Unbundling the Incumbent and Entry into Fiber: Evidence from France

Transcription

1 Unbundling the Incumbent and Entry into Fiber: Evidence from France Marc Bourreau Lukasz Grzybowski Maude Hasbi Preliminary version. Please, do not cite Abstract We use panel data on 36,080 municipalities in metropolitan France over the period to estimate two models of entry into local markets by: (i) the incumbent and two other operators using fiber technology and (ii) operators seeking wholesale access on the legacy network via local loop unbundling (LLU). We find that local market presence of LLU operators has a positive impact on entry by fiber operators. First, we observe in the data that in the case of two fiber entrants, the deployment of FttH is always preceded by entry via LLU. Moreover, the estimation results suggest that the decision to deploy fiber by the incumbent and competitors is positively influenced by the presence of other LLU operators. High speed fiber broadband allows firms to differentiate their offering from DSL-based services. Moreover, the presence of upgraded cable network stimulates fiber deployment. Firms using DSL technology deploy FttH to avoid being preempted by cable operator. We also find that the deployment of VDSL technology, which allows higher Internet speed on the copper network, slows down fiber deployment. Firms may choose to upgrade copper lines on the legacy network instead of investing in fiber networks. We use the estimates to calculate entry thresholds into local markets, which are substantially lower for broadband provision via LLU than via fiber and decrease over time. Fiber deployment becomes cheaper over time but remains unprofitable for the vast majority of municipalities in France. Telecom ParisTech, Department of Economics and Social Sciences, 46 rue Barrault, Paris, France. marc.bourreau@telecom-paristech.fr Telecom ParisTech, Department of Economics and Social Sciences, 46 rue Barrault, Paris, France. lukasz@mushroomski.com Corresponding author: Telecom ParisTech, Department of Economics and Social Sciences, 46 rue Barrault, Paris, France. maude.hasbi@telecom-paristech.fr 1

2 1 Introduction Universal access to superfast broadband is on top of the public policy agenda in the European Union. Policymakers in all industrialized countries have made investments in next generation access networks a priority. The roll-out of fiber-optic networks, delivering superfast broadband access to the Internet, is expected to improve productivity and stimulate growth and job creations (see Czernich et al. 2011; Ahfeldt et al. 2015). In the US, the National Broadband Plan aims at providing 100 Mbit/s connections to 100 million households by In Europe, the Commission has set as a target that half of European households should have the ability to access the Internet at speeds of 100 Mbit/s or more by In September 2016, the Commission further announced that by 2025 all European households should have access to connections with speed of at least 100 MB/s. 3 The roll-out costs of next generation access networks are, however, significant. For example, the European Commission has estimated that it would cost between 181 and 268 billion euros to achieve Digital Agenda s target. 4 Private operators deploying fiber also face competition from the previous broadband technology, which has been promoted in Europe through specific regulations. The contribution of this paper is to analyze how broadband provision on the incumbent s legacy copper network via local loop unbundling (LLU) influences entry and deployment of fiber infrastructures. 5 To the best of our knowledge, this is the first paper which at a granular local level analyzes the deployment of fiber infrastructures and the impact of legacy broadband networks on the deployment of next generation access infrastructures. Our results shed light on the ongoing debate on the impact of LLU and the optimal policies to achieve universal access to high speed broadband. Telecommunications industry in the EU has been subject to regulation since the opening to competition in the 1990 s, to limit the exercise of market power by incumbent operators and allow competition to emerge. In particular, to foster entry and competition in the broadband 1 Connecting America: The National Broadband Plan, Federal Communications Commission, March A Digital Agenda for Europe, European Commission, COM(2010) Source: State of the Union 2016: Towards a Better Europe - A Europe that Protects, Empowers and Defends 4 See COM(2010) 472 final. 5 Local loop unbundling is the regulatory process of allowing multiple telecommunications operators to use connections from the telephone exchange to the customer s premises. 2

3 market, the European Commission has implemented in the early 2000 s wholesale access to the incumbents local network, a regulatory policy known as local loop unbundling (or LLU). LLU requires incumbents to grant access to their physical local copper infrastructure, at regulated prices, to enable entrants to provide DSL ( digital subscriber line ) broadband services. The introduction of LLU has generated heated policy debates. While LLU can facilitate entry of alternative operators and allow new entrants to acquire market experience progressively, some (in particular incumbent) operators have argued that it may undermine investment incentives for both incumbent and entrant operators. Recently, the regulatory debate has centered on the impact of LLU on firms incentives to roll-out next generation access networks. Due to their high deployment costs, fiber-optic access infrastructures are likely to coexist with the legacy copper networks for some time. A Report commissioned by the European Competitive Telecommunication Association (WIK 2011) argued that promoting LLU via low access prices would foster investments in fiber, while another Report written for the European Telecommunications Network Operators (Plum 2011) concluded that it would rather discourage fiber deployment. Therefore, governments and regulators face a complex trade-off between promoting competition on the legacy copper network via LLU and providing industry players with the proper incentives to roll-out fiber networks. However, even though this question is critical for both policymakers and academics, so far there exist only limited and mostly aggregated empirical evidence on the impact of regulation of legacy network on the roll-out of fiber networks. In this paper, we analyze the impact of LLU on fiber investments using panel data on 36,080 municipalities in metropolitan France for the years The French market is interesting for such analysis for the following reasons. First, the regulation of wholesale broadband access via LLU in France has promoted entry and competition since early 2000 s. Second, in the past few years fiber broadband has been deployed not only by the former telecommunications incumbent Orange, but also its two main rivals, SFR and Free, which are also active in the provision of mobile services. Third, there is a single cable network operator, Numericable, which has been upgrading its network to a hybrid fiber-coaxial architecture allowing for Internet speeds comparable to fiber. Fourth, since 2013 Orange and other DSL operators have been upgrading 3

