Spectrum auction design: Simple

auctions for complex sales

Martin Bichler, Jacob Goeree, Stefan Mayer, Pasha
Shabalin

Aditya Chhabra, Ashwina Jha, Sumit Chaudhary

Delhi School of Economics

May 13, 2020

Introduction
• This paper focuses on auctions as a means of allocating radio
spectrum and the effect of bid language and payment rule on the
efficiency of auction and its revenue.
• Bichler et al construct a lab experiment with treatment variations in
simple versus complex bid language and simple versus complex
payment rule, for both ascending and sealed-bid formats to measure
how auction revenue varies.
• We then go on to evaluate the auctioning mechanism in place in India
and the issues around it
What is spectrum
• Communication airwaves, also known as radio frequency spectrum are an
important resource for mobile communication technologies.
• However, radio spectrum is not uniformly applicable, physical and natural
conditions can constrain its application to some technologies. To optimize its
use radio spectrum is divided into bands of varying frequencies
• Usually, low frequency spectrum is preferred for better propagation
characteristics, while high frequency spectrum is deployed to push greater
volumes of information in each frequency band
• Need to license the use of spectrum - While spectrum is not depletable,
technology can limit the extent to which it can be beneficially utilized at a
point in time.
Ways of allocation
• The three basic models for regulating spectrum are
• (i) a command and control model (e.g. assignment of bands for public
service use)
• (ii) a market-oriented model (i.e. through licensed auctions)
• (iii) a generic licensing or common use model (i.e. any user can access
the band provided that and for as long as the user complies with the
technical specifications set out in the generic license).
Command and control
• Under command and control, the government allocates spectrum
based on network, rollout, coverage and technology through what has
come to be called a “beauty contest”
• In several countries this model was deployed in the early stages of
telecom liberalization and was considered the norm until the success
of the 1994 PCS Auction by the FCC
• Pros - financially strong bidders might have advantages over weaker
bidders in an auction
• Cons – inefficient and monitors use closely, subjective and opaque
Common use model or Spectrum sharing
• The common use model completely liberalises the use of spectrum.
• Multiple users are allowed to share access to a single frequency band.
• This may be more technologically and economically efficient and the
world is moving towards it.

• Example - Licensed shared access (LSA) and authorized shared access (ASA) illustrate the concept. Such
sharing allows licensed services to share spectrum in a band with new users without disrupting existing
users, while still increasing the amount of spectrum available for other users. The new form of licensing is
under formal review in the United States and European Union, and will be addressed by the International
Telecommunications Union.
Auction
SMRA(Simultaneous Multiple Round
Ascending Auction)

• The PCS spectrum in 1994 was sold via a simultaneous multi-round auction,
considered a success which raised 600 million dollar revenue.
• It auctions multiple licenses for sale in parallel and uses simple activity rules
which force bidders to be active from the start.
• Exposure risk - This exposes bidders to risk when they bid aggressively for a
desired combination but end up winning an inferior subset at higher prices -
bidders may act cautiously with adverse effects for revenue and efficiency.
• When bidders rationally anticipate this exposure problem, competition will
be suppressed with adverse consequences for the auction’s performance.
Illustration of exposure risk
• In a scenario where circles A and B are complements and only stand-
alone bids are allowed, a bidder may overbid for A in the fear of being
outbid for B.
• In India – During the February 2014 auctions, actual revenue raised
was 41 per cent higher than what was estimated by the auctioneer;
thus suggesting some degree of over-bidding. Typically, in subsequent
rounds of auction, a player’s willingness to participate may get
overpowered by his fear of being subjected to a winner’s curse, leading
to lower levels of participation and thus inefficient trade. In this
context, packaged bidding as used in combinatorial auctions could be
an alternative.
CCA(Combinatorial Clock Auctions)
• The Combinatorial Clock Auctions (CCA) format was proposed by Ausubel, Cramton and
Milgrom (2006) to overcome the shortcomings of the SMRA. bidders can demand
combinations of licenses as well as individual licenses
• Bid on generic lots of spectrum rather than individual lots and the band plan for any
frequency is determined by the synergies and the presence of complementarities among
the components on offer. This overcomes the winner’s curse.
• Since 2008, the combinatorial clock auction (CCA) has been used by regulators in various
countries such as the Austria, Australia, Canada, Denmark, Ireland, the Netherlands, and
Switzerland to sell spectrum.
• Combinatorial auctions allow for bids on combinations of licenses and thus hold the
promise of improved performance. Recently, a number of countries worldwide have
switched to the combinatorial clock auction to sell spectrum. This two-stage auction uses
a core-selecting payment rule.
CCA Cons
• High Communication complexity –
This two-stage auction uses a core-selecting payment rule. The number of possible
packages a bidder can submit grows exponentially with the number of licenses, which
adds complexity to the auction. For larger auctions with dozens of licenses bidders
cannot be expected to reveal all their valuations during such an auction.
• “Core-payment rule”-
This payment rule is sufficiently complex that it generally does not allow for a game-
theoretic analysis and its outcomes can appear non-transparent as small changes in
the package bids selected by the bidders can lead to substantial variations in the
payments Moreover, the payments are not known until after the auction, which
precludes bidders from reporting to management about the progress of the auction
and about expected payments. These issues do not arise with a simple pay- as-bid
payment rule as used,for instance,in the Romanian spectrum auction in 2012.
Reason for core-payment rule

