Auction Design (4) - UACG(99)17
This paper reports, and responds to, comments received on paper UACG(99)8 which gave an account of two versions of an auction design, Modified-FCC version 1 and Modified-FCC version 2. These versions were developed to be suitable for an auction of five Third Generation (3G) mobile communications licences of unequal size and value with one reserved for a new entrant. The versions have now been experimentally trialled at University College London (UCL) and this paper gives an account of the trials and the results obtained. There will be an opportunity to discuss the trials and this paper with Professor Binmore and the Auction Team at the 14 May UACG meeting. One version, Modified-FCC version 2 (with ratcheted reserve price), is being taken forward and incorporated in the draft Notice UACG(99)13 that will issue shortly. Comments are invited on the auction design expressed in the draft Notice, and on this paper.
2. Comments should be received by noon on Tuesday 8 June 1999 by:
Please note that the RA returns to its headquarters in Docklands with effect from 1 June. Telephone, fax and email contact details will be unchanged.
3. 12 responses were received to UACG(99)8 and some requested confidentiality. The main points arising from the responses are summarised in general terms in the following list:
- a majority welcomed the basic Modified-FCC design
- some concerns were expressed over version 1 and version 2 and their consequences but there was also some support for version 2
- two respondents suggested a combinatorial auction design
- there was concern that 3 waivers and 2 postponements (referred to as recesses below and in the draft Notice to issue shortly) will slow the auction
- various more detailed points of auction design were raised and are being taken into account in ongoing work. Comments on some points are given in the Section "Auction Design - comments on detailed aspects" below.
Response to comments received
4. We have taken account in our developing plans of the broad support for the basic Modified-FCC design coupled with concerns over version 1 and version 2. The trials confirm the efficiency of both versions and a later Section of this paper gives our reasons for proceeding with version 2.
5. Two respondents favoured a combinatorial auction design allowing bidders to select within limits the amount of spectrum per licence as part of the auction. We believe such a design would be unlikely to generate benefits as the number of licences awarded would be likely to be the minimum allowed under the rules. Also there is a risk of other inefficient outcomes eg a licensee obtaining non-contiguous spectrum.
6. Concerns were expressed that the auction could be lengthy particularly with 3 waivers and 2 recesses per bidder. As the auction progresses it is believed that, from time to time, bidders will want extra time to consult within their own organisation or with their advisers. The ratcheted deposit (more information is given in UACG(99)8 paragraph 16) may present a particular need for bankers to be consulted, although we hope that bidders will anticipate this need as far as possible. For these reasons 3 waivers and 2 recesses have been retained in the draft Notice that will issue shortly. However we would welcome further discussion and comment on this point, as we share the view that the auction should not be overlong.
7. One way of speeding the auction could be to implement multiple non-discretionary bidding in which, say, one of nine bids is permitted each being a "minimum raise multiple" of the minimum bid and each being represented by a tick-box. Comments are sought on this and any other ideas for increasing the speed of the auction.
8. As mentioned in the introduction Professor Binmore will attend the UACG meeting on 14 May to answer questions. In addition the RA Auction Team will be pleased to hold individual meetings with respondents on request.
Auction Trials - introduction
9. Trials of both versions of the auction design have been carried out under the direction of Professor Binmore. As for the earlier set of trials for the four licence auction design (details in paper UACG(98)16) student subjects were recruited. They were paid £10 for turning up, plus an extra amount proportional to the amount they saved for their fictitious company as a consequence of acquiring a licence at less than their company's estimate of its value (this extra amount could be negative if they paid more than the value of the licence to their company). It is, of course, important that the subjects in such experiments be adequately motivated if they are to serve as adequate models of real-life players.
10. For the purposes of the experiment, the subjects were told a story about buying computer chip factories in Hong Kong, and the valuations were quoted in millions of dollars. They learned that five factories were for sale at the auction and that there were major bidders and minor bidders. They were given their own valuations of the five factories, but had only restricted information about the valuations of the other nine subjects. This restricted information consisted of the average value assigned to each factory by the major bidders and the average value assigned to each factory by the minor bidders. In addition, they were told the extent of the potential spread around these average values that might occur.
