What Really Matters in Auction Design

Paul Klemperer

A uctions have become enormously popular in recent years Governmentsare now especially keen using auctions to sell mobile-phone licensesoperate decentralized electricity markets privatize companies and for

many other purposes The growth of e-commerce has led to many business-to-business auctions for goods whose trade was previously negotiated bilaterally

Economists are proud of their role in pushing for auctions for example Coase(1959) was among the rst to advocate auctioning the radio spectrum But manyauctionsmdashincluding some designed with the help of leading academiceconomistsmdashhave worked very badly

For example six European countries auctioned off spectrum licenses forldquothird-generationrdquo mobile phones in 2000 In Germany and the United Kingdomthe spectrum sold for over 600 euros per person ($80 billion in all or over2 percent of GDP) But in Austria the Netherlands Italy and Switzerland therevenues were just 100 170 240 and 20 euros per person respectively To be sureinvestors became more skeptical about the underlying value of the spectrum during2000 (and they are even more skeptical today) But this is just a fraction of the storyThe Netherlands auction was sandwiched between the UK and German auctionsand analysts and government of cials predicted revenues in excess of 400 euros perperson from the Italian and Swiss auctions just a few days before they began(Michelson 2000 Roberts 2000 Total Telecom 2000 Klemperer 2002) Theseother auctions were ascoes primarily because they were poorly designed

So what makes a successful auctionWhat really matters in auction design are the same issues that any industry

y Paul Klemperer is the Edgeworth Professor of Economics Oxford University England Hise-mail address is paulklemperereconomicsoxacuk and his website is httpwwwpaulklempererorg

Journal of Economic PerspectivesmdashVolume 16 Number 1mdashWinter 2002mdashPages 169ndash189

regulator would recognize as key concerns discouraging collusive entry-deterringand predatory behavior In short good auction design is mostly good elementaryeconomics

By contrast most of the extensive auction literature (as summarized in forexample Klemperer 1999a 2000a) is of second-order importance for practicalauction design The auction literature largely focuses on a xed number of bidderswho bid noncooperatively and it emphasizes issues such as the effects of riskaversion correlation of information budget constraints and complementaritiesAuction theorists have made important progress on these topics from which othereconomic theory has bene ted and auction theory has also been fruitfully appliedin political economy nance law and economics labor economics and industrialorganization often in contexts not usually though of as auctions (Klemperer2001a) But most of this literature is of much less use for actually designingauctions

This paper will list and give examples of some critical pitfalls in auction designand discuss what to do about them We show that ascending and uniform-priceauctions are both very vulnerable to collusion and very likely to deter entry into anauction We consider including a nal sealed-bid stage into an otherwise-ascendingauction to create an ldquoAnglo-Dutchrdquo auction and we emphasize the need forstronger antitrust policy in auction markets

Collusion

A rst major set of concerns for practical auction design involves the risk thatparticipants may explicitly or tacitly collude to avoid bidding up prices Consider amultiunit (simultaneous) ascending auction (This is just like the standard auctionused for example to sell a painting in Sothebyrsquos or Christiesmdashthe price starts lowand competing bidders raise the price until no one is prepared to bid any higherand the nal bidder then wins the prize at the nal price bid However in this caseseveral objects are sold at the same time with the price rising on each of themindependently and none of the objects is nally sold until no one wishes to bidagain on any of the objects) In such an auction bidders can use the early stageswhen prices are still low to signal who should win which objects and then tacitlyagree to stop pushing up prices

For example in 1999 Germany sold ten blocks of spectrum by a simultaneousascending auction with the rule that any new bid on a block had to exceed theprevious high bid by at least 10 percent Mannesmanrsquos rst bids were 1818 milliondeutschmarks per megahertz on blocks 1ndash5 and 20 million DM per MHz on blocks6ndash10 the only other credible biddermdashT-Mobilmdash bid even less in the rst roundOne of T-Mobilrsquos managers then said (Stuewe 1999 p 13) ldquoThere were noagreements with Mannesman But [T-Mobil] interpreted Mannesmanrsquos rst bid asan offerrdquo The point is that 1818 plus a 10 percent raise equals approximately 20It seems T-Mobil understood that if it bid 20 million DM per MHz on blocks 1ndash5

170 Journal of Economic Perspectives

but did not bid again on blocks 6ndash10 the two companies would then live and letlive with neither company challenging the other on the otherrsquos half Exactly thathappened So the auction closed after just two rounds with each of the biddersacquiring half the blocks for the same low price ( Jehiel and Moldovanu 2001Grimm Riedel and Wolfstetter 2001)

Ascending auctions can also facilitate collusion by offering a mechanism forpunishing rivals The threat of punishment may be implicit for example it wasclear to T-Mobil that Mannesman would retaliate with high bids on blocks 1ndash5 ifT-Mobil continued bidding on blocks 6ndash10 But an ascending auction can alsoallow more explicit options for punishment

In a multilicense US spectrum auction in 1996ndash1997 US West was compet-ing vigorously with McLeod for lot number 378 a license in Rochester MinnesotaAlthough most bids in the auction had been in exact thousands of dollars USWest bid $313378 and $62378 for two licenses in Iowa in which it had earliershown no interest overbidding McLeod who had seemed to be the uncontestedhigh bidder for these licenses McLeod got the point that it was being punished forcompeting in Rochester and dropped out of that market Since McLeod madesubsequent higher bids on the Iowa licenses the ldquopunishmentrdquo bids cost US Westnothing (Cramton and Schwartz 1999)

A related phenomenon can arise in one special kind of sealed-bid auctionnamely a uniform-price auction in which each bidder submits a sealed bid statingwhat price it would pay for different quantities of a homogenous good likeelectricity (that is it submits a demand function) and then the good is sold at thesingle price determined by the lowest winning bid In this format bidders cansubmit bids that ensure that any deviation from a (tacit or explicit) collusiveagreement is severely punished each bidder bids very high prices for smallerquantities than its collusively agreed share Then if any bidder attempts to obtainmore than its agreed share (leaving other rms with less than their agreed shares)all bidders will have to pay these very high prices However if everyone sticks totheir agreed shares then these very high prices will never need to be paid As aresult deviation from the collusive agreement is unpro table1

The electricity regulator in the United Kingdom believes the market in whichdistribution companies purchase electricity from generating companies has fallenprey to exactly this kind of ldquoimplicit collusionrdquo (Of ce of Gas and ElectricityMarkets 1999 pp 173ndash174) ldquoFar from being the success story trumpeted aroundthe world the story of the UK generation market and the development ofcompetition has been something of a disasterrdquo reported Power UK (1999 see also

1 Since with many units the lowest winning bid in a uniform-price auction is typically not importantlydifferent from the highest losing bid this auction is analogous to an ascending auction (in which everywinner pays the runner-uprsquos willingness-to-pay) The ldquothreatsrdquo that support collusion in a uniform-priceauction are likewise analogous to the implicit threats supporting collusion in an ascending auctionCollusion in a uniform-price auction is harder if supply is uncertain since this reduces the number ofpoints on the bid schedule that are inframarginal and can be used as threats (Klemperer and Meyer1989 Back and Zender 1993 1999)

Paul Klemperer 171

von der Fehr and Harbord 1998 Newbery 1998 Wolfram 1998 1999) In addi-tion a frequently repeated auction market such as that for electricity is particularlyvulnerable to collusion because the repeated interaction among bidders expandsthe set of signaling and punishment strategies available to them and allows them tolearn to cooperate (Klemperer 2002)

Much of the kind of behavior discussed so far is hard to challenge legallyIndeed trying to outlaw it all would require cumbersome rules that would restrictbiddersrsquo exibility and might generate inef ciencies without being fully effectiveIt would be much better to solve these problems with better auction designs

Entry Deterrence and Predation

The second major area of concern of practical auction design is to attractbidders since an auction with too few bidders risks being unpro table for theauctioneer (Bulow and Klemperer 1996) and potentially inef cient Ascendingauctions are often particularly poor in this respect since they can allow somebidders to deter the entry or depress the bidding of rivals

In an ascending auction there is a strong presumption that the rm that valueswinning the most will be the eventual winner because even if it is outbid at an earlystage it can eventually top any opposition As a result other rms have littleincentive to enter the bidding and may not do so if they have even modest costs ofbidding

Consider for example Glaxorsquos 1995 takeover of the Wellcome drugs companyAfter Glaxorsquos rst bid of 9 billion pounds Zeneca expressed willingness to offerabout 10 billion pounds if it could be sure of winning while Roche considered anoffer of 11 billion pounds But certain synergies made Wellcome worth a little moreto Glaxo than to the other rms and the costs of bidding were tens of millions ofpounds Eventually neither Roche nor Zeneca actually entered the bidding andWellcome was sold at the original bid of 9 billion pounds literally a billion or twoless than its shareholders might have received Wellcomersquos own chief executiveadmitted ldquothere was money left on the tablerdquo (Wighton 1995a b)

While ascending auctions are particularly vulnerable to lack of entry otherauction forms can result in similar problems if the costs of entry and the asymme-tries between bidders are too large

The 1991 UK sale of television franchises by a sealed-bid auction is a dramaticexample While the regions in the South and Southeast Southwest East Wales andWest Northeast and Yorkshire all sold in the range of 936 to 1588 pounds perhead of population the onlymdashand therefore winningmdashbid for the Midlandsregion was made by the incumbent rm and was just one-twentieth of one penny ()per head of population Much the same happened in Scotland where the onlybidder for the Central region generously bid one-seventh of one penny per capitaWhat had happened was that bidders were required to provide very detailed

172 Journal of Economic Perspectives

region-speci c programming plans In each of these two regions the only bidder gured out that no one else had developed such a plan2

Another issue that can depress bidding in some ascending auctions is theldquowinnerrsquos curserdquo This problem applies when bidders have the same or close to thesame actual value for a prize but they have different information about that actualvalue (what auction theorists call the ldquocommon valuesrdquo case) The winnerrsquos cursere ects the danger that the winner of an auction is likely to be the party who hasmost greatly overestimated the value of the prize Knowing about the winnerrsquos cursewill cause everyone to bid cautiously But weaker rms must be especially cautioussince they must recognize that they are only likely to win when they have overes-timated the value by even more than usual Therefore an advantaged rm can beless cautious since beating very cautious opponents need not imply one hasoverestimated the prizersquos value Because the winnerrsquos curse affects weak rms muchmore than strong ones and because the effect is self-reinforcing the advantagedbidder wins most of the time And because its rivals bid extremely cautiously it alsogenerally pays a low price when it does win (Klemperer 1998)

The bidding on the Los Angeles license in the 1995 US auction for mobile-phone broadband licenses illustrates this problem While the licensersquos value washard to estimate it was probably worth similar amounts to several bidders ButPaci c Telephone which already operated the local xed-line telephone businessin California had distinct advantages from its database on potential local custom-ers its well-known brand-name and its familiarity with doing business in CaliforniaThe auction was an ascending one The result was that the bidding stopped at a verylow price In the end the Los Angeles license yielded only $26 per capita InChicago by contrast the main local xed-line provider was ineligible to competeand it was not obvious who would win so the auction yielded $31 per capita eventhough Chicago was thought less valuable than Los Angeles because of its lowerhousehold incomes lower expected population growth and more dispersed pop-ulation (Klemperer 1998 Bulow and Klemperer 2002) For formal econometricevidence for the FCC auctions more broadly see Klemperer and Pagnozzi (2002)

Of course the ldquowinnerrsquos curserdquo problem exacerbates the problem that weakerbidders may not bother to participate in an ascending auction GTE and BellAtlantic made deals that made them ineligible to bid for the Los Angeles licenseand MCI failed to enter this auction at all Similarly takeover battles are essentiallyascending auctions and there is empirical evidence that a rm that makes atakeover bid has a lower risk of facing a rival bidder if the rm has a largershareholding or ldquotoeholdrdquo in the target company (Betton and Eckbo 2000)

Because outcomes in an ascending auction can be dramatically in uenced bya seemingly modest advantage developing such an advantage can be an effectivepredatory strategy An apparent example was the 1999 attempt by BSkyB (RupertMurdochrsquos satellite television company) to acquire Manchester United (Englandrsquos

2 While I have advised the UK government on several auctions I have never had anything to do withtelevision licenses

What Really Matters in Auction Design 173

most successful soccer club) The problem was the advantage this would give BSkyBin the auction of football television rights Since Manchester United receives7 percent of the Premier Leaguersquos television revenues BSkyB would have received7 percent of the price of the leaguersquos broadcasting rights whoever won them SoBSkyB would have had an incentive to bid more aggressively in an ascendingauction to push up the price of the rights and knowing this other potential bidderswould have faced a worse ldquowinnerrsquos curserdquo and backed off BSkyB might have endedup with a lock over the television rights with damaging effects on the televisionmarket more generally Largely for this reason the UK government blocked theacquisition3

A strong bidder also has an incentive to create a reputation for aggressivenessthat reinforces its advantage For example when Glaxo was bidding for Wellcomeit made it clear that it ldquowould almost certainly top a rival bidrdquo (Wighton 1995b)Similarly before bidding for the California phone license Paci c Telephoneannounced in the Wall Street Journal that ldquoif somebody takes California away fromus theyrsquoll never make any moneyrdquo (Cauley and Carnevale 1994 p A4) Paci cTelephone also hired one of the worldrsquos most prominent auction theorists to giveseminars to the rest of the industry to explain the winnerrsquos curse argument thatjusti es this statement and it reinforced the point in full-page ads that ran in thenewspapers of the cities where its major competitors were headquartered (Koselka1995 p 63) It also made organizational changes that demonstrated its commit-ment to winning the Los Angeles license

Predation may be particularly easy in repeated ascending auctions such as ina series of spectrum auctions A bidder who buys assets that are complementary toassets for sale in a future auction or who simply bids very aggressively in earlyauctions can develop a reputation for aggressiveness (Bikhchandani 1988) Poten-tial rivals in future auctions will be less willing to participate and will bid lessaggressively if they do participate (Klemperer 2002)

Finally because an ascending auction often effectively blocks the entry ofldquoweakerrdquo bidders it encourages ldquostrongerrdquo bidders to bid jointly or to collude afterall they know that no one else can enter the auction to steal the collusive rents theycreate In the disastrous November 2000 Swiss sale of four third-generation mobile-phone licenses there was considerable initial interest from potential bidders Butweaker bidders were put off by the auction formmdashat least one company hiredbidding consultants and then gave up after learning that the ascending-biddingrules would give the company very little chance against stronger rivals Moreoverthe government permitted last-minute joint-bidding agreementsmdash essentially of -cially sanctioned collusion In the week before the auction the eld shrank fromnine bidders to just four bidders for the four licenses Since no bidder was allowed

