4 Frequency allocation: a comparison with other countries

4.1 Introduction

The amount of spectrum held by overseas operators may provide a useful comparison with the situation in New Zealand. We have undertaken some comparative analysis of a sample of operators, firstly collecting information on the quantity of spectrum held by the operators, and secondly undertaking a statistical analysis to identify if there are any factors which may have a relationship with the amount of spectrum held by the operators.

While we have selected a sample, the members of which have various points of similarity with Telecom and Vodafone New Zealand (described in Section 4.2), key differences still exist. There may be certain characteristics which explain why one operator may appear to have only a small amount of spectrum, or another operator a particularly generous allocation. The purpose of the statistical analysis is to adjust the sample data for these significant factors. This then allows us to compare the allocation of spectrum for Telecom and Vodafone New Zealand with that of the other operators within our sample.

4.2 Selecting a sample of operators

Mobile networks reflect the characteristics of the local environment. These characteristics include:

• population distribution and density
• mix of urban, suburban and rural areas
• coverage area
• terrain
• traffic levels
• the amount of spectrum available.

Every country is unique, which makes direct comparisons difficult, nevertheless it is still possible to use information from other countries to draw conclusions about the use of spectrum in New Zealand.

We have collected information across several countries which have operators of similar size in terms of subscriber numbers to Telecom and Vodafone New Zealand:

• Denmark
• Finland
• Hong Kong
• Ireland
• the Netherlands
• Norway.

While most of these countries have a number of points of similarity with New Zealand, in terms of demographic or geographic characteristics, Hong Kong represents a very different environment. It is one of the most competitive mobile markets in the world. With five operators (and 14 different networks) serving a population of just under 7 million within a small area characterised as an extremely high density urban environment, it offers great challenges in radiofrequency planning. Hong Kong thus provides an interesting illustration of an extreme case: namely, what can be achieved with a given amount of spectrum across multiple operators with high traffic densities.

In addition to all the mobile operators within the above countries, we have supplemented our sample with a small selection of operators from Canada, United States, Japan, United Kingdom, Germany, Spain and Hungary. This enabled the inclusion of several CDMA operators, as well as introducing more variation within the sample in terms of operator and country characteristics. AsTop

Exhibit 4.1 shows, in terms of subscriber base both Telecom and Vodafone New Zealand are positioned at around the midpoint of our sample of operators.

Exhibit 4.1: Mobile subscribers by operator, 2005 [Source: regulators, operators]

All of the countries within our sample have extensive mobile networks, with coverage in excess of 94% of the population, although there are individual operators in some countries which do not offer national coverage, or extend their coverage via roaming agreements with other operators.

Most of the New Zealand population live in non-rural areas, with the United Kingdom, Denmark, the Netherlands and Norway also displaying similar urbanisation levels (Exhibit 4.2). The urbanisation levels in the various spectrum licence areas for the United States and Canada can vary from national data. For example in Aliant’s licence area – encompassing the Canadian provinces of Newfoundland and Labrador, Prince Edward Island, Nova Scotia and New Brunswick – only 54% of the population live in non-rural areas.Top

Exhibit 4.2: Proportion of the population living in non-rural areas for selected countries [Source: World Bank]

In terms of land area, New Zealand is just below the midpoint of the countries within our sample (Exhibit 4.3 – this graph excludes the United States and Canada as the resultant scale would mask variation in the smaller countries). Note that our analysis uses land area as a proxy for coverage area, and in the cases of the regional US and Canadian operators we have used the spectrum licence areas rather than the national land area. There is little information available on the geographic coverage area for mobile operators – most report coverage only in terms of percentage of population covered. Coverage is either low or nonexistent in unpopulated areas.

Exhibit 4.3: Land area of selected countries [Source: World Bank, Hong Kong C&SD]

New Zealand has a similar population density to both Finland and Norway (Exhibit 4.4 – note that Hong Kong was omitted due to its effect on the scaling for this graph). However, caution should be used if inferring characteristics of mobile traffic density (for the purposes of network dimensioning) from a national figure. All countries will have areas of high density population (typically in the central business districts of major urban centres), areas of medium population density (such as in suburbs) and low density areas (rural regions), which means that subscribers, and the traffic, will be spread unevenly over the entire coverage area. Furthermore, for regional operators population density may also differ from the national figure.Top

Exhibit 4.4: Population density for selected countries [Source: Network Strategies]

Most of the countries in our sample have mature mobile markets, with more than 90 subscriptions per 100 persons, with the exception of Canada, the United States and Japan (Exhibit 4.5). In such markets, the opportunity for subscriber growth becomes more limited, as this translates into a proportion of the population taking up more than one service. Indeed, penetration has already exceeded 100% in Denmark, the United Kingdom, Ireland, Norway and Hong Kong.

