Journalof Economic Perspectives—Volume 16,Number 1—Winter 2002— Pages 169 – 189

What Really Matters inAuction Design

PaulKlemperer

uctions have becomeenormously popular inrecent years. Governments arenow especiallykeen, using auctions tosell mobile-phone licenses, A operatedecentralized electricity markets, privatize companies and for many otherpurposes. Thegrowth of e-commerce has ledto many business-to- business auctions forgoods whosetrade was previouslynegotiated bilaterally. Economistsare proud oftheirrole in pushing forauctions; forexample, Coase (1959)was among theŽ rstto advocate auctioning theradio spectrum. But many auctions—including some designed with the help of leading academic economists—have workedvery badly. Forexample, six European countries auctioned offspectrum licenses for “third-generation”mobile phones in2000.In Germanyand theUnited Kingdom, thespectrum sold forover 600 euros per person ($80billion in all, or over 2percentof GDP). Butin Austria, theNetherlands, Italy and Switzerland,the revenueswere just 100,170, 240 and 20eurosper person, respectively.To besure, investorsbecame more skeptical about theunderlying value of the spectrum during 2000(and theyare even more skeptical today). Butthis isjust afractionof the story. TheNetherlands auction was sandwiched betweenthe U.K. and Germanauctions, and analysts and governmentofŽ cials predicted revenues in excessof 400 euros per person fromthe Italian and Swissauctions just afewdays beforethey began (Michelson, 2000;Roberts, 2000; Total Telecom, 2000; Klemperer, 2002). These otherauctions wereŽ ascoes primarilybecause theywere poorly designed. Sowhat makesa successful auction? What reallymatters in auction designare the same issues that any industry y PaulKlemperer istheEdgeworth Professor of Economics,Oxford University, England. His e-mail addressis [email protected] ,andhis website is http://www. paulklemperer.org . 170Journal of Economic Perspectives

regulatorwould recognize as keyconcerns: discouragingcollusive, entry-deterring and predatorybehavior. In short, good auction designis mostly good elementary economics. Bycontrast, mostof the extensive auction literature(as summarizedin, for example,Klemperer, 1999a, 2000a) is of second-order importance for practical auction design.The auction literaturelargely focuses on a Ž xednumber of bidders who bidnoncooperatively, and itemphasizes issues such as theeffects of risk aversion,correlation of information, budget constraints and complementarities. Auction theoristshave madeimportant progress on thesetopics fromwhich other economictheory has bene Ž ted,and auction theoryhas also beenfruitfully applied inpoliticaleconomy, Ž nance, lawand economics, laboreconomics and industrial organization,often in contexts not usuallythough ofas auctions (Klemperer, 2001a).But most of this literatureis of much lessuse foractually designing auctions. This paperwill list and giveexamples of somecritical pitfalls in auction design and discuss what todo about them.We show that ascending and uniform-price auctions areboth veryvulnerable to collusion and verylikely to deter entry into an auction. Weconsiderincluding a Ž nal sealed-bidstage into an otherwise-ascending auction tocreate an “Anglo-Dutch ” auction, and weemphasize the need for strongerantitrust policy in auction markets.

Collusion

A Ž rstmajor setof concerns forpractical auction designinvolves the risk that participants mayexplicitly or tacitlycollude to avoidbidding up prices.Consider a multiunit(simultaneous) ascending auction. (This isjust likethe standard auction used, forexample, to sell a paintingin Sotheby ’sorChristies —theprice starts low, and competingbidders raise the price until no oneis preparedto bid any higher, and the Ž nal bidderthen winsthe prize at the Ž nal pricebid. However,in this case, severalobjects aresold at thesame time, with the price rising on each ofthem independently,and none ofthe objects is Ž nallysold untilno onewishes to bid again on any ofthe objects.) In such an auction, bidderscan usethe early stages, when pricesare still low, to signal who should winwhich objects and then tacitly agreeto stop pushing up prices. Forexample, in 1999,Germany sold tenblocks of spectrumby asimultaneous ascending auction withthe rule that any newbid on ablockhad toexceed the previoushigh bidby at least10 percent.Mannesman ’s Ž rstbids were18.18 million deutschmarks permegahertz on blocks1 –5and 20millionDM perMHz on blocks 6 –10;the only other credible bidder —T-Mobil— bideven less in the Ž rst round. One ofT-Mobil ’smanagersthen said (Stuewe,1999, p. 13): “Therewere no agreementswith Mannesman. But[T-Mobil] interpreted Mannesman ’s Ž rst bid as an offer.” Thepoint isthat 18.18plus a10percent raise equals approximately 20. It seemsT-Mobil understood that ifit bid20 million DM perMHz onblocks1 –5, PaulKlemperer 171

but didnot bidagain on blocks6 –10,the two companies wouldthen liveand let livewith neither company challengingthe other on theother ’shalf. Exactlythat happened. Sothe auction closedafter just tworounds witheach ofthe bidders acquiringhalf theblocks for the same low price ( Jehieland Moldovanu, 2001; Grimm,Riedel and Wolfstetter,2001). Ascending auctions can also facilitatecollusion by offering a mechanism for punishing rivals.The threat of punishment maybe implicit; for example, it was clearto T-Mobil that Mannesman wouldretaliate with high bids on blocks1 –5 if T-Mobilcontinued biddingon blocks6 –10.But an ascending auction can also allowmore explicit options forpunishment. In amultilicenseU.S. spectrumauction in1996 –1997,U.S. Westwas compet- ingvigorously with McLeod for lot number 378: a licensein Rochester,Minnesota. Although mostbids inthe auction had beenin exact thousands ofdollars, U.S. Westbid $313,378 and $62,378for two licenses in Iowa inwhich ithad earlier shown no interest,overbidding McLeod, who had seemedto be the uncontested high bidderfor these licenses. McLeod got the point that itwas beingpunished for competingin Rochester and dropped out ofthat market.Since McLeod made subsequent higherbids on theIowa licenses,the “punishment” bids cost U.S. West nothing (Cramton and Schwartz,1999). Arelatedphenomenon can arisein one special kind of sealed-bid auction, namely a uniform-price auction inwhich each biddersubmits a sealedbid stating what priceit would pay fordifferent quantities of a homogenous good, like electricity(that is, itsubmitsa demand function), and then thegood issold at the singleprice determined by the lowest winning bid. In this format,bidders can submitbids that ensurethat any deviationfrom a (tacitor explicit) collusive agreementis severely punished: each bidderbids veryhigh pricesfor smaller quantitiesthan itscollusively agreed share. Then, ifany bidderattempts to obtain morethan itsagreed share (leavingother Ž rmswith less than theiragreed shares), allbidders will have topay thesevery high prices.However, if everyone sticks to theiragreed shares, then thesevery high priceswill never need to be paid. As a result,deviation from the collusive agreement is unpro Ž table.1 Theelectricity regulator in the United Kingdom believes the market in which distributioncompanies purchase electricityfrom generating companies has fallen preyto exactly this kindof “implicitcollusion ” (OfŽ ceof Gas and Electricity Markets,1999, pp. 173 –174). “Farfrom being the success storytrumpeted around theworld, the story of the U.K. generationmarket and thedevelopment of competitionhas beensomething of a disaster, ” reported Power U.K.(1999;see also

1 Since,with many units, the lowestwinning bid in a uniform-priceauction is typicallynot importantly differentfrom the highest losingbid, this auctionis analogousto an ascendingauction (in whichevery winnerpays the runner-up ’swillingness-to-pay). The “threats” that supportcollusion in a uniform-price auctionare likewise analogous to the implicitthreats supportingcollusion in an ascendingauction. Collusionin auniform-priceauction is harder if supplyis uncertain, sincethis reducesthe numberof pointson the bidschedule that areinframarginal and can beused as threats (Klempererand Meyer, 1989;Back and Zender,1993, 1999). 172Journal of Economic Perspectives

von derFehr and Harbord, 1998;Newbery, 1998; Wolfram, 1998, 1999). In addi- tion, afrequentlyrepeated auction marketsuch as that forelectricity is particularly vulnerableto collusion, becausethe repeated interaction among biddersexpands theset of signaling and punishment strategiesavailable to them and allowsthem to learnto cooperate (Klemperer, 2002). Much ofthe kind of behavior discussed so faris hard tochallenge legally. Indeed, tryingto outlaw it allwould require cumbersome rules that wouldrestrict bidders’  exibilityand mightgenerate inef Ž ciencies,without being fully effective. It wouldbe much betterto solve these problems with better auction designs.