11. The subjects participated in four auctions, one "for practice" and three "for real". In the real auctions, they were paid in proportion to the profit they made for their company. In each set of auctions, the valuations were the same, but the subjects represented different companies. The assignment of roles was fixed so that each subject was a major bidder in at least one real auction. After a succession of pilot experiments that were used to refine the manner in which the data was presented to the subjects, trials of both versions to gather usable data began. Sufficient auctions were run to give confidence in the experimental results.
12. Valuations were selected having regard both to the need to challenge the auction design in the most difficult circumstances and to the potential range of valuations of bidders in the UMTS auction. As the valuations were informed inter alia by our advisers modelling, it would be inappropriate to reveal them
Auction Trials - results
13. The paragraphs below extract major lessons from the trials .
14. The experimental results turned out very much in line with theoretical predictions made by UCL. Both versions of the design proved to be very efficient for valuations considered to be likely for the five licence scenario, and for the valuations used in the previous round of experiments for the four licence design. This means that in these circumstances the designs performed very well in putting licences with those who valued them the most.
15. Before the trials UCL used a theoretical rule-of thumb to predict the endogenous reserve prices (produced by a Qualifying Stage in version 1 and by a ratcheted reserve price in version 2 - more details in paper UACG(99)8). For version 1 the endogenous reserve price was predicted to be the 4th highest valuation by a new entrant for the reserved licence, for version 2 the endogenous reserve price (for a "small" 2x10+5 MHz licence) was predicted to be 5/6 the second highest valuation for the reserved licence when that is 2x15 MHz, and 5/8 the second highest valuation for the reserved licence when that is 2x20 MHz. The observed experimental results were remarkably close to the theoretical predictions.
16. The versions were also tested with "wild-card" valuations believed to be very unlikely in practice. In some cases with these valuations one licence was unsold at the end of the auction. We believe this to be unlikely in practice, but if it did occur we expect that the unsold licence would be made available by auction at least a year later.
Auction Trials - conclusions
17. Overall there was no statistically significant outcome from the trials to direct us to a particular version; both performed very efficiently (putting the licences in the hands of those who value them the most) over the range of valuations considered likely in practice and both performed similarly under the very unlikely "wild-card" valuations.
Version 1 and Version 2
18. Taking account of the results of the trials and the views of industry in the context of the auction aim and objectives, we have decided to take forward implementation of Modified-FCC version 2 (with ratcheted reserve price) as the UK 3G auction design. This design will provide an efficient and competitive auction including a ratchet to ensure the price paid for the reserved licence is not disproportionate to that paid for the other licences. An advantage of version 2 over version 1 (with Qualifying Stage) is its relative simplicity presenting bidders with a one-stage auction and more straightforward strategic issues in developing their bidding strategy. Without the ratchet (and either with or without the Qualifying Stage of version 1) a possible consequence of limited bidding on the non-reserved licences and strong bidding on the reserved licence would be a large difference in the price paid per MHz between a new entrant and the incumbents.
Auction Design - comments on detailed aspects
19. - Once a bid has been made it will not be possible to withdraw the bid. To allow bidders to withdraw bids would risk inefficiency as other bidders might for example have withdrawn in the meantime or bid against an effectively bogus competitor on the basis of misleading information. It is an integral part of the auction design that highest bidders can only bid for another licence if their bid is topped, for similar reasons.
- It has been suggested that the current high bidder on a licence should be permitted to raise their own bid. Comments are invited on the value of this.
- We do not plan to provide bidders with details (value and bidder name) of all bids, partly because we believe this information would not be of legitimate use and partly because it would provide scope for bid signalling. The identity of highest bidders (and bid values) is made known to other bidders as part of the auction design to assist their strategic planning.
- Proposed penalties are set out in the draft Notice to issue shortly.
Auction Design - conclusions
20. In view of the successful trials, the responses to UACG(99)8, and the auction aim and objectives, we are developing plans for Modified-FCC version 2 (with ratcheted reserve price) as the UK 3G auction design. This is being incorporated in a draft of the Notice that will be issued for informal consultation shortly as UACG(99)13.
10 May 1999