4 copper network towards VDSL technology, which can provide Internet speed of up to 50 Mbp/s to some consumers. 6 We use panel data on 36,080 municipalities in metropolitan France over the period to estimate two models of entry into local markets by: (i) the incumbent and two other operators using fiber technology and (ii) operators seeking wholesale access to the copper legacy telecommunications network via local loop unbundling (LLU). In the model of fiber entry, we allow the entry decision to depend on the local market presence of competing LLU operators. We also consider the impact of local market presence of upgraded cable modem and upgraded DSL lines to VDSL. When estimating the models, we take into account the role of local market characteristics such as market size, population density, income and unemployment. We find that local market presence of LLU operators has a positive impact on entry by fiber operators. In the data we observe that fiber entrants SFR and Free always enter via LLU first. However, based on the estimation results, the decision to deploy fiber by the incumbent Orange and the two competitors is also positively influenced by the presence of other LLU operators. High speed fiber broadband allows firms to differentiate their offer from DSL-based services. Moreover, the presence of upgraded cable stimulates fiber deployment. Firms using DSL technology deploy FttH to avoid being preempted by cable operator Numericable. This result suggests that firms recognize that consumers increasingly care about connection speed. Moreover, being first in the market matters for building consumer base and recouping investment costs. In addition, we find that the deployment of VDSL technology, which provides higher Internet speed on the copper network, slows down fiber deployment. Thus, firms may choose to upgrade copper lines on the legacy network instead of investing in fiber networks. Our results have important policy implications since they show that LLU regulation does not impede fiber entry. We find that the main variable that influences fiber deployment is market size. The market size required for the first operator and additional operators to enter is decreasing over time, which implies that it becomes easier to enter less densely populated municipalities. Entry thresholds for LLU operators are much smaller than for fiber operators, 6 The highest speed of DSL connection is 8 MB/s. ADSL2+ and VDSL provide higher speed in the range between 8 MB/s and 50MB/s for premises close to the exchange. Thus, these advanced technologies are not available to all consumers. 4

5 which reflects the scale of investment required to enter the market. Since LLU operators to a large extent rely on the incumbent s infrastructure, the scale of initial investment is much smaller. Our results also indicate that fiber provision will remain unprofitable in the vast majority of smaller municipalities in France. There is therefore a critical role for public policy to provide incentives to telecommunications firms to deploy fiber and enter less densely populated municipalities by means of co-investment agreements. The remainder of the paper is organized as follows. In Section 2, we review the relevant literature on regulation and investment in the telecommunications industry. Section 3 introduces the broadband industry in France. Section 4 presents the data. Section 5 introduces the econometric framework. Section 6 presents the estimation results. Finally, Section 7 concludes. 2 Literature Review The link between access regulation and investment incentives has long been recognized in the theoretical literature (see for example, Valletti 2003; Hori and Mizuno 2006; Klumpp and Su 2010). 7 A few recent studies look more specifically at the effect of access regulation on investment incentives when the new infrastructure co-exists with the legacy infrastructure (Inderst and Peitz 2012; Bourreau, Cambini and Doǧan 2012 and 2014). They show that access regulation of the legacy network (via LLU, for example) has ambiguous effects on the incentives to roll-out new infrastructure. The empirical literature on investments in fiber networks is rather scant. Most studies have analyzed the effects of regulation and competition between technologies on adoption of DSL broadband technology using aggregate country-level data. 8 Only few studies rely on more fine-grained data. In particular, Dauvin and Grzybowski (2014) use NUTS 1 level data for the EU countries, while Nardotto, Valletti and Verboven (2015) use micro-data at the local level for the UK. In general, competition is found to have a significant and positive impact 7 See also Guthrie (2006) and Cambini and Jiang (2009) for comprehensive surveys of literature. The relation between access and investment has generated heated policy debates, and a large number of policy papers which also have addressed this question; see, among others, Crandall, Ingraham and Sidak (2004), Hausman and Sidak (2005) and Hazlett and Bazelon (2005). 8 See, among others, Distaso, Lupi and Manenti (2006), Denni and Gruber (2007), Lee and Brown (2008), Bouckaert, Dijk and Verboven (2010), Lee et al. (2011), Gruber and Koutroumpis (2013). 5

