Does net neutrality contribute to the migration to FTTH? a lesson from Japan’s experience Toshifumi Kuroda** Abstract Does net neutrality contribute to the migration to fiber to the home (FTTH)? In this paper, we estimate the indirect network effects on broadband demand of internet content services by using a set of panel data for the Japanese market. From our estimated results, we find three implications of indirect network effects. First, a consumer’s migration from asymmetric digital subscriber line (ADSL)to FTTH is explained not by the attributes of internet access services (i.e., price and speed) but by the indirect network effects of content services (i.e., the number of music distribution, video sharing, and online games.) Second, there is a large upward bias in the estimates of price elasticity if indirect network effects are ignored. Finally, the ADSL service provider has the strongest incentive to foreclosure in broadband content services. We conclude that net neutrality helps consumers migrate to FTTH and maintains the consumer welfare derived from indirect network effects.

Keywords: Broadband, Indirect Network effects, Net neutrality, Price elasticity, Vertical Foreclosure.

JEL classification: D120, L510, L860 1. Introduction In recent years, fast broadband migration from ADSL to FTTH has been taking place in Japan. According to the Broadband Statistics published in June 2007 by the Organization for Economic Co-operation and Development (OECD), 36% of Japan’s broadband subscribers use fiber connections, and Japan’s diffusion of fiber infrastructure is the widest in the world. In fact, the value of broadband access services arises from the use of broadband content services. Still, if broadband subscribership is small, no content provider would supply broadband contents services. Such interdependency is called indirect network effect. Gandal (2002) points out the importance of compatibility, standardization, and network effects in the analysis of network goods such as broadband. Indirect network

Assistant Professor Faculty of Economics, Tokyo Keizai University, 1-7-32, Minami-cho, Kokubunji-shi, Tokyo 185-8502, Japan Tel./Fax: +81 42 328 7886 E-mail: [email protected] (T. Kuroda).

**

effects have a particularly important role in the demand analysis of broadband. There exists some econometric analyses of demand for broadband: Kridel, Rappoport, and Taylor (2001), Dufy-Deno (2003), Scott and Donald (2005), Ida and Kuroda (2006), Ida and Sakahira (2008), and Cardona et al (2009) all have analyzed the demand for broadband access, by using price, speed, and consumer attributes. Hitt and Tambe (2007) have analyzed content consumption driven by the subscriptions to broadband access services. In these papers, broadband subscription is analyzed not by the indirect network effects from broadband content, but by consumer attributes. In the existing literature pertaining to indirect network effects,Gandal, Kende and Rob (2000) have analyzed the role of indirect network effects on the diffusion of the compact disc player. Nair, Chintagunta and Dubé (2004) analyze the competition between Windows CE and Palm OS in the U.S. Personal Data Assistant (PDA) market. Clements and Ohashi (2005) analyze the competition between consumer game machines in the U.S., however, there is still no econometric analysis of the network effects on the demand for broadband access services. In this paper, we analyze the indirect network effects on the demand for broadband access services, and show that the migration to FTTH was caused by the indirect network effects derived from content services. The paper proceeds as follows. In section 2, we briefly describe Japan’s broadband market and “the second migration”, that is from ADSL to FTTH. Section 3 introduces a estimation model of demand for goods that have indirect network effects. Section 4 describes the data and instruments used in the estimation. Section 5 discusses the estimation results. Section 6 assesses broadband access providers incentive to vertical foreclosure and the results of net neutrality. Section 7 concludes our findings. 2. Japan’s broadband market expansion. In this section, we give a brief summary of the broadband internet history in Japan, and the importance of considering the network effects of broadband content. Broadband demand cannot be explained by the attributes of access services (i.e., price and speed), but by the indirect network effects of content services (i.e., music, movies, and online games). We will also describe Japan’s broadband content market, and the issue of network neutrality between access providers and content providers. 2.1 The broadband access market. The Japanese Ministry of Internal Affairs and Communications (MIC) has adopted competition promotion policies, such as NTT group’s obligation of unbundling, the abolishment of entry barriers, and so on. Ida (2006) attributes the success of Japan’s

telecommunications industry to four factors: the unintended result of competition policy, the intended result of competition policy, the challenge of new entrants, and the efforts by incumbent operating companies. With the challenge of new entrants, there was a rapid decline in price and an increase in speed. According to MIC,, “the result of census on price differences in telecommunication services between internal price and foreign price” shows that Japan’s DSL services charge is the lowest in six years since 20011. Figure 1 The progress of broadband subscribers in Japan Figure 1 shows the progress in the number of broadband internet access service subscribers from June 2002 to December 2008. Until 2005, the number of ADSL subscribers had been steadily grown leading the broadband diffusion in Japan. However, the number of ADSL subscribers began to decline in March 2006. On the other hand, the number of FTTH and CATV subscribers has steadily increased. By 2002, the number of subscribers with CATV Internet connections was overtaken by those with FTTH, and ADSL was overtaken in 2008. Ida and Sakahira (2008) call this “the second migration,” and analyze the types of users who initially migrated from ADSL to FTTH. Table 1 Monthly fixed charges of major broadband access services Table 1 shows the monthly charges of representative access services. Until 2004, declines in price and acceleration of connection speed actively occurred, but there were no major improvements in price or connection speed since 20052. Hence, changes in price and speed do not explain “the second migration.” To explain “the second migration,” we must investigate alternative factors3. 2.2 Broadband content market. Table 2 Experience of purchasing online content

