Faculty of Economics and Business
MSc. Thesis in Economics
Programme: Industrial Organisation, Regulation and
Competition Policy
The Effect of Incumbent
Restrictions on Revenue in
European LTE Auctions
Author:
Peter Dubovsk´
y
Supervisor:
dr. A.M. Onderstal
Statement of Originality
This document is written by Student Peter Dubovsk´y who declares to take full
responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Contents
1 Introduction 3
2 Background 4
2.1 Radio spectrum and its auctioning . . . 5
2.2 Auction types . . . 5
2.3 Set-asides and spectrum caps . . . 6
3 Literature review 7 4 Theory 9 5 Data 11 5.1 Features of the auctions . . . 12
5.2 Auction revenues and price index derivation . . . 14
5.2.1 Categories of bands . . . 14
5.2.2 Dealing with the combinatorial auctions . . . 15
5.2.3 Dealing with annual fees . . . 16
5.2.4 Duration adjustment and finalizing the price index . . . 16
5.3 Normalizing revenues across spectrum bands . . . 17
5.4 Independent variable . . . 20
5.5 First check of the hypothesis . . . 21
6 Model and regression results 21 6.1 Model . . . 21
6.2 Results . . . 22
7 Conclusion 24
1
Introduction
Since its introduction in 1994 in the United States, spectrum auctioning has grown in popularity and become a preferred way of allocating spectrum rights for the govern-ments of many countries. Auctioning of radio frequencies has several advantages in comparison to other means of selling spectrum rights, like the administrative selec-tion process, lotteries or the ‘first-come first-served’ method. The specific advantages of spectrum auctions in comparison to other methods include higher transparency, higher efficiency, and greater revenue formation (McMillan, 1995).
Experience with the first spectrum auctions pointed at some issues that could arise during an auctioning process. According to Klemperer (2002), one of the biggest disadvantages of spectrum auctions is entry deterrence. He describes it as a phenomenon that occurs when companies which value auctioned spectrum less than other bidder(s) are discouraged from participating in an auction due to the high probability that they would not win any spectrum. Ascending auctions are commonly used for spectrum auctioning because of their ability to assign the prize to highest-valuing bidder. But this also makes them vulnerable to entry deterrence, as higher-valuing bidders always have the option of outbidding their competitors during such auctions (Klemperer, 1998, 2002).
Entry deterrence in spectrum auctions is a great concern in practice because incumbents typically place higher value on auctioned spectrum than do companies that want to enter a mobile telecommunication market (“entrants”). The reason is that if incumbents gain all the spectrum, they safeguard their market power and, hence, higher expected profits. In effect, incumbents bid for higher expected profits than entrants. Incumbents’ valuation of spectrum is also raised by other factors: An incumbent typically has an established mobile network, a customer base, a well-known brand and higher economies of scale, as well as easier access to capital markets (Ayres & Cramton, 1996; Cramton et al., 2011; Hoppe et al., 2006; Kim et al., 2012). Entry deterrence arising from valuation asymmetry can lead to a lower number of participants in an auction, affecting both entry into the mobile telecommunications market and the auction’s revenue. According to Klemperer (1998), even a small disadvantage can substantially decrease a bidder’s chance to win spectrum in an auction (p. 758).
In order to encourage entry into the auction, auctioneers (typically governments) have applied two tools in auction design: set-asides and spectrum caps. Setting aside a part of auctioned spectrum and reserving it for entrant bidders encourages their participation, as they can be sure to either acquire the spectrum immediately or compete for it with other similar bidders that do not possess the advantages of incumbents. The other way of diminishing deterrence is by capping the amount of spectrum that a single bidder can win. While set-asides reserve a part of spectrum exclusively for entrant bidders, spectrum caps limit the amount of spectrum that a single operator can acquire. If the capping is tight enough, implementation of this tool can effectively reserve spectrum that incumbents are unable to win due to the spectrum they already hold. Unless otherwise stated, the term “spectrum caps” is used from here onwards to refer solely to spectrum caps preventing incumbents from gaining all the spectrum in an auction. A band or an auction in which incumbents
cannot win all spectrum due to set-asides and spectrum caps is referred to as being subject to incumbent restriction.
The possible existence and direction of the effect of these tools on auction revenue is valuable knowledge for auction designers. Auction revenue, being one of the least distortionary means for governments to raise money, is an important benefit of
the auctioning process.1 The importance of this subject is further emphasized by
the great size that spectrum revenue can attain. For instance, the UK spectrum
auction in the year 2000 raised£22.5 billion (see e.g. Binmore & Klemperer, 2002),
which highlighted the ability of spectrum auctioning to raise substantial amounts of money. Hence, the revenue creation is implicitly pushed for even due to political reasons—raising low revenue could backfire on a government (McMillan, 1994). This thesis therefore seeks to answer the following question: What is the effect of set-asides and spectrum caps on spectrum auction revenue?
This study uses data from the spectrum auctions that were run in 16 European countries, from 2008 to 2013, in order to test the hypothesis that the presence of set-asides and spectrum caps in an auction leads to higher revenue in that auction. Over the last few years, technological development in the telecommunications industry, particularly the advent of LTE, has triggered a wave of auctions across Europe. In the series of auctions, several were designed with set-asides and spectrum caps.
The ordinary least squares (OLS) estimates do not allow me to reject the null hypothesis that revenue remains unaffected by the presence of set-asides and spec-trum caps. This implies that set-asides and specspec-trum caps have no effect on auction revenue. As the data indicates, an explanation for the zero effect is that incumbent restriction might be unable to attract entrant bidders.
The rest of this thesis is organized as follows: Section 2 describes the subject area of the thesis—that is, spectrum auctions—and explains some basic concepts, including the design features of set-asides and spectrum caps. Section 3 presents an overview of the related literature. In Section 4, using a theoretical model, the expla-nation of why set-asides and spectrum caps might affect auction revenue is given. Section 5 presents the gathered data and explains necessary data modifications. In Section 6, an econometric analysis is conducted and its results discussed. Section 7 concludes the study.
2
Background
This section explains in brief what radio spectrum is as well as why and how it is auctioned. Subsequently, the most popular types of auctions are described. Descrip-tions of set-asides and spectrum caps and their workings are also provided.
1Distortionary effect is the effect that causes deviation from the pareto-optimal allocation of
resources in an economy. In case of taxes, which are the main source of a state’s income and are typically distortionary, the distortion created is in the form of the change in relative prices of commodities, work, etc.(see e.g. McGrattan, 1994)
2.1
Radio spectrum and its auctioning
Radio spectrum is an indispensable resource for the mobile telecom industry. It is a specific part of the electromagnetic spectrum with a wave frequency 3kHz–3GHz, which allows a signal to travel long distance through the air without being absorbed by the atmosphere. This trait has determined utilization of the radio spectrum after the discovery of wireless telecommunication in the 19th century. Radio spectrum has been essential for telecommunication companies ever since the birth of the mobile telecommunication industry and the mobile revolution in the 1990s, when a decrease in mobile-phone technology costs triggered a massive spread of mobile phones that have since been used for the transmission of voice and data.