6 on broadband diffusion but with some differences with respect to the role of inter- and intraplatform competition. 9 Only very few papers analyze investments in high speed broadband networks. In particular, Bacache, Bourreau and Gaudin (2014) analyze the migration from LLU to fiber in 15 European Member States over the period In another paper, Briglauer (2015) studies the determinants of fiber investments for the 27 EU Members over the period These two studies use aggregate country-level data. There are only two recent papers which use local market data. In a recent paper, Minamihashi (2012) uses municipality-level data for Japan in years and finds that LLU regulation imposed on the Japanese incumbent operator has prevented entrants from deploying new broadband infrastructure. According to counter-factual exercise, LLU regulation leads to 24% decrease in the roll-out of new infrastructure. However, the incumbent s NGA investments were not hindered by the LLU regulation. In another paper, Fabritz and Falck (2013) use data on local exchange areas in the UK for the years to analyze how the introduction of geographically differentiated regulation of the wholesale broadband access influences investment in NGA network by the incumbent. They find that deregulation has a positive effect on the roll-out of fiber. There is also a growing body of literature on the so called ladder of investment (LoI) which refers to a regulatory approach aiming to reconcile service-based competition (or intra-platform competition) and facility-based competition (or inter-platform competition). The LoI approach suggests that regulating access to the incumbent firm s local loop infrastructure, in order to create service-based competition, is not only pro-competitive by reducing barriers to entry but it is also an indirect device to promote facility-based competition. In this manner, service-based and facility-based competition are no longer substitutes, but become complements. A number of empirical studies put the LoI approach to the test. Using data on US States, Hazlett and Bazelon (2005) find a negative effect of unbundled access on the development of facility-based competition. Hausman and Sidak (2005) focus on five countries, the US, the UK, New Zealand, Canada, and Germany, and find that mandatory unbundling failed to spur 9 Inter-platform competition: when entrants build their own infrastructure; intra-platform competition: when entrants lease access to incumbents facilities. 10 See also Grajek and Röller (2012), who study the impact of pro-entry regulation on incumbents and entrants aggregate investments. 6

7 infrastructure investments by incumbents or entrants. Using data from the European broadband market, Crandall and Sidak (2007) conclude that there is weak or no evidence of a ladder of investment. Grajek and Röller (2009) use data from 20 EU Member States over a 10 year period and find that promoting market entry by means of regulated access undermines incentives to invest in facility-based competition. In another paper, Waverman et al. (2007) use data from 27 European countries and find that the intensity of access regulation measured by LLU prices has a negative impact on investment in alternative and new access infrastructure. Using data from 15 European Member States, Bachache et al. (2010) find no LoI effect between LLU and new infrastructure connections but a LoI effect between bitstream access and LLU connections. It thus appears that entrants climb the ladder of investment to LLU level but do not reach the next level to build their own infrastructure. 11 The literature on entry is mature and excellently reviewed by Berry and Reiss (2007). There are only few studies which analyze entry into telecommunications markets, which is due to a lack of appropriate local-level data. Most of the studies focus on entry into local telecommunications markets in the US prior to the FCC decision in 2004 to reverse its open access policy (see Greenstein and Mazzeo 2006; Economides, Seim and Viard 2008; Xiao and Orazem 2009 and 2011; Goldfarb and Xiao 2011), at a time when broadband was still in its infancy. In a more recent study, Molnar and Savage (2015) analyze the impact of entry using data on about 5,500 households in the US to analyze the relationship between broadband Internet market structure and product quality. Finally, Nardotto et al. (2015) study the deployment of broadband technology in the last decade in the UK. However, they do not observe entry in high-speed broadband (fiber) in their period of analysis. 3 Broadband Industry in France In France, five main operators compete in the retail broadband market: Orange, SFR, Free, Bouygues Telecom and Numericable. Orange is the historical fixed-line incumbent operator owning the legacy copper network, which is used to provide DSL broadband services. SFR, 11 Another growing body of literature analyzes competition between broadband technologies from a consumer perspective. See for instance Cardona, Schwarz, Yurtoglu and Zulehner (2009) and Grzybowski, Nitsche, Verboven and Wiethaus (2014) 7

