COMPETITION IN ONLINE MARKETS

9.2.1 Key Characteristics of Online Markets

The intensity of competition in Internet markets is often (but not always) influenced by direct and indirect network effects and switching costs (e.g., Evans and Schmalensee, 2007; Alexandrov et al., 2011).

In contrast, indirect network effects arise if the number of users on one side of the market attracts more users on the other market side. Hence, users on one side of the market do not directly benefit from an increase in the number of users on their market side, but only indirectly, as an increase in users on their market side attracts more poten­tial transaction partners on the other market side.

Apart from eBay and Amazon Marketplace, prominent online platforms that exhibit indirect network effects are Uber, Lyft and similar ride-sharing platforms; Airbnb, Expedia, Booking.com and other travel-related booking platforms; Google, Bing, and other search engines; Craigslist, file-sharing networks and many other platforms and applications.

From a competition policy point of view, it is important to note that network effects often require large platform sizes to achieve efficient utilization of the platform. Hence, high concentration levels cannot simply be interpreted in the same manner as in conven­tional markets without network effects (e.g., Wright, 2004; Alexandrov et al., 2011; Evans and Schmalensee, 2015). In fact, the existence of one large market place may often be efficient, as it helps to reduce search costs for potential trading partners compared to a situation with a larger number of smaller market places.

From a business perspective, two-sided markets pose the challenge that it is not suf­ficient for the platform operator to convince only users of one market side to join the platform, as there is an interrelationship between the user groups on both market sides. Neither the buyer side nor the seller side of the market can be attracted to join the plat­form if the other side of the market is not sufficiently large. This is a realization of the well-known chicken-and-egg problem, where both sides of the market affect each other and no side can emerge without the other (Caillaud and Jullien, 2003). Consequently one side of the market is often ‘subsidized’ by the other (Wright, 2004; Parker and Van Alstyne, 2005). Products such as the Acrobat Reader, Microsoft’s MediaPlayer, or the RealPlayer are available free of charge for consumers, as is searching with search engines and shopping on online trading platforms. These services are ‘subsidized’ by the market side that is less price sensitive than the other (see, e.g., Wright, 2004; Rysman, 2009; Kaiser and Wright, 2006; Weyl, 2010). As a result, platform operators generate most of their profits on the market side with the lower price elasticity of demand.

9.2.2 Concentration of Online Markets

As a consequence of indirect network effects platform markets may be more concen­trated than other industries. However, this does not imply that every digital platform market is automatically highly concentrated (see, e.g., Haucap and Heimeshoff, 2014).

Even if multiple platforms are not associated with a duplication of fixed costs, the existence of multiple platforms may not be efficient due to the presence of indi­rect network effects. As Caillaud and Jullien (2003) and Jullien (2006) have shown, a

Table 9.1 Determinants of concentration in two-sided markets

Source: Evans and Schmalensee (2008).

monopoly platform can be efficient because network effects are maximized when all agents manage to coordinate over a single platform. Hence, strong network effects can easily lead to highly concentrated market structures, but strong network effects also tend to make these highly concentrated market structures efficient (see Weyl, 2010; Chandra and Collard-Wexler, 2009).

It is relatively straightforward and immediately plausible that indirect network effects and economies of scale lead to increasing concentration. But it is difficult to draw general conclusions as to how indirect network effects influence market concentration, as their strength varies from platform to platform. The second driver of concentration is econo­mies of scale, which are often the outcome of the cost conditions of online businesses. Many two-sided markets are characterized by a cost structure that combines a relatively high share of (fixed) set-up and maintenance costs with a relatively low share of vari­able costs (e.g., Jullien, 2006). For companies such as eBay, Expedia, and Booking.com most of the costs arise from managing the respective databases, while additional transac­tions within the capacity constraints of the databases cause very low additional costs. Increasing returns to scale are, therefore, not unusual, but rather typical for two-sided markets in the online world. While network effects and economies of scale both have a positive effect on market concentration levels, there are also three countervailing forces that facilitate market competition.