Radio spectrum is a scarce resource too. The scarcity is caused by the necessity of having an exclusive bandwidth for successful communication—for instance, two radios (in same region) cannot broadcast on the same frequencies. With a rising number of devices using the spectrum and a rising amount of data being transferred, the fact of its scarcity has become increasingly evident. Although technology is advancing in the area of spectrum utilization, promising vast growth of its capacity,
substantial increase can be anticipated in wireless traffic too.2
In order to prevent communication conflicts and encourage efficient use of the spectrum, governments started to regulate the utilization of spectrum by licensing it for exclusive use. A license would guarantee a company an exclusive right to use specific frequencies for a fixed period. Telecommunication companies have been those with the highest demand for the spectrum. Governments have followed various procedures to assign the frequencies to companies. These mostly included admin-istrative processes (“beauty contests”), lotteries, “first-come first-served” methods and auctions (McMillan, 1995).
For several reasons, auctions emerged as the best among these procedures: These “are transparent and fair; generate revenue for the government; reveal the firms’ estimates of license values; assign licenses to firms quickly and economically; can be designed to incorporate a wide range of public policy goals”(McMillan, 1995, p. 199). McMillan (1994) summarizes the auction objectives of the US government set out in the US legislation before the spectrum auctions were launched. These are intensive and efficient use of spectrum, easy implementation of new technologies, avoiding high concentration, attaining a wide variety of spectrum winners, and recovery of the spectrum value for public use (raising revenue). As spectrum auctions have been gaining popularity in the US and in many countries across the world, it is reasonable to believe that these are generally the goals of any government that decides to auction radio spectrum.
2.2
Auction types
Simultaneous multi-round ascending auctions (SMRA) and combinatorial clock auc-tions (CCA) have been the most frequently used auction formats in Europe. The dataset for this thesis comprises 11 SMRA auctions and nine CCA auctions (see also Section 5).
2For example, according to Cisco: “Global mobile data traffic will increase nearly 11-fold
The SMRA auction type is the older and simpler format. It was created as an expansion of the English auction design for the purpose of selling multiple items like, for example, the licenses in spectrum auctions. In an SMRA auction, bidders bid in a series of ascending auctions—one auction for each spectrum lot. In each round, the bidders increase their bids on the individual lots in which they are interested. An auction ends in the round when no more bids are placed; winners pay a price derived from their final bids in accordance with the auction rules.
The CCA type is more recent and more complex of the two formats. It normally consists of two stages: The first is an allocation phase consisting of clock rounds and a final supplementary round; the second is an assignment phase, in which winners of the previous phase bid for specific positions of the spectrum lots that they won in the first phase. In the clock rounds, bidders bid for the amount of lots that they are interested in at the announced price. Spectrum lots are generic—their exact position is not known; hence, they are equivalent. If the demand for lots surpasses supply, a new round starts with a higher price. When the demand is at the level of supply (or below), clock rounds end and a supplementary round follows. In this single sealed-bid round, bidders can increase their bids, as well as place bids on other combinations of lots. At the end of the supplementary round, winners are selected based on the combination of bids bringing the highest revenue. Subsequently, the winners bid for a specific position of their lots in an assignment phase. The final price they pay is a sum of the prices of both phases, derived from their bids according
to auction rules.3 Unlike in an SMRA auction, only the revenue for a bidder’s whole
package of won frequencies is known.
2.3
Set-asides and spectrum caps
Set-asides and spectrum caps are some of the tools that are used for promoting effi-ciency in auctioning; their main function is to ease entry into a market and increase competition within an auction (Cramton et al., 2011). Greater efficiency in spectrum
use is arguably the most important reason to auction radio spectrum.4 Auctioning
cannot always maximize this attribute by itself; therefore, auction design tools are
used to enhance this ability.5
A set-aside is the part of spectrum in a spectrum auction that is reserved for designated companies. This means that only the designated companies are allowed to submit bids on this part of the spectrum. Such companies are specified by a government depending on its particular goal. In practice, this has been either helping
the companies of historically disadvantaged owners6 to participate in a market, or
assuring higher competition via entry of a new company. In the European spectrum
3Usually a second-price rule
4For example, according to Cramton et al. (2011), the goal of an auctioneer is an efficient use
of spectrum or the “assignment of licenses that maximizes the consumer value of wireless services less the cost of producing those services”(p. 169).
5Besides the set-asides and the spectrum caps, the efficiency-enhancing instruments are bidding
credits, band plan, auction design and antitrust enforcement (Cramton et al., 2011).
6Companies owned by groups that have been economically (and otherwise) disadvantaged in
auctions examined here, the set-asides are normally reserved for entrants.7 The
reason for supporting entrant bidders is their disadvantaged position in an auction. Spectrum auctions are asymmetrical with respect to the value of spectrum because the incumbents’ valuation is generally higher (Klemperer, 1998)(see also Section 4). A spectrum cap is a restriction on the amount of spectrum that a mobile operator can hold. In each auction, it is defined whether the limitation applies to all of the spectrum held by the operator or just to the spectrum offered in an auction, as well as which spectrum bands are counted in. The spectrum caps can create a “space” in the auctioned spectrum, which cannot be reached by incumbents. Therefore, a strict enough capping can have a similar effect in an auction as a set-aside.
Both set-asides and spectrum caps are used to limit market concentration. This might be good or bad for a government as lower concentration does not always translate into higher efficiency—for example, due to the inability to attain opti-mal aggregation of spectrum (Cramton et al., 2011). Nevertheless, this thesis does not measure their impact on efficiency; the focus on the set-asides and spectrum caps solely concerns the incumbent restriction they create, and its effect on auction revenue.
3
Literature review
This section outlines the research that has been done on the topic of set-asides
and their effect on auction revenue.8 First, the paper that this thesis is closest to is
described, followed by other articles—initially, those that observe positive correlation between the presence of set-asides and the size of revenue and, later, those that did not find such a relation. The final part of this section explains how the thesis contributes to existing research in this area.
Out of the scholarly articles examined for this purpose, the subject of my thesis is mostly related to that of Ayres & Cramton (1996). They argue that affirmative action (subsidizing designated companies in an auction) can increase competition and lead to lower costs of the subsidy or even raise the revenue of an auction. This might be surprising for many because such actions used to be denounced as “giveaways”. In the specific case of set-asides, the revenue-raising effect is caused by two means: first, by decreasing the number of licenses available for non-designated companies, and, second, by the crossover bidding of designated companies on non-designated lots. Ayres & Cramton support their theory by using bidding data from an FCC spectrum auction, in which 40% bidding credit as well as softer rules of financing allowed the designated bidders to pay only about a half of their bids on the designated blocks. This was a large enough advantage to effectively create set-asides on those blocks. The designated bidders still had an advantage of around 16% on the rest of the spectrum due to cheaper installments. Ayres & Cramton (1996) argue that the affirmative action motivated a high demand of entrants, which eventually spilled to other licenses and led to an increase in revenue by roughly 25%.
7The one exception is the UK 2013 auction, where the set-aside was reserved for the weakest
incumbent.
In addition, Kim et al. (2012) got a result that is in line with the finding of Ayres & Cramton (1996). The former create an experimental agent-based computational
model. Within it, they run a variation of the ascending-bid auction developed
by Ausubel (2004). They reach the conclusion that the revenues of ascending-bid auctions increase when the number of set-asides decreases and that of entrant bidders increases. They find that the revenues increase since less spectrum is available for incumbents, which are then forced to compete for a smaller piece of spectrum. They also find that these results hold if there are more entrants than set-asides.
