This paper, available here, builds on draft sections of a forthcoming book on tech giants and public policy. It lays down the rudiments of a descriptive theory of competition among the digital tech platforms known as “FANGs” (Facebook, Amazon, Netflix and Google).
The paper begins by addressing the debate over whether FANGs are monopolies. One school argues that they are indeed monopolies, reflecting FANG’s control of a large share of output in relevant product(s) or service market(s), high barriers to entry, lateral integration and strong network effects. Some of these works also discuss (novel) theories of harm such as reductions in privacy, labour market monopsony and distortions of the democratic process. A different current argues that traditional monopoly harms are not manifest in FANGs. To the contrary, FANGs would outperform textbook monopolies by observable metrics of prices, output, labour or innovation. In addition, the tech industry is arguably rife with examples of once dominant later irrelevant companies like AOL, MySpace or Yahoo!, inviting caution against anticipative monopoly findings.
The author concludes that textbook models of monopoly are of limited usefulness when applied to FANGS. Instead, the paper seeks to develop a theory of competition that embraces the business uncertainty typical of a number of network markets concerning the short-term market equilibrium. This requires focusing on exploitative practices in markets which have tipped (which is different from large market shares), and on practices that reduce business uncertainty by limiting competition in both core and non-core markets.
The paper is structured as follows:
It begins with a discussion of how (poorly) FANG’s match with traditional models of monopoly.
Under economic theory, a monopolist is a dictator with absolute powers. No rival, entrant, input seller or buyer can influence its decisions. In order to maximise its profits, monopolies are expected to grow output and lower prices up to the level where marginal revenue (“MR”) equals marginal cost (“MC”). Put differently, the monopolist decides – without constraints – to produce an extra quantity of output if (and only if) this yields a revenue greater than the costs incurred to produce an additional marginal unit.
Under this model, monopolies are constrained solely by falling revenue and increasing (or constant) costs – i.e. marginal revenue decreases as quantity increases, because the monopolist is confronted with a falling demand curve for his product. To sell more output, the monopolist must lower the price to get people to buy more units of output. A logical implication is that the monopolist’s marginal profit will decrease up to the point of equality between marginal revenue and marginal cost. From a social welfare standpoint, the profit maximising equilibrium level of output leads to a monopoly price level greater than marginal cost, which in turn leads to reduced output. The monopoly equilibrium also imposes a loss on society because some customers ready to pay a price lower than the monopoly price but higher than marginal costs are not served. This allocative inefficiency of the monopoly is coupled with a variety of other harms stemming from insulation from competition. In the familiar parade of horribles come cost inefficiency, low innovation and rent seeking.
FANG’s look like monopolies. Each FANG holds a large share of output in a market where entry is limited. The fact that high prices, low output and reduced innovation are not manifest in FANG is irrelevant for these purposes. However, by studying a dataset covering FANG’s Securities and Exchange Commission (SEC) 10K filings, the author finds that these companies seem to operate in a manner inconsistent with the textbook monopoly model. Under this model, we should observe decreasing marginal revenues and prices at firm level as output increases. While we have no readily available quantitative measure of output that entitles us to compute the revenues, costs and profits of FANGS on a marginal basis, there are some useful available metrics out there: Amazon’s number of active customer accounts, Facebook’s number of daily active users, Netflix’s streaming subscribers and Google’s core search queries. By crossing SEC data and this data, one can estimate the incremental evolution of each FANG’s average revenue, costs and profits on each customer account, user, subscriber or core search query as output grew year-on-year. The results are that not only have the margins of all FANGs been rising, but marginal price has been increasing too.
FANGs thus seem to violate two essential conditions of the textbook monopoly model. This raises an intriguing question: if both marginal revenue and marginal price increase, is a short-term equilibrium even possible? A logical implication of a simultaneous increase in marginal revenue and marginal price is that marginal cost will remain close to constant (as is the case with Facebook), decreases or rises more slowly than marginal revenue (as with Google). In all three cases, this means that there is no convergence between marginal revenue and marginal cost.
In other words, FANGs are not behaving as short-term profit-maximising monopolists. This has two logical implications. First, profit maximisation by FANG firms – inasmuch as this is a realistic proposition –must take place on a long-term perspective. Second, FANG firms’ short-term profit maximisation approach must be about something else. One possibility – supported by increasing marginal revenue – is that FANGs face an upward sloping or shifting demand curve in the short term. This suggests that FANGs’ short-term goal should be solely to grow output. This is consistent with anecdotal observations of early loss-making by FANGs such as Amazon, and with arguments to the effect that the marginal cost is zero in digital markets.