3 Although the term ldquotoehold effectrdquo coined by Bulow Huang and Klemperer (1999) and Klemperer(1998) in the related context of takeover battles (see above) entered the popular press and thesepapers were cited by the UK Monopolies and Mergers Commission (1999) report which effectivelydecided the issue neither I nor my coauthors had any involvement in this case

174 Journal of Economic Perspectives

to take more than one license the sale price was determined by the reserve pricewhich was just one-thirtieth of the UK and German per capita revenues andone- ftieth of what the Swiss had once hoped for

Other Pitfalls

Reserve PricesMany of the disasters above were greatly aggravated by failure to set a proper

reserve price (the minimum amount the winner is required to pay) Take theprevious example It was ridiculous for the Swiss government to set its reserve at justone-thirtieth of the per capita revenue raised by the German and UK governmentsfor similar properties Since the governmentrsquos own spokesman predicted just vedays prior to the auction that twenty times the reserve price would be raised whatwas the government playing at

Inadequate reserve prices also increase the incentives for predation and mayencourage collusion that would not otherwise have been in all biddersrsquo interests Astronger bidder in an ascending auction has a choice between either tacitly collud-ing to end the auction quickly at a low price or forcing the price up to drive outweaker bidders The lower the reserve price at which the auction can be concludedthe more attractive is the rst option This factor may have been an importantcontributor to several of the ascoes we have discussed

Political ProblemsSerious reserve prices are often opposed not only by industry groups but also

by government of cials for whom a very embarrassing outcome is that the reserveprice is not met the object is not sold and the auction is seen as a ldquofailurerdquo

Similarly standard ( rst-price) sealed-bid auctionsmdashin which the bidders si-multaneously make ldquobest and nalrdquo offers and the winner pays the price hebidmdash can sometimes be very embarrassing for bidders as BSCH (Spainrsquos biggestbank) found out when Brazil privatized the Sao Paulo state bank Banespa Whenthe bids were opened BSCHrsquos managers were horri ed to learn that their bid ofover 7 billion reals ($36 billion) was more than three times the runner-uprsquos bid andthat they were therefore paying 5 billion reals ($25 billion) more than was neededto win In other auctions meanwhile losers who have just narrowly underbid thewinners have found it equally hard to explain themselves to their bosses andshareholders So rms or at least their managers can oppose rst-price auctions

On the other hand a second-price sealed-bid auctionmdashin which the winner paysthe runner-uprsquos bidmdash can be embarrassing for the auctioneer if the winnerrsquos actualbid is revealed to be far more than the runner-uprsquos even if the auction design wasboth ef cient and maximized expected revenue McMillan (1994) reports asecond-price New Zealand auction in which the winner bid NZ $7 million but paidthe runner-uprsquos bid of NZ $5000 New Zealand should have set a minimum reserveprice that the winner had to pay but even if that had been politically possible the

Paul Klemperer 175

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

regulator would recognize as key concerns discouraging collusive entry-deterringand predatory behavior In short good auction design is mostly good elementaryeconomics

By contrast most of the extensive auction literature (as summarized in forexample Klemperer 1999a 2000a) is of second-order importance for practicalauction design The auction literature largely focuses on a xed number of bidderswho bid noncooperatively and it emphasizes issues such as the effects of riskaversion correlation of information budget constraints and complementaritiesAuction theorists have made important progress on these topics from which othereconomic theory has bene ted and auction theory has also been fruitfully appliedin political economy nance law and economics labor economics and industrialorganization often in contexts not usually though of as auctions (Klemperer2001a) But most of this literature is of much less use for actually designingauctions

This paper will list and give examples of some critical pitfalls in auction designand discuss what to do about them We show that ascending and uniform-priceauctions are both very vulnerable to collusion and very likely to deter entry into anauction We consider including a nal sealed-bid stage into an otherwise-ascendingauction to create an ldquoAnglo-Dutchrdquo auction and we emphasize the need forstronger antitrust policy in auction markets

Collusion

A rst major set of concerns for practical auction design involves the risk thatparticipants may explicitly or tacitly collude to avoid bidding up prices Consider amultiunit (simultaneous) ascending auction (This is just like the standard auctionused for example to sell a painting in Sothebyrsquos or Christiesmdashthe price starts lowand competing bidders raise the price until no one is prepared to bid any higherand the nal bidder then wins the prize at the nal price bid However in this caseseveral objects are sold at the same time with the price rising on each of themindependently and none of the objects is nally sold until no one wishes to bidagain on any of the objects) In such an auction bidders can use the early stageswhen prices are still low to signal who should win which objects and then tacitlyagree to stop pushing up prices

For example in 1999 Germany sold ten blocks of spectrum by a simultaneousascending auction with the rule that any new bid on a block had to exceed theprevious high bid by at least 10 percent Mannesmanrsquos rst bids were 1818 milliondeutschmarks per megahertz on blocks 1ndash5 and 20 million DM per MHz on blocks6ndash10 the only other credible biddermdashT-Mobilmdash bid even less in the rst roundOne of T-Mobilrsquos managers then said (Stuewe 1999 p 13) ldquoThere were noagreements with Mannesman But [T-Mobil] interpreted Mannesmanrsquos rst bid asan offerrdquo The point is that 1818 plus a 10 percent raise equals approximately 20It seems T-Mobil understood that if it bid 20 million DM per MHz on blocks 1ndash5

170 Journal of Economic Perspectives

but did not bid again on blocks 6ndash10 the two companies would then live and letlive with neither company challenging the other on the otherrsquos half Exactly thathappened So the auction closed after just two rounds with each of the biddersacquiring half the blocks for the same low price ( Jehiel and Moldovanu 2001Grimm Riedel and Wolfstetter 2001)

Ascending auctions can also facilitate collusion by offering a mechanism forpunishing rivals The threat of punishment may be implicit for example it wasclear to T-Mobil that Mannesman would retaliate with high bids on blocks 1ndash5 ifT-Mobil continued bidding on blocks 6ndash10 But an ascending auction can alsoallow more explicit options for punishment

In a multilicense US spectrum auction in 1996ndash1997 US West was compet-ing vigorously with McLeod for lot number 378 a license in Rochester MinnesotaAlthough most bids in the auction had been in exact thousands of dollars USWest bid $313378 and $62378 for two licenses in Iowa in which it had earliershown no interest overbidding McLeod who had seemed to be the uncontestedhigh bidder for these licenses McLeod got the point that it was being punished forcompeting in Rochester and dropped out of that market Since McLeod madesubsequent higher bids on the Iowa licenses the ldquopunishmentrdquo bids cost US Westnothing (Cramton and Schwartz 1999)

A related phenomenon can arise in one special kind of sealed-bid auctionnamely a uniform-price auction in which each bidder submits a sealed bid statingwhat price it would pay for different quantities of a homogenous good likeelectricity (that is it submits a demand function) and then the good is sold at thesingle price determined by the lowest winning bid In this format bidders cansubmit bids that ensure that any deviation from a (tacit or explicit) collusiveagreement is severely punished each bidder bids very high prices for smallerquantities than its collusively agreed share Then if any bidder attempts to obtainmore than its agreed share (leaving other rms with less than their agreed shares)all bidders will have to pay these very high prices However if everyone sticks totheir agreed shares then these very high prices will never need to be paid As aresult deviation from the collusive agreement is unpro table1

The electricity regulator in the United Kingdom believes the market in whichdistribution companies purchase electricity from generating companies has fallenprey to exactly this kind of ldquoimplicit collusionrdquo (Of ce of Gas and ElectricityMarkets 1999 pp 173ndash174) ldquoFar from being the success story trumpeted aroundthe world the story of the UK generation market and the development ofcompetition has been something of a disasterrdquo reported Power UK (1999 see also

1 Since with many units the lowest winning bid in a uniform-price auction is typically not importantlydifferent from the highest losing bid this auction is analogous to an ascending auction (in which everywinner pays the runner-uprsquos willingness-to-pay) The ldquothreatsrdquo that support collusion in a uniform-priceauction are likewise analogous to the implicit threats supporting collusion in an ascending auctionCollusion in a uniform-price auction is harder if supply is uncertain since this reduces the number ofpoints on the bid schedule that are inframarginal and can be used as threats (Klemperer and Meyer1989 Back and Zender 1993 1999)

Paul Klemperer 171

von der Fehr and Harbord 1998 Newbery 1998 Wolfram 1998 1999) In addi-tion a frequently repeated auction market such as that for electricity is particularlyvulnerable to collusion because the repeated interaction among bidders expandsthe set of signaling and punishment strategies available to them and allows them tolearn to cooperate (Klemperer 2002)

Much of the kind of behavior discussed so far is hard to challenge legallyIndeed trying to outlaw it all would require cumbersome rules that would restrictbiddersrsquo exibility and might generate inef ciencies without being fully effectiveIt would be much better to solve these problems with better auction designs

Entry Deterrence and Predation

The second major area of concern of practical auction design is to attractbidders since an auction with too few bidders risks being unpro table for theauctioneer (Bulow and Klemperer 1996) and potentially inef cient Ascendingauctions are often particularly poor in this respect since they can allow somebidders to deter the entry or depress the bidding of rivals

In an ascending auction there is a strong presumption that the rm that valueswinning the most will be the eventual winner because even if it is outbid at an earlystage it can eventually top any opposition As a result other rms have littleincentive to enter the bidding and may not do so if they have even modest costs ofbidding

Consider for example Glaxorsquos 1995 takeover of the Wellcome drugs companyAfter Glaxorsquos rst bid of 9 billion pounds Zeneca expressed willingness to offerabout 10 billion pounds if it could be sure of winning while Roche considered anoffer of 11 billion pounds But certain synergies made Wellcome worth a little moreto Glaxo than to the other rms and the costs of bidding were tens of millions ofpounds Eventually neither Roche nor Zeneca actually entered the bidding andWellcome was sold at the original bid of 9 billion pounds literally a billion or twoless than its shareholders might have received Wellcomersquos own chief executiveadmitted ldquothere was money left on the tablerdquo (Wighton 1995a b)

While ascending auctions are particularly vulnerable to lack of entry otherauction forms can result in similar problems if the costs of entry and the asymme-tries between bidders are too large

The 1991 UK sale of television franchises by a sealed-bid auction is a dramaticexample While the regions in the South and Southeast Southwest East Wales andWest Northeast and Yorkshire all sold in the range of 936 to 1588 pounds perhead of population the onlymdashand therefore winningmdashbid for the Midlandsregion was made by the incumbent rm and was just one-twentieth of one penny ()per head of population Much the same happened in Scotland where the onlybidder for the Central region generously bid one-seventh of one penny per capitaWhat had happened was that bidders were required to provide very detailed

172 Journal of Economic Perspectives

region-speci c programming plans In each of these two regions the only bidder gured out that no one else had developed such a plan2

Another issue that can depress bidding in some ascending auctions is theldquowinnerrsquos curserdquo This problem applies when bidders have the same or close to thesame actual value for a prize but they have different information about that actualvalue (what auction theorists call the ldquocommon valuesrdquo case) The winnerrsquos cursere ects the danger that the winner of an auction is likely to be the party who hasmost greatly overestimated the value of the prize Knowing about the winnerrsquos cursewill cause everyone to bid cautiously But weaker rms must be especially cautioussince they must recognize that they are only likely to win when they have overes-timated the value by even more than usual Therefore an advantaged rm can beless cautious since beating very cautious opponents need not imply one hasoverestimated the prizersquos value Because the winnerrsquos curse affects weak rms muchmore than strong ones and because the effect is self-reinforcing the advantagedbidder wins most of the time And because its rivals bid extremely cautiously it alsogenerally pays a low price when it does win (Klemperer 1998)

The bidding on the Los Angeles license in the 1995 US auction for mobile-phone broadband licenses illustrates this problem While the licensersquos value washard to estimate it was probably worth similar amounts to several bidders ButPaci c Telephone which already operated the local xed-line telephone businessin California had distinct advantages from its database on potential local custom-ers its well-known brand-name and its familiarity with doing business in CaliforniaThe auction was an ascending one The result was that the bidding stopped at a verylow price In the end the Los Angeles license yielded only $26 per capita InChicago by contrast the main local xed-line provider was ineligible to competeand it was not obvious who would win so the auction yielded $31 per capita eventhough Chicago was thought less valuable than Los Angeles because of its lowerhousehold incomes lower expected population growth and more dispersed pop-ulation (Klemperer 1998 Bulow and Klemperer 2002) For formal econometricevidence for the FCC auctions more broadly see Klemperer and Pagnozzi (2002)

Of course the ldquowinnerrsquos curserdquo problem exacerbates the problem that weakerbidders may not bother to participate in an ascending auction GTE and BellAtlantic made deals that made them ineligible to bid for the Los Angeles licenseand MCI failed to enter this auction at all Similarly takeover battles are essentiallyascending auctions and there is empirical evidence that a rm that makes atakeover bid has a lower risk of facing a rival bidder if the rm has a largershareholding or ldquotoeholdrdquo in the target company (Betton and Eckbo 2000)

Because outcomes in an ascending auction can be dramatically in uenced bya seemingly modest advantage developing such an advantage can be an effectivepredatory strategy An apparent example was the 1999 attempt by BSkyB (RupertMurdochrsquos satellite television company) to acquire Manchester United (Englandrsquos

2 While I have advised the UK government on several auctions I have never had anything to do withtelevision licenses

What Really Matters in Auction Design 173

most successful soccer club) The problem was the advantage this would give BSkyBin the auction of football television rights Since Manchester United receives7 percent of the Premier Leaguersquos television revenues BSkyB would have received7 percent of the price of the leaguersquos broadcasting rights whoever won them SoBSkyB would have had an incentive to bid more aggressively in an ascendingauction to push up the price of the rights and knowing this other potential bidderswould have faced a worse ldquowinnerrsquos curserdquo and backed off BSkyB might have endedup with a lock over the television rights with damaging effects on the televisionmarket more generally Largely for this reason the UK government blocked theacquisition3