Nakabayashi (2009) finds that the set-asides can be beneficial to a government even when it has the role of a purchaser instead of a seller—specifically, in pro-curement auctions. He argues that the set-asides for small and medium enterprises (SMEs) decrease government costs in procurement auctions (where a winner has the smallest bid). In his findings, there are two counteracting factors: the effect increas-ing the costs by not allowincreas-ing bigger, more cost-efficient companies to bid for the set-aside projects, and the effect decreasing the costs due to the higher competition of SMEs. Nakabayashi writes that the second effect prevails and, therefore, overall,
the set-aside program decreases the procurement costs.9
Some of the research, however, implies a rather ambiguous effect of set-asides. For example, Meng (2011) finds that the effect of set-asides on the revenue in spec-trum auctions varies—when a market is big or a cost asymmetry low, the set-asides decrease the revenue; in other situations, as his model and simulation show, the set-aside auctions result in higher revenue.
Other economists argue that set-asides have a negative effect on auction revenue. Rothkopf et al. (2003) oppose Ayres & Cramton (1996), arguing that they failed to make a distinction between two separate features of the design of auctions they observed. As a result, they misinterpreted the effect of subsidies as an effect of set-asides. According to Rothkopf et al., contrary to subsidies, the set-asides do not have a positive effect on auction revenue because they “reduce the incentive for disadvantaged bidders to compete with first-line bidders, while subsidies enhance this incentive”(p. 83).
Katz (2012) reasons, in an article commissioned by Verizon10, that set-asides
reduce expected revenues from an auction by restricting the highest-valuing bidders which would be able to pay the most. He cites Ausubel et al. (1997), who have
found that increasing the number of bidders in PCS11 auctions raised its revenues.
According to Katz (2012), an artificially created scarcity that increases revenues, as stated by Ayres & Cramton (1996), might not compensate for the losses. Katz provides an example of four bidders: X and Y as incumbents, and S and T as entrants. If their valuation is X > S > Y > T , then a set-aside would produce
revenue Y + T ,12 which is lower than in the case without a set-aside when the
revenue would be 2Y (Katz, 2012, p. 10–11, note 28).13
9Decreasing the costs in a procurement auction is equivalent to increasing revenue in a spectrum
auction.
10U.S. telecommunications company
11Broadband Personal Communications Service 12The winner pays the second-highest bid.
13Building on this idea, it is possible to draw an implication that if the valuation is X > Y >
With respect to the research presented above, the main contribution of this the-sis lies in the different methodological approach. It is based on statistical modeling upon cross-sectional data from several auctions. The approach is empirical and uses a comparably wide array of auctions, which makes it stand apart from previous research on the topic. Moreover, the presented economic literature mainly concen-trates on the effect of set-asides on revenue. This thesis works with a concept of incumbent restriction, or, in other words, a combined effect of set-asides and spec-trum caps. Given that the possible effect of set-asides and specspec-trum caps is yet to be sorted out, the results of this thesis can be considered as another piece of evidence in this debate. Last but not least, current data is used, which makes this analysis one of the first assessments of European LTE auctions as a whole.
4
Theory
This section explains the following three issues: a) why incumbents generally value spectrum more than entrants—a simplified version of Hoppe et al.’s (2006) model
is used for this purpose14—b) how this can lower an auction revenue, and c) how
set-asides and spectrum caps might affect the revenue.
The assumptions in the simplified model are the following: In a post-auction market, n incumbents and s entrants are active; the number of companies in the market is z, which means z = n + s; there is one license to be auctioned and, hence, s ∈ {0, 1}. According to standard oligopoly theory, total industry profits are decreasing in z. The value of spectrum for a company is given by the profit the
company can gain with its use:15
an unsuccessful entrant’s profit is zero
a successful entrant’s profit is we(z)
an unsuccessful incumbent’s profit is π(z)
a successful incumbent’s profit is wi(z)
The incumbent’s intrinsic valuation is defined as v(z) ≡ wi(z) − π(z). This is
the difference between winning additional spectrum and forgoing it to the other incumbent, even as the number of companies does not change.
The preemptive willingness to pay is given by π(n) + v(n) − π(n + 1). This is the difference between the profit with the new spectrum on the one hand, and the profit without the spectrum and with another firm in the market on the other.
The valuation of the entrant is we(n + 1).
There can be three possible scenarios:
[see Section 4]), it means that a set-aside increases revenue if (Y − S) > (S − T )—that is, when the difference between the value of the second incumbent and of the first entrant is higher than that between the two entrants—which is also very plausible. Thus, if Katz’s assumptions are substituted with the more plausible ones, the results are opposite to his.
14Hoppe et al. (2006) include features to their model to allow the analysis of how a number of
auctioned licenses influences the chance of entry. In the present thesis, only those features of the model are used which are needed to explain the higher valuation of spectrum by incumbents.
15The profit is assumed to be a single number; it might be interpreted, for example, as the net
1. π(n) + v(n) − π(n + 1) < we(n + 1)
2. we(n + 1) < v(n)
3. v(n) < we(n + 1) < π(n) + v(n) − π(n + 1)
This shows that in order for an entrant to win the spectrum license, his or her valuation must be higher than the preemptive willingness to pay of an incumbent.
Such a scenario has a very low chance of occurrence in practice.16
Since an incumbent is almost always willing to bid more in an auction, entrants
have no reason to participate if there is even a small participation cost.17 In an
ascending auction, winners pay the value of the highest-valuing non-winning bidder (e.g. in a case of one license, this is a bidder with the second-highest value in an auction). The expected value of the highest losing bidder grows with the number of bidders in an auction (see e.g. Matthews, 1995). It follows that revenue in an auction with fewer participants is lower on average, ceteris paribus.
The presence of a set-aside or spectrum caps in an auction may increase its revenue. Such an effect of set-asides and spectrum caps could be working through several channels:
Decreasing supply of spectrum available to incumbents by setting aside a part of the spectrum could raise revenue in case of unsatisfied demand of incumbents (Ayres & Cramton, 1996; Kim et al., 2012; Nakabayashi, 2009). This can be illustrated using the example given by Ayres & Cramton (1996), which describes an auction in which two incumbents and two entrants bid for two licenses. Each bidder can win one license at most. Their valuations are the
following: $110 and $90 for the two incumbents, and $60 and $40 for the two
entrants. Without a set-aside, the revenue is slightly more than $120—both
incumbents pay$60 (the value of the stronger entrant) and a small increment.
On the other hand, with one of the licenses set aside, the incumbents have to
compete for a single license and the auction’s revenue is$90+$40=$130, which
is more than in the previous situation.
Designated companies18 with unsatisfied demand could eventually start
plac-ing bids on the non-reserved parts of the auctioned spectrum. Such an action is called crossover bidding. It creates additional competition to incumbents and drives the revenues up, according to Ayres & Cramton (1996). The latter illustrate this via a practical example of a specific spectrum auction run in the US. In the auction, the competition among designated bidders raised the price sufficiently, so that it was reasonable for some of the designated companies to
16The possibility of an entrant having a higher value might arise, for example, in case of better
technology or superior managerial skills, or due to diseconomies of scale of an incumbent (e.g. Gilbert & Newbery, 1982). Such cases, however, are improbable in the mobile telecommunication industry, especially in the European context, where large multinational companies (with strong industrial know-how) are present in most of the markets. The presence of diseconomies of scale in the telecom industry is also not plausible.
17Such costs may be in the implicit form of transaction costs of participation in an auction—for
example, a necessity to negotiate with investors, a necessity to create a business plan, lost earnings on deposit in an auction, etc.
place bids on non-designated spectrum.19 In spite of eventually being overbid
by incumbents in the non-designated parts, the revenues ended up at a much
higher level.20
Reserving a part of the spectrum increases the expected profits of designated bidders —the consequent level of competition could be higher than in the case where only incumbents would bid for it. Indeed, as can be seen in the example of Ayres & Cramton (1996) in the previous point, the price level for designated parts in the auction rose above the level of the non-designated parts (which triggered the crossover bidding).