It is widely observed access service provider’s campaign for a penetration pricing. Hendel and Nevo (2006) shows estimates that confuse temporally discount and permanent discount are overestimating price elasticities. In this paper, we consider a permanent price change. 2 A few large CATV internet providers have raised connection speeds since 2007, but the number of CATV subscribers has not increased along with it. 3 This can be explained by a consumer lag in changing services. Ida and Sakahira (2008) point to lock-in effects and measure them. It is true that lock-in effects have an important role in broadband migration, but there is no available data that observes individual choice history. Extending a model to dynamic choice is a further issue. 1

Figure 2 shows the results of a user survey published in “Internet hakusyo 2007” by Impress R&D. About 30% of the respondents had bought on line content. Music files (61.6%), online games (28.4%), e-books (26.7%), and videos (20.0%) were the four major categories among purchased content services. Music, video, and online gaming are representative content services for broadband internet, however, because of widespread, free, ad-supported content, the broadband subscribers experience of buying online content is limited. Figure 2 Progress of broadband content services in Japan Figure 2 shows the progress of broadband content services in Japan. The number of content services is increasing, making it necessary to examine the indirect network effects. 2.3 Network neutrality arguments in Japan Due to the increase in income of ad-supported content services along with it’s users, content service providers would not limit their services to particular access services. However, an access provider’s income would not increase along with the increasing use of content because of the lack of a metering plan. In 2006, MIC held “a table of Network neutrality” where it defined the three principles of Japan’s net neutrality: (1) IP-based networks should be accessible to users and easy to use, allowing ready access to content and application layers, (2) IP-based networks should be accessible and available to any lawful terminal that meets the relevant technical standards, and should support terminal-to-terminal (or “end-to-end”) communication, (3) Users should be provided with equality of access to telecommunications and platform layers at a reasonable price and should be considered as a part of competition promotion policy. “Net neutrality”, as defined in Japan, prevents content discrimination4. However, in case that the access market is highly concentrated and the content market is competitive, it is easy for an access provider to compensate the content provider’s loss revenue by vertical foreclosure, and they might agree to perform conduit discrimination. Hence, it is worth examining the possibility of conduit discrimination under the Rubinfeld and Singer (2002) pointed out two access provider’s incentives to foreclose. Conduit discrimination is insulating its own conduit from competition by limiting rival platform distribution of its affiliated content and services, and content discrimination is the insulation of its own affiliated content from competition by blocking or degrading the quality of outside content.

4

prohibition of content discrimination. 3. The model of network subscription with indirect network effects In this section, we briefly outline the nature of indirect network effects and explain our model. We analyze network effects between broadband access services and broadband content services. There are three broadband access service alternatives and one outside option: ADSL, CATV, FTTH, and not subscribing. Consumers choose one access service that maximizes utility. Consumer’s utility derived from broadband access service is determined by a constant term, each alternative attributes such as bandwidth and monthly fixed charge, and the number of content services available. In the case of Japan's broadband market, almost all contents services can be used with any access service, because “Net neutrality” ensures compatibility between platforms. Hence “Net neutrality” allows investments in content services to spill over to the rival’s network. Since access providers cannot internalize the network effects, their investment in content services is not necessarily beneficial. 3.1 Network subscription equation Consumers choose one alternative from a set of four choices. Following Ida and Kuroda(2006), we use a nested logit model that gather ADSL, CATV, and FTTH in one nest. Because of “Net neutrality,” there is no difference in the number of contents between access services; hence we must use time series or panel data to identify network effects. We follow the formulation of Clements and Ohashi (2005) derived from Berry (1994) to estimate the indirect network effects by pooled data5. We fixed the utility of the not subscribing alternative at 0. Consumer’s utility from a given broadband service takes a random utility function of the form

uij = x j β − α p j + ∑ l lωlj + ξ j + ε ij = V j + ε ij where

(1)

uij is consumer i’s utility when alternative j is chosen, x j is an observable

attribute of alternative j, p j is the price of alternative j, ξ j is an unobservable attribute of alternative j, ε ij is an error term from the unobservable utility.  l is the number of type l content services, and the amount of the indirect network effects of type l content is

Following Doganoglu and Grzybowski(2007), we assume the a consumer’s utility doesn’t depend on former choice.

5

captured by  lωlj . Because access services are compatible in the broadband market, there is no difference in the number of contents services. But, the parameter of network effects

ω j might differ between alternatives by a technological nature of each type of

access service. We later test the hypothesis of equality of

ωj .

