Contact Information
Darin Lee, Ph.D.
LECG-Cambridge
350 Massachusetts Ave, Suite 300
Cambridge, MA 02139
+1-617-761-0108
Darin_Lee@lecg.com
Lessons from the Nigerian GSM Auction
Darin Lee, LECG Perspectives, January 2002
Of the recent spectrum auctions worldwide, those for third-generation (3G) spectrum throughout Europe during 2000 have received the most attention. Indeed, the enormous sums of money paid by bidders in the U.K. and Germany not only motivated many other governments worldwide to allocate their third-generation spectrum via auction, but may also be to blame — at least in part — for the current dismal financial state of many wireless telecommunications firms.[1] Although the various 3G auctions throughout Europe have been the primary focus of attention for scholars and industry analysts, one of the most novel recent spectrum auctions — for three second-generation (GSM) licenses in Nigeria — went largely unnoticed. Nevertheless, the Nigerian GSM auction should be of particular interest to auction theorists and regulatory bodies alike, as it was the first time ever that a variation of the so called “Anglo-Dutch” auction format, which combines elements of ascending and sealed-bid formats, has been used to auction radio spectrum.[2] Moreover, the Nigerian GSM auction was characterized by an unusual number of important rule changes in the days leading up to the auction, and thus serves as an interesting example of how regulatory authorities need to be prepared to change their auction rules — possibly at the last moment — to accommodate the concerns of bidders.
This article discusses some of the novel aspects of the Nigerian GSM auction. Moreover, I discuss some last-minute rule changes in the context of the stated objectives of the Nigerian Communication Commission (NCC). I then consider what lessons can be drawn from the auction and its design.
Spectrum Auction Formats
Virtually all recent spectrum auctions, including most U.S. FCC auctions and all of the European 3G auctions conducted in 2000, were based on variations of the ascending-auction format.[3] Ascending auctions are multiple-round auctions in which prices continue to rise until supply equals demand. Ascending formats have been widely adopted because the allocations that they generate tend to be highly efficient. That is, ascending auctions have the desirable property that they usually allocate scarce resources to those bidders who value them the most and will therefore make the best economic use of them. Since ascending auctions give bidders the opportunity to top their rivals’ bids, bidders who fail to win a license in an ascending auction do so because they refuse to pay the asking price. Ascending-auction formats also have the property that they reveal a great deal of information during the bidding process, and thus, tend to reduce the “winner’s curse.”[4]
Although most recent spectrum auctions have tended to favor variations of the ascending-auction format, disappointing results in some of the European 3G auctions (e.g., Italy, the Netherlands, and Austria) have prompted some auction theorists to explore alternative designs.[5]
One alternative auction format is the so called “Anglo-Dutch format.”[6] An Anglo-Dutch auction for N homogenous objects begins with a traditional ascending auction, which continues until N+i bidders remain.[7] The N+i bidders then participate in a sealed-bid phase with the minimum bid equal to the price from the final round of the ascending phase. It has been suggested that the Anglo-Dutch format retains virtually all of the benefits of both the ascending and sealed-bid auction formats while mitigating most of their shortcomings. In particular, some argue that most of the efficiency and information-revelation properties of the ascending auction are retained (through the use of the first phase), while the presence of a final sealed-bid phase encourages entry and discourages collusive behavior.
The Nigerian GSM Auction
The Nigerian GSM auction was the first auction for radio spectrum to utilize a variation of the Anglo-Dutch format. The auction took place during January 17–19, 2001, in Abuja, Nigeria, with five bidders competing for three licenses. The auction rules upon which bidders based their preparations were published by the NCC on December 8, 2000, in their Information Memorandum.
The Ascending-Bid Phase
The ascending phase of the Nigerian auction was somewhat novel in that an ascending-clock format was chosen.[8] In an ascending-clock auction, the auctioneer announces a common price for all bidders, and the bidders elect whether or not they are willing to accept that price. In the Nigerian GSM auction, the Auction Control Team (ACT) announced a new (higher) price each round — the announced price — and bidders were required to submit one of three possible responses. The possible responses were: Yes, which indicated that the bidder was willing to pay that price; No, which indicated that the bidder was not willing to pay that price; or Waive, which indicated that the bidder was not willing to reveal whether or not it was willing to pay that price. Once a bidder bid No, it was no longer permitted to bid in subsequent rounds of the ascending-bid phase of the auction. Each bidder was permitted to use a maximum of three waivers and the initial reservation price for each of the licenses was U.S. $100 million.
