Designing the 700 MHz Auction
Pasadena, Calif.--It's been called beachfront property. Wireless companies are clamoring for pieces of it. The auction that will parcel it out will be the biggest of the next decade, with reserve prices set at $10 billion.
The property is in the sky, and it constitutes the most valuable communications spectrum that will hit the open market in the foreseeable future. It is the 700 megahertz (MHz) frequency, which until now has been the exclusive domain of broadcast television. On January 24, on the road toward ubiquitous digital television, the Federal Communications Commission (FCC) will auction off licenses to use swaths of this spectrum.
Potential newbies to the wireless grid, like Google and Cox Cable, will bid alongside entrenched companies like Verizon and AT&T and start-ups like Frontline Wireless. They'll all use a bidding system designed by California Institute of Technology economics professor Jacob Goeree and economics professor Charles Holt from the University of Virginia. The system was tested and refined through a series of laboratory experiments in which more than 200 Caltech undergraduates participated over the course of two years.
Right now, a few companies dominate the U.S. wireless market. This auction is the last chance for new entrants to create a national footprint. The 700 MHz frequency is particularly appealing for wireless--the signal can penetrate walls, and each tower broadcasting in this range can cover at least four times as many square miles as conventional cell-phone towers. This means fewer towers, at less expense, to any potential bidder. To cell-phone users, it may also mean goodbye to the roaming signal.
When the FCC approached Goeree and Holt in 2004, it asked them to test auctioning software the FCC had already built. The two rebuilt a downsized version of that software and commissioned Caltech undergrads to test it. The students came into the computer lab on weekends, placing bids in simulated auctions for hours at a time. They competed with each other so realistically that, says Goeree, their intelligent bidding mimicked professional auction behavior. "They even found bugs in the software because they're so smart," he says.
All of the bidding for the 700 MHz spectrum will take place online. This auction will run the same way that art or real estate auctions do--interested parties make their offers, and then compete in a bidding war until a winner is declared. But a communications-spectrum auction can quickly get complicated because of the volume of goods on the block and the number of parties with varying levels of interest.
To start, the FCC has divided the spectrum into several bands--A through E--that occupy different frequencies within the 700 MHz range, and divided each of these bands into several regions. Band A consists of 176 licenses for the frequencies between 698-704 and 728-734 MHz, for example, and band C, considered the most commercially attractive, has only twelve licenses and occupies the 746-757 and 776-787 MHz frequency ranges.
Each bidder could be interested in different bands and in different geographic regions; they don't necessarily want nationwide coverage. For example, Verizon may be particularly attracted to the C band, and they may only want it in Texas and New Mexico, where their coverage is spotty. T-Mobile may want to buy regions of the same band for coverage from North Dakota all the way south through Texas and east as far as Kansas. And Google, in a bid to establish a nationwide network, might want a package combining bands B and C through several regions to collectively cover the entire country.
The FCC wants all buyers to be able to compete equally and wants to make the most money in the process. But the program it had devised was too complex. "We tested the plane the FCC built and it didn't fly that well," says Goeree. The FCC's program was inefficient--it allowed too many potential combinations of bandwidths and geographic regions. It didn't maximize profits, or potential wins for the bidders. Most of all, it alienated bidders with its intricacies. Goeree and Holt tested related auction designs but these didn't fly well either and were still too complex. So they created a new method.
"We had a very simple idea for how to do it," says Goeree. In December 2006 they called the FCC and presented their approach, called Hierarchical Package Bidding (HPB). It groups the available licenses for all bands into packages according to a hierarchy with a fixed number of levels or tiers. For band C, for example, there could be three levels that comprise different bundles of regions. At the bottom level, 12 individual licenses would correspond to 12 different regions--Region 1 is the Northeast, for example; Region 4 is the Mississippi Valley; Region 12 is the Gulf of Mexico. On level two, the 12 regions could merge into three packages made of four individual regions each. At the top, it would be winner take all, meaning that if the highest bid was at level one, that bidder would take home the national package consisting of all 12 licenses for band C.
The HPB method gives small players a prayer at winning in the high-stakes game. In a three-level system, participants can opt to bid on any of the three levels simultaneously, though as the bidding war progresses they may find themselves squeezed into a single level. Although ultimately the FCC wants to turn the maximum profit, it also wants bidders who value a certain license or package the most to win it. Verizon, for example, can probably afford bidding for countrywide coverage, but they might not want it. They may just want one smaller license. Less wealthy bidders can also afford to compete at a local level, where individual licenses will be cheaper. Collectively, individually smaller bids at the local level could add up to more than what is bid at the nationwide level.
At the close of each bidding round, the software calculates the total money bid at each level. In a three-level system, say the bids at level three total $12 billion, at level two they total $11 billion, and the level one bid is $12.4 billion. The software would then advise bidders on their next move should they want to stay in the game. Each of the 12 bidders on level three would be alerted to increase their bid by about $34 million. The four bidders at level two would need to fork over more than $350 million each to stay in the game. Level one could sit tight, until the next round. There would be no need to calculate; the bidders would just make sure they could afford the suggested bid. "It solves the complexity for them," says Goeree. It also means that if each bidder at level two or three follows the advice in unison, they can all move on to the next round. Of course, whoever can't follow the suggestion will get shut out.
In an October 2007 public notice, the FCC declared, "The HPB auction format was chosen in part because it mitigates issues inherent in some other package bidding formats that give bidders interested in large packages an advantage over bidders interested in individual licenses."
The FCC chose a two-level hierarchy for the upcoming auction. On level two, 12 individual licenses will be available. Level one is more complex and consists of three packages: a 50-state package will constitute eight of the 12 licenses, another two of the 12 licenses will be made of other U.S. Pacific territories, and an Atlantic package will combine the U.S. Virgin Islands, the Gulf of Mexico, and Puerto Rico for the final two licenses. Testing by Goeree and Holt showed that even a simple two-tiered format performed dramatically better than the FCC's previous setup.
Goeree's main concern about his hierarchical system was that making packages of licenses without information about bidder preferences would fail to meet bidder interests. The HPB offers prepackaged units, but companies might only be interested in an intermediate choice that matched their needs more closely. They might therefore refuse to bid on Goeree's setup. "It turned out I was wrong. The HPB auction actually performed better," Goeree says. He found that allowing his volunteer testers to build their own packages resulted in overlapping regions and too much extra complexity.
"In fact, we will use HPB in part because the mechanism for calculating [prices] is significantly simpler than other package bidding pricing mechanisms," the FCC reported in October. "In addition, we find that . . . HPB procedures in general strike a careful balance between permitting bidders adequate bidding flexibility and discouraging insincere and anticompetitive bidding behavior."
Given the stakes and the number of licenses, the auction will likely last for several weeks. There will be several bidding opportunities per day, but just like in art auctions, the bidding ends only when the money runs out.