Introduction

Australian regulation of fixed broadband networks was broadly consistent with other developed economies until 2009, when NBN Co was established as a government-owned monopoly. Australian consumers are currently less satisfied with Internet services than with mobile phone or fixed line telephone services, and are particularly dissatisfied with Internet data speeds, service costs and the speed of fault repairs (ACMA_2017, pp. 60-61). Successive government policies are regularly criticised as contributing to poor Australian outcomes, but how much impact does policy actually have, compared to the impact of other economic and demographic factors

This article explores this question through an international comparison. It reviews Australian policy across the era of broadband Internet, and then reviews research about the impact of such regulation internationally. Models of broadband pricing, penetration and quality are formed by applying this research to recent data from developed economies. These models are used to explore Australian broadband outcomes, and to trace the effects of regulatory policies and decisions. Implications for future broadband policy are also discussed.

History of Australian Broadband Regulation

Ownership separation, competition and intermediate compromises have been features of Australia's telecommunications regulatory policy for decades. Ownership separation refers to detaching the ownership of monopoly infrastructure from the ownership of firms using that infrastructure to deliver services (Baldwin, Cave & Lodge, 2011, p. 467). Ownership separation, referred to as structural separation, is the solution currently adopted in Australia (Telecommunications Act, 1997). It was adopted after long experimentation with other remedies for market failure, including competition and functional separation.

In the late twentieth century, competition, initially in long distance and international calls, was preferred to monopoly regulation of Australian telecommunications. Such competition would encourage the expansion of Australia's telecommunications infrastructure and industry (Brown, 1990). Enabling this competition initially required regulated access to Telecom Australia's (later Telstra's) local network, and AUSTEL was established as an industry regulator (Telecommunications Act, 1991). Beginning in 1997, the need for public oversight increased as Telstra was privatised in three stages (Telstra (Dilution of Public Ownership) Act, 1996). Over time, competition in telephony expanded to include infrastructure, when Optus and Telstra constructed rival hybrid fibre coaxial (HFC) networks, carrying pay TV alongside telephone calls. And as networks were expanding, new applications were emerging, notably the Internet (Richardson, 1997).

Internet access, especially email and the World Wide Web, emerged as a significant third use for telecommunications networks in the 1990s. Dialup Internet protocol (IP) services could be delivered to households over the existing (copper or HFC) networks, with new equipment required only on the customers' and the Internet service providers' (ISP) premises. Dialup Internet services had limited bandwidth and could not be used concurrently with telephone calls. HFC networks offered superior Internet access and enabled the "triple play" of telephone, pay TV and Internet services to utilize and repay the investment in HFC infrastructure. But accessibility and low cost meant dialup remained the most common mode of Australian household Internet access until 2006 (Figure 1). For households without HFC access, the alternative to dialup was a digital subscriber line (DSL) service delivered over existing copper telephone lines. DSL had at least 17 times the bandwidth of dialup services, up to 1500 kbps, and bandwidth increased further as technology improved. Increasing bandwidth over the copper network enabled Internet services such as video streaming, multiplayer computer games and file sharing, without significant new network construction.

Without overbuilding, entrepreneurial ISPs could only resell services delivered over Telstra's monopoly copper network. They could not easily compete with Telstra's and Optus's vertically integrated products, and captured little DSL market share. Figure 1 shows the initially high but rapidly falling retail market share (as percentage of total subscriptions) of dialup access from 2005. Telstra and Optus combined had 90% of retail broadband (DSL and HFC) subscribers in 2007. (In accordance with the Australian Bureau of Statistics, broadband here is defined as an 'always on' internet connection with an access speed of 256kbps or higher (ABS 2017).) iiNet was the most successful start-up with only 6% of subscribers. Competition was not thriving.

Figure 1 Subscriber concentration in the Australian residential Internet market. Company and Other broadband shares are given for broadband subscriptions according to the ABS definition, dialup subscriptions are consolidated.  (Source: Author's calculation from ABS data and companies public reports and investor presentation)

Australian law gives the Australian Competition and Consumer Commission (ACCC) authority to declare certain telecommunications services, that is, to establish regulated access to declared services (Telecommunications Act, 1997, Clause 350A). Telstra's local line monopoly concerned the ACCC, and various Telstra operations were declared in 1997, which allowed other firms to access Telstra services on a wholesale basis at set prices (ACCC, 1997). Sharing of the copper lines to premises (local loop unbundling or LLU) was initially declared in 2002 (ACCC, 2002), but the regulated prices proved unattractive to competitors, and direct sale and resale of Telstra services remained the dominant mode. The basis for LLU pricing was altered in 2007 (ACCC, 2007a), allowing rival ISPs to connect their own DSL equipment to Telstra's local copper lines for $14.30 instead of $30 per month (ACCC, 2008b). The residential market responded by increasing the market share of Optus and fringe ISPs at Telstra's expense. Between 2007 and 2010 Telstra's number of broadband customers fell by 6% while the total number of Australian fixed broadband subscriptions grew by 70% (Figure 1). The regulator's determinations were often vigorously contested by Telstra, and a plan to upgrade much of the local copper to a higher bandwidth fibre to the node (FTTN) network was halted in 2006 as the ACCC and Telstra failed to reach agreement on competitive access pricing (Burgess, 2006).

Declaring and regulating Telstra's infrastructure increased retail competition and allowed customers meaningful choice within the infrastructure duopoly. However, access to, and the quality of, Telstra-owned exchanges and copper remained problematic for rivals, and there was no apparent path to an upgrade of the fixed telecommunications network (iiNet Pty Ltd, et al., 2007). Furthermore, regulatory requirements discouraged Telstra investment. Telstra instead directed investment to more profitable and less heavily regulated operations associated with mobile telephone and data services (Telstra, 2010). In 2007, the G9 consortium of telecommunication firms (excluding Telstra) proposed building a FTTN network, structurally separated from the investing firms and from Telstra, but also failed to reach agreement with the ACCC on access pricing (ACCC, 2007b).

Network quality became a significant political issue in the 2007 Australian federal election, when the challenging Labor party promised to invest up to 4.7 billion dollars upgrading most of the copper network to FTTN (Farnsworth, 2007). But Telstra, whose cooperation was required to install and connect the new fibre to the existing network, refused to participate without guarantees regarding future regulation (McGauchie, 2008). The government was unwilling to offer such guarantees; Telstra was particularly concerned about forced separation, but was also concerned about future regulatory consistency.

In response, the Australian Government revised its policy in 2009 and established a government-owned company to build a primarily fibre to the premises (FTTP) network, at an estimated cost of 43 billion dollars. The policy initially envisaged the company (NBN Co) being financed with majority government equity plus some private funding, but private investors could not be found and the new company began operations as entirely government-owned. The intention to eventually privatise the company, to recoup the public investment, remained (Wong & Conroy, 2010, p. 12), requiring NBN Co to operate as a long-run profit-making firm.

The new FTTP network would replace Telstra's local copper lines with optic fibre, although it would rent access to Telstra's ducts and other passive physical infrastructure on a wholesale basis. The NBN would only sell wholesale access to retail ISPs, thus achieving structural separation at the local network level. Points of Interconnection (POI) between the NBN and the existing backhaul network were located only where competing backhaul services were already present, to preserve backhaul competition. Telstra agreed to work with NBN Co on access and transition arrangements, and other ISPs broadly welcomed the policy.

Difficulties emerged with the building of the FTTP network, and project delays and concerns about the eventual total cost were issues in the 2013 federal election. A change in government brought a focus on costs and speed of deployment, and in accordance with the recommendations of a cost-benefit study (Australian Government Department of Communications and the Arts Panel of Experts, 2014) the design of the network was modified to include a mix of technologies including FTTP, FTTN and the existing HFC lines. NBN Co accordingly purchased the HFC networks of Telstra and Optus, thus eliminating the fixed infrastructure competition to the FTTN/FTTP network. NBN Co agreed to progressively assume ownership of passive Telstra infrastructure as the NBN replaced Telstra's copper network and DSL broadband services.

Role of Regulation and Industry Concentration

The quantity of broadband units purchased is an important measure of the industry's economic contribution. In the early twenty-first century, many governments supported increased quantity by promoting affordable home broadband. For instance, a key objective of the eEurope initiative was to bring every home into the digital age and online (Liikanen, 2000). Many studies of these policies investigate how government and industry actions affected broadband penetration (the typical measure of quantity). A common finding is that broadband adoption is driven by competition between different technological platforms (inter-platform) and not, or less, by competition on a single platform (intra-platform) (Bouckaert, van Dijk, & Verboven, 2010; Cincera, Estache, & Dewulf, 2012; Distaso, Lupi, & Manenti, 2006). Cincera et al. (2012) also found that larger incumbent market shares are associated with slower broadband diffusion. There is some evidence of regulated local loop unbundling (LLU) having slightly positive effects, and other forms of access regulation having small, insignificant or negative effects. Cave (2014), whose work informed much regulatory practice, reviewed results from the period of DSL's technological dominance.

As fixed broadband penetration approached saturation in developed economies (see Figure 2) interest shifted from broadband to superfast next generation network penetration. Briglauer (2014) found strong regulation of DSL networks hinders adoption of next generation networks. He has argued that competitive firms invest to escape competition, and so refrain from investing if their competitors will share the rewards. Australia apparently witnessed this effect when neither Telstra nor the G9 consortium were able to agree with regulators on conditions to privately build a FTTN network. Australian regulators must have recognised the benefits of upgrading the network, but were concerned about the effect on consumer prices.

Figure 2. Fixed broadband penetration in OECD nations. (Source: Author's calculation from collated OECD data)

Costs of supply and consumer demand together drive broadband prices. But institutional factors, such as price controls, or market imperfections allowing firms to exploit their market power, may also affect prices. The effect of imperfect competition on price has been detected in particular markets. Correa and Crocioni (2012) used a range of broadband prices in Ireland and the Netherlands to detect which firms in those markets have market power, mitigated by three levels of access regulation. Fageda, Rubio-Campillo, and Termes-Rif (2014) studied prices in Spain, and found intra-platform competition affects price, while inter-platform competition does not reversing the common finding for broadband penetration. They also found service-level competition (bitstream access) is significant, results replicated across Europe by Calzada and Martnez-Santos (2014). Studying prices is typically complicated by problems of selecting which broadband price to study from the often large number of offers in the market, and by the need to take account of varying quality.

For the purposes of this article, I consider quality to include all aspects of a product that consumers are willing to pay a higher price for, except quantity. Here quantity is measured as the number of broadband subscriptions, so a contract with a higher download cap, for example, is considered higher quality, not greater quantity. Some aspects of broadband quality are difficult for consumers and analysts to assess (Nirmalathas & Lodders, 2016). Service reliability, data transmission speeds, fault resolution and bundled services all influence consumers product choice, but they are difficult to compare within a market, and even more difficult between markets. Cross-national data on customer satisfaction or complaints regarding home Internet services is not available, so is not considered here. This article accounts for download caps (which are also a direct proxy for advertised download speeds in the data set) in national prices, and for differences in broadband network quality internationally.

Empirical Data and Analysis

In this article I analyse data from the OECD Digital Economy Outlooks (OECD, 2015, 2017) supplemented by a range of other sources covering the OECD economies (Central Intelligence Agency, 2016; OECD, 2013, 2016; Transparency International, 2016; World Bank, 2016). Price data is from a single survey in 2014, so the analysis does not attempt to detect trends over time. It does use some 2016 data, without prices, to check model robustness. The particular value of this data is the inclusion of cross-national broadband prices for comparable bundles (bundles here being combinations of advertised speed and data download caps it excludes bundles of Internet services with telephone, pay TV or other services). The data is also relatively recent, coming from generally mature broadband markets serviced by DSL, fibre and HFC technology.

Using the survey's price data does introduce limitations. Broadband consumers typically do not simply select from a range of connection speeds or download caps at single prices. Rather, firms market broadband subscriptions bundled with complementary products, and at differing service levels. Australian firms commonly maintain multiple brands which support price discrimination, as in Telstra's Bigpond and Belong, and TPG's ongoing use of the iiNet and Internode brands from merged firms. Haucap, Heimeshoff, and Lange (2016) address this by testing price diversity as a factor influencing broadband penetration (finding diversity positively affects penetration) but this method is not transferable to the consideration of price as the dependent variable. This article does not attempt to explain price differences within markets, but does consider underlying price differences between markets.

Prices are supplied for 18 standardised bundles per nation, characterised as combinations of advertised speed and data download cap. Not every bundle is available in every nation, and the survey sometimes includes bundles with lower speed or download caps at the same price as bundles offering more. These lower offerings have been excluded from the data, so retained prices represent the most generous bundle offered at that price in that market. The OECD collected advertised price data from the incumbent telecommunications firm, and the largest cable firm, and one alternate firm (all subject to availability), for offers available in the nation's largest city. The survey reports only one price per bundle per nation. In theory, we expect negligible price differences between firms if we assume broadband bundles are interchangeable commodities but, as discussed above, prices in fact show significant variation. In this article, I assume the OECD's data collection method accurately reports incumbent pricing, and that international differences in incumbent pricing (corrected for the incumbents market share) reflect underlying price differences across markets.

I imputed data to fill occasional gaps in demographic data using methods recommended by Hair (2014), but did not impute missing market share data. Instead, where market share data was unavailable, I excluded the affected nations from parts of the analysis addressing market concentration. Dependent variable distributions from this subset are not substantially different to the whole sample.

I use Ordinary Least Squares (OLS) regression to estimate a control equation for broadband price, initially excluding regulatory and market share information. Regulatory and, separately, market share measures are then added to the specification to see if they make a statistically significant improvement to the control equations accuracy, i.e. its ability to estimate price. Given sufficient variation of these measures with the sample, finding significant effects would confirm a general capacity for broadband service retailers in more concentrated markets to maintain higher prices through coordinated conduct, and a capacity for access regulation to curtail the effects of coordinated conduct. Coordinated conduct involves firms implicitly or explicitly coordinating their pricing, output or related commercial decisions (ACCC, 2008a, p. 30) and may be more sustainable when the number of competitors is fewer. Conversely, the absence of effects would suggest that coordinated conduct is not prevalent, or has effects too small to detect. The analysis is repeated for broadband penetration, which is expected to decrease in situations of oligopoly pricing, and the same factors are checked for any contribution to network quality.

I test market share and concentration using several measures, including those used by regulators (ACCC, 2008a). I calculate the broadly used Herfindal-Hirschman Index (HHI) for retailers number of subscriptions, independent of connection type, and for connection types, independent of retailers. Other concentration indicators tested included the combined market share of the one, two, three and four largest firms, and Anbarci and Katzmans (2015) industry concentration index, classifying one, two and three firms as dominant. Results using other concentration indices are not substantially different to those using HHI and the incumbents market share, and are therefore not reported.

More importantly, the only drivers and indicators of market power considered are market share and concentration measures, regulatory settings, and the potential substitutability of wireless broadband for fixed. Barriers to entry other than network access, consumers countervailing power, the availability of information, and more broadly defined substitutes may all be relevant to market power within particular markets, but due to the difficulty of classifying and comparing these they are beyond the scope of this cross-national analysis.

Principles of parsimony and the modest sample size require including all available useful data while excluding variables which do not add significant explanatory power to the regression equations. I consolidate related factors to single indicators for national population concentration, public investment in telecommunications, national broadband speed, the technological quality of the network, the technological diversity of the network, and the market power of the five largest firms. Table 1 lists the raw factors and Table 2 the consolidated factors tested in the analysis. All factors are naturally numerical, except those representing regulatory options, which I code as dummies indicating presence (1) or absence (0). Three nonexclusive options corresponding to successive rungs on the ladder of investment (see OECD (2013) page 10 for a nontechnical description) are included, for low level bitstream access, mid-level unbundled local loop (LLU) access and high level direct (naked) access to the local loop. Mutual interaction terms are also generated for these dummies.

I generate OLS regression equations including all remaining independent variables and factors. I progressively eliminate the least significant variable, and repeat the regression estimation, until the returned Sum of Squares due to Error (SSE) stops decreasing with subsequent eliminations. This stepwise elimination of covariates can produce misleading predictions under some circumstances, so the selection of variables for the control equations was verified using Lasso estimation (Tibshirani, 1996). The confirmation of selected variables by Lasso estimation suggests the retained variables do indeed have consistent effects compared to those eliminated. I note one exception in the discussion of results.

A further robustness check uses the penetration equation generated from 2014 data to estimate penetration in 2016, and compares it to actual penetration. (An updated price survey is unavailable.) The 2014 equation predicts 2016 outcomes quite consistently, with the coefficient of determination (R-squared) only falling from 0.665 in 2014 to 0.577 in 2016.

Table 1 Test variables used in the OLS regressions

Table 2 Composite factors tested in OLS regressions