ε ij is identically and independently distributed (IID) among

Assuming

broadband alternatives but not IID between broadband alternatives and not subscribers, we can obtain the choice probability from the nested logit model. V j /(1−σ )

s j = s j|BB sBB =

(Σ

e

k∈BB

Vk /(1−σ )

e

(Σ e ) ∑ (Σ e k∈g

g

)

Vk /(1−σ ) 1−σ

k∈g

)

δ k /(1−σ ) (1−σ )

(2)

Following Berry (1994), we derive a linear regression model from a two-stage nested logit model.

ln( s j ) − ln( s0 ) = x j β − α p j + ∑ l lωlj + σ ln( s j|BB ) + ξ j

(2)

, where s j is the share of alternative j and s0 is the share of non subscribers, s j |BB is the

σ shows the degree of the error term's correlation between broadband using alternatives. If σ = 0 , share of alternative j within broadband using alternatives. The parameter

the error terms of all alternatives are IID and the choice structure reduces to the conditional logit model. Since the price, share within broadband access services and number of content services are endogenous variables, we need instruments6. Notice that estimating equation (2) without the indirect network effects may confuse a direct price effect with an indirect network effect because the number of content services depends on the number of broadband subscribers. When price declines, the number of broadband subscriber increases by the direct price effect. Then, the number of content services might increase by interdependency. In the case of the broadband market, estimates without the indirect network effects may have bias toward excess estimation7. It is easy to show the endogeneity of these variables. Clements and Ohashi (2005) illustrate one condition. 7 It is desirable to estimate a simultaneous equation model that includes network subscription equations and content service equations like Clements and Ohashi (2005). We use the number of subscribers in only ten periods in each prefecture to increase a number of observations. However, there are no differences in number of content services. 6

4. Data Quarterly data on each prefecture’s number of access services subscribers was obtained from MIC. We further calculated the subscription shares by dividing MIC’s number of subscribers by the number of households in each prefecture8. We analyze the indirect network effects of the most representative broadband content services, such as music distribution, video sharing, and online games. We then collected the number of music distribution firms, number of video sharing services, and number of online games for broadband content services from reliable sources. 4.1 The share of broadband access services We calculated shares of each prefecture from MIC’s “The number of subscribers of broadband services” and the number of households from the “National Census of Heisei 17.” The prefecture data in “The number of subscribers of broadband services” was released from June 2005 to December 2007. Before June 2005, MIC did not have the authority to require broadband providers to report the number of regional subscribers, so our data consists of 10 periods and 47 regions. 4.2 Price and Speed of Broadband access services There are several different broadband access services within one alternative in each prefecture, but we are unable to observe shares within one alternative. Hence, the price attribute of access services in each prefecture is represented by the service whose monthly charge per bandwidth is the lowest, and is available at the prefecture’s government office9. We collected the availability of service data from the web site “RBB Today,” run by “IRI Commerce and Technology,” who provides broadband access service information and the price and speed of services from the access provider’s web page. Monthly charges contain a fixed charge of access service, ISP service, equipment rental fee, and, in case of ADSL, NTT’s access charge for line sharing. In the case of FTTH, the charge of service for single-family homes and for multi-unit dwelling is different. So, we use a weighted average of the two charges calculated by the rate of households who live Hence, a reliable estimation of the content service equation is impossible. Structural estimation is one further issue. 8 MIC’s statistics include the subscribers of Fixed Wireless Access (FWA) service. In Japan. FWA is used in some rural areas. But, the number of subscribers of FWA in December 2007 is only 12.585 and a service plan is not available. So, we ignore FWA in our study. 9 Because price and speed are included in linear additive form in equation (1), that means a rational consumer chooses the service whose monthly charge per bandwidth is the lowest among the alternatives.

in apartment buildings according to “The statistic census of House and Land of Heisei 15.” We do not account for bundle discounts for broadcasting or telephone, temporary discounts for new subscribers, or discounts for large number of subscribers in apartment buildings. In Japan, almost all access service providers offer a free installation campaign, so we do not take into account this initial cost10. We use the consumer price index to adjust the nominal monthly charges. Following Ida and Kuroda (2006), we use nominal speeds of services. Figure 3 is the average of real price and nominal speed of all regions weighted by regional subscribers in a given period. Figure 4 is the average of real price and nominal speed of all periods in a representative prefecture in which the incumbent power company has a head office11. Table 3 Real price and nominal speed by periods Table 4 Real price and nominal speed by representative prefectures 4.3 Number of broadband content services In this paper, we estimate the indirect network effects from music distribution, video sharing, and online game services. For music services, we use the number of firms that have an agreement with the “Japanese Society for Rights of Authors, Composers and Publishers (JASRAC)” for music copyrights. JASRAC defines digital music delivery service as an “interactive delivery,” and it contains a service for fixed internet users and mobile internet users12. Hence, we use the share of fixed internet delivery in the sales figures of digital music delivery released by the “Recording Industry Association of Japan (RIAJ)” to estimate the number of firms for Internet delivery. For video sharing services, we use a list of video sharing services recorded in “Internet Hakusyo 2007” published by Impress R&D. For online games, we count the number of titles in commercial service from the list of “2007 Online Game Hakusyo,” published by Media Create Co., Ltd. It provides a list until March 2007. We complement the list using a database of online gaming web news, “4Gamer.net” run by Aetas, Inc. Table 5 shows the descriptive statistics and Table 6 shows the correlation matrix. To analyze initial costs and former periods status, it is necessary to estimate a dynamic model with stock variables. Keane and Wolpin (1997) is a pioneer work of dynamic discrete choice model. Since, it requires large data and heavy computation, it is left as a further issue. 11 In Japan, many incumbent power companies entered into the telecommunication market and supply FTTH services. 12 According to RAIJ’s statistics, 93.3% of digital music is delivered through mobile phones in 2007. 10

