Online Platforms, Economics of

Abstract

Following the Internet’s widespread adoption, much economic work has studied ‘online platforms’: firms that mainly interact with consumers in cyberspace. This article surveys such work, focusing on the ways in which traditional economic models have been adapted to incorporate novel aspects made relevant by the Internet. This literature can be divided roughly into two categories: broad-brush study of the competition between platforms and more fine-grained study of the ways in which users and platforms interact with one another. The former focuses on extending oligopoly theory to include ‘consumption externalities’; the latter extends auction and search theory to a world of precisely measureable actions.

Introduction

The objective of this article is to articulate as clearly as possible the ways in which more traditional economic theory has been adapted in order to inform the study of Internet-connected intermediaries, or ‘online platforms’. As Varian (2005, p. 12) remarks, ‘Recent literature that aims to understand the economics of information technology is firmly grounded in the traditional literature. As with technology itself, the innovation comes not in the basic building blocks, but rather the ways in which they are combined’.

The building blocks whose combination this article focuses on are the following. First, it analyses the incorporation of consumption externalities into monopoly and oligopoly theory, selectively reviewing the recent two-sided markets literature and its precursors. In doing so, it presents a broad-brush picture that is useful for understanding the fundamental similarities and differences between online platforms and traditional firms, particularly regarding their pricing incentives. It then considers finer-grained perspectives of search engines and electronic commerce, surveying research that mixes elements of auction and search theory to understand how intermediaries sell advertisements to firms and to what extent they seek to match the latter with consumers in a frictionless way.

A Bird’s Eye View: Oligopoly with Consumption Externalities

The theory of ‘network effects’, including the recently developed theory of ‘two-sided markets’ or ‘multi-sided platforms’, can be seen as attempting to extend classical oligopoly theory to incorporate features that are both prevalent among and, to some extent, novel to online platforms. A particularly salient point of distinction between traditional firms and online platforms is that, while the former are more likely to sell goods (or services) whose quality or performance depends largely on the way the firm itself produces it, the latter are more likely, one way or another, to sell connections between different economic agents who potentially benefit from interacting with one another. For example, while a traditional firm might sell shirts or plumbing services, an online platform might offer to link job seekers with employers or to provide the technical means to connect video game developers with gamers. Thus, consumers’ perceptions of the quality of an online platform depend on the set of connections it offers as well as on more traditional factors.

In view of this ‘connecting’ role that they frequently play, a natural class of models for studying online platforms turns out to be that which generalises one version or another of existing oligopoly models by allowing for ‘consumption externalities’. Whereas traditional oligopoly models assume that the utility each consumer derives from purchasing a good depends solely on the price of the good and, in some cases, on certain production choices the seller has made, platform models also allow this utility to depend on the consumption choices of other consumers.

The first literature studying the economics of network effects is not about ‘online platforms’ per se, as it was written before the Internet came of age, in the 1970s and 1980s. Instead, it largely considered issues such as ownership of telephone networks and standards for videocassettes and computer keyboards (Rohlfs 1974; Katz and Shapiro 1985; Farrell and Saloner 1985; David 1985). Nevertheless, the study of online platforms owes a significant debt to this literature, as it provides a fundamental building block for the subsequent literature on multi-sided platforms that is more explicitly focused on online industries. Moreover, it is worth noting that, while the telephone and home video industries may not have been ‘online’ at the time that the above literature was written, such goods are increasingly furnished using the Internet; consider, for example, Skype’s Internet voice service or Netflix’s streaming video service.

A Toy Model of a Platform

The crucial building block established by the earlier literature on network effects is the formalisation of the aforementioned idea that each consumer ‘cares about’ the choices made by other consumers. To see, in the simplest way possible, the form that such a formalisation takes, consider the following very stylised model of a monopoly social network. Here (unlike what one observes with Facebook, for example), assume that the social network charges a monetary price to each consumer who chooses to become a member.

