NS Report Peer Review

Peer review of report: Renewal of management rights for cellular services (800/900 MHz): What is the optimal spectrum quantity?

1.1. Overview of draft report

1.2. Assessment of benchmarking and spectrum regression model

In 2011 and 2012, licenses for spectrum currently utilised for cellular voice and data communications expire. The Ministry of Economic Development ("the Ministry") is to oversee the renewal or auctioning of rights to this spectrum.

The Ministry has received a draft report from their consultants Network Strategies ("the consultant")¹ in response to the Ministry's request for advice on:

• The optimal quantity of spectrum required by the current management right holders in the 800 MHz and 900 MHz bands over the period 2011-2031, and
• The optimal quantity of 800 or 900 MHz spectrum that a new entrant would require in the 800 MHz and 900 MHz bands to operate a nationwide cellular network over the period 2011-2031.

The Ministry indicated the optimal amount of spectrum under both recommendations should seek to maximise the value of spectrum to society as a whole. The recommendations should seek to balance the policy objectives of:

• Optimising incentives to invest for both current holders and new entrants;
• Promoting competition in telecommunications markets;
• Minimising risk of stranded investment by current holders;
• Minimising risk of supply discontinuity by current holders; and
• Any objectives the Ministry directs to be considered in the course of the Project. The Ministry also noted:
• The report is expected to be primarily based on a desktop study of the amount of spectrum that could reasonably be forecast to be required by the relevant players over the period 2011-2031; and
• If the respondent considered it feasible, the response could also consider modeling to support the recommendations.

 The Ministry has received a draft report and a draft model from the consultant and has asked us to provide a peer review of this analysis. The Ministry has requested that we assess the consultant's analysis and submit a short report on:

• The completeness of the consultant's report vis-à-vis the tasks listed above;
• The appropriateness of the assumptions used;
• The logic of the arguments presented;
• The reasonableness of the conclusions; and
• Provide other comments, as may be appropriate, that might enhance the robustness of the report.

 Top
1. Assessment of the report

1.1. Overview of draft report

The question asked of the consultant is a particularly challenging one: how should spectrum be allocated to maximise the value of spectrum to society over the period 2011-2031?

In their draft report the consultant addresses this question by developing two analyses:

1. A benchmarking study of comparable operators in other countries incorporating a multivariate regression used to explain variations in spectrum use in the sample. The model is then used to predict a benchmark (but not optimal) spectrum allocation for the two major New Zealand mobile providers. It is noted that actual holdings of these two providers exceeds this predicted value, leading the consultant to conclude that "in comparison to the operators within our sample, both Telecom and Vodafone New Zealand have a relatively generous spectrum allocation within the 800/900 MHz bands".
2. A bottom-up model of spectrum requirements and network costs for two operators during the period 2011-2031. The model is used to derive a cost-minimising allocation of 20 MHz of spectrum in the 800/900 MHz range between two operators.

We consider these approaches separately.

1.2. Assessment of benchmarking and spectrum regression model

In section two of their draft report the consultant compares the spectrum holdings of New Zealand mobile operators with comparable countries overseas, and then develops a multivariate regression model to explain a significant proportion (over 80%) of the variation in spectrum holdings per carrier.

The benchmarking comparison of New Zealand operators against other carriers is presumably conducted to give an idea of the spectrum holdings required to operate a cellular network in conditions comparable to New Zealand. As evidence of the spectrum required for the period 2011-2031, however, the current approach has at least two drawbacks.

One, the consultant does not draw any distinction between incumbent and entrant carriers. As we note in a later section, Exhibit 2.7 illustrates a sharp distinction between entrants, almost entirely located in the top half of the table, and incumbent operators, mostly located in the bottom half. Incumbents have legacy platforms and may therefore have greater spectrum requirements. Without highlighting this distinction, a simple comparison of Telecom and Vodafone spectrum holdings with single-technology overseas entrant operators is comparing apples with oranges.

Two, we are not convinced that without controls for spectrum requirements of technology change and technology transition, changes in services and population a simple cross-sectional benchmarking study of existing spectrum allocation is any guide to future spectrum requirements over so extended a timeframe.