• Inefficiency of VCG(Vickrey-Clarke-Groves) Mechanism –

Also referred to as a second price auction, it gives the bidder discount equalling their
marginal contribution to the overall revenue which can lead to outcomes where the
winners pay less then what losing bidders are willing to pay with their bids.
• Avoiding uncompetitively low revenue as is possible in a VCG mechanism was one of
the original design goals of core- selecting payment rules
• Example of Inefficiency - Examination of bidding behavior in CCA reflects a striking degree of heterogeneity among
participants. For example, in the Canadian auction of 700 MHz in 2014, of the three most active bidders, two (Bell
Canada and Telus) submitted final bids for a large number of different license packages (close to 500) at essentially
the maximum amount allowed by the activity rules, whereas the third (Rogers), which ended up paying much more
for the licenses it won, submitted only a single final round bid, with which it increased its bid for its winning
package. While CCA is arguably better than the simultaneous ascending auctions, it is complex and requires a high-
level of bidder sophistication. It can also result in widely varying prices depending on the strategies adopted by
bidders.
CCA Cons (cont)
• Revenue is important but is rarely a design goal. Efficiency cannot be
analysed where bidder’s valuations stay private. Additional design
goals – transparency and law of one price. Trade-off between this and
efficiency and revenue. Both, CCA and VCG don’t follow law of one
price.
• Example – In Switzerland, one of the bidders had to pay almost 482
million Swiss Francs and another bidder close to 360 million Francs
although they won a similar set of licenses
Languages
Atomic bids
• Also called Single Minded Bids.
• (S,P) says that the bidder chooses S, a subset of available items, for
price P: For all S′ ̸= S, v(S′) = 0. v(S) = P
• Obviously, atomic bids are not expressive enough to express majority
of symmetric/asymmetric valuation patterns, e.g. they can not specify
“simple additive valuation”.
• An additive valuation: Value of item j = v and v(S) = Σj∈Sv .
•   OR bids
• Bidder chooses multiple bids.
S = (S1,p1) OR (S2,p2) OR ... OR (Sk,pk)
• The valuation of bids for a certain bidder consists of only disjoint bids.
• v(S)=Max( pi), Si,Sj ∈ W⇒Si∩Sj=∅
• The real problem with OR bids is that they don’t support
substitutability.
• Also, It is not to possible to express simple unit demand valuation.
• The simple unit demand valuation: v(S) = 1 for all S ̸= ∅.
Consider the following OR bid:
• S = {({1}, 5 ) OR ({2}, 4 ) OR ({1,2}, 7 ))}
• The bidder is interested in having both items 1 and 2, only if she Pays
7. (Sub-additive- hence substitutes)
• We can’t express such constraints with OR bids.
• XOR is expressive enough to express all valuation variations. Given a
XOR bid:
• S = (S1,p1) XOR (S2,p2)XOR ...XOR (Sk,pk)
• While XOR is more expressive than OR, there are valuations that can
be specified more succinctly by OR.
• Consider the simple additive valuation. OR can specify that in size m
while XOR require 2m clauses.
Alternate language
OR-of-XORs
• Bids consist of clauses that entirely consist of XOR bids and such
clauses are connected by ORs.
(...XOR...)OR(...XOR...)OR ... OR(...XOR...)
• In the experiments (later) bids could be submitted on 2, 4, and 6 lots
only in each of the bands and at most one of the bids within a band
could become winning.
• However, a bidder could win multiple bids in different bands, i.e., we
use an OR bid language across bands
 OR∗ Bids
• The idea is to express XOR with OR.
• This can be achieved by using dummy items.
• Dummy items don’t have any values but can constrain bids to represent XOR.
(S1, p1) XOR (S2, p2)
or
(S1 ∪ {dummy}, p1) OR (S2 ∪ {dummy}, p2)
• This language can simulate all bidding languages discussed so far.
• (Nisan 2000) Any valuation that can be represented by OR/XOR formula of size
