In the last few months, antitrust authorities around the world have been focusing their attention on hi-tech superstar firms. Examples include the recent EU report on Competition Policy for the digital era (Crémer et al. 2019) the UK launch of an “Online platforms and digital advertising market study” and, in the US, the recent announcement by several state attorneys of separate antitrust probes of large tech companies such as Alphabet’s Google and Facebook.1 Following the recent cases in the EU where the DG Comp heavily fined Google’s parent company for multiple abuses of its dominant position in online markets, the ongoing investigations will seek to detect and correct possible abuses committed by the tech giants.
For the so-called FAANGs (Facebook, Amazon, Apple, Netflix and Google), the sale of online ad space is (or is becoming) a key source of revenues. The situation of Google is emblematic: in 2018, advertising made up 85% of its annual revenue, which totalled $136.22 billion. Interestingly, this amount is the sum of millions and millions of small transactions originating from auctions through which the search engine both allocates the available ad slots to the advertisers (i.e. bidders) and determines their payments. Two landmark studies by Edelman et al. (2007) and Varian (2007) provided the game-theoretic analysis of these auctions, which are known as generalised second price (GSP) auctions.2
Intermediaries and coordinated bids in sponsored search auctions
The market for online space is rapidly changing, however, and specialised intermediaries – the digital era equivalent of marketing agencies – have emerged to support advertisers in their bidding activities. As shown by Decarolis et al. (2019), when the same intermediary concentrates demand from multiple advertisers, auctions might become a flawed selling system leading to large revenue losses for the seller. In particular, by extending the game-theoretical analysis of GSP auctions to the case in which a subset of bidders delegates their bid to a shared intermediary, they show how weak the GSP auction is to this form of bid coordination. Such forms of tacit price collusion can take place through the pricing algorithms handled by the intermediary, posing significant and non-trivial questions about their legal and economic impacts, as discussed in related work by Calvano et al. (2019).
Consider a situation in which there is one ad slot up for sale and three advertisers interested in it (Figure 1), either with three different intermediaries each handling a different advertiser (panel A) or with two intermediaries, one of which handles the two advertisers with the highest value for the slot (panel B). In the former case, the original bids are passed on by the intermediary to the search engine (q wins and pays 3). In the latter case, however, when two advertisers bid through the same intermediary, the losing bid among the coalition members is suppressed – or substantially reduced – leaving the winner unchanged but abating the winning bid (q wins and pays 1). With multiple slots for sale the mathematics becomes more complex, but the intuition remains identical. The key idea is that in the GSP auction all bids are interlinked in equilibrium. Therefore, reducing even just a single bid triggers a chain reaction in which also advertisers outside the coalition reduce their own bid.3
From theory to data: Quantifying countervailing buyer power
But is this phenomenon of coordinated bidding a mere theoretical curiosity or is it something that truly affects online ad auctions? And in the latter is true, is the bid reduction large enough to offset all the potential benefits that intermediated bidding might produce for the search engines, in terms of both bringing in new advertisers and making them bid on vast amounts of (carefully selected) keywords? These are the empirical questions that we address in our new study (Decarolis and Rovigatti 2019) through a novel dataset where we link 6,000 large advertisers to both the keywords they bid on in the Google search auctions and their intermediary, if any.
Using data on nearly 40 million Google keyword auctions held in the US market between 2014 and 2017, we first apply machine learning algorithms to group the keywords into homogeneous thematic groups. This is a crucial step to evaluate the effects of intermediaries on the search engine revenues. Indeed, sophisticated intermediaries might ease the price competition not just by lowering bids for a given keyword auction, but also by splitting similar keywords among different clients. The algorithm developed pools together groups of keywords mimicking what antitrust authorities do to determine the relevant markets on the basis of both demand and supply information.
Provided with these markets, we develop an empirical strategy to quantify whether increases in intermediaries’ concentration in a market cause a decline in the revenues of the search engine.4 We find that greater concentration of intermediaries lowers the search engine’s revenues. Under our baseline model, a change in the Herfindahl-Hirschman Index (HHI) of 200 points – the threshold typically used to identify mergers likely to enhance market power – leads to a decrease of 8.04% in revenues. The decline in revenues appears to be driven by a decline in the price of the chosen keywords – demand concentration is negatively associated with the average cost-per-click, but not with the number of keywords or their associated search volumes.