The consultant then develops a multivariate regression model to explain spectrum allocations. We have two main reservations with this study.

First, as a top-down model of the current explanators of spectrum use, the regression approach has only limited appeal either as a model of the drivers of current spectrum use or as a predictor of future spectrum requirements. As noted above, spectrum requirements depend on a host of factors including technology, the timing and frequency of transitions between old and new technologies, and the range, type and timing of value-added services that may be introduced. These are not easily captured in a top-down model, and not captured at all in the consultant's regression model.

Second, the regression model, even if used to produce a forecast of spectrum requirements, is unlikely to produce predictions that outperform those available from a bottom-up model of the type developed by the consultant in the second half of their report. The reason is that a bottom-up model readily incorporates country-specific features such as terrain and population distribution, technology change, spectrum requirements during the transition to new technology, and time.

We believe the multivariate regression would be usefully developed to produce predicted spectrum requirements over the period 2011-2031, with explicit in-model controls or post-estimation adjustments for spectrum demand resulting from new technology and technology transition. These results might be used as an independent check of the spectrum requirements calculated by the bottom-up model.

1.2.1. Conclusion on benchmarking and spectrum regression model

In its current form, we do not believe either the benchmarking study or the multivariate regression model satisfies or contributes to any the Ministry's objectives. The Ministry learns nothing about spectrum requirements during the period 2011-2031 from these analyses in their present form. We have suggested how the regression model could be developed to predict future spectrum requirements for use as a check on results derived from the bottom-up model in section 3 of the consultant's report.

1.3. Assessment of network modelTop

In section 3 of their report, the consultant develops a bottom-up model to address the Ministry's request for advice on optimal spectrum requirements. We believe this is a correct approach to the problem they have been asked to address, in part because bottom-up modeling permits a wide range of country-specific factors to be incorporated into the analysis while remaining tractable, and in part because technology, competition, and investment incentives are more readily modeled bottom-up.

Our review of the analysis indicates its results are driven primarily by assumed constraints. We suggest ways that the model can be usefully extended to meet the Ministry's objectives.

1.3.1. Methodology

The Ministry has asked for advice on:

• The optimal quantity of spectrum required by the current management right holders in the 800 MHz and 900 MHz bands over the period 2011-2031, and
• The optimal quantity of 800 or 900 MHz spectrum that a new entrant would require in the 800 MHz and 900 MHz bands to operate a nationwide cellular network over the period 2011-2031.

To address this, the consultant develops a bottom-up model to determine the optimal amount of spectrum in the 800/900 MHz bands to allocate between two operators. The consultant defines "optimal" as the amount of spectrum that results in the lowest total cost for both operators.

The analysis proceeds by assuming one entrant into each 20 MHz block of spectrum to operate alongside a single incumbent. In view of the Ministry's objectives, we believe other entry assumptions are appropriately tested. However, relaxing this one-entrant assumption has two major implications. One, since the model output follows directly from assuming a single entrant in each of the 20 MHz bands, alternative entry scenarios automatically produce different optimal allocations. Two, when entry is permitted to vary, the cost minimisation standard as it is currently defined is unable to detect an optimum because it does not control for the effects of differences in competition under each entry scenario. It is therefore unable to select between the calculated optimum allocations across different entry scenarios.

To illustrate this second point, consider a plausible (though not necessarily correct) response to a claim that zero entry maximises spectrum's value. A response is that a single operator of each spectrum block is unlikely to maximise the value of the available spectrum because there is less competition in the market. Without assumptions about the relationship between competition and cost, cost minimisation is unable to verify this claim. The optimal spectrum allocation is unresolved, and the Ministry's objectives are not satisfied by the existing cost minimisation standard.

The question of what degree of entry maximises the value of spectrum to society is complex. It is almost certainly not properly informed by calculating and comparing the expected cost of building and operating a network under different spectrum allocations.