Table 5 Descriptive statistics Table 6 Correlation matrix 4.4 Instruments The network subscription equation contains five endogenous variables, price, share within access service, and three broadband content services. We employ two instruments from the cost structure. One is the access charge of dry copper for DSL service, and the other is the access charge of dark fiber. We employ three instruments from the offline content market. Offline content services have no effect on broadband subscription, but have an effect on content services. Specifically, we employ the amount of production units of audio software (CD, cassette, analog disks, and others) and video software (DVD, Tape and others) from the statistics of RIAJ and the amount of sales of video game software from the statistics of “TV Games Sangyou Hakusyo,” published by Media Create Co., Ltd. 5. Estimated results This section presents the estimated results of the network subscription equation. We estimate one OLS model and three two-stage least square models. 2SLS (1) is the model that excludes indirect network effects, 2SLS (2) is the model with indirect network effects that is constrained to have the same parameters throughout each alternative, and 2SLS (3) is the model with indirect network effects. The Hauseman test supports fixed effects in all estimated models. Table 7 shows the estimated results. Table 7 Estimated results of the network subscription equation 5.1 Results of price and speed parameters The estimated parameters of price and speed are significant at a 95% confidence level in 2SLS (1) and 2SLS (3), but those of OLS and 2SLS (2) are not significant. Because the direct price effect and the indirect network effects are not distinguished, the speed and price parameters of 2SLS (1) are larger than those of 2SLS (3). Table 8, and 9 show the price and speed elasticities of 2SLS (3) computed by the “method of sample enumeration”13. In these tables, the rows indicate the alternative of

13

The price elasticity for alternative j over change of alternative k’s price is

ε p| j ,k = p j /(1 − σ )∂V / ∂p j 1(k = j ) − σ sk |B (t )=1 − (1 − σ ) sk  . The speed elasticities

price change, and the columns indicate the alternative of choice probability change. No subscribe is the alternative of not signing up for any alternative. Since there is high price elasticity between broadband alternatives, a price increase in one access service will not make a consumer stop using broadband, but merely change to another broadband alternative. We can calculate the willingness to pay (WTP) for speed by dividing the speed parameter over the price parameter. The WTP to speed is 18.13yen/Mbps on average.14 Table 8 Price Elasticities Table 9 Speed Elasticities There are many analytical tools of competition policy that use price elasticities (i.e., market definition, merger analysis, and so on). Hausman, Leonard and Zona (1995) show that the use of overestimated elasticity induces an underestimation of price increase from mergers and defining larger markets. Hendel and Nevo (2006) point to a bias of price cost margin (PCM) estimates by confusing direct price effect (substitution between goods) and indirect price effect (substitution between times.) Table 10 shows the estimated price cost margin under the assumption that each type of broadband access service is supplied by a hypothetical monopolist. Table 10 Estimates of price cost margin under assumption of hypothetical monopolist. In the case of broadband market, network effects have large influence on consumer subscription. Hence, there is a large upward bias by confusing a direct price effect and an indirect one. From our estimation, the price coefficient of 2SLS (1) is 20 times as large as that of 2SLS (3). From a small difference of

σ , there is a difference of

20 times between price elasticities. In the case of a market that is sensitive to indirect network effects, it is preferable to use a model that can distinguish between direct price effect and indirect price effect by controlling for indirect network effects. 5.2 Results of the indirect network effects parameters calculate same way. In logit models, elasticity at sample average differs from average of elasticity at each sample. Louviere, Hensher and Swait(2000) recommends to use “method of sample enumeration“ that weighing average of a each sample elasticity by choice elasticity. 14 Because of figure of price is larger than other’s, we rescale a price to thousand yen. So, in our data, WTP is WTP = β s / β p *1000 .

We turn now to the results of content services. We performed an F test for linear equality restrictions on the parameters of content services between alternatives. The null hypothesis was rejected at the 1% significant level in all models. Consequently, there is a difference in indirect network effects from content services between broadband access services which affects the share of access services. Hereafter, we use estimates of 2SLS (3). Despite the fact that the sign of the coefficients shows an increase or decrease in the utility of each alternative, the sign of the coefficients does not show a share of each alternative, since these alternatives depend on the relative size of utility. In order to understand the estimated results, we calculate the content service elasticity. Table 11 Content services elasticities Table 11 shows the average content services elasticities computed by a method of sample enumeration. In our model, we can divide the elasticities of content services by the share of alternatives in two parts15. Between elasticities show the substitution of broadband subscription (“the first migration”),and within elasticities show the substitution among broadband access services (“the second migration”). From the structure of our nested logit model, between elasticities are almost the same for each broadband access services. On the other hand, within elasticity shows a substitution of broadband services that reflects a difference of indirect network effects between alternatives. To see the first migration, the share of FTTH is increased by all contents services. The share of ADSL is increased by music distribution services and movie sharing services but decreased by online game services. The share of CATV is increased by music distribution services and online game services but decreased by movie sharing services. The share of broadband services is increased by all contents services, especially by online game services. To see the second migration, the share of ADSL within broadband services is decreased by all contents services, that of CATV is decreased by movie sharing services and online game services, and that of FTTH is increased by all contents services.

Service elasticities differ from attributes elasticities (price or speed), because when the number of services change, all alternative’s utilities simultaneously change. The 15

service elasticities equation is ε lj =  l /(1 − σ )

{∑

Within elasticities equation is ε lj|BB =  l /(1 − σ )

k∈BB

(ω

lj

}

ωlk 1(k = j ) − σ sk |BB − (1 − σ ) sk  .