Entry Deterrence and Predation

Thesecond major areaof concern ofpractical auction designis to attract bidders,since an auction withtoo few bidders risks being unpro Ž tablefor the auctioneer(Bulow and Klemperer,1996) and potentiallyinef Ž cient.Ascending auctions areoften particularly poor in this respect,since they can allowsome biddersto deter the entry, or depress the bidding, of rivals. In an ascending auction, thereis a strongpresumption that the Ž rmthat values winningthe most will be the eventual winner, because even if it isoutbidat an early stage, itcan eventuallytop any opposition. As aresult,other Ž rmshave little incentiveto enter the bidding and maynot do so iftheyhave evenmodest costs of bidding. Consider, forexample, Glaxo ’s1995takeover of the Wellcome drugs company. After Glaxo’s Ž rstbid of 9 billionpounds, Zeneca expressedwillingness to offer about 10billionpounds ifit could besureof winning, whileRoche consideredan offerof 11 billionpounds. Butcertain synergies made Wellcome worth a littlemore toGlaxothan tothe other Ž rms,and thecosts ofbiddingwere tens ofmillions of pounds. Eventually,neither Roche nor Zeneca actuallyentered the bidding, and Wellcomewas sold at theoriginal bid of 9 billionpounds, literallya billionor two lessthan itsshareholders might have received.Wellcome ’sown chiefexecutive admitted “therewas moneyleft on thetable ” (Wighton, 1995a,b). Whileascending auctions areparticularly vulnerable to lack of entry, other auction formscan resultin similarproblems if thecosts ofentry and theasymme- triesbetween bidders are too large. The1991 U.K. saleof televisionfranchises bya sealed-bidauction isa dramatic example.While the regions in theSouth and Southeast, Southwest, East, Walesand West, Northeastand Yorkshireall sold inthe range of 9.36 to 15.88 pounds per head ofpopulation, theonly —and thereforewinning —bidfor the Midlands regionwas madeby the incumbent Ž rmand was just one-twentiethof one penny (!) perhead ofpopulation. Much thesame happened inScotland, wherethe only bidderfor the Central region generously bid one-seventh of one penny percapita. What had happened was that bidderswere required to provide very detailed WhatReally Mattersin Auction Design 173

region-speci Ž cprogrammingplans. In each ofthese two regions, the only bidder Ž guredout that no oneelse had developedsuch aplan. 2 Anotherissue that can depressbidding in some ascending auctions isthe “winner’s curse.” This problemapplies when biddershave thesame, orcloseto the same, actual valuefor a prize,but theyhave differentinformation about that actual value(what auction theoristscall the “common values ” case). Thewinner ’s curse re ectsthe danger that thewinner of an auction islikelyto be the party who has mostgreatly overestimated the value of theprize. Knowing about thewinner ’s curse willcause everyoneto bid cautiously. Butweaker Ž rmsmust beespecially cautious, sincethey must recognizethat theyare only likely to win when theyhave overes- timatedthe value by even more than usual. Therefore,an advantaged Ž rm can be lesscautious, sincebeating very cautious opponents neednot implyone has overestimatedthe prize ’svalue.Because the winner ’scurseaffects weak Ž rms much morethan strongones, and because theeffect is self-reinforcing, the advantaged bidderwins most of the time. And because itsrivals bid extremely cautiously, italso generallypays alowprice when itdoes win(Klemperer, 1998). Thebidding on theLos Angeles license in the1995 U.S. auction formobile- phone broadband licensesillustrates this problem.While the license ’s value was hard toestimate, it was probablyworth similar amounts toseveral bidders. But PaciŽ cTelephone,which alreadyoperated the local Ž xed-linetelephone business inCalifornia,had distinctadvantages fromits database onpotentiallocal custom- ers,its well-known brand-name and itsfamiliarity with doing business inCalifornia. Theauction was an ascending one. Theresult was that thebidding stopped at avery lowprice. In theend, theLos Angeles license yielded only $26 per capita. In Chicago, bycontrast, themain local Ž xed-lineprovider was ineligibleto compete, and itwas not obviouswho wouldwin, so theauction yielded$31 per capita even though Chicago was thought lessvaluable than LosAngeles because ofits lower household incomes,lower expected population growthand moredispersed pop- ulation(Klemperer, 1998; Bulow and Klemperer,2002). For formal econometric evidencefor the FCC auctions morebroadly, see Klemperer and Pagnozzi(2002). Of course, the “winner’s curse” problemexacerbates the problem that weaker biddersmay not botherto participate in an ascending auction. GTEand Bell Atlanticmade deals that madethem ineligible to bid for the Los Angeles license, and MCIfailedto enter this auction at all.Similarly, takeover battles are essentially ascending auctions, and thereis empirical evidence that a Ž rmthat makesa takeoverbid has alowerrisk of facing a rivalbidder if the Ž rmhas alarger shareholdingor “toehold” inthetarget company (Bettonand Eckbo,2000). Becauseoutcomes in an ascending auction can bedramatically in  uenced by aseeminglymodest advantage, developingsuch an advantage can bean effective predatorystrategy. An apparent examplewas the1999 attempt by BSkyB (Rupert Murdoch’ssatellitetelevision company) toacquireManchester United (England ’s

2 WhileI have advisedthe U.K. governmenton severalauctions, Ihave neverhad anything todo with televisionlicenses! 174Journal of Economic Perspectives

mostsuccessful soccerclub). Theproblem was theadvantage this wouldgive BSkyB inthe auction offootball television rights. Since Manchester United receives 7percentof thePremier League ’stelevisionrevenues, BSkyB would have received 7percentof the price of the league ’sbroadcasting rights,whoever won them.So BSkyBwould have had an incentiveto bid more aggressively in an ascending auction topush up theprice of the rights, and knowingthis, otherpotential bidders wouldhave faced aworse “winner’s curse” and backedoff. BSkyBmight have ended up witha lockover the television rights, with damaging effects on thetelevision marketmore generally. Largely for this reason, theU.K. governmentblocked the acquisition. 3 Astrongbidder also has an incentiveto create a reputationfor aggressiveness that reinforcesits advantage. Forexample, when Glaxowas biddingfor Wellcome, itmade it clear that it “wouldalmost certainly top arivalbid ” (Wighton, 1995b). Similarly,before bidding for the California phone license,Paci Ž c Telephone announced inthe WallStreet Journal that “ifsomebody takes California away from us, they’llnever make any money ” (Cauleyand Carnevale,1994, p. A4).Paci Ž c Telephonealso hiredone of the world ’smostprominent auction theoriststo give seminarsto the rest of the industry to explain the winner ’scurseargument that justiŽ esthis statement,and itreinforced the point infull-pageads that ran inthe newspapers ofthecities where its major competitorswere headquartered (Koselka, 1995,p. 63).It also madeorganizational changes that demonstratedits commit- mentto winning the Los Angeles license. Predationmay be particularly easy in repeatedascending auctions, such as in aseriesof spectrum auctions. Abidderwho buys assets that arecomplementary to assets forsale in a futureauction orwho simplybids veryaggressively in early auctions can developa reputationfor aggressiveness (Bikhchandani, 1988).Poten- tialrivals in future auctions willbe less willing to participate and willbid less aggressivelyif they do participate(Klemperer, 2002). Finally,because an ascending auction ofteneffectively blocks the entry of “weaker” bidders,it encourages “stronger” biddersto bidjointly or to collude; after all,they know that no oneelse can enterthe auction tostealthe collusive rents they create.In thedisastrous November2000 Swiss sale of four third-generation mobile- phone licenses,there was considerableinitial interest from potential bidders. But weakerbidders were put offby the auction form —at leastone company hired biddingconsultants and then gaveup afterlearning that theascending-bidding ruleswould give the company verylittle chance against strongerrivals. Moreover, thegovernment permitted last-minute joint-bidding agreements — essentiallyof Ž - ciallysanctioned collusion. In theweek before the auction, the Ž eldshrank from ninebidders to just fourbidders for the four licenses! Since no bidderwas allowed

3 Although the term “toeholdeffect, ” coinedby Bulow, Huang and Klemperer(1999) and Klemperer (1998) inthe relatedcontext of takeoverbattles (see above), entered the popularpress, and these paperswere cited by the U.K. Monopoliesand MergersCommission (1999) report,which effectively decidedthe issue,neither I normy coauthorshad any involvementin this case. PaulKlemperer 175

totakemore than onelicense, the sale price was determinedby the reserve price, which was just one-thirtiethof the U.K. and Germanper capita revenuesand one-Ž ftiethof what theSwiss had once hoped for!