8 Bouygues Telecom and Free are entrants which do not possess their own copper network. They provide broadband services by leasing access to the incumbent s (Orange) local access network via local loop unbundling (LLU). 12 Broadband can be delivered with different technologies: DSL, fiber, cable modem, Wifi and mobile broadband. In 2014, the respective market shares were 88% for the DSL technology, 3.6% for fiber, 6.6% for cable modem, and 1.8% for other broadband technologies, such as WiFi, mobile broadband and satellite DSL In France and in the European Union, the wholesale access to the copper local loop is regulated. 14 LLU has played a critical role in promoting broadband diffusion in France by enabling competition between the incumbent and alternative operators which do not own developed local telecommunications infrastructure. Based on the LoI theory, a phase of service-based competition is a prerequisite for alternative operators to enter the broadband market. The LLU entrants can first acquire market knowledge, build customer base and then they should have stronger incentive to develop their own network infrastructure. Figure (2) presents the architecture of LLU, which also referred to as wholesale local access. The copper network can typically be divided into three parts: access, backhaul connections and core networks. The access network (often called local loop or the last mile ) establishes the connection between the end user s premises and a main distribution frame (MDF), passing through a first network aggregation point which is called street cabinet. These elements belong to the incumbent operator. In order to provide Internet services, the alternative operator builds its own backhaul connection and core network down to the incumbent s MDF. Then it connects its customers to its own broadband equipment (DSLAM) which is located within the MDF, next 12 There is also a competitive fringe of small private or public broadband operators, which often focus specific market segments such as business consumers. 13 Source: ARCEP observatory - High and very-high speed Internet Retail market; SFR-Numericable 2014 annual results. 14 Pursuant to the 2000 European Community Regulation on unbundled access to the local loop, the local loop is the physical twisted metallic pair circuit in the fixed public telephone network connecting the network termination point at the subscriber s premises to the main distribution frame or equivalent facility. 8

9 to the incumbent s DSLAM. SFR and Free are the most active with regard to unbundling. 15 In France, almost 90% of metropolitan population has access to LLU offers. In order to increase the speed delivered through the copper network, a new technology, called VDSL2 has been implemented, which requires deploying fiber to the sub local loop between the MDF and street cabinet, where the DSLAM is then located. 16 The implementation of VDSL2, which can deliver speed superior or equal to 30 MB/s, has been authorized by the French telecommunications regulator ARCEP in October VDSL2 is cheaper to deploy than FttH network because the copper local loop is still used between the street cabinet and the customer s premises (see Figure 1). According to ARCEP, in the fourth quarter of 2014, more than half of the MDFs, covering 91% of the population, have been equipped with VDSL Optical Fiber From 2010 onwards, DSL operators have started to invest in Fiber to the Home network (FttH) in order to provide high speed broadband services on the retail market. Fiber optic networks are rolled-out up to the customer s premises and can carry video, data, voice and interactive video-telephone services. Orange and SFR are the most active in FttH providers. In January 2010, Orange and SFR signed co-investment agreement to deploy FttH in areas which are less densely populated. 19 In October 2014, Numericable aquired SFR threatening the viability of the co-investment agreement since some areas are also covered by the cable network of Numericable In comparison, the third biggest alternative DSL operator, Bouygues Telecom, unbundled only 985 MDFs. Unlike SFR and Free, Bouygues Telecom favored DSLAM renting either from the incumbent or LLU operator. 16 VDSL stands for very high bit rate digital subscriber line. This technology is also called Fiber to the Neighborhood (FttN). 17 In 2013, the authorization to implement VDSL2 only concerned lines in direct distribution, i.e, lines which are directly connected to a MDF. Since October 2014, ARCEP generalized the use of VDSL2 to all eligible lines, i.e., all lines connected to a street cabinet. 18 Another solution which has been implemented by Orange to improve broadband speed in rural areas, where some lines were not eligible to ADSL, was the creation of MDF shadow zones. The idea was to reduce the line distance by creating a new MDF next to the street cabinet, where the DSLAM would be placed. However, this solution has been stopped with the development of VDSL. 19 In frame of a co-investment agreement, there is one leading operator, which deploys the FttH network. The other operator participates to the financial cost in proportion of the market share it wants to obtain. However, we could not get information on the local market share targeted by the co-investors. We also do not have information on the municipalities covered by any of the co-investment agreements. 20 The merger has been cleared by the Competition Authority in October

10 Numericable, is the cable-operator. It provides broadband services on the retail market using its co-axial cable network. The cable network covers only 30% of the population living mostly in urban areas. From 2007 onwards, Numericable has started to upgrade its cable network by using DOCSIS 3.0 standard. Numericable does not deploy fiber network to consumers premises but brings it only to the last amplifier (Fiber to the Last Amplifier or FttLA). It then keeps relying on its co-axial local loop to provide broadband services to its customers. Numericable is at present, the biggest operator in the high speed broadband segment. Free deploys its own FttH network in very high density areas and co-invests in less densely populated areas. In July 2011, Free signed a co-investment agreement with Orange to deploy FttH in 60 agglomerations located in less densely populated areas. In 2014, another 20 municipalities have been added to the list. Since August 2012, Free has almost stopped deploying its optical distribution frames (ODF). At the same time Free has intensified its unbundling strategy and transfers its DSL subscriber towards the VDSL2 protocol. Bouygues Telecom does not deploy its own FttH network. It offers retail fiber services by subscribing to Numericable s bitstream offer on its upgraded cable or FttLA network. In addition, in December 2010, Bouygues Telecom signed a co-investment agreement with SFR to cover few high density areas. In January 2012, it has also signed a co-investment agreement with Orange concerning FttH deployment in both less densely populated and very high density areas. As of December 2014, Bouygues Telecom fiber offers were commercially available in 6 agglomerations, namely Paris, Lyon, Marseille, Toulouse, Bordeaux and Nice. Bouygues Telecom keeps also unbundling Orange s MDFs. Bouygues Telecom is not considered in this analysis due to a lack of detailed data on coverage and much smaller scale of activities. Other smaller private operators concentrate their investments in less densely populated areas. Many municipalities or departments across the country deploy FttH network or a mix of technologies, such as FttH, VDSL2, satellite or wireless local access to provide high speed Internet connection in rural areas. However, the analysis of public investment and its impact on private investment is out of the scope of the paper. The roll-out of FttH networks is in progress. According to ARCEP observatory, in December 2014, 4,064,000 premises were eligible to FttH and 356,000 households had subscribed to an FttH 10

11 offer, which compared to December 2013 represents an increase by 37% and 91% respectively. Besides, 8,707,000 premises were eligible to Numericable s FttLA offer. All fiber operators have chosen to add approximately 5 Euros mark up as compared to the price of DSL broadband offers. 4 The Data As discussed in Section 3, there are four main market players in French broadband market: Orange, SFR, Free and Numericable, which differ with respect to the broadband technology deployed and the timing of entry. The previous fixed-line incumbent operator, Orange, covered the whole country with xdsl technology by 2007 and deploys FttH alone or in co-investment with SFR or Free. SFR provides xdsl services based on the incumbent s copper network via LLU and deploys FttH alone or in co-investment with Orange or Free. Similarly, Free provides xdsl services via LLU and deploys fiber network alone or in co-investment with Orange or SFR. Finally, Numericable owns the cable network which has been upgraded to provide broadband services and also provides xdsl services on the business segment via LLU. Moreover, in April 2014 which is the last year of our data, Numericable acquired SFR. Table (1) shows the number of municipalities in which the first three operators entered with FttH and Numericable upgraded its cable network to provide high speed broadband. The remaining firms on the market provide xdsl broadband via LLU or mobile broadband. In this paper, we use data on copper infrastructure and the deployment of FttH and cable upgrade in municipalities in metropolitan France (excluding Corsica) in years , which was collected from different proprietary and public data sources. From Orange we received information on the state of deployment of FttH network at each concentration point within a municipality. 21 There are three possible states of FttH deployment: deployed, in deployment and planned. For year 2014 information was extracted on all concentration points having status deployed and ignored concentration points having status in deployment and planned since no timeline on their actual deployment is given. We consider that FttH is deployed in a municipality if there is at least one concentration point with status deployed. The other data received from 21 What is concentration point? 11

12 Orange includes information on copper network infrastructure in years We know the location and number of lines at each MDF as well as the number of lines upgraded to VDSL. We also know the number and identity of firms which unbundled particular MDFs. The information on deployment of FttH and cable upgrade by the other firms is publicly available. The information of FttH deployment by SFR, Free and Numericable was collected from the websites of these operators. SFR publishes a map of France listing municipalities in which it has deployed or will deploy FttH network. Thus, for each municipality we know whether SFR is planning to deploy its network alone or in co-investment with another operator but there is no timeline of deployment available. The information on deployment was verified using press releases published on its website which list municipality where households can subscribe to FttH as of October The data on FttH deployment by Free was collected from an unofficial website updated by Free s users community. This information is consistent with information available on other websites and with Free s annual reports. For each municipality in metropolitan France we know whether Free has active FttH connections. Furthermore, data on Numericable s cable upgrade to FttLA was collected from its website for years 2010 to For each municipality, we know whether Numericable has upgraded its cable network to provide high speed broadband services. The information on fiber and copper networks was completed with selected municipality-level socio-demographic statistics from the French National Institute for Statistics and Economics Studies (INSEE). In particular, we use municipality-level information on population size and density, number of flats and houses and unemployment rate. This information is published with a delay and available only until Since no better statistical information is available to the firms, we consider that they make decisions based on values with two-years lag. Moreover, information on the average fiscal income per municipality in years was collected from the website of General Direction of Public Finance (DGFIP). These data sets were merged using unique municipality INSEE code. However, since the code is not provided for municipality-level data collected from the websites of SFR, Free and Numericable, we assigned to each municipality INSEE code based on the name and department. 22 We collected historical information using wayback machine. Source: 12

