How Does an Auction Work? Four Formats and Why the Winner Often Loses

Almost every real auction is one of four types or a variant, and they differ in what information is revealed and when:

• •The English or ascending auction, familiar from the saleroom, where bids rise openly until one bidder remains, which reveals information as it proceeds and lets bidders learn from each other
• •The Dutch or descending auction, where the price starts high and falls until someone accepts, used for flowers in the Netherlands and for fish, and valued for speed since one decision ends it
• •The first-price sealed bid, where each bidder submits one number privately and the highest wins and pays their bid, which is standard in construction procurement
• •The second-price sealed bid, or Vickrey auction, where the highest bidder wins and pays the second highest bid, which has the elegant property that bidding your true value is optimal regardless of what anyone else does
• •Variants include the all-pay auction, where everyone pays regardless of winning, which models lobbying, litigation and research races, and the combinatorial auction, where bidders bid on packages of items rather than individually

William Vickrey's analysis in 1961, for which he shared a Nobel Prize in 1996, produced a counterintuitive theorem: under a set of specific conditions, all four standard formats yield the same expected revenue for the seller. The conditions are that bidders are risk-neutral, that each knows their own value privately and independently of others, that the highest bidder always wins and that a bidder valuing the item at zero pays nothing. The intuition is that bidders adjust their behaviour to the rules, so that in a first-price auction they shade their bids below their true value by exactly enough to offset the difference from a second-price format. The theorem matters mostly because of what happens when its conditions fail, since it turns the question from which format is better into which assumption is broken in this particular case, and the answer determines the design. When values are interdependent rather than private, when bidders are risk-averse, when they can collude or when budgets are constrained, the formats diverge sharply.

When bidders are estimating a common value rather than expressing a personal taste, as when oil companies bid for a drilling tract whose contents are the same whoever wins, a systematic problem arises. Each bidder makes an estimate with some error, and the bidder who wins is the one whose estimate was highest, which means the winner is disproportionately likely to be the one who overestimated most. Winning is therefore bad news about your own estimate. The effect was identified by engineers at Atlantic Richfield in 1971 examining why offshore oil leases had produced disappointing returns across the industry, and it has since been documented in construction contracts, in corporate acquisitions, in book advances and in sports transfers. The rational correction is to shade bids downward by an amount that increases with the number of competitors, since more bidders means the winning estimate is further out on the tail, and the persistent empirical finding is that experienced professionals in real markets do this insufficiently.

The practical design problems are about behaviour the simple models leave out. Collusion is the largest, since bidders who can coordinate will divide the lots and keep prices down, which has been prosecuted repeatedly in construction, and open ascending formats make it easier because bidders can signal to each other through their bids, including by using the trailing digits of a bid to indicate which lot they want, a tactic documented in an American spectrum auction. Entry matters enormously, because a weak bidder who expects to lose will not participate and the auction becomes uncompetitive, so good designs actively encourage entry. The spectrum auctions designed from 1994 introduced the simultaneous multiple round format, in which many licences are auctioned in parallel over many rounds, letting bidders assemble the combinations they need, and later designs added combinatorial bidding on packages. Where the design has been done badly the results have been visible, with several European third-generation spectrum auctions in 2000 raising enormous sums that contributed to a subsequent telecoms collapse, and others raising almost nothing through weak competition.

An auction extracts information about willingness to pay that no buyer would volunteer, and the four standard formats differ in what is revealed and when, with the second-price sealed bid making honest bidding optimal. Vickrey showed that all four yield the same expected revenue under specific assumptions, which makes the design question one of identifying which assumption fails. Where bidders estimate a common value, the winner is systematically the one who overestimated most, and real designs must handle collusion and encourage weaker bidders to enter at all.