A strong bidder also has an incentive to create a reputation for aggressivenessthat reinforces its advantage For example when Glaxo was bidding for Wellcomeit made it clear that it ldquowould almost certainly top a rival bidrdquo (Wighton 1995b)Similarly before bidding for the California phone license Paci c Telephoneannounced in the Wall Street Journal that ldquoif somebody takes California away fromus theyrsquoll never make any moneyrdquo (Cauley and Carnevale 1994 p A4) Paci cTelephone also hired one of the worldrsquos most prominent auction theorists to giveseminars to the rest of the industry to explain the winnerrsquos curse argument thatjusti es this statement and it reinforced the point in full-page ads that ran in thenewspapers of the cities where its major competitors were headquartered (Koselka1995 p 63) It also made organizational changes that demonstrated its commit-ment to winning the Los Angeles license

Predation may be particularly easy in repeated ascending auctions such as ina series of spectrum auctions A bidder who buys assets that are complementary toassets for sale in a future auction or who simply bids very aggressively in earlyauctions can develop a reputation for aggressiveness (Bikhchandani 1988) Poten-tial rivals in future auctions will be less willing to participate and will bid lessaggressively if they do participate (Klemperer 2002)

Finally because an ascending auction often effectively blocks the entry ofldquoweakerrdquo bidders it encourages ldquostrongerrdquo bidders to bid jointly or to collude afterall they know that no one else can enter the auction to steal the collusive rents theycreate In the disastrous November 2000 Swiss sale of four third-generation mobile-phone licenses there was considerable initial interest from potential bidders Butweaker bidders were put off by the auction formmdashat least one company hiredbidding consultants and then gave up after learning that the ascending-biddingrules would give the company very little chance against stronger rivals Moreoverthe government permitted last-minute joint-bidding agreementsmdash essentially of -cially sanctioned collusion In the week before the auction the eld shrank fromnine bidders to just four bidders for the four licenses Since no bidder was allowed

3 Although the term ldquotoehold effectrdquo coined by Bulow Huang and Klemperer (1999) and Klemperer(1998) in the related context of takeover battles (see above) entered the popular press and thesepapers were cited by the UK Monopolies and Mergers Commission (1999) report which effectivelydecided the issue neither I nor my coauthors had any involvement in this case

174 Journal of Economic Perspectives

to take more than one license the sale price was determined by the reserve pricewhich was just one-thirtieth of the UK and German per capita revenues andone- ftieth of what the Swiss had once hoped for

Other Pitfalls

Reserve PricesMany of the disasters above were greatly aggravated by failure to set a proper

reserve price (the minimum amount the winner is required to pay) Take theprevious example It was ridiculous for the Swiss government to set its reserve at justone-thirtieth of the per capita revenue raised by the German and UK governmentsfor similar properties Since the governmentrsquos own spokesman predicted just vedays prior to the auction that twenty times the reserve price would be raised whatwas the government playing at

Inadequate reserve prices also increase the incentives for predation and mayencourage collusion that would not otherwise have been in all biddersrsquo interests Astronger bidder in an ascending auction has a choice between either tacitly collud-ing to end the auction quickly at a low price or forcing the price up to drive outweaker bidders The lower the reserve price at which the auction can be concludedthe more attractive is the rst option This factor may have been an importantcontributor to several of the ascoes we have discussed

Political ProblemsSerious reserve prices are often opposed not only by industry groups but also

by government of cials for whom a very embarrassing outcome is that the reserveprice is not met the object is not sold and the auction is seen as a ldquofailurerdquo

Similarly standard ( rst-price) sealed-bid auctionsmdashin which the bidders si-multaneously make ldquobest and nalrdquo offers and the winner pays the price hebidmdash can sometimes be very embarrassing for bidders as BSCH (Spainrsquos biggestbank) found out when Brazil privatized the Sao Paulo state bank Banespa Whenthe bids were opened BSCHrsquos managers were horri ed to learn that their bid ofover 7 billion reals ($36 billion) was more than three times the runner-uprsquos bid andthat they were therefore paying 5 billion reals ($25 billion) more than was neededto win In other auctions meanwhile losers who have just narrowly underbid thewinners have found it equally hard to explain themselves to their bosses andshareholders So rms or at least their managers can oppose rst-price auctions

On the other hand a second-price sealed-bid auctionmdashin which the winner paysthe runner-uprsquos bidmdash can be embarrassing for the auctioneer if the winnerrsquos actualbid is revealed to be far more than the runner-uprsquos even if the auction design wasboth ef cient and maximized expected revenue McMillan (1994) reports asecond-price New Zealand auction in which the winner bid NZ $7 million but paidthe runner-uprsquos bid of NZ $5000 New Zealand should have set a minimum reserveprice that the winner had to pay but even if that had been politically possible the

Paul Klemperer 175

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

but did not bid again on blocks 6ndash10 the two companies would then live and letlive with neither company challenging the other on the otherrsquos half Exactly thathappened So the auction closed after just two rounds with each of the biddersacquiring half the blocks for the same low price ( Jehiel and Moldovanu 2001Grimm Riedel and Wolfstetter 2001)

Ascending auctions can also facilitate collusion by offering a mechanism forpunishing rivals The threat of punishment may be implicit for example it wasclear to T-Mobil that Mannesman would retaliate with high bids on blocks 1ndash5 ifT-Mobil continued bidding on blocks 6ndash10 But an ascending auction can alsoallow more explicit options for punishment

In a multilicense US spectrum auction in 1996ndash1997 US West was compet-ing vigorously with McLeod for lot number 378 a license in Rochester MinnesotaAlthough most bids in the auction had been in exact thousands of dollars USWest bid $313378 and $62378 for two licenses in Iowa in which it had earliershown no interest overbidding McLeod who had seemed to be the uncontestedhigh bidder for these licenses McLeod got the point that it was being punished forcompeting in Rochester and dropped out of that market Since McLeod madesubsequent higher bids on the Iowa licenses the ldquopunishmentrdquo bids cost US Westnothing (Cramton and Schwartz 1999)

A related phenomenon can arise in one special kind of sealed-bid auctionnamely a uniform-price auction in which each bidder submits a sealed bid statingwhat price it would pay for different quantities of a homogenous good likeelectricity (that is it submits a demand function) and then the good is sold at thesingle price determined by the lowest winning bid In this format bidders cansubmit bids that ensure that any deviation from a (tacit or explicit) collusiveagreement is severely punished each bidder bids very high prices for smallerquantities than its collusively agreed share Then if any bidder attempts to obtainmore than its agreed share (leaving other rms with less than their agreed shares)all bidders will have to pay these very high prices However if everyone sticks totheir agreed shares then these very high prices will never need to be paid As aresult deviation from the collusive agreement is unpro table1

The electricity regulator in the United Kingdom believes the market in whichdistribution companies purchase electricity from generating companies has fallenprey to exactly this kind of ldquoimplicit collusionrdquo (Of ce of Gas and ElectricityMarkets 1999 pp 173ndash174) ldquoFar from being the success story trumpeted aroundthe world the story of the UK generation market and the development ofcompetition has been something of a disasterrdquo reported Power UK (1999 see also

1 Since with many units the lowest winning bid in a uniform-price auction is typically not importantlydifferent from the highest losing bid this auction is analogous to an ascending auction (in which everywinner pays the runner-uprsquos willingness-to-pay) The ldquothreatsrdquo that support collusion in a uniform-priceauction are likewise analogous to the implicit threats supporting collusion in an ascending auctionCollusion in a uniform-price auction is harder if supply is uncertain since this reduces the number ofpoints on the bid schedule that are inframarginal and can be used as threats (Klemperer and Meyer1989 Back and Zender 1993 1999)

Paul Klemperer 171

von der Fehr and Harbord 1998 Newbery 1998 Wolfram 1998 1999) In addi-tion a frequently repeated auction market such as that for electricity is particularlyvulnerable to collusion because the repeated interaction among bidders expandsthe set of signaling and punishment strategies available to them and allows them tolearn to cooperate (Klemperer 2002)

Much of the kind of behavior discussed so far is hard to challenge legallyIndeed trying to outlaw it all would require cumbersome rules that would restrictbiddersrsquo exibility and might generate inef ciencies without being fully effectiveIt would be much better to solve these problems with better auction designs

Entry Deterrence and Predation

The second major area of concern of practical auction design is to attractbidders since an auction with too few bidders risks being unpro table for theauctioneer (Bulow and Klemperer 1996) and potentially inef cient Ascendingauctions are often particularly poor in this respect since they can allow somebidders to deter the entry or depress the bidding of rivals

In an ascending auction there is a strong presumption that the rm that valueswinning the most will be the eventual winner because even if it is outbid at an earlystage it can eventually top any opposition As a result other rms have littleincentive to enter the bidding and may not do so if they have even modest costs ofbidding

Consider for example Glaxorsquos 1995 takeover of the Wellcome drugs companyAfter Glaxorsquos rst bid of 9 billion pounds Zeneca expressed willingness to offerabout 10 billion pounds if it could be sure of winning while Roche considered anoffer of 11 billion pounds But certain synergies made Wellcome worth a little moreto Glaxo than to the other rms and the costs of bidding were tens of millions ofpounds Eventually neither Roche nor Zeneca actually entered the bidding andWellcome was sold at the original bid of 9 billion pounds literally a billion or twoless than its shareholders might have received Wellcomersquos own chief executiveadmitted ldquothere was money left on the tablerdquo (Wighton 1995a b)

While ascending auctions are particularly vulnerable to lack of entry otherauction forms can result in similar problems if the costs of entry and the asymme-tries between bidders are too large

The 1991 UK sale of television franchises by a sealed-bid auction is a dramaticexample While the regions in the South and Southeast Southwest East Wales andWest Northeast and Yorkshire all sold in the range of 936 to 1588 pounds perhead of population the onlymdashand therefore winningmdashbid for the Midlandsregion was made by the incumbent rm and was just one-twentieth of one penny ()per head of population Much the same happened in Scotland where the onlybidder for the Central region generously bid one-seventh of one penny per capitaWhat had happened was that bidders were required to provide very detailed

172 Journal of Economic Perspectives

region-speci c programming plans In each of these two regions the only bidder gured out that no one else had developed such a plan2

Another issue that can depress bidding in some ascending auctions is theldquowinnerrsquos curserdquo This problem applies when bidders have the same or close to thesame actual value for a prize but they have different information about that actualvalue (what auction theorists call the ldquocommon valuesrdquo case) The winnerrsquos cursere ects the danger that the winner of an auction is likely to be the party who hasmost greatly overestimated the value of the prize Knowing about the winnerrsquos cursewill cause everyone to bid cautiously But weaker rms must be especially cautioussince they must recognize that they are only likely to win when they have overes-timated the value by even more than usual Therefore an advantaged rm can beless cautious since beating very cautious opponents need not imply one hasoverestimated the prizersquos value Because the winnerrsquos curse affects weak rms muchmore than strong ones and because the effect is self-reinforcing the advantagedbidder wins most of the time And because its rivals bid extremely cautiously it alsogenerally pays a low price when it does win (Klemperer 1998)

The bidding on the Los Angeles license in the 1995 US auction for mobile-phone broadband licenses illustrates this problem While the licensersquos value washard to estimate it was probably worth similar amounts to several bidders ButPaci c Telephone which already operated the local xed-line telephone businessin California had distinct advantages from its database on potential local custom-ers its well-known brand-name and its familiarity with doing business in CaliforniaThe auction was an ascending one The result was that the bidding stopped at a verylow price In the end the Los Angeles license yielded only $26 per capita InChicago by contrast the main local xed-line provider was ineligible to competeand it was not obvious who would win so the auction yielded $31 per capita eventhough Chicago was thought less valuable than Los Angeles because of its lowerhousehold incomes lower expected population growth and more dispersed pop-ulation (Klemperer 1998 Bulow and Klemperer 2002) For formal econometricevidence for the FCC auctions more broadly see Klemperer and Pagnozzi (2002)

Of course the ldquowinnerrsquos curserdquo problem exacerbates the problem that weakerbidders may not bother to participate in an ascending auction GTE and BellAtlantic made deals that made them ineligible to bid for the Los Angeles licenseand MCI failed to enter this auction at all Similarly takeover battles are essentiallyascending auctions and there is empirical evidence that a rm that makes atakeover bid has a lower risk of facing a rival bidder if the rm has a largershareholding or ldquotoeholdrdquo in the target company (Betton and Eckbo 2000)

Because outcomes in an ascending auction can be dramatically in uenced bya seemingly modest advantage developing such an advantage can be an effectivepredatory strategy An apparent example was the 1999 attempt by BSkyB (RupertMurdochrsquos satellite television company) to acquire Manchester United (Englandrsquos

2 While I have advised the UK government on several auctions I have never had anything to do withtelevision licenses

What Really Matters in Auction Design 173

most successful soccer club) The problem was the advantage this would give BSkyBin the auction of football television rights Since Manchester United receives7 percent of the Premier Leaguersquos television revenues BSkyB would have received7 percent of the price of the leaguersquos broadcasting rights whoever won them SoBSkyB would have had an incentive to bid more aggressively in an ascendingauction to push up the price of the rights and knowing this other potential bidderswould have faced a worse ldquowinnerrsquos curserdquo and backed off BSkyB might have endedup with a lock over the television rights with damaging effects on the televisionmarket more generally Largely for this reason the UK government blocked theacquisition3

A strong bidder also has an incentive to create a reputation for aggressivenessthat reinforces its advantage For example when Glaxo was bidding for Wellcomeit made it clear that it ldquowould almost certainly top a rival bidrdquo (Wighton 1995b)Similarly before bidding for the California phone license Paci c Telephoneannounced in the Wall Street Journal that ldquoif somebody takes California away fromus theyrsquoll never make any moneyrdquo (Cauley and Carnevale 1994 p A4) Paci cTelephone also hired one of the worldrsquos most prominent auction theorists to giveseminars to the rest of the industry to explain the winnerrsquos curse argument thatjusti es this statement and it reinforced the point in full-page ads that ran in thenewspapers of the cities where its major competitors were headquartered (Koselka1995 p 63) It also made organizational changes that demonstrated its commit-ment to winning the Los Angeles license

Predation may be particularly easy in repeated ascending auctions such as ina series of spectrum auctions A bidder who buys assets that are complementary toassets for sale in a future auction or who simply bids very aggressively in earlyauctions can develop a reputation for aggressiveness (Bikhchandani 1988) Poten-tial rivals in future auctions will be less willing to participate and will bid lessaggressively if they do participate (Klemperer 2002)

