Electronic money and the network externalities theory - Springer Link

Netnomics 1 (1999) 137–171

137

Electronic money and the network externalities theory: lessons for real life ∗ Leo Van Hove Postdoctoral Fellow of the Fund for Scientific Research – Flanders, Money and Finance Research Group and Vesalius College, Free University of Brussels, CFEC-M414, Pleinlaan 2, B-1050 Brussels, Belgium E-mail: [email protected] The aim of this paper is to show that the network externalities theory provides a useful framework to analyse the introduction and further development of the new electronic payment instruments currently being launched. To that end the paper presents a pragmatic (and selective) reading of the network externalities literature; i.e., it screens the literature in search of both theoretical insights and empirical results which can be transposed to the case of electronic payment instruments. In so doing, the paper concentrates on the so-called electronic money products and, especially, on electronic purses. Specifically the paper shows that the network externalities literature provides a number of useful insights concerning consumer reactions to the introduction of electronic purses (and the ways in which card issuers can anticipate these reactions) and concerning the strategies card issuers may follow in a competitive market with incompatible electronic purses. All this is substantiated by multiple references to real-life situations. The first section of the paper defines the concept of network externalities. The third section demonstrates that payment cards in general are indeed network goods. It also points out which kind(s) of network effects apply to the electronic purses currently available. The paper then goes on to answer five questions in depth. Firstly, what can we learn from the network externalities literature as far as the ‘chicken-and-egg’ problem is concerned? – the problem being: merchants will not invest in terminals without a sufficient number of potential users, while the general public will not use electronic purses unless there is sufficient acceptance. Secondly, is there room for more than one incompatible electronic purse, or is the electronic purse market prone to ‘tipping’ and ‘lock-in’? Thirdly, can an electronic purse issuer gain a (decisive) first-mover advantage by entering the market before others do? Fourthly, under which conditions will card issuers be inclined to make their electronic purses compatible? And finally, what is the optimum pricing strategy for an electronic purse issuer?

1.

Introduction

All over the world a host of new so-called electronic money products1 are undergoing testing or limited application trials. Some of these products are on the verge ∗

An earlier version of this paper was presented at the 1st Berlin Internet Economics Workshop, Berlin, Germany, October 24–25, 1997. I wish to thank Mady Decrick and Aloys Prinz for helpful comments. 1 Following Groeneveld and Visser [49, pp. 69–70] I use the term electronic money for “products with an information carrier, e.g., a microchip or a computer hard disk, containing prepaid value to be used as a multipurpose means of payment. This definition covers prepaid cards (electronic purses) as well  Baltzer Science Publishers BV

138

L. Van Hove / Electronic money and the network externalities theory

of nationwide roll-out (or global roll-out, where Internet payment instruments are concerned). Still others, like the Belgian Proton card or Digicash’s Ecash, have been launched for some time already. When launching these payment instruments, issuers are faced (or will be) with a ‘chicken-and-egg’ problem: merchants will be reluctant to invest in new equipment or software needed to accept payments unless sufficient consumers show their interest, while consumers, on the other hand, will not use the new means of payment as long as they can only pay with it in a few places. This dilemma originates from the fact that electronic money products – just like telephones, facsimile machines, e-mail, etc. – are ‘network goods’; that is, products that are subject to the so-called network externalities. This paper aims to show that the network externalities theory provides an interesting framework to analyse the above and other problems related to the introduction and proliferation of electronic money products. In doing so, the paper concentrates on the case of electronic purses, but most of the insights hold equally well in the case of electronic cash. The remainder of the paper is organised as follows. In section 2, I first set out the concept of network externalities. In section 3, I point out which kind(s) of network effects apply to the electronic purses currently available. In the body of the paper – section 4 and its subsections – I then screen the network externalities literature in search of insights that are of practical relevance. More specifically, I will answer five questions in detail: (1) What does the literature teach us about the above mentioned ‘chicken-and-egg’ problem? (2) Is there enough room for two or more incompatible electronic purses, or is the electronic purse market prone to ‘tipping’ and ‘lock-in’? (3) Does an electronic purse issuer get a decisive head start when being the first on the market? (4) When will card issuers be inclined to make their electronic purses compatible with others? and (5) Can a well thought-out pricing strategy be of help in breaking the vicious circle mentioned above? 2.

Network externalities

Microeconomic theory defines an externality or spill-over to be a situation in which the production or consumption activities of one individual or firm have an impact on the production or utility functions of other (not directly involved) economic agents, without this effect being accounted for via the price mechanism2 (see, e.g., [12, p. 38]). Externalities can be beneficial or detrimental. The activities of a bee-keeper, for example, may involuntarily increase the utility of a neighbour who grows apples in his orchard. Air and water pollution, on the other hand, are classic textbook examples of negative externalities. Network externalities are in fact nothing more than a particular form of the externalities just set out. Network externalities are positive externalities, they are consumption externalities (rather than production externalities), and their most distinctive 2

as software products that use computer networks such as Internet (digital cash)”. Some authors use the term ‘not market-mediated’ to emphasize the latter.


139

feature is that they occur (strongest) for a well-defined category of consumption goods. More specifically, network goods are “products for which the utility that a user derives from consumption of the good increases with the number of other agents consuming the good” [51, p. 424]. The key requirement for network externalities to arise is thus clearly a certain complementarity and/or interaction between the goods of the individual consumers [38, p. 3; 35, pp. 3, 6, 9]3 . Put differently, the consumers and their goods have to be part of the same explicit network or, in the case of industries that are not explicitly networks, one has to be able to think of the consumers and their goods as being components of the same virtual network. Typical network goods have little or no value in isolation; they derive their value solely from the ‘connection’ with other goods. The examples below will illustrate this amply. Network externalities can be direct or indirect. Direct network externalities are also called demand network externalities. Typical examples include phones, fax machines, and a number of other communication technologies. It is self-evident that your fax machine is quite useless if you are the only possessor of a fax machine. It is equally self-evident that the utility of your fax machine increases as the apparatus becomes more widespread. Besides such network externalities in their ‘purest’ state – “generated through a direct physical effect of the number of purchasers on the quality of the product” ([51, p. 424]; my emphasis)4 – there are also network externalities of a more indirect nature. These are also labelled supply network externalities as they are not situated on the demand side (as direct network externalities are), but rather on the supply side. In this respect, Katz and Shapiro [51,54] use the term ‘hardware/software’ systems, “where one user’s adoption has no direct impact on the utility of other users, but may have lagged, indirect effects through the provision of software. . .” [54, note 2, p. 96]). That is: as more units of a given type of hardware are sold, the supply (read: the variety5 ) of software for use with this hardware will increase, which in turn 3

Cf. also Economides [34, p. 2]): “In general, network externalities arise out of the complementarity between the various pieces of the network”. 4 Where “quality” has to be interpreted broadly as ‘utility for the consumer’. 5 Another effect may be a fall in the price of software. This, however, does not classify as an externality in a narrow interpretation. In this respect, it is interesting to note that Liebowitz and Margolis [59,60] argue that the definition of Katz and Shapiro (and other authors) is not accurate enough since it can be interpreted to comprise pecuniary externalities – “external effects that work through the price system” [59, p. 137]. Liebowitz and Margolis argue that “almost any product with increasing or decreasing costs can be considered a network, as network is being used in the current literature: Additional consumption may raise or lower the cost of a product to other consumers and it may raise or lower the cost of substitutes and complements. [. . .] Any network externality that is ‘market mediated’, meaning that the size of the network influences the price of inputs to a firm, or goods and services to a consumer, is the same as the pecuniary external economies and diseconomies that so perplexed Marshall, Pigou and at least some in the generations of economists that followed” [60, p. 3]. Liebowitz and Margolis therefore make a distinction between network effects (broad) and network externalities (narrow interpretation). In this paper I do not make this distinction. The terms network externalities and network effects are used interchangeably and they both refer exclusively to ‘real’ externalities (that is, exclusive of pecuniary externalities). Note in this respect that the definition of externalities that I used as a stepping stone to

140


increases the utility of the hardware. This hardware/software paradigm holds not only in its literal interpretation – for home computers – but also for video recorders, hifi equipment, etc. A final source of indirect network effects relates to the postpurchase service for durable goods: “the quality and availability of postpurchase service for the goods (may) depend on the experience and size of the service network, which may in turn vary with the number of units of the goods that have been sold” [51, p. 424]. An obvious example is the service network of a particular automobile brand6 . As the examples show, there is another important distinctive feature of network goods besides the required complementarity and interaction: network externalities are more self-contained than other forms of externality [81, p. 26]. That is, “in order to enjoy the effects of the externality, one must join the network. The benefits to network participation, although partially external to the individual participant, are entirely internal to the network as a group”, whereas “pollution is an external cost that has effects beyond the set of people engaged in the polluting activity” (ibidem, my emphasis).

3.

Payment cards and network externalities

All goods mentioned so far have one feature in common: “the utility that a given user derives from the good depends upon the number of other users who are in the same ‘network’ as is he or she” [51, p. 424]. Clearly this is also the case for payment cards (see [69, p. 23] for a real-life illustration). As a matter of fact, the network externalities literature often explicitly cites payment cards as an example of goods that are subject to network effects7 . This said, the network effects involved will be indirect rather than direct, since the cards cannot be used for payments between consumers (the so-called ‘personto-person’ or ‘peer-to-peer payments’). Encaoua et al. [43], for example, mention introduce the concept of network effects/externalities is also a narrow one. Note that the so-called “Marketing Science literature” [14, p. 3] studies a particular form of positive consumption externalities that are generally called experience effects. The essence of such effects can be summarised as follows: when a new product is introduced, there is uncertainty among potential consumers about the product’s attributes, especially if it is a complex and/or high-tech product. As a result, risk averse consumers will rate the product lower than they would if they were perfectly informed. As more consumers adopt the innovation, word of mouth gradually reduces the uncertainty. This will lead risk averse consumers to increase their product valuation, resulting in an increased demand [85, p. 909]. For further references on experience effects – which will not be discussed further here – the reader is referred to the two articles cited. In my view, an important difference in comparison with the network effects in this paper seems to be that with experience effects the adoption of the product by new consumers does not increase its value for existing consumers (see also below). 7 See, for example, [33,34,38, p. 4; 39, note 4 p. 2,; 40, p. 105; 43, p. 52; 44, p. 267; 54, p. 94; 74]. In contrast, the payment systems literature until recently paid only little attention to network effects. Exceptions were [6,24,67], and especially [19,20,65]. However, the interest in the impact of network externalities on payment systems seems to be increasing, cf. [1,11,50,68, pp. 6 and 18–20; 73, pp. 18–21; 75, p. 9; 81; 83, pp. 36–37]. 6


141

credit cards as an example of a situation “where a consumer of some network values directly the size of the network instead of the total number of users” ([43, p. 55]; my emphasis)8 . Probably more enlightening is the explanation of Katz and Shapiro [54, p. 94], who stick with their hardware/software terminology and state that where credit cards are concerned, the card itself is the hardware, and merchant acceptance of the card the software. The network effects caused by the entry of an extra card holder clearly are on the software side: the more consumers use a given card, the greater the chance that additional merchants will decide to accept it (note that this is what Encaoua et al. mean by ‘the size of the network’). This in turn will increase the value of the card for card holders – both old and new. A third way of explaining why, in the case of payment cards, the network effects are indirect rather than direct, is by making use of the terminology of Economides and Himmelberg [38]. Economides and Himmelberg make a distinction between one-way and two-way networks, depending on whether there is a sense of direction9 . Telephone and facsimile networks, e.g., are thus two-way networks in which “customers are identified with components and the externality is direct” [38, p. 3]. Put differently: in typical two-way networks, the size of the network equals the number of consumers (because the consumers are the network nodes), and hence the externalities are of a direct nature. The connection of an extra telephone, for example, yields direct benefits for all other subscribers in the network because the number of possible communication links increases10 . In typical one-way network – such as broadcasting television – on the other hand, there is no direct interaction between consumers, and thus benefits will by definition be indirect. Payment card networks clearly also fall in this category11 . For the majority of the electronic purses currently available or on the way – the Belgian Proton card, the Dutch Chipknip, the Finnish Avant card, the Danish Danmønt card, etc. – the analysis is completely identical to the one I have just presented for the case of debit and credit cards. There is, however, a maverick: the Mondex card, originally developed in the UK. A unique feature of the Mondex technology is that it not only allows for payments from consumers to merchants, but also – unlike any other electronic purse – for transfers between consumers. This can be done by inserting Mondex cards into a so-called electronic wallet – an apparatus somewhat resembling a 8

Compare with Saloner and Shepard [74, p. 481] on the subject of ATM networks: “Although each user is largely unaffected by the number of other users of the same network, each is better off the greater the number of outlets from which she can access the network”; a similar remark can be found in [64, p. 1116]. 9 This terminology was originally introduced by Economides and White [41,42]. 10 Note that because the number of users is of direct importance, direct network effects are sometimes also called club effects; see, e.g., [72]. 11 Cf. Economides and Himmelberg [39, note 4, p. 2]: “As a rule, direct effects occur in two-way networks, such as telephone and road networks, where reciprocity is present (a phone call from A to B is distinct from a phone call from B to A, and they are both feasible and demanded) and consumers are identified with network nodes. Indirect network externalities typically occur in one-way networks (such as an Automatic Teller Machine network) where two different types of components (ATM machines and bank accounts) are combined to create a demanded good, and reciprocity is not present”.