Other Pitfalls

ReservePrices Many ofthe disasters above were greatly aggravated by failure to seta proper reserveprice (the minimum amount thewinner is required to pay). Takethe previousexample. It was ridiculousfor the Swiss government to set its reserve at just one-thirtiethof the per capita revenueraised by the German and U.K. governments forsimilar properties. Since the government ’sown spokesman predictedjust Ž ve days priorto the auction that twentytimes the reserve price would be raised, what was thegovernment playing at? Inadequate reserveprices also increasethe incentives for predation and may encouragecollusion that wouldnot otherwisehave beenin allbidders ’ interests.A strongerbidder in an ascending auction has achoicebetween either tacitly collud- ingto end theauction quicklyat alowprice or forcing the price up todrive out weakerbidders. The lower the reserve price at which theauction can beconcluded, themore attractive is the Ž rstoption. This factormay have beenan important contributorto several of the Ž ascoes wehave discussed.

PoliticalProblems Seriousreserve prices are often opposed not onlyby industry groups, but also bygovernmentof Ž cialsfor whom a veryembarrassing outcome is that thereserve priceis not met,the object is not sold, and theauction isseen as a “failure.” Similarly,standard ( Ž rst-price)sealed-bid auctions —inwhich thebidders si- multaneouslymake “best and Ž nal” offers,and thewinner pays theprice he bid— can sometimesbe very embarrassing for bidders, as BSCH(Spain ’s biggest bank) found out when Brazilprivatized the Sao Paulo statebank Banespa. When thebids wereopened, BSCH ’smanagerswere horri Ž edto learn that theirbid of over7 billionreals ($3.6 billion) was morethan threetimes the runner-up ’s bid and that theywere therefore paying 5 billionreals ($2.5 billion) more than was needed towin. In otherauctions, meanwhile,losers who have just narrowlyunderbid the winnershave found itequally hard toexplain themselves to their bosses and shareholders. So Ž rms,or at leasttheir managers, can oppose Ž rst-priceauctions. On theother hand, a second-price sealed-bidauction —inwhich thewinner pays therunner-up ’s bid— can beembarrassing for the auctioneer if the winner ’s actual bidis revealed to be far more than therunner-up ’s, evenif theauction designwas both efŽ cientand maximizedexpected revenue. McMillan (1994) reports a second-priceNew Zealand auction inwhich thewinner bid NZ $7millionbut paid therunner-up ’sbidof NZ $5,000.New Zealand should have seta minimumreserve pricethat thewinner had topay, but evenif that had beenpolitically possible, the 176Journal of Economic Perspectives

winnerwould probably have bidmore than ithad topay, so this mighthave been an economicallybut not politicallysensible auction.

Loopholes In somecases, theauction rulesmay leave gaping loopholes for behavior to gamethe auction. In 2000,Turkey auctioned twotelecom licenses sequentially, withan additional twistthat setthe reserve price for the second licenseequal to the sellingprice of the Ž rst. One Ž rmthen bidfar more for the Ž rstlicense than it could possiblybe worth if the Ž rmhad tocompete in the telecom market with a rivalholding the second license.But the Ž rmhad rightly Ž guredthat no rivalwould bewillingto bidthat high forthe second license,which thereforeremained unsold, leaving the Ž rmwithout a rivaloperating the second license! As another example,McMillan (1994) reports an Australian auction for satellite-televisionlicenses in which twobidders each madelarge numbers of differentsealed bids on thesame objects and then, afterconsiderable delays, defaultedon those bids theydid not likeafter the fact —sincethe government had neglectedto impose any penaltiesfor default. More recently, the U.S. spectrum auctions have beenplagued by bidders “winning” licensesand subsequentlyde- faultingon theircommitments, often after long delays. (Spectrum auctions inIndia also recentlyfell into the same trap.) If defaultcosts aresmall, then biddersare bidding for options on prizesrather than theprizes themselves. Furthermore, if smaller,under Ž nanced Ž rmscan avoidcommitments through bankruptcy, then an auction actuallyfavors these bidders over better- Ž nanced competitorswho cannot default.

Credibilityof the Rules It maynot becredible for the auctioneer to punish abidderviolating the auction ruleswhen just onebidder needs to be eliminated to end an auction, because excludingthe offending bidder would end theauction immediately,and it mightbe hard toimpose Ž nes largeenough tohave aseriousdeterrent effect. Fines ofhundreds ofmillions or even billions of dollars might have beenrequired to deterimproper behavior in someof theEuropean third-generation mobile-phone licenseauctions. In theNetherlands sale, for example, six bidders competed for Ž ve licensesin an ascending auction inwhich bidderswere permitted to win just one licenseeach. One bidder,Telfort, sent aletterto another, Versatel,threatening legalaction fordamages ifVersatel continued tobid! Telfort claimed that Versatel “believesthat itsbids willalways be surpassed by[others ’ ...so it]must bethat Versatelis attempting to either raise its competitors ’ costs orto get access to their. ..networks. ” Many observersfelt Telfort ’sthreatsagainst Versatelwere outrageous. However,the government took no action —not evenan investigation. As aresult,Versatel quit the auction, and thesale raised less than 30percent of what theDutch governmenthad forecastbased on theresults of the United Kingdom’ssimilarauction just threemonths earlier. Ascending auctions areparticularly vulnerable to rule breaking by the bidders, WhatReally Mattersin Auction Design 177

sincethey necessarily pass through astagewhere there is just oneor a fewexcess bidders,and theascending structureallows a cheat timeto assess thesuccess ofits strategy(Klemperer, 2001b, 2002). Sealed-bid auctions, bycontrast, maybe more vulnerableto rule changing bythe auctioneer. For example, excuses for not acceptinga winningbid can oftenbe found iflosing bidders are willing to bid higher.The famous RJR-Nabisco salewent through severalsupposedly Ž nal sealed- bidauctions (Burroughand Helyar,1990). But if, after a sealed-bidauction, the auctioneercan reopenthe auction tohigher offers, the auction isreally an ascending-bid auction and needsto be recognized as such. In fact, genuinesealed- bidauctions maybe dif Ž cultto run intakeover battles, especially since a director who turns down ahigherbid for his company afterrunning a “sealed-bidauction ” maybe vulnerable to shareholder lawsuits. Sealed-bidauctions can also beespeciallyhard tocommit to if the auctioneer has any association witha bidder,as, forexample, would have beenthe case inthe U.K. footballtelevision rights auction discussed earlierif BSkyB (a bidder)had takenover Manchester United (an in  uentialmember of the football league, which was theauctioneer). Committingto future behavior may be aparticularproblem for governments. Forexample, it may be dif Ž cultto auction alicenseif the regulatory regime may change, but bindingfuture governments (or even the current government) to a particularregulatory regime may prove dif Ž cult. Thecredibility of reserve prices is of special importance. If areserveprice is not agenuinecommitment not tosell an objectif it does not reach itsreserve, then ithas no meaning,and bidderswill treat it as such. Forexample, returning to the Turkishtale of woe, the government is now consideringnew arrangements to sell thesecond license,but at what cost tothe credibility of its future auctions? 4

MarketStructure In someauctions, forexample, of mobile-phone licenses, the structure of the industrythat willbe created cannot beignored by the auction designer.It is temptingsimply to “letthe market decide ” theindustry structure by auctioning many smallpackages ofspectrum,which individual Ž rmscan aggregateinto larger licenses.But the outcome of an auction isdriven by bidders ’ proŽ ts, not bythe welfareof consumers orsociety as awhole. Themost obvious possible distortion is that since Ž rms’ joint proŽ ts in a marketare generally greater if fewercompetitors are in themarket, it is worthmore toany groupof Ž rmsto prevent entry of an additional Ž rmthan theadditional Ž rm iswillingto pay toenter. As aresult,too few Ž rmsmay win a share ofspectrum,and thesewinners may each wintoo much, injust thesame way as a “hands-off” policy