s, can be represented by OR* bids of size s, using at most s2 dummy items.
Experimental design
The value model
• A multi-band value model which has four bands with 6 licenses each.
• Within a band, each individual block has the same value for bidders so that there are
essentially 74 -1=2400 different packages.
• The structure of the value model and the distribution of the block valuations of
• All bands are known to all bidders. In particular, band A is of high value to all bidders
and bands B, C, and D are less valuable. Bidders receive base valuations for items in
each band.
• Base valuations are uniformly distributed: vA was in the range of [100, 300] while vB,
vC, and vD were in the range of [50, 200]. Furthermore, bidders have complementary
valuations for bundles of blocks within bands, but not across bands.
• Synergies - In all bands, bundles of two blocks resulted in a bonus of 60%
on top of the base valuations, while bundles of three or more blocks
resulted in a bonus of 50% for the first three blocks.
• Bid languages - They use an OR-of-XOR bid language, which draws on the
observation that typically there are high synergies among licenses within a
band, but lower synergies across bands.
• no cross-band synergies: The use of OR-of-XOR bid language, which draws
on the observation that typically there are high synergies among licenses
within a band, but lower synergies across bands.
• Goal:
The experiments allow us to estimate the differences in efficiency of a
compact bid language compared to an XOR bid language, which has been
used in spectrum auctions so far.
Treatment structure
• They analyze two variations, simple (S) and complex (C), of the bid
language and payment rule. In particular, we consider a compact bid
language versus a fully expressive bid language, and a pay-as-bid
versus a bidder-optimal core-selecting payment rule. We do so for
both ascending (A) and sealed-bid (SB) auctions.
• The different treatments are denoted as FLP where F ={A, SB} denotes
the format and the subscripts L = {S,C} and P={S,C} indicate the bid
language and payment rule respectively.
• ACC = CCA, and
• SMRA, instead of ACS
Procedures and organization
• They used the same sets of value draws (waves) across treatments to
reduce performance differences due to the random draws.
• Each wave was used to run four different auctions, which combined to
define one session.
• A tool was provided that showed a simple list of available bundles,
which could be sorted by bundle size, bidder individual valuations, or
payoffs based on current prices in the ascending auction formats.
• At the start of each auction, subjects received their individual value
draws, information about the value distributions and their synergies
for certain bundle
Results (Comprising in two parts)
• Efficiency and revenue of the different auction formats
• Individual bidder behaviour
Result 1
• Formats with compact bid language are more efficient than those
with a fully expressive language.
Result 2
• Formats with a pay-as-bid payments rule yields higher revenue than
those with a core-selecting payment rule.
• Impact of big language, payment rule and auction format on revenue
Result 3
• Bidder behaviour in ascending auctions –
Bidders in an ascending auction with a compact bid language select their bundles mainly based
on payoff.
Bidders did not only bid on their highest valued bundles but also on most of the bundles with a
positive payoff
Result 4
• Bidder behaviour in sealed-bid auctions -
Bidders in core selecting sealed-bid auctions with a compact bid
language on all possible bundles.
However, bid sharing was more with the pay-as-bid payment rule
compared to the core-selecting payment rule.
Conclusions
• CCA is being used increasingly worldwide but the problem of missing values
– when bidders bid on a few subsets of packages amongst a huge number of
it letting the value of packages not bid on be 0 - can have adverse effects for
the auction's efficiency and revenue. Caused due to use of fully expressive