Finally because an ascending auction often effectively blocks the entry ofldquoweakerrdquo bidders it encourages ldquostrongerrdquo bidders to bid jointly or to collude afterall they know that no one else can enter the auction to steal the collusive rents theycreate In the disastrous November 2000 Swiss sale of four third-generation mobile-phone licenses there was considerable initial interest from potential bidders Butweaker bidders were put off by the auction formmdashat least one company hiredbidding consultants and then gave up after learning that the ascending-biddingrules would give the company very little chance against stronger rivals Moreoverthe government permitted last-minute joint-bidding agreementsmdash essentially of -cially sanctioned collusion In the week before the auction the eld shrank fromnine bidders to just four bidders for the four licenses Since no bidder was allowed

3 Although the term ldquotoehold effectrdquo coined by Bulow Huang and Klemperer (1999) and Klemperer(1998) in the related context of takeover battles (see above) entered the popular press and thesepapers were cited by the UK Monopolies and Mergers Commission (1999) report which effectivelydecided the issue neither I nor my coauthors had any involvement in this case

174 Journal of Economic Perspectives

to take more than one license the sale price was determined by the reserve pricewhich was just one-thirtieth of the UK and German per capita revenues andone- ftieth of what the Swiss had once hoped for

Other Pitfalls

Reserve PricesMany of the disasters above were greatly aggravated by failure to set a proper

reserve price (the minimum amount the winner is required to pay) Take theprevious example It was ridiculous for the Swiss government to set its reserve at justone-thirtieth of the per capita revenue raised by the German and UK governmentsfor similar properties Since the governmentrsquos own spokesman predicted just vedays prior to the auction that twenty times the reserve price would be raised whatwas the government playing at

Inadequate reserve prices also increase the incentives for predation and mayencourage collusion that would not otherwise have been in all biddersrsquo interests Astronger bidder in an ascending auction has a choice between either tacitly collud-ing to end the auction quickly at a low price or forcing the price up to drive outweaker bidders The lower the reserve price at which the auction can be concludedthe more attractive is the rst option This factor may have been an importantcontributor to several of the ascoes we have discussed

Political ProblemsSerious reserve prices are often opposed not only by industry groups but also

by government of cials for whom a very embarrassing outcome is that the reserveprice is not met the object is not sold and the auction is seen as a ldquofailurerdquo

Similarly standard ( rst-price) sealed-bid auctionsmdashin which the bidders si-multaneously make ldquobest and nalrdquo offers and the winner pays the price hebidmdash can sometimes be very embarrassing for bidders as BSCH (Spainrsquos biggestbank) found out when Brazil privatized the Sao Paulo state bank Banespa Whenthe bids were opened BSCHrsquos managers were horri ed to learn that their bid ofover 7 billion reals ($36 billion) was more than three times the runner-uprsquos bid andthat they were therefore paying 5 billion reals ($25 billion) more than was neededto win In other auctions meanwhile losers who have just narrowly underbid thewinners have found it equally hard to explain themselves to their bosses andshareholders So rms or at least their managers can oppose rst-price auctions

On the other hand a second-price sealed-bid auctionmdashin which the winner paysthe runner-uprsquos bidmdash can be embarrassing for the auctioneer if the winnerrsquos actualbid is revealed to be far more than the runner-uprsquos even if the auction design wasboth ef cient and maximized expected revenue McMillan (1994) reports asecond-price New Zealand auction in which the winner bid NZ $7 million but paidthe runner-uprsquos bid of NZ $5000 New Zealand should have set a minimum reserveprice that the winner had to pay but even if that had been politically possible the

Paul Klemperer 175

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

von der Fehr and Harbord 1998 Newbery 1998 Wolfram 1998 1999) In addi-tion a frequently repeated auction market such as that for electricity is particularlyvulnerable to collusion because the repeated interaction among bidders expandsthe set of signaling and punishment strategies available to them and allows them tolearn to cooperate (Klemperer 2002)

Much of the kind of behavior discussed so far is hard to challenge legallyIndeed trying to outlaw it all would require cumbersome rules that would restrictbiddersrsquo exibility and might generate inef ciencies without being fully effectiveIt would be much better to solve these problems with better auction designs

Entry Deterrence and Predation

The second major area of concern of practical auction design is to attractbidders since an auction with too few bidders risks being unpro table for theauctioneer (Bulow and Klemperer 1996) and potentially inef cient Ascendingauctions are often particularly poor in this respect since they can allow somebidders to deter the entry or depress the bidding of rivals

In an ascending auction there is a strong presumption that the rm that valueswinning the most will be the eventual winner because even if it is outbid at an earlystage it can eventually top any opposition As a result other rms have littleincentive to enter the bidding and may not do so if they have even modest costs ofbidding

Consider for example Glaxorsquos 1995 takeover of the Wellcome drugs companyAfter Glaxorsquos rst bid of 9 billion pounds Zeneca expressed willingness to offerabout 10 billion pounds if it could be sure of winning while Roche considered anoffer of 11 billion pounds But certain synergies made Wellcome worth a little moreto Glaxo than to the other rms and the costs of bidding were tens of millions ofpounds Eventually neither Roche nor Zeneca actually entered the bidding andWellcome was sold at the original bid of 9 billion pounds literally a billion or twoless than its shareholders might have received Wellcomersquos own chief executiveadmitted ldquothere was money left on the tablerdquo (Wighton 1995a b)

While ascending auctions are particularly vulnerable to lack of entry otherauction forms can result in similar problems if the costs of entry and the asymme-tries between bidders are too large

The 1991 UK sale of television franchises by a sealed-bid auction is a dramaticexample While the regions in the South and Southeast Southwest East Wales andWest Northeast and Yorkshire all sold in the range of 936 to 1588 pounds perhead of population the onlymdashand therefore winningmdashbid for the Midlandsregion was made by the incumbent rm and was just one-twentieth of one penny ()per head of population Much the same happened in Scotland where the onlybidder for the Central region generously bid one-seventh of one penny per capitaWhat had happened was that bidders were required to provide very detailed

172 Journal of Economic Perspectives

region-speci c programming plans In each of these two regions the only bidder gured out that no one else had developed such a plan2

Another issue that can depress bidding in some ascending auctions is theldquowinnerrsquos curserdquo This problem applies when bidders have the same or close to thesame actual value for a prize but they have different information about that actualvalue (what auction theorists call the ldquocommon valuesrdquo case) The winnerrsquos cursere ects the danger that the winner of an auction is likely to be the party who hasmost greatly overestimated the value of the prize Knowing about the winnerrsquos cursewill cause everyone to bid cautiously But weaker rms must be especially cautioussince they must recognize that they are only likely to win when they have overes-timated the value by even more than usual Therefore an advantaged rm can beless cautious since beating very cautious opponents need not imply one hasoverestimated the prizersquos value Because the winnerrsquos curse affects weak rms muchmore than strong ones and because the effect is self-reinforcing the advantagedbidder wins most of the time And because its rivals bid extremely cautiously it alsogenerally pays a low price when it does win (Klemperer 1998)

The bidding on the Los Angeles license in the 1995 US auction for mobile-phone broadband licenses illustrates this problem While the licensersquos value washard to estimate it was probably worth similar amounts to several bidders ButPaci c Telephone which already operated the local xed-line telephone businessin California had distinct advantages from its database on potential local custom-ers its well-known brand-name and its familiarity with doing business in CaliforniaThe auction was an ascending one The result was that the bidding stopped at a verylow price In the end the Los Angeles license yielded only $26 per capita InChicago by contrast the main local xed-line provider was ineligible to competeand it was not obvious who would win so the auction yielded $31 per capita eventhough Chicago was thought less valuable than Los Angeles because of its lowerhousehold incomes lower expected population growth and more dispersed pop-ulation (Klemperer 1998 Bulow and Klemperer 2002) For formal econometricevidence for the FCC auctions more broadly see Klemperer and Pagnozzi (2002)

Of course the ldquowinnerrsquos curserdquo problem exacerbates the problem that weakerbidders may not bother to participate in an ascending auction GTE and BellAtlantic made deals that made them ineligible to bid for the Los Angeles licenseand MCI failed to enter this auction at all Similarly takeover battles are essentiallyascending auctions and there is empirical evidence that a rm that makes atakeover bid has a lower risk of facing a rival bidder if the rm has a largershareholding or ldquotoeholdrdquo in the target company (Betton and Eckbo 2000)

Because outcomes in an ascending auction can be dramatically in uenced bya seemingly modest advantage developing such an advantage can be an effectivepredatory strategy An apparent example was the 1999 attempt by BSkyB (RupertMurdochrsquos satellite television company) to acquire Manchester United (Englandrsquos

2 While I have advised the UK government on several auctions I have never had anything to do withtelevision licenses

What Really Matters in Auction Design 173

most successful soccer club) The problem was the advantage this would give BSkyBin the auction of football television rights Since Manchester United receives7 percent of the Premier Leaguersquos television revenues BSkyB would have received7 percent of the price of the leaguersquos broadcasting rights whoever won them SoBSkyB would have had an incentive to bid more aggressively in an ascendingauction to push up the price of the rights and knowing this other potential bidderswould have faced a worse ldquowinnerrsquos curserdquo and backed off BSkyB might have endedup with a lock over the television rights with damaging effects on the televisionmarket more generally Largely for this reason the UK government blocked theacquisition3

A strong bidder also has an incentive to create a reputation for aggressivenessthat reinforces its advantage For example when Glaxo was bidding for Wellcomeit made it clear that it ldquowould almost certainly top a rival bidrdquo (Wighton 1995b)Similarly before bidding for the California phone license Paci c Telephoneannounced in the Wall Street Journal that ldquoif somebody takes California away fromus theyrsquoll never make any moneyrdquo (Cauley and Carnevale 1994 p A4) Paci cTelephone also hired one of the worldrsquos most prominent auction theorists to giveseminars to the rest of the industry to explain the winnerrsquos curse argument thatjusti es this statement and it reinforced the point in full-page ads that ran in thenewspapers of the cities where its major competitors were headquartered (Koselka1995 p 63) It also made organizational changes that demonstrated its commit-ment to winning the Los Angeles license

Predation may be particularly easy in repeated ascending auctions such as ina series of spectrum auctions A bidder who buys assets that are complementary toassets for sale in a future auction or who simply bids very aggressively in earlyauctions can develop a reputation for aggressiveness (Bikhchandani 1988) Poten-tial rivals in future auctions will be less willing to participate and will bid lessaggressively if they do participate (Klemperer 2002)

Finally because an ascending auction often effectively blocks the entry ofldquoweakerrdquo bidders it encourages ldquostrongerrdquo bidders to bid jointly or to collude afterall they know that no one else can enter the auction to steal the collusive rents theycreate In the disastrous November 2000 Swiss sale of four third-generation mobile-phone licenses there was considerable initial interest from potential bidders Butweaker bidders were put off by the auction formmdashat least one company hiredbidding consultants and then gave up after learning that the ascending-biddingrules would give the company very little chance against stronger rivals Moreoverthe government permitted last-minute joint-bidding agreementsmdash essentially of -cially sanctioned collusion In the week before the auction the eld shrank fromnine bidders to just four bidders for the four licenses Since no bidder was allowed

3 Although the term ldquotoehold effectrdquo coined by Bulow Huang and Klemperer (1999) and Klemperer(1998) in the related context of takeover battles (see above) entered the popular press and thesepapers were cited by the UK Monopolies and Mergers Commission (1999) report which effectivelydecided the issue neither I nor my coauthors had any involvement in this case

174 Journal of Economic Perspectives

to take more than one license the sale price was determined by the reserve pricewhich was just one-thirtieth of the UK and German per capita revenues andone- ftieth of what the Swiss had once hoped for

Other Pitfalls

Reserve PricesMany of the disasters above were greatly aggravated by failure to set a proper

reserve price (the minimum amount the winner is required to pay) Take theprevious example It was ridiculous for the Swiss government to set its reserve at justone-thirtieth of the per capita revenue raised by the German and UK governmentsfor similar properties Since the governmentrsquos own spokesman predicted just vedays prior to the auction that twenty times the reserve price would be raised whatwas the government playing at

Inadequate reserve prices also increase the incentives for predation and mayencourage collusion that would not otherwise have been in all biddersrsquo interests Astronger bidder in an ascending auction has a choice between either tacitly collud-ing to end the auction quickly at a low price or forcing the price up to drive outweaker bidders The lower the reserve price at which the auction can be concludedthe more attractive is the rst option This factor may have been an importantcontributor to several of the ascoes we have discussed

Political ProblemsSerious reserve prices are often opposed not only by industry groups but also

by government of cials for whom a very embarrassing outcome is that the reserveprice is not met the object is not sold and the auction is seen as a ldquofailurerdquo

Similarly standard ( rst-price) sealed-bid auctionsmdashin which the bidders si-multaneously make ldquobest and nalrdquo offers and the winner pays the price hebidmdash can sometimes be very embarrassing for bidders as BSCH (Spainrsquos biggestbank) found out when Brazil privatized the Sao Paulo state bank Banespa Whenthe bids were opened BSCHrsquos managers were horri ed to learn that their bid ofover 7 billion reals ($36 billion) was more than three times the runner-uprsquos bid andthat they were therefore paying 5 billion reals ($25 billion) more than was neededto win In other auctions meanwhile losers who have just narrowly underbid thewinners have found it equally hard to explain themselves to their bosses andshareholders So rms or at least their managers can oppose rst-price auctions

On the other hand a second-price sealed-bid auctionmdashin which the winner paysthe runner-uprsquos bidmdash can be embarrassing for the auctioneer if the winnerrsquos actualbid is revealed to be far more than the runner-uprsquos even if the auction design wasboth ef cient and maximized expected revenue McMillan (1994) reports asecond-price New Zealand auction in which the winner bid NZ $7 million but paidthe runner-uprsquos bid of NZ $5000 New Zealand should have set a minimum reserveprice that the winner had to pay but even if that had been politically possible the

Paul Klemperer 175

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

region-speci c programming plans In each of these two regions the only bidder gured out that no one else had developed such a plan2