On the other hand, there are possible reasons why the channels in which set-asides and spectrum caps could affect auction revenue, as described above, might not work. Indeed, they are based upon assumptions that might not be valid universally or even in the majority of auctions. In particular:
The demand of incumbents could be satisfied by the blocks that are not set aside
Competition among designated bidders for set-aside spectrum might not be strong enough
In addition, some economists argue that there is also a downright negative effect of set-asides on auction revenues (Katz, 2012; Nakabayashi, 2009). This is because bidders with the highest value are not allowed to bid on a reserved part of the spectrum, so it can be sold for less.
It is apparent that the theory does not show a clear effect of the set-asides and spectrum caps on auction revenue. To find out whether set-asides and spectrum caps could have an effect on the revenues of the European LTE auctions, the null hypothesis will be tested to determine whether the band revenues are the same in those LTE European auctions where a set-aside or spectrum caps or both are present as in those where neither a set-aside nor spectrum caps are present.
5
Data
This section presents the data used in this research; the data is cross-sectional and comes from 23 auctions that were run in the period from 2008 to 2013 in 16 European
countries (see Table 1).21 Overall, this amounts to 50 band-level observations, which
are used to test the hypothesis. For the purposes of this research, auctioned spectrum was partitioned into four categories of adjacent frequencies (the “bands”). Therefore, depending on the range of spectrum auctioned, there are one to four observations per auction.
19The designated bidders still had the benefit of cheaper financing of licenses (imposed by
auction rules), estimated to be roughly equivalent to a 12% bidding subsidy.
20Ayres & Cramton (1996) claim 30% of the final price was due to the bidding of designated
bidders on non-designated parts of spectrum.
21In some of the countries, there were two or more auctions for different bands while other
The relevant variables for each observation are the following: a band, revenue, the presence of a set-aside in an auction, the presence of spectrum caps in an auc-tion, numbers of bidders by their type (incumbent/entrant) in an aucauc-tion, and a number of entrant winners in an auction. For each observation, revenue and set-aside/spectrum caps presence is unique; the other variables are the same for all the observations/bands within an auction. Data about each country’s GDP per capita
in PPS22 is also used here as a control for purchasing power which affects auction
revenue.
The rest of this section is divided into several subsections. The first of these presents the data on auction design and gives an overview of the auctions with the indicated presence of set-asides and spectrum caps, and the numbers of bidders and entrant winners. The focus of the second and third subsections is on the auctions’ revenues: They explain the way in which a suitable price measure was created for the research out of the raw auction revenues data. In addition, descriptive statistics are provided on the revenues and present graphs for each of the bands (a country on the x-axis, revenue for a band on the y-axis). The last two subsections present the independent dummy variable identifying incumbent restriction in a band and also provide the first check of the research hypothesis.
5.1
Features of the auctions
The choice of countries with similar economic background in this study aims to decrease the chance of omitted-variable bias. The variability of auction setups, mar-kets and operators is an inevitable obstacle that one must expect to deal with when analyzing a number of national markets. The lack of uniformity cannot be curbed in its entirety; nevertheless, this thesis does so partially by focussing on a sample of auctions predominantly from western Europe. In spite of many differences, these markets share similarities in size, stage of development and companies operating in
the field. Most importantly, these markets are economically integrated and
regu-lated by similar policies.23 Indeed, it was the common European policy of digital
television transition that initiated the series of auctions run in the last few years.24
There are two main types of auctions in the present sample: simultaneous multi-round ascending auctions (SMRA) and combinatorial clock auctions (CCA). Besides these, there is also a clock auction (Sweden 10/2011), a combinatorial first price sealed bid auction (Norway 12/2013) and a tender (France 12/2011). Research is not clear on the difference in revenue-generating potential depending on a particular
auction type,25 nor does the present data show any difference between the types of
22Purchasing Power Standards—an artificial measure of Eurostat for the purpose of controlling
for different price levels in countries, calculated as a quotient of GDP in a national currency, and a purchasing power parity.
23Norway and Switzerland are members of the European Free Trade Association, while the rest
are the European Union member states.
24The 800–900 MHz frequencies that had been used for analogue broadcasting have been freed,
and offered to mobile operators to provide for the increasing demand for data services.
25Some economists argue that SMRA is superior to CCA in revenue generation. For example,
Bichler et al. (2012) conducted an experiment with teams of people acting as bidders and found that revenue is lower in CCA than in SMRA auctions. Other research indicates that the opposite
Table 1: Overview of the auctions and participants
Country Date Type Bands Set-aside Caps Incumbent Bidders
Entrants Winners Entrant Winners
Sweden 5/2008 SMRA 2600F,T No No 4 1 5 1 Finland 11/2009 SMRA 2600F,T No No 3 2 4 1 Germany 4/2010 SMRA 800, 1800/2100, 2600F,T No No 4 0 4 0 Netherlands 4/2010 CCA 2600F Yes Yes 3 2 5 2 Denmark 5/2010 CCA 2600F,T No No 4 0 4 0 Austria 9/2010 CCA 2600F,T No No 4 0 4 0 Sweden 3/2011 SMRA 800 No No 3 2 3 0 Spain 7/2011 SMRA 800, 2600F No No 3 0 3 0 Italy 9/2011 SMRA 800, 1800/2100, 2600F,T No No 4 0 4 0 Sweden 10/2011 Clock auction 1800/2100 No No 3 0 2 0 Belgium 11/2011 SMRA 2600F,T No Yes 3 2 4 1 Portugal 12/2011 SMRA 800, 1800/2100, 2600F,T No Yes 3 0 3 0 France 12/2011 Tender 800 No No 4 0 3 0 Switzerland 2/2012 CCA 800, 1800/2100, 2600F,T No No 3 0 3 0 Denmark 6/2012 CCA 800 No No 3 0 2 0 Ireland 11/2012 CCA 800, 1800/2100 No No 4 0 4 0 Netherlands 12/2012 CCA 800, 1800/2100, 2600T Yes No 3 2 4 1 United Kingdom 3/2013 CCA 800, 2600F,T No No 4 3 5 1 Austria 10/2013 CCA 800, 1800/2100 Yes No 3 0 3 0 Finland 10/2013 SMRA 800 No No 3 0 3 0 Belgium 11/2013 SMRA 800 No No 3 0 3 0 Czech Republic 11/2013 SMRA 800, 1800/2100, 2600F Yes Yes 3 0 3 0 Norway 12/2013 CFPSB 800, 1800/2100 No Yes 3 1 3 1 Note: CFPSB—Combinatorial first-price sealed-bid auction; F—FDD; T—TDD; Bands 800 and 1800/2100 always as FDD
auctions;26therefore, the auction type indicator is disregarded in the statistical tests
conducted as part of the present thesis.
Rather, the significance might lie in those (additional) features that might cre-ate an uneven playing field in an auction with respect to whether a bidder is an incumbent or an entrant. This study considers set-asides for entrants and spectrum capping rules. Any of these could ease an entry and, hence, make an auction more attractive for bidders and affect the revenue of such auction. There are 10 observa-tions where either a set-aside and/or spectrum caps are present. Besides this data, the numbers of bidding incumbents, entrants, winners, and entrant winners were also collected. This information is required to form an impression about a level of competition in an auction. It is not used for the primary analysis of this study, but adds considerably to the whole picture of an auction.