4 Reauctioningwith alowerreserve price after adelaymay sometimes be sensible,to allow further entry if thereare high costs of enteringthe auction(Burguet and Sakovics,1996; McAfee and McMillan, 1988), butin this casethe auctioneershould make clear in advancewhat willhappen if the reserveis not met. 178Journal of Economic Perspectives

tomergercontrol will tend tocreate an overlyconcentrated industry. TheTurkish Ž asco discussed earlierwas aspectacular exampleof how an auction can bebiased towardgenerating a monopoly. 5 Butthis outcomeis not theonly socially suboptimal possibility. A Ž rm with a largedemand mayprefer to reduceits demand toend theauction at alowprice, ratherthan raisethe price to drive out itsrivals, even when thelatter course would besocially more ef Ž cient(Ausubel and Cramton, 1998).There can also betoo many winnersif Ž rmscollude to divide the spoils at alowprice. In theAustrian third-generationmobile spectrum sale, for example, six Ž rmscompeted for twelve identicallots in an ascending auction and, not surprisingly,seemed to agree to dividethe market so each Ž rmwon twolots each at not much morethan thevery lowreserve price. Perhaps sixwinners was theef Ž cientoutcome. But we certainly cannot tellfrom the behavior in the auction. It was rumoredthat thebidding lasted onlylong enough tocreatesome public perception of genuinecompetition and to reducethe risk of the government changing therules. Thus, itmay sometimes be wiser to predetermine the number of winners by auctioning offfewer larger licenses, but limitingbidders to one license apiece, ratherthan toauction many licensesand toallow bidders to buy as many as they wish.

Whenis Auction Design Less Important? Thefact that collusion, entrydeterrence and, moregenerally, buyer market poweris thekey to auction problemssuggests that auction designmay not matter verymuch when thereis alargenumber of potential bidders for whom entry to the auction iseasy. Forexample, though much inkhas beenspilt on thesubject of governmentsecurity sales, auction designmay not mattermuch foreither price or efŽ ciencyin this case. Indeed, theU.S. Treasury ’srecentexperiments with different kinds ofauctions yieldedinconclusive results (Simon, 1994; Malvey, Archibald and Flynn, 1996;Nyborg and Sundaresan, 1996;Reinhart and Belzar,1996; Ausubel and Cramton, 1998),and thebroader empirical literature is also inconclusive.Of course, evensmall differences in auction performancecan besigni Ž cant when such largeamounts ofmoney are involved, and collusionhas beenan issuein some governmentsecurity sales, so furtherresearch is still warranted. 6

Solutions

Makingthe Ascending Auction More Robust Much ofourdiscussion has emphasizedthe vulnerability of ascending auctions tocollusion and predatorybehavior. However, ascending auctions have several

5 Similarly,the recentJuly 2001 Greek second-generation spectrum auction led to a moreconcentrated telecommarket than seemslikely to be socially ef Ž cient. 6 Theseviews are personal. I have advisedU.K. governmentagencies on the relatedissue of the saleof gold.See Klemperer (1999b) for more discussion. PaulKlemperer 179

virtues,as well.An ascending auction isparticularly likely to allocatethe prizes to thebidders who valuethem the most, sincea bidderwith a highervalue always has theopportunity to rebid to top alower-valuebidder who mayinitially have bid moreaggressively. 7 Moreover,if there are complementarities between the objects forsale, a multiunitascending auction makesit more likely that bidderswill win efŽ cientbundles than inapuresealed-bid auction inwhich theycan learnnothing about theiropponents ’ intentions. Allowingbidders to learn about others ’ valua- tions duringthe auction can also makethe bidders more comfortable with their own assessments and lesscautious, and itoften raises the auctioneer ’srevenuesif informationis “afŽ liated” inthe sense of Milgrom and Weber(1982). Anumberof methods tomake the ascending auction morerobust areclear enough. Forexample, bidders can beforced to bid “round” numbers, theexact incrementscan beprespeci Ž ed, and bids can bemade anonymous. Thesesteps makeit harderto usebids tosignalother buyers. Lots can beaggregated into larger packages tomake it harder for bidders to dividethe spoils, and keepingsecret the numberof bidders remaining in theauction also makescollusion harder (Cramton and Schwartz,2000; Salant, 2000).Ausubel ’s(1998)suggested modi Ž cation ofthe ascending auction mitigatesthe incentive of bidders to reduce their demands to end theauction quicklyat alowprice. Sometimes it is possible to pay biddersto enteran auction; forexample, “whiteknights ” can beoffered options toenter a takeoverbattle against an advantaged bidder. Butwhile these measures can beuseful, they do not eliminatethe risks of collusionor of too few bidders. An alternativeis to choose adifferenttype of auction.

UsingSealed-Bid Auctions In astandard sealed-bidauction (or “Ž rst-price” sealed-bidauction), each biddersimultaneously makes a single “best and Ž nal” offer.As aresult, Ž rms are unable toretaliate against bidderswho failto cooperatewith them, so collusionis much harderthan inan ascending auction. Tacitcollusion is particularlydif Ž cult since Ž rmsare unable touse the bidding to signal. True,both signalingand retaliationare possible in aseriesof sealed-bidauctions, but collusionis stillusually harderthan ina seriesof ascending auctions. Fromthe perspective of encouraging more entry, the merit of a sealed-bid auction isthat theoutcome is much lesscertain than inan ascending auction. An

7 This appliesin many “commonvalues ” and “privatevalues ” settings (Maskin, 1992), butis not necessarilythe sameas maximizingef Ž ciency.When biddersare Ž rms, it ignoresconsumer welfare (which islikelyto favor amorewidely dispersed ownership than Ž rmswould choose), and, ofcourse, it ignoresgovernment revenue. We assume governments (as wellas otherauctioneers) care about revenue becauseof the substantial deadweightlosses (perhaps 33cents perdollar raised) of raising government funds throughalternative methods (Ballard, Shovenand Whalley, 1985). Resaleis not aperfect substitute foran ef Ž cientinitial allocation, because even costless resale cannot usuallyensure an efŽ cientoutcome in the presenceof incomplete information (Myerson and Satterthwaite, 1983; Cramton,Gibbons and Klemperer,1987). 180Journal of Economic Perspectives

advantaged bidderwill probably win a sealed-bidauction, but itmust makeits single Ž nal offerin the face of uncertainty about itsrivals ’ bids, and because itwants toget abargain, itssealed-bid will not bethe maximum it could bepushed toin an ascending auction. So “weaker” biddershave at leastsome chance ofvictory,even when theywould surely lose an ascending auction (Vickrey,1961, appendix III). It followsthat potentialentrants arelikely to be more willing to enter a sealed-bid auction than an ascending auction. Asealed-bidauction mighteven encourage bidders who enteronly to resell, furtherincreasing the competitiveness of theauction. Such biddersseem less likely toenter an ascending auction, sinceit is generally more dif Ž cult to proŽ t from resellingto Ž rmsone has beatenin an ascending auction. Becausesealed-bid auctions aremore attractive to entrants, theymay also discourageconsortia fromforming. If thestrong Ž rmsforma consortium, theymay simplyattract other Ž rmsinto the bidding in thehope ofbeating the consortium. So strong Ž rmsare more likely to bid independently in a sealed-bidauction, makingthis auction much morecompetitive. Consistent withall this, thereis some evidence from timber sales that sealed- bidauctions attract morebidders than ascending auctions do and that this makes sealed-bidauctions considerablymore pro Ž tablefor the seller, and this seemsto be believedin this industry(Mead and Schneipp, 1989;Rothkopf and Engelbrecht- Wiggans, 1993),even though conditional on thenumber of bidders, sealed-bid auctions seemonly slightly more pro Ž tablethan ascending auctions (Hansen, 1986). Furthermore,in the “common values ” case that biddershave similaractual valuesfor a prize,the “winner’s curse” problemfor a weakerbidder is farless severe inasealed-bidauction. Winning an ascending auction means theweaker bidder is payinga pricethat thestronger rival is unwillingto match —which should makethe weakerbidder very nervous. Butthe weaker player has achance ofwinning a sealed-bidauction at apricethe stronger rival would bewilling to match, but didn ’t. Sincebeating the stronger player isn ’tnecessarilybad newsin a sealed-bidauction, theweaker player can bidmore aggressively. So auction priceswill be higher,even fora givennumber of bidders (Klemperer, 1998; Bulow, Huang and Klemperer, 1999).8 Butwhile sealed-bid auctions have many advantages, theyare not without  aws. Mainly,by giving some chance ofvictory to weaker bidders, sealed-bid auctions are lesslikely than ascending auctions tolead to ef Ž cientoutcomes. Moreover,in standard sealed-bidauctions inwhich winnerspay theirown bids, biddersneed to have good informationabout thedistribution of their rivals ’ valuesto bid intelli- gently(Persico, 2000). By contrast, inan ascending oruniform-price auction the