142


pocket calculator. By using such a wallet, money can be transferred from one card to another. Current Mondex technology is also able to make person-to-person payments via telephones equipped with a card reader (so-called smart phones), and at some point in the future it will also become possible to exchange Mondex electronic cash over the Internet [9, p. 6] and even via mobile phones [71] (for additional information on Mondex, see [78,80]). Especially the latter two options make it abundantly plain that the Mondex network – unlike any other electronic purse network – is in fact a twoway network. Hence, the Mondex card is subject to both indirect (where ‘over the counter’ payments are concerned) as well as direct network effects (for peer-to-peer payments)12 .

4.

Lessons for real life

In the previous section I have demonstrated that payment cards in general and electronic purses in particular are indeed network goods. I have also pointed out which kind(s) of network effects apply to the electronic purses currently available. An obvious question then is: what can be learnt from the network externalities literature for the specific case of electronic purses? And especially: can this literature be of help in identifying factors that could either enhance or inhibit the adoption of electronic purses? Before I present the results of my pragmatic reading of the network externalities literature in the following subsections, let me first stress that the literature to date has been largely theoretical; for the present, empirical studies can be numbered on the fingers of one hand13 . Hence, there are almost no studies at hand that deal specifically with payment cards. However, as I will show, it is possible to transpose selected conclusions of the general-theoretical literature to the specific case of electronic purses. 12

In view of the widely documented resistance of (specific layers of) the general public against the ongoing ‘electronisation’ of the payment system, it can in my opinion be safely assumed that electronic purses will also be subject to the so-called experience effects (see footnote 6). This paper does not, however, go further into this matter. 13 An empirical study which is of special interest for the subject of the present paper is the article by Saloner and Shepard [74]. This because it underpins – albeit indirectly – the existence of network effects in ATM networks. The reasoning behind the approach followed in [74] is, in summary, the following: the network externalities literature suggests that the value of an ATM network for a (potential) bank member’s depositors increases in the number of locations it serves. Assuming that banks are able to extract part of the benefits accruing to their depositors (by charging a higher price, . . .), one would expect banks with the potentially largest ATM network to adopt sooner. Saloner and Shepard test this hypothesis with data about the adoption of ATMs by U.S. financial institutions in different states over the 1972–1979 period. In doing so, they control for economies of scale and other differences between banks that could affect the decision to adopt ATMs (labor expenses, degree of market concentration, . . .). Their regressions show that, all else equal, banks with a large number of branches – the variable that serves as a proxy for the (unobservable) expected size of their ATM network – will indeed install ATMs sooner. Saloner and Shepard stress that this behaviour is consistent with the existence of network effects. Moreover, the estimated network effects are quite strong: adding an extra branch increases the probability of adoption by far more than adding enough depositors to equal an average-sized branch.


143

It should be clear that I will not let the entire network literature pass under review here. To start with, entire chunks are plainly irrelevant for the specific case studied in this paper. One example concerns articles on the behaviour of ‘software’ firms and/or on the vertical integration of software and hardware firms (see, e.g., [22,23,40]) – simply because of the fact that in the case of electronic purses the ‘software’ is of a special nature (as was set out in section 3) and cannot be ‘produced’ as a separate ‘product’ by other firms. Moreover, I have also been selective in my reading of those parts of the literature that could, in principle, be of relevance. For one, the paper deliberately keeps aloof from a discussion of the welfare impacts of the adoption of one or other electronic purse, the possible social inefficiencies that may arise, the need for government intervention, etc.; the paper studies the introduction of electronic purses (almost) exclusively from a business-policy perspective. Moreover, only those cases are studied that correspond best to the situation prevailing in real life. Unlike Farrell and Saloner [45,46], Katz and Shapiro [52,53], and other authors, the paper, for example, steers clear from the issue of intergenerational rivalry – being a situation in which “an older technology with an installed base competes against a newer technology lacking such a base” [53, p. 57] – but rather concentrates on situations where “competing technologies become commercially viable at approximately the same time” [53, p. 72]14 . Throughout the paper I also (implicitly) assume that the competing electronic purses are (relatively) homogeneous goods, in the sense that they are ‘functionally identical’. I will therefore not discuss the impact of differences in ‘intrinsic quality’15 . Finally, it is not the paper’s intention to discuss the theoretical models in detail. 4.1. Critical mass The first important contribution of the network externalities literature concerns the formalisation – through the introduction of the critical mass concept – of the intuitive chicken-and-egg-paradox hinted at in the introduction: “. . . consumption goods with network externalities are often characterized by the existence of a critical mass point. That is, an equilibrium market for the good does not exist unless the installed base is greater than a minimum level” [38, p. 1]. And further: “. . . for many network goods, the critical mass is of significant size, and therefore for these goods small market coverage will never be observed – either their market does not exist or it has significant coverage” [38, p. 5]. 14

Katz and Shapiro [53], e.g., study new-product introduction in markets with ongoing technological progress (and network externalities). In this situation, firms have a strategic reason to delay entry: the later they enter, the better their products and/or the cheaper they are to produce. Under these circumstances, the timing of the introduction is obviously a crucial strategic decision. Where electronic purses are concerned, however, the market is characterised by projects which are past the development stage and which are on the verge of being introduced (if they have not already been launched). Hence, in my view the different electronic purses can reasonably be thought of as being part of the same generation. 15 See [44] for a model that relates to so-called vertically differentiated goods.

144


Not surprisingly, expectations play an important role in the modelling of the above. The theoretical models assume without exception that consumers are rational surplus-maximising individuals who decide to purchase a product (or not) by weighing its utility against its cost. If the difference is positive, the consumer buys the network good; if there is no surplus, he stays out of the market. In the least complex, static models, the consumer’s purchase decision relates to a single period, and the price is given. The size of the benefit, on the other hand, is uncertain as it is assumed that the consumer must make his decision at the beginning of the period, before the actual network size is known. Consumers will thus have to form expectations about the size of the network during the period. In the models, the utility of the network good is usually denoted by an expression of the general form a + b(ne ). In this expression the constant a represents the ‘stand-alone benefit’ or ‘autarky value’ of the product16 ; that is, its value in a network of size 0. For the so-called pure network goods – such as telephones – a equals 0 [38, p. 18]. The second part of the expression, b(ne ), captures the actual network benefit. This is an increasing function of the expected network size ne with b(0) = 0. It is further assumed that consumers are, to a certain degree, heterogeneous in their preferences (and thus in their basic willingness to pay for a given product). This can be built into the model by distinguishing a number of consumer types, by adding an index to the expression a + b(ne ), or by specifying the utility function in the style of u(y, ne ) = y(a + b(ne )) (see [38, p. 6] for an example of the latter approach). Under the latter specification, the utility that consumers receive from owning the network good is proportional to their income, y, but in principle it is perfectly feasible to incorporate one or more other relevant consumer characteristics besides (or in the place of) income [38, note 5, p. 6]17 . Finally, it should be stressed that the models, on the other hand, do place limits on the heterogeneity of consumers by assuming that they are homogeneous in their valuations of the network externality b(ne ), and by assuming that all consumers have identical expectations of network sizes [51, p. 426]. It will be clear that in models with the above features, the expectations of the general public will be of crucial importance for the success of a network good. And this all the more so when an entirely new product is introduced – as is the case for electronic purses18 . What is important is that for a given price and a given set of utility functions, multiple equilibria may exist – depending on the expectations of the public (see a.o. [38,39,51,54]). If these expectations are low, the product will not succeed in attracting the critical mass needed and, hence, the network will be bogged down. Formally this can be explained as follows [35,38]. Obviously, aggregate demand equals the total 16

These terms are used by Saloner and Shepard [74, p. 481], and Liebowitz and Margolis [61], respectively. 17 For simplicity it is then assumed that the consumer attribute in question is uniformly distributed, which in fact boils down to assuming a linear demand curve; see, e.g., [31, p. 337; 51, p. 426]. 18 If the product is already on the market, consumers can base their expectations ne in part on the current size of the network. If, however, the product is only just being launched, they lack such a ‘clue’ and, hence, ne will be based on pure expectations alone.


145

Figure 1. Construction of the fulfilled expectations demand curve (a = 0). Source: based on Economides and Himmelberg [39, figure 2, p. 9]).

number of consumers willing to purchase the good given the market price p and given the size of the installed base ne ; put differently, the total number of consumers for whom y(a + b(ne )) > p. Aggregate demand thus takes the general form n = f (ne , p). By inverting this, one obtains an expression for the price that the marginal consumer is willing to pay, given the number of people who demand the good, n, and given ne . This expression is of the form p = p(n, ne ) (see [38, note 6, p. 7] for more details). The curve p(n, ne1 ) in figure 119 , e.g., shows the willingness to pay for a varying quantity n, given an expected network size of ne1 . Note that this demand function is downward sloping in price all right – just like a demand curve for any ‘normal’ good20 . What makes network goods a special case, however, is that the demand curve 19

Note that this figure is drawn for the special case where a = 0. Figure 2 contains more general demand curves. Note also that n and ne have been normalised so that they represent market shares rather than absolute quantities. 20 As Economides [35, p. 6] points out, the usual definition of the network externalities concept may be confusing in this respect: “(A network externality) signifies the fact that the value of a unit of the

146


shifts outwards with increasing expectations; compare p(n, ne2 ), p(n, ne3), and so. Put differently: p(n, ne ) is a negative function of the first determinant, but the willingness to pay does increase with ne (in consequence of the network effect). Now, obviously, in equilibrium n = ne ; cf. [35, p. 7]: “. . . the level of network demand obviously determines the size of the installed base if the market clears” (cf. also [36, p. 8; 51, p. 426]). By imposing this equilibrium condition it becomes possible to construct a more tractable, ‘classic’ demand curve p(n, n) – in which ne has disappeared. This is referred to as the fulfilled expectations demand [35,51]. To construct this curve, one simply has to take those points from each demand curve p(n, ne1) at which the corresponding expectation is fulfilled; that is, the points at which n = nei , cf. E1 , E2 , etc. In other words, the fulfilled expectations demand curve is simply the set of points p(ne , ne ). Now, what is fundamental is that “for goods with network externalities, the (fulfilled expectations) demand-price schedule may not slope downward everywhere” ([38, p. 5], my emphasis21 ). If this is the case, as in figure 1, there will be a critical mass point. The latter is defined as the minimal nonzero equilibrium size (market coverage) n0 of a network good or service (for any price) (ibidem). In the figure, the critical network size corresponds to the top of the demand curve. Clearly, at all marginal costs c > c0 , the only equilibrium is n = 0; that is, the network will simply not take off22 . This is why networks with a critical mass point will typically display a discontinuous expansion path: “In such markets, as costs decrease (in the comparative statics sense) we may observe discontinuous expansions in sales rather than the smooth expansion along a downward sloping demand curve. In particular, we may observe a discontinuous start of the network: as costs decrease, the network starts with a significant market coverage (say 10% of the market) rather than starting with 0.1% coverage” [39, pp. 4, 5]. In figure 1: as soon as c falls to c0 , the network will all at once reach a size n0 0 23 . good increases with the number of units sold. To economists, this fact seems quite counterintuitive, since they all know that, except for potatoes in Irish famines, market demand slopes downwards”. As the above wording may be somewhat misleading, Economides argues that one should better rephrase it as follows: “the value of a unit of a good increases with the expected number of units to be sold” (ibidem). 21 Compare also with Dhebar and Oren [31, figure 1, p. 340]. Note especially that the model underlying this figure does not include expectations. Dhebar and Oren (in their most simple model) consider consumers who are completely myopic, and who therefore base their decisions on the size of the existing network [31, p. 338]. This is nevertheless sufficient to obtain a ‘bell-shaped’ demand curve and, thus, a situation where “if the network is to reach some desired stable steady-state equilibrium, then it must grow beyond a ‘critical mass’ ” [31, p. 399]. The explanation for this is simple: it suffices that the utility of the good increases with the size of the network – either via b(n) or via b(ne ). In combination with the assumed consumer heterogeneity, this will cause the demand curve to shift upward when the size of the network increases. 22 Note that p(n, n) also includes the entire vertical axis, which is drawn thicker on purpose. Note also that for now constant marginal cost are assumed. 23 Note that for c < c0 there are multiple fulfilled expectations equilibria. In this respect, Economides and Himmelberg [39, p. 5] note: “Often, there are multiple fulfilled expectations equilibria. Consumers and producers can coordinate to reach any one of them. We will assume that they will reach the


