*Pages 1--30 from Microsoft Word - 48349.doc* Comparing the FCC’s Combinatorial and Non- Combinatorial Simultaneous Multiple Round Auctions: Experimental Design Report Prepared for the Federal Communications Commission Jacob K. Goeree, Caltech Charles A. Holt, University of Virginia This Version: April 27, 2005 The authors wish to acknowledge FCC staff Mark Bykowsky, William Sharkey, and Martha Stancill for input and helpful comments. Any views expressed in this paper are those of the authors and are not purported to reflect those of Caltech, the University of Virginia, the Federal Communications Commission, Wireless Telecommunications Bureau, or other members of the Commission’s staff. 1 2 I. Introduction The Federal Communications Commission (FCC) has decided to use laboratory experiments to compare the performance properties of its simultaneous multiple round (SMR) auction design with its newly developed simultaneous multiple round auction that permits combinatorial, or package, bidding (SMRPB). The purpose of this comparison is to shed light on the conditions under which the FCC should employ one auction form rather than the other for the assignment of spectrum licenses. Experiments are a natural way to compare the different auction formats, since many aspects of package bidding have proven to be extremely difficult to analyze using economic theory. Experiments may also be able to capture behavioral elements (risk aversion, winner’s curse effects, bounded rationality) that are hard to incorporate in standard theoretical models. These behavioral elements may have important effects on auction performance, and these effects may be mitigated or exacerbated by the auction form used. Furthermore, controlled laboratory experiments allow the researcher to examine the extent to which the tested auction design assigns licenses to the bidders which value them most highly. Because the researcher assigns bidders’ values for licenses and packages, at the close of the auction it is possible to evaluate the economic efficiency of the allocation. In contrast, field data obtained from actual auctions do not permit one to make a direct inference about the efficiency of the final allocation since real bidders’ values are unknown. Finally, experimentation offers the opportunity to run numerous auctions under controlled conditions while changing a single design feature. This replication and control allows the researcher to spot systematic differences across auction formats that are not due to random variations from one auction to another. The FCC has only limited experience with package bidding. Therefore, the experiments may help the Commission to avoid any procedural problems and unintended side effects that become apparent, and will facilitate the effective use of package bidding. An important first step in this process is the development of an “experimental design,” which specifies the economic environment, the experimental procedures, and the criteria used to measure performance. The purpose of this paper is to propose an experimental design. In addition to permitting a comparison between the SMR and SMRPB auctions, the experimental design can also be used as a benchmark for possible future evaluations of the performance properties of other auctions (e. g., price- driven “clock” auctions). While the FCC has sponsored an earlier experimental study of auctions that allow for package bidding, the currently proposed SMRPB design has not yet been evaluated in the laboratory. 1 Furthermore, the proposed experimental design includes several new features that supply additional realism (e. g., bidders may face budget constraints and their valuations consist of both private and common- value components). Finally, besides revenue and efficiency, the performance measures used to evaluate the different designs 1 Among previous studies, the most relevant is Cybernomics (2000), “An Experimental Comparison of the Simultaneous Multi- Round Auction and the CRA Combinatorial Auction.” Available at http:// wireless. fcc. gov/ auctions/ conferences/ combin2000/ releases/ 98540191. pdf. 2 3 now include other factors, such as the effects on small bidders, and the extent to which collusion is suppressed. Section II of this report discusses a method for generating bidder valuations for licenses based on common and private- value elements, population covered by the licenses, and economies of scale. Several treatment conditions are outlined in Section III based upon structural conditions that might make one auction form more attractive than another. Section IV describes experimental procedures such as the number of experimental sessions, and the number and experience level of subjects. Section V details the performance measures by which the auction formats are evaluated and compared. Appendix A reviews the definitions of competitive equilibrium and the core. Appendix B suggests some specific parameter values for different treatments. The auction rules are specified in detail in Appendices C and D. II. Modeling Bidder Valuations 1. Private and Common Values Auction models typically fall into one of two categories. In private value auctions, bidders know their own value for the object but are unsure about others' valuations, as in the purchase of a painting for personal enjoyment. Common value auctions pertain to situations where the object is worth the same to everyone but bidders have different information about its true value, as in the auctioning of oil drilling rights. While this dichotomy is useful from a theoretical viewpoint, real world auctions, including spectrum auctions, are likely to exhibit both characteristics. For instance, when a painting is auctioned, it may be resold in the future and the resale price will be the same for all bidders, which adds a common value element. And in the oil drilling example, private value differences may arise because of differences in production technologies. One way to capture both private and common value elements is provided by Milgrom and Weber’s (1982) “affiliated- values” approach. 