8 In Milgromand Weber ’s(1982) model,sealed-bid auctions are less pro Ž tablethan ascendingauctions if signals are “afŽ liated.” But they assumesymmetric bidders, and the effect doesnot seemlarge in practice(Riley and Li,1997). Sealed-bidauctions are generally more pro Ž tableif biddersare risk averse orbudgetconstrained (Klemperer, 2000a). WhatReally Mattersin Auction Design 181

beststrategy of abidderwho knows itsown valueis just tobidup tothat value,and winners’ payments aredetermined by the bids ofnonwinners. So “pay-your-bid ” sealed-bidauctions maydiscourage potential bidders who have onlysmall amounts totrade and forwhom the costs ofobtaining market information might not be worthpaying. Forexample, in March 2001,the U.K. electricityregulator replaced theproblematic uniform-price auction wedescribedearlier by an exchangemarket followedby a “pay-your-bid ” sealed-bidauction, which makescollusion harder, becausebids can no longerbe used as costlessthreats. Buta major concern isthat thenew trading arrangements may deter potential entrants frominvesting the sunk costs necessaryto enter the electricity market. 9 However,the entry problem in many-unit auctions ismuch lessserious if small bidderscan buy fromlarger intermediaries who can aggregatesmaller bidders ’ de- mands and bidin their place, as, forexample, occurs inauctions ofTreasury bills. And theentry problem is also alleviatedif smaller bidders are permitted to make “noncom- petitivebids, ” that is, tostate demands for Žxedquantities for which theypay the averagewinning price, as isalso thecase insome Treasury bill auctions.

TheAnglo-Dutch Auction Asolutionto the dilemma of choosing betweenthe ascending (oftencalled “English”)and sealed-bid(or “Dutch”)formsis to combinethe two into a hybrid, the “Anglo-Dutch, ” which oftencaptures thebest features of both and was Ž rst describedand proposed inKlemperer (1998). Forsimplicity, assume asingleobject is to be auctioned. In an Anglo-Dutch auction, theauctioneer begins by running an ascending auction inwhich priceis raisedcontinuously untilall but twobidders have dropped out. Thetwo remaining biddersare then each requiredto make a Ž nal sealed-bidoffer that isnot lower than thecurrent asking price, and thewinner pays thewinning bid. Theprocess is much likethe way houses areoften sold, although, unlikein many house sales, the procedurethe auctioneer will follow in an Anglo-Dutch auction isclearlyspeci Ž ed in advance. Anotherauction withsimilar features —and probablysimilar motivations to the Anglo-Dutch —isW.R. Hambrecht ’s OpenBook auction forcorporate bonds. The earlybidding is publicand ascending, but bidderscan make Ž nal sealedbids inthe last hour. Although allbidders are permitted to make Ž nal bids, higherbidders in the Ž rststages are given an advantage that isevidently large enough toinduce seriousbidding early on (Hall, 2001,p. 71). Theprocess also has somesimilarity to auctions on eBay(by far the world ’s most successful e-commerceauctioneer), which areascending auctions, but witha Žxed endingtime so that many biddersoften bid only in the last few seconds inessentially sealed-bidstyle. eBay attracts farmore bidders than itsrival, Yahoo, which runs a

9 Also, the newarrangements may not fullyresolve the collusionproblem anyway sincethe marketis so frequentlyrepeated (Klemperer, 1999b). 182Journal of Economic Perspectives

standard ascending auction witha traditional “going,going, gone ” procedurethat does not closethe auction untilthere have been no bids for10 minutes. Themain valueof the Anglo-Dutch procedurearises when onebidder (for example,the incumbent operator of a licensethat istobe reauctioned) is thought tobe stronger than potentialrivals. Potential rivals might be unwilling to enter a pureascending-bid auction against thestrong bidder, who wouldbe perceived to bea surewinner. But the sealed bid at the Ž nal stageinduces someuncertainty about which ofthe two Ž nalists willwin, and entrants areattracted by the knowl- edgethat theyhave achance tomakeit tothis Ž nal stage.So theprice may easily behighereven by the end ofthe Ž rstascending stageof theAnglo-Dutch auction than ifa pureascending auction wereused. TheAnglo-Dutch should capturethe other advantages ofthe sealed-bid auc- tiondiscussed inthe previous section. Collusionwill be discouraged because the Ž nal sealed-bidround allows Ž rmsto renege on any dealswithout fear of retaliation and because theAnglo-Dutch auction eliminatesthe stage of the ascending auction when just oneexcess bidder remains, at which point therules against collusionand predationmay not becredible. Consortiumformation will also bediscouraged. Imaginethere are two strong biddersfor an item.In an ascending auction theyare unlikely to be challenged if theyform a consortium, so theyhave an incentiveto do so. Butin an Anglo-Dutch auction, formingthe consortium would open up an opportunityfor new entrants who wouldnow have achance tomake it to the Ž nal sealed-bidstage. Sothestrong Ž rmsare much lesslikely to bid jointly. Butthe Anglo-Dutch should also capturemuch ofthe bene Ž tofan ascending auction. It willbe more likely to sell to the highest valuer than apuresealed-bid auction, both because itdirectly reduces the numbers allowed into the sealed-bid stageand also becausethe two Ž nalists can learnsomething about each other ’s and theremaining bidders ’ perceptionsof theobject ’svaluefrom behavior during the ascending stage. When theAnglo-Dutch auction isextended to contexts in which individual biddersare permitted to win multiple units and thereare complementarities betweenthe objects, theascending stagemakes it morelikely that bidderswill win efŽ cientbundles than ina puresealed-bid auction. Finally,I conjecturethat theascending stagesof the Anglo-Dutch auction may extractmost of the information that wouldbe revealed by a pureascending auction, raisingrevenues if bidders ’ informationis “afŽ liated,” whilethe sealed-bid stagemay do almostas wellas apuresealed-bid auction incapturing extrarevenues dueto the effects of bidders ’ riskaversion, budget constraints and asymmetries. This suggeststhe Anglo-Dutch auction mayoutperform ascending and sealed-bid auctions evenif it attracts no additional bidders. In short, theAnglo-Dutch auction oftencombines the best of both theascend- ingand thesealed-bid worlds. PaulKlemperer 183

Antitrust Effectiveantitrust iscritical to Ž ghtingcollusion and predationin auctions. But antitrust enforcementin the context of auctions seemsmuch lighterthan in “ordinary” economicmarkets. TheU.S. Departmentof Justice has pursued someauction signalingcases, but thelegal status ofmany ofthe kinds ofbehavior discussed inthis articleremains ambiguous, and collusionin takeover battles for companies islegalin theUnited States. Europeanantitrust has beeneven weaker, as evidencedby T-Mobil ’s willing- ness to conŽ rmexplicitly the signaling behavior described earlier. True, when apparentlysimilar behavior was observedin the more recent German third- generationspectrum auction, Ž rmsrefused to con Ž rm ofŽ ciallythat theywere signalingto rivals to end theauction. Evenso, the FinancialTimes reportedthat “[o]neoperator has privatelyadmitted to altering the last digitof its bid in a semi-seriousattempt to signal to other participants that itwas willingto accept [fewerlots to end theauction] ” (Robertsand Ward, 2000,p. 21).This kindof signalingbehavior could perhaps bechallenged as an abuse of “joint dominance ” underEuropean law. But European regulators have showed no interestin pursuing such matters. Firmsare also permittedto make explicit statements about auctions that would surelybe unacceptable ifmadeabout a “normal” economicmarket. For example, beforethe Austrian third-generationspectrum auction, TelekomAustria, thelarg- estincumbent and presumablythe strongest among thesix bidders, said it “would be satisŽ edwithjust twoof the12 blocks of frequency on offer ” and “ifthe[5 other bidders]behaved similarly it should bepossible to getthe frequencies on sensible terms,” but “itwould bid for a thirdfrequency block if one of its rivals did ” (Crossland, 2000).It seemsinconceivable that adominant Ž rm in a “normal” marketwould be allowed to make the equivalent offer and threatthat it “would be satisŽ edwitha marketshare ofjust one-sixth ” and “if the other Ž ve Ž rmsalso stick toone-sixthof the market each, itshould bepossibleto sell at high prices, ” but “it wouldcompete aggressively for a largershare, ifany ofits rivals aimed for more than one-sixth. ”10 Just as damaginghas beenthe European authorities ’ acceptance ofjoint- biddingagreements that are,in effect, open collusion. Combinations that are arrangedvery close to the auction date(as intheexample of Switzerland discussed earlier)should beparticularly discouraged since they give no timefor entrants to emergeto threaten the new coalition. One viewis that auction participants should