The analysis would usefully enhanced by:

• Expanding the number of entry scenarios considered; and
• Developing the definition of cost minimisation to incorporate competition, or conduct the analysis using an alternative proxy for the Ministry's objective such as total welfare maximisation.

1.3.2. Dynamic vs. Static Efficiency

A well-known result in economics is that long run efficiency is achieved not by minimising costs or maximising welfare at any one point in time but by doing so over time. A fundamental problem with the consultant's model is that it applies a static view of costs to a dynamic market. Specifically, the model does not account for technology transitions, during which time two networks are operated alongside the other. Spectrum demand is substantially higher when multiple mobile technologies are being operated. The model also fails to account for uncertainty, which we discuss later.

Technology transition and the heavy demands it places on spectrum is a fact of life for incumbent mobile providers. Exhibit 2.7 in the consultant's report illustrates the point. Firms with relatively small spectrum allocations in the top half of the table are recent entrants into cellular telephony operating with relatively limited spectrum. The heavy spectrum users in the bottom half of the table are almost entirely incumbent providers. It is these providers in the lower half of the table who may be operating legacy networks.

Legacy networks affect demand for spectrum. In its submission dated 4 September 2006,² Telecom reports it is using its current allocation of 20 MHz in support of its CDMA and AMPS networks. According to Telecom, this will change following the switching off of its AMPS network in April 2007, but it is likely that at some point in the period 2011-2031 a new network will be constructed alongside its CDMA network and Telecom will enter another transition period lasting several years. Vodafone is also delivering its services via two mobile technologies. Top

The bottom-up model would be usefully enhanced by:

1. explicitly recognising the co-existence of multiple technologies at any one point in time in a given block of spectrum;
2. adjusting incumbents' spectrum requirements to reflect the effect of legacy on spectrum demand; and
3. adjusting entrants' spectrum requirements to reflect the effect of legacy on spectrum demand as they eventually transition to new mobile technologies.

1.3.3. Investment under uncertainty

In preferring a static cost model, the consultant overlooks the effect of uncertainty in investment decisions. In a dynamic market like mobile telephony and data services, cost minimisation at any one point in time is unlikely to produce dynamic efficiency, meaning the maximisation of society welfare over time. At any one time under an efficient mobile network design, there will almost certainly be slack in the use of spectrum and other resources, because some slack is optimally reserved for future technology transitions, and for responding to uncertainties such as the introduction of new and unforeseen services that is a normal part of competition.

The question of optimal slack in spectrum use is complex, but an assessment of it is certainly within the scope of the consultant's brief given the Ministry's objectives of a) optimising incentives to invest for both current holders and new entrants, and b) estimating the amount of spectrum that could reasonably be forecast to be required by the relevant players over the period 2011-2031. Part of the cost of depriving an incumbent of some of its spectrum must include the cost of reduced flexibility and options, and in our view a question on the optimal allocation of spectrum is appropriately informed by these costs.

The model would be usefully improved by developing a qualitative analysis of the effect of uncertainty on an optimal spectrum allocation, and adjusting the model to account for uncertainty in line with this analysis. 

1.3.4. Certainty attached to entrant's likelihood of entry

In the analysis it is assumed that entry will occur with certainty, but the Ministry's objectives would be better satisfied by an analysis that took into account the possibility that a potential entrant holding spectrum rights may choose not to enter. This uncertainty is particularly relevant given the timeframe considered in the analysis.

With any future event, the expected value of it is the event's "payoff" if it occurs multiplied by the probability of it occurring. The consultant effectively assumes that when spectrum is allocated to an entrant then entry will occur with certainty. This is an unrealistic assumption for two reasons. One, firms operate under uncertainty and as this uncertainty is resolved plans for entry may change. Two, spectrum is an asset and firms may rationally purchase spectrum for reasons other than entering the market. An entirely plausible investment strategy, and indeed one that may have occurred in New Zealand, is that firms purchase spectrum rights then adopt a wait-and-see approach to entry.

