Standards Battles and Public Policy

Transcription

1 Standards Battles and Public Policy Luís M B Cabral New York University and CEPR Tobias Kretschmer London School of Economics Aril 004 Abstract We examine the effectiveness of ublic olicy in a context of cometing standards with network externalities. We show that, if the olicymaker is very imatient, then it is otimal to suort the leading standard; whereas, if the olicymaker is very atient, then it is otimal to suort the lagging standard. We also consider the timing for otimal intervention and rovide sufficient conditions under which it is otimal to delay or not to delay intervention. Keywords: standards, network externalities, ublic olicy. JEL Code Nos.: L13, L51, O33 lcabral@stern.nyu.edu, t.kretschmer@lse.ac.uk. We are grateful to Lorenz Schneider and various seminar articiants for comments and suggestions. The usual disclaimer alies.

2 1 Introduction VHS vs. Betamax VCRs; Ale MacIntosh vs. PC DOS microcomuters; discrete vs. matrix quadrahonic systems. These are three of a long list of examles from recent history where two or more) alternative versions of a new standard battled for market dominance. One asect common to most of these standards is the imortance of network effects: the fact that many users buy a DOS-based microcomuter increases the utility of buying a DOSbased microcomuter among other reasons because the amount of software, technical suort, etc., available for DOS users will be better and more widely available). Given the imortance of network industries, it is surrising that little attention has been aid to the role of ublic olicy. 1 Consider the cases of high-definition television HDTV) and mobile telecommunications. Public olicy towards these industries differed greatly between Euroe and the U.S.: Whereas the Euroean Commission E.C.) was rimarily concerned with early standardization, the U.S. s Federal Communications Commission F.C.C.) adoted the more atient aroach of letting cometition decide the winning standard. At first sight, the Euroean aroach seems referable in that it takes better advantage of network effects. The U.S. aroach, in turn, is more to likely lead to a higher-quality standard. Our urose in this aer is to analyze the different trade-offs involved in the olicymaker s decisions with resect to standardization in network industries. Secifically, there are at least two questions that a olicymaker should address. First, the decision of which standard or standard to suort, if any. Second, the decision of when to intervene. Regarding the first question, we show that, if the olicymaker is sufficiently atient, then it is otimal to favor the lagging standard. Conversely, if the olicymaker is sufficiently imatient, 1 Excetions include David 1986), Stoneman and David 1986), Katz and Shairo 1986), Cowan 1991), and Choi 1994). 1

3 then it is otimal to favor the leading standard. Regarding the second question, we show that, if the olicymaker is sufficiently atient, then it is otimal to delay intervention. Conversely, if the olicymaker is sufficiently imatient, then it is otimal not to delay intervention. In our analysis, we consider the extreme cases of infinitely atient and an infinitely imatient olicymaker. An infinitely atient olicymaker is one who cares exclusively about the welfare of future adoters, whereas an infinitely imatient olicymaker cares exclusively about current adoters. We consider these extremes for illustrative uroses only; reality is likely to fall somewhere in between. There are two factors that determine the degree of olicymaker atience in each articular case. One is the olicymaker s references: witness, for examle, the contrast between Euroe and the U.S. in wireless telecommunications, or the contrast between Jaan and the U.S. in HDTV. More imortantly, the degree of atience is likely to reflect the nature of the technology in question. Take for examle the case of color television in the 1950s. Given the success of monochrome TV and the absence of a likely substitute for TV, a olicymaker should take a long-term view of the standardization rocess: whatever solution is achieved, it is likely to stay for a long time and be used by a great number of future adoters. Our infinitely atient olicymaker assumtion tries to cature this feature. By contrast, digital audio tae DAT) is a good examle of a technology with a relatively short exected life san, considering the raid advancements in storage and recording devices such as CDs. In such a situation, a olicymaker is more likely to concentrate on the existing set of adoters and the standardization roblems they face. Our infinitely imatient olicymaker assumtion tries to cature this situation. In addition to issues of time horizon and atience, our model treats the In the case of DAT, there were two different standards, DDS and DataDAT.