10 Similarly,during the Germanthird-generation spectrum auction, MobilComtold a newspaperthat “should[Debitel] fail tosecure a license[it could]become a ‘virtualnetwork operator ’ usingMobil- Com’snetworkwhile saving onthe costof the license ” (Benoit, 2000,p. 28). This translates roughlyto a Ž rm in a “normal” marketsaying it “wouldsupply a rivalshould it chooseto exit the market, ” but MobilCom’sremarkswent unpunished. 184Journal of Economic Perspectives

generallybe restricted to entities that existwhen theauction is Ž rstannounced, although exceptionswould clearly be necessary. Theantitrust agencies ’ responseto predationin auction marketshas also been feeble.Dominant bidderssuch as Glaxoand Paci Ž cTelephonein the examples aboveare apparently allowed to make open threatsthat theywill punish new entrants. Forexample, Glaxo ’slettingit be known that it “wouldalmost certainly top arivalbid, ” wouldroughly translate to an incumbent Ž rm in a “normal” economicmarket saying it “wouldalmost certainly undercut any newentrant ’s price.”1 1 Regulatorsshould takesuch threatsseriously and treatauction marketsmore like “ordinary” economicmarkets.

Tailoring Auction Design to the Context

Good auction designis not “one size Ž ts all.” It must besensitiveto the details ofthe context. A good exampleof this lesson —and ofour other principles —is affordedby the recent European third-generation (UMTS) mobile-phone license auctions. TheUnited Kingdom, which ran the Ž rstof theseauctions, originallyplanned to sell just four licenses.12 In this case, thepresence of exactly four incumbent operatorswho had theadvantages ofexisting brand names and networkssuggested that an ascending auction mightdeter new Ž rmsfrom bidding strongly in the auction oreven from entering at all.So thegovernment planned an Anglo-Dutch auction. Anascending stagewould have continued untiljust Ž vebiddersremained, afterwhich the Ž vesurvivorswould have madesealed bids, requiredto be no lower than thecurrent price level, for the four licenses. 1 3 Thedesign performed ex- tremelywell in laboratoryexperiments in both ef Ž ciencyand revenuegeneration. Butwhen itbecame possible to sell Žve licenses,an ascending auction made moresense. Becauseno bidderwas permittedto winmore than onelicense, at least onelicense had tobe sold toa newentrant. This wouldbe a suf Ž cientcarrot to

11 Similarly,Paci Ž c Telephone’sremarkthat “ifsomebodytakes Californiaaway fromus, they ’ll never makeany money ” seemsto correspond to threatening that “ifanyonetries to compete with us, we ’ll cut the priceuntil they losemoney. ” Further, Paci Ž c Telephone’shiringof an auctiontheorist to explain the winner’scurseto competitors might correspond to hiring an industrialeconomist to explain the theoryof the dif Ž cultiesof entering new markets to potential entrants. 12 Iwas the principalauction theorist advising the U.K. government ’sRadiocommunicationsAgency, whichdesigned and ran the recentU.K. mobile-phonelicense auction. KenBinmore had aleadingrole and supervisedexperiments testing the proposeddesigns. Other academic advisors included Tilman Borgers,Jeremy Bulow, Philippe Jehiel and JoeSwierzbinski. 13 It was proposedthat allfour winners would pay the fourth-highest sealedbid. Since the licenseswere not quiteidentical, a Ž nal simultaneousascending stage wouldhave followedto allocate them more efŽ cientlyamong the winners.The sealed-bid stage couldbe run using an ascendingmechanism that wouldhide the actual bidseven from the auctioneer,if this wouldreduce political problems. See Klemperer(1998, 2001b,2002), RadiocommunicationsAgency (1998a, b) and Binmoreand Klemperer (2002) formore details. WhatReally Mattersin Auction Design 185

attract severalnew entrants intheU.K. contextin which itwas veryunclear which newentrant(s) mightbe successful. 14 Becauselicenses could not bedivided, bidderscould not colludeto divide the market without resort to side payments. As aresult,the problems of collusion and entrydeterrence were minimal, and a versionof an ascending auction was thereforeused foref Ž ciencyreasons. The auction was widelyjudged asuccess; ninenew entrants bidstrongly against thein- cumbents, creatingintense competition and record-breakingrevenues of 22.5 bil- lionpounds. TheNetherlands ’ salecame next. Their key blunder was tofollow the actual Britishdesign when theyhad an equalnumber ( Ž ve)of incumbents and licenses. It was not hard topredict (indeed, prior to theauction, an earlydraft of this paper, quotedin theDutch pressand Maasland, 2000, did predict)that veryfew entrants wouldshow up. Netherlandsantitrust policy was as dysfunctional as theauction design,allowing the strongest potential entrants tomake deals with incumbent operators.In theend, just oneweak new entrant (Versatel) competed with the incumbents. As wehave alreadydiscussed, withjust oneexcess bidder in an ascending auction, itwas unsurprisingwhen theweak bidder quit early amid allegationsof predation, at lessthan 30percent of the per capita U.K. prices.Six months later,the Dutch parliamentbegan an investigationinto the auction process. Aversionof the Anglo-Dutch designwould probably have workedbetter in the Netherlandscontext. There are reasons tobelieve Versatel would have bidhigher inthe sealed-bid stage than theprice at which itquit the ascending auction. In addition, thefear of this wouldhave madethe incumbents bidhigher. Further- more, the “hope and dream ” that asealed-bidstage gives weaker bidders might have attractedmore bidders and discouragedthe formation of the joint-bidding consortia. TheItalian governmentthought ithad learnedfrom the Netherlands Ž asco. It also chose roughlythe U.K. design,but stipulatedthat ifthere were no more “serious” bidders(as de Ž ned byprequali Ž cation conditions) than licenses,then the numberof licenses could, and probablywould, bereduced. At Ž rstglance, this seemeda cleverway to avoid an uncompetitiveauction, but (as Iand others argued)the approach was fundamentally  awed. First,it isputtingthe cart before thehorse to create an unnecessarilyconcentrated mobile-phonemarket to make an auction lookgood. Second, ourearlier discussion demonstratesthat arulethat allowsthe possibility that therewill be just onemore bidder than licensedoes not guaranteea competitiveascending auction! Also, itwas clearthat thenumber of likelyentrants intoan ascending auction was much smallerthan ithad beenfor the UnitedKingdom, in largepart because weakerpotential entrants had Ž gured out

14 In largepart, this was becausethe United Kingdomran the Ž rst third-generationauction. Goingto market Ž rst was adeliberatestrategy ofthe auctionteam, and the sustained marketingcampaign was also important. TheU.K. auctionattracted 13bidderswho then learntabout others ’ strengths, and noneof the eightsubsequent auctions had morethan sevenbidders. 186Journal of Economic Perspectives

fromthe earlier auctions that theywere weaker and that theytherefore had little chance ofwinning such an auction. In theevent, just sixbidders competed for Ž ve licenses,and theauction endedamid allegations of collusion after less than two days ofbiddingwith per capita revenuesbelow 40 percent of the U.K. level,about half theamount thegovernment was expecting.Again, an Anglo-Dutch orpure sealed-biddesign would probably have performedbetter. Klemperer(2001b, 2002) discusses the2000 –2001European spectrum auc- tions inmuch moredetail.