13 After merging the data sets we have information on 36,080 municipalities (99.7%) for years There are 112 municipalities out of the 36,192 counted in 2014 which are missing due to administrative changes in years Some municipalities have been split into two different ones and some others merged resulting in changes in names and INSEE codes. Table (??) shows summary statistics for the data used in the analysis. 5 Econometric Model In this section, we introduce the model of entry, which we use to estimate the determinants of entry into local markets via LLU and FttH. As discussed in Section 3, an entrants who wants to provide DSL broadband services via LLU has to builds its own backhaul connection and core network down to the incumbent s MDF and then has to locate in the MDF its own broadband equipment to which its broadband consumers are connected. The entry via FttH requires deploying fiber optic from the backbone to every single house which is much higher capital investments. 23 In the previous literature on entry into broadband markets, both Xiao and Orazem (2011) in the study for US and Nardoto et al. (2015) for UK consider that investments made by LLU operators are sunk. The identification of sunk costs is based on a comparison of entry thresholds for markets which experienced entry with thresholds for markets without entry. There is a difference in the decision problem of the entrant and incumbent. The entrants will only enter when market size is big enough to cover entry costs. The incumbent firms, when deciding whether to continue operations do not need to take entry costs into account anymore. The existence of sunk entry costs implies that it takes less demand for an incumbent to continue operations than to support a new entrant. As shown in Table (5), in the data we observe a large number of LLU entries and exists. At the same time as Table (4) shows, there was a relatively 23 In general, the infrastructure fixed costs can be divided into country- and local market level costs. Countrylevel costs are related to the deployment and maintenance of the backbone network, administration, marketing, etc. Local market costs are related to network deployment and administration within the municipality. These costs depend on factors such as the size of geographic area to be covered, number of premises and spatial distribution of population. After the infrastructure is deployed, the marginal costs are relatively small and include factors such as modem rental expenditures and customer service. There are therefore economies of scale which imply that only a certain number of entrants can profitably operate in a municipality of certain characteristics. 13

14 small number of FttH entries and there were no exists at all. First, we model decisions of firms to enter via LLU assuming that the market demand is stochastic. At the end of each time period, firms decide whether to enter in the next period into new local markets and whether to continue operations in the old local markets. They form expectations about market demand, cost levels and competition with other firms. These expectations are fulfilled in the equilibrium. The number of LLU entrants in municipality i at time t is denoted by N it = n, where n = {0, 1, 2, 3, 4, 5+}. 24 Following Xiao and Orazem (2011) and Nardotto et al. (2015), in the presence of n-1 competitors, the discounted value of future profits of n th firm in market i at time t can be written as: π n it = α t lns it + X it β t µ n I(N it = n) + ɛ it π n it + ɛ it (1) where S it is the potential market size approximated by the number of households, X it is a vector of other demographic and geographic characteristics of municipalities which are potential profit determinants, such as the population density, the share of flats in the total number of premises, income, the unemployment rate and the number of firms in the local market, µ nt denotes the negative effect on profits from the presence n th firm. We also include a set of regional dummy variables to account for any other market specific effects which may play a role. Since the LLU regulation in France has been in place since 2000 and LLU entry in general took place before FttH entry and before upgrading cable to high speed broadband by Numericable, we do not consider that LLU entry is influenced by competition with these technologies. The reduced form profit specification given by equation (1) does not allow to distinguish how the number of competitors affects the variable profits function and the fixed costs, as in Bresnahan and Reiss (1991). The model also does not allow for simultaneous entry and exit. Instead, it is focused on the expected profitability from operating in the next period of the marginal entering or exiting firm. The profits πit n include the non-sunk part of fixed costs. Firms may also incur a sunk cost SC to enter a market, which cannot be recouped when they exit. We assume that all entrants incur the same entry costs regardless of the order of entry. Profits are 24 Since there was a small number of markets with more than five entries, we truncated the number of entries with five. 14

15 unobserved with πit n being a latent variable. We draw inferences on the profit determinants by assuming free entry equilibrium, where firms enter only if such decision is profitable. There are three situations in which we may observe that at time t in market i there are n active firms. In the first case, in period t 1 there were fewer than n firms and in period t one or more firms have entered, so that N it > N it 1. In such case, for the n th firm, the expected discounted value of entry must exceed the sunk cost of entry, but not for the n + 1 st firm, which can be expressed as: Case 1, net entry: N it > N it 1 if π n it SC and π n+1 it < SC (2) In the second case, no firm has entered or exited a market in which there were n firms, so that N it = N it 1. In such case, in period t 1 the n th firm has decided to stay in the market because its expected discounted value of continuation exceeds 0, while for the n + 1 st the expected discounted value of entry must not exceed the sunk cost of entry in period t: Case 2, inaction: N it = N it 1 if π n it 0 and π n+1 it < SC (3) In the third case, there were more than n firms at time t 1 and one or more firms have exited the market in period t, so that N it < N it 1. In such case, the market becomes unprofitable when more than n firms keep operating. The marginal n + 1 st firm expected that it would be unprofitable to remain in the market and when this firm left, the n th firm expected otherwise: Case 3, net exit: N it < N it 1 if π n it 0 and π n+1 it < 0 (4) Using the profit specification (1), the above inequalities can be combined to obtain the following probability of observing N it = n entrants in market i at time t: P r(n it = n) = Φ(π n it SC I + it ) Φ(πn+1 it SC (I + it + I0 it)) (5) where Φ(.) denotes the cumulative normal distribution function, and I + it I(N it > N it 1 ) 15

