Dynamic Entry and Investment in New Infrastructures: Empirical Evidence from the Fixed Broadband Industry

Service-based competition takes place when entrants rent access to the incumbents’ networks (via local loop unbundling, for example), whereas there is facility-based competition when the entrants build their own infrastructures to provide services to end customers. See Sect. 2 for more details on the industry background.

See, for instance, Crandall et al. (2004).

See Bourreau et al. (2010) for a critical review of the ladder-of-investment approach. See also Bourreau and Drouard (2012), who study the effect of a phase of experience acquisition on an entrant’s investment incentives.

Note that the ladder-of-investment approach is very similar to the stepping-stone hypothesis that has been hotly debated in the United States (e.g., see Rosston and Noll 2002).

For example, according to the French telecommunications regulatory authority (Autorité de régulation des communications électroniques et des postes, ARCEP) “the development of competition in France since 1998 is a good illustration of the theory of the ladder of investment” (see ARCEP 2007, p. 36). The European Regulatory Group (ERG) also argues that there has been a positive relationship between the implementation of the ladder of investment approach and the pace of development of the broadband market: “[the ladder of investment] explains recent developments in European broadband markets quite well and can serve as a good regulatory model” (ERG 2005, p. 1). As a final example, the European Commission (EC) cited the ladder approach in a decision on squeeze tests conducted by the Italian telecommunications regulatory authority (Autorità per le Garanzie nelle Comunicazioni, AGCOM): “AGCOM’s approach [...] may fail to take sufficient account of alternative operators’ financial leeway to climb up the ladder of investment throughout the national territory” (Commission decision concerning Case IT/2010/1103: Margin squeeze test guidelines, 6 August 2010).

Here, the “incumbents” refer to the historical fixed-line operators or former monopolies (e.g., France Telecom in France, or Telecom Italia in Italy), and therefore, there is only one incumbent operator in each European country. The “entrants” are the incumbent’s competitors.

See: Regulation (EC) No 2887/2000 of the European Parliament and of the Council of 18 December 2000 on unbundled access to the local loop, Official Journal of the European Communities, December 2000, the 30th, available at http://​eur-lex.​europa.​eu/​LexUriServ/​LexUriServ.​do?​uri=​OJ:​L:​2000:​336:​0004:​0008:​EN:​PDF.

In this vein, Distaso et al. (2006) provide empirical evidence that facility-based competition has been the main driver of broadband adoption in 14 European countries, from 2000 to 2004. Similarly, Nardotto et al. (2012) demonstrate that local loop unbundling had no positive impact on broadband adoption in the U.K. between 2005 and 2009, in contrast to facility-based competition from cable operators. However, a different conclusion is reached by Gruber and Koutroumpis (2013), who study a large dataset of 167 broadband markets over 11 years, and find that facility-based competition has delayed broadband adoption. Höffler (2007) also shows that for the period 2000–2004 in western Europe, the costs that were associated with the inefficient duplication of existing infrastructures outweighed the benefits of increased broadband penetration.

In addition to these technologies, one could include mobile broadband technologies, such as 3G. However, as our focus is on the fixed broadband market, we do not take into account mobile broadband technologies.

An entrant can use any technology when it invests in a network from the outset, without climbing the investment ladder.

Several elements in the architectures of DSL and cable DOCSIS (Data Over Cable Service Interface Specification) networks cause these two broadband access technologies to be incompatible. For example, a DSL access network is based on ATM (Asynchronous Transfer Mode) transmission standards at the layer 2 in the OSI model, whereas a cable network uses Ethernet or IP-based forwarding for connections between the network interface of cable modem termination systems and wider area networks. By contrast, fibre access networks, for instance, are based on ATM transmission standards, and, hence, are compatible with the architecture of DSL networks. Finally, note that the assumption that we make (i.e., to ignore cable as an investment option on the ladder) is in line with the positions of regulators: When they have set up a ladder-of-investment, they never argue that cable is an option for the entrants that climb that ladder.

See Valletti (2003), Guthrie (2006), and Cambini and Jiang (2009) for a survey.

See Cambini and Jiang (2009) for a survey of this empirical literature.

See Arrow (1962). A monopolist has less incentive to innovate than does a new entrant, because the former replaces “itself”.

WLL stands for Wireless Local Loop, and PLC stands for Power Line Communications. As we explained in Sect. 2, we exclude cable lines and satellite connections.