Conclusion

Much ofwhat wehave said about auction designis no morethan an applica- tionof standard antitrusttheory. The key issues inboth Ž eldsare collusion and entry.The signaling and punishment strategiesthat support collusionin auctions arefamiliar from “ordinary” industrialmarkets, as are Ž rms’ verbalencouragement tocollude and thepredatory threats they make. Our point that evenmodest biddingcosts maybe a seriousdeterrent to potential bidders is analogous tothe industrial-organizationpoint that thecontestability of a marketis nonrobust to evensmall sunk costs ofentry. We also arguedthat because an ascending auction ismore likely than asealed-bidauction tobe won bythe strongest Ž rm, the ascending auction maytherefore be lessattractive to bidders and maytherefore be less proŽ tablethan asealed-bidauction; this isjust an exampleof the standard industrialorganization argument that amarketthat isin principle more compet- itive(for example, “Bertrand” rather than “Cournot”)islessattractive to enter and so mayin fact beless competitive. A particularfeature of auction marketsis that “winner’s curse” effectsmay mean that sealed-bidand Anglo-Dutch auctions not onlyattract more Ž rmsthan ascending auctions, but mayalso leadto better outcomesfor the auctioneer for a givennumber of Ž rms.But there is no justi Ž ca- tionfor the current feebleness of antitrust policy in auction markets:regulators should treatthem much morelike “ordinary” economicmarkets. However,none ofour examples of auction failuresshould betaken as an argumentagainst auctions ingeneral. Most auctions workextremely well. Occa- sionally—forexample, when thereare too fewpotential bidders or large costs of supplyingnecessary information to bidders —aformof structured negotiations may bebetter, but an auction isusually more attractive to potential buyers, who are crucialto asale ’ssuccess (Bulowand Klemperer,1996). Even relatively unsuccess- fulauctions, such as theNetherlands and Italian spectrumauctions, wereprobably moresuccessful than the “beautycontest ” administrativehearings used toallocate third-generationspectrum in several other European countries. For example, the Spanish beautycontest yieldedjust 13eurosper head ofpopulation, but generated considerablepolitical and legalcontroversy and awidespreadperception that the outcomewas both unfairand inef Ž cient,all problems that aretypical of such procedures(Binmore and Klemperer,2002; Klemperer, 2000b). The dif Ž culties PaulKlemperer 187

withthe French beauty contest mean that Francehas not onlymissed its govern- ment’soriginallyplanned datefor allocation of the spectrum (already by a yearat thetime of writing), but also missedEuropean Union deadlines. In conclusion, themost important features of an auction areits robustness against collusionand itsattractiveness to potential bidders. Failure to attend to theseissues can leadto disaster.Furthermore, anyone settingup an auction would befoolishto followpast successful designsblindly; auction designis not “one size Ž ts all.” Whilethe sealed-bid auction performswell in some contexts, and theAnglo- Dutch auction isidealin othercontexts, theascending auction has also frequently beenused verysuccessfully. In thepractical design of auctions, localcircumstances matter,and thedevil is in the details. y Iwasthe principal auction theorist advising the U.K. government ’sRadiocommunications Agency, whichdesigned and ran the recent U.K. mobile-phonelicense auctiondescribed here, andhave advised several otherU.K. government agencies,but the views expressed inthis paper aremine alone.Although some observers thoughtsome ofthe behavior described above warrantedinvestigation, I donotintend to suggestthat any of it violates any applicablerules orlaws.I amvery gratefulto many colleagues,including Sushil Bikhchandani, Nils-Henrik von derFehr, Tim Harford, Emiel Maasland,Margaret Meyer, MikeRothkopf, David Salant, RebeccaStone, Timothy Taylor, Chuck Thomas, Tommaso Valletti, Michael Waldman, Mark Williamsand especially my coauthorsJeremy Bulow andMarco Pagnozzi, for helpful advice.

References

Ausubel,Lawrence M. 1998. “An EfŽ cient Ballard,Charles L., JohnB. Shovenand John Ascending-BidAuction for Multiple Objects. ” Whalley. 1985. “GeneralEquilibrium Computa- Mimeo,University ofMaryland. tions ofthe MarginalWelfare Costs of Taxesin Ausubel,Lawrence M. andPeter Cramton. the United States. ” AmericanEconomic Review. 1998. “DemandReduction and Inef Ž ciency in March, 75:1, pp. 128 –38. Multi-Unit Auctions. ” Mimeo,University of Benoit,Bertrand. 2000. “BiddersWarned in Maryland. German3G Phone Auction. ” FinancialTimes. Ausubel,Lawrence M., PeterCramton, Pres- August 2, p.28. tonMcAfee and John McMillan. 1997. “Syner- Betton,Sandra and Espen B. Eckbo. 2000. giesin Wireless Telephony: Evidence from the “Toeholds,Bid Jumps, and ExpectedPayoffs in BroadbandPCS Auction. ” Journalof Economics Takeovers.” Reviewof Financial Studies. Winter, andManagement Strategy. Fall, 6:3, pp. 497 –527. 13:4, pp. 841 – 82. Back,Kerry and Jaime F.Zender. 1993. “Auc- Bikhchandani,Sushil. 1988. “Reputation in tions ofDivisible Goods. ” Reviewof Financial Stud- RepeatedSecond-Price Auctions. ” Journalof Eco- ies.Winter, 6:4, pp. 733 – 64. nomicTheory. October,46:1, pp. 97 –119. Back,Kerry and Jaime F.Zender. 1999. “Auc- Binmore,Ken and Paul D. Klemperer. 2002. tions ofDivisible Goods with EndogenousSup- “TheBiggest Auction Ever: The Sale of the Brit- ply.” WorkingPaper, Washington University in ish 3GTelecom Licences. ” EconomicJournal. St. Louisand University ofArizona. Forthcoming. 188Journal of Economic Perspectives

