Monday, February 23, 2009

Market design from the other side of the fence

Market design is both about designing markets (design as a verb) and about understanding the design of existing markets (design as a noun). This latter activity is the way game theorists originally got into the business; by looking at the existing 'rules of the game,' and figuring out how they worked, and could be made to work.

I'm reminded of this by two working papers on auctions, one focused on offering consulting advice to a bidder, and the other on criminal collusion among bidders.

The consulting story is by the A-team of spectrum auction consultants, and recounts some of their recent experience giving bidding advice: Winning Play in Spectrum Auctions by Jeremy Bulow, Jonathan Levin and Paul Milgrom. They focus on the information flows in multi-round auctions, and how the amounts bid in early rounds can help forecast final prices, which is an aid in deciding on which lots to bid:

"In major spectrum auctions, even large corporations need to raise or put aside money in advance to finance their spectrum purchases. Many of these companies also have a broad set of target licenses. If these licenses are substitutes and the budget constraint is binding, the bidder's optimal purchase will involve spending its whole budget or nearly so. Of course not every bidder falls into this category. For bidders with tight budgets and narrow interests, or for entrants with all-or-nothing goals, rising prices could lead them to spend zero once the prices of target licenses rise too high.

"If bidders in the first category account for enough of the money in the auction, a previously unexplored pattern becomes identifiable in the data. Define a bidder's exposure to be the sum of all of its bids in a given round, including its standing high bids from the prior round and all of its new bids in the current round, whether provisionally winning or not. This is the largest amount that a bidder might have to pay if all of its bids were to become winning. If a bidder faces a binding budget constraint and has broad interests, then as prices increase from round to round, its total exposure will eventually level off at an amount approximating its budget. If all bidders were to fall in this category, then the total exposure of all bidders in the auction would rise to the level of the aggregate bidder budgets and level off, forecasting the final auction prices. As prices rise, bidders will narrow the set of licenses on which they bid, the identities of the provisionally winning bidders on various licenses will change, and total winning bids will continue to rise, but final total winning bids will be forecast early and well by total exposure."

The account of criminal collusion is John Asker's A Study of the Internal Organisation of a Bidding Cartel, which tells of a long lived cartel of stamp dealers who agreed in advance which of their members would bid on each lot that they were collectively interested in. (He obtained the data from the prosecution records of the Antitrust Bureau of the New York Attorney General’s Department.) They coordinated among themselves by first holding a "knockout auction" (of roughly the kind that Graham, Marshall, and Richard 1990 described among bidders in New Jersey machine tool auctions), to determine which of the cartel members would be allocated the right to bid on each lot, and what sidepayments would be made following a successful bid. The detailed data allow Asker to estimate the costs that the cartel imposed on sellers and on other bidders who were not members of the cartel.

No comments: