Spectrum auctions

Introduction

I Cell phone carriers need a license to operate (a license to use

a certain spectrum of radio frequencies).
I Allocating licenses is a complex problem:
I How many licenses?
I How to split the spectrum? (in some cases some frequencies
are better than others)
I What price for a license?
The allocation of spectrum is one the most exciting and succesful
story in auction design.
A very complex problem (much more than a keyword auction):
No ready-to-use theoretical model. . .
I There’s a gap between the theory and the reality
(Some innocuous details may matter.)
I Designing spectrum auctions amounts to travel to the
unknown.

After 10-15 years, the successes and failures gave insight about
what is important (and what is not).
How to allocate spectrum?

A first solution was to run a lottery:

I strong incentives to participate: large number of participants.
I high risk of misallocation: licenses not assigned to the best
carriers, yielding poor service to consumers.

An alternative is to use a beauty contest:

I Public authority spells out the specifications required for the
operators
I Candidate submit a project
I A selection is made by a committee
Issues with beauty contest

Governments usually asks

I guaranteeing lowest prices to consumers.
I How guaranteeing prices for 5–20 years for a products that
don’t yet exist?

I Minimal infrastructure investment

I How to monitor the investments? How to enforce the
commitments made by the operators?
The use of a beauty contest is controversial:
I An auction can extract an information (about how worth are
the licenses) that is unavailable to the government, less so
with a beauty contest.
I Beauty contest can easily lead to favoritism/corruption

I An auction can potentially raise substantial revenues for the

government (by eliciting bidders’ willingness to pay).
I Carriers argue that beauty contest is cheaper and thus it
benefits consumers (less need to charge high prices).
I Flawed argument: The license is a fixed cost, so it does not
affect the profit maximization.
Designing an auction

In a large country like the US it may be desirable to have:

I Several nationwide carriers
I Regional carriers
Regional licenses are not identical (population density, size, etc).
So the problem is to allocate multiple, not identical licenses where
some carriers may acquire several regional licenses.
We would need a combinatorial auction.
I A VCG auction would be the ideal choice. . . in theory.
I We need to design a multi-item auction that is “easy” to run.
Additional objectives

I Efficiency (allocate the licenses to those who value them the

most) and revenue maximization.

I Favor entry (otherwise a game between incumbents with low

competition and thus low revenues).

I Avoid collusion among bidders.

Pitfalls

In a spectrum auction several items are for sale. There are 2 ways:
I combinatorial auction (e.g., VCG)

I multi-ascending auction (ascending auctions running in

parallel).

Choosing either auction format means facing the following

tradeoff:
I Exposure problem

I Winner’s curse
Exposure problem

When several items are for sale, bidders may view them as
I complements (item A is valuable only if I have item B)
Example: A regional operator wants a license for Wisconsin
only if it has the license for Illinois
I substitutable (I want A or B, but not both).
Example: A regional operator wants a license for Virginia or
Louisiana.
I some combination of the two.
Example: A regional operator wants a license for New York
and: either Pennsylvania or Massachusetts (but not the three
states).
Combinatorial vs. Multi-ascending:
I With a combinatorial auction, bidders can express through
their bids more precise valuations over combinations of items.
I With a multi-ascending this is more difficult:
Example:
I A carrier wants one license: Iowa or Minnesota.
I In the Iowa license auction the outside option is the outcome
of the auction for the Minnesota license.
I In the Minnesota license auction the outside option the
outcome of the auction for the Iowa license.
I In a multi-ascending auction the carrier would have to bid
simultaneously for the two licenses with the risk of winning
both.
Winner’s curse

The winner’s curse is the possibility for the winner of an auction to

pay more than what the bidder’s (correct) valuation.
This occurs when bidders only have partial information about their
true valuations.
Example:
I I know that item A is worth between 10 and 30
I A preliminary study suggests my valuation is 25.
I I win and pay 20: Yippee!
I Ex-post, the real value is in fact 15: final payoff negative.
Combinatorial vs. Multi-ascending:
In spectrum auctions valuations are typically interdependent: if a
New York license is highly valuable for AT&T it is also likely to be
highly valuable for T-Mobile, Verizon and Sprint.
I Multi-ascending auction: observing competitors’ bids reveal
information about how much they value the license.
This yields better estimate of the true value of a license,
minimizing the winner’s curse.
I Combinatorial auction: carriers cannot learn bidders’ bids
(and thus adjust their valuation).
Simultaneous ascending auction

A natural generalization of the English auction for many goods,

defined by Paul Milgrom, Robert Wilson and Preston McAfee.
I all items for sale simultaneously

I Each bidder can bid on any of the items

I Bidding continue until no bidder raises the price on any of the

items
I Final allocation:
I A bidder wins an item if she has the highest bid for that item
I Each bidder pays the last posted price for any items won (like
in the English auction).
Advantages
I Bidders can observe the tentative prices.

I Bidders can easily condition subsequent bid based on current

information.
I Price information can help bidders concentrate their valuation
efforts on the relevant price range.
⇒ exposure problem mitigated.
The US ’94 PCS broadband auction (2G)

PCS = Personal Communication Service.

Two competing technologies at that time:
I PCS: operates in the “1900 MHz band” (1850-1990 MHz).
Low power, cells are small (about 1 mile radius).
I Cellular: operates in the 700 MHz band.
High power, cells are big (about 8 miles radius). More
adapted to highly mobile services (typical for cars).
In 1994 the Federal Communication Commission launched an
auction for 99 CPS licenses.
I Auction lasted 3+ months

I $7 billion raised.

I 30 bidders, 18 winners.

Licenses to be sold:
I For 48 regions, 2 types of licenses (A and B), covering
adjacent 30MHz blocks
I For NYC, LA and Washington DC (only licenses of type B for
sale).
Auction rules

I For each round, for each item, bidders were allowed to bid if
they were “sufficiently active” in the previous round
This avoided bidders from waiting (not bidding until they see
their opponents’ strategies).

I Perfect information: at each round, all bids (with identity of

the bidders) was disclosed.
I Once bids are disclosed bidders could withdraw a bid (only if
the highest).
I Penalty for withdrawing bids.
I Reduces even further the exposure problem.
How did it go?

I Revenues in US auctions was substantial, it typically exceeded

industry and government estimates.
I Simultaneous ascending auction may be a cause:
I Information revelation reduces the winners’ curse
I Bidders can then bid more aggressively.
I Revenue also increases when bidders can piece together
complementary licenses.

I Prices generated are close to market prices: similar items sold

for similar items.
Efficient license aggregation

I For narrowband, bidders managed to buy adjacent bands.

The buffer between bands can be used, increasing capacity.

I For bidders with nationwide interest, bidders managed to

I get a license for each region
I Get the same band across regions (easier to avoid disruption
when crossing state lines).

I licenses with local synergies. Higher bidding when competing

with a bidder holding neighboring licenses:
Bidders bid for synergistic gains (and obtained them).
Efficiency

Difficult to measure, we do not observe firms’ true valuations.

⇒ Efficiency is almost impossible to assess.

Some economists argue that efficiency is not an issue if resale is

allowed: post-auction transactions would fix inefficiencies.
I But transaction costs are not zero and may reduce
opportunity of resale.
I Nevertheless, no resale observed
Except for GTE, but resold at the price it paid what it paid
(most probably due to a corporate strategy shift from PCS
toward Cellular).
Demand reduction

In single-item auctions, each bidder has a dominant strategy and

bid its private valuation (if using the “right” auction format).
In multi-item auctions, there’s no dominant strategy, and there is
scope for tacit collusion via demand reduction.
I Bidding on more items raises the price on all items.

I So, there’s an incentive to bid on fewer items.

I Less bidding = lower demand ⇒ lower prices.

I More likely if bidders prefer few units at a low price than

many units at a high price.
Example
Two items, A and B, and two bidders, i and j. Both bidder value
either item at 100, and having both A and B at 200.

Strategy:
I If not the highest bidder on any item, bid the minimum
increment for the lowest-priced item (if both prices are
identical, select at random).
I If the highest bidder for one item, do not submit any bid.

Suppose the current bids are 20 (for Am by i) and 25 (for B, by j).

The above strategy implies:
I i does not bid further on B
I j does not bid further on A.
The auction ends with low prices.
The German ’99 GSM auction

I 10 blocks of spectrum, up to 5 licenses (at least 2 blocks for a

license).
to Germany auctioned off spectrum while the US auctioned
off licences (attached to a spectrum).
I 2 dominant firms (T-Mobil & Mannesmann), and 2 small
operators (E-Plus & Viag).
(T-Mobil = subsidiary of Deutsche Telekom).
I Both T-Mobil and Mannesmann had a good idea of each
other’s valuations and that of the small providers.
I Auction opens at 10:15, Oct. 28, 1999, minimum bid set to
DM 1 Million. Bidders had 30’ to post a bid.
 I At 10:23 Mannesman jump bids to:
I DM 36.36 Million for licenses 1–5
I DM 40.00 Million for licenses 6–9
I DM 56.00 Million for license 10 (bigger license).

I Second round: T-Mobile bids DM 40.01 Million for licenses

1–5.
I Third round: No further bidding.

Revenue: about DM 400 Million (≈ e200 Million).

(A year later for the 3G/UMTS auction Germany would raise
about e55 billion)
Problems with the German design:
I Maximum number of licenses = number of incumbent.
⇒ Less incentives for entry, less competition.

I Number of licenses is endogenous.

⇒ Incentives for large bidders (T-Mobil & Mannesman) to
outbid smaller bidders (E-Plus & Viag).
When competition ends

The German auction is an example of demand reduction:

A simultaneous ascending auction ends as soon as:

I number of bidders = number of licenses for sale;

I Each bidder is the top bidder for one license.

It’s like when there is an auction for one item only and there’s only
one 1 bidder.
⇒ There’s an incentive to “coordinate” and stop bidding. This is
what happened in the 1999 German GSM auction.

Lesson: we need more bidders than there are licences.

The 2000 German UMTS auction

I Simultaneous ascending auction

I 12 blocks of spectrum. A license = 2 or 3 blocks (so ranges
from 4 to 6 licenses).
I Only 7 bidders.

Collusion:
I MobilCom made a (public) collusive offer to Debitel (inviting
not to bid and use MobilCom’s network as a virtual operator).
I German government did not respond: going to 6 bidders
would have been an invitation to end the auction in the 1st or
2nd round.
I But Debitel eventually stopped bidding early.
For the 2 largest incumbent (Deutsche Telekom &
Vodafone-Mannesman) two options remained after Debitel
dropped:
I Raise bidding to force exit of the weaker firms among the
remaining.
I “Collude” and reduce demand (and thus stop the auction).
I Vodafone-Mannesman signaled through its bid the “collusive
strategy.”
I Deutsche Telekom played deaf and continued to bid, but
stopped before excluding weaker firms.
I Outcome: assignment same as when Debitel dropped, but at
much higher prices.
The UK 3g auction

In March–April 2000 the UK ran an auction for 3G licenses.

Issues
I 4 incumbents with a 2G license
I 4 very similar 3G licenses to be sold.

To promote competition (and favor entry), the British government

planned to run a Anglo-Dutch auction.
Anglo-Dutch auction

I First an ascending auction is run until 5 bidders remain.

Objectives were:
I To promote entry;
I To gather information (mitigate the winner’s curse).

I Then a sealed-bid auction, where all bidders would pay the

4th highest bid.
⇒ Increase competition, bidders would bid aggressively.

In the end, 5 licenses were sold, so the Anglo-Dutch format was

abandoned and they went for a simultaneous ascending auction.
The simultaneous ascending auction provides
I a more transparent process

I competitive prices

I simple bidders’ strategies

I (nearly) efficient outcome

Simple strategies
I Suppose every bidder puts a positive probability (could be
small), at any time, that each rival make no further bid.
I So each bid can possibly be the last one.

I At each round a bidder must then make the bid that would
maximize its profit:
choose the license and bid that maximize
value - (current highest bid + minimal increment)
At the end, the prices are such that:
I each bidder chooses to bid exactly what it ends up with.

I So, given prices, no reallocation of licenses can increase the

surplus of any bidder.
I Since prices are just a transfer between the bidders and the
seller, there’s no other allocation and other prices that can
give a higher surplus.
I So the outcome is efficient (it maximizes the total valuations,
i.e., it maximizes the total surplus).

The auction generated £22.5 billion.

3G auctions in Europe

Year 2000 Year 2001

Austria 100 Belgium 45
Germany 615 Denmark 95
Italy 240 Greece 45
Netherlands 170
Switzerland 20
UK 650

Revenues in euro/capita.

Cell phone contract in Europe: typically around e30/month.

Take-away

I Auctioning spectrum licenses is a challenge for auction theory.

I An auction is a better alternative to lotteries or beauty

contest.
I Combinatorial auctions (like VCG) may not be feasible in
practice (computationally too hard to run).
I An auction should aim at:
I Efficiency and revenue maximization;
I Favor entry;
I Avoid collusion.

I Proposed design: the multi-ascending auction (MAE).

I Crucial problems are:
I Exposure problem: VCG better.
I Winner’s curse: MAE better.
I Demand reduction: MAE better.

I First use of the MAE: the US ’94 PCS auction: success!

I Too few bidders (or endogenous number of licenses) can make

the auction fail (German ’99 GSM auction).
I Anglo-Dutch auction can be an option when there are too few
bidders.