Another issue that can depress bidding in some ascending auctions is theldquowinnerrsquos curserdquo This problem applies when bidders have the same or close to thesame actual value for a prize but they have different information about that actualvalue (what auction theorists call the ldquocommon valuesrdquo case) The winnerrsquos cursere ects the danger that the winner of an auction is likely to be the party who hasmost greatly overestimated the value of the prize Knowing about the winnerrsquos cursewill cause everyone to bid cautiously But weaker rms must be especially cautioussince they must recognize that they are only likely to win when they have overes-timated the value by even more than usual Therefore an advantaged rm can beless cautious since beating very cautious opponents need not imply one hasoverestimated the prizersquos value Because the winnerrsquos curse affects weak rms muchmore than strong ones and because the effect is self-reinforcing the advantagedbidder wins most of the time And because its rivals bid extremely cautiously it alsogenerally pays a low price when it does win (Klemperer 1998)

The bidding on the Los Angeles license in the 1995 US auction for mobile-phone broadband licenses illustrates this problem While the licensersquos value washard to estimate it was probably worth similar amounts to several bidders ButPaci c Telephone which already operated the local xed-line telephone businessin California had distinct advantages from its database on potential local custom-ers its well-known brand-name and its familiarity with doing business in CaliforniaThe auction was an ascending one The result was that the bidding stopped at a verylow price In the end the Los Angeles license yielded only $26 per capita InChicago by contrast the main local xed-line provider was ineligible to competeand it was not obvious who would win so the auction yielded $31 per capita eventhough Chicago was thought less valuable than Los Angeles because of its lowerhousehold incomes lower expected population growth and more dispersed pop-ulation (Klemperer 1998 Bulow and Klemperer 2002) For formal econometricevidence for the FCC auctions more broadly see Klemperer and Pagnozzi (2002)

Of course the ldquowinnerrsquos curserdquo problem exacerbates the problem that weakerbidders may not bother to participate in an ascending auction GTE and BellAtlantic made deals that made them ineligible to bid for the Los Angeles licenseand MCI failed to enter this auction at all Similarly takeover battles are essentiallyascending auctions and there is empirical evidence that a rm that makes atakeover bid has a lower risk of facing a rival bidder if the rm has a largershareholding or ldquotoeholdrdquo in the target company (Betton and Eckbo 2000)

Because outcomes in an ascending auction can be dramatically in uenced bya seemingly modest advantage developing such an advantage can be an effectivepredatory strategy An apparent example was the 1999 attempt by BSkyB (RupertMurdochrsquos satellite television company) to acquire Manchester United (Englandrsquos

2 While I have advised the UK government on several auctions I have never had anything to do withtelevision licenses

What Really Matters in Auction Design 173

most successful soccer club) The problem was the advantage this would give BSkyBin the auction of football television rights Since Manchester United receives7 percent of the Premier Leaguersquos television revenues BSkyB would have received7 percent of the price of the leaguersquos broadcasting rights whoever won them SoBSkyB would have had an incentive to bid more aggressively in an ascendingauction to push up the price of the rights and knowing this other potential bidderswould have faced a worse ldquowinnerrsquos curserdquo and backed off BSkyB might have endedup with a lock over the television rights with damaging effects on the televisionmarket more generally Largely for this reason the UK government blocked theacquisition3

A strong bidder also has an incentive to create a reputation for aggressivenessthat reinforces its advantage For example when Glaxo was bidding for Wellcomeit made it clear that it ldquowould almost certainly top a rival bidrdquo (Wighton 1995b)Similarly before bidding for the California phone license Paci c Telephoneannounced in the Wall Street Journal that ldquoif somebody takes California away fromus theyrsquoll never make any moneyrdquo (Cauley and Carnevale 1994 p A4) Paci cTelephone also hired one of the worldrsquos most prominent auction theorists to giveseminars to the rest of the industry to explain the winnerrsquos curse argument thatjusti es this statement and it reinforced the point in full-page ads that ran in thenewspapers of the cities where its major competitors were headquartered (Koselka1995 p 63) It also made organizational changes that demonstrated its commit-ment to winning the Los Angeles license

Predation may be particularly easy in repeated ascending auctions such as ina series of spectrum auctions A bidder who buys assets that are complementary toassets for sale in a future auction or who simply bids very aggressively in earlyauctions can develop a reputation for aggressiveness (Bikhchandani 1988) Poten-tial rivals in future auctions will be less willing to participate and will bid lessaggressively if they do participate (Klemperer 2002)

Finally because an ascending auction often effectively blocks the entry ofldquoweakerrdquo bidders it encourages ldquostrongerrdquo bidders to bid jointly or to collude afterall they know that no one else can enter the auction to steal the collusive rents theycreate In the disastrous November 2000 Swiss sale of four third-generation mobile-phone licenses there was considerable initial interest from potential bidders Butweaker bidders were put off by the auction formmdashat least one company hiredbidding consultants and then gave up after learning that the ascending-biddingrules would give the company very little chance against stronger rivals Moreoverthe government permitted last-minute joint-bidding agreementsmdash essentially of -cially sanctioned collusion In the week before the auction the eld shrank fromnine bidders to just four bidders for the four licenses Since no bidder was allowed

3 Although the term ldquotoehold effectrdquo coined by Bulow Huang and Klemperer (1999) and Klemperer(1998) in the related context of takeover battles (see above) entered the popular press and thesepapers were cited by the UK Monopolies and Mergers Commission (1999) report which effectivelydecided the issue neither I nor my coauthors had any involvement in this case

174 Journal of Economic Perspectives

to take more than one license the sale price was determined by the reserve pricewhich was just one-thirtieth of the UK and German per capita revenues andone- ftieth of what the Swiss had once hoped for

Other Pitfalls

Reserve PricesMany of the disasters above were greatly aggravated by failure to set a proper

reserve price (the minimum amount the winner is required to pay) Take theprevious example It was ridiculous for the Swiss government to set its reserve at justone-thirtieth of the per capita revenue raised by the German and UK governmentsfor similar properties Since the governmentrsquos own spokesman predicted just vedays prior to the auction that twenty times the reserve price would be raised whatwas the government playing at

Inadequate reserve prices also increase the incentives for predation and mayencourage collusion that would not otherwise have been in all biddersrsquo interests Astronger bidder in an ascending auction has a choice between either tacitly collud-ing to end the auction quickly at a low price or forcing the price up to drive outweaker bidders The lower the reserve price at which the auction can be concludedthe more attractive is the rst option This factor may have been an importantcontributor to several of the ascoes we have discussed

Political ProblemsSerious reserve prices are often opposed not only by industry groups but also

by government of cials for whom a very embarrassing outcome is that the reserveprice is not met the object is not sold and the auction is seen as a ldquofailurerdquo

Similarly standard ( rst-price) sealed-bid auctionsmdashin which the bidders si-multaneously make ldquobest and nalrdquo offers and the winner pays the price hebidmdash can sometimes be very embarrassing for bidders as BSCH (Spainrsquos biggestbank) found out when Brazil privatized the Sao Paulo state bank Banespa Whenthe bids were opened BSCHrsquos managers were horri ed to learn that their bid ofover 7 billion reals ($36 billion) was more than three times the runner-uprsquos bid andthat they were therefore paying 5 billion reals ($25 billion) more than was neededto win In other auctions meanwhile losers who have just narrowly underbid thewinners have found it equally hard to explain themselves to their bosses andshareholders So rms or at least their managers can oppose rst-price auctions

On the other hand a second-price sealed-bid auctionmdashin which the winner paysthe runner-uprsquos bidmdash can be embarrassing for the auctioneer if the winnerrsquos actualbid is revealed to be far more than the runner-uprsquos even if the auction design wasboth ef cient and maximized expected revenue McMillan (1994) reports asecond-price New Zealand auction in which the winner bid NZ $7 million but paidthe runner-uprsquos bid of NZ $5000 New Zealand should have set a minimum reserveprice that the winner had to pay but even if that had been politically possible the

Paul Klemperer 175

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

most successful soccer club) The problem was the advantage this would give BSkyBin the auction of football television rights Since Manchester United receives7 percent of the Premier Leaguersquos television revenues BSkyB would have received7 percent of the price of the leaguersquos broadcasting rights whoever won them SoBSkyB would have had an incentive to bid more aggressively in an ascendingauction to push up the price of the rights and knowing this other potential bidderswould have faced a worse ldquowinnerrsquos curserdquo and backed off BSkyB might have endedup with a lock over the television rights with damaging effects on the televisionmarket more generally Largely for this reason the UK government blocked theacquisition3

A strong bidder also has an incentive to create a reputation for aggressivenessthat reinforces its advantage For example when Glaxo was bidding for Wellcomeit made it clear that it ldquowould almost certainly top a rival bidrdquo (Wighton 1995b)Similarly before bidding for the California phone license Paci c Telephoneannounced in the Wall Street Journal that ldquoif somebody takes California away fromus theyrsquoll never make any moneyrdquo (Cauley and Carnevale 1994 p A4) Paci cTelephone also hired one of the worldrsquos most prominent auction theorists to giveseminars to the rest of the industry to explain the winnerrsquos curse argument thatjusti es this statement and it reinforced the point in full-page ads that ran in thenewspapers of the cities where its major competitors were headquartered (Koselka1995 p 63) It also made organizational changes that demonstrated its commit-ment to winning the Los Angeles license

Predation may be particularly easy in repeated ascending auctions such as ina series of spectrum auctions A bidder who buys assets that are complementary toassets for sale in a future auction or who simply bids very aggressively in earlyauctions can develop a reputation for aggressiveness (Bikhchandani 1988) Poten-tial rivals in future auctions will be less willing to participate and will bid lessaggressively if they do participate (Klemperer 2002)

Finally because an ascending auction often effectively blocks the entry ofldquoweakerrdquo bidders it encourages ldquostrongerrdquo bidders to bid jointly or to collude afterall they know that no one else can enter the auction to steal the collusive rents theycreate In the disastrous November 2000 Swiss sale of four third-generation mobile-phone licenses there was considerable initial interest from potential bidders Butweaker bidders were put off by the auction formmdashat least one company hiredbidding consultants and then gave up after learning that the ascending-biddingrules would give the company very little chance against stronger rivals Moreoverthe government permitted last-minute joint-bidding agreementsmdash essentially of -cially sanctioned collusion In the week before the auction the eld shrank fromnine bidders to just four bidders for the four licenses Since no bidder was allowed

3 Although the term ldquotoehold effectrdquo coined by Bulow Huang and Klemperer (1999) and Klemperer(1998) in the related context of takeover battles (see above) entered the popular press and thesepapers were cited by the UK Monopolies and Mergers Commission (1999) report which effectivelydecided the issue neither I nor my coauthors had any involvement in this case

174 Journal of Economic Perspectives

to take more than one license the sale price was determined by the reserve pricewhich was just one-thirtieth of the UK and German per capita revenues andone- ftieth of what the Swiss had once hoped for

Other Pitfalls

Reserve PricesMany of the disasters above were greatly aggravated by failure to set a proper

reserve price (the minimum amount the winner is required to pay) Take theprevious example It was ridiculous for the Swiss government to set its reserve at justone-thirtieth of the per capita revenue raised by the German and UK governmentsfor similar properties Since the governmentrsquos own spokesman predicted just vedays prior to the auction that twenty times the reserve price would be raised whatwas the government playing at

Inadequate reserve prices also increase the incentives for predation and mayencourage collusion that would not otherwise have been in all biddersrsquo interests Astronger bidder in an ascending auction has a choice between either tacitly collud-ing to end the auction quickly at a low price or forcing the price up to drive outweaker bidders The lower the reserve price at which the auction can be concludedthe more attractive is the rst option This factor may have been an importantcontributor to several of the ascoes we have discussed

Political ProblemsSerious reserve prices are often opposed not only by industry groups but also

by government of cials for whom a very embarrassing outcome is that the reserveprice is not met the object is not sold and the auction is seen as a ldquofailurerdquo

Similarly standard ( rst-price) sealed-bid auctionsmdashin which the bidders si-multaneously make ldquobest and nalrdquo offers and the winner pays the price hebidmdash can sometimes be very embarrassing for bidders as BSCH (Spainrsquos biggestbank) found out when Brazil privatized the Sao Paulo state bank Banespa Whenthe bids were opened BSCHrsquos managers were horri ed to learn that their bid ofover 7 billion reals ($36 billion) was more than three times the runner-uprsquos bid andthat they were therefore paying 5 billion reals ($25 billion) more than was neededto win In other auctions meanwhile losers who have just narrowly underbid thewinners have found it equally hard to explain themselves to their bosses andshareholders So rms or at least their managers can oppose rst-price auctions

On the other hand a second-price sealed-bid auctionmdashin which the winner paysthe runner-uprsquos bidmdash can be embarrassing for the auctioneer if the winnerrsquos actualbid is revealed to be far more than the runner-uprsquos even if the auction design wasboth ef cient and maximized expected revenue McMillan (1994) reports asecond-price New Zealand auction in which the winner bid NZ $7 million but paidthe runner-uprsquos bid of NZ $5000 New Zealand should have set a minimum reserveprice that the winner had to pay but even if that had been politically possible the

Paul Klemperer 175

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

to take more than one license the sale price was determined by the reserve pricewhich was just one-thirtieth of the UK and German per capita revenues andone- ftieth of what the Swiss had once hoped for

Other Pitfalls

Reserve PricesMany of the disasters above were greatly aggravated by failure to set a proper

reserve price (the minimum amount the winner is required to pay) Take theprevious example It was ridiculous for the Swiss government to set its reserve at justone-thirtieth of the per capita revenue raised by the German and UK governmentsfor similar properties Since the governmentrsquos own spokesman predicted just vedays prior to the auction that twenty times the reserve price would be raised whatwas the government playing at

Inadequate reserve prices also increase the incentives for predation and mayencourage collusion that would not otherwise have been in all biddersrsquo interests Astronger bidder in an ascending auction has a choice between either tacitly collud-ing to end the auction quickly at a low price or forcing the price up to drive outweaker bidders The lower the reserve price at which the auction can be concludedthe more attractive is the rst option This factor may have been an importantcontributor to several of the ascoes we have discussed

Political ProblemsSerious reserve prices are often opposed not only by industry groups but also

by government of cials for whom a very embarrassing outcome is that the reserveprice is not met the object is not sold and the auction is seen as a ldquofailurerdquo