2 Roughly speaking, bidders’ assessments of their values, or “signals”, are affiliated when one bidder’s high (low) signal makes it more likely that others’ signals are high (low) as well. This approach is able to reproduce the independent private values model and the pure common- value model as limit cases. However, in this setup, each bidder receives a single signal (and in equilibrium, the bidder with the highest signal wins), which precludes an analysis of the possible inefficiencies that arise when separate private and common values are present. To illustrate, consider the case of spectrum licenses where bidders’ valuations involve private value elements (e. g., geographic preferences) and common value elements (e. g., future consumer demand for wireless services). Here, an inefficient allocation results when a bidder with a relatively high private value for a particular license but who is relatively pessimistic about future demand is outbid by an overly optimistic bidder with a 2 Milgrom and Weber (1982) “A Theory of Auctions and Competitive Bidding,” Econometrica, 50, 1089- 1122. 3 5 may exacerbate winner’s curse effects since a bidder’s high estimate for one license would imply high estimates for all other licenses. For example, with a single common value, a bidder who is optimistic about the demand for wireless services in one geographic area is equally optimistic about the demand in all areas. 2. Packages: Substitutes and Complements The value a bidder places on a package of licenses need not be equal to the sum of the values for the individual licenses. For instance, if licenses A and B are considered substitutes, the value of the package AB may be less than the values of A and B added together. Alternatively, if licenses A and B are considered complements, the value of the package AB may exceed the values of A and B added together. The degree of complementarity or substitutability for licenses or packages can be generated from assumptions about the distributions of the private and common values within the bidder population. As an example, consider the case of pure private values (i. e. CV = 1) and suppose bidders’ values for licenses A and B are uniformly distributed on [100, 200]. The relationship between license values may indicate either substitutes or complements if the AB package value is distributed on [200, 500]. When the value draw for the AB package is towards the lower end (close to 200) the package value will generally be less than the sum of the license values (although higher than each of the license values individually). Conversely, when the value draw for the AB package is towards the upper end (close to 500), the package value exceeds the sum of the individual license values, indicating synergies between the two licenses. Through the use of such distributional assumptions, together with assumptions regarding license preference overlap, a wide variety of efficient license assignments can be created. For example, assume there are two bidders with strong local preferences, one for A and the other for B. Each bidder’s value for its “own” location is drawn from [100, 200], with no value for the other location. Assume, further, that there is a single national bidder whose single item values for licenses A and B are drawn from [100, 200], and whose value for the AB package is drawn from [200, 500]. Depending on the draws, the efficient allocation may involve assigning both licenses to one bidder while, in other instances, it would involve assigning each license to different bidders. 3. A Parametric Approach to Defining Valuations Another consideration is how to scale up the experiment (in terms of the number of bidders and packages) without overloading subjects with information about distributional assumptions on private values (for each license and package), the common value, and common- value signals. One approach is to take a setup like the one above for two licenses and a single package and replicate it – increasing the numbers of licenses and bidders, with values drawn from the same distributions as before. We prefer instead to use a formula- based approach that includes parametric terms to capture market size, geographic location, and complementarities. In this manner, the 5 7 4. Budget Constraints Along with a set of valuations for individual licenses and packages of such licenses, each bidder will be assigned prior to bidding a budget that limits the total payment the bidder can make at the close of the auction for the licenses it has won. The bidder’s objective is to maximize its net profit (i. e., license or package valuation minus payment) subject to this budget constraint. The experiments should, at a minimum, include some situations in which budget constraints are binding for winning bidders. However, budget constraints should not be so tight as to eliminate opportunities to observe bidding behavior related to the financial exposure issue. For example, we are interested in whether bidders will risk exposure by bidding above the stand- alone values for a group of licenses with superadditive values in the SMR auction, but they will not take the risk if their budget is inadequate. If bidders default (i. e. when their winning bids total more than their allotted budgets), they must default on all winning bids, forfeiting the licenses and incurring the penalties specified in the auction rules. In the actual FCC auctions, a bidder who defaults is required to pay the difference between the withdrawn winning bids and the amount that the licenses sell for subsequently as well as an additional amount based on a percentage of the winning bids. This percentage is higher if the default occurs after a package bidding auction. In the experiment, this process can be mimicked by having the subsequent sale prices be determined as a random draw from a known distribution. 