Public Interest Test

Allocating spectrum to mobile operators by auction is the mechanism preferred by most regulators, and allows a scarce resource to be assigned to those who value it the most, so helping to identify the highest value use and users. However, much depends on the auction format and auction rules, which are usually designed to deliver policy objectives, but there are aspects of these formats and rules that present a high risk of adverse consequences, as they distort spectrum auctions. The common distortions are:

• The use of set-asides or spectrum caps
• Excessively high reserve prices and licence payment terms
• A choice of auction format that does not suit the market conditions in a particular country.

PUBLIC INTEREST CONSIDERATIONS

The public good is best served by policies that advance the following objectives:

• Efficient allocation: a distribution of spectrum that maximises social utility
• Transparency of the award process
• Increased broadband access including for rural and remote customers
• Innovation and network investment
• Undistorted competition
• A high net economic return to the public.

Inefficient spectrum allocations may lead to a reduction in total industry capacity, potentially leading to greater network congestion, and an opaque allocation process could lead to slow or curtailed network deployment and reduced consumer choice. These criteria are well accepted in most countries. However, there is ample scope to disagree on what constitutes a high net return to the public. Policymakers often seem to emphasise the direct auction proceeds a government will receive, at the expense of wider social and economic benefits; as Thomas Hazlett and colleagues observe in ‘What really matters in spectrum allocation design’:

“The ratio of social gains [is of] the order of 240-to-1 in favour of services over licence revenues… Delicate adjustments that seek to juice auction receipts but which also alter competitive forces in wireless operating markets are inherently risky. A policy that has an enormous impact in increasing licence revenues need impose only tiny proportional costs in output markets to undermine its social utility.”

Higher retail prices and/or reduced network investment, whether caused by artificially high spectrum costs or by reduced competition, undermine the adoption and consumption of services, bearing negatively on digital participation and economic productivity. Moreover, once they are lost, these economic benefits cannot be recovered.

“There is ample scope to disagree on what constitutes a high net return to the public.”

SAFEGUARDS BY SET-ASIDES OR SPECTRUM CAPS

Spectrum set-asides, floors and/or spectrumacquisition caps are designed to prevent excessive spectrum concentration, and need to be determined with great care to avoid unduly distorting outcomes. The main risks include inefficient spectrum use and, in the case of spectrum reservations, potentially higher spectrum costs on aggregate, resulting in higher retail prices and/or reduced network investment. Recent examples of unfavourable outcomes caused by spectrum reservations or caps include the following:

• The 700 MHz APT band plan auction (September 2014) in Brazil was set up assuming four national bidders, despite the fact that consolidation was certain to take place. Of the 2 x 45 MHz, one near national block (lot 4) of 2 x 10 MHz was unsold. The government did not achieve its objectives in terms of revenue from the auction, fostering competition, and maximising quality of rural connectivity. It will be difficult to sell lot 4 competitively at a later stage, because the winner of the adjacent lot 2 (TIM Brazil) would be able to attach a much higher value to lot 4 than the winners of the non-adjacent lots 1 and 3.
• Spectrum acquisition caps in the 2.6 GHz auction in the Netherlands in 2010 were so tight as to reflect a de facto set aside for new entrants. Since only two new entrants participated in the auction (instead of five expected), the entire TDD band and several blocks of FDD spectrum were unsold.
• In Belgium, the packaging of lots in 2011 reflected a de facto reservation for an entrant of 2 x 15 MHz at 2.6 GHz, which was unsold. It will be more difficult to remedy this inefficiency than was the case with unsold 2.6 GHz in the Netherlands, because the current band plan is enshrined by royal decree.
• The applicable spectrum caps in Chile effectively excluded the three incumbents from the AWS auction in 2009, leading to an award of 2 x 45 MHz to two new entrants. Over four years later, this spectrum serves less than 1% of the market. This inefficient use of spectrum is likely to be difficult to reverse. The economic impact of this inefficiency is assessed below.
• The 2008 AWS spectrum auction in Canada included 2 x 45 MHz. One 2 x 5 MHz and two 2 x 10 MHz blocks were available to all bidders, as well as two 2 x 5 MHz blocks and one 2 x 10 MHz block that had been set aside for new entrants. There were five significant new entrants (Shaw, Wind, Mobilicity, Videotron and Eastlink) which won spectrum. Shaw, an existing fixed operator, has not deployed five years after the auction. Accordingly, the public lost out due to lower auction receipts, while Mobilicity is in bankruptcy administration and Wind is near insolvent. The government’s attempt to create a fourth national carrier therefore failed, with great cost to the public.
• During the 2012 multiband auction in the Netherlands, 2 x 10 MHz at 800 MHz was set aside for the recent entrants. The net effect of this reservation was to artificially increase contention among the three incumbents (KPN, Vodafone and T-Mobile) for the remaining 2 x 20 MHz, leading to exceptionally high spectrum prices. KPN suffered a cut in its credit rating following the auction, which is likely to hamper investment. The high burden of spectrum fees on the industry is almost certain to be felt, indirectly, by consumers.