4 olicymaker s actions in a stylized way. Secifically, we assume that the olicymaker has the otion to tilt the system in favor of one standard or the other. In reality, this may come about through direct subsidies e.g., HDTV in Jaan), government regulations e.g., wireless in Euroe), or direct adotion decisions by the olicymaker when the latter is a large user e.g., nuclear reactors in the U.S.). The aer is organized as follows. In the following section, we extend Arthur s 1989) model of standard adotion to consider the ossibility of ublic olicy intervention. Next we consider the direction of otimal intervention in the case when the olicymaker is very imatient Section 3) or atient Section 4). In Section 5, we look at the otimal timing for intervention. Section 6 includes a discussion of some of the results in the context of several recent standards battles. Section 7 concludes the aer. Basic model Our analysis dearts from Arthur s 1989) seminal model of standard adotion. Suose there are two unsonsored standards, A and B, available to consumers at constant marginal cost which we normalize to zero). In each eriod, one new consumer arrives in the market and buys one unit of one of the standards. Some consumers favor standard A, some standard B; all benefit from the size of the network they link into. Secifically, by choosing standard i, a consumer receives, at time t, utility v i + wn it, where v i is stand-alone utility and n it standard i s network size at time t. We assume that v i {0, 1} and that v j = 1 v i. Following Arthur, we assume that, in each eriod, consumers make adotion decisions based on that eriod s utility levels. Under this assumtion, it can be shown that, if standard i is chosen sufficiently more often than standard j, then all future adotions are directed to i, even by consumers who, absent network effects, would refer standard j v j > v i ). Arthur et al. 1983) have 3

5 shown that the above stochastic rocess of technology adotion ends u in one of these absorbing barriers in finite time with robability one. The secific condition for lock-in to standard i is n i w > n j w + 1, or i > N 1, where w i n i n j is the difference in installed bases. The values N, N are called absorbing barriers since, once crossed i < N, i > N), they are never crossed again. Arthur et al. s result can then be rehrased as: lock-in to one technology occurs in finite time with robability one. Consider now the following extension of Arthur s model: suose that consumers are unevenly distributed: a fraction > 1 refers one of the standards. Since standards are otherwise symmetric, it follows that the standard with > 1 fans is the better standard.3 A crucial assumtion in our analysis is that the olicymaker knows the above information as well as the rior distribution of, which we assume is symmetric around 1. However, the olicymaker does not know the exact value of. Consider now the roblem faced by a welfare maximizing olicymaker. Since marginal cost is constant and identical for both standards, a sufficient statistic for social welfare is discounted consumer surlus: t W = δ t uτ, t), t=1 τ=1 where uτ, t) is eriod t utility of the consumer who joined the network at time τ, and δ the discount factor. We will consider the following olicy instrument: At a given oint in time, the olicymaker has the otion of forcing the next s i adotions of standard i. These forced adotions can be interreted in various ways. One is to assume the olicymaker subsidizes adotion by rivate agents. An alternative interretation is that the olicymaker is itself a large adoter see Section 6 for examles). As we will see, the direction of the otimal olicy deends crucially 3 Standardization, i.e., lock-in to one standard, is otimal in our model as it is in Arthur s 1989), a result that deends on the assumtion of a linear utility function. Farrell and Saloner 1986) and Bassanini and Dosi 1998) develo models where this assumtion is relaxed and find that standardization need not be otimal. 4

6 on the olicymaker s discount factor. We will consider two extreme cases: a very imatient, or myoic, olicymaker; and a very atient olicymaker. 3 The case of an imatient olicymaker We start with the case of a very imatient olicymaker. Our main result is that such a olicymaker should favor the leading standard. Proosition 1 If δ is close to zero, and given that olicy intervention takes lace in state i, it is otimal to favor standard i if i is sufficiently greater than zero. Proof: If δ is close to zero, then all eriods after the next are of secondorder imortance. The network benefits added to the current base of users are given by wn i. The difference between the two standards is thus w i. If i is sufficiently high, then the benefits on the existing users outweigh the benefits received by the new user, and the result follows. An alternative version of the result is as follows. Suose that the olicymaker has the otion of offering a subsidy to the new user at time t. Then the subsidy to standard i is ositive if and only if i > 0. This result corresonds to the classical case of an externality. Since the discount factor is close to zero, there is no informational issue; that is, the olicymaker is not concerned with the value of and how it will influence the exected attern of future adotions beyond the next eriod). The main thing the olicymaker is concerned with is how the next adoter will affect the revious adoters. If the i installed base is greater than the j installed base, then the externality is greater when an i adotion takes lace, and thus the olicymaker is better off subsidizing this standard. 5