Finally, the auction design data27 used for this research includes the annual
license fees together with the data on duration of spectrum licenses. The following subsection describes how they are implemented in calculating price indices.
5.2
Auction revenues and price index derivation
Amounts paid by the winning bidders for specific spectrum bands represent the essential information for this research. Price per megahertz of spectrum per capita
(price/MHz/pop) is a commonly used unit of revenue from a spectrum auction.28 To
a certain extent, this measure allows a comparison of spectrum revenues in different countries, as it accounts for a population size (besides an amount of spectrum) which affects spectrum value for an operator.
This subsection describes how the price index is formed in order to allow for a more exact comparison of band revenues across auctions: First, I explain how and why four band categories are specified. Thereafter, the focus moves to how the revenues generated in auctions where amounts were paid for whole packages of frequencies are handled. The next step presents how the annual fees were added to the revenues and how the revenues were corrected for the length of a license validity. Finally, the normalization of revenues within each category of bands is carried out (described in the following subsection).
5.2.1 Categories of bands
The different qualities of spectrum predispose different bands for different tasks and their values vary too—it is important to account for this when making a price com-parison featuring different bands. The necessity to create different categories is given by the fact that the auctioned spectrum does not have equal intrinsic value. The
is true. For instance, in another experiment with volunteers acting as bidders, Damic et al. (2012) found that the SMRA type is outperformed by CCA in several categories, including the revenues.
26The OLS estimates from the research data do not allow rejecting the hypothesis that there
is no difference between SMRA and CCA auctions (p-value = 0.571; clock auction and CFPSB counted as CCA, while tender observation was not included).
27As opposed to the revenues data described in the next subsection. 28See, for example, Bulow et al. (2009)
value of spectrum depends on its wavelength29: Frequencies with longer wavelengths
travel further and have better capability to penetrate through walls of buildings. On the other hand, frequencies with shorter wavelengths are more efficient at data transfer; this ability has been augmented in spectrum auctions by the fact that these bands could more readily be obtained in consecutive slots (as there were many), thereby improving the capacity of the spectrum.
In most of the auctions, such bands were sold that could be sorted into several groups: 800 MHz and 900 MHz frequencies, bands in the range from 1800 MHz to 2100 MHz and, finally, 2500 MHz and 2600 MHz frequencies. Some of the bands in the 1800/2100 MHz and 2600 MHz groups were offered as single spectrum (TDD) or paired spectrum (FDD). Unfortunately, it was not possible to estimate the revenue of TDD bands in the 1800/2100 MHz group as there were not enough suitable
obser-vations.30 The spectrum was divided into these sections arbitrarily as a result of the
compromise between the necessity to recognize the qualities of different frequencies and to have a sufficient number of observations. Therefore, the wider the range of
a group, the fewer times the bands within a group were auctioned.31 As a result,
there are four groups of bands in the present analysis labeled: 800 FDD, 1800/2100 FDD, 2600 FDD and 2600 TDD.
5.2.2 Dealing with the combinatorial auctions
For some of the auctions, the identification of a price paid for a specific band is
not straightforward. There are 10 auctions32in the present sample where the bands
were sold as a package (usually in a combinatorial auction). In such an instance, a price is known only of entire batch purchased by an operator.
For this reason, the ratio of values of different bands were estimated based on the auctions where prices were band-specific. I use the band-specific prices from the
competitively33 auctioned bands of my auctions sample.34 There are nine
observa-tions of the 800 FDD group, five observaobserva-tions of the 1800/2100 FDD group, eight observations of the 2600 FDD group and five observations of the 2600 TDD group. The values arrived at by calculating a weighted average for each group (using the amount of MHz as a weight) were subsequently divided by the average of the least valuable group (2600 TDD), which serves as a “unit of value” to which it is possible
to relate all the groups.35
29Longer wavelength translates to smaller frequency, i.e. less hertz.
30There were only two cases in the current sample of these frequencies being sold; this type
of band was not in demand and went unsold in several auctions—Netherlands 4/2010, Portugal 12/2011 and Switzerland 2/2012.
31As is evident from the name of the category, 1800/2100 FDD has the widest range and its
specific bands were auctioned most infrequently.
32Denmark 5/2010, Austria 9/2010, Sweden 3/2011, Switzerland 2/2012, Ireland 11/2012,
Netherlands 12/2012, Austria 10/2013, Norway 12/2013, United Kingdom 3/2013, Czech Republic 11/2013
33That is, those auctioned bands, where the reserve prices are below the prices paid for it. 34Germany 2010, Sweden 2008, Sweden 3/2011, Sweden 10/2011, Finland 2009, Finland 2013,
Czech Republic 2013, France 9/2011, France 12/2011, Italy 2011, Spain 2011, Denmark 2012, Netherlands 2010, Belgium 2011. (The France 9/2011 auction is not used for the primary analysis because it was not possible to establish its “spectrum caps status”).
The resulting value ratio of the four groups is roughly: 15 to 3 to 1.5 to 1 (going
from 800 FDD to 2600 TDD).36 For example, my estimates say that 5 MHz of the
1800/2100 FDD group has a similar value as a megahertz of the 800 FDD group, or that a megahertz of the 2600 FDD group is valued 50% more than its TDD counterpart. These ratios, although for slightly different groups, are similar to those estimated in the report by PA Consulting Group (2010), which, interestingly, was a result of an analysis based on the costs of rolling out a network as opposed to the revenue- or demand-based approach that I follow.
The value ratios alone cannot serve directly as the weight for estimating band prices from a package price. In addition to the different value, the weight has to account for different length of licenses of different bands within one package as well as amounts of megahertz in each group. The weight used in this study is, therefore, calculated as a product of the duration of license (rounded to complete years), the amount of MHz and a value ratio.
5.2.3 Dealing with annual fees
In several auctions, winners were required to pay annually a fee for the use of spectrum. It is necessary to account for this too and, therefore, where relevant,
I add a net present value of the fees to previously established band revenues.37
The net present values of annual fees were calculated using the standard formula for calculation of the present value of annuity: N etP resentV alue = AnnualF ee ∗
1−(1+i)−n
i , where n is the duration of a license in years and i is an interest rate. Here
and throughout this thesis, the weighted average cost of capital of 8.86% is used as an equivalent of an interest rate, the value used in the report for Ofcom by Aetha &
DotEcon (2012).38Moreover, here and throughout my analysis, whenever it becomes
necessary to change the currency to Euro, an exchange rate is applied as stated by
the European Central Bank for the date when an auction ended.39
5.2.4 Duration adjustment and finalizing the price index
After having accounted for annual fees where necessary, one needs to adjust license revenues according to their duration. I proceed here in similar fashion to Aetha
& DotEcon (2012).40 The length of license duration taken as a benchmark is 20
years and all the licenses with other durations are re-calculated as if they were
in detail in Subsections 5.2.3 and 5.2.4.
36For the case where a package consisted of the band of 1800/2100 TDD too, which was not
analyzed due to insufficient number of observations, its value was estimated as being 0.2 or 1/5 of the value of 2600 TDD. This prevents its price being assigned to other bands.
37In case an annual fee was to be paid for used spectrum only, I assume usage of all the spectrum
that was auctioned.
38According to the report, this value was used by Ofcom for its termination caps adjustment as
well.