− Σ k∈BBωlk sk |BB ) . The sum of within

elasticities and between elasticities is total elasticity ( ε within + ε between = ε ).

Therefore, music distribution services stimulate the migration from ADSL to FTTH, and movie sharing services and online game services stimulate the migration from ADSL and CATV to FTTH. The online games category has the strongest effect among contents services. To show the migration factor to FTTH, we specify the change of each alternative’s utility from June 2005 to December 2007. Figure 3 shows that the migration to FTTH is caused by the change of indirect network effects from content services and not by a change of the alternative’s attributes (price and speed). Figure 3 Detailed utility change from 2005 to 2007 6. Broadband access provider’s vertical foreclosure and net neutrality. In this section, we describe the simulated results of vertical foreclosure by a hypothetical monopolist that monopolizes one type of access service. Based on the simulation results, we assess the effect of net neutrality regulation. 6.1 Incremental profit of foreclosure Broadband content service providers offer their services to all broadband users, no matter what access service they use. From the estimated results, because the indirect network effects on FTTH from content services are larger than on other access services, the diversity of broadband content service results in an increase of FTTH subscribers, hence FTTH providers have an incentive to invest in content services without vertical foreclosure. On the other hand, ADSL providers and CATV providers have the incentive to integrate with content services to attempt to seize FTTH customers. Table 10 shows the simulated incremental access service profits of hypothetical monopolists in each access service derived from vertical foreclosure in content service providers –in order not to provide their content to other access services16-. Table 12 Incremental access service profit per month from vertical foreclosure in content service providers Incremental profits derived from vertical foreclosure depend on profits per customer and the number of customers taken away from other competitors. As subscribers of broadband services increase, the number of customers potentially seized increases, hence profit from vertical foreclosure augments. Incremental profit of hypothetical ADSL monopolists on the last quarter of 2007 is the largest among access Per customer incremental revenue is calculated by deducting from price the estimated hypothetical monopolist’s cost shown in table.8. 16

providers. Because FTTH service’s utility gain from content services is larger than that of ADSL and CATV, the utility loss from rivals’ foreclosure is larger too. Hence, the FTTH provider’s customer loss is larger than that of ADSL and CATV. To see the effect on welfare, hypothetical ADSL monopolist’s vertical foreclosure makes CATV users and FTTH users lose utility. The monthly utility loss from music distribution services amounts to 9.03 million yen, that from movie sharing services amounts to 408 million yen, and that from online game services amounts to 265 million yen. 6.2 Likelihood of Foreclosure In order to assess a provider’s incentive for vertical foreclosure, we must consider not only the incremental profit from the foreclosure but also the reduction in content services revenue. Following Rubinfield and Singer (2001), we examine the likelihood of foreclosure by using market data. In the last quarter of 2007, a hypothetical ADSL monopolist’s incremental revenue from foreclosing in movie sharing services was 408 million yen per month, and from foreclosing in online game services is 265 million yen per month. On the other hand, the revenue per month of “nikoniko douga,” a representative movie sharing service in Japan, was about 100–150 million yen in the same period. In addition, the average online games revenue per month, calculated by dividing the online game market’s sales in 2007 by the number of titles, is about 100 million yen17. Because the share of ADSL in the broadband market is 47%, if the ADSL provider could compensate for half of the content provider’s revenue, his proposal on foreclosure with the content provider would come to an agreement. From this data, the hypothetical ADSL monopolist’s incremental profit is large enough to compensate for the expected reduction of content service revenue. Hence, vertical foreclosure is profitable for the hypothetical ADSL monopolist. In fact, there are various ADSL providers in the Japanese broadband market, and profit from vertical foreclosure in the real world is smaller than that of a hypothetical ADSL monopolist. However, as broadband subscribers increase, incremental profits from foreclosure also rise, and the concentration of ADSL providers increases as well (At the end of 2006, HHI was 2,320 and in 2007 it increased to 2.404. Moreover, the 4th and 5th largest providers plan to merge in 2009). According to this trend, the incremental profit of foreclosure may become large enough in the future. Therefore, net neutrality, that prohibits vertical foreclosure, helps consumers migrate to FTTH.

We use data on sales of the online game market in 2007 extracted from “Famitsu game hakusyo 2008.” 17

7. Conclusion In this paper, we estimated the demand for broadband access services with indirect network effects. From the estimated results, we come to three main conclusions. First, a consumer’s migration from ADSL to FTTH is explained by the indirect network effects of services diversification (i.e., the number of music distribution, video sharing, and online games services), not by the attributes of access services (i.e., price and speed). Second, there is a large upward bias in the estimates of price elasticity if indirect network effects are ignored. Finally, the ADSL service providers have the strongest incentive to foreclose with broadband content service firms. Therefore, net neutrality helps consumers migrate to FTTH, and maintains the consumers welfare derived from indirect network effects. In the Japanese broadband market, only the NTT group provides both ADSL and FTTH service. NTT group’s share of FTTH service was 71.4% at the end of 2007 and is growing. After the second migration, the importance of competition between FTTH providers may increase, which will require further consideration of the effect of net neutrality. It is true that a prohibition of vertical foreclosure helps “the second migration, but it sacrifices competition between ADSL and FTTH. If competition within FTTH does not work well, the second migration might be harmful. Therefore, it is desirable to promote competition within FTTH providers when promoting “the second migration” by net neutrality. Because choice structure analysis requires data availability on the choice of type of broadband service, examination of each firms profits derived from vertical foreclosure is a further issue. References Berry, S. (1994, Summer). Estimating Discrete-Choice Models of Product Differentiation.

