In this paper, which can be found here, the authors develop a model of digital platforms as attention brokers that have proprietary information about their users’ product preferences and sell targeted ad space to upstream industries. The paper argues that increased concentration among attention brokers can lead to reduced entry, higher prices and less product variety in upstream industries.
In a nutshell, the argument runs as follows. A monopolistic attention broker has an incentive to create an attention bottleneck by reducing the supply of targeted advertising. If an attention broker reduces the number of ads it sells, it will reduce the number of upstream firms that have access to consumers, thus increasing their market power. This bottleneck strategy can generate higher total profits for the upstream industry that are partly captured by the platform through higher total ad revenue. However, under standard conditions, this supply reduction hurts consumers who face less choice and higher prices. A corollary of this argument is that the right measure of platform concentration is at the level of each individual user. In a world where platforms obtain personal information and can tailor ads to individual users, what matters is the number of platforms that upstream firms can use to reach particular users. Thus, a meaningful concentration index for attention brokers must be built from individual attention data.
The paper is structured as follows:
Section 1 contains an introduction and a literature review. Section 2 introduces a model with three layers of actors: non-strategic consumers, attention brokers (digital platforms), and upstream producers.
Attention markets are central to recent competition cases and policy work. Wu in particular has introduced the idea that social media companies are first and foremost attention brokers: they capture the attention of their users and sell it to advertisers. The authors seek to determine whether attention bottlenecks may be leveraged by incumbent firms to protect their position against entrants. It is important to keep in mind that this model zeroes in on only one effect of mergers between digital platforms, regarding the creation of an attention bottleneck. There are a number of other potential positive and negative effects that this paper does not analyse.
In the authors’ model, producers sell goods to consumers and can advertise their production through digital platforms. There are multiple digital platforms. Consumers differ in their platform usage and may multi-home among platforms (usage is not affected by which ads they are shown). Platforms use competitive selling mechanisms to allocate ads to firms, which can buy non-targeted ads from traditional mass-media or targeted ads from digital platforms. The latter have two advantages: they have better knowledge of the product preferences of their users; and they can sell tailored ads targeted to users that want a particular product.
Ad-purchasing firms may be incumbents or entrants. The products of incumbents are more familiar to consumers than those of entrants. Any type of advertising can help entrants close this informational gap. The authors assume that a consumer’s surplus is higher if he is aware of the entrant’s product, but this reduces the incumbent’s revenue.
Section 3 applies the model to the sale of targeted platform ads.
The authors identify a strategic asymmetry between incumbents and entrants. Entrants want the consumer to become aware of their product and can achieve this through any of the available platforms. Incumbents, who want to make sure that entrant products do not become known to consumers, must prevent entrant access on all available platforms. This is a strategic advantage for the entrant and it means that he has to pay less to make his product known than the incumbent has to pay to keep the consumer in the dark.
Section 4 expands the model to the whole social platform industry.
In short, the probability that the consumer becomes aware of the entrant’s product increases with the number of independently owned platforms utilised by that consumer. An increase in platform concentration leads to an increase in the probability that an entrant will be kept out of one or more market segments, in turn leading to a consumer welfare loss. In other words, welfare is higher if consumers use multiple platforms because this multiplicity of platforms makes it more expensive for incumbents to keep consumers uninformed about entrant products.
Section 5 characterises the welfare effect of a merger between two existing platforms.
A merger is beneficial to the merging platforms in a given segment when it allows them to increase equilibrium revenues by restricting supply. In certain cases, a merged entity will exploit its newfound market power by coordinating sales across the two platforms: in practice, this means selling one ad rather than two. This in turn induces the incumbent to choose a monopoly strategy and keep entrants out.
The model has a number of implications for competition authorities in merger control. Firstly, a merger between attention oligopolists can hurt consumers through an anticompetitive effect on downstream product markets. In particular, a merger can make it easier for incumbents to keep out entrants, thus lowering consumer surplus. Secondly, this negative effect depends on the extent of usage overlap between the merging platforms. The effect is nil if the two platforms have no common users before the merger, and it increases in tandem with the common usage share. Thirdly, this potential negative effect of a merger is in principle measurable through platform usage. However, it is not enough to know the usage rates of the various platforms or some other aggregate form of market share; one must also know the overlap between these platforms.
A regulator who knows usage shares but not usage overlaps risks making highly inaccurate decisions. This is particularly evident for usage share values just below 50%. The authors demonstrate this by applying their model to US usage data of three major social media platforms: Facebook, Instagram, and Twitter.
Section 6 moves beyond targeted platform ads and looks at the impact of non-targeted mass media ads.
The authors show that non-targeted mass media ads are more likely to be employed by larger, mainstream industries, because the advertising fee is spread over a larger consumer base. Instead, targeted platform ads are more likely to be chosen by niche industries. Therefore, the merger effects of digital platforms are also more likely to be felt in such niche industries.
- Section 7 conducts a robustness check considering alternative ad selling specifications, while section 8 concludes.
Time/attention of users is a scarce resource controlled by platforms that can become bottlenecks for producers who want to reach users that utilise a small set of platforms. Attention bottlenecks typically occur when online platforms are concentrated, and they manage to set prices for ads that select incumbent producers to the detriment of entrants. In this sense, online platform market power is not good for innovative entrants (who need advertisements to be known by users), nor is it for users (who end up buying at a high price from incumbent producers in downstream markets). With many competing platforms, it becomes very expensive for incumbents to bid out entrants. Entrants get to be known, consumers are typically well off and marginal mergers would not matter. As online platforms become more concentrated, competition for ad scarcity works to the advantage of incumbent producers. A merger between online platforms allows them to better control this scarcity, tilting the game even more in favour of incumbents.
The crucial element in an online merger assessment (as in any other merger) is to look at the overlaps of users across platforms. If consumers multi-home, scarcity is more likely to disappear, making entry also more likely. A problematic merger is therefore one between concentrated online platforms with overlapping users. Standard metrics that ignore these overlaps and just concentrate on usage can lead to large biases. Even more so, metrics that only focus on the supply-side (market shares) are inappropriate in these markets. In contrast, existing individual-level platform usage data can be used to account for overlaps and obtain more meaningful estimates. Lastly, given the importance of advertising on traditional media for general markets, potential competition problems related to targeted ads on online platforms are more likely to be related to niche product markets for whom targeted ads are the ideal channel to reach consumers.
This paper is well beyond my ability to discuss. I would merely make a number of disclaimers. Firstly, this is only a working paper. Secondly, the model seems to rely on extreme simplification, so its practical application would likely have to be heavily caveated to ensure that the effects identified are relevant and not outweighed by other effects.