Similarly standard ( rst-price) sealed-bid auctionsmdashin which the bidders si-multaneously make ldquobest and nalrdquo offers and the winner pays the price hebidmdash can sometimes be very embarrassing for bidders as BSCH (Spainrsquos biggestbank) found out when Brazil privatized the Sao Paulo state bank Banespa Whenthe bids were opened BSCHrsquos managers were horri ed to learn that their bid ofover 7 billion reals ($36 billion) was more than three times the runner-uprsquos bid andthat they were therefore paying 5 billion reals ($25 billion) more than was neededto win In other auctions meanwhile losers who have just narrowly underbid thewinners have found it equally hard to explain themselves to their bosses andshareholders So rms or at least their managers can oppose rst-price auctions

On the other hand a second-price sealed-bid auctionmdashin which the winner paysthe runner-uprsquos bidmdash can be embarrassing for the auctioneer if the winnerrsquos actualbid is revealed to be far more than the runner-uprsquos even if the auction design wasboth ef cient and maximized expected revenue McMillan (1994) reports asecond-price New Zealand auction in which the winner bid NZ $7 million but paidthe runner-uprsquos bid of NZ $5000 New Zealand should have set a minimum reserveprice that the winner had to pay but even if that had been politically possible the

Paul Klemperer 175

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

winner would probably have bid more than it had to pay so this might have beenan economically but not politically sensible auction

LoopholesIn some cases the auction rules may leave gaping loopholes for behavior to

game the auction In 2000 Turkey auctioned two telecom licenses sequentiallywith an additional twist that set the reserve price for the second license equal to theselling price of the rst One rm then bid far more for the rst license than itcould possibly be worth if the rm had to compete in the telecom market with arival holding the second license But the rm had rightly gured that no rival wouldbe willing to bid that high for the second license which therefore remained unsoldleaving the rm without a rival operating the second license

As another example McMillan (1994) reports an Australian auction forsatellite-television licenses in which two bidders each made large numbers ofdifferent sealed bids on the same objects and then after considerable delaysdefaulted on those bids they did not like after the factmdashsince the government hadneglected to impose any penalties for default More recently the US spectrumauctions have been plagued by bidders ldquowinningrdquo licenses and subsequently de-faulting on their commitments often after long delays (Spectrum auctions in Indiaalso recently fell into the same trap) If default costs are small then bidders arebidding for options on prizes rather than the prizes themselves Furthermore ifsmaller under nanced rms can avoid commitments through bankruptcy then anauction actually favors these bidders over better- nanced competitors who cannotdefault

Credibility of the RulesIt may not be credible for the auctioneer to punish a bidder violating the

auction rules when just one bidder needs to be eliminated to end an auctionbecause excluding the offending bidder would end the auction immediately and itmight be hard to impose nes large enough to have a serious deterrent effect Finesof hundreds of millions or even billions of dollars might have been required todeter improper behavior in some of the European third-generation mobile-phonelicense auctions In the Netherlands sale for example six bidders competed for velicenses in an ascending auction in which bidders were permitted to win just onelicense each One bidder Telfort sent a letter to another Versatel threateninglegal action for damages if Versatel continued to bid Telfort claimed that Versatelldquobelieves that its bids will always be surpassed by [othersrsquo so it] must be thatVersatel is attempting to either raise its competitorsrsquo costs or to get access totheir networksrdquo Many observers felt Telfortrsquos threats against Versatel wereoutrageous However the government took no actionmdashnot even an investigationAs a result Versatel quit the auction and the sale raised less than 30 percent ofwhat the Dutch government had forecast based on the results of the UnitedKingdomrsquos similar auction just three months earlier

Ascending auctions are particularly vulnerable to rule breaking by the bidders

176 Journal of Economic Perspectives

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

since they necessarily pass through a stage where there is just one or a few excessbidders and the ascending structure allows a cheat time to assess the success of itsstrategy (Klemperer 2001b 2002) Sealed-bid auctions by contrast may be morevulnerable to rule changing by the auctioneer For example excuses for notaccepting a winning bid can often be found if losing bidders are willing to bidhigher The famous RJR-Nabisco sale went through several supposedly nal sealed-bid auctions (Burrough and Helyar 1990) But if after a sealed-bid auction theauctioneer can reopen the auction to higher offers the auction is really anascending-bid auction and needs to be recognized as such In fact genuine sealed-bid auctions may be dif cult to run in takeover battles especially since a directorwho turns down a higher bid for his company after running a ldquosealed-bid auctionrdquomay be vulnerable to shareholder lawsuits

Sealed-bid auctions can also be especially hard to commit to if the auctioneerhas any association with a bidder as for example would have been the case in theUK football television rights auction discussed earlier if BSkyB (a bidder) hadtaken over Manchester United (an in uential member of the football league whichwas the auctioneer)

Committing to future behavior may be a particular problem for governmentsFor example it may be dif cult to auction a license if the regulatory regime maychange but binding future governments (or even the current government) to aparticular regulatory regime may prove dif cult

The credibility of reserve prices is of special importance If a reserve price isnot a genuine commitment not to sell an object if it does not reach its reserve thenit has no meaning and bidders will treat it as such For example returning to theTurkish tale of woe the government is now considering new arrangements to sellthe second license but at what cost to the credibility of its future auctions4

Market StructureIn some auctions for example of mobile-phone licenses the structure of the

industry that will be created cannot be ignored by the auction designer It istempting simply to ldquolet the market deciderdquo the industry structure by auctioningmany small packages of spectrum which individual rms can aggregate into largerlicenses But the outcome of an auction is driven by biddersrsquo pro ts not by thewelfare of consumers or society as a whole

The most obvious possible distortion is that since rmsrsquo joint pro ts in amarket are generally greater if fewer competitors are in the market it is worth moreto any group of rms to prevent entry of an additional rm than the additional rmis willing to pay to enter As a result too few rms may win a share of spectrum andthese winners may each win too much in just the same way as a ldquohands-offrdquo policy

4 Reauctioning with a lower reserve price after a delay may sometimes be sensible to allow further entryif there are high costs of entering the auction (Burguet and Sakovics 1996 McAfee and McMillan1988) but in this case the auctioneer should make clear in advance what will happen if the reserve is notmet

What Really Matters in Auction Design 177

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

to merger control will tend to create an overly concentrated industry The Turkish asco discussed earlier was a spectacular example of how an auction can be biasedtoward generating a monopoly5

But this outcome is not the only socially suboptimal possibility A rm with alarge demand may prefer to reduce its demand to end the auction at a low pricerather than raise the price to drive out its rivals even when the latter course wouldbe socially more ef cient (Ausubel and Cramton 1998) There can also be toomany winners if rms collude to divide the spoils at a low price In the Austrianthird-generation mobile spectrum sale for example six rms competed for twelveidentical lots in an ascending auction and not surprisingly seemed to agree todivide the market so each rm won two lots each at not much more than the verylow reserve price Perhaps six winners was the ef cient outcome But we certainlycannot tell from the behavior in the auction It was rumored that the bidding lastedonly long enough to create some public perception of genuine competition and toreduce the risk of the government changing the rules

Thus it may sometimes be wiser to predetermine the number of winners byauctioning off fewer larger licenses but limiting bidders to one license apiecerather than to auction many licenses and to allow bidders to buy as many as theywish

When is Auction Design Less ImportantThe fact that collusion entry deterrence and more generally buyer market

power is the key to auction problems suggests that auction design may not mattervery much when there is a large number of potential bidders for whom entry to theauction is easy For example though much ink has been spilt on the subject ofgovernment security sales auction design may not matter much for either price oref ciency in this case Indeed the US Treasuryrsquos recent experiments with differentkinds of auctions yielded inconclusive results (Simon 1994 Malvey Archibald andFlynn 1996 Nyborg and Sundaresan 1996 Reinhart and Belzar 1996 Ausubeland Cramton 1998) and the broader empirical literature is also inconclusive Ofcourse even small differences in auction performance can be signi cant when suchlarge amounts of money are involved and collusion has been an issue in somegovernment security sales so further research is still warranted6

Solutions

Making the Ascending Auction More RobustMuch of our discussion has emphasized the vulnerability of ascending auctions

to collusion and predatory behavior However ascending auctions have several

5 Similarly the recent July 2001 Greek second-generation spectrum auction led to a more concentratedtelecom market than seems likely to be socially ef cient6 These views are personal I have advised UK government agencies on the related issue of the sale ofgold See Klemperer (1999b) for more discussion

178 Journal of Economic Perspectives

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

virtues as well An ascending auction is particularly likely to allocate the prizes tothe bidders who value them the most since a bidder with a higher value always hasthe opportunity to rebid to top a lower-value bidder who may initially have bidmore aggressively7 Moreover if there are complementarities between the objectsfor sale a multiunit ascending auction makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction in which they can learn nothingabout their opponentsrsquo intentions Allowing bidders to learn about othersrsquo valua-tions during the auction can also make the bidders more comfortable with theirown assessments and less cautious and it often raises the auctioneerrsquos revenues ifinformation is ldquoaf liatedrdquo in the sense of Milgrom and Weber (1982)

A number of methods to make the ascending auction more robust are clearenough For example bidders can be forced to bid ldquoroundrdquo numbers the exactincrements can be prespeci ed and bids can be made anonymous These stepsmake it harder to use bids to signal other buyers Lots can be aggregated into largerpackages to make it harder for bidders to divide the spoils and keeping secret thenumber of bidders remaining in the auction also makes collusion harder (Cramtonand Schwartz 2000 Salant 2000) Ausubelrsquos (1998) suggested modi cation of theascending auction mitigates the incentive of bidders to reduce their demands toend the auction quickly at a low price Sometimes it is possible to pay bidders toenter an auction for example ldquowhite knightsrdquo can be offered options to enter atakeover battle against an advantaged bidder

But while these measures can be useful they do not eliminate the risks ofcollusion or of too few bidders An alternative is to choose a different type ofauction

Using Sealed-Bid AuctionsIn a standard sealed-bid auction (or ldquo rst-pricerdquo sealed-bid auction) each

bidder simultaneously makes a single ldquobest and nalrdquo offer As a result rms areunable to retaliate against bidders who fail to cooperate with them so collusion ismuch harder than in an ascending auction Tacit collusion is particularly dif cultsince rms are unable to use the bidding to signal True both signaling andretaliation are possible in a series of sealed-bid auctions but collusion is still usuallyharder than in a series of ascending auctions

From the perspective of encouraging more entry the merit of a sealed-bidauction is that the outcome is much less certain than in an ascending auction An

7 This applies in many ldquocommon valuesrdquo and ldquoprivate valuesrdquo settings (Maskin 1992) but is notnecessarily the same as maximizing ef ciency When bidders are rms it ignores consumer welfare(which is likely to favor a more widely dispersed ownership than rms would choose) and of course itignores government revenue We assume governments (as well as other auctioneers) care about revenuebecause of the substantial deadweight losses (perhaps 33 cents per dollar raised) of raising governmentfunds through alternative methods (Ballard Shoven and Whalley 1985) Resale is not a perfectsubstitute for an ef cient initial allocation because even costless resale cannot usually ensure anef cient outcome in the presence of incomplete information (Myerson and Satterthwaite 1983Cramton Gibbons and Klemperer 1987)

Paul Klemperer 179

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

advantaged bidder will probably win a sealed-bid auction but it must make its single nal offer in the face of uncertainty about its rivalsrsquo bids and because it wants to geta bargain its sealed-bid will not be the maximum it could be pushed to in anascending auction So ldquoweakerrdquo bidders have at least some chance of victory evenwhen they would surely lose an ascending auction (Vickrey 1961 appendix III) Itfollows that potential entrants are likely to be more willing to enter a sealed-bidauction than an ascending auction

A sealed-bid auction might even encourage bidders who enter only to resellfurther increasing the competitiveness of the auction Such bidders seem less likelyto enter an ascending auction since it is generally more dif cult to pro t fromreselling to rms one has beaten in an ascending auction

Because sealed-bid auctions are more attractive to entrants they may alsodiscourage consortia from forming If the strong rms form a consortium they maysimply attract other rms into the bidding in the hope of beating the consortiumSo strong rms are more likely to bid independently in a sealed-bid auctionmaking this auction much more competitive

Consistent with all this there is some evidence from timber sales that sealed-bid auctions attract more bidders than ascending auctions do and that this makessealed-bid auctions considerably more pro table for the seller and this seems to bebelieved in this industry (Mead and Schneipp 1989 Rothkopf and Engelbrecht-Wiggans 1993) even though conditional on the number of bidders sealed-bidauctions seem only slightly more pro table than ascending auctions (Hansen1986)

Furthermore in the ldquocommon valuesrdquo case that bidders have similar actualvalues for a prize the ldquowinnerrsquos curserdquo problem for a weaker bidder is far less severein a sealed-bid auction Winning an ascending auction means the weaker bidder ispaying a price that the stronger rival is unwilling to matchmdashwhich should make theweaker bidder very nervous But the weaker player has a chance of winning asealed-bid auction at a price the stronger rival would be willing to match but didnrsquotSince beating the stronger player isnrsquot necessarily bad news in a sealed-bid auctionthe weaker player can bid more aggressively So auction prices will be higher evenfor a given number of bidders (Klemperer 1998 Bulow Huang and Klemperer1999)8

But while sealed-bid auctions have many advantages they are not without awsMainly by giving some chance of victory to weaker bidders sealed-bid auctions areless likely than ascending auctions to lead to ef cient outcomes Moreover instandard sealed-bid auctions in which winners pay their own bids bidders need tohave good information about the distribution of their rivalsrsquo values to bid intelli-gently (Persico 2000) By contrast in an ascending or uniform-price auction the

8 In Milgrom and Weberrsquos (1982) model sealed-bid auctions are less pro table than ascending auctionsif signals are ldquoaf liatedrdquo But they assume symmetric bidders and the effect does not seem large inpractice (Riley and Li 1997) Sealed-bid auctions are generally more pro table if bidders are risk averseor budget constrained (Klemperer 2000a)

180 Journal of Economic Perspectives

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

best strategy of a bidder who knows its own value is just to bid up to that value andwinnersrsquo payments are determined by the bids of nonwinners So ldquopay-your-bidrdquosealed-bid auctions may discourage potential bidders who have only small amountsto trade and for whom the costs of obtaining market information might not beworth paying For example in March 2001 the UK electricity regulator replacedthe problematic uniform-price auction we described earlier by an exchange marketfollowed by a ldquopay-your-bidrdquo sealed-bid auction which makes collusion harderbecause bids can no longer be used as costless threats But a major concern is thatthe new trading arrangements may deter potential entrants from investing the sunkcosts necessary to enter the electricity market9