147

From the quotations in the first paragraph of this section, the reader may already have gathered, and correctly so, that not all network goods are faced (to the same extent) with the existence of a critical mass point. The question thus is: under what conditions do networks exhibit critical mass? And also: for which network goods will this problem be largest? To begin with, Economides and Himmelberg [38] show that neither the existence nor the size of the minimum feasible network depends on market structures. The critical threshold is the same for everyone, even for a monopolist [38, p. 12]. This said, one qualifying remark is in order: a monopolist stands a better chance at influencing the expectations of consumers (see below), and this should increase the probability that the threshold level will be exceeded (ibidem). Economides and Himmelberg [38] also indicate which factors do determine whether the critical mass phenomenon will raise its head. In order to avoid going into too much detail, let me just state that this depends on the strength of the network effects and/or the size of the stand-alone benefit of the good under consideration (for more details, see [38, p. 10; 39, pp. 9–11]. More formally: the greater the slope of b(n) – especially in the lower regions of the function (that is, for small network sizes) – and/or the smaller the a, the greater the probability that there will be a (significant) minimum threshold. It will be these conditions that determine whether the fulfilled expectations demand curve will slope downward everywhere or not. More specifically, strong network externalities and/or a low value for a will cause p(n, n) to be upward sloping for small n and to have an overall reversed U-shape with a peak that corresponds to the critical network size. For weaker network externalities, p(n, n) will slope downward everywhere, so that the curve “exhibits no qualitative difference to an ordinary demand curve” [38, p. 8]. This means that there will be no critical mass point and, hence, no discontinuous start of the network. See figures 2(a) and 2(b) for a graphical representation of the fulfilled expectations demand curve in the case of strong and weak network externalities, respectively. So far my reasoning has been confined to static one-period models. It is obvious that such models are very simplifying and that not all network goods can be adequately modelled within such a framework. For durable network goods such as phones and fax machines, for example, one will clearly have to switch to (more complex) dynamic multi-period models. In such models, the consumer will need to predict, for each and every period, both the price of the network good and the size of the network, and this for the entire expectation of life of the good under consideration. When making his decision, the rational consumer will then weigh the present value of all future benefits against the present value of all costs involved. In view of the purpose of this paper, I will refrain from discussing the nuts and bolts of multi-period models (the interested reader is referred to [39, p. 20]); the more so because the conclusions which can be drawn from these models are for the greater part equilibrium of the largest network size. Thus, when more than one network size is supported by the same price, we select as the equilibrium the highest network size supported by that price; this network size Pareto dominates the other network sizes supported by the same price”; see also [52, p. 827].

148


(a)

(b)

Figure 2. The fulfilled expectations demand curve with strong (a) and weak (b) network externalities. Source: based on Economides and Himmelberg [39, figure 2, p. 9].

analogous to the conclusions of one-period models24 . There is, however, one specific (restrictive) assumption that can hardly be left unmentioned. In the vast majority of the models – static as well as dynamic – it is assumed that supply is infinitely elastic. Under this assumption a network would be able to jump instantaneously from zero to its maximum size, provided that costs drop sufficiently. Clearly, this is rather unrealistic25 . It is intuitively clear that past a certain point, production costs simply will have to increase with the speed at which the network is expanded. In real life one will therefore never observe fully grown networks emerging out of the blue. This is why Economides and Himmelberg [38,39], in an attempt to make their model more 24

Cf. [38, p. 14]: “. . . we are able to show a one-to-one correspondence between a dynamic durable goods problem under perfect competition to a single-period problem”; see also [39, pp. 22, 23]. Moreover, one could argue that for the specific case of the introduction of electronic purses, a one-period model covering one year would suffice. For the use of their purse consumers typically will have to pay an annual fee, as is currently the case for debit cards. This implies that consumers can reconsider their decision at the end of each year, or, put differently, that their initial decision relates to a one-year period. The level of the annual fee can reasonably be considered to be fixed over that period. Moreover, it will also be known at the beginning of the period. In order to make their decisions, consumers will therefore only have to predict the size of the network during the first year. Hence, what one has here is just a one-period model (cf. the body of the text). Building a model that would not only cover the introductory stage, but also later periods seems much more complex, as one would also have to take into account the ‘repeated purchase’ – nature of the electronic purse (and this is still virgin territory in the network literature). In connection with this, it should be stressed that, in this paper, the term ‘electronic purses’ refers solely to reloadable purses. The reader will understand that disposable cards are an entirely different problem all together. 25 Cf. [39, p. 23]: “The discontinuity of the time path is empirically counterfactual. This is because it implies an infinite instantaneous rate of supply at the discontinuity”.


149

realistic, introduce a finitely elastic supply curve, thus precluding discrete jumps26 . Nevertheless, even in such a model, the adoption path still has a different shape compared to goods that are not subject to network externalities; that is, the ‘classic’ Sshape turns out to be much more pronounced: “. . ., the growth path . . . displays a much steeper slope and reaches the maximum network size much faster” [39, p. 28]27 . This is due to the self-reinforcing nature of the demand for network goods: once the adoption takes off, a ‘bandwagon effect’ comes into being28 . According to Economides and Himmelberg [38,39], the explosive growth of the market for facsimile machines in the U.S. during the second half of the 80s is a good illustration of this: “[T]his tremendous surge in demand was not driven as much by outside shifts in consumer demand and price reductions as much as it was driven by the ‘feedback’ effect induced by both past increases and anticipated future increases in the size of the installed base” [38, p. 16]29 . A final question that needs answering in this subsection is whether an electronic purse is a good that will have to succeed in generating an initial critical mass, for it to be viable. The theory suggests that this question can be answered affirmatively without a shadow of doubt, as the electronic purse satisfies both necessary conditions mentioned above. For one, the stand-alone utility of an electronic purse clearly equals zero (a = 0). This in itself is already a sufficient condition for the existence of critical mass (see also figure 1). Moreover, it is intuitively clear that the network externalities associated with the use of an electronic purse are fairly strong, so that the second condition – a sharply upward sloping b(ne ) – is also satisfied. Obviously, the theory is silent on the actual level of the critical minimum threshold. It does, however, underpin the intuition that the success of an electronic purse can become self-reinforcing once the adoption exceeds the threshold: the more people use the card, the more merchants will accept it, the more interesting it becomes for as yet unconvinced consumers to start using it, and so on. 26

“Intuitively, this modification rules out discrete jumps in equilibrium since such jumps could only be supported by infinite prices” [39, p. 23]. 27 See also [13]. Note that the slope of the adoption path is positively correlated with the (assumed) degree of supply elasticity; cf. [39, p. 28]: “Additional simulations . . . confirm that as the slope of the industry supply curve approaches zero, the slope of the growth path over the initial region approaches infinity”. 28 This snowball effect seems explosive, and indeed would be, except for the inherent downward slope of the demand curve [35, p. 9]. 29 Economides and Himmelberg underpin this by calibrating their model to the U.S. fax market (in essence they estimate a simple demand function in which a price term appears on the right-hand side of the equation together with a measure of the (expected) installed base, which functions as a demand shifter). The goodness of the calibration fit turns out to be high and the estimated network effect is (very) strong. Note that Economides and Himmelberg control explicitly for the dramatic fall in prices on the U.S. fax market. Hence, their exercise appears to be less liable to the criticism of Liebowitz and Margolis [59], who argue (and rightly so) that anecdotal evidence – the sort of evidence that was predominant until then – about exponentially growing networks is simply insufficient proof for the existence of network externalities, because the influence of (a fall in prices as a result of) technological progress is not separated from the possible influence of network externalities: “Some phenomena that look like they are network effects are simply manifestations of technological progress” [59, p. 149].

150


4.2. The importance of the installed base, and the probability of tipping The previous subsection has shown that the size of the installed base is of crucial importance for the adoption of a network good30 . A logical corollary is that the installed base is equally important in situations where two or more incompatible networks compete on the same market. Rational consumers will choose the ‘brand’ for which their surplus is maximised. And for network goods, the size of the network – both at present and in the future – has a large weight in this calculation. In the terminology introduced earlier: consumers will buy good i for which y(ai + b(nei )) − pi is largest31 . In short, as Economides [35] puts it: “. . . in markets with network externalities, when firms and consumers interact in more than one period, history matters” [35, pp. 26, 27]. Stronger still: especially the early periods in the product life cycle may have a decisive impact on the final outcome of the rivalry. Under certain circumstances, the feedback nature of the demand for network goods may magnify an insignificantly small initial advantage into an almost unbridgeable gap32 . A classic reference in this respect is Arthur’s article on the diffusion of technological standards [2]. In his initial and most simple model, Arthur considers two technologies A and B which compete for adoption on a market made up of consumers who have a natural preference for one or other of the technologies. However, the returns to choosing A or B realised by any consumer also depend upon the number of previous adopters. Arthur shows that if this relationship is positive – that is, if there are increasing returns to adoption33 – the market is inherently unstable, meaning that “. . . the two technologies cannot coexist indefinitely: one must exclude the other” [2, p. 121]. At a certain point the competitive balance will inevitably (and irrevocably) tilt in favour of one of the two: notwithstanding their natural preference, ‘A consumers’ will start opting for B too. This phenomenon is called tipping and creates a lock-in in favour of B, which becomes the de facto standard. It is also important to note that it is impossible to predict the market outcome beforehand (read: on the basis of exante knowledge of the technologies’ intrinsic qualities). In Arthur’s terminology: the process is ‘path-dependent’ – that is, “it depends on the cumulation of random events that occur as the process unfolds” [2, p. 124]. And also: “[S]mall chance events early in the history of an industry or technology can tilt the competitive balance” (Arthur [3, p. 80]). Hence the importance of an ‘initial advantage in adoptions’ or ‘early lead’ [2, p. 116]. 30

Katz and Shapiro [53] even use the terms installed-base effects and network externalities as synonyms: “It is commonly believed that the presence of these installed-base effects, known as network externalities. . .” ([53, p. 56], my emphasis). 31 Obviously, if this expression is negative for all i, consumers will buy none of the brands. 32 Cf. [3, p. 80]: “. . . positive feedback magnifies the effects of small economic shifts . . . If one product or nation in a competitive marketplace gets ahead by ‘chance’, it tends to stay ahead and even increase its lead” (ibidem). 33 Note that the learning effects studied by Arthur are not the only possible source of increasing returns to adoption; another one is network externalities [2, note 7, p. 126; 26, p. 1083; 27, pp. 5, 6].


