem theories of harm’ to capture the idea that the collection of a conglomerate firm’s existing assets and capabilities may matter when assessing the acquisition of an unrelated asset. This second in a series of two columns describes how network economics can help develop formal models to understand systemic effects in settings where agents act strategically – thus potentially enriching the industrial organisation approach which has long been the foundation of antitrust analysis.

In a previous column (Caffarra et al. 2023), we discussed how new approaches are needed to articulate how an existing constellation of assets and capabilities may matter to the analysis of an acquisition by a large digital conglomerate. In this column, we discuss how combining network economics with industrial organisation (IO) approaches may provide a promising avenue. Network economics is attracting attention from antitrust economists and enforcers interested in “ecosystem issues’ beyond what IO has had to offer (at least to date).  The ‘network economics’ perspective thinks of a firm as operating within two relevant networks. One is the network of capabilities where the nodes are the set of capabilities and the set of firms, and a connection between a firm and a capability indicates that the firm has that capability (Chen et al. 2021, Boa et al. 2023). The other is the network of products the company operates, and into which the capabilities are projected. A connection here captures substitutability/complementarity of products from the perspective of the firm operating them (Galeotti et al. 2020). These connections might be derived from, for example, vertical relations, or network effects across products (like software and hardware, or like multiple sides in a multi-sided platform).

The current analytical interest is to see how traditional economic models of oligopolistic competition can be integrated within the network approach, with the aim of (a) understanding the effect of these networks on market outcomes while going beyond the normal approach of defining markets and focusing attention within those markets; and (b) describing these effects still in a language (i.e., the IO language) that is familiar to practitioners (Elliott and Galeotti 2019).

In this framework, a conglomerate acquisition can change the network of capabilities – and in particular, the capabilities available to the buyer. This can in turn create competitive advantages for the buyer, who can become stronger in markets where it was already strong, but may also lead to the introduction and development of innovative products (including products that neither the acquirer nor the target was initially present in, and others that might not yet exist). The extent to which such ‘market disruptions’ will occur depends on the network of products. 

In this conceptual approach, changes in the network of capabilities resulting from an acquisition may lead to changes in the level of competition across markets.  Thus the acquisition of a firm with complementary capabilities to the acquirer’s may create some competitive advantage.  The question is whether this approach can help to identify in practice a chain of capability spillovers creating competitive advantages across product markets, and importantly establish when this chain reaction can lead to consumer harm. While the potential impact of bringing together complementary capabilities is articulated qualitatively in familiar ways (strengthening the acquirer’s position in some markets, increasing barriers to entry in the presence of switching costs and/or network externalities, entering new markets with strong first-mover advantage and creating an early lead, ‘hoarding’ capabilities/assets required to compete and denying then to current/potential competitor, weakening the innovation incentives of rivals), the theory is plugging a hole in developing models which can generate these effects from the combination of capabilities. 

But this is still theory. What is then the path to implementable analyses? Can we go from as-yet-tentative theoretical analyses to something that stands a chance of being implementable?  This requires developing an approach to address  at a minimum three broad questions: how to think of the relevant ‘capabilities’ and how to ‘capture’ them; how to measure them and their relationship – at a basic level, the degrees of complementarity and substitutability; and how to at least tentatively identify the threshold at which their combination can become problematic. The short answer is we don’t have answers yet. 

IO theory and network theory have just started to ’talk’ to each other in this space. Only very recently are we seeing incipient attempts to incorporate the richness of networks into IO – and even then we are only talking about taking very basic IO models (often hyper-simplified with symmetric players, a few substitute products, or a simple vertical production structure) and trying to introduce richness in the product space (potentially leading to strategic competition across multi-product firms) and the production space (embedding firm’s behaviour and choice in the supply chain, and taking into account intangible assets like data/ capabilities).  Unsurprisingly, there is no practical guide to implementation when even the theory is still not well formulated and not well understood for the purpose.

The starting point for implementation would have to be a way of identifying the key capabilities of the target: for instance, proprietary software, brand, customer-base, data. Then one would need to understand how they fit and compare with the existing capabilities of the buyer: again technologies, algorithms to predict consumer preferences, data, customer base, technologies, also brand, operating systems, cloud and computing capabilities, additional data from other sources/devices. In particular, one would want to establish how they relate to one another. At the most basic level, are they substitutes or complements? Or a bit of both? Think for example of data assets: is the target bringing additional data assets that the buyer does not currently have, or is there at least in part an overlap between the ‘signals’ that can be extracted from their respective data assets? Are these correlated in some way?  In particular, how are the data collected by the target valuable for additional uses by the buyer? What is their incremental value? Do the data lend themselves to be used together with other complementary data to power up the buyer’s businesses?  Note: we are not interested in just complementarities between data assets owned respectively by the parties; but in the degree of potential complementarity between data assets acquired, and the broader set of assets the buyer owns – e.g. devices connected to the internet, cloud, etc.