Bulow,Jeremy I.andPaul D. Klemperer. its Applications. ” EuropeanEconomic Review. May, 1996. “Auctionsversus Negotiations. ” American 42:3-5, pp. 757 –69. EconomicReview. March, 86:1, pp. 180 –94 Klemperer,Paul D. 1999a. “AuctionTheory: A Bulow,Jeremy I.andPaul D. Klemperer. Guideto the Literature. ” Journalof Economic 2002. “Pricesand the Winner ’s Curse.” Rand Surveys. 13:3, pp. 227 –86. Alsoreprinted in The Journalof Economics. Forthcoming. CurrentState of Economic Science, Volume 2. 1999. Bulow,Jeremy I., MingHuang and Paul D. ShriBhagwan Dahiya, ed.Rohtak, India: Spell- Klemperer. 1999. “Toeholdsand Takeovers. ” bound,pp. 711 –66. Journalof Political Economy. June,107:3, pp. 427 – Klemperer,Paul D. 1999b. “ApplyingAuction 54. Theoryto Economics. ” OxfordDepartment of Burguet,Roberto and Jozsef Sakovics. 1996. EconomicsDiscussion Paper, April. “ReservePrices Without Commitment. ” Games Klemperer,Paul D. ed. 2000a TheEconomic andEconomic Behavior. August, 15:2, pp. 149 – 64. Theoryof Auctions. Cheltenham,U.K.: Edward Burrough,Brian and John Helyar. 1990. Bar- Elgar. bariansat the Gate: The Fall ofRJR Nabisco. Lon- Klemperer,Paul D. 2000b.“ Spectrumon the don: Arrow. Block.” Wall StreetJournal (Asia). May 10, p. 8. Cauley,Leslie and Mary LuCarnevale. 1994. Also at http://www.paulklemperer.org . “WirelessGiants, SomeSurprise Players to Seek Klemperer,Paul D. 2001a. “Why EveryEcon- NewGeneration of Licenses. ” Wall StreetJournal. omistShould Learn Some Auction Theory. ” October31, p.A4. Forthcomingin Advancesin Economicsand Econo- Coase,Ronald H. 1959. “TheFederal Com- metrics:Invited Lectures to Eighth World Congressof municationsCommission. ” Journalof Law and theEconometric Society. M.Dewatripont,L. Hansen Economics. October,2, pp. 1 – 40. and S.Turnovksy, eds.Cambridge, U.K.: Cam- Cramton,Peter and Jessie A. Schwartz. 1999. bridgeUniversity Press.Also at http://www. “CollusiveBidding in the FCCSpectrumAuc- paulklemperer.org . tions.” WorkingPaper, University ofMaryland. Klemperer,Paul D. 2001b. “What ReallyMat- Cramton,Peter and Jesse A. Schwartz. 2000. tersin Auction Design. ” WorkingPaper version, “CollusiveBidding: Lessons from the FCCSpec- NufŽ eldCollege, Oxford University Discussion trumAuctions. ” Journalof Regulatory Economics. Paper.Also at http://www.paulklemperer.org . May, 17:3, pp. 229 –52. Klemperer,Paul D. 2002.“ How(Not) toRun Cramton,Peter, Robert Gibbons and Paul D. Auctions: TheEuropean 3G Telecom Auctions. ” Klemperer. 1987. “Dissolvinga PartnershipEf Ž - EuropeanEconomic Review. Forthcoming.Also at ciently.” Econometrica. 55:3, pp. 615 –32. http://www.paulklemperer.org . Crossland,David. 2000. “Austrian UMTS Auc- Klemperer,Paul D. andMargaret A. Meyer. tionUnlikely to Scale Peaks. ” Reuters. October 1989. “SupplyFunction Equilibriain Oligopoly 31. Availableat http://www. totaltele.com . UnderUncertainty. ” Econometrica. November, Fehr,Nils-Henrik von der and David Harbord. 57:6, pp. 1243 –277. 1998. “Competitionin Electricity Spot Markets: Klemperer,Paul D. andMarco Pagnozzi. EconomicTheory and International Experi- 2002. “Advantaged Biddersand SpectrumPrices: ence.” MemorandumNo. 5/ 1998,Department An EmpiricalAnalysis. ” Forthcoming. ofEconomics,University ofOslo. Koselka,Rita. 1995. “PlayingPoker with Craig Grimm, Veronika,Frank Riedel and Elmar McCaw.” Forbes. July3, pp. 62 –3. Wolfstetter. 2001. “LowPrice Equilibrium in Maasland,Emiel. 2000. “Veilingmiljarden Multi-Unit Auctions: TheGSM SpectrumAuc- Zijn eenFictie (Billions from Auctions: Wish- tionin Germany. ” WorkingPaper, Humboldt ful Thinking). ” EconomischStatistisch eBerichten. Universitat zu Berlin. June9, p. 479.Translation availableat http:// Hall,Robert E. 2001. DigitalDealing. New www.paulklemperer.org . York:W. W.Norton. Malvey,Paul F., ChristineM. Archibaldand Hansen,Robert G. 1986. “Sealed-Bidversus SeanT. Flynn. 1996. “Uniform-PriceAuctions: OpenAuctions: TheEvidence. ” EconomicInquiry. Evaluationof the TreasuryExperience. ” Work- January, 24:1, pp. 125 – 42. ingPaper, U.S. Treasury. Jehiel,Phillipe and Benny Moldovanu. 2001. Maskin,Eric S. 1992. “Auctionsand Privatiza- “TheUMTS/ IMT-2000License Auctions. ” tion,” in Privatization. H.Siebert,ed. Tubingen: WorkingPaper, University CollegeLondon and Mohr,pp. 115 –36. University ofMannheim. McAfee,R. Prestonand John McMillan . 1988. Klemperer,Paul D. 1998. “AuctionsWith Al- “SearchMechanisms. ” Journalof Economic Theory. mostCommon Values: The ‘Wallet Game’ and February,44:1, pp. 99 –123. WhatReally Mattersin Auction Design 189

McMillan,John. 1994. “SellingSpectrum tionChoice: A NumericalAnalysis. ” Mimeo, Rights.” Journalof Economic Perspectives. Summer, University ofCalifornia at LosAngeles. 8:3, pp. 145–62. Roberts,Dan. 2000. “PhoneNumbers that Mead,Walter J. andMark Schneipp. 1989. CouldWell Result in Panic. ” FinancialTimes. “CompetitiveBidding for Federal Timber in Re- October19, p. 38. gion6, An Update: 1983 –1988.” Communityand Roberts,Dan and Andrew Ward. 2000. “Little OrganizationResearch Institute, University of Goldat the Endof the Spectrum. ” Financial California,Santa Barbara,Contractor Report, Times. November3, p.21. USDAAward No. 40-3187-8-1683, June. Rothkopf,Michael H. andRichard Engel- Michelson,Marcel. 2000. “Swiss3G Auction brecht-Wiggans. 1993. “MisapplicationsReviews: Setto Become Battle ofGiants. ” Reuters. Novem- Gettingthe ModelRight: TheCase of Compet- ber9. Availableat http://www.totaltele.com . itiveBidding. ” Interfaces. May, 23:3, pp. 99 –106. Milgrom,Paul R. andRobert J. Weber. 1982. Salant, David. 2000. “Auctionsand Regula- “ATheoryof Auctions and CompetitiveBid- tion: Reengineeringof Regulatory Mechanisms. ” ding.” Econometrica. September,50:5, pp. 1089 – Journalof Regulatory Economics. May, 17:3, pp. 122. 195–204. Myerson,Roger B. andMark A. Satterthwaite. Simon,David P. 1994. “TheTreasury ’s Exper- 1983. “EfŽ cientMechanisms for Bilateral Trad- imentwith Single-PriceAuctions in the Mid- ing.” Journalof Economic Theory. April,29:2, pp. 1970s:Winner ’sorTaxpayer ’s Curse?” Review of 265– 81. Economicsand Statistics. November,76:4, pp. Newbery,David M. 1998. “Competition,Con- 754– 60. tracts, and Entry inthe ElectricitySpot Market. ” Stuewe,Heinz. 1999. “Auktionvon Telefon- RANDJournal of Economics. 29:4, pp. 726 – 49. frequenzen:Spannung biszur letzten Minute. ” Nyborg,Kjell and Suresh Sundaresan. 1996. FrankfurterAllgemeine Zeitung. October 29. “Discriminatoryversus Uniform Treasury Auc- TotalTelecom. 2000. “Italy’sUMTS Auction tions: Evidencefrom When-Issued Transac- toStart October. ” Reuters. October12. Available tions.” Journalof Financial Economics. September, at http://www.totaltele.com . 42:1, pp. 63–104. U.K.Monopolies and Mergers Commission. OfŽce of Gas andElectricity Markets. 1999. 1999. BritishSky BroadcastingGroup and Manches- “TheNew Electricity Trading Arrangements, July.” At http://www.open.gov.uk/offer/reta. terUnited: A Reporton the Proposed Merger. Cm htm . 4305.London: The Stationery Of Ž ce. Persico,Nicola. 2000. “InformationAcquisi- Vickrey,William. 1961. “Counterspeculation, tionin Auctions. ” Econometrica. 68:1, pp. 135 – 48. Auctions, and CompetitiveSealed Tenders. ” Power U.K. 1999. “TheProblems with the Journalof Finance. 16, pp. 8–37. Pool.” August 31, 66, p. 14. Wighton,David. 1995a. “WellcomeAccepts RadiocommunicationsAgency. 1998a. “UMTS GlaxoBid and CriticisesTrust. ” FinancialTimes. AuctionDesign. ” UMTS AuctionConsultative March8, p. 27. GroupPaper 14 of 1998. Available as UACG(98)14 Wighton,David. 1995b. “WellcomeStill Smart- at http://www.spectrumauctions.gov.uk . ingOver Handling of Trust ’s Stake.” Financial RadiocommunicationsAgency. 1998b. “UMTS Times. March8, p.32. AuctionDesign 2. ” UMTS AuctionConsultative Wolfram,Catherine D. 1998. “StrategicBid- GroupPaper 16 of 1998. Available as UACG(98)16 dingin aMultiunit Auction: An EmpiricalAnal- at http://www.spectrumauctions.gov.uk . ysis ofBidsto Supply Electricity in England and Reinhert,Vincent and Gregory Belzer. 1996. Wales.” RANDJournal of Economics. 29:4, pp. 703 – “SomeEvidence on BidSharing and the Useof 25. Informationin the U.S. Treasury ’sAuctionEx- Wolfram,Catherine D. 1999. “Measuring periment.” WorkingPaper, Board of Governors DuopolyPower in the BritishElectricity Spot ofthe FederalReserve System. Market.” AmericanEconomic Review. September, Riley,John G. andHuagang Li. 1997. “Auc- 89:4, pp. 805 –26.