RAND Journal of Economics, 25(2), 242–262. Cardona, M., Schwarz, A., Yurtoglu, B., & Zulehner, C. (2009, February). Demand Estimation and Market Definition for Broadband Internet Services. Journal of

Regulatory Economics, 35(1), 70–95. Clements, M., & Ohashi, H. (2005, December). Indirect Network Effects and the Product Cycle: Video Games in the U.S., 1994–2002. Journal of Industrial Economics, 53(4), 515–542. Church, J., & Gandal, N. (2005). Platform Competition in Telecommunications.

Handbook of Telecommunications Economics. Volume 2. Technology Evolution and the Internet (pp. 117–153). Amsterdam and San Diego:. Enterbrain. (2008). Famitsu game hakusyo 2008. Tokyo: Enterbrain. Gandal, N. (2002, Spring). Compatibility, Standardization, and Network Effects: Some

Policy Implications. Oxford Review of Economic Policy, 18(1), 80–91. Gandal, N., Kende, M., & Rob, R. (2000, Spring). The Dynamics of Technological Adoption in Hardware/Software Systems: The Case of Compact Disc Players. RAND

Journal of Economics, 31(1), 43–61. Hausman, J., Leonard, G., & Zona, J. (1994, April). Competitive Analysis with Differenciated Products. Annales d'Economie et de Statistique, no. 34: 159–180. Hitt, L., & Tambe, P. (2007, October). Broadband Adoption and Content Consumption.

Information Economics and Policy, 19(3–4), 362–378. Hendel, I., & Nevo, A. (2006, November). Measuring the Implications of Sales and Consumer Inventory Behavior. Econometrica, 74(6), 1637–1673. Ida, T. (2006). Broadband, Information Society, and the National System in Japan.

Global Broadband Battles: Why the U.S. and Europe Lag while Asia Leads (pp. 65–86). Innovation and Technology in the World Economy series. Ida, T., & Kuroda, T. (2006, January). Discrete Choice Analysis of Demand for Broadband in Japan. Journal of Regulatory Economics, 29(1), 5–22. Ida, T., & Sakahira, K. (2008, October). Broadband Migration and Lock-In Effects: Mixed Logit Model Analysis of Japan's High-Speed Internet Access Services.

Telecommunications Policy, 32(9–10), 615–625. Zaidanhoujin Internet Kyoukai (Ed.) 2007. Internet hakusyo 2007. Impress R&D. Media Create. (2007). 2007 Online Game Hakusyo. Tokyo: Media Create. Media Create. (2006). 2006 TV Games Sangyou Hakusyo. Tokyo: Media Create. Media Create. (2007). 2007 TV Games Sangyou Hakusyo. Tokyo: Media Create. Media Create. (2008). 2008 TV Games Sangyou Hakusyo. Tokyo: Media Create. Nair, H., Chintagunta, P., & Dube, J. (2004, March). Empirical Analysis of Indirect Network Effects in the Market for Personal Digital Assistants. Quantitative Marketing

and Economics, 2 (1), 23–58. Rubinfeld, D., & Singer, H. (2001, September). Vertical Foreclosure in Broadband Access?. Journal of Industrial Economics, 49(3), 299–318. Savage, S., & Waldman, D. (2005, September). Broadband Internet Access, Awareness, and Use: Analysis of United States Household Data. Telecommunications Policy, 29(8), 615–633. Train, K. E. (2003). Discrete choice methods with simulation. Cambridge: Cambridge University Press.

Figure 1 The progress of broadband subscribers in Japan

Subscriber

The progress of broadband subscribers in Japan

16,000,000 14,000,000 12,000,000 10,000,000 8,000,000 6,000,000 4,000,000 2,000,000 0 2002Q2

2003Q2

2004Q2

2005Q2

ADSL

2006Q2 CATV

2007Q2 FTTH

2008Q2

Table 1 The progress of price and speed of broadband access services Monthly fixed charge of major broadband access services Film NTT east NTT east NTT east SoftbankBB SoftbankBB USEN B Flet's Yahoo!BB Yahoo!BB BROAD– Name of services Flet's ADSL Flet's ADSL Hyper 8M 50M GATE01 Technology ADSL ADSL FTTH ADSL ADSL FTTH Speed 1.5Mbps 47Mbps 100Mbps 8Mbps 50Mbps 100MBps December 26, 2000 4,050 6,100 July 16, 2001 3,800 6,100 November 8, 2002 3,100 9,000 2,830 4,800 December 1, 2002 2,600 5,800 2,830 4,800 April 1, 2003 2,600 4,500 2,830 4,200 April 28, 2004 2,600 4,500 2,830 2,980 August 6, 2004 2,600 2,940 4,500 2,830 2,980 September 1, 2004 2,600 2,940 4,500 2,830 3,957 2,980 February 1, 2007 2,600 2,940 4,100 2,830 3,957 2,980 March 1, 2008 2,600 2,940 4,100 2,830 3,957 2,980 *1 NTT east's Flet's service requires the ISP charge and the modem retal fee *2 Charges of Softbank and USEN are including the ISP charge and the modem rental fee *3 Ignoring initial costs and temporal discount Source: "WHITE PAPER 2005 Information and Communications in Japan" and Provider's Web site