One important countervailing force is capacity constraints. While in offline two-sided markets such as shopping centers, trade fairs and nightclubs space is physically limited,² this does not necessarily hold for digital two-sided markets.

Directly related to the platforms’ heterogeneity is the degree of product differentia­tion between platforms. For dating sites, magazines, and newspapers it is almost always evident that consumer preferences are heterogeneous, so some product differentiation emerges. Such differentiation can be vertical (e.g., high-income users may be more interesting for the advertising industry than a low-income audience) and horizontal (e.g., people interested in sailing, people interested in golf).

The higher the degree of heterogeneity among potential users and the easier it is for platforms to differentiate, the greater the diversity of platforms that will emerge in the marketplace and the lower the level of market concentration. The finding that increas­ing returns to scale foster market concentration while product differentiation and het­erogeneity of user preferences work in the opposite direction is well known from other markets (e.g., Dixit and Stiglitz, 1977; Krugman, 1980). On two-sided markets increas­ing concentration will be driven by indirect network effects, but capacity limits, product differentiation, and the potential for multi-homing (i.e., the parallel usage of different platforms) will decrease concentration levels. How easy it is for consumers to multi-home depends, among other things, on (1) switching costs (if they exist) between platforms and (2) whether usage-based tariffs or positive flat rates are charged on the platform.

To illustrate this idea, consider online travel agencies such as Expedia. Switching from one online travel agency to another is usually associated with relatively low switching costs. Multi-homing is also simple, as travelers can easily search for flights, hotels, and so on, over more than one platform before actually booking. Likewise, airlines, hotels, and so forth, can easily be listed on more than one platform. With respect to search engines, users can also easily and without major costs switch away from Google to another general search engine such as Bing or even to specialized searches over Amazon, TripAdvisor, social networks (for people), library catalogs, travel sites, restaurant guides and so on if a switch appears to be attractive. In contrast, switching costs between social networks such as Facebook are generally much higher because of strong direct network effects and the effort needed to coordinate user groups. While there are no significant direct network effects for Google (i.e., it does not directly matter how many other people use Google) this is not true for social networks, such as Facebook, where the number of users is a very important factor for users’ utility. Still, entry into the search engine business is not easy, due to the indirect network effects described above and the economies of scale that are (1) at least partly based on learning effects, which depend on the cumulative number of searches made over the network in the past, and (2) caused by substantial fixed costs of the technical infrastructure that result in decreasing average costs over a wide output range.

There is another form of switching cost on platforms such as eBay or Airbnb, where, apart from indirect network effects, the user’s reputation is also highly relevant (e.g., Melnik and Alm, 2002; Bajari and Hortaςsu, 2004). As a user’s reputation is a function of the number of transactions already conducted over the platform, it is typically platform specific (e.g., for eBay), so that changing platforms involves high switching costs, as it is difficult, if not impossible, to transfer one’s reputation from one platform to another.

9.2.3 Market Definition for Platform Markets

Having discussed the determinants of concentration in two-sided markets, let us now discuss the peculiarities of defining markets for platform services, as the delineation of the relevant product market is typically the first step in any antitrust proceeding. Market definition concepts are based on actual and potential substitution patterns in order to determine the products and companies that actually or potentially compete with each other. The market definition process aims at revealing the products and companies that are likely to be affected by, for instance, a merger or an abuse of market power. A popular approach in the academic literature is the small but significant non-transitory increase in price (SSNIP) test. If a firm was (hypothetically) in the position to profitably and sustain­ably raise its price by 5 to 10 percent above the competitive price level, it is considered not to be effectively constrained by competition. If, in contrast, such a price increase is unprofitable, for example, because consumers switch to alternative products that they consider to be sufficiently good substitutes, these alternative products are considered to belong to the same product market. Hence, if a 5-10 percent price increase is estimated to be unprofitable there must be other products or firms in the relevant product market.