“Artificially high reserve prices are prone to distort outcomes and harm the public interest.”

EXCESSIVE RESERVE PRICES ARE NOT IN THE PUBLIC INTEREST

Artificially high reserve prices are prone to distort auction outcomes and harm the public interest in a number of ways:

• Spectrum may be left unsold and unutilised, representing a productivity loss to the state
• An unnecessarily high cost-burden may be imposed on the industry, leading to adverse downstream consequences
• National imbalances in spectrum holdings may be exacerbated.
• High reserve prices carry the risk that valuable spectrum resources are left unsold. Depriving mobile communications markets of key inputs comes at significant economic and social cost. If spectrum is left idle during a period when it could have been used productively, that economic loss is permanent, as the examples below highlight:
• In New Zealand the APT band plan 700 MHz auction (October 2013) had spectrum caps so that each of the three incumbent operators would be able to obtain 2 x 15 MHz. However, reserve prices were so high that 2degrees, the smallest operator, could only afford to acquire 2 x 10 MHz. As a result the company is burdened with a higher deployment cost per MHz of LTE, and competition is weakened.
• In India, successive failures to fully allocate the available 1800 MHz spectrum between 2012 and 2014 have led to delays in its deployment. This issue might not have occurred if moderate reserves had been applied from the outset, allowing the market to determine the final prices paid.
• Due to the adoption of high reserve prices in the 2013 Australian digital dividend auction ($1.25/MHz/pop for 700 MHz), a third of the prime 700 MHz band was left unsold. This has created an intractable problem for the regulator, particularly as the government indicated prior to the auction that operators should not expect unsold spectrum to be made available at a lower price in the future. It is therefore possible that the unsold 700 MHz will lie fallow for a considerable period, to the detriment of the Australian economy and consumers.
• In Bangladesh, a total of three blocks of 2×5 MHz at 2100 MHz failed to attract any bids during the 3G auction in 2013. This represents 37.5% of the 40 MHz available during the auction. The high reserve prices in relative terms ($0.07/MHz/pop for a market generating less than $2 per customer a month) were likely to be the main cause, albeit ongoing tax disputes and other regulatory uncertainties may have contributed to this outcome.

CHOOSING THE RIGHT AUCTION FORMAT

If a single licence is available, regulators tend to rely on either a Sotheby’s style rising clock auction, or a sealed-bid format. A sealed-bid auction may either use a first or a second-price rule. Under the firstprice rule, winners pay the amount of their winning bids. This can lead to a ‘winner’s curse’: paying more than would have been required to win, which threatens auction efficiency. A first-price rule also yields an uncomfortable trade-off between minimising the risk of losing, and minimising the winner’s curse.

The 1998 Brazilian auction for 2G licences provides an example of this problem: BellSouth paid $1 billion more than the next highest bidder for the São Paulo metro licence. The introduction of a second-price rule eliminates this particular problem. Under the latter, the winner pays the amount of the next highest bid. A Sotheby’s style English auction approximates the second-price rule: prices stop rising when the second-highest bidder drops out. This incentivises truthful bidding, as there is no penalty for bidding full value, because winners pay no more than the minimum required to justify their allocation. If all participants bid their true values, the auction will be efficient in identifying the highest value user of the resource.

Moreover, if the two highest valuations are relatively close, this may generate higher proceeds than a first-price auction in which participants ‘shade’, ie. reduce their bids in order to avoid the winner’s curse. Because they promote allocation efficiency, second-price formats are vastly to be preferred when awarding single licences.

The main formats used to auction multiple spectrum blocks are the combinatorial clock auction (CCA), which uses a generalised secondprice rule, and the simultaneous multiple round auction (SMRA). Sealed bid formats are sometimes adopted, such as the combinatorial second-price auction and the combinatorial first-price auction.

“No format is perfect and policymakers need to consider potential sources of distortion.”

AUCTION FORMATS IMPACTING OUTCOME

Under a given set of circumstances, each of the above formats may produce widely differing outcomes. While no format appears to be perfect, policymakers need to consider the following potential sources of distortion:

1. Inefficient allocations caused by fixed budget constraints in conjunction with a lack of visibility over price exposure
2. Material price disparities between bidders for comparable packages
3. Risk of predatory bidding strategies designed to raise the costs incurred by rivals
4. Inefficiencies caused by difficulties in aggregating optimal spectrum packages
5. Inefficiencies caused by demand moderation strategies.