39Available online on https://www.ecb.europa.eu/stats/exchange/eurofxref/html/index.en.html 40The slight difference is in the fact that Aetha & DotEcon (2012) calculate a net present value
as if profit cash flows were coming from the beginning of license validity—that is, they add up the cash flows from period 0 to period n − 1; the present thesis, on the other hand, assumes profits at the end of a year and, therefore, sums up cash flows from period 1 to period n.
lasting 20 years. For this, the net present valuation of annuity is used once again
and an equal size of yearly cash flows is assumed.41 The formula is the following:
AdjustedRevenue = Revenue ∗1−(1+i)1−(1+i)−20−n, where n is the duration of license in years
and i is an interest rate.
The final step to obtain the price index is to divide band revenues by a number
of MHz within a band42 and by the population size of a country in an auction year.
The outcome of these and all other procedures described above in this section are the prices for four different groups of bands comparable across countries. They are summarized in Table 2 and presented as graphs in Figures 1 to 4.
Table 2: Descriptive statistics of revenues per megahertz per capita in EUR Band N Mean Std. Dev. Min Max
800 FDD 16 0.670 0.362 0.311 1.551 1800/2100 FDD 10 0.163 0.110 0.031 0.327 2600 FDD 13 0.055 0.053 0.001 0.177 2600 TDD 11 0.047 0.035 0.005 0.109 Sources: Telecommunications regulators (revenues, annual fees), European Central Bank (exchange rates), World Bank (population size)
5.3
Normalizing revenues across spectrum bands
In order to use all 50 observations in a single statistical analysis, the revenues cal-culated in the previous section were rescaled using one scale for all bands, which permits a comparison of the revenues across bands. The revenues were transformed in two ways: first, turning the revenues within a band into a standard score, which I call standardization, and second, by compressing revenues within a band on a scale
from zero to one, which I call normalization.43
Rescaling revenues to a common scale in each of the four bands filters out the part of the price given by a category of spectrum (e.g. 800 MHz is generally more valuable than 1800 MHz). This allows for examination of revenues from the viewpoint of their distribution within a category and their position, relative to other prices of the same category. Based on this relative position, one can compare, for example, a megahertz of 800 FDD and a megahertz of 2600 TDD, and identify which of the two is relatively more expensive.
Using two kinds of scales for rescaling results in having two dependent variables adds to the model’s robustness. The two processes have in common that resulting
41That is, at the assumption of equal annual profits coming from a license—e.g. a 20-year
license is less than twice as valuable as a 10-year one, because each latter cash flow has a lower net present value.
42To have, at most, one price per band category in a country, I averaged the prices of bands
belonging to one group in a country with the number of MHz as weights.
43This nomenclature was chosen to facilitate referring to the two processes within this text.
In practice, the two names (standardization and normalization) may refer to a range of ways of rescaling variables.
€ -€ 0,200 € 0,400 € 0,600 € 0,800 € 1,000 € 1,200 € 1,400 € 1,600 € 1,800
Figure 1: Price/MHz/pop of group 800 FDD
€ -€ 0,050 € 0,100 € 0,150 € 0,200 € 0,250 € 0,300 € 0,350
Austria Netherlands Italy Norway Switzerland Ireland Sweden Germany Czech Rep. Portugal
€ -€ 0,020 € 0,040 € 0,060 € 0,080 € 0,100 € 0,120 € 0,140 € 0,160 € 0,180 € 0,200
Figure 3: Price/MHz/pop of group 2600 FDD
€ -€ 0,020 € 0,040 € 0,060 € 0,080 € 0,100 € 0,120
Netherlands Denmark Belgium Switzerland Sweden UK Germany Austria Portugal Finland
values indicate where a price stands, relative to other prices in the same group of bands across the countries within the data sample.
Standardized value shows by how many standard deviations a price is distanced from an average of a category (e.g. Gujarati (2004, p. 173)).
Standp = (P − Avgp)/Sdp (1)
where Standp is the standardized variable price/MHz/pop, P is the price/MHz/pop variable, Avgp is an average of the price/MHz/pop for relevant group of bands and Sdp is a standard deviation of the price/MHz/pop variable in that group.
Normalization sets a distance between maximum and minimum prices of a cate-gory to one; the normalized value indicates where a price lies on an axis from zero to one, where zero is the minimum of a category and one is the maximum of a category (e.g. Aksoy & Haralick (2001)).
N ormp = (P − M inp)/(M axp − M inp) (2)
where N ormp is the rescaled variable price/MHz/pop, P is the price/MHz/pop variable, M inp is the smallest price/MHz/pop in a given category of frequencies, and M axp is the largest one.
5.4
Independent variable
Table 3: Auctioned bands with incumbent restriction
Country Date Band Set-aside Spectrum Caps Entrant Winners Netherlands 4/2010 2600 FDD Yes No 2 Belgium 11/2011 2600 FDD No Yes 0 Portugal 12/2011 1800/2100 FDD No Yes 0 Portugal 12/2011 2600 TDD No Yes 0 Netherlands 12/2012 800 FDD Yes No 1 Austria 10/2013 800 FDD Yes No 0 Czech Republic 11/2013 800 FDD Yes No 0 Czech Republic 11/2013 2600 FDD No Yes 0 Norway 6/2013 800 FDD No Yes 1 Norway 6/2013 1800/2100 FDD No Yes 1 Sources: Telecommunications regulators; Ofcom (2015)
To test the hypothesis of no effect of incumbent restriction on revenue, the indi-cator of presence of either a set-aside or spectrum caps, or both, are employed. The incumbent restriction is present in 10 out of 50 bands auctioned, whereas, in four cases, this is due to a set-aside, and in six cases, due to spectrum caps. In four out
of these cases, an entrant won a incumbent restricted band (see Table 3).44
44In Belgium 2011, there was incumbent restriction and an entry too; however, the entrant won
Besides the primary explanatory variable, I also use a variable measuring the GDP in the purchasing power standards (explained in the beginning of this section) of a country in an auction year (or the latest known). This serves to control for pos-sible bias caused by different strengths of demand for operators’ services in different countries.
5.5
First check of the hypothesis
Looking closer at the four cases where incumbent restriction and an entry were simultaneously present (Table 4), we see that the relevant revenues vary in size. Two of these are below the averages of their respective bands’ revenues and one of them is the smallest of revenues in its category. On the other hand, the other two revenues are above their categories’ averages and one of them is even the second highest in its category. The effect of incumbent restriction seems to be ambiguous or not present at first sight, which is in line with the null hypothesis.
Table 4: Revenues for bands where entrants won incumbent restricted spectrum
Country Date Band Entrant Winners Standp Normp Pctmaxp Netherlands 4/2010 2600 FDD 2 -1.02 0.00 0.7 Netherlands 12/2012 800 FDD 1 2.19 0.93 94.2 Norway 6/2013 800 FDD 1 -0.41 0.17 33.5 Norway 6/2013 1800/2100 FDD 1 0.51 0.64 67.1 Notes: Standp—standardized revenues by band category; Normp—normalized re-venues by band category; Pctmaxp—percentage of the highest revenue within a band
6
Model and regression results
6.1
Model
The hypothesis was tested with regard to the effect of incumbent restriction on auction revenue by estimating the following ordinary least squares model:
Revenueik = β0+ β1IncumbentRestrictionik+ β2GDPk+ uik (3)
Where:
Revenueik Dependent variable; in each of the model’s variations, it represents one
of two variables: Standp and N ormp for band i, in auction k
IncumbentRestrictionik Primary explanatory variable; a dummy variable that
equals one if there is a set-aside and/or spectrum caps in band i, in auction k
GDPk Control variable; gross domestic product per capita in PPS in a country
β1 Intercept coefficient
β1 Coefficient of interest that indicates the size of difference in revenue between the
means of observations with incumbent restriction and those without
β2 Coefficient indicating the size and direction of impact of GDP on the dependent
variable
uk Error term45; assumption: uk is independently and identically distributed as
∼ N (0, σ2
u)
The model specifications had to be varied as several bands within the sample
were sold at the level of reserve price,46 from which it can be inferred that there was
no competitive bidding for such bands. Therefore, two of my models use only the 44 observations where the price was higher than the reserve level. Thus, there are four specifications: one with 50 and one with 44 observations for each of the two dependent variables.