16 are indicator variables to denote whether entry increased (+) or remained constant (0). The parameter vector θ = [α t, β t, µ n ] is estimated by maximizing the following log-likelihood function: LL(θ) = M T N y it ln(p r(n it = n) θ) (6) i=1 t=1 n=1 where y it takes value of 1 if N it = n and zero otherwise. Since there was relatively small number of FttH entries and no exists at all, we are not able to identify sunk costs in the case of FttH entry. Without sunk costs, SC = 0, the model for FttH entry is static and reduces to standard ordered logit or probit. The discounted value of future profits for FttH entrants may be written in analogous way to (1), but in this case n = {0, 1, 2, 3}. In the estimation, we use the same set of demographic and geographic characteristics as determinants of profits of LLU entrants. Importantly, the profits of FttH operators are also affected by competition with other technologies. First, we create a variable which represents the number of local entrants via LLU, excluding SFR and Free operators. In the data, we observe that SFR and Free deploy fiber only in markets in which they also provide broadband services via LLU. Therefore, the data would tell us that there is a positive effect of entry via LLU on FttH deployment by the same firm. 25 Instead, we analyze whether the number of other LLU operators has positive or negative impact on FttH entry. For instance, a negative effect may be due to lower incentives of Orange to invest in areas in which it has to provide LLU access to many firms, from which it receives a revenue on the wholesale market. At the same time, SFR and Free may also have lower incentives to invest in areas in which there are many LLU competitors. On the contrary, a positive effect may be due to incentives for vertical differentiation by providing higher quality services. In the analysis, we consider that the number of LLU entrants is exogenous because entry via LLU by SFR, Free and other firms took place before the deployment of fiber. A situation in which LLU entry does not take place because fiber is being deployed in the local market is unlikely. Apart from competition with ADSL technology, the incentives to deploy fiber may be influ- 25 The effect of LLU entry by SFR and Free may be negative on fiber deployment by Orange but in this model we can not separate this. 16

17 enced by availability of VDSL technology which offers higher speed than ADSL. Since consumers having access to VDSL may be less willing to switch to FttH, firms may be also less willing to deploy fiber in municipalities in which VDSL is available. We have information on the share of VDSL lines in municipalities and create a dummy variable which takes value one for observations with nonzero shares. Importantly, the deployment of VDSL started only in We also treat this variable as exogenous, i.e, as independent on the deployment of fiber, even though the firms which deploy fiber may decide to upgrade ADSL lines instead of investing in FttH. Finally, the presence of cable network may influence the decision to deploy fiber. On the one hand upgraded cable offers comparable speed levels and may discourage FttH entry since current cable subscribers will be less willing to switch to FttH. On the other hand, firms may deploy fiber to secure transition of consumers from xdsl to FttH, who otherwise could be lost to cable. In most municipalities cable was deployed before fiber deployment took place. Since, as shown in Table 3, Orange is the market leader in FttH deployment, we also estimate a model in which the 0-1 entry decision by Orange is regressed on the same set of explanatory variables. In principle, Orange is the first mover and there are only few municipalities in which there is one fiber operator which is not Orange. We analyze whether the deployment of fiber by Orange is discouraged in areas in which: (i) there are many LLU entrants; (ii) there are ADSL lines upgraded to VDSL; (iii) there is present cable operator Numericable. 6 Estimation Results Table (7) shows the estimation results for the model of LLU entry. Model I is estimated without sunk costs (ordered logit model) and Model II allows for non-zero sunk costs. Based on the value of log-likelihood function, the preferred model is the model with sunk costs. The variables which significantly determine LLU entry are the market size and the density of population, which is measured by the number of households divided by the surface. The interaction terms of density with time dummy variables are significant but do not suggest a particular increasing or decreasing trend. At the same time, the coefficients of time dummies increase over time. The share of flats in the total number of residencies is not significant. Furthermore, we find that a higher level of income has a positive impact on LLU entry. Surprisingly, a higher unemployment 17