5. Additional Considerations An important question is whether aids should be provided to experimental subjects to help them make thoughtful decisions in complex environments. This issue typically gets clarified after some pilot experiments. At present, we envision providing subjects with an interface that contains a “Package information” table, a “Bidding basket,” and “Bidding History” plus a separate “Auction Result” page, see Figure 1. In addition, it would be useful to incorporate a message area, which the experimenter can use to send general messages (e. g. “the experiment will start in a few seconds”). In the subject’s “Package Information” table there would be a row for each license/ package for which the subject has positive values. For example, in Figure 1, bidder 1 has positive values for packages A, B, C, AB, AC, BC, and ABC. The columns of the “Package information” table would identify the license or package, the current price, the private value component, and the common value signal. 8 Subjects can place bids on licenses/ packages they are interested in by moving them into the “Bidding basket,” which they can submit after all relevant licenses/ packages have been entered. In each round, bidders are informed if they are the provisional winner on a license or package through the use of colors: in Figure 1, for instance, the grey cells indicate that bidder 1 is the provisional winner on package BC. When licenses/ packages are added to 8 Figure 1 does not show a column for the common value signal. 7 8 the bidding basket, bidders will see the minimum bids that need to be submitted in the next round: for non- provisional winners, it is the current price plus a pre- specified increment. 9 In the right bottom panel, the “Bidding history” shows the bidding activity during the period on the packages relevant to the subject. The experimenter should be able to choose different information feedback conditions, e. g. show all bids and identities, show all bids (but not identities), or show only own bids. 10 Finally, by clicking on the “Auction Result” tab on the top- left panel, subjects can view previous auction results. The corresponding screen will appear automatically after an auction has ended, and subjects will get some time to interpret the results before a new auction starts and the screen changes back to the one shown in Figure 1. Of course, other experimenters are likely to have different ideas about how to best represent all the information to subjects, so the interface presented in Figure 1 should be seen as a suggestion not a requirement. Figure 1. Example of Proposed Client Interface In any case, the experiment software would have to be extremely simple from the subject’s point of view, without many of the extra details required in the actual FCC 9 Alternatively the “Current Price” column could be shown in the “Bidding Basket” table next to the “Bid” column. 10 The experimenter should also be able to set budget constraints and activity rules if desired. In the example given in Figure 1 no such constraints were imposed (see the top panel). 8 9 auctions. In contrast, the software must provide the experimenter with all of the options needed for the treatments discussed below. In particular, the setup pages should allow the experimenter to specify the auction format, the protocol for matching subjects in auction groups (random or fixed), the common and/ or private value distributions, the information feedback conditions, the enforcement of bidding budget and activity rule constraints, etc. Moreover, the software must provide for instructions and easy to understand earnings calculations at the end of each auction period. III. Proposed Treatments Before we discuss the treatments that are most relevant to test the SMR auction vis- à- vis the SMRPB auction, we first review the attributes of package bidding. We also discuss elements likely to affect both institutions. 1. Attributes of Package Bidding The principal reason to consider the SMRPB auction is to allow bidders to express their preference for a combination of licenses without risking financial exposure. The “exposure problem” arises when a bidder obtains part, but not all, of the preferred package but spends more for the obtained pieces than they are worth to him. For example, suppose a bidder values the package AB at 10 but has no value for either A or B separately. In an SMR auction with no package bidding, a bidder who attempts to acquire both licenses but is outbid for one will incur a financial loss. To avoid the loss, the bidder may decide not to bid aggressively on either item, resulting in an inefficient allocation when the total value of licenses A and B to others is less than 10. In contrast, in the SMRPB auction the bidder would be willing to bid up to 10 on the AB package, and this bidder will efficiently be assigned the package if the total value of licenses A and B to others is less than 10. Intuitively, the exposure problem of the SMR auction will become more severe as the degree of complementarity between the licenses rises and bidder license preference overlap increases. The ability to express preferences for packages may thus enhance efficiency and revenues. However, the