7 4 The case of a atient olicymaker Consider now the oosite case with resect to the revious section, namely that of a very atient olicymaker. From an otimization oint of view, this is the rather more interesting case. The olicymaker s roblem is that, while knowing that one of the standards is suerior higher ), it does not know which one is which. All that the olicymaker knows is the rior distribution on, which we assume is symmetric around 1. In other words, the two standards look the same at the start of the rocess. Naturally, as the adotion rocess unfolds, the olicymaker acquires more information, secifically, the number of adotions of each standard. Our main result is that the olicymaker s otimal olicy is to favor the lagging standard: Proosition If δ is close to one, and given that olicy intervention takes lace in state i, it is otimal to favor standard j by s = 1 i. Proof: See the Aendix. This result states that the otimal intervention intensity is to ull the leading standard halfway back to the symmetric state i = 0). Intuitively, moving the rocess halfway back takes into account the tradeoff between keeing the rocess away from the absorbing barriers for some more time which imlies suorting the lagging standard) and making use of the information gained from the rocess rior to intervention which suggests that the leading standard is leading for a reason: it is more likely that it is indeed the right one). Notice that the result does not deend on the articular distribution of ; the only restriction is that the distribution is symmetric, i.e., the two standards have a riori an equal chance of being the otimal standard. Secifically, consider the extreme case of a binomial distribution and suose that is very close to 1. In other words, suose that each of the stan- 6

8 dards is equally likely to be favored by a fraction of the oulation, where is close to one. Even then, the otimal olicy would be to delay the lock-in rocess. This may at first seem counterintuitive: if so many adoters have chosen standard i reviously, then it is very likely that this is the right standard. But recisely because is close to one and the olicymaker is very atient, favoring the lagging standard is an otimal olicy: in the likely) event that the leading standard is the right standard, then favoring the lagging standard won t do much harm; most likely, the leading standard will eventually win anyway. Broadly seaking then, Proosition seems consistent with David s 1987) rescrition that one thing that ublic olicy could do is to try to delay the market from committing. 5 Otimal timing for ublic intervention So far, we have addressed the question: given that the olicymaker must make a decision at time t, which standard should the olicymaker favor? The natural next ste is to ask when the olicymaker should intervene. We will address a somewhat more secific question: given that the olicymaker must choose a single time at which to intervene, what is the otimal time t? The main result in this section states that a atient olicymaker should wait, whereas an imatient one should act soon. Proosition 3 Suose that the olicymaker must choose a single time at which to intervene. If the discount factor δ is close to one, then it is otimal to wait until i = N. If δ is close to zero and i is large, then it is otimal to intervene right away. Proof: See Aendix. 7

9 The intuition for the imatient olicymaker case is similar to that of Proosition 1. Regarding the atient olicymaker case, the question might be asked: Why should a atient olicymaker wait until i = N? The answer is, the closer to N we are the more information the olicymaker has. Since N is achieved in finite time with robability one, and the olicymaker is infinitely atient, there is no cost of waiting. Waiting for longer than i = N does the olicymaker no good: once we hit an absorbing barrier, no additional information is gained. 6 Examles The results resented in the revious sections are as tentative as the model they are based on is stylized. Real world examles are far more comlicated than simle models. Still, we believe the theoretical analysis allows us to make some qualitative oints about ublic olicy. In this section, we resent a few examles of ublic olicy in industries with strong network effects. These examles illustrate the structure and assumtions of our model. Second generation mobile telehony. Second generation wireless standard setting rovides an interesting testing ground for the economic theory of ublic olicy. The U.S. and the E.U. took very different aroaches to the roblem. Whereas in the U.S. the Federal Communications Commission FCC) followed a hands-off aroach, in Euroe the Euroean Commission EC) mandated a standard from very early on. 4 Standard-setting in Euroe was regarded as a success story, esecially in the early stages of G technology: early diffusion was faster in Euroean countries than in the U.S. at roughly comarable rices, 5 and roaming i.e., using one s cellhone outside 4 For a qualitative assessment of U.S. and Euroean decisions, see Gandal et. al. 003). Toivanen 00) uses a decision-theoretic framework to analyze 1G standard choice in 85 countries. 5 See, e.g. Gruber and Verboven 001) or Koski and Kretschmer 00) for emirical 8