All this cautions against comparing FANG with textbook monopolies. Both their rational and predicted response to the specificities of the demand curve suggests absence of monopoly harms. In plain words, FANGs do not do the bad things we normally expect monopolists to do. This, in turn, leaves also open a critical question: if FANG do not act like monopolies in the short term, how do they compete?
A second section focuses on how FANGs compete.
Since the 1970s, economics has studied firm behaviour in markets with “network effects”, where users’ willingness to pay increases with quantity demanded. In such markets, each user’s marginal benefit is based not only on the value of the service’s functional attributes, but also on the number of (expected) users of the network. A consequence of this is that the demand curve may contain an upward-sloping segment in network effects markets.
Paradoxically, business seems more risky when demand is increasing than when it is falling, because there is no clear market equilibrium. Economists use diverse concepts to refer to the uncertain environment of firms that operate with an upward slopping or shifting demand curve: “disequilibrium”, “out of equilibrium”, “multiplicity of equilibria”, or “unstable equilibria”. A direct consequence of the existence of multiple equilibria is unpredictability. Both the firm, its competitors, and external observers face uncertainty as to how the market will behave. Moreover, prices lose relevance, for the market can sustain different network sizes for the same price.
As much as non-equilibrium seems to generate uncertainty, network effects markets display another property that seems, at least facially, to work in reverse – “critical mass”. In short, firms that reach a critical mass of users can expect to arrive at a high participation equilibrium. By contrast, firms that do not reach critical mass of users can expect their network collapse. This is also known as ‘tipping effect’. The problem for businesses is that there is no critical mass threshold or tipping point corresponding to a set number of users against which firms can assess network adoption performance. Rather, there is a range of numbers that define a zone in which tipping is likely, even if this can be extremely difficult to compute. Another difficulty is that the critical range/tipping zone is not fixed, since market potential keeps changing as firms grow.
In any event, companies can try to make markets tip to their advantage, and economists have developed a number of insights regarding how this may occur and how such markets operate. This is a complex topic to which the author devotes some attention. One particular insight is that it is markets, not firms, that tip, and that investments into critical mass and tipping are imperfectly appropriable. In other words, competitors can free ride on rivals’ network specific investment, which can work both to the benefit of late entrants and incumbents. Another insight is that network effects can lead to companies being displaced very quickly in markets that have already tipped, if some conditions occur.
One consequence of the absence of equilibrium is that firms behave under uncertainty. This is supported by FANGs’ disclosures in their reports, as well as by industry-specific fact patterns. FANGs routinely report fear of disruption, which is reflected in investments that led to a significant share of their profits (and therefore of their ability to grow) coming from accidental discoveries: Amazon discovered the lucrative cloud services’ market accidentally. and Netflix did not initially believe in online streaming. Digital industries also face a universe of risk factors, such as uncertainty as to what users want (which explains the importance and prevalence of venture capital in this sphere), particularly below critical mass.
On the other hand, when a network has overcome the critical mass constraint, there is less uncertainty. Even then, such networks still face significant risks. Network markets in non-equilibrium dynamics offer the highest profit opportunities, which creates significant incentives for attempted entry. Complements to existing networks are also a source of uncertainty, in that they can make monetisation more complex, allow competition at different levels of a market and can, on occasion, completely overturn industry structures. Last, firms in non-equilibrium markets may not only be victims but also active agents of uncertainty. Because third-party entry dissipates the likelihood of long-term equilibrium profits, incumbents in network markets are incentivised to look for emerging or future non-equilibrium markets – which leads
FANGs to compete with one another in non-core markets on occasion. For example, nowadays we see many FANGs entering online payments, entertainment or wearable devices like connected glasses or watches.
A third section draws the implications for competition and public policy.
From a public policy perspective, FANG markets represent a challenge. In a non-equilibrium environment, the firm is more fragile, which may be procompetitive. FANGs display patterns of behaviour more consistent with well-functioning competitive markets than with monopoly power that leads to consumer harm. Believers in the benefits of FANGs, and FANG firms themselves, often point to high-level data points as diverse as investments in R&D, slow monetisation strategies and cost reduction plans. At the same time, because the firm is more fragile, it may have incentives to take steps to remove uncertainty, which may be anticompetitive. Several patterns of observed behaviour are more consistent with the textbook monopoly model, like fee increases, ad cluttering or planned obsolescence. Besides these polar examples, non-equilibrium strategies are difficult to categorize as pro or anticompetitive. Think about cross platform integration of complements through M&A (e.g. Facebook’s acquisition of Instagram), preferential treatment (e.g. Google’s integration of maps on its search engine and mobile OS Android), bundling (e.g. Netflix’s bundle of DVD and streaming subscriptions), imitation (e.g. Amazon’s cloning of merchants’ products) or exclusive dealing (e.g. app stores’ bans on third party distribution). In a non-equilibrium environment, cross platform integration is a well-accepted strategy to grow network effects. At the same time, cross platform integration reduces reversibility, increases switching costs and exacerbates lock-in.