SMRPB auction may create problems for “small bidders,” which we define here to be those bidders interested only in small packages or even a single license. This is known as the “threshold problem.” Suppose in the above example that there are two other small bidders who have values of 8 for license A or license B, but no value for the package AB. Furthermore, suppose the current prices in the SMRPB auction are 2 for A, 2 for B, and the package bidder is the provisional winner by bidding 9 for AB. The small bidders could overthrow the package bid by both increasing their bids to 5, but both would prefer a smaller increase in their own bid hoping the other will make up for the difference. In other words, the positive effect of an increase in one small bidder’s bid for other small bidders creates incentives for small bidders to “free ride,” i. e. to let others bear the cost of outbidding the package bidder. Moreover, small bidders face a coordination problem in that their combined total bid needs to rise to overthrow the package bid, but an increase that overshoots the minimum required amount is wasteful 9 10 from their point of view. (The situation the small bidders face is therefore akin to a step-level public goods game. 11 ) Intuitively, this coordination problem may get worse as more small bidders are needed to overthrow the package bid. Note that the threshold problem is a problem in the sense that free riding and/ or failure to coordinate bid increases may result in an inefficient allocation (i. e., the licenses included in the package are worth more to two or more non- winning bidders than to the winning bidder). In addition to the exposure and threshold issues which are likely to vary according to whether a package or non- package format is used, the winner’s curse and tacit collusion may also affect the outcomes of the alternative auction designs differentially. The winner’s curse effect is due to a failure of a bidder to recognize that winning is more likely when one’s common value signal is relatively high. In other words, even though the value signal is unbiased ex ante, there is an upward bias conditional on winning. The failure to realize that winning is an informative event may cause bidders to submit bids that are too high for given signals, and as a consequence, the winner may end up paying more than the license turns out to be worth to him. When there is a single common value that applies to all licenses, winner’s curse effects may be stronger for larger packages since the adjacency factor in equation (1.2) exaggerates any misjudgments that bidders may make regarding the level of the common value. Experimentation provides an ideal vehicle to test how these winner’s curse effects interact with the auction format. Tacitly collusive outcomes may result from coordinated attempts to divide the market at low prices. By tacit collusion, we mean mutual pricing restraint that is a result of a “meeting of the minds” without any explicit communication or bid agreements, which would have to be reported under current FCC rules. Intuitively, tacit collusion is more likely to be observed if each bidder has a local advantage for one license or, more generally, when bidders’ preferred packages show little overlap, and if complementarities are low. 12 The lower the complementarities, the smaller the economic value a cooperative bidder surrenders when agreeing to a collusive strategy that splits the market. In addition to the valuation environment, auction rules may have an important effect on the ability of bidders to identify a collusive outcome. For example, in order to divide up the market, bidders must signal to other bidders their desired collection of licenses. This is facilitated if bidders have information regarding the identities of the submitters of the provisionally winning and non- winning bids. Moreover, such information is needed in order for cooperative bidders to impose, in the form of a retaliatory bid, a cost upon bidders that behave non- cooperatively. Finally, the ability to make publicly announced bids on packages may facilitate bidders’ abilities to lay claims on non- overlapping 11 See, for instance, Isaac, R. M., D. Schmidt, and J. Walker (1989) “The Assurance Problem in a Laboratory Market,” Public Choice, 62( 3), 216- 236; Ledyard, J. O. (1995): "Public Goods: A Survey of Experimental Research," in A Handbook of Experimental Economics, ed. by A. Roth, and J. Kagel. Princeton: Princeton University Press, 111- 194. 12 There is some evidence that in an ascending auction without package bidding, collusive outcomes are less likely when there are strong complementarities between licenses. See Anthony Kwasnica and Katerina Sherstyuk, “Collusion and equilibrium selection in auctions”. Presented at the FCC, September 2004. 10 11 segments of the market. 13 An important auction design question is whether bidders can successfully divide up a market and whether changing the information environment and allowing for package bids would enhance their ability to do so. The suggested treatment structure in Appendix B is designed to focus on the effects of auction rules on tacit collusion, in environments where collusive outcomes are more likely to appear. Finally, both the SMR auction and the SMRPB auction described in Appendices C and D provide price feedback on individual licenses that may help bidders decide where and how to adjust their bids in the course of an auction. When the auction closes and no bidder wishes to further raise their bids at the current prices, the resulting set of final prices essentially corresponds to a competitive equilibrium in which all licenses are allocated. This process may be unstable when