J:COM J:COM NET CATV 30Mbps

5,775

Table 2 Experience in purchasing online content

Experience of purchase of an online contents Music files Online games e-books Videos Pictures Businss information Education News/Special site Sports Broadcasting Others n=645 Source: "2007 wthite paper of Internet"

61.6% 28.4% 26.7% 20.0% 8.4% 5.9% 5.1% 4.2% 2.0% 3.9%

Figure 2 –Progress of broadband content services in Japan

Progress of broadband content services in Japan Number 800 700 600 500 400 300 200 100 0 Sep-05

Jan-06

May-06

Sep-06

Number of online music services Number of online games

Jan-07

May-07

Sep-07

Number of video sharing services

Table 3 Real price and nominal speed by period

Sep-05 Dec-05 Mar-06 Jun-06 Sep-06 Dec-06 Mar-07 Jun-07 Sep-07 Dec-07

Price(Real/yen) ADSL CATV FTTH 3,153 5,227 4,722 3,154 5,227 4,725 3,160 5,239 4,726 3,145 5,192 4,688 3,137 5,177 4,675 3,145 5,190 4,690 3,162 5,232 4,720 3,147 5,202 4,698 3,141 5,196 4,694 3,130 5,267 4,683

Speed(Mbps) Price per bit （yen/Mbps) ADSL CATV FTTH ADSL CATV FTTH 50 39 199 62.47 134.68 23.74 50 39 197 62.48 134.96 23.93 50 39 195 62.60 135.48 24.27 50 39 193 62.30 134.41 24.27 50 39 191 62.15 133.72 24.46 50 39 190 62.31 134.06 24.67 50 41 189 62.65 126.73 24.91 50 41 188 62.36 125.85 24.94 50 51 188 62.24 102.79 24.92 50 56 188 62.01 93.82 24.89

Table 4 Real price and nominal speed by representative prefectures

Price(Real/yen) ADSL CATV FTTH Hkkaido 3,147 5,770 4,088 Miyagi 3,140 5,023 4,781 Tokyo 3,147 4,984 3,864 Toyama 3,147 5,548 5,333 Aichi 3,135 5,759 3,836 Osaka 3,144 5,354 5,235 Hirodhima 3,138 4,507 4,492 Kagawa 3,152 5,160 5,134 Fukuoka 3,142 5,773 4,346 Okinawa 3,145 4,181 4,554

Speed(Mbps) Price per bit （yen/Mbps) ADSL CATV FTTH ADSL CATV FTTH 51 30 100 62.33 192.32 40.88 51 35 100 62.18 143.52 47.81 51 110 100 62.32 45.31 38.64 51 30 100 62.31 184.95 53.33 51 30 100 62.08 191.98 38.36 51 50 455 62.25 106.16 11.51 51 30 100 62.13 150.23 44.92 51 15 100 62.42 344.01 51.34 51 30 100 62.22 192.42 43.46 51 30 100 62.27 139.38 45.54

Table 5 Descriptive statistics

Mean

Std.Dev.

Minimum

Maximum

Variables Price(thousand \) 4.46802 1.10357 3.09546 7.47039 Speed (Mbps) 83.0348 123.391 8 780.634 Music distribution 467.2 133.648 169 696 Movie sharing 12 8.57042 0 23 Online game 120.8 26.6958 74 154 Log(Sj| BB) -1.37082 0.847969 -5.89783 -0.16288 Instruments Offline Music 88859.4 10699.1 71701 104966 Offline Movie 28165.1 7499.76 19442 45088 Offline Games 15263900 2896140 10891400 20678300 Dry copper 1.72804 0.129303 1.56051 1.92573 Dark fiber 1365.44 40.7681 1290.74 1411.4

Price(thousand \) Speed (Mbps) Music distribution Movie sharing Online game Log(Sj| BB) Offline Music Offline Movie Offline Games Dry copper Dark fiber

Price Speed Music Movie Online Offline Offline Offline (thousand \) (Mbps) distribution sharing games Log(Sj| BB) Music Movie Game Dry copper Dark fiber 1 0.41673 -0.00752 -0.00933 -0.00718 -0.58734 0.00697 0.01104 0.00025 0.00445 0.01862 0.41673 1 0.00873 0.00836 0.00849 0.10705 -0.00118 -0.00315 0.00157 -0.00891 0.12715 -0.00752 0.00873 1 0.86195 0.91659 0.05505 -0.30754 -0.46096 0.41346 -0.84953 -0.16934 -0.00933 0.00836 0.86195 1 0.96574 0.05727 -0.42371 -0.63082 0.19306 -0.89796 -0.16137 -0.00718 0.00849 0.91659 0.96574 1 0.05837 -0.46146 -0.59374 0.2476 -0.95477 -0.15474 -0.58734 0.10705 0.05505 0.05727 0.05837 1 -0.0296 -0.03863 0.01597 -0.05414 0.08418 0.00697 -0.00118 -0.30754 -0.42371 -0.46146 -0.0296 1 0.89026 0.56716 0.57187 0.09694 0.01104 -0.00315 -0.46096 -0.63082 -0.59374 -0.03863 0.89026 1 0.44353 0.64685 0.15419 0.00025 0.00157 0.41346 0.19306 0.2476 0.01597 0.56716 0.44353 1 -0.0617 -0.03559 0.00445 -0.00891 -0.84953 -0.89796 -0.95477 -0.05414 0.57187 0.64685 -0.0617 1 0.1554 0.01862 0.12715 -0.16934 -0.16137 -0.15474 0.08418 0.09694 0.15419 -0.03559 0.1554 1