Consumer i’s expected payoff from joining the social network is given by ${v}_i+\beta \widehat{N}-P,$ where vi denotes the component of i’s valuation for the network that is independent of the choices of other consumers, $\widehat{N}$ denotes the total number (more formally, the measure) of consumers expected to join the network, β captures the strength of i’s valuation for ‘interaction’ with other consumers, and P denotes the price. For now, assume that β takes on the same value for all consumers, whereas vi is allowed to vary from one consumer to another and is distributed according to a continuous density function, f(v). Meanwhile, assume that the social network’s profits are given by PN−c(N), where N denotes the number of consumers that indeed do join, and the continuous and increasing function c(N) denotes the cost to the network of serving N consumers. The timing is such that, first, the social network announces P, and, second, each consumer decides whether or not to join.

For a given price, P, the set of consumers that choose to join are those for whom the inequality ${v}_i\ge P-\beta \widehat{N}$ holds. Therefore, the social network’s demand can be written as

$$ N\left(P,,,\widehat{N}\right)={\int}_{P-\beta \widehat{N}}^{\infty }f(x) dx. $$

(1)

To proceed, it is convenient to assume that all consumers correctly anticipate the demand level that the network seeks to attract, so that $\widehat{N}=N$. (Justification for this assumption is considered below.) Using this assumption and the demand function in expression (1), it is relatively straightforward to derive the optimal price for the profit-maximising social network, given by

$$ P={c}^{\prime }+\frac{N}{-\frac{\partial N}{\partial P}}-\beta N. $$

(2)

To interpret the price prescribed by Eq. (2), it is useful to compare it with the price that maximises total surplus, which can be derived in a similar fashion, given by

$$ P={c}^{\prime }-\beta N. $$

(3)

According to (3), it is socially optimal for price to differ from marginal cost by βN Thus, if, as makes sense in this example, β > 0, meaning that consumers positively value the presence of others on the social network, then it is socially optimal to offer a ‘discount’ to each consumer, compared with traditional marginal cost pricing, equal to the total value that the consumer adds to the network: this total value is βN, because each consumer gains β from the presence of another consumer and there are N consumers. (This follows closely in the spirit of Pigou (1912).)

Turning to the formula in (2), for the network’s optimal price, the only difference is the ‘markup’ term, $\frac{N}{-\frac{\partial N}{\partial P}}$, which is positive since $\frac{\partial N}{\partial P}<0$, and which is precisely the same as the markup term that appears in the traditional monopoly pricing formula (Cournot 1838). This reflects the standard trade-off of infra-marginal gain versus marginal loss that faces a firm with market power when it raises its price. As is the case for a total surplus-maximising social planner, the profit-maximising network finds it optimal to discount its price by βN, since, for each additional consumer that joins, the network can extract the value this new consumer creates by raising its price by β, while still retaining N consumers.

Therefore, in this model that incorporates consumption externalities into the traditional monopoly model in the simplest way possible, no additional distortion arises between socially and privately optimal pricing. Such lack of additional distortion, however, is driven crucially by the assumption that β is the same for all consumers. The following discussion illustrates another distortion that arises when β varies across consumers, but first, it considers a more basic form of heterogeneity among a platform’s users.

A Richer Model: Diverse Groups of Platform Consumers

An important feature of many online platforms, which may seem to undermine their resemblance to the model described above, is the fundamental difference between the different types of agents that they connect to one another. For example, while, to a first approximation, one may think of a social network as a platform that connects users to each other, all of whom fall into some common category, in the examples given above of an employment or gaming platform, the assumption is inaccurate. Job seekers look especially for open positions, but not so much for other job seekers. Gamers may value both the opportunity to play new games and the ability to play them with other gamers, but their valuations for these two things cannot reasonably be conflated into a single valuation for interacting with other ‘consumers’.

The literature on multi-sided platforms, pioneered especially by Caillaud and Jullien (2003), Evans (2003), Parker and Van Alstyne (2005) and Rochet and Tirole (2003), extends the type of model discussed above, allowing it to be applicable in a much broader set of environments in which consumers fall into many different categories and have differential valuations for interacting with one another. The basic model assumes that there are s groups or ‘sides’ of consumers. For example, consider a simple extension representing an employment platform and assume that s = 2, where side w represents workers seeking jobs and side e represents employers looking to fill positions. Here, the payoff to worker i of joining the platform is given by ${v}_i+{\beta}^w{\widehat{N}}^{\mathrm{e}}-{P}^w,$ where β^w denotes the marginal impact that the presence of an additional employer has on workers’ valuations for joining the platform, ${\widehat{N}}^e$ denotes the number of employers that workers expect to join, and P^w denotes the price that the platform charges workers. Employers’ payoffs are, analogously, given by ${v}_i+{\beta}^e{\widehat{N}}^w-{P}^e$. The platform’s profits are now given by P^wN^w + P^eN^e−c(N^w, N^e).