18 rate also has a positive effect. Table (6) shows the estimation results for the model of FttH entry. Model I is estimated without sunk costs. We were not able to estimate a model with sunk costs due to convergence issues, which may be caused by the particularities of FttH entry data. Model II is an alternative specification of Model I, in which Numericable is considered to be one of the fiber entrants. Due to a merger between SFR and Numericable in April 2014, we consider that in this year one of the firms exited the market in locations in which both of them operate. Model III is a 0-1 logit model for fiber entry by Orange, which is the market leader and in general enters first in local markets. We find that the local market presence of competing technologies has a significant impact on fiber deployment. In particular, a greater number of LLU entrants has a significant and positive impact on FttH entry. In municipalities with many LLU entrants, Orange, Free and SFR have more incentives to deploy fiber networks. We also observe in the data that fiber deployment by SFR and Free is always preceded by LLU entry. Thus, the data and estimation support the ladder of investment (LoI) theory suggesting that the main operators start providing broadband services in selected municipalities using incumbent s infrastructure via LLU, and then follow up with deploying FttH networks. Furthermore, all three fiber operators have incentives to differentiate their offer to provide consumers with a higher speed FttH access, in competitive xdsl markets. 26 We also find that FttH entry is more likely in areas in which the sole cable operator Numericable upgraded its network to provide high speed broadband services. According to industry experts, the marketing strategy of fiber operators to gain customers is more aggressive in municipalities which are covered by cable. This result suggests that fiber operators recognize that consumers care about the connection speed. Moreover, it suggests that being first in the market is critical to build a consumer base and recoup the investments. Firms deploy fiber to secure transition of their consumers from xdsl to FttH, who may be otherwise lost to cable. Moreover, upgrading DSL lines to VDSL technology, which started in 2013, has a negative impact on fiber deployment. The negative effect suggests that FttH and VDSL connections are substitutes. 26 We also estimated a model using lagged number of LLU entrants which yields similar results. 18

19 Thus, if fiber operators invest in VDSL technology they may have less incentives to deploy fiber networks. But also when LLU competitors upgrade to VDSL technology, fiber operators may have less incentives to enter with FttH. Unfortunately, we do not know the identity of firms which use VDSL technology to separate these two effects. The FttH entry is determined by the demographic and geographic characteristics of local markets. The main variable which influences entry is the market size measured by the number of households in the local market. Also, the population density, which is measured by the number of households divided by the surface, has a significant positive impact on FttH entry which decreases over time. This means that FttH is more profitable in densely populated areas, but over time it becomes easier to enter areas with lower density of population. The share of flats in the total number of residencies has a positive impact on FttH entry which may reflect lower cost of deploying fiber in areas with many apartment blocks. Furthermore, we find that a higher level of income has a positive impact on FttH entry, while unemployment is insignificant. These two variables approximate the quality of demand in terms of purchasing power. The estimation results of logit model for Orange s decision to deploy FttH confirm the results from our FttH entry model. There is a significant and positive impact of the number of LLU entrants on entry via fiber, which indicates that Orange tries to differentiate the offer from DSL-based competitors. The presence of Numericable offering very high speed broadband via upgraded cable also has a positive impact on deployment of fiber. As discussed above, Orange may try to avoid being preempted by upgraded cable. Moreover, fiber deployment is slowed down by the process of upgrading xdsl lines to VDSL. The local market characteristics also have similar impact on Orange s decision to deploy fiber. We use the estimates to compute entry thresholds for LLU and FttH operators and their changes over time. For each market i at period t we calculate entry threshold for n = 1, 2, 3, 4, 5+ LLU operators by solving for the critical market size which sets the mean profits to zero: Sit n = exp((x it β t µ n t I(N it = n))/α t ). The thresholds are market specific and show large variation. The average entry thresholds for LLU are shown in Table (9). The average entry thresholds for FttH are shown in Table (8). The entry thresholds for LLU operators are much smaller than for fiber operators, which reflects the scale of investment required to enter the market. 19

20 Since LLU operators rely mainly on the incumbent s network the investments required for entry are substantially lower. We found that fiber entry is positively influenced by the number of LLU entrants. We conduct counterfactual simulation and compute FttH entry thresholds in the absence of LLU access regulation. The average entry thresholds increase by about XX%, as shown in Table 8. 7 Conclusion We analyze the impact of LLU on fiber investments using panel data on 36,080 municipalities in metropolitan France for the years We estimate two models of entry into local markets by: (i) the incumbent and two other operators using fiber technology and (ii) operators seeking wholesale access to the legacy telecommunications network via local loop unbundling (LLU). We also estimate the determinants of incumbent s decision to deploy fiber. We find that local market presence of competing LLU operators has a positive impact on entry by fiber operators. High speed fiber broadband allows firms to differentiate their offer from DSL-based services. We also observe in the data that fiber deployment by SFR and Free is always preceded by entry via LLU. Thus, our estimates overall support the Ladder of Investment hypothesis. Both the incumbent and entrants have incentives to deploy fiber technology in the presence of LLU access regulation. Moreover, the presence of upgraded cable has a positive impact on fiber deployment. Firms using DSL technology fear being preempted by the cable operator and losing consumers, which encourages them to invest in fiber in areas covered by cable. This result suggests that firms recognize that consumers care about connection speed. It also implies that being first in the market is critical to build consumer base and recoup investments. In addition, we find that the deployment of VDSL technology, which provides higher Internet speed on the copper network, slows down fiber deployment. Thus, firms choose to upgrade copper lines on the legacy network instead of investing in fiber networks. But also when LLU competitors upgrade to VDSL technology, fiber operators may have less incentives to enter with FttH. Unfortunately, we do not know the identity of firms which use VDSL technology to separate these two effects. Our results have important policy implications as they show that LLU regulation does not 20