XOR language.
• Trade off between SMRA(Exposure risk) and CCA(Communication
complexity).
• Approach a middle-ground solution – OR-of-XOR
• We find that auction revenue is substantially higher with the simpler pay-as-
bid rule. The pay-as-bid rule avoids uncertainty about how much a bidder
has to pay for a bid at the end of an auction, if this bid becomes winning.
India
• 2010 represented a watershed in spectrum management policies in
India. Until then spectrum was administratively assigned by command
and control. In 2010 , for the first time, spectrum in the 2100 MHz and
2300 MHz bands was assigned through an online auction designed to
discover a market price for 3G/ BWA spectrum.
• The underlying basis was to enable efficient use of spectrum, avoid
hoarding, stimulate competition and promote the roll out of 3G and
broadband services. The government also used the auction to raise
revenue. The 2010 auction was successful in achieving these objectives
with all of it sold above the reserve prices determined by the regulator.
Auction outcomes
• Between 2012 and 2016, the Department of Telecommunication (DoT) conducted
several rounds of spectrum auctions using the Simultaneous Multiple Round
Ascending (SMRA) method. The auctions for different circles were conducted
simultaneously in continuous cycles known as clock rounds.
Shares of
spectrum and
revenue
collected
Issues with existing Auction mechanism
• Spectrum acquisition increased costs to operators despite competition.
• Over the six auctions held during the period 2010 to 2016, the government
has auctioned only sub-sets of the total frequencies and the average
reserve price in every subsequent auction has witnessed an upward
revision. The outcomes for most spectrum auctions are discouraging. There
is a lack of enthusiasm among operators due to unrealistic pricing.
• The number of operators has declined from a peak of 12 operators per
circle to an average of 5. In order to ease the pressure on operators the
government is considering moderation of spectrum trading rules and
regulatory charges. The financial stress on individual operators affects the
appetite to invest in upgradation of technology.
In India
• To implement package bidding, the regulator needs to gather information
from bidders on how they value different packages of spectrum and create
bundles accordingly. Each player then bids for these packaged circles. Based
on true revelation of valuation by bidders, this mechanism mitigates
aggregation risks. Additionally, it induces contiguous spectrum outcomes as
aggregated band plans tend to emerge endogenously in the bidding process.
Need for contiguous spectrum has most recently been recognized as a pre-
requisite for the fast emerging data market. It would also lead
to consolidated outcomes, i.e. lesser number of players operating in each
circle, which may help overcome the industry challenges due to spectrum
fragmentation. Packaged bidding would also facilitate procurement
of national licenses and creation of a single pan-India license.
Conclusion
• A simplistic change in design would not guarantee efficiency
alone. Prior information regarding availability of spectrum would be
required to develop and plan robust strategies. In the absence of
which the industry resorts to knee-jerk reactions, creating
inefficiencies. It is vital to have a more dynamic auction
mechanism with timely modifications to the auction rule in
accordance to changes in technology and market conditions. Most
importantly, re-prioritization of objectives with greater focus on
efficiency is imperative to bring about the envisioned change.
References
• WHAT IS SPECTRUM SHARING, AND WHY DOES IT MATTER?By Gary Kim  Posted 
November 16, 2015  In Business Model, Internet Access, Mobile, Spectrum, 
Spectrum Futures Conference
• Evaluating Spectrum Auctions in India, Rajat Kathuria et al,2019,ICRIER
• Spectrum auctions in India: lessons from experience, R.S. Jain,2001, Telecommunications Policy