Then one would ideally need to establish how the new network of capabilities post-acquisition would affect the competitive advantages of the acquirer across product markets. This analysis would try to anticipate how these additional assets could be leveraged across different services/markets (including markets that the parties are not currently present in).  Note that an acquisition of new capabilities which do not perform similar functions to the acquirer’s capabilities, and/or are used to serve markets unrelated to the acquirer’s products, does not preclude in principle that substantial possible changes might be triggered in the buyer’s competitive advantage across different markets.  This is hard to do though, at least if one wants to formalise the connections between these capabilities to create some prediction of how combining different assets would impact firms’ competitive places in the market.

One would also want to understand how easily competitors can obtain alternative capabilities/assets to those being acquired.  Think for example of a deal that would confer to the merged entity control over ‘critical flows’ of data, where those flows result from interactions between networks of users (e.g. developers and consumers).  There is then a threat to competition if others can be deprived of some critical nexus on that flow of data.

Again, how could one go about this in practice?

One consideration is that with advances in machine-learning (ML) and AI, it will be increasingly possible for algorithms to be written to extract a mapping of capabilities from text analysis of a variety of data sources (Hoberg and Phillips 2010, 2016, 2018).  These could include company filings and analyst reports, but also the large volumes of internal documents that agencies elicit from merging parties when examining a deal. Patents and formal IP might also be one avenue for gathering information on capabilities (and tend to be publicly available), though aspects such as ‘critical data flows’ between ‘networks of users’ will not be captured in a patent or formal IP document.

A key question is whether we will find ways (that are ideally scalable), to systematically identify and analyse data assets.  Data assets are generally recognized as a key ‘ingredient’ in the creation and expansion of market power, and there is a developing literature studying how data-rich firms acquire and preserve power, and how data affects risk, firm size and the composition of the goods firms produce, all of which affect markups (e.g. Eeckhout and Veldkamp 2022). One immediate objective would then be to formalise how those ‘critical data flows’ operate, then work toward how we capture those relationships as graph representations, maybe through expert analysis of product descriptions (automated by algorithm). One could perhaps start with a comprehensive list of consumer characteristics, and then map data assets onto these characteristics on the basis of how a data set is able to predict the different consumers’ traits. Building up a picture of firms through the lens of their data assets would tell us something about which data sources are complements (i.e. informative about different characteristics) and which are substitutes (informative about the same characteristics). Then a merger that allows a firm to acquire substitute dataset on characteristics that its competitors do not have information about could entrench its competitive advantage (i.e. be an application of asset hoarding), while a merger that allows a firm to acquire a complementary dataset could – especially if there are no good substitute datasets available to competitors for the one being acquired – increase the acquirer’s strength across a variety of markets. 

Google/Fitbit is an obvious example here. The European Commission did not recognise this concern, but the main theory of harm in the public debate was around Google already controlling a large data firehose and multiple other assets, now also acquiring a dataset with specific characteristics (vital signs, etc.) that could be combined with other assets and leveraged into new products and services. And as mentioned the relevant complementarities are not just among data assets, they may also be between data and other capabilities/assets: devices that collect data, storage of data, advertising tools. 

What would be a limiting principle? We would also obviously need a way of quantifying the threshold above which a combination causes competitive concerns versus where it does not. This is even harder. Perhaps one way to go about it as a first approximation would be to worry about circumstances where the new data assets are absorbed into a company with an existing significant position in some core markets that are related – e.g. digital advertising, or intermediation. That is, we could say again as a first approximation: “If the merged entity has to start with a strong position in a particular product/ service/ set of capabilities, then we would regard as potentially problematic any further deal that gave the buyer access to an incrementally valuable data flow which could be used to power its presence in the same or other markets”. Of course this is vague, but it is a start. To use the Google/Fitbit example again, there the incremental value of the Fitbit data was indeed not so much in digital advertising, and indeed the issue was not Google using the data to do more digital advertising; but the potential use of the new data – combined with existing assets – to entrench its position as an intermediary connecting people to products and services while allowing it to establish a position of potential power in new markets that use health data.

More work on both theory and implementation is obviously needed, but this is a relevant and highly topical area for research.

References

Boa, I, M Elliott, and D Foster (20230, "A capability approach to merger review”, mimeo.

Caffarra, C, M Elliott and Andrea Galeotti (2023),”'Ecosystem’ theories of harm in digital mergers: New insights from network economics, part 1”, VoxEU.org, 5 June.

Chen, J, M Elliott and A Koh (2021), "Capability accumulation and conglomeratization in the information age", Available at SSRN 2753566.

Eeckhout, J and L Veldkamp (2022), “Data and Market Power”, NBER Working Paper 30022.

Elliott, M and A Galeotti (2019), “The role of networks in antitrust investigations”, Oxford Review of Economic Policy 35(4): 614-637.

Galeotti, A, B Golub, and S Goyal (2020), "Targeting interventions in networks", Econometrica 88(6): 2445-2471.

Hoberg, G and G Phillips (2010), “Product market synergies and competition in mergers and acquisitions: A text-based analysis,” The Review of Financial Studies 23(10): 3773–3811.

Hoberg, G and G Phillips (2016), “Text-based network industries and endogenous product differentiation”, Journal of Political Economy 124(5): 1423–1465.

Hoberg, G and G Phillips (2018), “Conglomerate industry choice and product language”, Management Science 64(8): 3735–3755.

1,237 Reads