Table 6 Correlation matrix

OLS 1410 0.9993 0.9992 β -0.0071 0.0001

2SLS(1) 1410

S.E β S.E Price (thousand yes) 0.0089 -6.0000 ** 1.6860 Speed (Mbps) 0.0001 0.0957 ** 0.0236 Music Music * ADSL 0.0002 ** 0.0000 Music * CATV 0.0002 ** 0.0000 Music * FTTH 0.0003 ** 0.0000 Movie sharing Movie sharing * ADSL 0.0074 ** 0.0006 Movie sharing * CATV 0.0076 ** 0.0006 Movie sharing * FTTH 0.0088** 0.0006 Online games Online games * ADSL 0.0029 ** 0.0002 Online games * CATV 0.0029 ** 0.0002 Online games * FTTH 0.0034 ** 0.0002 Ln(Sj|BB) 0.9201 ** 0.0146 0.8281 ** 0.2050 * 　　Significance at the 95-percent confidence level. ** 　　Significance at the 99-percent confidence level.

N R-squared Adjusted R-squared

Table 7 Estimation results of the network subscription equation

**

0.0004

1.0071 ** 0.0083

0.0027 ** 0.0001

0.0091

β S.E 0.0991 0.1008 0.0012 0.0009 0.0002 ** 0.0000

2SLS(2) 1410

0.0034 ** 0.0036 ** 0.0049 ** 0.8320 **

0.0008 0.0003 0.0009 0.0936

0.0059 ** 0.0008 0.0048 ** 0.0012 0.0078 ** 0.0012

0.0001 ** 0.0001 0.0001 ** 0.0000 0.0002 ** 0.0000

β S.E -0.2438 ** 0.1097 0.0044 * 0.0011

2SLS(3) 1410

Table 8– Price Elasticities Price elasticities ADSL ADSL -2.6113 CATV 1.9111 FTTH 1.9913

CATV 3.2395 -4.2465 3.2951

FTTH No subscribe 3.1881 0.0351 3.0649 0.0237 -4.2194 0.0320

Table 9– Speed Elasticities Speed elasticities ADSL ADSL 0.7561 CATV -0.5556 FTTH -0.5784

CATV -0.3538 0.5243 -0.4066

FTTH No subscribe -2.0543 -0.0102 -1.7603 -0.0036 2.9441 -0.0276

Table 10 – Estimates of price cost margin under assumption of hypothetical monopolist. ADSL CATV FTTH Price 3147.3 5214.9 4702.1 MC 2SLS(1) 3096.2 5164.5 4657.0 2SLS(3) 1942.0 3986.8 3587.7 PCM 2SLS(1) 0.0162 0.0097 0.0096 2SLS(3) 0.3829 0.2355 0.2370

Table 11–Content services elasticities

Contents elasticities ADSL

CATV

FTTH

No subscribe

Music Movie sharing Online games Music Movie sharing Online games Music Movie sharing Online games Music Motion picture Online games

Within Between -0.0240 0.0349 -0.0299 0.0383 -0.3258 0.2420 0.0039 0.0347 -0.1014 0.0384 -0.1749 0.2377 0.0509 0.0362 0.1279 0.0461 0.8283 0.2501

Total 0.0109 0.0084 -0.0838 0.0386 -0.0630 0.0627 0.0871 0.1740 1.0785 -0.6403 -2.7638 -9.8002

Figure 3 –Detailed utility change from 2005 to 2007 Detailed utility change from 2005 to 2007 dV 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1

ADSL Price

CATV Speed

Music

FTTH Movie

Online Game

Table 12 – Incremental access service profit per month from vertical foreclosure in content service providers

2005Q3 2005Q4 2006Q1 2006Q2 2006Q3 2006Q4 2007Q1 2007Q2 2007Q3 2007Q4

Music Movie sharing Online games ADSL CATV FTTH ADSL CATV FTTH ADSL CATV FTTH 5,033 2,439 2,806 211,668 111,782 119,136 141,674 65,141 70,613 5,568 2,608 3,291 236,869 119,898 139,484 157,833 70,093 82,782 6,141 2,720 3,897 265,229 125,551 164,977 175,683 73,692 97,975 6,702 2,831 4,514 293,221 131,219 190,850 193,253 77,326 113,423 7,278 2,881 5,160 322,534 134,030 218,206 211,522 79,245 129,603 7,900 2,967 5,930 354,470 138,666 250,441 231,361 82,336 148,845 8,114 3,035 6,223 365,392 142,054 262,473 238,167 84,496 156,144 8,679 3,128 6,969 394,505 146,947 293,516 256,226 87,729 174,738 8,767 3,798 7,216 395,349 178,395 300,719 257,566 106,917 180,829 9,033 3,941 7,581 408,342 185,379 315,453 265,781 111,271 189,927 （thousand yen）