Continuing with the same form of analysis discussed above, consider the privately and socially optimal prices. On the workers’ side of the market, these are, respectively,

$$ {P}^w=\frac{\partial c}{\partial {N}^w}+\frac{N^w}{-\frac{\partial {N}^w}{\partial {P}^w}}-{\beta}^e{N}^e $$

(4)

and

$$ {P}^w=\frac{\partial c}{\partial {N}^w}-{\beta}^e{N}^e $$

(5)

Note, moreover, that the corresponding expressions for prices charged to employers simply have the e and w indices reversed.

These pricing formulae follow the same logic as those of the first example, with one crucial modification. The ‘discount’ that workers receive, with respect to the analogous prices in a model with no consumption externalities, is given by β^eN^e: employers’ valuations for the presence of an additional worker times the number of people that join the platform. In the case of the socially optimal price given by (5), β^eN^e is the relevant quantity, because, in this example, it measures the total externality that a worker has on other consumers, as only employers ‘care’ about how many workers are present. For the same reason, in the case of the privately optimal price of Eq. (4), β^eN^e measures the total additional profit that the platform can earn from its ‘other’ consumers, by virtue of serving one more worker. In a more general setting, with numerous groups of consumers and more complex externalities from one group to another, or within groups, the above formulae can be readily extended, in accordance with these general principles.

One particularly relevant insight that a two-sided model gives (but that a model with just one group of consumers does not) is the fact that it can be optimal for a profit-maximising platform to charge one group a negative price, i.e., to pay one type of consumer to join. This occurs when one group’s own demand for the platform is relatively elastic compared to the positive externality that their presence has on the other group. An oft-cited, albeit brick-and-mortar, example of such a phenomenon is nightclub pricing, whereby women are charged a negative price in the form of free entry and complementary drinks, thus attracting more women and allowing the nightclub to extract a higher cover charge from men. A similar phenomenon occurs on many online platforms, although, in practice, it is often not feasible to charge one group a strictly negative price, so, instead, a price of zero is offered to the consumers that generate a high positive externality. One such example is that of search engines, such as Google and Microsoft’s Bing, which are free for web users, whose presence in large numbers increases advertisers’ willingness to pay to appear among the search results.

Rich Heterogeneity Within Groups

As mentioned above, another issue arises when consumers within a given group have heterogeneous valuations for externalities. For simplicity, reconsider the social network example with just one group of consumers. However, following Weyl (2010), who extends the model of Rochet and Tirole (2006), assume that consumer i’s payoff is given by ${v}_i+{\beta}_i\widehat{N}-P$. Here, not only vi but also βi varies across consumers, and they are distributed according to a continuous joint density function, f(v, β). Under this setup, consumers who choose to join the network are those for whom ${v}_i\ge P-{\beta}_i\widehat{N}$ holds, giving rise to demand,

$$ N\left(P,,,\widehat{N}\right)={\int}_{-\infty}^{\infty }{\int}_{P-y\widehat{N}}^{\infty }f\left(x,y\right) dxdy. $$

The expression for the network’s privately optimal price then becomes

$$ P={c}^{\prime }+\frac{N}{-\frac{\partial N}{\partial P}}-\widehat{\beta}N, $$

(6)

where $\widehat{\beta}\equiv E\left[{\beta}_i|{v}_i=P-{\beta}_iN\right]$, while the expression for the socially optimal price is given by

$$ P={c}^{\prime }-\overline{\beta}N, $$

(7)

where $\widehat{\beta}\equiv E\left[{\beta}_i|{v}_i\ge P-{\beta}_iN\right]$.

In economic terms, $\widehat{\beta}$ is the average valuation among the set of marginal consumers, i.e., those consumers who are indifferent between joining the network or not, for the externality created by one more consumer. In contrast, $\overline{\beta}$ is the analogous quantity averaged over the entire set of consumers that join the network. Thus, when consumers are allowed to be heterogeneous in their valuations for externalities, a second source of distortion arises between privately and socially optimal pricing incentives. On the one hand, regarding total surplus maximisation, as (7) illustrates, it is still optimal to offer consumers a discount, with respect to marginal cost, equal to the total externality they create, measured here by $\overline{\beta}N$.