10 the rovider s coverage area) was clearly better in Euroe early on. As the technology matured however, these differences became less relevant. Currently, diffusion is at similar levels and roaming is virtually seamless in both markets. Finally, as third-generation technology enters the icture, it is interesting to note that the cometing standards are both based on CDMA, the standard that survived the battle for suremacy in the olicy-neutral U.S. ground. Our theoretical analysis Proositions 1 3) suggests that a very atient olicymaker should wait and favor the lagging technology before the market sets on to a articular standard; whereas a very imatient olicymaker should favor the leading technology early on. Moreover, by continuity, Proositions 1 and suggest that, for intermediate values of δ, the otimal government olicy is not to favor any of the technologies. This result is strengthened if we consider additional sources of uncertainty for examle, uncertainty regarding ayoff levels), or if we consider more than two tyes of adoters. In other words, the best olicy may in fact be not to have a olicy. The contrasting aroaches taken by the U.S. and the E.C. suggest that either one of them made the wrong decision, or else that they started from different utility functions. The latter may be accounted by different erceived time horizons or different weights laced on early adoters. Wide-body aircraft: DC-10 vs. B-747. Over a eriod of time during the mid 1970s, the U.S. Air Force ordered about sixty military cargo and tanker aircraft. It was seen as a no brainer that the U.S.A.F. would select the Boeing roosal on the grounds of the technical secifications of their lanes. As it turned out, the McDonnell Douglas KC-10 was selected. The KC-10 is the military version of the DC-10 and shares many features with the latter. The Air Force decision thus had the effect of keeing the DC-10 rogram alive for a while longer. In the end, the indirect network effect created studies of the effect of standardization on the evolution of mobile telehony markets. 9

11 by the learning curve, as well as a series of DC-10 crashes in the late 1970s, led to a shar decline in orders for the McDonnell Douglas lane; roduction was discontinued in This examle illustrates how the olicymaker can intervene as a large adoter. Nuclear ower reactors. By the late 1950s, there were about a dozen relevant technologies for nuclear ower reactors. Of these, the main contenders were light water, heavy water and gas grahite. None of the technologies was erceived as clearly suerior, and early adotion figures indicated that consumers were divided in their references. Due to strong learning and network effects, exerts redicted that one of the technologies would eventually dominate. One imortant event in the race was the U.S. Navy s decision to adot the light water technology in their nuclear submarines. Eventually, when a market for civilian nuclear ower emerged, the light water absorbing barrier had been crossed and the industry was locked-in to this technology. According to Cowan 1990), light water is considered inferior to other technologies, yet it dominates the market for nuclear reactors. This examle thus illustrates, among other things, how subotimal outcomes may take lace in the standard setting rocess with ublic intervention. Pest control technology. For a limited eriod of time, the U.S. Deartment of Agriculture sonsored one of the alternative technologies for est control: Integrated Pest Management IPM). Individual farmers have little incentive to deviate from the common ractice in the vicinity, which imlies a network effect similar to the one we consider in our theoretical model. For this reason, while the government intervention was temorary, its effects were ermanent: the industry got locked in to IPM, which, according to Cowan and Gunby 1996), was the welfare maximizing outcome. This examles illustrates 6 For more on McDonnell Douglas, see htt:// 10

12 that early ublic intervention may have a determinant effect even if limited in its extent and duration. Linux vs. Windows. Recently, several government agencies in the U.S., Euroe and Asia have decided to adot the Linux oerating system at various levels. 7 Acting as large and influential customers, governments may suort the lagging technology with the aim of delaying the outcome of the standards battle, or simly to lower the deendency on single software vendors. 7 Conclusion The above examles illustrate the variety of situations where standards battles take lace and government intervention is a ossibility. Sometimes the olicymaker acts by law, sometimes by offering adotion incentives, sometimes by acting as a lead adoter. Notwithstanding the secificities of each situation, one thing that is common to all cases is the olicymaker s dilemma of which standard to favor, if any, and when. We thus think that our model, stylized as it is, addresses an imortant set of ublic olicy issues. 7 The list includes: in the U.S., the Air Force, the Federal Aviation Administration, the Postal Service, and the Deartments of Defense, Agriculture and Energy; in Euroe, the Euroean Commission, various government offices in Germany, and France s Ministries of Culture, Defense and Education; and China s Post Office. Sources: htt:// on 04/06/00; htt:// on 30/05/00. 11