However the entry problem in many-unit auctions is much less serious if smallbidders can buy from larger intermediaries who can aggregate smaller biddersrsquo de-mands and bid in their place as for example occurs in auctions of Treasury bills Andthe entry problem is also alleviated if smaller bidders are permitted to make ldquononcom-petitive bidsrdquo that is to state demands for xed quantities for which they pay theaverage winning price as is also the case in some Treasury bill auctions

The Anglo-Dutch AuctionA solution to the dilemma of choosing between the ascending (often called

ldquoEnglishrdquo) and sealed-bid (or ldquoDutchrdquo) forms is to combine the two into a hybridthe ldquoAnglo-Dutchrdquo which often captures the best features of both and was rstdescribed and proposed in Klemperer (1998)

For simplicity assume a single object is to be auctioned In an Anglo-Dutchauction the auctioneer begins by running an ascending auction in which price israised continuously until all but two bidders have dropped out The two remainingbidders are then each required to make a nal sealed-bid offer that is not lowerthan the current asking price and the winner pays the winning bid The process ismuch like the way houses are often sold although unlike in many house sales theprocedure the auctioneer will follow in an Anglo-Dutch auction is clearly speci edin advance

Another auction with similar featuresmdashand probably similar motivations to theAnglo-Dutchmdashis WR Hambrechtrsquos OpenBook auction for corporate bonds Theearly bidding is public and ascending but bidders can make nal sealed bids in thelast hour Although all bidders are permitted to make nal bids higher bidders inthe rst stages are given an advantage that is evidently large enough to induceserious bidding early on (Hall 2001 p 71)

The process also has some similarity to auctions on eBay (by far the worldrsquos mostsuccessful e-commerce auctioneer) which are ascending auctions but with a xedending time so that many bidders often bid only in the last few seconds in essentiallysealed-bid style eBay attracts far more bidders than its rival Yahoo which runs a

9 Also the new arrangements may not fully resolve the collusion problem anyway since the market is sofrequently repeated (Klemperer 1999b)

What Really Matters in Auction Design 181

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

standard ascending auction with a traditional ldquogoing going gonerdquo procedure that doesnot close the auction until there have been no bids for 10 minutes

The main value of the Anglo-Dutch procedure arises when one bidder (forexample the incumbent operator of a license that is to be reauctioned) is thoughtto be stronger than potential rivals Potential rivals might be unwilling to enter apure ascending-bid auction against the strong bidder who would be perceived tobe a sure winner But the sealed bid at the nal stage induces some uncertaintyabout which of the two nalists will win and entrants are attracted by the knowl-edge that they have a chance to make it to this nal stage So the price may easilybe higher even by the end of the rst ascending stage of the Anglo-Dutch auctionthan if a pure ascending auction were used

The Anglo-Dutch should capture the other advantages of the sealed-bid auc-tion discussed in the previous section Collusion will be discouraged because the nal sealed-bid round allows rms to renege on any deals without fear of retaliationand because the Anglo-Dutch auction eliminates the stage of the ascending auctionwhen just one excess bidder remains at which point the rules against collusion andpredation may not be credible

Consortium formation will also be discouraged Imagine there are two strongbidders for an item In an ascending auction they are unlikely to be challenged ifthey form a consortium so they have an incentive to do so But in an Anglo-Dutchauction forming the consortium would open up an opportunity for new entrantswho would now have a chance to make it to the nal sealed-bid stage So the strong rms are much less likely to bid jointly

But the Anglo-Dutch should also capture much of the bene t of an ascendingauction It will be more likely to sell to the highest valuer than a pure sealed-bidauction both because it directly reduces the numbers allowed into the sealed-bidstage and also because the two nalists can learn something about each otherrsquos andthe remaining biddersrsquo perceptions of the objectrsquos value from behavior during theascending stage

When the Anglo-Dutch auction is extended to contexts in which individualbidders are permitted to win multiple units and there are complementaritiesbetween the objects the ascending stage makes it more likely that bidders will winef cient bundles than in a pure sealed-bid auction

Finally I conjecture that the ascending stages of the Anglo-Dutch auction mayextract most of the information that would be revealed by a pure ascendingauction raising revenues if biddersrsquo information is ldquoaf liatedrdquo while the sealed-bidstage may do almost as well as a pure sealed-bid auction in capturing extra revenuesdue to the effects of biddersrsquo risk aversion budget constraints and asymmetriesThis suggests the Anglo-Dutch auction may outperform ascending and sealed-bidauctions even if it attracts no additional bidders

In short the Anglo-Dutch auction often combines the best of both the ascend-ing and the sealed-bid worlds

182 Journal of Economic Perspectives

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

AntitrustEffective antitrust is critical to ghting collusion and predation in auctions But

antitrust enforcement in the context of auctions seems much lighter than inldquoordinaryrdquo economic markets

The US Department of Justice has pursued some auction signaling cases butthe legal status of many of the kinds of behavior discussed in this article remainsambiguous and collusion in takeover battles for companies is legal in the UnitedStates

European antitrust has been even weaker as evidenced by T-Mobilrsquos willing-ness to con rm explicitly the signaling behavior described earlier True whenapparently similar behavior was observed in the more recent German third-generation spectrum auction rms refused to con rm of cially that they weresignaling to rivals to end the auction Even so the Financial Times reported thatldquo[o]ne operator has privately admitted to altering the last digit of its bid in asemi-serious attempt to signal to other participants that it was willing to accept[fewer lots to end the auction]rdquo (Roberts and Ward 2000 p 21) This kind ofsignaling behavior could perhaps be challenged as an abuse of ldquojoint dominancerdquounder European law But European regulators have showed no interest in pursuingsuch matters

Firms are also permitted to make explicit statements about auctions that wouldsurely be unacceptable if made about a ldquonormalrdquo economic market For examplebefore the Austrian third-generation spectrum auction Telekom Austria the larg-est incumbent and presumably the strongest among the six bidders said it ldquowouldbe satis ed with just two of the 12 blocks of frequency on offerrdquo and ldquoif the [5 otherbidders] behaved similarly it should be possible to get the frequencies on sensibletermsrdquo but ldquoit would bid for a third frequency block if one of its rivals didrdquo(Crossland 2000) It seems inconceivable that a dominant rm in a ldquonormalrdquomarket would be allowed to make the equivalent offer and threat that it ldquowould besatis ed with a market share of just one-sixthrdquo and ldquoif the other ve rms also stickto one-sixth of the market each it should be possible to sell at high pricesrdquo but ldquoitwould compete aggressively for a larger share if any of its rivals aimed for morethan one-sixthrdquo10

Just as damaging has been the European authoritiesrsquo acceptance of joint-bidding agreements that are in effect open collusion Combinations that arearranged very close to the auction date (as in the example of Switzerland discussedearlier) should be particularly discouraged since they give no time for entrants toemerge to threaten the new coalition One view is that auction participants should

10 Similarly during the German third-generation spectrum auction MobilCom told a newspaper thatldquoshould [Debitel] fail to secure a license [it could] become a lsquovirtual network operatorrsquo using Mobil-Comrsquos network while saving on the cost of the licenserdquo (Benoit 2000 p 28) This translates roughly toa rm in a ldquonormalrdquo market saying it ldquowould supply a rival should it choose to exit the marketrdquo butMobilComrsquos remarks went unpunished

Paul Klemperer 183

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

generally be restricted to entities that exist when the auction is rst announcedalthough exceptions would clearly be necessary

The antitrust agenciesrsquo response to predation in auction markets has also beenfeeble Dominant bidders such as Glaxo and Paci c Telephone in the examplesabove are apparently allowed to make open threats that they will punish newentrants For example Glaxorsquos letting it be known that it ldquowould almost certainlytop a rival bidrdquo would roughly translate to an incumbent rm in a ldquonormalrdquoeconomic market saying it ldquowould almost certainly undercut any new entrantrsquospricerdquo11

Regulators should take such threats seriously and treat auction markets morelike ldquoordinaryrdquo economic markets

Tailoring Auction Design to the Context

Good auction design is not ldquoone size ts allrdquo It must be sensitive to the detailsof the context A good example of this lessonmdashand of our other principlesmdashisafforded by the recent European third-generation (UMTS) mobile-phone licenseauctions

The United Kingdom which ran the rst of these auctions originally plannedto sell just four licenses12 In this case the presence of exactly four incumbentoperators who had the advantages of existing brand names and networks suggestedthat an ascending auction might deter new rms from bidding strongly in theauction or even from entering at all So the government planned an Anglo-Dutchauction An ascending stage would have continued until just ve bidders remainedafter which the ve survivors would have made sealed bids required to be no lowerthan the current price level for the four licenses13 The design performed ex-tremely well in laboratory experiments in both ef ciency and revenue generation

But when it became possible to sell ve licenses an ascending auction mademore sense Because no bidder was permitted to win more than one license at leastone license had to be sold to a new entrant This would be a suf cient carrot to

11 Similarly Paci c Telephonersquos remark that ldquoif somebody takes California away from us theyrsquoll nevermake any moneyrdquo seems to correspond to threatening that ldquoif anyone tries to compete with us wersquoll cutthe price until they lose moneyrdquo Further Paci c Telephonersquos hiring of an auction theorist to explainthe winnerrsquos curse to competitors might correspond to hiring an industrial economist to explain thetheory of the dif culties of entering new markets to potential entrants12 I was the principal auction theorist advising the UK governmentrsquos Radiocommunications Agencywhich designed and ran the recent UK mobile-phone license auction Ken Binmore had a leading roleand supervised experiments testing the proposed designs Other academic advisors included TilmanBorgers Jeremy Bulow Philippe Jehiel and Joe Swierzbinski13 It was proposed that all four winners would pay the fourth-highest sealed bid Since the licenses werenot quite identical a nal simultaneous ascending stage would have followed to allocate them moreef ciently among the winners The sealed-bid stage could be run using an ascending mechanism thatwould hide the actual bids even from the auctioneer if this would reduce political problems SeeKlemperer (1998 2001b 2002) Radiocommunications Agency (1998a b) and Binmore and Klemperer(2002) for more details

184 Journal of Economic Perspectives

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

attract several new entrants in the UK context in which it was very unclear whichnew entrant(s) might be successful14 Because licenses could not be dividedbidders could not collude to divide the market without resort to side payments Asa result the problems of collusion and entry deterrence were minimal and aversion of an ascending auction was therefore used for ef ciency reasons Theauction was widely judged a success nine new entrants bid strongly against the in-cumbents creating intense competition and record-breaking revenues of 225 bil-lion pounds

The Netherlandsrsquo sale came next Their key blunder was to follow the actualBritish design when they had an equal number ( ve) of incumbents and licensesIt was not hard to predict (indeed prior to the auction an early draft of this paperquoted in the Dutch press and Maasland 2000 did predict) that very few entrantswould show up Netherlands antitrust policy was as dysfunctional as the auctiondesign allowing the strongest potential entrants to make deals with incumbentoperators In the end just one weak new entrant (Versatel) competed with theincumbents As we have already discussed with just one excess bidder in anascending auction it was unsurprising when the weak bidder quit early amidallegations of predation at less than 30 percent of the per capita UK prices Sixmonths later the Dutch parliament began an investigation into the auctionprocess

A version of the Anglo-Dutch design would probably have worked better in theNetherlands context There are reasons to believe Versatel would have bid higherin the sealed-bid stage than the price at which it quit the ascending auction Inaddition the fear of this would have made the incumbents bid higher Further-more the ldquohope and dreamrdquo that a sealed-bid stage gives weaker bidders mighthave attracted more bidders and discouraged the formation of the joint-biddingconsortia

The Italian government thought it had learned from the Netherlands asco Italso chose roughly the UK design but stipulated that if there were no moreldquoseriousrdquo bidders (as de ned by prequali cation conditions) than licenses then thenumber of licenses could and probably would be reduced At rst glance thisseemed a clever way to avoid an uncompetitive auction but (as I and othersargued) the approach was fundamentally awed First it is putting the cart beforethe horse to create an unnecessarily concentrated mobile-phone market to make anauction look good Second our earlier discussion demonstrates that a rule thatallows the possibility that there will be just one more bidder than license does notguarantee a competitive ascending auction Also it was clear that the number oflikely entrants into an ascending auction was much smaller than it had been for theUnited Kingdom in large part because weaker potential entrants had gured out

14 In large part this was because the United Kingdom ran the rst third-generation auction Going tomarket rst was a deliberate strategy of the auction team and the sustained marketing campaign was alsoimportant The UK auction attracted 13 bidders who then learnt about othersrsquo strengths and none ofthe eight subsequent auctions had more than seven bidders

What Really Matters in Auction Design 185

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

from the earlier auctions that they were weaker and that they therefore had littlechance of winning such an auction In the event just six bidders competed for velicenses and the auction ended amid allegations of collusion after less than twodays of bidding with per capita revenues below 40 percent of the UK level abouthalf the amount the government was expecting Again an Anglo-Dutch or puresealed-bid design would probably have performed better

Klemperer (2001b 2002) discusses the 2000ndash2001 European spectrum auc-tions in much more detail

Conclusion

Much of what we have said about auction design is no more than an applica-tion of standard antitrust theory The key issues in both elds are collusion andentry The signaling and punishment strategies that support collusion in auctionsare familiar from ldquoordinaryrdquo industrial markets as are rmsrsquo verbal encouragementto collude and the predatory threats they make Our point that even modestbidding costs may be a serious deterrent to potential bidders is analogous to theindustrial-organization point that the contestability of a market is nonrobust toeven small sunk costs of entry We also argued that because an ascending auctionis more likely than a sealed-bid auction to be won by the strongest rm theascending auction may therefore be less attractive to bidders and may therefore beless pro table than a sealed-bid auction this is just an example of the standardindustrial organization argument that a market that is in principle more compet-itive (for example ldquoBertrandrdquo rather than ldquoCournotrdquo) is less attractive to enter andso may in fact be less competitive A particular feature of auction markets is thatldquowinnerrsquos curserdquo effects may mean that sealed-bid and Anglo-Dutch auctions notonly attract more rms than ascending auctions but may also lead to betteroutcomes for the auctioneer for a given number of rms But there is no justi ca-tion for the current feebleness of antitrust policy in auction markets regulatorsshould treat them much more like ldquoordinaryrdquo economic markets