Implications for competition policy
These results call for a third way in competition policy between the two polar extremes of ‘do nothing’ and actively stepping in to fine and, possibly, break up the tech giants. This alternative is based on a detailed understanding of the functioning of such complex markets, with the aim of monitoring, and possibly blocking, those specific actions that players such as Google are currently undertaking to limit the ability of intermediaries to lower prices. This is the case, for instance, of the recent ‘disintermediation’ campaigns held by Google, aimed at steering advertisers away from third-party intermediaries and towards adopting Google’s own intermediary bidding service.
Many important questions remain open. First of all, is the activity of bid coordination legitimate, or could it be a form of buying cartels that are considered a violation of competition under certain jurisdictions? Furthermore, is the activity of the intermediaries placing a disproportionate burden on those smaller ad sellers that, in contrast to Google, have limited means to react? Our findings leave these questions open and, more generally, do not provide a definite answer about whether ‘the market is fixing the market’. In fact, what our findings cannot tell is whether the revenue lost by the search engine is benefiting the intermediaries or the advertisers (and, indirectly, consumers). The high degree of concentration among intermediaries makes it plausible to assume that they are able to capture at least a part of the gains. It is certainly noteworthy, nonetheless, that a mechanism that has been at play in the past in many other markets – the emergence of countervailing buyer power – exists and is quantitatively important in the online ad market, too.
Calvano, E, G Calzolari, V Denicolò, and S Pastorello (2018), “Algorithmic Pricing and Collusion: What Implications for Competition Policy?”, CEPR Discussion Paper 13405.
Crémer, J, Y-A de Montjoye and H Schweitzer (2019), Competition policy for the digital era, European Commission.
L. Dafny, M. Duggan, and S. Ramanarayanan (2012) “Paying a premium on your premium? Consolidation in the US health insurance industry.” American Economic Review, 102(2), 1161-85.
Decarolis, F, M Goldmanis, and A Penta (2019), “Marketing Agencies and Collusive Bidding in Online Ad Auctions”, Management Science, forthcoming.
Decarolis, F and G Rovigatti (2019), “From Mad Men to Maths Men: Concentration and Buyer Power in Online Advertising”, CEPR Discussion Papers 13897.
Edelman, B, M Ostrovsky, and M Schwarz (2007), “Internet Advertising and the Generalized Second-Price Auction: Selling Billions of Dollars’ Worth of Keywords”, American Economic Review 97(1): 242-259.
Varian, H (2007) “Position auctions”, International Journal of Industrial Organization 25(6): 1163-1178.
 Whenever a user of Google’s search engine clicks on one of the links marked as “ad” – which are often located at the very top of the search outcome page – this action triggers a payment to the search engine. In fact, not all links appearing on the screen after running a keyword query are the same: in the online market jargon, organic links are those that appear in the outcome page because the search algorithm has established that they are a good match for the user’s query (and are ranked by their relevance), while sponsored links are those listed on the page because advertisers had placed a bid to appear on searches involving that keyword (and are ranked proportionally to their bid). The bid is never just for the keyword, but for a more complex combination of elements (location, time, etc.) that, together with the keyword, allow a very precise targeting of the ad to the consumer. This is the essential source of value of the online ad and their perceived superiority for advertisers relative to ad placed on traditional media, such as TV or newspapers.
 They also show that for this reason, the alternative auction mechanism adopted to sell ad space by Facebook (called Vickrey–Clarke–Groves, or VCG) outperforms the GSP auction in the presence of bid coordination. Under its different payment system, there is no chain reaction in the VCG auction, so that all bidders outside the common intermediary have no incentive to revise their bid in response to the bids placed by the intermediary.
 The method is an instrumental variable strategy along the lines of Dafny et al. (2012), where the instrument builds on the changes over time in the proprietary structure of the intermediaries.