complementarities in valuations preclude the existence of a set of individual- license prices that clear the market. Experiments can show if in these circumstances the extra pricing flexibility provided by package bidding results in improved performance of the auction. Therefore, it is essential that the experimental environment include some cases where no competitive equilibrium exists. See Appendix A for a detailed discussion of the non- existence of a set of competitive equilibrium prices in the presence of license value complementarities (i. e., non-convexities) and the related concept of the core. 2. Experimental Treatments To evaluate all four effects (i. e., financial exposure, threshold problem, winner’s curse, and tacit collusion) we propose to allow for both common and private value elements, and for bidders of different sizes and license preferences (i. e., small bidders interested in a single license and large bidders interested in sets of licenses). Within this framework, we need to define the variables under experimental control. These are: (i) The degree of complementarities among licenses. (ii) The amount of overlap in bidders’ preferences. (iii) The “strength” of small bidders vis- à- vis large bidders (i. e., the difference between the sum of the valuations small bidders place on a set of licenses and the value the highest value large bidder places on those same licenses as a package). (iv) The amount of information bidders receive during the course of the auction regarding the identities of the provisionally winning and non- winning bids. We propose to include two cases for each variable: low/ high degree of complementarity, low/ high overlap in bidder preferences, “weak/ strong” small bidders, and full/ incomplete bidder information disclosure. The combinations of variables, or “treatments,” are categorized in the table below. Each would be tested in both the SMR and SMRPB formats. If desired, additional treatments, including, for example, varying degrees of 13 Of course, revealing information about all bids and bidder identities may have benefits, as well, such as making the auction more “transparent” and informing bidders about demand for the licenses on which they are bidding. 11 12 common- value uncertainty, could be considered. Also, depending upon results, it may be unnecessary to include all of the treatments listed. Treatments Degree of Complementarity Preference Overlap Small Bidder Strength Information Disclosure Low High Low High Weak Strong Full Partial LC, OL, SW, IF X X X X LC, OL, SS, IF X X X X LC, OH, SW, IF X X X X LC, OH, SS, IF X X X HC, OH, SS, IF X X X HC, OH, SW, IF X X X HC, OL, SS, IF X X X HC, OL, SW, IF X X X LC, OL, SW, IP X X X X LC, OL, SS, IP X X X X LC, OH, SW, IP X X X X Each of the eleven treatments will require six groups (six independent data points) for a total of 66 groups for the 2 basic auction formats, which results in a grand total of 2* 11* 6= 132 groups. Appendix B gives a specific proposal for the parameterizations of these sessions. IV. Experimental Procedures It is anticipated that adequate data collection will involve about 6- 12 subjects 14 for each of the 132 groups, and another 150 subjects for debugging and pilot studies, for a total between 942 and 1,734 subjects. These numbers are only an approximation, and may be negotiated with FCC staff as the project proceeds. The sessions will probably be somewhat long, perhaps 2 hours or more, due to the need to explain complex procedures and obtain enough replications. Financial motivation should be high enough to merit serious attention, and therefore, it is anticipated that earnings will be about $50 per person, including show- up fee. The auction rules should be based closely on the description of the two alternative auctions, contained in Appendix C, which was provided by FCC staff. The actual FCC auction framework will be done on- line, and therefore, the experiments will also be computerized. The FCC will provide the needed pricing and assignment algorithms (see Appendix D), but it is the responsibility of the experimenter to develop the software that would be used for the experiments (not for the actual FCC auction). The software developed by the FCC for the actual auction may be of limited use to run the experiments, although it may be useful to do some comparability testing to assure that the 14 Larger groups can result from scaling up the basic six- bidder design in Appendix B or from randomly reconfiguring a group of twelve subjects among two smaller groups of six. 12 13 experimental and FCC interfaces induce similar results. In particular, it is necessary to provide the experimenter with the flexibility to run alternative treatments, induce a desired array of private and common values, enforce budget and activity constraints, and allow for a variety of information feedback conditions. Furthermore, the software should provide subjects with an easy- to- understand interface for instructions, valuations and signals, bidding decisions and earnings. Of course, all information collected in the experiment (current and past bids, values, signals, earnings, etc.) should be recorded and reported to the FCC. In addition, it is desirable to report and control any prior experience that particular bidders have had in prior auction experiments, along with some demographic information. This information can be collected with a questionnaire that should be done ex post. The structure of the survey is up to the experimenter. V. Performance Measures Economic efficiency, the degree to which the auction assigns the licenses to the bidders with the highest valuations, is the most commonly used performance measure in economic experiments. Efficiency