6.2
Results
Based on the results of my estimation (see Table 5), the null hypothesis about the incumbent restriction having no effect on the revenues cannot be rejected at reasonable levels of significance (5% or 10%). In other words, the results indicate that there is no correlation between the presence of a set-aside or spectrum caps within a band on the one hand, and the size of the revenue for the band, on the other hand, ceteris paribus.
Table 5: Ordinary least squares estimates of incumbent restriction effect on revenue Standardized Revenue Normalized Revenue
(1) (2) (3) (4) Incumbent Restriction 0.036 0.305 -0.021 0.030
(0.923) (0.516) (0.830) (0.815) GDP 4.47e-05 2.79e-05 1.60e-05 1.21e-05
(0.050)* (0.216) (0.029)** (0.108) Constant -1.403 -0.861 -0.145 -0.014 (0.042)** (0.243) (0.477) (0.953) R2 0.100 0.055 0.118 0.059 N 50 44 50 44 P-values in parentheses; *** p<0.01, ** p<0.05, * p<0.1
45The regression errors are clustered at the auction level because the band revenues within an
auction might be correlated.
46These are Portugal 12/2011 (all four categories), the Czech Republic 11/2013 (2600 FDD)
The model does not test for spillover bidding of entrants and incumbents on the other bands within an auction. Despite the low number of the auction-level obser-vations, it is still of interest to test for it on this level: The estimated significance,
however, is still weak (P-value: 0.320 [n=23]; 0.238 [n=21]47). The estimates
indi-cate that the revenues of the other, non-designated bands within an auction with the incumbent restriction were not affected.
There could be several reasons for these results. One of the possible causes could be that set-asides and spectrum caps do not prove to be strong enough incentives for entrants to participate in an auction. Indeed, the study sample comprised eight entrants and 10 bands with incumbent restriction; however, only in four cases did an entrant win a band that is a part of incumbent-restricted spectrum (see Table 4).
To find out whether set-asides and spectrum caps are weak incentives for en-trant participation in an auction, the hypotheses were tested to determine whether incumbent restriction does not increase the probability of entrants’ participation in an auction (one specification), nor the probability of them winning spectrum (another two specifications for the band and auction levels). The Probit model es-timates show that the effect is significant (see Table 6), and incumbent restriction increases the chances of entrants’ participation (or winning spectrum) by roughly 27–46 percentage points. Although this seems to be a large effect, the chance of en-trants’ participation (or winning) in an auction with set-asides and spectrum caps is still only roughly 40–66%. This is quite a low probability and it confirms the assumption that incumbent restriction is generally not a strong enough incentive for entrant companies to enter an auction.
One of the reasons for the zero effect48 proposed in the previous research is that
revenue could be pressed downwards because incumbents (with high valuation) can-not bid on reserved spectrum. Nevertheless, on looking closer at the seven auctions where an entry occurs, it becomes clear that this is not the case—entrants gener-ally pay similar prices as incumbents. This is partly because they bid on and win the non-designated parts of spectrum too. Surprisingly enough, the entrants paid
competitive prices even for incumbent restricted spectrum.49 An OLS regression of
4721 auction-level observations are the equivalent to 44 band-level observations—in other words,
observations filtered out of those where the revenues were at the level of a reserve price.
48In this chapter, I explain possible reasons for “no effect”. These are the factors affecting the
revenues downwards. But in this context they are considered as weakening the positive effect and, therefore, I refer to these factors as possible reasons for the “zero effect”.
49The Netherlands 4/2010 auction achieved generally low revenues, so that it was not too costly
for the entrants to pay similar prices as the incumbents. In the Netherlands 12/2012 auction, even though the entrant paid a price slightly below the mean (800FDD), this was more than compensated for by the incumbents. In the Norway 6/2013 auction, the entrant paid more than some of the incumbents. In other auctions, the entrants won spectrum for which incumbents were able to bid; therefore, it is not surprising that the prices paid did not deviate from the average of an auction: In the UK 3/2013 auction, the entrant paid a price above the average for given bands in the auction. In Sweden 5/2008, the entrant paid roughly 90% of the average price. In Finland 11/2009, the price paid by the entrant did not differ from that of the incumbents either, even though the revenues were some of the lowest from all of the auctions in my dataset. Lastly, in the Belgium 11/2011 auction, the entrant paid a roughly similar average price for 2600 TDD as incumbents paid for 2600 FDD—considering the higher valuation of FDD spectrum, the entrant paid more than the incumbents.
revenues on the “entrant winners” dummy on the band level does not show any effect (P-value 0.553 [n=50]; P-value 0.532 [n=44]). Moreover, regression of reve-nues on the “entrant participation” dummy in an auction level does not show any effect either (P-value 0.999 [N=23]; P-value 0.945 [N=21])—this implies that the participation of entrants (including those that do not win spectrum) in an auction does not affect the revenues.
Another potential reason why the analysis indicates a zero effect of set-asides and spectrum caps on revenues might be the possible endogenous choice of governments to employ them. Governments might only impose spectrum caps and set-asides if they expect them to increase revenue or facilitate competition in the market for mobile telecommunications. As a consequence, the observed effect might be zero on average across auctions, despite being positive or negative in some specific instances. Lastly, a possible reason for the absence of any effect on the revenues could be that the incumbents’ demand for spectrum might either have been satisfied by other bands or they might have already had the same band in their possession—in such a scenario, the decreased supply of spectrum (in a band with restriction) to incumbents would not have an effect on competition and revenues.
To sum up, the most important implications of the analysis are that:
there is no effect of set-asides and spectrum caps on revenues on the band level, and, probably, neither on the auction level
the probable reason is that the incumbent restriction does not prove to be strong incentive for entrants to participate in an auction
the incumbent restriction does not play an important role in an entrant’s decision to participate in an auction—in many instances, entrants bid in the auctions without any reserved spectrum
7
Conclusion
The thesis explores the effect of restricting incumbents in an auction by set-asides and spectrum caps on revenue paid for the spectrum bands by mobile operators. The data sample comes from the LTE auctions run from 2008 to 2013 in countries with similar market and competition policy backgrounds (mostly in western Eu-rope). The OLS estimates do not allow rejecting the hypothesis that set-asides or spectrum caps have no effect on auction revenues. As the data analysis implies, this is probably because incumbent restriction might not be a decisive factor for entrants to participate in an auction. On the other hand, the results also indicate that the revenues are not smaller in those auctions where set-asides and spectrum caps are used. As such, it can serve as a cheap way to increase the attractiveness of an auction for entrants (at least to some extent). The results can be considered as a contribution to the ongoing discussion about the effect of auction design tools on auction revenue, as well as a review of the recent LTE auctions in Europe from the perspective of their revenues.