Bidders that submitted responses of Yes or Waive were considered to be active at the end of an ascending-bid round; bidders that submitted a response of No were deemed inactive. The response of each bidder was made public to all of the bidders at the completion of each round. The ascending-bid phase continued so long as there were four or more active bidders at the end of a round. If a round of bidding ended with exactly three active bidders, those three bidders were deemed “successful” and the auction ended with each successful bidder paying the announced price in the last round in which all successful bidders each responded Yes. Note that since a bidder responding Waive was considered active at the end of a round, this need not be the last announced price. If an ascending-bid round ended with strictly fewer than three active bidders, the auction proceeded to the sealed-bid phase.
The Sealed-Bid Phase
If an ascending-bid round ended with two or fewer active bidders, the auction entered the sealed-bid phase and bidders were divided into three subsets. All bidders who bid Yes in the final round were deemed grant-stage bidders (i.e., one of the three licenses was reserved for them at a price to be determined during the sealed-bid phase, but they were excluded from any further bidding).[9] Bidders who bid Waive or No in the final round of the ascending-bid phase were permitted to bid in the sealed-bid phase, and they competed for the remaining licenses (i.e., those licenses not already allocated to the grant-stage bidders). Bidders who had bid No in a round prior to the final round were not permitted to bid in the sealed-bid phase and were not eligible to win a license.
Each eligible bidder in the sealed-bid phase was required to submit a final bid of at least that amount which it last bid Yes to in the ascending phase, and no more than the final announced price from the ascending phase. Hence, there was a maximum sealed bid common to all bidders. To determine the winners and common final price paid by all winning bidders (including the grant-stage bidders), sealed bids were ordered from highest to lowest and the final license price was defined as the lowest successful sealed-bid amount. For example, if there were two licenses available in the sealed-bid phase, the final price to be paid by all successful bidders would be the second-highest sealed bid. In the event of a tie during the sealed-bid phase, successful bidders were to be determined by the toss of a coin.
Concerns of Bidders Need to Be Considered in Designing Auction Rules
The Nigerian GSM auction rules were designed with two normative goals in mind: (1) no successful bidder should have to pay more than it explicitly said it was willing to (either by bidding Yes in the ascending-bid phase or through a sealed bid), and (2) all successful bidders should pay the same price for a license. However, it soon became clear that these policy objectives depended crucially on one aspect of the auction design which made bidders extremely uncomfortable: the use of the coin toss as a tie-breaking mechanism.
From the standpoint of allocative efficiency, the coin toss was not problematic, since it would only be used in situations where bidders effectively had the same valuation. Therefore, it should not come as a complete surprise that the auction’s designers did not expect this to be problematic. Nevertheless, the possibility of not being allocated a license based on the outcome of a coin toss was extremely unsettling for a number of bidders, and thus the NCC needed to quickly revise the rules to accommodate these concerns.[10] Consequently, four days before the auction was scheduled to begin, the NCC altered the auction rules such that in the sealed-bid phase, there would no longer be a maximum bid amount. Moreover, in the event of a tie, the tied bidders would re-bid, with the new minimum bid being set at the tied bid amount, thus eliminating the need to use a coin toss.
What the NCC and its auction designers failed to realize was that these seemingly innocuous rule changes put into potential conflict the two normative objectives of the auction. By eliminating the maximum bid during the sealed-bid phase, it was now possible that the price to be paid by all three winning bidders exceeded the last-announced price to which grant-stage bidders said Yes — and hence could be expected to pay. This was in direct conflict with the first policy objective – that no successful bidder should have to pay more than it explicity said it was willing to. In order to resolve this conflict, different prices could be used (i.e., winning bidders from the sealed-bid phase could be required to pay more than grant-stage bidders if necessary), but this would have violated the second policy objective of the auction – that all bidders pay the same final price. In the end, the NCC opted to abandon its second policy objective by requiring sealed bidders to pay their bid amount, a rule which they imposed less than 12 hours before the auction was to begin.[11] As it turns out, the auction ended when the number of active bidders fell from four to three in the last round and, hence, the auction never entered the sealed-bid phase.
Even Carefully Designed Rules Can Have Loopholes
While some design issues in the original Nigerian auction rules led to inconsistencies with the stated objectives of the auction, others had the potential to drastically reduce the amount of revenue raised. For example, a less obvious design issue which remained unmodified even in the final auction rules was the determination of the final price in the event that the ascending phase ended with exactly three active bidders. In this case, the final price paid by the three successful bidders was defined as the announced price in the last round in which all three successful bidders each responded Yes. Since a bid of Waive maintained a bidder’s status as an active bidder, there was a strong incentive to use waivers even when the announced price was less than a bidder’s bid ceiling. This strategic incentive to use waivers was balanced by the desire to preserve them for situations in which they were needed for non-strategic reasons, such as requiring additional time to make a decision. Although in many endgame scenarios, this rule for determining the final price produces the “correct” price, it is not hard to conceive of examples in which the Nigerian Government would have received significantly less revenue than they were entitled to because of this oversight. Consider the following example:
In the example above, the auction ends in round 8 since the number of active bidders falls from four to three. It would not be unreasonable to expect the three successful bidders (A, B and C) to each pay $160 million, since all three of these bidders said Yes to an announced price of at least this much. According to the auction rules however, each of the winning bidders would only be required to pay the first round price of $100 million, since this is the only price at which they all simultaneously responded Yes. A more appropriate rule would have been to require each successful bidder to pay the last announced price at which all three successful bidders said Yes to that price or a price higher than it. In the example above, this would have increased the Nigerian Government's auction revenue by 60% from $300 million to $480 million, assuming that all three licenses were allocated and paid for.[12] In the actual auction, the NCC was fortunate that this oversight did not impact the amount of revenue they collected. Nevertheless, future auctions adopting a similar format will no doubt attempt to close this potentially costly loophole.