The CCA format is generally vulnerable to the first three of these forms of distortion, SMRA to the last two, and combinatorial first-price auctions to the first, second and last.

Following the UK’s multiband award in 2013, which used a CCA mechanism, the UK National Audit Office (NAO) observed that two bidders appeared to be subject to fixed budget constraints, and that this may have prevented them from achieving all their objectives. This potentially limited the efficiency of the auction.

The issue in a CCA is that participants have limited visibility over their actual price exposure, as the bids of rivals, which determine the price they pay if they win, are not disclosed. This introduces a potentially intractable strategic dilemma for bidders that are constrained by a fixed budget, rather than by their spectrum valuations. To minimise the risk of being knocked out, such bidders may choose to bid full valuations on smaller packages. However, in doing so they squeeze the extra amount they can bid for additional lots, while staying within their overall budget limit. This reduces their chances of winning the extra lots.

As a result, they can fail to secure the larger package that, given the relative valuations and constraints of each participant, they actually ought to win. The opposite strategy open to management is to bid the available war chest on a larger package and reduce the bid values for smaller packages. This would increase the chance of winning a more ambitious prize, but also the risk of walking away with nothing. Given the resulting threat to efficiency, the NAO recommended that UK regulator Ofcom “select designs for future auctions that take account of the circumstances of likely bidders, such as the likelihood that they will be subject to budget constraints”.

Since prices in a CCA are determined by the marginal bids of rivals, there is potential for pricing asymmetries. For example, Swisscom secured a larger package, yet paid a lower amount than Sunrise during the Swiss multiband auction in 2012. The limited feedback between what one bids and what one pays may also invite bid strategies designed to impose higher costs on competitors.

The SMRA format introduces different threats to efficiency. First, since participants bid on individual lots rather than on entire combinations, it is possible to win an unwanted subset of a target package. Subsequently, management’s need to manage this risk may distort bid behaviour. If the lots reflect specific frequency ranges rather than generic quantities within a given band, there is also a risk of fragmented assignment, which is technically inefficient. Neither issue arises in a CCA or in a first-price sealed-bid auction, because participants in these either win a whole package bid, or nothing at all.

Second, SMRA may be construed as a tacit negotiation between the auction participants: the quicker everyone agrees on who gets what, the less bidders will pay collectively. While demandmoderation strategies are rational, these may impact spectrum auction revenues. However, if an efficient allocation is the prime objective, policymakers should be less concerned about the impact on auction receipts – which will be, in part, self-correcting through corporate taxation. Nor does demand moderation necessarily lead to a suboptimal distribution of resources: if bidders are realistic and correctly gauge relative spectrum valuations, the outcome may be the most efficient.

CCA and SMRA both have disadvantages. However, the combinatorial first-price auction introduces even greater risks, which also share the following key drawbacks with the CCA format:

• Increased knock-out risk, because bidders are unable to respond to unforeseen events (this risk is further exacerbated by the incentive to shade bid values in order to minimise the winner’s curse)
• Risk of price disparities
• Risk of distortion caused by budget constraints.

Combinatorial first-price auctions are the most distorting of the main auction approaches and should be avoided. Opinions will differ on whether CCA or SMRA is more prone to distortions overall. The relative levels of risk will also depend on circumstances as well as the detailed rules. Regulators should examine their national circumstances and reflect feedback from operators in their ultimate choice of format.

ECONOMIC COST OF INEFFICIENT SPECTRUM ALLOCATIONS

The economic cost of inefficient spectrum allocations can readily be quantified by analysing their effect on total industry capacity. Comparing the latter with potential (ie. unconstrained) demand allows us to express the inefficiency in terms of foregone consumption. This can in turn be used to gauge the impact on consumer and producer surplus (CS and PS):

• India’s 850 MHz and 1800 MHz awards (2012–14): 15% of the available 850 MHz plus 1800 MHz bandwidth is unutilised. The CS plus PS foregone: $3.6 billion a year.
• Bangladesh’s 3G auction (2013): 37.5% of the spectrum on offer was unsold. The potential GDP growth impact amounts to $1 billion a year.
• Chile’s AWS auction (2009): 43% of industry bandwidth was reserved for entrants, serving 0.9% of the market five years on. The GDP growth foregone amounts to $0.4–$3.4 billion a year.

Where spectrum is left unsold due to high reserves, these costs sometimes far exceed the extra auction proceeds that may reasonably be attributed to them, as was the case in India. As this highlights, the social and economic cost of inefficient allocations is often substantial, especially if spectrum is left fallow or is underutilised for prolonged periods, making it vital that governments research and choose the right auction format.