Overall, characterising firms’ strategies in non-equilibrium markets probably requires adjustments of competition law and policy’s frameworks. A competition policy framework committed to consumer welfare should place priority on equilibrium markets that have tipped and where exploitative conduct is more likely. Early theorists of network effects have long supported a policy distinction between markets that have tipped, and others that have not. This has implications concerning the identification of market power and anticompetitive conduct. Under a structural analysis, many firms in network effects markets are likely to be deemed dominant on the ground that they control a large share of output, even though their environment is one of non-equilibrium where uncertainty is highest. The problem here is that the conventional definition of dominance makes no sense in a non-equilibrium market with uncertainty – where companies are utterly dependent on their competitors, customers and consumers. The concept of dominance should appropriately distinguish the properties of procompetitive and anticompetitive non-equilibrium markets by focusing on the competitive pressure that companies are subject to. A potential metric for this is whether a firm hustles to move away from its current position. A range of metrics may be considered, such as entry and exit choices, R&D expenditures and intensity, rate of product introduction, change in business methods and strategy, or capital allocation choices.
Regarding anticompetitive conduct, and acknowledging the ambiguity of much FANG conduct, the author suggests that one start from the fact that uncertainty is a threat to firms but good for society. In short, competition enforcement is justified if firm conduct in non-equilibrium markets reduces uncertainty. Thus, strategies that implement anti-connection, imitation and rent seeking strategies should be investigated. Once tipping has occurred and uncertainty subsided, competition law can also focus on catching rapacious monetisation strategies that exploit locked in users, including ad cluttering, fee introduction in zero price markets or even planned obsolescence.
This approach also has implications for merger control. The critique of killer mergers relies on an implicit assumption that the buyer has correctly anticipated the “but for” world and eliminated its competitors. Given the inherent uncertainty in which start ups operate, however, little can be predicted of their odds of success. Further –and, to my mind, more importantly – the author argues that this assumption mischaracterises the possible motives for M&A in non-equilibrium markets – most of which are pro-competitive (e.g. change the overall market structure; provide the buyer with a key, non-reproducible/defensible asset; or dramatically change some aspect of your business (cost structure, distribution channel, etc.)). A possible approach is to develop a standard of review that focuses on: (i) whether the target is a competitive force of disruption in a relevant market or in adjacent, neighbouring, or complementary markets; and (ii) whether the acquiring firm’s incentives are to discard the product or service. As part of this second test, competition law and policy should consider alternative scenarios, including “raiser” – when the purported merged entity’s strategy is to grow the target’s product – and “catch-up” mergers – where the purported merged entity’s strategy is to increase its ability to compete with rival companies.
It is rare to read something genuinely original in this field. I cannot vouch that this is the case here, but this paper introduced me to a new way of thinking about these matters. Whether the proposed approach is useful is, of course, another matter. While I must emphasise that the arguments in this paper will be better developed in a forthcoming book, there are some points that left me unconvinced or that I felt required additional development.
For example, I understand that these are markets with network effects where the companies’ imperative is to grow – this is something that is clear from the business literature and the way venture capitalists operate. However, at some point the market matures – after all, network industries are quite common outside the digital real, and quite stable – and then uncertainty diminishes significantly. There are good reasons why the business literature, venture capitalists and these companies repeatedly refer to ‘moats’ or ‘defensible positions’, and focus on monetisation once they achieve such a status. If anything, these companies would seem to operate is an environment of reduced business uncertainty when compared to traditional businesses – in effect, the level of their dominance is exactly why we even have acronyms like FANG or GAFA. It thus feels a bit odd to use these companies as an example of why we should carefully consider whether a market has tipped, and of why we should carefully consider the implications of uncertainty in industries where demand curves are upward-sloping. This angle might make sense when discussing competition between these companies – something the author does quite well with his concept of ‘moligopolies’ – but it strikes me as being out of place here.
I would also like better to understand the practical implications of the author’s proposals. For example, as regards dominance I am not aware of investigations against digital giants that have not sought to establish that a dominant position has been held for years – instead, the durability of market shares is an important criterion to establish dominance. I wonder whether there are other criteria that one can use to identify dominance to the requisite standard in these markets. Finally, as regards theories of harm, I would really have enjoyed a more detailed discussion of what they may entail, and of how to determine whether a company is operating under a ‘sufficient’ level of uncertainty.
Overall, and as should be clear by the extent of this review I highly recommend this paper – and I look forward to the publication of the book where these arguments will be fully developed.