151

Obviously, the above conclusions are too extreme and need to be qualified. One only has to take a look at reality to see that not all markets where network externalities – or, more generally, increasing returns to adoptions – are present, are characterised by an all-or-nothing outcome. For one, Liebowitz and Margolis [61] show that the existence of network effects34 is in se neither a necessary, nor a sufficient condition for the occurrence of an ‘either/or equilibrium’: “The nature of the equilibrium, either as a mixed format or as an either/or equilibrium, depends on the slopes of the net value curves, and synchronization effects are only part of the story”. The ‘net value’ referred to by Liebowitz and Margolis is simply the consumer surplus a + b(ne ) − p. Liebowitz and Margolis show that if this surplus increases with ne , then the outcome will indeed be an “either/or choice that is often argued to be the expected outcome for standards” [61] – but only then. It is important to note here that the slope of the surplus curve is influenced at once by network effects – via b(ne ) – and by economies of scale in production – via p. This implies that surplus curves can be upward sloping in the absence of network effects too and, thus, that economies of scale alone may suffice to give rise to an all-or-nothing outcome35 . At the same time, the above also implies – and this is what Liebowitz and Margolis are hammering at – that it is possible to conceive of situations in which the mere existence of network effects is not sufficient: “The existence of synchronization effects, the raison d’être of standardization, [. . .] does not rule out the possibility of downward sloping net value curves, and the resulting efficient coexistence of formats” [61]. Liebowitz and Margolis underline that “synchronization effects may coexist with increasing, decreasing or constant returns to scale” and that it is quite feasible that the network effects may be overpowered by decreasing returns in production36 . In short, the shape of the cost curve is also of importance. In close connection with this, it has to be noted – and this is a second qualifying remark – that the shape of b(ne ) is important too. Whether there is room for two or more incompatible networks or not, depends in part on the answer to the question whether the use of the good under consideration is subject to ‘inframarginal externality’. 34

In their article, Liebowitz and Margolis use the term ‘synchronization effects’ instead, as they specifically consider the case of the so-called standard-embodying goods like video recorders. 35 Besen and Farrell [8, p. 118] point out that this analogy between economies of scale and network effects should come as no surprise since network effects can be thought of as demand-side economies of scale; similar remarks can be found in [52, p. 824] and [47, p. 125 and note 16, p. 128]. 36 In another paper, Liebowitz and Margolis [60] extensively criticise the network literature on the ground that the theoretical models mostly not only assume a given level of fixed production costs, but also constant marginal costs – the combination of which installs inexhaustible potential economies of scale. In the opinion of Liebowitz and Margolis, this emphasis on economies of scale is excessive (even for the high-tech products on which the network literature focuses): “Without investigation, it is unreasonable to accept that the law of diminishing marginal product somehow takes a vacation in new-technology industries” [60]. Moreover, Liebowitz and Margolis also argue that the emphasis on economies of scale leads to biased results: “Our point here is not that there cannot be economies to scale, but rather that many of the results associated with network externalities are anchored in the assumption of inexhaustible scale economies” [60]. And also: “. . . special cases are too readily taken for the general network problem”.

152


Liebowitz and Margolis [59, p. 140] define this concept as follows: “For inframarginal externality, the marginal utility of the external activity is zero. Very simply, the affected party is not affected by marginal changes in the externality-causing activity”. Or, more specifically: “Many activities require a critical mass but are not much helped by participation beyond that level . . . . [T]he fact that other people use the same sort of VCR that we use makes a tape rental market available to us, but the marginal benefits of increasing the number of households that own our kind of VCR are likely exhausted now that businesses that rent videotapes are about as prevalent as ones that sell milk” (ibidem). If the positive network effects associated with the use of the good are limited or ‘exhaustible’ in this manner, then a de facto standardisation as in Arthur’s model is no longer inevitable: “. . . where marginal gains of network size are exhaustible at network sizes that are small relative to the market, there is no impediment to the coexistence of multiple networks” [59, pp. 140, 141]37 . A third remark has to do with the fact that network models are, by definition, a simplification of the real world. This inspires Liebowitz and Margolis [59, p. 145] to come out with the following (justified) caveat: “While it is inevitable and probably desirable that economists work with restricted models, we should avoid the presumption that the things that are excluded from these models are unimportant or nonexistent”. Liebowitz and Margolis [61] argue, for example, that consumer tastes may differ (greatly); that is, that their valuation of both the stand-alone utility and the network benefit of a product may diverge38 . All else equal, this increases the probability that rival networks may be able to survive: “Consumer heterogeneity and product differentiation tend to limit tipping and sustain multiple networks. If the rival systems have distinct features sought by certain consumers, two or more systems may be able to survive by catering to consumers who care more about product attributes than network size” [54, p. 105], cf. also [8, note 2, p. 118]: “. . . rival standards may coexist if the disadvantages of being on a small network are, for some users, more than offset by a technology’s intrinsic advantages. An example is the continuing role for the Apple computer operating system in a world largely dominated by MS-DOS” 39 . 37

For the sake of completeness, I should add that Arthur himself – who, as was noted earlier, models the effects of learning-by-doing rather than network effects – already recognised this: “Dominance by a single technology is no longer inevitable, however, if the improvement function r is bounded, as when learning effects become exhausted. [. . .] Increasing returns, if they are bounded, are in general not sufficient to guarantee eventual monopoly by a single technology [2, p. 126]. He also points this out in a more vulgarizing article, cf. the following example on industrial location: “If some location by good fortune attracts more firms than the others in the early stages of [the] evolution, the probability that it will attract more firms increases. Industrial concentration becomes reinforcing. [. . .] If the attractiveness exerted by the presence of other firms always rises as more firms are added, some regions will always dominate and shut out all others. If the attractiveness levels off, other solutions, in which regions share the industry, become possible” [3, p. 83]. 38 In the notation used in this paper, this would imply that both a and b(ne ) would have to carry some sort of consumer group index. 39 Dalle [26, p. 1085] also points out the structural lack of consumer heterogeneity in Arthur’s model. Dalle argues that ‘losing’ technologies may be able to survive in specific market segments. As an


153

A fourth remark is that a situation of ‘lock-in with captive consumers’ can only occur if the decisions of the latter are characterised by a certain degree of irreversibility [72, p. 62]; that is, if consumers incur significant switching costs when switching to a competing product. Obviously, this will especially be the case for durable goods: an early replacement causes the consumer to lose (part of) his investment. It should however be stressed that the concept of switching costs is broader than this; switching costs comprise the so-called relation-specific investments (including investments in human capital) and also possible transaction costs [47, p. 123; 54, p. 94; 56, p. 375]. As pointed out by Klemperer [57, p. 99], a consequence of (significant) switching costs is that “products that are ex ante homogeneous become, after the purchase of one of them, ex post heterogeneous”. Put differently, switching costs impose asymmetric prices between users with and without any sunk investment in an existing technology (Yang [86, p. 7]). Consequently, “. . . rational consumers display brand loyalty when faced with a choice between functionally identical products” (ibidem)40 . A fifth and final criticism on Arthur’s model is due to Dalle [26]. Dalle doubts whether consumers can be perfectly informed about the installed base of competing technologies [26, p. 1087]. Specifically, Dalle argues that consumers can hardly have illustration he points out that while VHS has driven out Betamax, this has not been the case in Colombia, nor for professional users [26, p. 1087]. 40 It should however be noted that – and this, in fact, is a qualification of a qualification – even such brand loyalty induced by switching costs does not necessarily imply that a market once it is occupied, will remain forever closed to newcomers. The models of Klemperer [57] and Farrell and Shapiro [47] show that under certain circumstances switching costs may even stimulate the entry of newcomers. The intuition behind this result is worded by Klemperer [57, p. 100] as follows: “In a growing market [. . .] the incumbent’s first-mover advantage must be set against a loss of flexibility. Having built up a customer base as a monopolist, the incumbent is on the horns of a dilemma. Should it charge a high price that capitalises on its repeat purchasers but restricts it to a small share of the growing market, or should it charge a low price that attracts new customers but does not take advantage of its current customer base?”. The higher the switching costs, the more profitable the first option for the monopolist, but the easier it becomes for potential newcomers (ibidem). In short, as Klemperer puts it lapidary: “Switching costs not only lock in customers but may also lock in an incumbent to its current customer base” [57, p. 115]. Farrell and Shapiro [47, p. 124] call this the “fat-cat effect”: the former monopolist is content with serving (read: exploiting) his existing clients, leaves the majority of new consumers to the new entrant, and thus gradually loses market share. However, this scenario only holds under certain conditions. For one, both models referred to assume (1) that the market is growing (so that there is a steady supply of new, unattached consumers), and (2) that there is no (or insufficient) possibility for companies to discriminate between new and existing clients on price. The cost structure of the industry under consideration, and more specifically the intensity of the economies of scale, is, once more, of crucial importance, too; cf. [47, p. 125]: “When economies of scale are great [. . .] there is no entry in equilibrium. The switching costs protect the incumbent from the entrant’s competition for attached buyers, while the economies of scale make it unattractive for the entrant to enter and serve only the unattached buyers as well as unattractive for the incumbent to set a price that encourages him to do so”. Farrell and Shapiro add that this holds equally well when the term ‘economies of scale’ is replaced by ‘network externalities’ – its counterpart on the demand side [47, p. 133]. Summing up: switching costs alone do not form a barrier to entry, only in combination with economies of scale – or network externalities – do they enable an incumbent firm to exclude competitors.

154


a global picture, implying that their information will be chiefly of a ‘local’ nature. Consumers will thus base their purchase decision on what their ‘neighbours’ – read: the other consumers within their horizon – have done. Dalle therefore presents a diffusion model (of a completely different inspiration from Arthur’s), in which he tries to add a spatial dimension to Arthur’s time dimension. In Dalle’s simulations both standards survive: each occupies a number of ‘zones’ from which it has completely expelled the other41 . The above string of qualifying remarks on Arthur’s model clearly shows that the initial, extreme conclusion must be lowered to a more tentative “all else equal, markets characterised by the existence of network effects are more prone to tipping”. The big question then is: is the electronic purse market prone to tipping, or not? 42 To address this question, let me use my string of remarks as a checklist. In my view, the first condition, brought forward by Liebowitz and Margolis [61], needs little reflection. It needs no argument that payment systems exhibit significant economies of scale; one need only think of the central computer infrastructure [65, p. 11]. On top of this, electronic purses also exhibit strong network effects. Hence, the slope of the surplus curves will undoubtedly be positive. The second condition – the network effects are not exhausted quickly – appears to be satisfied too. Electronic purses are meant to be an alternative to currency (and are also marketed as such). Hence, consumers will expect to be able to use their purses almost everywhere43 . Put differently, if each time he wants to buy a paper at a newsstand, the holder of an electronic purse has to wonder whether his electronic purse will indeed be accepted, and if his hopes are regularly deceived, then his valuation of the card will certainly diminish. Having to search for a store that accepts the card is frustrating at the least44 . The alternative would be to carry around some cash at all 41

Note that the concepts of ‘neighbours’ and ‘zones’ need not be interpreted strictly in their spatial sense as Dalle’s example about the use of Betamax in the television industry (footnote 39) shows. In this case, the intense exchanges – rather than the spatial proximity – make the group of professional users a coherent (minority) market segment [26, p. 1093]. 42 I should stress that in this paper, electronic purses comprise only the so-called multi-purpose cards which can be used to pay for a wide range of goods and services on a large geographical scale and in an open system – a system with multiple service providers and one or more independent card issuers [79]. It goes without saying that the so-called single-purpose cards, which can only be used to pay for one specific service (telephone, public transport, . . .), can coexist without difficulty; although they do of course render less attractive a ‘real’ electronic purse. 43 Cf. [66]: “In payment systems, as with local telephone service, consumers demand ‘universal service’ ”. Empirical evidence to underpin this can be found in the results of a survey on Internet money that was carried out in 1995 among Internet users. Wide acceptability proved to be significantly more important than any other attribute of money [76]. 44 Cf. Worthington [84, p. 33]: “To warrant [charging consumers], any electronic purse is going to have to offer consumers a truly ubiquitous acceptance, for otherwise consumers are being asked to pay for the privilege of loading their own cash on to an electronic purse that cannot offer them the same ubiquity as their cash”. Note that some analysts have argued that a regional introduction of an electronic purse – as in the case of the Proton card, which was first introduced in the cities of Louvain and Wavre – is not the best option. Rather they argue that a ‘function-by-function’ introduction (e.g., a card that can be


155

times, but this would mean that one of the main purported advantages of electronic purses – no need to walk around with bulky pockets full of change – would disappear. In my opinion, it is thus safe to presume that the marginal utility created by an additional outlet that accepts the card will not drop to zero quickly – and certainly not “at network sizes that are small relative to the market” (cf. supra). Seen from a different angle: if not for this reason, why would one have made currency legal tender?45 Thirdly, it seems rather doubtful whether differences in consumer tastes can increase the possibility that two or more rival electronic purses may continue to share the market. Card issuers might of course add a number of extra functions to their electronic purse – and, in this way, differentiate it from the competing purse(s) – but the primary function of an electronic purse will always remain the payment function; and for this, what counts is the size of the installed base. My fourth remark is related to switching costs. For merchants, the investment in hardware may perhaps not be all that big, but it is not insignificant either46 : a Proton terminal, e.g., costs some 15.000 BEF. For consumers direct costs remain altogether limited: most banks charge a fixed annual fee of 200 BEF. However, as was noted above, switching costs comprise more than just the direct investment; transaction costs etc. are also important. From the literature about credit cards it can be gathered that total switching costs for consumers are indeed significant in the case of electronic purses. In the credit cards literature, switching costs are generally considered to be an important explanation for the stickiness of credit card interest rates [4,15,16]. According to Ausubel [4, p. 69], a (potentially) dissident consumer is faced with the following search and switch costs in the credit card industry: “(a) the information cost of discovering which banks are offering lower interest rates; (b) the cost in time, effort, and emotional energy in filling out an application for a new card (and possibly getting rejected); (c) the fact that the card fee is usually billed on an annual basis, so that if one switches banks at the wrong time, one foregoes some money; (d) the perception that one acquires a better credit rating or a higher credit limit by holding the same bank’s card for a long time; and (e) the time lag between applying for a card and receiving one”. (Items (b), (c), and (e) are also mentioned by Calem [15, p. 12].) Except for item (d) this list can easily be transposed to the case of electronic purses. Moreover, one has to keep in mind that a (reloadable) electronic purse has to be connected in one way or another to a current account. In certain configurations, used at all parking meters throughout the country) would result in faster adoption (“ ‘Chipknip’ krijgt tegenwind in Nederland”, Financieel Economische Tijd, October 24, 1995). In other words, a complete geographical coverage of one (limited) function would be better than an incomplete coverage of multiple functions. The underlying reasoning seems to be that in the first option the consumer’s uncertainty is reduced. 45 In this respect it is interesting to point out that under the Danish ‘Payment Card Act’ of 1984 a card accepting telephone must have next to it a coin accepting telephone, and that the issuer of Danmønt – the Danish electronic purse – had to apply for exemptions from this regulation before being able to embark on a trial [28, p. 73, 75]. 46 This is especially true for small businesses [29, p. 39].