We tested our model with additional lags for \(\textit{LLU}lines\), and obtained similar results.

We also do not have any information on the number of entrants per country.

In Sect. 4, we will introduce additional variables to conduct some robustness checks.

As the variables that represent a number of subscribers (i.e., \({ Newlines}\), \(\textit{LLU}lines\), and \(BAlines\)) can take values of zero, \(\log ({ Newlines})\) is actually computed as \(\log (Newlines+1)\).

Note that the British incumbent operator did not own any mobile operations for the period under study. Similarly, the Irish incumbent sold its mobile operations in 2001 and re-entered the mobile market in 2005 through an acquisition.

We also tried to use other variables to control for market power, and in particular the Hirschmann–Herfindahl index (HHI) for the broadband market. Our results remained unchanged. Note however that the HHI for the broadband market could be an endogenous variable.

See Bourreau and Doğan (2006) for a theoretical argument along this line.

Source for the GDP per capita: OECD. For population: OECD (except for France, source: INSEE; see the footnote on Table 4). For mobile penetration: ICT Eye—ITU.

Source: OECD Factbook 2009 for population (except for France, source: INSEE), and Eurostat for the size of the country (except for France, source: CIA World Factbook).

Due to lack of appropriate data, we cannot control directly for the cost of fibre. However, we believe that our analysis does not suffer from a huge omitted variable problem. Indeed, the cost of a fibre infrastructure in a given country typically depends on the cost of fibre equipment and on factors that determine infrastructure costs in that country. The cost of fibre equipment (fibre cables, optical routers, etc.) is roughly the same in all European countries, and therefore is captured through the time fixed effects in our estimations. As for factors that affect infrastructure costs, we control for population density in each country (which is often used as a proxy for infrastructure costs). Finally, note that the cost of a fibre network is not huge, relative to the cost of a mobile network. For example, in France, it would cost between 5 and 10 billion Euros to build a national mobile network, and 8–10 billion Euros to cover 60–70 % of France with fibre (Datar 2010; Maurey 2010).

The specific patterns that we observe for Denmark, the Netherlands, and Portugal can be explained by some industry facts that are unrelated to the ladder-of-investment. First, we observe a fast decrease in the number of LLU lines in Denmark (in 2009) and in the Netherlands (in 2007), followed by a growth in LLU lines. In both countries, the decrease in the number of LLU lines is explained by the acquisition of an entrant by the incumbent operator. Second, in Portugal, we observe an important increase in cable lines followed by a decrease in LLU lines and a take-up in new access lines. This is explained by the fact that, in 2007, the incumbent, which owned at that time a local copper network and a cable network, had to sell its entire cable activity to competitors. This justifies the important increase in cable lines that are controlled by entrants. At that time, therefore, facility-based competition between cable and the incumbent’s copper network suddenly emerged. As a response, LLU entrants either exited the market or started to replace their LLU-based offers by investing in new lines in order to remain competitive.

In Sect. 4, we will take cable into account as a robustness check.

Moreover, in Ireland, the growth in \(\textit{LLU}lines\) starts after the growth in \({ Newlines}\) has already begun. This is not consistent with the ladder-of-investment hypothesis, which claims that the development of LLU fosters investment in \({ Newlines}\).

In the difference-GMM, the instrumental variables are therefore (i) the number of \(BAlines\), (ii) all lags of \({ Newlines}\) (starting from the 1-period lag), and (iii) all lags of \(\textit{LLU}lines\) (starting from the 2-period lag).

We also ran OLS and IV regressions (the estimation results are available upon request from the authors), and performed a Hausman test to compare the estimates and to determine whether the difference in coefficients was statistically significant. In a first stage, we regressed \(\log (LLUlines)\) on all exogenous variables, including the lagged values of \(\log (BAlines)\), and predicted the residuals. In a second step, we regressed \(\log ({ Newlines})\) on \(\log (LLUlines)\), other exogenous variables (excluding \(\log (BAlines)\)), and the residuals of the first-stage estimation. The Hausman test is equivalent to a \(t\) test on the coefficient of the residuals. The coefficient of the residuals was equal to \(-\)0.546 and the \(t\) test was equal to \(-\)2.75; hence, we rejected the hypothesis at the 1 % level. We can conclude that \(\log (LLUlines)\) is not exogenous.