is measured as a ratio of the sum of the valuations winning bidders place on the obtained items to the maximum valuation placed by bidders under the efficient allocation of items, and is independent of prices paid. For example, suppose two bidders compete for a single license; one of them values the license at 10 while the other bidder values it at 6. Awarding the license to the lower- valuing bidder would result in an efficiency level of 6/ 10, or 60%. One of the main benefits of laboratory experiments is that bidders’ valuations are induced and known to the researcher, which makes it possible to examine efficiency and to compare average efficiency levels across different auction formats. The FCC may be interested in other measures as well. The experimental data should also include sales revenue comparisons between auctions run with and without package bidding. The reports should distinguish between revenue resulting from sales which were profitable to bidders, and revenue from sales in which bidders “overbid” due to exposure problems or a “winner’s curse.” Since the length of FCC auctions is important to actual FCC bidders, the experiments should report the number of rounds for each auction, and compare treatments. In addition, there should be some measure of the extent to which the time patterns of bidding activity vary. One simplistic measure would be in terms of cumulative numbers of bid changes, by round, which would give a picture of the extent to which bidding is concentrated towards the beginning or end of the auction. A more economically relevant measure would take bid amounts into account and would attempt to ascertain the extent to which bids begin to reveal final valuations during the auction process. To this end, the provisional winning bids in each round would be calculated, summed, and expressed as a fraction of the maximum- value allocation or as a fraction of the final sales revenue. 13 14 It may also be useful to try to evaluate the extent to which the various auction formats and treatments produce stable prices and outcomes. Loosely speaking, final auction prices can be thought of as market- clearing prices, since no new bids are being made. However, the final prices and allocation may not comprise a competitive equilibrium; indeed, a competitive equilibrium may not exist in some of the more complex fitting environments that these experiments are designed to test. In this context, it is important to consider whether or not such equilibria are possible, and if so, whether the final experimental allocations and prices correspond to competitive equilibrium outcomes. 15 Given that in theory there are situations in which competitive equilibria do not exist in non- package bidding auctions but are possible in package bidding auctions, 16 we are interested in the extent to which the auction format contributes to the potential for attaining a competitive equilibrium outcome. The researcher should note whether competitive equilibria exist, and whether the experimental outcomes correspond to those equilibria. A weaker measure of the stability and desirability of an outcome is to examine whether an experimental auction results in prices and allocations which are in the “core.” 17 The researcher should report the percentage of times that the final allocations are in the core. Of course, as with competitive equilibria, noise and other variations in bidding behavior are likely to result in inefficiencies which may preclude core outcomes, even if such outcomes are predicted in theory. However, differences which vary systematically by auction format are of interest. Finally, the FCC may be interested in assessing whether the success of small bidders in winning licenses, relative to the success of large bidders, varies according to auction format, taking into account valuations and budgets. One way to examine this effect is to consider the efficient allocation of licenses (with some provision for ties), and to calculate the sum of the license valuations that all small bidders obtain in the efficient allocation, V*. Then let V represent the sum of all prize valuations obtained by small bidders in the actual allocation that results from the auction. These valuations would include the complementary effects but would not include prices paid. Then the ratio, V/ V*, provides a license- based measure of bidding success, which can be compared across auction formats. This ratio, however, does not incorporate profit information based on prices paid. An alternative is to compare ratios of small bidders’ earnings to large bidders’ earnings across auction formats. These and possibly other measures may be used to help the FCC assess performance in this dimension. 15 See Appendix A. 16 See, for example, Bykowsky, Cull and Ledyard. “Mutually Destructive Bidding: The FCC Auction Design Problem,” Journal of Regulatory Economics, 17( 3), May 2000. 17 The FCC will provide an algorithm to determine whether or not a final allocation is in the Core. 14 20 Appendix C: Auction Rules The following describes the rules of the SMR and SMRPB auctions. 