156


switching to a competing electronic purse might thus imply having to change banks too (provided that one wants to avoid holding accounts at two banks)47 . It is self-evident that changing banks entails significant transaction costs [15, p. 12; 56, p. 375; 57, p. 99].48 . Finally, concerning Dalle’s [26] criticism that the installed base is only imperfectly observable to the average consumer, it can be noted that firms know all too well that the visibility of their network is of the utmost importance and that they thus will leave nothing undone to actively promote it. This is certainly true for payment systems; cf. Matutes and Padilla [64, p. 1113]: “A cursory look at financial magazines, newspapers, and banks’ own brochures reveals that banks heavily advertise the size of the [ATM] network that their customers can access” 49 . Dalle’s remark about the importance of the spatial dimension in the diffusion of a technology, on the other hand, does raise two questions. The first is: to what extent might the market for an electronic purse be geographically limited? And, thus, the second question is: would it not be possible after all to see two or more electronic purses operating in the same country, each within a separate geographical area? Judging from the following remark by Linkens et al. [63], one would be inclined to think that this might indeed be possible: “Rappelons [. . .] la vraie nature du porte-monnaie e´ lectronique: les achats de tous les jours (pain, ticket de bus, journal, . . .), par des habitants dans leur voisinage proche” ([63, p. 279]; my emphasis). Translated literally: ‘Recall . . . the true nature of an electronic purse: payments for everyday purchases (bread, bus tickets, newspapers, . . .) by residents in their nearby neighbourhood’. The big question then of course is: what exactly should one understand by ‘voisinage proche/nearby neighbourhood’? In this respect it can be noted that Banksys [7, p. 26] – the interbank organisation that developed the Belgian Proton card – underscores that “all opinion surveys show that 99% of electronic purse 47

Compare with what McAndrews [66] writes on the subject of ATM networks: “. . . the cost to a consumer of changing ATM network affiliation may not exceed the benefit because ATM network access is a relatively small consideration for a consumer of a bundle of banking services, which may consist of both savings and demand deposits, certificates of deposits, and auto and home loans. If one is unhappy with the ATM network to which one has access, but happy with all the other services of one’s bank, there is a large cost to getting access to the rival network since the customer would have to incur the cost of changing banks or, at the very least, establishing an account at a different bank (and, therefore, holding accounts at two banks). This cost may exceed the inconvenience of the ATM network that one’s bank offers, and so the consumer may not switch to the better ATM network”; cf. also [67, p. 57]. 48 I should also mention that in the case of electronic purses issuers clearly can find ways to introduce price discrimination between new and existing consumers. It thus seems rather improbable that a monopolist would change into a ‘fat cat’ as time goes by (see footnote 40). 49 See also Katz and Shapiro [54, p. 107]: “A firm in a systems market has strong incentives to build up consumer beliefs about its own system, and to tear down consumer beliefs about rival systems, in trying to tip the market in their favor”. Katz and Shapiro take one of their examples from the payments card industry: “. . . Visa has had a long-running advertising campaign telling consumers that Visa cards are accepted ‘everywhere you want to be’, whereas merchants ‘don’t take American Express’ ” (ibidem); see also [68, p. 20]. Besen and Farrell [8, p. 122] point out that in some markets there is scope for puffery, “since appearances may count as much as does reality”.


157

transactions will occur within 50 miles of the cardholder’s house”. Banksys seizes upon this result to argue that it makes little sense to equip an electronic purse with a multi-currency functionality for use abroad (as Mondex has done); cf. Linkens [62, p. 79] and Linkens et al. [63, p. 279]. In short, according to Banksys, the market for an electronic purse does not exceed, or hardly so, the borders of a country. However, the question I need to answer here goes in the opposite direction. Providing a definitive answer would require data (or forecasts) on the relative importance of card use within the radii smaller than the 50 miles mentioned above. Nevertheless, in view of the ever increasing mobility and the importance of commuter traffic, and considering the fact that most electronic purse projects see pay phones, parking meters, public transport, and catering services as interesting markets, it can in my opinion be concluded that even in a spatial sense there seems to be little or no room for several full-fledged and incompatible electronic purses – at least not within the borders of a small country like Belgium50 . Summing up, all elements – important economies of scale, strong network effects, no inframarginal externality, largely homogeneous goods, fair visibility of the installed base, and a minimum required geographical scale – point in the same direction: the electronic purse market does indeed appear to be a ‘0/1-market’. Obviously, in such a market competition will be of a special nature and will require issuers to follow special strategies, primarily aimed at building up an early lead (Perrot [72, pp. 60–62]), since “an early-start technology may already be locked in, so that a new potentially-superior arrival cannot gain a footing” (Arthur [2, p. 123]). 4.3. A first-mover advantage? The first, and most obvious strategy to get an early lead simply consists of entering the market before others do. However, the rational consumers of the network models are not easily roped in. In a market where switching costs are important, consumers are very afraid to join the ‘wrong’ network, because “buyers who join what turns out to be a losing network must either switch, which may be costly, or else content themselves with smaller network externalities than those associated with the winner” (Besen and Farrell [8, p. 118]). This implies that a first-mover strategy will only be effective if the general public is convinced that the network will eventually attain sufficient coverage or, in an all-or-nothing market, that the network is able to get a 50

A serious indication consistent with this conclusion is that an evaluation of the Proton pilot project pointed out that two out of every four inhabitants of the cities of Louvain and Wavre had the intention to procure a Proton card, but that they preferred to wait until it was launched on a national scale. The reason for this was that they primarily do their shopping outside of the two pilot cities (Linkens [62, p. 78]). The Mondex pilot in the Canadian city of Guelph seems to encounter the same difficulty; cf. Guelph city administrator [David] Creech who notes “that one of the big problems with a pilot test in a place like Guelph is that many people who work in the city live elsewhere, and many who live in town work in Toronto, 80 kilometres away. To them, Mondex isn’t ubiquitous enough to make it attractive” (Blackwell [10]).

158


hold of the entire market51 . Conversely, “if consumers expect a seller to be dominant, then [. . .] it will, in fact, be dominant” (Katz and Shapiro [51, p. 425]) – even if the seller enters the market a bit later52,53 . In short, a firm’s reputation is of crucial importance; cf. Katz and Shapiro [54, p. 104]: “In markets where network effects are present, a firm may benefit from having a reputation for selling ‘successful’ products. Casual observation suggests that one reason that the IBM PC was so successful is that consumers expected the product to succeed since it was backed by IBM” (see also [51, p. 439]). According to Besen and Farrell [8, p. 118], a good reputation may even compensate for possible handicaps: “. . . , victory need not go to the better or cheaper product: an inferior product may be able to defeat a superior one if it is widely expected to do so. For example, the initial success of MS-DOS is usually attributed not to any technical superiority, but to the fact that it was supported by IBM”. In the specific case of the electronic purse market, it is clear that thanks to the success of their debit card networks, companies such as Banksys (in Belgium), Interpay (in the Netherlands), etc. have an advantage over possible initiatives taken by, e.g., the retail sector (such as the Dutch Primeur Card)54 . Obviously, another strength of large interbank organisations such as Banksys and Interpay is that they have the necessary means and infrastructure to back up a large-scale launch of their electronic purses. According to Rosalie Zobel of the European Commission, a “relatively fast and massive deployment” would, for that matter, be the only way to come out of the chicken-and-egg-deadlock [86, p. 5]. Zobel is also convinced that one bank on its own will not be able to carry things to a happy conclusion: “If the [electronic purse] is to 51

Cf. Caskey and Sellon [19, p. 84]: “[A] consequence of network externalities is that many potential users of the product might decide to wait for it to attain some initial success before entering the market. This delay occurs because early adopters will see few benefits from the product until its use is widespread. If a sufficient number of consumers adopt a wait-and-see attitude, there may be insufficient demand to launch the product successfully”. 52 See also Katz and Shapiro [54, p. 94]: “. . . , systems that are expected to be popular – and thus have widely available components – will be more popular for that very reason”. 53 Note that new technologies or products do not necessarily enter the market unannounced. It is conceivable that consumers are aware – perhaps as a result of the so-called product preannouncement – of the fact that the first entrant will not remain the only one on the market for long. This can compromise the first mover’s success: “. . . a product preannouncement in an industry with strong network externalities, . . . , can deter entry by preventing an alternative network from gaining a large enough base of members to make it an attractive alternative. Such an announcement can work in this way if it succeeds in convincing enough participants to delay joining any network other than the preannounced one. In other words, if enough participants anticipate the bandwagon effect in the preannounced product, they can reduce the possible bandwagon for competing products” (McAndrews [65, p. 15]); see [46, pp. 948–949] for a model of this. Note also that in the United States there have been lawsuits in which companies such as IBM were accused of using the so-called premature or predatory announcements in an effort to block competitors’ ways [46, p. 942; 65, p. 15]. 54 Cf. Choi et al. [21]: “. . . , reputation built in physical markets can be transferred to the electronic marketplace. As a result, the company with reputation as a dominant and large firm in physical markets enjoys a form of size advantage in electronic commerce as well. In this vein, SET, the electronic payment system for credit cards favored by Visa and MasterCard, has the advantages of these firms’ reputation and size in physical markets”.


159

be successful as a mass market retail application, it is unlikely that any single bank or financial institution alone will be able to gain any ‘first mover’ advantage. As a result, being an early entrant will not provide significant tangible financial benefits. The financial community will need alliances, within the industry and probably across industries, to generate critical masses of investments to create a market from scratch” (ibidem). 4.4. Compatibility In close connection with the above, it can be noted that a proliferation of incompatible systems may very well prove detrimental: it creates uncertainty among consumers and merchants [86, p. 4]55 and it cuts up the market; cf. Caskey and Sellon [19, p. 84]: “. . . , if there are several producers of a product subject to positive network externalities, people may resist using the product unless some or all producers agree to use compatible technological standards. Without such compatibility, the network of users for each producer’s product might be too small to make the product sufficiently useful”. Tijdens [77] is on the same wavelength; she points out that, in payment systems, cooperation on a sectoral level may be a conditio sine qua non for an innovation to be successful. According to Tijdens, this is especially true when standardisation of the product is a prerequisite to penetrate the market [77, p. 13]. A comparison with the introduction of debit cards may clearly be of some guidance here. Where Belgium is concerned, it is clear that the genesis, in the late 70s and early 80s, of two separate nationwide networks – Bancontact and Mister Cash – was anything but conducive for the adoption of this new means of payment [30, pp. 28 and 30; 32, p. 135; 48, p. 285]. Conversely, the compatibility agreement between Bancontact and Mister Cash has clearly given a big push to the use of debit cards in Belgium56 . Other countries have had similar experiences. Caskey and Sellon [19], for example, argue that the limited success of debit cards in the United States is largely due to the fact that the cards could initially only be used within the proprietary network of the bank where the card holder had its current account. And concerning New Zealand, Ledingham [58, p. 348] states that “the original EFTPOS developments demonstrated that competing alternatives can sometimes delay progress because retailers – and others – do not want to have to deal with a number of incompatible systems” 57 . Where Belgium is concerned, there is little reason to fear that a battle between incompatible electronic purses would inhibit the adoption of the new means of payment. So far the Proton card is the only one on the market and the situation is likely to stay like this because Banksys is an interbank organisation comprising the majority of 55

See also West [82, p. 3]: “Th[e] multiplicity [in designs] is a major barrier to development. It makes selection difficult and encourages players to wait rather than risk choosing the wrong design”. 56 One example is the relative increase in use of debit cards in the Delhaize supermarkets over the period 1987–1989 [32, figure 1, p. 138]. 57 See also Horvitz and White [50, p. 11]: “Merchants may be willing to accept stored-value cards issued by many different banks or non-banks, but not if they need a different terminal for each card”.