Nevertheless, the analysis has some weaknesses as well. It does not control for the incumbents’ previously owned bands; therefore, it is not clear whether their
demand in a specific band is strong or weak, which would be one of the channels through which revenues can be affected by the aforementioned design tools. This would help answer why set-asides and spectrum caps do not lead to higher revenues. Moreover, my analysis of the possible spillover bidding on the non-designated bands might give only weak implications, as it depends on a few observations due to the
requirement of using auction-level observations. The estimated effects could be
also biased by the possible existence of endogeneity in the choice of governments whether to restrict incumbents or not. Besides this, a large part of the data on revenues consists of estimated values (those in the combinatorial auctions). This might distort the estimates, as the ratio between the specific unitary band prices is the same in all of these bands.
I envision the following avenues for future research on this subject: The research will have to resolve the issue of the package prices, as combinatorial auctions are seeming to grow in popularity. However, further studies can benefit from more observations (from single economic area) in order to find more reliable estimates by analyzing a sample on the auction level instead of the band level. Such research can give more reliable answers to whether the spillover effect of set-asides and spectrum caps works or not. Similarly, future research could be improved by including the control for bands owned by incumbents before an auction and, possibly, by an estimation of their demand.
Table 6: Probit model estimates of incumbent restriction effect on dummy dependent variables
Non-inc. Bidder(s) (auction level) Non-inc. Winner(s) (auction level) Non-inc. Winner(s) (band level) (1) (2) (3) (4) (5) (6) IncRestr(auction) 0.855 1.054 1.067 1.272 (0.145) (0.096)* (0.075)* (0.049)** IncRestr (band) 0.897 1.282 (0.059)* (0.018)** Constant -0.674 -0.623 -0.887 -0.842 -1.150 -1.102 (0.048)** (0.073)* (0.014)** (0.023)** (0.000)*** (0.000)*** P r(Y = 1|X = 0) 25% 26.7% 18.8% 20% 12.5% 13.5% P r(Y = 1|X = 1) 57.1% 66.7% 57.1% 66.7% 40% 57.1% Marginal effect 32.1pp 40pp 38.4pp 46.7pp 27.5pp 43.6pp (0.137) (0.074)* (0.069)* (0.033)** (0.093)* (0.025)** Pseudo R2 0.073 0.103 0.116 0.153 0.075 0.128 Correctly classified 69.6% 71.4% 73.9% 76.2% 82% 81.8% N 23 21 23 21 50 44 P-values in parentheses; *** p<0.01, ** p<0.05, * p<0.1
Bibliography
[1] Aetha & DotEcon, ‘Spectrum value of 800MHz, 1800MHz and
2.6GHz.’Technical report, DotEcon Ltd, 2012.
[2] Aksoy, Selim & Robert M Haralick, ‘Feature normalization and
likelihood-based similarity measures for image retrieval.’ Pattern Recognition Letters, 22 (5), pp. 563–582, 2001.
[3] Ausubel, Lawrence M, ‘An efficient ascending-bid auction for multiple
objects.’ American Economic Review, pp. 1452–1475, 2004.
[4] , Peter Cramton, R Preston McAfee, & John McMillan,
‘Synergies in wireless telephony: Evidence from the broadband PCS auctions.’ Journal of Economics & Management Strategy, 6 (3), pp. 497–527, 1997.
[5] Ayres, Ian & Peter Cramton, ‘Deficit reduction through diversity: How
affirmative action at the FCC increased auction competition.’ Stanford Law Review, pp. 761–815, 1996.
[6] Bichler, Martin, Pasha Shabalin, & J¨urgen Wolf, ‘Efficiency,
auc-tioneer revenue, and bidding behavior in the combinatorial clock auction.’ In Auctions, Market Mechanisms, and Their Applications: Springer, pp. 38–39, 2012.
[7] Binmore, Ken & Paul Klemperer, ‘The biggest auction ever: the sale
of the British 3G telecom licences.’ The Economic Journal, 112 (478), pp. C74–C96, 2002.
[8] Bulow, Jeremy, Jonathan Levin, & Paul Milgrom, ‘Winning play in
spectrum auctions.’Technical report, National Bureau of Economic Research, 2009.
[9] Cisco, ‘Cisco Visual Networking Index: Global Mobile Data Traffic Forecast
Update, 2013-2018.’Technical report, Cisco, 2014.
[10] Cramton, Peter, Evan Kwerel, Gregory Rosston, & Andrzej
Skrzypacz, ‘Using spectrum auctions to enhance competition in wireless services.’ Journal of Law and Economics, 54 (4), pp. S167–S188, 2011.
[11] Damic, Lina, Nenad Peteh, & Ana Petric, ‘An experimental analysis
of Simultaneous Multiple Round Auction formats and Combinatorial Clock Auction formats.’ In Software, Telecommunications and Computer Networks (SoftCOM), 2012 20th International Conference on, pp. 1–9, IEEE, 2012.
[12] Gilbert, Richard J & David MG Newbery, ‘Preemptive patenting and
the persistence of monopoly.’ The American Economic Review, pp. 514–526, 1982.
[14] Hoppe, Heidrun C, Philippe Jehiel, & Benny Moldovanu, ‘License auctions and market structure.’ Journal of Economics & Management Strat-egy, 15 (2), pp. 371–396, 2006.
[15] Katz, Michael L., ‘An Economic Analysis of Auction Set-Asides.’ May,
2012, Paper was commissioned by Verizon Communications Inc. The views expressed are those of the author and do not necessarily reflect the views of Verizon.
[16] Kim, Junwhan, Achla Marathe, Guanhong Pei, Sudip Saha,
BS Subbiah, & Anil Kumar S Vullikanti, ‘Analysis of policy instru-ments for enhanced competition in spectrum auction.’ In Dynamic Spectrum Access Networks (DYSPAN), 2012 IEEE International Symposium on, pp. 89–96, IEEE, 2012.
[17] Klemperer, Paul, ‘Auctions with almost common values: TheWallet
Game’and its applications.’ European Economic Review, 42 (3), pp. 757–769, 1998.
[18] , ‘What really matters in auction design.’ The Journal of Economic
Perspectives, 16 (1), pp. 169–189, 2002.
[19] Matthews, Steven A, A technical primer on auction theory I: Independent
private values: Center for Mathematical Studies in Economics and Manage-ment Science, Northwestern University, 1995.
[20] McGrattan, Ellen R, ‘The macroeconomic effects of distortionary
taxa-tion.’ Journal of Monetary Economics, 33 (3), pp. 573–601, 1994.
[21] McMillan, John, ‘Selling spectrum rights.’ The Journal of Economic
Per-spectives, pp. 145–162, 1994.
[22] , ‘Why auction the spectrum?’ Telecommunications policy, 19 (3), pp.
191–199, 1995.
[23] Meng, Xin, ‘The Effect of Set-Aside Auctions.’ In Advances in Applied
Eco-nomics, Business and Development: Springer, pp. 294–299, 2011.
[24] Nakabayashi, Jun, ‘Small business set-asides in procurement auctions: an
empirical analysis.’ Typescript, Department of Economics, The Ohio State University, 2009.
[25] Ofcom, ‘Annual licence fees for 900 MHz and 1800 MHz spectrum,
An-nex 8, Recent European Awards.’ http://stakeholders.ofcom.org.uk/ binaries/consultations/annual-licence-fees-900-MHz-1800-MHz/ annexes/Annex_8.pdf, 2015, Accessed: 2015-02-27.
[26] PA Consulting Group, ‘Study on comparability of frequency bands in
[27] Rothkopf, Michael H, Ronald M Harstad, & Yuhong Fu, ‘Is sub-sidizing inefficient bidders actually costly?’ Management Science, 49 (1), pp. 71–84, 2003.