1. SMR Auction a. Simultaneity and Bid Structure All licenses are put up for bid simultaneously. Participants have the opportunity to submit bids on individual licenses, for as many licenses as they wish. b. Iterative The auction consists of discrete, successive rounds in which buyers have the opportunity to place bids on their desired licenses. Following each round, the high bid for each license is posted. These high bids then become the standing bids for the next round of bidding. c. Minimum Opening Bid The minimum opening bid is defined as the minimum price the Commission demands in exchange for selling the license. d. Minimum Acceptable Bid In the first round and until a bid has been placed on a license, an acceptable bid must be equal to or exceed the minimum opening bid. Subsequently, in order to be acceptable, a bid must exceed, by a specified percentage, the provisionally winning bid for the license. In order to move the auction along more quickly, the increment percentage can be calculated on a per- license basis, as an increasing function of the number of bids placed on the license in previous rounds. If speed of the auction is a concern in these experiments, we may want to use this feature – known as the exponential smoothing formula. e. Bid Increments Bidders are given the choice of making the minimum acceptable bid, or of making one of eight incrementally higher bids. f. Bid Withdrawal In a limited number of rounds (usually two), participants are permitted to withdraw any of their provisionally winning bids. After the withdrawal, the Commission becomes the provisionally winning bidder for the withdrawn license and the minimum acceptable bid in the following round equals the second highest bid received on the license (which may be less than or equal to, in the case of tied bids, the amount of the withdrawn bid). A withdrawing bidder pays a penalty equal to the maximum of zero or the difference between the price at which the bidder withdrew its bid and the final sale price in the current auction or in a subsequent auction in which the license is sold. If the license remains unsold in the current auction, the withdrawing bidder pays an interim payment of 3% of its withdrawn bid. 20 21 g. Bidding Eligibility and Activity Each license is assigned bidding units typically based on its bandwidth and the size of the population “covered” by the geographic area of the license. A bidder’s upfront payment determines the total number of bidding units a bidder can “bid on” at any one time. The total number of bidding units available to the bidder establishes the bidder’s maximum “eligibility” to bid in the auction. To encourage active bidding, the auction has rules that impose a cost upon bidders if they fail to display a minimum amount of bidding activity. In each round, bidders must be active on at least a fixed percentage of the total bidding units available to the bidder. High standing bids for licenses from the previous round and new bids in the current round are considered “active” bids for purposes of measuring bidding activity. The failure of a bidder to satisfy the activity rule leads to a reduction in his bidding eligibility sufficient to bring him into compliance with the rule (unless the bidder uses an activity rule waiver, as described below). Therefore, in subsequent rounds, the maximum number of bidding units on which the bidder may be active is below the original number. The bidder may also actively reduce his eligibility, if he does not wish to maintain the minimum bidding activity required by the activity rule. While a bidder’s eligibility determines the maximum number of bidding units on which a bidder can bid on and the activity rule defines the minimum level of bidding activity the bidder must exhibit in order to maintain its current bidding eligibility, the activity and eligibility rules do not restrict the amount of the bids. h. Activity Rule Waiver Each bidder is granted a limited number of opportunities (usually 3- 5) to avoid being subject to the activity rule. If the bidder does not meet his minimum activity requirement for a round, the system will automatically apply a waiver, if the bidder has one remaining, or the bidder may actively submit a waiver. An activity rule waiver applies to an entire round, not to particular licenses (i. e., by using an activity rule waiver, a bidder is essentially taking a “time out” for the round). i. Closing Rule The auction closes after the first round in which no new bids were placed, no bids were withdrawn, and no waivers were actively submitted. j. Information Environment Prior to entering the auction, bidders have information showing the number of bidding units associated with each license and the number of bidding units each bidder is able to “bid on.” Following the end of each round, participants receive information on provisionally winning bids, the identity of each high bidder, the name of each non- winning bidder and their bids, the identity of any bidders using activity waivers, the eligibility requirement for each bidder in the subsequent round, and the minimum acceptable bid for each license for the subsequent round. After each round, bidders also know the identity of the 21 22 bidders that have withdrawn bids and the license( s) on which they have withdrawn bids. k. Payment Default Rule If a winning bidder defaults on payment for the licenses he won, he is liable for a deficiency payment, equal to the difference between the amount of the bidder’s bid and the amount of the winning bid the next time the licenses are won in an auction, plus an additional payment equal to a percentage of the defaulter’s bid or of the subsequent winning bid, whichever is less. In instances in which the amount of a default payment cannot yet be determined, the Commission assesses an initial default deposit of a percentage of the defaulted bid amount. (If a bidder defaults, he must default on all of his winning bids.) 2. SMRPB Auction a. Simultaneity and Bid Structure All licenses are put up for bid simultaneously. Participants have the opportunity to submit sets of bids, each of which may be for one or more licenses. Multiple bids submitted by each bidder will have an “exclusive OR” (i. e., “XOR”) relationship. For example, bids of the form “AB XOR C” would be interpreted as the bidder’s desire to acquire A and B or C, but not A or B separately or the set A, B, and C. Implementing this structure requires that only one of a bidder’s bids can be included in the provisionally winning set of bids. b. Iterative The auction consists of discrete, successive rounds in which buyers have the opportunity to place bids on their desired licenses. Following the submission of bids, an algorithm identifies the provisionally winning assignment of licenses through revenue maximization of gross bids. In addition, an algorithm estimates a set of individual item prices (i. e., “current price estimates” (“ CPEs”)) that is consistent with this assignment. 