In online markets, this market definition process becomes much more complicated for two reasons. First, in many online markets consumers do not pay a positive price, at least not in monetary terms. Instead, consumers pay an implicit price in the form of personal data and/or attention (see Evans, 2013). Platforms compete for consumers’ data and their attention in order to sell it to advertisers (tailor-made based on personal data). Clearly, the SSNIP test scenario of a 5-10 percent price increase cannot be computed as long as the starting price (in money) is zero. Even if one were to consider a 5-10 percent increase in the implicit price that consumers pay, namely their disclosure of personal data and/ or their exposure to advertising, it is unclear what such an increase of advertising expo­sure or data disclosure would mean in practical terms and how it could be measured in a meaningful way. This highly relevant practical problem of defining two-sided markets has thus far been largely ignored in the literature.

More attention has been paid to a second problem in defining two-sided markets, namely that the profitability of a price increase on one side of the market also depends on user reactions on the other side and the feedback effects induced as a consequence of the indirect network effects. As Evans (2003, p. 325) has pointed out, in two-sided markets ‘market definition and market power analyses that focus on a single side will lead to analytical errors’. A price increase on one side of the market cannot be analyzed in isolation from the other side, as such separate treatment may define the relevant market too narrowly. A price increase that may be profitable on one side of the market - if looked at in isolation - may no longer be profitable once user reactions on the other market side are accounted for. To provide a simple example: it may appear profitable for an online shopping platform to increase the commission charged to the sellers listed if the additional revenues generated from the price increase exceed the loss in revenues that results from some sellers leaving the platform. However, having fewer listed sellers reduces the platform’s value for buyers, so they may switch to a different online platform, in turn reducing the value of the platform for sellers. In total, the price increase may thus be unprofitable once feedback effects are accounted for, highlighting the importance of defining the market more broadly.

The two-sided market structure causes another problem for competition authori­ties. Since a platform sets prices (explicit or implicit) to at least two customer groups (e.g., advertisers and users) it is not clear which price(s) should hypothetically be increased in a market definition exercise. Should only the price on one market side be increased or all prices simultaneously? This problem is especially severe in situations with asymmetric substitution patterns. Advertisers may regard platforms as closer substitutes than users and may respond to a price increase for advertising more quickly than the user side. An alternative approach to market definition may be to predict how a price increase on either side will impact on the platform’s transaction volume.

Argentesi and Filistrucchi (2007) and Filistrucchi et al. (2014) discuss the applicabil­ity of the SSNIP test in two-sided markets and propose modifying the test.⁴ In order to measure market power, it is necessary to compute price-cost margins while taking into account its two-sided nature. For instance, on online news pages the publisher’s optimal behavior depends on four different elasticities: the elasticity of readers’ demand with respect to the price to access an article; the elasticity of readers’ demand with respect to the quantity of advertising; the elasticity of advertising demand with respect to adver­tising prices (which are typically charged on a pay-per-click basis); and the elasticity of advertising demand with respect to the click conversion rate. In order to compute the price structure an empirical model has to include demand estimations on all sides of the markets taking interactions between the sides into account. This puts high requirements on the amount and detail of data needed and on the estimation techniques.

While the interrelatedness of the markets may, in theory, be resolved via more complex versions of the SSNIP test, data needs put practical limits on its use. Even more chal­lenging is the fact that many Internet platforms are (seemingly) free for users, so that a 5-10 percent price increase can often not be calculated, as users do not pay with money, but with their data and their attention to the advertising shown. It is difficult to opera­tionalize a hypothetical 5-10 percent increase of data disclosure requirements in practice (even though the issue may be examined theoretically if the simplifying assumptions that users are homogeneous and hold the same valuation for privacy are accepted). The value of personal data or privacy varies heavily in terms of monetary equivalents between users (e.g., Bendorf et al., 2015). Moreover, even a theoretical solution is unlikely to work in practice for antitrust agencies, given the enormous data requirements. Some relief may come from surveys about hypothetical consumer reactions and conjoint analysis techniques. Their major drawback, however, is that they use stated rather than revealed preferences and are therefore less reliable than data on observed consumer behavior.

9.3