160


Belgian banks. In other countries, however, competition between, say, Mondex and Visa Cash might very well result in years of slow adoption – unless they agree on a common standard. If neither company pulls ahead within the next few years, a collaborative Mondex/Visa standard seems rather inevitable. In the Netherlands, the two local rival purses – interbank organisation Interpay’s Chipknip and the Chipper card backed by PTT Telecom and the Postbank – have recently agreed to make use of the same infrastructure58 . The network literature pays a great deal of attention to the motives that may induce companies to make their products compatible with other products on the market. This should hardly be surprising as this is a strategic decision of the utmost importance59 . Note, however, that not all procedures studied in the literature are relevant to the specific case of the electronic purse. In some markets, a company can act unilaterally to make its product compatible (for example, by constructing an adapter or a converter). It goes without saying that in the case of payment systems such a procedure will not work: the Belgian Postomat network, for example, cannot start issuing debit cards that can be read by Bancontact/Mister Cash terminals without having the approval of the latter to do so60,61 . In the case of payment card systems, compatibility can only be attained via a joint decision [64, p. 1118]. When taking this decision, a firm has to weigh two opposing effects on its profitability [33, pp. 8, 9; 85, p. 916]. The positive effect is that after the unification of the networks, the firm’s product will be valued on the basis of total installed base rather than on its individual installed base alone62,63 . As a result the consumers’ willingness to pay will increase. This is the network effect. The negative effect is that the firm will have to face a more intense price 58

“Samenwerking tussen chipknip en chipper in Nederland”, Financieel Economische Tijd, January 22, 1998. 59 Cf. Katz and Shapiro [51, p. 434]: “When the network externalities are large, the choice of whether to make the products compatible will be one of the most important dimensions of market performance”. 60 A good real-life illustration is the attempt of Argenta (a small Belgian savings bank) to provide its customers with cards that would have two magnetic stripes: one for the Banksys network and one for the Postomat network. This, however, ran up against Banksys’s veto – supposedly for security reasons (“Argenta wil met Postomat monopolie Banksys doorbreken”, Financieel Economische Tijd, April 30, 1994; “Banksys weigert Argenta-kaart met dubbele magnetische strook”, Financieel Economische Tijd, December 29, 1994). For points of view on the antritrust implications of payment networks mergers and on the welfare implications of refusals to open up such networks for members of other networks, see [5,6,17,18,34,66,67]. 61 See [51, pp. 436–437], and [46] for a theoretical analysis of the decision to fit one’s product with a converter. Economides [33] and Encaoua et al. [43] also model situations where a company can unilaterally make its products compatible (and where the other company cannot prevent this). 62 “The central feature of the market that determines the scope of the relevant network is whether the products of different firms may be used together” (Katz and Shapiro [51, p. 424]). 63 In this respect the definition of network effects given by Katz and Shapiro [51] (cf. supra) is not entirely accurate. The definition given by Church and Gandal [23], for example, explicitly indicates that it need not necessarily concern the same good: “The benefit received from the consumption of a particular good often depends on the aggregate number of consumers who elect to purchase compatible goods” ([23, p. 239], my emphasis).


161

competition due to the lower product differentiation (or, put differently, the increased substitutability)64 . This is named the substitution effect or the competitive effect65 . In the network literature, the trade-off between network and substitution effects is analysed predominantly in a duopolistic framework. The main conclusions of the different duopolistic models [45,47,51,52] can be summarised as follows. The first important conclusion is that a distinction has to be made between the situation where the two firms are of equal strength (the symmetric case) and the situation where this is not the case (the asymmetric case). In the symmetric case, both firms will always benefit from a move to compatibility. The size of the potential benefit – and this will determine whether the firms will effectively make their products compatible – depends on the strength of the network externalities. If the externalities are weak, the move to compatibility will only have a modest impact on the consumers’ willingness to pay. Hence, the game will not be worth the candle (since a move to compatibility obviously entails costs too). In the asymmetric case, the two firms have conflicting interests: the strongest firm – the firm with the largest network, with a better reputation, or with a clear technological advantage – will oppose compatibility, whereas the weaker firm will be in favour of compatibility66 . If compatibility can be achieved only through a joint decision – as is the case for electronic purses – the logical outcome will be that the two networks will remain incompatible67 . The underlying intuition is selfevident: the dominant firm is in favour of incompatibility because a merger with the smaller network will have little or no effect on the valuation of its product, so that the network effect will be outweighed by the adverse effect of increased competition. In Belgium, for example, Banksys has little to gain (and a lot to lose) by merging its Bancontact/Mister Cash debit network with the (much smaller) Postomat network68 . 64

Cf. Besen and Farrell [8, p. 121]: “. . . , although economic theory is not conclusive, most analysts believe that price competition is more intense when vendor’s products are compatible, both because product variety is reduced and because users are less likely to be locked-in to a single firm’s product”. 65 Note that when companies offer a bundle of goods – banks being a typical example – the more intense price competition need not happen solely on the market of the product that has been made compatible. There may be downward pressure on the prices of other goods in the bundle too. Matutes and Padilla [64], for example, model the decision of banks to share their proprietary ATM networks with competitors. Matutes and Padilla point out that “when two banks have a compatible network, depositors can credibly threaten their banks to switch to a rival offering a better [deposit] rate and still benefit from their ATM services” [64, p. 1120]. In short, as banks have become better overall substitutes, the elasticity of the supply of deposits increases, which may lead to more intense price competition on the deposits market. 66 Katz and Shapiro [54, p. 111] point out that this may also be the case for products that are not yet on the market: “. . . since systems competition is prone to tipping, there are likely to be strong winners and strong losers under incompatibility. Therefore, if a firm is confident it will be the winner, that firm will tend to oppose compatibility”. This confidence may be based on the firm’s reputation, on the fact that consumers have a strong brand preference for its products, etc. 67 Note that the outcome will be different when non-cooperative behaviour is possible (a situation which is not really relevant here); cf., e.g., Encaoua et al. [43]. 68 Note that these conclusions apparently continue to hold when one explicitly allows, as Esser and Leruth [44] do, for a difference in ‘basic’ quality between the goods of the two firms – a difference which may

162


It should however be stressed that it was assumed throughout the preceding analysis that firms cannot make side payments to one another. It is intuitively clear that if they can make side payments – and this certainly is the case for payments systems69 – compatibility will be achieved more often. For this to happen, it suffices that “the change of the profits within the set that can make side payments to one another exceeds the joint costs of compatibility” (Katz and Shapiro [51, p. 434], my emphasis), whereas in the absence of side payments the products of two firms will only be made compatible if either firm benefits from the move (ibidem)70 . A number of other papers study not so much a duopolistic market but rather point out that (challenged) monopolists have to make compatibility choices too. Economides [36] tries to determine whether – in a market with network externalities – it cannot be profitable for a monopolist to invite one or more competitors on his market, even though he is the exclusive holder of a technology. Xie and Sirbu [85] want to know what the optimum strategy of a monopolist/incumbent is when challenged by a new entrant. Needless to say, there are large parallels between these problems and those mentioned above. The conclusions are thus fairly similar. To start with the article by Xie and Sirbu, it will not come as a surprise to learn that a new entrant is better off if its products are compatible with those of the incumbent, and this especially when the network externalities are strong and the installed base of the incumbent is large. A somewhat less intuitive result is that, under certain circumstances, the incumbent as well may be better off by permitting compatibility. The latter depends – once again – on the relative strenghts of the network and substitution effects. The simulation results of Xie and Sirbu show that the first effect will overshadow the second if the product under consideration exhibits strong network externalities and if the entry takes place before the incumbent has built a large installed base. This is because early in the product life cycle [85, p. 924] the network externalities can still play to the full. In all other cases, the former monopolist will realize lower profits under compatibility71 . Economides [36] shows that it is quite possible that a monopolist/innovator does not succeed in generating sufficient sales and sufficient network externalities on its own (the reader is referred to [36, p. 3] and [51, p. 431] for further explanations on the theoretical background of this conclusion). Provided that the network effects are sufficiently strong, it can therefore be profitable for a monopolist to license its technology (without be magnified by “an externality given by the number of consumers of each good” [44, p. 252]. Esser and Leruth show that the strongest firm will prefer incompatibility because “[it] benefits so much from the externality that the (large) size of its own market is sufficient to insure high profits” [44, p. 265]. 69 In shared ATM networks, for example, member banks charge the so-called interchange fees for use of their ATMs by depositors of other member banks. 70 Here too it is assumed that compatibility can only be achieved via a joint decision. If firms can act unilaterally to make their products compatible and if side payments are not feasible, then it is sufficient – in a duopolistic market – that one of the two firms finds the move profitable to achieve a situation of full compatibility. 71 When transposing these results to the case of electronic purses, one caveat has to be kept in mind: Xie and Sirbu consider a market for durable goods, without repeat purchases (cf. my remarks in footnote 24).


163

charge). In markets with very strong network externalities, he will even benefit by paying a subsidy to the invited followers. If there are more than two firms on the market, the compatibility problem becomes much more complex because coalitions can be formed. So far, very little research has been done on this issue [35, p. 18]. I am aware of two studies: an article by Matutes and Padilla [64] for the three firms case, and a paper by Economides and Flyer [37] for the n firms case. The latter paper, however, concentrates primarily on non-cooperative behaviour (which is irrelevant for the case studied in this paper), and will therefore not be discussed further here. Another reason why the article by Matutes and Padilla is more relevant is because it specifically deals with ATM networks of banks who are competitors in the market for deposits. In their model, Matutes and Padilla consider three banks with ATM networks of equal size. A first striking result, which contrasts sharply with the results of the duopoly models, is that the banks – despite the ex ante symmetry of their networks – will never agree on full compatibility: in equilibrium, either partial compatibility or total incompatibility will prevail72 . The explanation for this is the following: “If all banks share their networks none of them obtains a network advantage so that the only effect of compatibility is to reduce the effective degree of horizontal differentiation among banks. As a result, competition gets tougher and banks are unable to internalise the positive network externality which is then entirely appropriated by depositors. It follows that two compatible banks would always veto the entry of a third bank into their common network” [64, p. 1122]. Matutes and Padilla do, however, make mention of two qualifications. First is that “if banks were able to collude in deposit rates, full compatibility would be observed in equilibrium because banks could then fully extract the additional surplus generated by compatibility” [64, p. 1124]. A second remark is similar to a remark made for the duopoly framework: the existence of interchange fees increases the possibility of a full compatibility outcome, because by using such fees banks can limit the substitution effect of compatibility and appropriate part of the network externality that they generate on depositors [64, p. 1115]. The question of course remains which factors determine – in the absence of collusion and interchange fees – which of the two possible outcomes, partial compatibility or total incompatibility, will come about. The answer will not come as a surprise: it will depend on the extent to which depositors value the greater accessibility to their funds [64, p. 1123] – or, put differently, on the strength of the network externalities. If the latter are sufficiently strong, depositors will be willing to accept lower interest rates [64, p. 1114], or to pay a higher fee for their debit card; hence, the profits of the banks willing to share their networks will increase. 4.5. Pricing strategies In view of the above, it will be clear to the reader that in markets with network externalities the pricing in the early stages of the product life cycle is of crucial 72

Matutes and Padilla conjecture that this will also be the case with four banks; that is, at best two competing networks will coexist [64, note 22, p. 1124].