13 Aendix Proof of Proosition : We begin by assuming that the rior distribution of takes two values, and later generalize to the case of a symmetric distribution. In words, the olicymaker knows that one of the standards is referred by a majority > 1 of the oulation, knows the value of, but does not know which standard is which. Suose that at time t the system is in state i > 0. Let P i ) be the robability that the system will eventually get locked-in to i. Let Π i ) be the robability that standard i is the right standard, that is, the standard associated to > 1. Then the unconditional robability that the system gets locked-in to the right standard is simly π i ) = Π i )P i ) + Π i )P i ). 1) The olicymaker s goal is to maximize the robability that the right absorbing barrier is hit. Absent any intervention, that robability is given by π i ). We assume that the olicymaker has the otion of, starting at time t and over a eriod of time, forcing adotions in favor of one of the standards. Define ˆπ i, s) as the robability that the right standard is chosen given that, starting in state i, the next s adoters are forced to adot standard i. Derivation of P i ) and Π i ). Given the stationarity of the rocess, the robability that the system in state i will get locked-in to A satisfies the difference equation P i ) = P i + 1) + 1 )P i 1). Let N be the necessary lead for one of the standards to lock in so that the distance between barriers is N). The boundary conditions are then given by P 0) = 1 and P N) = 0. We thus get 1

14 P i ) = N N i 1 ) N+ i N 1 ) N. The robability that A is the right standard, Π i ), is defined by the robability that the current state is reached given that A is associated with > 1. If there have been t adotions and the robability of adoting A is, then the likelihood that state i is reached is m t+ i 1 ) t i, where m is the number of ossible combinations of us and downs which lead to state i. On the other hand, the robability of reaching state i given that A is associated with 1 ) is given by m t i 1 ) t+ i, The osterior robability that A is associated with is therefore given by m t+ i 1 ) t i m t+ i 1 ) t i + m t i 1 ) t+ i = i i + 1 ) i. Finally, substituting for P and Π in 1), we get the unconditional robability that the system, Y t), will hit the right barrier: π i ) = i N N i 1 ) N+ i) i + 1 ) i ) N 1 ) N ) i N N+ i 1 ) N i) + ) ). i + 1 ) i N 1 ) N Otimal intervention for secific. A olicymaker will maximize the robability that the right barrier is hit by forcing s adoters to adot either the leading or the lagging standard. Let s > 0 denote forced adotions of the 13

15 lagging standard and s < 0 adotions of the leading one. Note that s 0 influences only P i ). That is, the robability that a given barrier is the right one is not affected. We can now see that the new robability of achieving a desired outcome is ˆπ i ) = i N N i+s 1 ) ) N+ i s ) )+ i + 1 ) i N 1 ) N N N+ i s 1 ) ) N i+s + i i + 1 ) i ) N 1 ) N ). We now maximize ˆπ with resect to s: ˆπ i ) s i + i = N i +s ln)1 ) N+ i s + N i+s 1 ) N+ i s ln1 ) ) i + 1 ) i ) N 1 ) N ) N i +s ln)1 ) N i+s N+ i s 1 ) N i+s ln1 ) ) i + 1 ) i ) N 1 ) N ), or simly ˆπ i ) s = ψ i ) i + 1 ) i) + ψ i ) i + 1 ) i) ) ) ln1 ) ln) ) ) ), i + 1 ) i i + 1 ) i N 1 ) N where ψ i ) = i N i+s 1 ) N+ i s. The denominator is different from zero, and so is ln1 ) ln) for 1 ). A necessary condition for maximizing π i, s) is therefore that ψ i ) i + 1 ) i) + ψ i ) i + 1 ) i) = 0. which imlies s = i /. We take the second derivative to determine whether s is a maximum: 14