A decision to deprive an incumbent provider of spectrum may be costly in a way that the consultant's model does not capture. An incumbent that is deprived of, say, 10 MHz of spectrum will see its costs of provision increased (these are costs which the consultant captures in the current model). However, without entry into the block of re-allocated spectrum, there may be no benefit offsetting the incumbent's higher costs, possibly causing a reduction in total welfare.³

In our view, the model would be usefully enhanced by incorporating an adjustment to calculations that reflects uncertainty of entry by a potential entrant. 

1.3.5. Model logic

The model derives complimentary, convex cost curves for both the incumbent and new operator assuming that the available spectrum is limited in their model to 20 MHz. The net result is that the model determines not that the optimal spectrum is 10 MHz but rather that is half of whatever the assumed constraint is.⁴ While there are differences between incumbent and entrant that feed into identifying the cost-minimising allocation, these ultimately make no difference, and model output is not sensitive to changes in these differences. Given the granularity of spectrum bands being tested (5MHz blocks) this is a predictable outcome; the optimal allocation could effectively have been determined from the point that a) a representative negative convex cost curve was defined, b) the number of entrants was fixed, and c) the spectrum to be shared between two operators was fixed. 

1.3.6. Spectrum blockTop

In the documentation we have been provided, the Ministry does not constrain the analysis to consider only a fixed 20 MHz of spectrum. Rather than consider the optimal allocation of fixed spectrum, the consultant may usefully consider an optimal allocation under alternative constraints. For example, an alternative test could combine two 20 MHz blocks and consider optimal allocations of a 40 MHz block of spectrum between n providers.

1.3.7. Aggressive market share gains for entrant

The consultant assumes aggressive gains in market share for new entrants. Following entry in 2010, entrants gain 50% market share in voice five years and 50% market share in data and modem after 10 years (see Figure 1). Other things being equal, the effect of this fast gain in market share is to increase its share of spectrum under the cost-minimizing objective.
 

Figure 1: Assumed entrant market share voice/data/modem - first 10 years⁵

We would like to see slower market share gains for the entrant tested. Success of the kind postulated for this entrant is comparatively rare in telecommunications and in our view a less aggressive gain in market share is more reasonable. 

1.3.8. Conclusion on bottom-up model

While the overall approach to the problem is sound, the bottom-up model in its present form requires considerable development. The Ministry is asking a difficult question and in our view a more elaborate analysis that reflects important drivers of demand or spectrum, and which models key inputs such as entry rather than assumes them, is required.
 

2. Assessment of objectives

The consultant was asked to estimate the optimal quantity of spectrum that will be required by incumbent and entrant mobile operators over the period 2011-2031. The consultant has addressed this objective by developing a bottom-up model of network costs. We have, however, noted our reservations with important aspects of the draft analysis, and offered some suggestions on how it might be further developed. Our main concerns with the approach in the consultant's draft report are:

• Entry is fixed at one entrant per 20 MHz spectrum by assumption when other entry scenarios may plausibly optimise the value of spectrum to society;
• Cost minimisation is not a suitable proxy for spectrum value when entry is not fixed and an alternative objective function is required;
• Entry by a rights-holder is assumed to occur with certainty when theory and experience both indicate this is not always the case. This has important but complex implications for spectrum value;
• The model not allow for any "slack" in spectrum use and thus ignores uncertainty; and
• The effects of legacy systems and transitions between technologies on spectrum demand are ignored.

¹ The report we critique is Renewal of Management Rights for Cellular Services (800/900 MHz): What Is the Optimal Spectrum Quantity? Network Strategies Report Number 26019, dated 24 October 2006.

² Telecom New Zealand Submission on Ministry of Economic Development Discussion Paper, Renewal of Management Rights for Cellular Services, 4 September 2006.

³ This is another complex issue because holding the option to enter, and the competitive tension the threat of entry may bring to bear on other operators in the market, has value even if entry does not actually occur. Even without entry, the reallocation of spectrum may still offset high incumbent costs and thus be optimal.

⁴ This is, strictly speaking, inconsistent with the Ministry's directives as provided to us, which ask for an optimum without regard to existing allocations

⁵ On this chart, the line for Handheld (data) is entirely behind the line for Modem.