Conclusions
The Nigerian GSM auction, by virtually all reports, was considered to be a success. The auction raised U.S. $855 million (a final price of $285 million per license), substantially more than the government had expected. Indeed, given the somewhat disappointing results, (from the point of view of the governments involved), of the previous few 3G auctions in Europe, the Nigerian GSM auction surpassed virtually everyone’s revenue expectations and, based on “addressable” users, raised more revenue per capita than the 3G auctions in the Netherlands and Italy.[13]
Although the Nigerian GSM auction was perceived to be a complete success based on the revenues it raised, the outcome could have been much different, and many important lessons can be drawn from observing how its rules evolved. The experience of the Nigerian GSM auction provides valuable insight into some of the challenges facing auction designers and administrators. Bidders often have concerns which may be overlooked by auction designers, since they are unlikely to occur in equilibrium, or fail to have negative implications on issues which auction designers are typically the most concerned with, such as economic efficiency. Nevertheless, the experience of the Nigerian GSM auction is insightful since it demonstrates how the concerns of bidders may influence an auction’s final design and that despite their careful design, potential loopholes often still remain. Finally, given the high-stakes nature of these complex games, the numerous last-minute rule changes in the Nigerian GSM auction demonstrate why bidders need to be prepared for virtually all types of contingencies when participating in spectrum auctions.
Notes
[1] See, for example, “Wireless Web Woes: Telecom Players Have Spent Billions, but Profits are Still Far Off,” BusinessWeek Online, June 4, 2001, or “3G Operators Desperate For Help,” 3G Newsroom.com, May 18, 2001.
[2] Moreover, it was the first time that the “clock auction” format (one in which the price “ticks” up and bidders merely choose to accept or reject the price) of bidding has been used to auction radio spectrum.
[3] These auctions are often referred to as “simultaneous multiple round ascending auctions.” For brevity, I will adopt the term “ascending auction” to refer to auctions of this type.
[4] The “winner’s curse” describes the phenomenon that, in common-value auctions, the winners are necessarily the bidders who overestimate the true value of what is being auctioned by the largest amount.
[5] Whereas 3G auctions in the U.K. and Germany raised 129 and 103 Euros per capita, respectively, the Dutch, Austrian and Italian 3G auctions only raised 34, 18 and 42 Euros per capita, respectively.
[6] The Anglo-Dutch format was first suggested by Paul Klemperer in “Auctions with almost common values,” European Economic Review, volume 42, pages 757-769, 1998 and further elaborated in “What really matters in auction design,” unpublished manuscript, Nuffield College, Oxford University, 2000.
[7] Klemperer (1998) suggests that i=1.
[8] Although clock auctions have been used to allocate various types of commodities (flowers, treasury bills, etc.), the Nigerian auction was the first radio-spectrum auction ever to use an ascending-clock format.
[9] Note that by definition, there could be at most two grant-stage bidders.
[10] Indeed, some bidders were of the belief that not being allocated a license on the basis of losing a coin toss would have been the worst imaginable outcome, one which could not be justified, after the fact, to senior management, board members, and shareholders.
[11] It was interesting to note, however, that the NCC and its auction designer adamantly denied that the rules violated the second policy objective, insisting that all bidders still paid the same “price,” but some might be required to pay an additional “premium.”
[12] Although the author and his client felt compelled to inform the NCC of many design issues in the auction, we chose not to inform them of this oversight for obvious reasons.
[13] Based on 10 million potential Nigerian users, the Nigerian GSM auction raised roughly 95 Euros per capita, compared to 34 Euros per capita in the Netherlands and 42 Euros per capita in Italy. Moreover, if one were also to adjust for differences in income, the Nigerian auction raised more per capita than any other spectrum auction to date. It should be noted, however, that Nigeria has one of the lowest teledensities in the world, at .4 mainlines per 100 people, and the highest revenue per mainline, at $3,738 per year. Source: World Telecommunication Development Report 2000, International Telecommunication Union.