Exhibit 4.5: Mobile penetration in selected countries, 2005 [Source: Network Strategies]

The traffic levels experienced in New Zealand are substantially lower than for many of the other operators in our sample. There is a relationship between the number of subscribers and the traffic volumes (Exhibit 4.6), nevertheless there is some variability of average minutes of use per subscriber between operators.Top

Exhibit 4.6: The relationship between subscribers and annual outgoing traffic volumes [Source: operators, Network Strategies]

If we exclude the larger operators from the above graph (Exhibit 4.7), we see that both Telecom and Vodafone New Zealand have a relatively low level of traffic given the number of subscribers. Only the Dutch operator Telfort has less traffic, but this is due to the subscriber base being predominantly low-usage prepaid customers. A discussion of the reasons behind low traffic levels in New Zealand is beyond the scope of this project – it is a complex topic and would be an avenue for further research.

Exhibit 4.7: The relationship between subscribers and annual outgoing traffic volumes for smaller operators [Source: operators, Network Strategies]

4.3 How much spectrum do the operators in our sample hold?

The amount of 800/900MHz spectrum held by both Telecom and Vodafone New Zealand falls at the upper end of the allocations of the other operators within our sample (Exhibit 4.8). In the table below, row shading is used to denote those operators that use CDMA technology.

Exhibit 4.8: Summary of spectrum holdings of licensees with 800/900MHz spectrum, ranked by spectrum amount [Source: regulators, operators]

4.4 Comparative analysis of the sample dataTop

We have seen that both Telecom and Vodafone New Zealand have a relatively generous quantity of spectrum in the 800/900MHz bands (Exhibit 4.8), in comparison with the other operators within our sample, however such a simple comparison does not take into account any factors that may vary from operator to operator, and which may justify the amount of spectrum held.

So, the next step in the comparative analysis is to try to identify if there are any factors which have a significant relationship with the amount of spectrum, and then develop a model which may adjust the sample data for these factors. Such a model can then be used to estimate, based on the sample data, how much spectrum Telecom and Vodafone New Zealand would be expected to hold.

A statistical analysis of our sample data, using multiple regression, shows that there are four significant factors which have a relationship with the amount of spectrum held by the operator37:

FAdjusted R² = 0.419
F_4,12 = 3.890
Durbin-Watson statistic = 1.78

where:

• Spectrum is the amount of paired spectrum in the 800/900MHz bands, expressed in MHz
• Traffic is the outgoing traffic, in millions of minutes
• PopDensity is the population density, in persons per square kilometre
• Area is the land area, in square kilometres
• CDMA is a dummy variable which has the value of one if the operator uses CDMA or zero otherwise.

Clearly, our model only goes part way towards explaining the variability within the sample data. This is indicated by the relatively low adjusted R2 value, and the lack of statistical significance of the predictor variables (indicated by the t statistics). It is not surprising that we have not captured all the sources of variability within the data – the frequency allocation process varies from country to country, and certainly in some jurisdictions, such as Japan, there has been criticism that the process lacks transparency38 . Thus it would be difficult to capture in such a model all the key drivers influencing the various spectrum management agencies.

Nevertheless, the model can still be used as an indication of how the spectrum amounts in New Zealand compare with those in other countries. It should be noted that this model is based on a reduced data sample of 17 operators:

• In order to estimate the spectrum amount for Telecom and Vodafone New Zealand based on the characteristics of the overseas operators, data for the New Zealand operators needed to be excluded from the multiple regression – inclusion of the New Zealand data would bias the result.
• Outgoing traffic data was not available for all the operators within our sample, and for a number of operators outgoing traffic was estimated from the traffic data that was reported (either total incoming and outgoing traffic, or average minutes per subscriber). A number of operators were excluded from the analysis, as we were unable to obtain reliable current traffic information within the timeframe for this project.
• A small number of operators were removed from the sample due to being outliers, or extreme observations. The licence areas for TELUS and ALLTEL are extremely large in comparison with all other operators (1.6 and 2.6 million square kilometres respectively); the statistical analysis found that Sonera and Elisa were also outliers. It is standard practice to remove extreme observations in order to improve the fit of the statistical model.

One factor we did not examine – due to limited time – was the presence of any relationship between the spectrum allocation in the 800/900MHz bands and the amount of spectrum held in higher frequency bands.

It should be emphasised that this regression model does not estimate ‘optimal’ spectrum – it is unable to provide any information on whether the operators in our sample have an optimal allocation. It cannot be used to estimate the appropriate amount of spectrum for New Zealand operators – a far more complex task than can be captured in such a simple model, as described in Section 2 – nor can it be used to derive any conclusions regarding causality – for example are the traffic levels driving the amount of spectrum, or are operators seeking to maximise spectrum utilisation and are thus encouraging high traffic volumes?

The model does however suggest that in comparison to the operators within our sample, both Telecom and Vodafone New Zealand have a relatively generous spectrum allocation within the 800/900MHz bands, given adjustments for various factors that differ between New Zealand and the other countries. Applying New Zealand data to the regression model results in a estimated allocation of 2 × 9.4MHz for Telecom and 2 × 10.7MHz for Vodafone, which is less than the actual holdings of the operators. This confirms our finding based on the simple comparison of Exhibit 4.8.