164


importance. This holds both for a monopolist and for duopolists (or more generally: competing firms). The monopolist has to convince a critical mass of consumers to get the bandwagon rolling, whereas competing firms will want to establish an installed base ahead of their rivals. Let me start with the case of the monopolist. Both the models of Dhebar and Oren [31], Cabral et al. [14], and Xie and Sirbu [85] show that when a monopolist introduces a new product on the market that exhibits strong network externalities – as is the case of electronic purses – the monopolist has a clear incentive to use penetration pricing (below cost) or, less extreme, introductory pricing (lower markup than normal) in order to attract as many early adopters as possible; cf. Cabral et al. [14, p. 7]: “the seller has an incentive to lower first-period prices so as to discourage buyers from delaying first-period purchases”. As mentioned earlier, rational consumers are afraid of joining a network that in the end would prove to be too small. Put differently: “. . . inducement of a low first-period price is needed to compensate for the uncertainty of an early adoption” [14, p. 3]73 . Under certain circumstances, this low initial price can also serve as a signal of low cost, “thus raising early buyers’ expectations about the likelihood of future sales” (ibidem). Once the bandwagon gets rolling, consumers’ willingness to pay will increase substantially – in the case of strong network externalities, that is – so that in later periods the price can be raised74 . In short, the optimal strategy for a monopolist is an increasing price trajectory: “For producers of products with significant positive demand externalities, the optimum strategy in these markets is to price low – even below cost – initially and raise price over time as consumer valuation of product benefit increases with the installed base” (Xie and Sirbu [85, p. 924])75 . When externalities are weaker, the optimal strategy is a skimming strategy: “When demand externalities are weaker, the installed base has very little effect on the potential demand. Thus, the optimal strategy is to start pricing high, extracting surplus from those with higher willingness to pay, and then decreasing price monotonically over time” [85, p. 919]. The pricing strategy in the early stages is also of crucial importance in a com73

Katz and Shapiro [54, p. 102] point out that a monopolist could completely eliminate the fear of joining a losing network “by insuring potential buyers against the possibility of a small, low-value network. This would be accomplished by making the price paid by any one consumer contingent on the overall size of the network”. In the case of payment cards, however, this option is a purely theoretical one. 74 Cf. Katz and Shapiro [54, p. 102]: “In essence, offering users access to a larger network is like offering them a better product”. Perrot [72, p. 64] notes that a strategy of intertemporal price discrimination essentially internalises (part of) the network externalities (which are limited in the early stages, but increase gradually afterwards). 75 Perrot [72, p. 64] points out that network goods require ‘binomial’ (that is, two-part) fees rather than linear fees. She also argues that a reduction of the fixed component is the appropriate way to stimulate the growth of the network, because one has to stimulate not so much the intensity of use of the network but rather the number of users. In view of the fact that the network effects are of an indirect nature, I would argue that this does not hold in the case of electronic purses. Having a lot of consumers who possess an electronic purse is not sufficient; for consumer adoption to have an impact on merchant acceptance, people will have to use their cards.


165

petitive setting. In a market with network externalities, there will be an aggressive competition aimed at market share in the early periods. As Katz and Shapiro [51,52] show in a duopolistic model, “by winning in period 1, (a firm) faces a rival that is less attractive to consumers in period 2” [52, p. 835] – simply because the network of the rival firm will be smaller. Katz and Shapiro call this the “weakened-rival effect”. This implies that the latter (say firm Y), in order to be able to attract consumers, will have to lower its price until aY + b(neY ) − pY > aX + b(neX ) − pX , so that the network disadvantage is compensated for by a price advantage. If the network size advantage of firm X becomes too large, firm Y will become unprofitable and will disappear from the market. Reversely, seen from the angle of the winning firm, the above implies – as is shown formally by Xie and Sirbu [85, p. 916] – that “demand externalities and installed base combine to confer market power”, that is, “the ability to charge a higher price than one’s competitor while maintaining a higher sales rate”. This will hold all the more so in the presence of switching costs; cf. [56]. It is this prospect of possible collusive profits which causes the violent competition in the early periods; that is, in periods when the majority of consumers is still unattached. Once the market enters the maturity stage, competition will become less intense [72, p. 62]76 . Real-life examples of intertemporal price discrimination are easy to find. One example concerns the introduction of the Minitel system in France. Initially, Minitel terminals were given away for free. This created a critical mass of early adopters, which made the Minitel system an attractive medium for service providers. The ever increasing supply of services in turn, increased its attractiveness for new users and hence their willingness to pay. Consequently, after some time, Minitel terminals were rented to users [72, p. 64] (for other examples, see [14, p. 1] and [85, p. 924]). Issuers of electronic purses also seem to have recognised the importance of introductory pricing. During the pilot in Louvain and Wavre, for example, Banksys lowered by 50% both the price of Proton terminals merchant commission (the latter was reduced from 0.9% to 0.45%). Moreover, during the launch banks also lowered their tariffs for card holders: customers of the Kredietbank, e.g., got the Proton card for free, while customers of the BBL and the Cera Bank only had to pay a promotional price of 100 BEF (instead of 200 BEF). (Source: De Standaard, February 11, 1995.) In Portugal, some acquiring banks tried to induce merchants to install Multibanco Electronic Purse terminals by offering them a MSC-free period of one year (MSC stands for Merchant Service Charge; this charge normally is between 0.5% and 1.5% of the transaction value). Some banks also provided cards for free to small merchants to distribute to regular customers in their locality in order to encourage take-up [70]. In the Netherlands, Interpay also uses (some sort of) introductory pricing: each card holder receives an extra 5 guilders when loading his or her Chipknip for the first time [29, p. 39; 55, p. 20]. 76

Cf. also Farrell and Saloner [47, note 4, p. 124]: “Intense initial competition followed by feeble ex post competition occurs more generally when a first-period customer base confers a second-period advantage, for instance when learning-by-doing or network externalities are at work”.

166


To conclude this subsection, let me stress that price reductions are not the only way to remove consumers’ fear of joining (and being locked-in on) a ‘losing’ network. Goods can, for example, be rented rather than sold [54, p. 103]77 . In this manner, users need not be afraid of losing part of their capital. Transposed to the case of electronic purses, this approach would obviously relate primarily to terminals (rather than cards). Issuers of electronic purses appear to have recognised this too. During the trial of the Dutch Chipknip, for example, merchants were given the possibility to rent a terminal for 15 NLG per month – instead of buying one for approximately 900 NLG [29, p. 39]. In Belgium, Banksys committed itself to take back the terminals at full price from the merchants if the pilot in Louvain and Wavre would prove to be a failure. (Source: “Banksys loopt warm voor elektronische portemonnee”, FinancieelEconomische Tijd, February 15, 1995.) Note also that both Danmønt (in Denmark) and Avant (in Finland), unlike Banksys, started with disposable cards rather than reloadable cards [79]. It is self-evident that in this case consumers need not be afraid of being locked-in, the more so because disposable cards, by definition, are not linked to a bank account: “Disposable cards . . . give consumers the opportunity to try out the idea of prepayment without further commitment” [28, p. 163]. Note that in a competitive market (where electronic purses are issued by banks), this feature of disposable cards may increase the probability that depositors of other banks can also be tempted to use the purse – because in comparison with reloadable cards, they will not incur switching costs. Finally, let me also mention that issuers of electronic purses also have the potential to increase the switching costs of their card holders in an attempt to turn them into captive consumers. Obvious possibilities are rewarding card holders who actively use their cards by giving them some sort of electronic saving-stamps or cooperating with supermarkets in order to equip the purse with a customer loyalty program [79]78 . Under these circumstances, card holders will be less inclined to change purses because they would then lose the loyalty points accumulated so far. If the electronic purse issuers are banks, they also have the possibility to explicitly bundle their products79 . 77

Cf. also Liebowitz and Margolis [59, p. 143]: “Certainly the commitment problem would not apply to those networks where participants make a payment each period for their place in the network (or rentals for durable goods), since in this case first period consumers need not fear getting stuck with the wrong product in period two”. 78 Klemperer [56] would call this the creation of ‘artificial’ or ‘contractual switching costs’. The main difference between this type of switching costs and the two other types distinguished by Klemperer – learning costs and transaction costs (cf. supra) – is that the first type “arises entirely at firms’ discretion” [56, p. 376]. 79 A good real-life example of this approach seems to be the Full Electronic account of the Belgian Gemeentekrediet. For a fixed fee of 1,000 BEF (initially reduced to 400 BEF), the holder of this relatively new type of current account receives both a debit card, a credit card, and a Proton card. He or she also gets unlimited access to phone banking. As the name of the account shows, there is one additional condition: paper transactions are severely penalised (“Gemeentekrediet verleidt klanten tot volledig elektronisch bankieren”, Financieel Economische Tijd, February 9, 1996).


5.

167

Conclusion

In a recent paper, McAndrews [69, p. 16] argues that “an understanding of the economics of networks and the unique features of network goods gives insight into the organization of markets for these goods and provides the basis for formulating good business and public policy concerning these goods”. This paper has shown that the network externalities theory does indeed provide a useful framework to analyse the introduction and further proliferation of electronic purses. Specifically, the paper has shown that the network externalities literature provides a number of useful insights concerning, for example, consumer reactions to the introduction of electronic purses (and the ways in which card issuers can anticipate these reactions), and concerning the strategies card issuers may follow in a competitive market with incompatible electronic purses. Obviously, some of these ‘lessons’ can hardly be called new. One could even argue that a fair amount of common sense is sufficient to arrive at identical insights. The network literature, however, has the advantage that the insights can be presented in a more formal manner, and especially that all insights can be integrated in a coherent theoretical framework. References [1] Access to payment networks in the Canadian payments system, Background paper for discussion by the Payments System Advisory Committee, Discussion paper Nr. 3, Bank of Canada and the Department of Finance (July 1997). http://www.bank-banque-canada.ca/english/ lvtsgn.htm. [2] W.B. Arthur, Competing technologies, increasing returns, and lock-in by historical events, Economic J. 99 (1989) 116–131. [3] W.B. Arthur, Positive feedbacks in the economy, Sci. Amer. 262(2) (February 1990) 80–85. [4] L. Ausubel, The failure of competition in the credit card market, Amer. Economic Rev. 81(1) (1991) 50–81. [5] D.I. Baker, Shared ATM networks – the antitrust dimension, Federal Reserve Bank of St. Louis Rev. 77(6) (1995) 5–17. [6] D.A. Balto, Payment systems and antitrust: Can the opportunities for network competition be recognized?, Federal Reserve Bank of St. Louis Rev. 77(6) (1995) 19–40. [7] Proton: Belgian electronic purse, presentation of Banksys, in: Card Europe Conf. on Smart Cards in Business, Dublin (22–23 November, 1995). [8] S.M. Besen and J. Farrell, Choosing how to compete: Strategies and tactics in standardization, J. Economic Perspectives 8(2) (1994) 117–131. [9] D.W.G. Birch, Real electronic commerce – Smart cards on the superhighway, in: Conf. on The Future of Money, Palm Springs (January 1996). http://www.hyperion.co.uk/pub/. [10] R. Blackwell, Where cash isn’t king, The Financial Post (September 27, 1997). http://www. cmtcanada.com/FPTechnology/sep27 wherecashi.html. [11] Board of Governors of the Federal Reserve System, Report to the Congress on the Application of the Electronic Fund Transfer Act to Electronic Stored-value Products (March 1997). http://www. bog.frb.fed.us/boarddocs/RptCongress/default.htm#Electronic. [12] C.V. Brown and P.M. Jackson, Public Sector Economics, 4th ed. (Basill Blackwell, Oxford, 1990). [13] L. Cabral, On the adoption of innovations with ‘network’ externalities, Math. Social Sci. 19 (1990) 299–308.