As a robustness check, we considered an alternative and more extensive instrument, which is the sum of service-based lines that were opened through bitstream access and resale. Indeed, it could be argued that new entrants could earn some market experience through resale, and then climb directly to the LLU rung. When we use this more extensive instrument, our results are not affected. However, we consider that the number of bitstream access lines represents the best instrument for the number of unbundled lines, as we follow the rung-by-rung approach of the ladder-of-investment.

Telecommunications operators confirmed to us that this was the usual investment pattern that is observed in Europe. We also found some empirical evidence that this is indeed the pattern in France. In this country, the main entrant investing in fibre (excluding the cable network operator) is the company SFR. From SFR’s website, we found the cities where SFR has invested in fibre (as of November, 2011). For all of these cities, we then checked whether SFR was using local loop unbundling, or bitstream access. We found that it was using LLU in all of them and BA in none of them. In other words, we do not observe any switch from bitstream access to fibre for SFR, whereas SFR uses bitstream access (and not LLU) in a significant proportion of French cities.

We use \(\log (Pop)\) as a control variable, but we do not address the question of the partial effect of population on new lines. The correlation between some of the explanatory variables could lead to a high variance in the coefficient of \(\log (Pop)\). Some of the independent variables (e.g., \(\log (Pop)\) and \(\log (\textit{GPD}percapita)\)) are highly correlated; hence one cannot interpret the partial effect of each of them, though they are collectively significant.

However, with the development of smartphones, it might be that factors that influence the penetration of mobile phones also affect the demand for broadband services. As a robustness check, we therefore ran our estimations for the period before the launch of the first iPhone in Europe (2002–2007). Our qualitative results were not modified.

We could also consider a four-rung ladder that is composed of resale, bitstream access, local loop unbundling, and new lines. However, adding resale lines does not modify our results. Besides, the number of bitstream access lines is not determined by the number of resale lines, since new entrants can enter directly at the bitstream access rung.

The system GMM estimates give a non-significant coefficient for \(\log (BAlines).\) The system GMM estimator makes an additional assumption that can be tested using a difference-in-Hansen test of exogeneity. The difference-in-Hansen test checks the over-identification of the additional instruments of the level equation. This test equals one in the case of too many instruments. When we try to reduce the number of instruments (by reducing lags, or by using the collapse option in Stata), the difference-in-Hansen test gives a small p-value. Hence, we reject the hypothesis that the additional subset of instruments that are used in the system GMM estimation are exogenous. Thus, we keep the difference GMM estimation.

The institutional context may explain this result: The European Regulation 2887/2000 on unbundled access to the local loop imposed mandatory LLU for all member states. Hence, once a European country has transposed the European regulation, we expect no specific time effect: The time path of the number of unbundled lines in each country has no reason to vary significantly.

For example, ERG (2005) writes that “the complementary use of several access products may mean that both forms of access should be made available over a longer period”.

We also performed our analysis with other threshold values for each type of access, ranging from 5 to 15 % of DSL lines. The qualitative results remained the same.

Cardona et al. (2009) provide some empirical evidence that DSL and cable networks are indeed part of the same market.

We performed the same test using a mean split, and obtained similar results.

We also ran an estimation using the cable market share as a level variable. This did not qualitatively affect our results.

We built the Polynomics index from the Polynomics Regulation Index 2012 Dataset, following the tutorial by Zenhäusen et al. 2012). As is true for all regulatory indexes, the Polynomics index has weaknesses; and in particular, it captures only formal aspects of regulation. Thus, it should be interpreted with some caution. However, a previous version of this index (the Plaut Economics index; see Zenhäusen et al. 2007) has been used as a measure of regulatory intensity in other studies (for example, by Grajek and Röller 2012), and we therefore use it for our robustness check.

The estimation results are reproduced in Table 9. Note that each period \(t\) still represents a semester, but since the Polynomics index is available on an annual basis only, our database for the Polynomics regression is on an annual basis too. Therefore, for example, we lag the \({ Newlines}\) variable 2 periods rather than 1 period.

European Commission Implementation Reports No. 8, 9, 10, 12, 14, and 15.

We also tested a simplified specification by using a dummy variable that equals one if the LLU monthly fee increases. However, prices could decrease not only because the regulator applies the ladder-of-investment approach, but also for other macroeconomic reasons. Using the shift in prices in a specific country relative to the average allows us to take into account those unobserved factors.

The estimation results for the additional robustness checks presented in Sect. 4.5 are available upon request.