On the other hand, as (6) shows, the profit-maximising network has an incentive to offer such a discount only to the extent that it can recoup the loss that the discount provokes by increasing the rent it can extract from its entire set of N consumers who value the network more highly when there is an additional consumer. When adding a marginal consumer, in order to hold fixed the size of its demand at N, the network increases its price by an amount which will not incite an additional flow of users either into or out of the network. This amount is precisely $\widehat{\beta}$, the average of those consumers who are marginal, because the marginal set of consumers are the only ones who, in response to small price changes, are prone to reversing their decision of whether or not to join.

Note that $\widehat{\beta}$ may be either larger or smaller than $\overline{\beta}$. Thus, unlike the traditional markup distortion, which always pushes the privately optimal price to be higher than the socially optimal one, the distortion arising from heterogeneity in valuations for externalities can push prices to be either too high or too low. Moreover, as Weyl (2010) notes, this distortion, based on differences in valuations for a good’s characteristics between marginal and infra-marginal consumers, closely mirrors the one studied by Spence (1975) in his model of a traditional, quality-choosing monopolist. Veiga and Weyl (2012) explore this connection further, developing a model that allows consumers to contribute externalities to the network in a heterogeneous way.

Modelling Platform Competition: Challenges and Proposed Solutions

The above discussion focuses on monopoly platforms, and a detailed discussion of competition is beyond the scope of this article. A challenge facing the analysis of models of competition with consumption externalities is the presence of multiple equilibria, which arise from two sources. One source, touched on above (when the assumption that $\widehat{N}=N$ was posited), is the possibility of multiple equilibria in the game played by consumers after the platform has already set its prices. In the case of monopoly, this problem can be assumed away somewhat innocuously, because platforms can use contingent pricing to eliminate consumers’ coordination problems. (See Weyl’s (2010) discussion of ‘Insulating Tariffs’ as well as Ambrus and Argenziano (2009), who take an alternative approach that refines consumers’ possible reactions to given prices.) The other source of multiplicity arises in the price-setting game played by competing platforms. As a well-known article by Armstrong (2006) shows, if platforms compete with one another using arbitrary forms of contingent pricing, then the equilibrium of their strategic interaction with one another is severely underdetermined. White and Weyl (2012) propose Insulated Equilibrium as a joint solution to these two multiplicity problems, using it to analyse the impacts of consumer heterogeneity on pricing in a competitive environment. See Reisinger (2012) for an alternative solution to the indeterminacy in the price-setting game.

The above discussion illustrates some of the main issues that arise when consumption externalities are incorporated into oligopoly theory in order to make it useful for the broad-brush study of online platforms that connect consumers to one another. However, this discussion is by no means exhaustive. For example, in many circumstances dynamic considerations, such as those studied by Cabral (2011), are of great importance. In this category, one may include the study of time-dependent pricing strategies as well as the issue of which and how many platforms survive in their respective markets. Furthermore, this discussion ignores the possibility of a platform engaging in second-degree price discrimination, an issue that Gomes and Pavan (2012) concentrate on. Another interesting issue is the impact of consumers patronising multiple competing platforms (known in the literature as ‘multi-homing’) and potentially interacting multiple times (Athey et al. 2012). Finally, for a broad, recent survey of the multi-sided platforms literature, focusing on applications, see Rysman (2009).

Detailed Views of Interaction on and Via Online Platforms

The previous section takes a ‘bird’s eye view’; however, there are also many more detailed issues related to online platforms that can be better understood by ‘zooming in’. While attention to detail typically comes with a loss of general applicability, two topics have proved to be of especially broad interest. These are the ‘sponsored search’ auctions that search engines such as Google and Bing use to sell advertisements and the techniques that Internet sellers use to sustain profit margins in an environment that would seem to favour perfect price competition.