16 ˆπ i ) s = ψ i ) i + 1 ) i) + ψ i ) i + 1 ) i) ) ln1 ) ln) ) ) ) ). i + 1 ) i i + 1 ) i N 1 ) N Observe that all of the bracketed exressions on the right-hand side are ositive. It follows then that that the second derivative is negative and s is a global maximum. Generalization to any symmetric distribution. The above results readily generalize to any symmetric distribution of. The idea is that any distribution symmetric about 1 is the integral of a series of binomial distributions like the one we considered above. Since the otimal solution does not deend on, it follows that the same solution holds for any distribution of that is symmetric around 1. Proof of Proosition 3: The case when δ is close to zero is straightforward: anything that takes lace after the current eriod is of second-order imortance; and an intervention in the current eriod has a ositive effect on welfare. When δ is close to one, discounting is irrelevant or close to irrelevant). We need to find out in which eriod the imact of ublic olicy is greatest. Define ρ 1. Notice that, given our assumtion that > 1, it follows that 0 < ρ < 1. With this change in variables, we can simlify various revious exressions as follows: Π i ) = ρ i P i) = 1 ρn+ i 1 ρ N 15

17 Substituting in the exression for π i ) and simlifying, we get Φ 1 ρ N )π i ) = 1 ρ N. ) Moreover, since by Proosition we have s = 1 i, it follows that P i + s ) = 1 ρn+ 1 i 1 ρ N. Substituting in the exression of ˆπ i, s) and simlifying, we get Ψ 1 ρ N )ˆπ i ) = 1 + ρ N 1 i ρ N i ρ N ) ρ 1. 3) N i + ρ N Comaring ) and 3), we conclude that: a) exected value without olicy is indeendent of i ; b) exected value with olicy is increasing in i. We conclude that the exected incremental value from imlementing the otimal olicy is increasing in i. 16

18 References Arthur, W Brian 1989), Cometing Standards, Increasing Returns, and Lock-In by Historical Events, The Economic Journal 99, Arthur, W Brian, Yu Ermoliev, and Yu Kaniovski 1983), On Generalized Urn Schemes of the Polya Kind, Kybernetics 19, Bassanini, Andrea, and Giovanni Dosi 1998), Cometing Technologies, International Diffusion and the Rate of Convergence to a Stable Market Structure, IIASA Interim Reort IR Choi, Jay Pil 1994), Irreversible Choice of Uncertain Technologies with Network Externalities, Rand Journal of Economics 5, Cowan, Robin 1990), Nuclear Power Reactors: A study in Technological Lock-in, Journal of Economic History 50, Cowan, Robin 1991), Tortoises and Hares: Choice Among Technologies of Unknown Merit, Economic Journal 101, Cowan, Robin, and Phili Gunby 1996), Srayed to Death: Path Deendence, Lock-In and Pest Control Strategies, Economic Journal 106, David, Paul 1986), Narrow Windows, Blind Giants, and Angry Orhans: The Dynamics of Systems Rivalries and Dilemmas of Technology Policy, Technological Innovation Project Working Paer no. 10 CEPR, Stanford University). David, Paul 1987), Some New Standards for the Economics of Standardization in the Information Age, Ch. 8 in Economic Policy and Technological Performance, P. Dasguta and P. Stoneman eds.), Cambridge University Press

19 Farrell, Joseh, and Garth Saloner 1986), Standardization and Variety, Economics Letters 0, Gandal, Neil, David Salant, and Leonard Waverman 003), Standards in Wireless Telehone Networks, forthcoming in Telecommunications Policy. Gruber, Harald, and Frank Verboven 001), The Evolution of Markets under Entry and Standards Regulation - The Case of Global Mobile Telecommunication, International Journal of Industrial Organization 19, Katz, Michael, and Carl Shairo 1986), Technology Adotion in the Presence of Network Externalities, Journal of Political Economy 94, Koski, Heli, and Tobias Kretschmer 00), Entry, Standards and Cometition: Firm Strategies and the Diffusion of Mobile Telehony, ETLA Discussion Paer No. 84. Stoneman, Paul, and Paul David 1986), Adotion Subsidies vs. Information Provision as Instruments of Technology Policy, Economic Journal 96, Toivanen, Otto 00), Choosing Standards, Unublished Draft, Helsinki School of Economics, Aril. 18