168


[14] L. Cabral, D. Salant and G. Woroch, Monopoly pricing with network externalities (May 1994), Internat. J. Industrial Organization, to appear. http://elsa.berkeley.edu/users/woroch/. [15] P. Calem, The strange behavior of the credit card market, Business Rev., Federal Reserve Bank of Philadelphia (January/February 1992) 3–14. [16] P.S. Calem and L.J. Mester, Consumer behavior and the stickiness of credit card interest rates, Amer. Economic Rev. 85(5) (1995) 1327–1336. [17] D.W. Carlton and A.S. Frankel, The antitrust economics of credit card networks, Antitrust Law J. 63 (1995) 643–668. [18] D.W. Carlton and A.S. Frankel, Antitrust and payment technologies, Federal Reserve Bank of St. Louis Rev. 77(6) (1995) 41–54. [19] J.P. Caskey and G.H. Sellon, Is the debit card revolution finally here?, Economic Rev., Federal Reserve Bank of Kansas City 4 (1994) 79–95. http://www.kc.frb.org/publicat/ econrev/er94q4.htm#Caskey. [20] J.P. Caskey and S.S. Laurent, The Susan B. Anthony Dollar and the theory of coin/note substitution, J. Money Credit Banking 26(3) (August 1994) 495–510. [21] S.-Y. Choi, D.O. Stahl and A.B. Whinston, Cyberpayments and the future of electronic commerce, in: ICEC Cyberpayments ’97, Seoul, Korea (1997). http://cism.bus.utexas.edu/ works/articles/cyberpayments.html. [22] J. Church and N. Gandal, Network effects, software provision and standardization, J. Industrial Economics 40(1) (1992) 85–103. [23] J. Church and N. Gandal, Complementary network externalities and technological adoption, Internat. J. Industrial Organization 11 (1993) 239–260. [24] E.K. Clemons, MAC – Philadelphia National Bank’s strategic venture in shared ATM networks, J. Management Inform. Systems 7(1) (1990) 5–25. [25] K.P. Coyne and R. Dye, The competitive dynamics of network-based businesses, Harvard Business Rev. 76(1) (1998) 99–109. [26] J.-M. Dalle, Dynamiques d’adoption, coordination et diversité: La diffusion des standards technologiques, Rev. Economique 46(4) (1995) 1081–1098. [27] P. David and S. Greenstein, The economics of compatibility standards: An introduction to recent research, Economics of Innovation and New Technology 1(1) (1990) 3–41. [28] B. Dean, Prepayment cards in Europe, Financial Times Management Report, FT Financial Publishing (October 1995) 175. [29] P. de Gélis and A. Chabrol, Des circuits et moyens de paiement en mutation, Dossier Les Banques Néerlandaises, Banque 570 (May 1996) 36–39. [30] B. de Liedekerke Beaufort, Betaalsystemen in België, Aspecten en Documenten, Belgische Vereniging van Banken 40 (August 1985). [31] A. Dhebar and S.S. Oren, Optimal dynamic pricing for expanding networks, Marketing Sci. 4(4) (1985) 336–351. [32] M. Eeckhout, Le secteur de la distribution et les cartes, Revue de la Banque/Bank-en Financiewezen 3 (March 1990) 135–140. [33] N. Economides, Compatibility and the creation of shared networks, in: Electronic Services Networks: A Business and Public Policy Challenge, eds. M. Guerin-Calvert and S. Wildman (Praeger Publishing Inc., New York, 1991). http://edgar.stern.nyu.edu/networks/ papers.html. [34] N. Economides, Comments on ‘Antitrust economics of credit card networks’ by Carlton and Frankel, Federal Reserve Bank of St. Louis Rev. 77(6) (1995) 60–63. http://edgar.stern.nyu.edu/ networks/. [35] N. Economides, The economics of networks, Internat. J. Industrial Organization 14(6) (1996) 673– 699. http://edgar.stern.nyu.edu/networks/top.html.


169

[36] N. Economides, Network externalities, complementarities, and invitations, European J. Political Economy 12(2) (1996) 211–233. http://edgar.stern.nyu.edu/networks/papers. html. [37] N. Economides and F. Flyer, Technical standards coalitions for network goods, Discussion Paper EC-95-12, Stern School of Business, New York University (1995). http://edgar.stern. nyu.edu/networks/papers.html. [38] N. Economides and C. Himmelberg, Critical mass and network evolution in telecommunications, in: “Toward a Competitive Telecommunication Industry”, 1994 Telecommunications Policy Research Conf., ed. G. Brock (1995). http://edgar.stern.nyu.edu/networks/papers.html. [39] N. Economides and C. Himmelberg, Critical mass and network size with application to the U.S. fax market, Discussion Paper EC-95-11, Stern School of Business, New York University (1995). http://edgar.stern.nyu.edu/networks/papers.html. [40] N. Economides and S.C. Salop, Competition and integration among complements and network market structure, J. Industrial Economics 40(1) (1992) 105–123. [41] N. Economides and L.J. White, One-way networks, two-way networks, compatibility, and antitrust, Discussion Paper EC-93–14, Stern School of Business, New York University, 1993, eds. D. Gabel and D. Weimann, Opening Networks to Competition: The Regulation and Pricing of Access (Kluwer Academic Press, 1996). http://raven.stern.nyu.edu/networks/papers.html. [42] N. Economides and L.J. White, Networks and compatibility: Implications for antitrust, European Economic Rev. 38 (1994) 651–662. http://raven.stern.nyu.edu/networks/papers. html. [43] D. Encaoua, P. Michel and M. Moreaux, Network compatibility: Joint adoption versus individual decisions, Ann. Econom. Statist. 25/26 (1992) 51–69. [44] P. Esser and L. Leruth, Marketing compatible, yet differentiated products: In search of competitive equilibria when network externalities are at work, Internat. J. Res. Marketing 5 (1988) 251–270. [45] J. Farrell and G. Saloner, Standardization, compatibility, and innovation, RAND J. Economics 16 (1985) 70–83. [46] J. Farrell and G. Saloner, Installed base and compatibility: Innovation, product preannouncement, and predation, Amer. Economic Rev. 76(5) (1986) 940–955. [47] J. Farrell and C. Shapiro, Dynamic competition with switching costs, RAND J. Economics 19(1) (1988) 123–137. [48] L. Germentier, Les moyens de paiements des particuliers – Le point de vue du banquier, Reflets et ´ Perspectives de la Vie Economique 34(4) (1995) 283–294. [49] J.M. Groeneveld and A. Visser, Seigniorage, electronic money and financial independence of central banks, Banca Nazionale del Lavoro Quarterly Rev. 200 (March 1997) 69–88. [50] P.M. Horvitz and L.J. White, The challenges of the new electronic technologies in banking: private strategies and public policies, Working paper S-96-44, Salomon Center for Research in Financial Institutions and Markets, Stern School of Business, New York University (September 1996). [51] M.L. Katz and C. Shapiro, Network externalities, competition, and compatibility, Amer. Economic Rev. 75(3) (1985) 424–440. [52] M.L. Katz and C. Shapiro, Technology adoption in the presence of network externalities, J. Political Economy 94(4) (1986) 822–841. [53] M.L. Katz and C. Shapiro, Product introduction and network externalities, J. Industrial Economics 40(1) (1992) 55–83. [54] M.L. Katz and C. Shapiro, Systems competition and network effects, J. Economic Perspectives 8(2) (1994) 93–115. [55] R.F.M. Kieviet, Chipknip: het begin van een nieuw tijdperk, Bank-en Effectenbedrijf 44(10) (1995) 19–20. [56] P. Klemperer, Markets with consumer switching costs, Quarterly J. Economics 102 (May 1987) 375–394.

170


[57] P. Klemperer, Entry deterrence in markets with consumer switching costs, Economic J. 97 (1987) 99–117. [58] P. Ledingham, Pre-paid cards, Reserve Bank Bulletin, Reserve Bank of New Zealand 57(4) (1994) 346–349. [59] S.J. Liebowitz and S.E. Margolis, Network externality: An uncommon tragedy, J. Economic Perspectives 8(2) (1994) 133–150. [60] S.J. Liebowitz and S.E. Margolis, Are network externalities a new source of market failure?, Res. in Law and Economics 17 (1995) 1–22. http://wwwpub.utdallas.edu/∼liebowit/ netwextn.html. [61] S.J. Liebowitz and S.E. Margolis, Market processes and the selection of standards, Working paper (1996). http://wwwpub.utdallas.edu/∼liebowit/standard/standard.html. [62] A. Linkens, Proton: le porte-monnaie e´ lectronique en Belgique, Banque 560 (June 1995) 77–79. [63] A. Linkens, D. Hautain and Y. Moulart, L’évolution technologique, Reflets et Perspectives de la ´ Vie Economique 34(4) (1995) 263–282. [64] C. Matutes and A.J. Padilla, Shared ATM networks and banking competition, European Economic Rev. 38(6) (1994) 1113–1138. [65] J.J. McAndrews, The evolution of shared ATM networks, Business Rev., Federal Reserve Bank of Philadelphia (May–June 1991) 3–16. [66] J.J. McAndrews, Antitrust issues in payment systems: Bottlenecks, access, and essential facilities, Business Rev., Federal Reserve Bank of Philadelphia (September–October 1995). http://www. phil.frb.org/econ/br/brso95jm.html. [67] J.J. McAndrews, Comments on ‘Antitrust economics of credit card networks’ by Carlton and Frankel, Federal Reserve Bank of St. Louis Rev. 77(6) (1995) 55–59. [68] J.J. McAndrews, Banking and payment system stability in an electronic money world, Working paper Nr. 97-9, Federal Reserve Bank of Philadelphia (July 1997). http://www.libertynet.org/ ∼fedresrv/econ/wps/1997.html. [69] J.J. McAndrews, Network issues and payment systems, Business Rev., Federal Reserve Bank of Philadelphia (November/December 1997) 15–25. http://www.phil.frb.org/econ/br/ brnd97in.html. [70] R.F. Meneses and F. Lourenço, Porta Moedas Multibanco – PMB, an evolving electronic purse system (September 1996). http://cardshow.com/guide/card/porta-moedas multibanco.html. [71] Mobile phones to become cashpoint machines, press release, Mondex International (September 22, 1997). htt://www.mondex.com/mondex/cgi-bin/show.pl?english+global&../ english/documents/global/prel29539959.txt. [72] A. Perrot, Ouverture a` la concurrence dans les réseaux. L’approche stratégique de l’économie des ´ réseaux, Economie et Prévision 119 (1995) 59–71. [73] A. Prinz, Money makes the Web go round: e-money, c-money and the demand for cb-money, in: 1st Berlin Internet Economics Workshop, Berlin, Germany (October 24–25, 1997). [74] G. Saloner and A. Shepard, Adoption of technologies with network effects: An empirical examination of the adoption of automated teller machines, RAND J. Economics 26(3) (1995) 479–501. [75] G. Stuber, The electronic purse: An overview of recent developments and policy issues, Technical Report, Bank of Canada 74 (January 1996). http://www.bank-banque-canada.ca/ english/tr abs.htm#74. [76] Survey on Internet money, Imperial College Management School (1995). http://graph.ms. ic.ac.uk/analysis. [77] K. Tijdens, 25 jaar produkt-en procesinnovaties in het girale betalingsverkeer, Research Memorandum, University of Amsterdam, Department of Economics, Nr. 9207 (1992). [78] L. Van Hove, De nieuwe generatie betaalmiddelen (Electronic purse schemes in Europe – The next generation), Rev. de la Banque/Bank-en Financiewezen 3 (March 1996) 128–139.


171

[79] L. Van Hove, Op naar de ‘cashless society’? (Towards a cashless society?), Economisch en Sociaal Tijdschrift 51(1) (1997) 35–81. [80] L. Van Hove, A selected bibliography on electronic purses, Web site (1998). http://cfec. vub.ac.be/cfec/purses.htm. [81] J.A. Weinberg, The organization of private payment networks, Economic Quarterly, Federal Reserve Bank of Richmond 83(2) (1997) 25–43. http://www.rich.frb.org/generalinfo/ econinfo.html. [82] R.J. West, Electronic purse developments in Europe: An overview, in: CAFE Seminar “Electronic Consumer Payment Systems: The Next Generation”, Brussels (January 31, 1996). [83] L.J. White, Technological change, financial innovation, and financial regulation in the U.S.: The challenges for public policy, Working paper S-96-45, Salomon Center for Research in Financial Institutions and Markets, Stern School of Business, New York University, New York (September 1996). [84] S. Worthington, Smart cards and retailers – Who stands to benefit?, Internat. J. Retail Distribution Managm. 24(9) (1996) 27–34. [85] J. Xie and M. Sirbu, Price competition and compatibility in the presence of positive demand externalities, Managm. Sci. 41(5) (1995) 909–926. [86] Y.-N. Yang, An introduction to network externalities: A recent literature review, Working paper, Department of Economics, Utah State University (August 1997). [87] R. Zobel, Opening address, in: CAFE Seminar “Electronic Consumer Payment Systems: The Next Generation”, Brussels (January 31, 1996).