Sponsored Search Auctions

In practice, sponsored search auctions differ from ‘textbook’ auctions in two particularly important ways. First, even though the size of advertisers’ bids largely determines whether they appear in a more prominent slot near the top of the search engine’s results page or in a more obscure one near the bottom, the auction mechanisms dictate that the total payment that an advertiser makes depends on the number of times that users click on that particular advertiser’s link. Second, because the items being auctioned are ads, which are, in effect, opportunities for sellers to connect with web surfers, the interplay between the auction mechanism and users’ surfing behaviour matters, both for descriptive purposes and for evaluating the welfare associated with different mechanisms.

The first articles in economics to study the impact of using per-click or ‘Generalised Second-Price’ auction mechanisms are Edelman et al. (2007) and Varian (2007). While the exact details of the auction are both complex and proprietary to search engines, two stylised features of these auctions are the following.

• a. Auctions are conducted on a keyword-by-keyword basis. So, for example, an advertiser can bid one amount for an ad slot appearing among the results a user sees after searching for ‘shoes’ and a different amount for an ad slot a user sees after searching for ‘boots’.

• b. In each auction, the highest-bidding advertiser receives the most prominent slot and pays the amount bid by the second-highest bidder each time a user clicks on the former’s ad. The second-highest bidder receives the second-most prominent slot and pays the third-highest bid for each click it receives, and so on.

The aforementioned articles show that, despite the second-price ‘flavour’ of these auctions, they are not special cases of the Vickrey–Clarke–Groves (VCG) mechanism, which, in view of a famous result of Green and Laffont (1979), implies that it is not a dominant strategy for advertisers to bid their true valuations for clicks. They further show that such auctions have multiple equilibria, all of which yield at least as much revenue to the search engine as would a VCG mechanism. In more recent work, Athey and Nekipelov (2012) modify the model somewhat, relaxing the assumption that advertisers literally submit a separate bid in every single auction and showing that this pins down a unique equilibrium.

Both Athey and Ellison (2011) and Chen and He (2011) explicitly integrate user behaviour into models of sponsored search. A basic insight of these models is that the value for an advertiser of receiving a slot at the top of the search results page stems not necessarily from its intrinsic ‘prominence’. Instead, the value can come from surfers’ anticipation that advertisers who bid more in an auction are also more likely to have websites that are worth visiting, thus creating a positive feedback loop. Athey and Ellison (2011) further show that, unlike in auctions for traditional goods, in search auctions, reserve prices can be welfare-improving, as they screen out low-quality sites that would be a waste of users’ time to visit.

Price Competition and Obfuscation

As Ellison and Ellison (2009, p. 427) remark, ‘When Internet commerce first emerged, one heard a lot about the promise of “frictionless commerce.” Search technologies would have a dramatic effect by making it easy for consumers to compare prices at online and offline merchants’. However, many would argue that, in its current, more mature state, online shopping is sometimes rather complicated, with goods’ prices and characteristics often not disclosed in a transparent way.

Numerous articles consider different aspects of this issue. Notable examples include the relatively early paper by Baye and Morgan (2001) focusing on brick-and-mortar firms’ decisions about whether or not also to advertise online, Hagiu and Jullien (2011), who examine an intermediary’s incentives not to eliminate the search frictions of its users, and Ellison and Wolitzky (2012), who adapt the classic model of Stahl (1989) to consider the incentives facing the sellers of goods themselves to provoke such frictions. Also, on this issue, see numerous articles published in volume 121, Issue 556 of the Economic Journal, described in an introduction by Wilson (2011).

Further Issues

This article focuses on some of the ways in which traditional components of microeconomics have been combined to build theories that speak to a world in which many important firms are ‘online platforms’. It does not begin, however, to address many of the fascinating issues involving such firms that the economics literature has studied. In particular, it ignores a large body of empirical work that examines matters including regulation of online privacy (Goldfarb and Tucker 2011a), circumstances in which people substitute between online and offline platforms (Goldfarb and Tucker 2011b) and the effect of such substitution on broader social trends (Gentzkow and Shapiro 2011), the dynamics of pro-social behaviour in large online communities (Zhang and Zhu 2011), and the ways in which online sellers experiment (Einav et al. 2011), to name a few. Finally, for a more comprehensive survey, which expands on many of the topics discussed in this article, the reader is referred to Levin (2012).

See Also

• Computer Industry

• Internet, Economics of the

• Information Technology and the World Economy

• Network Goods (Empirical Studies)

• Open Source Software, a Brief Survey of the Economics of