However none of our examples of auction failures should be taken as anargument against auctions in general Most auctions work extremely well Occa-sionallymdashfor example when there are too few potential bidders or large costs ofsupplying necessary information to biddersmdasha form of structured negotiations maybe better but an auction is usually more attractive to potential buyers who arecrucial to a salersquos success (Bulow and Klemperer 1996) Even relatively unsuccess-ful auctions such as the Netherlands and Italian spectrum auctions were probablymore successful than the ldquobeauty contestrdquo administrative hearings used to allocatethird-generation spectrum in several other European countries For example theSpanish beauty contest yielded just 13 euros per head of population but generatedconsiderable political and legal controversy and a widespread perception that theoutcome was both unfair and inef cient all problems that are typical of suchprocedures (Binmore and Klemperer 2002 Klemperer 2000b) The dif culties

186 Journal of Economic Perspectives

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

with the French beauty contest mean that France has not only missed its govern-mentrsquos originally planned date for allocation of the spectrum (already by a year atthe time of writing) but also missed European Union deadlines

In conclusion the most important features of an auction are its robustnessagainst collusion and its attractiveness to potential bidders Failure to attend tothese issues can lead to disaster Furthermore anyone setting up an auction wouldbe foolish to follow past successful designs blindly auction design is not ldquoone size tsallrdquo While the sealed-bid auction performs well in some contexts and the Anglo-Dutch auction is ideal in other contexts the ascending auction has also frequentlybeen used very successfully In the practical design of auctions local circumstancesmatter and the devil is in the details

y I was the principal auction theorist advising the UK governmentrsquos RadiocommunicationsAgency which designed and ran the recent UK mobile-phone license auction described hereand have advised several other UK government agencies but the views expressed in thispaper are mine alone Although some observers thought some of the behavior described abovewarranted investigation I do not intend to suggest that any of it violates any applicable rulesor laws I am very grateful to many colleagues including Sushil Bikhchandani Nils-Henrikvon der Fehr Tim Harford Emiel Maasland Margaret Meyer Mike Rothkopf David SalantRebecca Stone Timothy Taylor Chuck Thomas Tommaso Valletti Michael Waldman MarkWilliams and especially my coauthors Jeremy Bulow and Marco Pagnozzi for helpful advice

References

Ausubel Lawrence M 1998 ldquoAn Ef cientAscending-Bid Auction for Multiple ObjectsrdquoMimeo University of Maryland

Ausubel Lawrence M and Peter Cramton1998 ldquoDemand Reduction and Inef ciency inMulti-Unit Auctionsrdquo Mimeo University ofMaryland

Ausubel Lawrence M Peter Cramton Pres-ton McAfee and John McMillan 1997 ldquoSyner-gies in Wireless Telephony Evidence from theBroadband PCS Auctionrdquo Journal of Economicsand Management Strategy Fall 63 pp 497ndash527

Back Kerry and Jaime F Zender 1993 ldquoAuc-tions of Divisible Goodsrdquo Review of Financial Stud-ies Winter 64 pp 733ndash64

Back Kerry and Jaime F Zender 1999 ldquoAuc-tions of Divisible Goods with Endogenous Sup-plyrdquo Working Paper Washington University inSt Louis and University of Arizona

Ballard Charles L John B Shoven and JohnWhalley 1985 ldquoGeneral Equilibrium Computa-tions of the Marginal Welfare Costs of Taxes inthe United Statesrdquo American Economic ReviewMarch 751 pp 128ndash38

Benoit Bertrand 2000 ldquoBidders Warned inGerman 3G Phone Auctionrdquo Financial TimesAugust 2 p 28

Betton Sandra and Espen B Eckbo 2000ldquoToeholds Bid Jumps and Expected Payoffs inTakeoversrdquo Review of Financial Studies Winter134 pp 841ndash82

Bikhchandani Sushil 1988 ldquoReputation inRepeated Second-Price Auctionsrdquo Journal of Eco-nomic Theory October 461 pp 97ndash119

Binmore Ken and Paul D Klemperer 2002ldquoThe Biggest Auction Ever The Sale of the Brit-ish 3G Telecom Licencesrdquo Economic JournalForthcoming

Paul Klemperer 187

Bulow Jeremy I and Paul D Klemperer1996 ldquoAuctions versus Negotiationsrdquo AmericanEconomic Review March 861 pp 180ndash94

Bulow Jeremy I and Paul D Klemperer2002 ldquoPrices and the Winnerrsquos Curserdquo RandJournal of Economics Forthcoming

Bulow Jeremy I Ming Huang and Paul DKlemperer 1999 ldquoToeholds and TakeoversrdquoJournal of Political Economy June 1073 pp 427ndash54

Burguet Roberto and Jozsef Sakovics 1996ldquoReserve Prices Without Commitmentrdquo Gamesand Economic Behavior August 152 pp 149ndash64

Burrough Brian and John Helyar 1990 Bar-barians at the Gate The Fall of RJR Nabisco Lon-don Arrow

Cauley Leslie and Mary Lu Carnevale 1994ldquoWireless Giants Some Surprise Players to SeekNew Generation of Licensesrdquo Wall Street JournalOctober 31 p A4

Coase Ronald H 1959 ldquoThe Federal Com-munications Commissionrdquo Journal of Law andEconomics October 2 pp 1ndash40

Cramton Peter and Jessie A Schwartz 1999ldquoCollusive Bidding in the FCC Spectrum Auc-tionsrdquo Working Paper University of Maryland

Cramton Peter and Jesse A Schwartz 2000ldquoCollusive Bidding Lessons from the FCC Spec-trum Auctionsrdquo Journal of Regulatory EconomicsMay 173 pp 229ndash52

Cramton Peter Robert Gibbons and Paul DKlemperer 1987 ldquoDissolving a Partnership Ef -cientlyrdquo Econometrica 553 pp 615ndash32

Crossland David 2000 ldquoAustrian UMTS Auc-tion Unlikely to Scale Peaksrdquo Reuters October31 Available at httpwww totaltelecom

Fehr Nils-Henrik von der and David Harbord1998 ldquoCompetition in Electricity Spot MarketsEconomic Theory and International Experi-encerdquo Memorandum No 51998 Departmentof Economics University of Oslo

Grimm Veronika Frank Riedel and ElmarWolfstetter 2001 ldquoLow Price Equilibrium inMulti-Unit Auctions The GSM Spectrum Auc-tion in Germanyrdquo Working Paper HumboldtUniversitat zu Berlin

Hall Robert E 2001 Digital Dealing NewYork W W Norton

Hansen Robert G 1986 ldquoSealed-Bid versusOpen Auctions The Evidencerdquo Economic InquiryJanuary 241 pp 125ndash42

Jehiel Phillipe and Benny Moldovanu 2001ldquoThe UMTSIMT-2000 License AuctionsrdquoWorking Paper University College London andUniversity of Mannheim

Klemperer Paul D 1998 ldquoAuctions With Al-most Common Values The lsquoWallet Gamersquo and

its Applicationsrdquo European Economic Review May423-5 pp 757ndash69

KlempererPaul D 1999a ldquoAuction Theory AGuide to the Literaturerdquo Journal of EconomicSurveys 133 pp 227ndash86 Also reprinted in TheCurrent State of Economic Science Volume 2 1999Shri Bhagwan Dahiya ed Rohtak India Spell-bound pp 711ndash66

Klemperer Paul D 1999b ldquoApplying AuctionTheory to Economicsrdquo Oxford Department ofEconomics Discussion Paper April

Klemperer Paul D ed 2000a The EconomicTheory of Auctions Cheltenham UK EdwardElgar

Klemperer Paul D 2000bldquo Spectrum on theBlockrdquo Wall Street Journal (Asia) May 10 p 8Also at httpwwwpaulklempererorg

Klemperer Paul D 2001a ldquoWhy Every Econ-omist Should Learn Some Auction TheoryrdquoForthcoming in Advances in Economics and Econo-metrics Invited Lectures to Eighth World Congress ofthe Econometric Society M Dewatripont L Hansenand S Turnovksy eds Cambridge UK Cam-bridge University Press Also at httpwwwpaulklempererorg

Klemperer Paul D 2001b ldquoWhat Really Mat-ters in Auction Designrdquo Working Paper versionNuf eld College Oxford University DiscussionPaper Also at httpwwwpaulklempererorg

Klemperer Paul D 2002ldquo How (Not) to RunAuctions The European 3G Telecom AuctionsrdquoEuropean Economic Review Forthcoming Also athttpwwwpaulklempererorg

Klemperer Paul D and Margaret A Meyer1989 ldquoSupply Function Equilibria in OligopolyUnder Uncertaintyrdquo Econometrica November576 pp 1243ndash277

Klemperer Paul D and Marco Pagnozzi2002 ldquoAdvantaged Bidders and Spectrum PricesAn Empirical Analysisrdquo Forthcoming

Koselka Rita 1995 ldquoPlaying Poker with CraigMcCawrdquo Forbes July 3 pp 62ndash3

Maasland Emiel 2000 ldquoVeilingmiljardenZijn een Fictie (Billions from Auctions Wish-ful Thinking)rdquo Economisch Statistische BerichtenJune 9 p 479 Translation available at httpwwwpaulklempererorg

Malvey Paul F Christine M Archibald andSean T Flynn 1996 ldquoUniform-Price AuctionsEvaluation of the Treasury Experiencerdquo Work-ing Paper US Treasury

Maskin Eric S 1992 ldquoAuctions and Privatiza-tionrdquo in Privatization H Siebert ed TubingenMohr pp 115ndash36

McAfee R Preston and John McMillan 1988ldquoSearch Mechanismsrdquo Journal of Economic TheoryFebruary 441 pp 99ndash123

188 Journal of Economic Perspectives

McMillan John 1994 ldquoSelling SpectrumRightsrdquo Journal of Economic Perspectives Summer83 pp 145ndash62

Mead Walter J and Mark Schneipp 1989ldquoCompetitive Bidding for Federal Timber in Re-gion 6 An Update 1983ndash1988rdquo Community andOrganization Research Institute University ofCalifornia Santa Barbara Contractor ReportUSDA Award No 40-3187-8-1683 June

Michelson Marcel 2000 ldquoSwiss 3G AuctionSet to Become Battle of Giantsrdquo Reuters Novem-ber 9 Available at httpwwwtotaltelecom

Milgrom Paul R and Robert J Weber 1982ldquoA Theory of Auctions and Competitive Bid-dingrdquo Econometrica September 505 pp 1089ndash122

Myerson Roger B and Mark A Satterthwaite1983 ldquoEf cient Mechanisms for Bilateral Trad-ingrdquo Journal of Economic Theory April 292 pp265ndash81

Newbery David M 1998 ldquoCompetition Con-tracts and Entry in the Electricity Spot MarketrdquoRAND Journal of Economics 294 pp 726ndash49

Nyborg Kjell and Suresh Sundaresan 1996ldquoDiscriminatory versus Uniform Treasury Auc-tions Evidence from When-Issued Transac-tionsrdquo Journal of Financial Economics September421 pp 63ndash104

Of ce of Gas and Electricity Markets 1999ldquoThe New Electricity Trading ArrangementsJulyrdquo At httpwwwopengovukofferretahtm

Persico Nicola 2000 ldquoInformation Acquisi-tion in Auctionsrdquo Econometrica 681 pp 135ndash48

Power UK 1999 ldquoThe Problems with thePoolrdquo August 31 66 p 14

Radiocommunications Agency 1998a ldquoUMTSAuction Designrdquo UMTS Auction ConsultativeGroup Paper 14 of 1998 Available as UACG(98)14at httpwwwspectrumauctionsgovuk

Radiocommunications Agency 1998b ldquoUMTSAuction Design 2rdquo UMTS Auction ConsultativeGroup Paper 16 of 1998 Available as UACG(98)16at httpwwwspectrumauctionsgovuk

Reinhert Vincent and Gregory Belzer 1996ldquoSome Evidence on Bid Sharing and the Use ofInformation in the US Treasuryrsquos Auction Ex-perimentrdquo Working Paper Board of Governorsof the Federal Reserve System

Riley John G and Huagang Li 1997 ldquoAuc-

tion Choice A Numerical Analysisrdquo MimeoUniversity of California at Los Angeles

Roberts Dan 2000 ldquoPhone Numbers thatCould Well Result in Panicrdquo Financial TimesOctober 19 p 38

Roberts Dan and Andrew Ward 2000 ldquoLittleGold at the End of the Spectrumrdquo FinancialTimes November 3 p 21

Rothkopf Michael H and Richard Engel-brecht-Wiggans 1993 ldquoMisapplications ReviewsGetting the Model Right The Case of Compet-itive Biddingrdquo Interfaces May 233 pp 99ndash106

Salant David 2000 ldquoAuctions and Regula-tion Reengineering of Regulatory MechanismsrdquoJournal of Regulatory Economics May 173 pp195ndash204

Simon David P 1994 ldquoThe Treasuryrsquos Exper-iment with Single-Price Auctions in the Mid-1970s Winnerrsquos or Taxpayerrsquos Curserdquo Review ofEconomics and Statistics November 764 pp754ndash60

Stuewe Heinz 1999 ldquoAuktion von Telefon-frequenzen Spannung bis zur letzten MinuterdquoFrankfurter Allgemeine Zeitung October 29

Total Telecom 2000 ldquoItalyrsquos UMTS Auctionto Start Octoberrdquo Reuters October 12 Availableat httpwwwtotaltelecom

UK Monopolies and Mergers Commission1999 British Sky Broadcasting Group and Manches-ter United A Report on the Proposed Merger Cm4305 London The Stationery Of ce

Vickrey William 1961 ldquoCounterspeculationAuctions and Competitive Sealed TendersrdquoJournal of Finance 16 pp 8ndash37

Wighton David 1995a ldquoWellcome AcceptsGlaxo Bid and Criticises Trustrdquo Financial TimesMarch 8 p 27

Wighton David 1995b ldquoWellcome Still Smart-ing Over Handling of Trustrsquos Stakerdquo FinancialTimes March 8 p 32

Wolfram Catherine D 1998 ldquoStrategic Bid-ding in a Multiunit Auction An Empirical Anal-ysis of Bids to Supply Electricity in England andWalesrdquo RAND Journal of Economics 294 pp 703ndash25

Wolfram Catherine D 1999 ldquoMeasuringDuopoly Power in the British Electricity SpotMarketrdquo American Economic Review September894 pp 805ndash26

What Really Matters in Auction Design 189