22 The CPEs then become the basis for establishing minimum bids for the next round of bidding. c. Minimum Opening Bid The minimum opening bid is defined as the minimum price the Commission demands in exchange for selling the license. It serves as the minimum acceptable bid on a license in the first round and in subsequent rounds if no bids have been placed covering that license. For a package, the minimum opening bid is the sum of the minimum opening bids for its component licenses. 22 Broadly speaking, the price estimate of a license corresponds to the monetary cost of not awarding that license to the bidder to whom it is provisionally assigned. Appendix B contains a detailed description of the pricing algorithm. 22 23 d. Minimum Acceptable Bid In all rounds subsequent to the first round in which a bid is placed on a license or on a package containing the license, the minimum acceptable bid for a license must exceed the CPE by a specified percentage. This percentage is either a flat rate or a license- by- license percentage based on bidding activity. The minimum acceptable bid for a package of licenses is the sum of the minimum acceptable bids for its component licenses. e. Bid Increment Bids can also be made in one of eight incrementally higher bid amounts. The increment is based on a fixed percentage of the minimum acceptable bid amount for the license or package. f. Bidding Activity Each license is assigned bidding units typically based on its bandwidth and the size of the population “covered” by the geographic area of the license. A bidder’s upfront payment determines the total number of bidding units a bidder can “bid on” at any one time. The total number of bidding units available to the bidder establishes the bidder’s maximum “eligibility” to bid in the auction. To encourage active bidding, the auction has rules that impose a cost upon bidders if they fail to display a minimum amount of bidding activity. In each round, bidders must be active on at least a fixed percentage of the total bidding units available to the bidder. A bidder's activity in a round is equal to the number of bidding units in the bidder’s largest (in terms of bidding units) active bid. Active bids include current provisionally winning bids, new bids and any bids from previous rounds which are at or above the current minimum acceptable bid. While a bidder’s eligibility determines the maximum number of bidding units on which a bidder can be active and the activity rule requires a minimum level of activity, the activity and eligibility rules do not restrict which licenses the bidder can bid on or the amounts of the bids. The failure of a bidder to satisfy the activity rule leads to a reduction in his bidding eligibility sufficient to bring him into compliance with the rule (unless the bidder uses an activity rule waiver, as described below). Therefore, in subsequent rounds, the maximum number of bidding units on which he may be active is below the original number corresponding to his upfront payment. He may also actively reduce his eligibility, if he does not wish to maintain the minimum bidding activity required by the activity rule. g. Activity Rule Waiver Each bidder is granted a limited number of opportunities (usually 3- 5) to avoid being subject to the activity rule. If the bidder does not meet his minimum activity requirement for a round, the system will automatically apply a waiver, if the bidder has one remaining, or the bidder may actively submit a waiver. 23 24 h. Closing Rule The auction closes when no new bids have been submitted and no proactive waivers applied in one or two consecutive rounds. Whether the auction closes after one round or two consecutive rounds with no new bids and no proactive waivers will be established prior to the auction. i. Information Environment Prior to entering the auction, bidders have information showing the number of bidding units associated with each license and the number of bidding units each bidder has available to assign. Following the end of each round, participants receive information on the provisionally winning set of bids, the identity of each provisionally winning bidder, the name of each non- winning bidder and their bids, the identity of any bidders using activity waivers, the eligibility requirement for each bidder, and the minimum acceptable bid for the next round for each license and for each package that has already received a bid. In addition, bidders are given the per- license CPEs. j. Payment Default Rule If a bidder defaults on payment for a winning bid, he is liable for a deficiency payment, equal to the difference between the amount of the bidder’s bid and the amount of the winning bid the next time licenses covering the same spectrum are won in an auction, plus an additional payment equal to 25 percent of the defaulter’s bid or of the subsequent winning bid, whichever is less. In instances in which the amount of a default payment cannot yet be determined, the Commission assesses an initial default deposit of between 3 percent and 20 percent of the defaulted bid amount. 24