The winner's curse

In the 1960s the United States government auctioned drilling rights on the Outer Continental Shelf by sealed bid. Companies surveyed a block of seabed, formed a private estimate of the oil beneath it, and submitted a number. The highest number won. At Atlantic Richfield, one sale went badly in an unusual way: the company won every block it bid on, and paying for all of them strained its budget for two years afterwards. Three of its engineers, Edward Capen, Robert Clapp and William Campbell, were asked to work out what had gone wrong.

Their answer, published in the Journal of Petroleum Technology in 1971, was that the problem sat in the structure of the auction rather than in anyone's geology. Ten companies survey the same block. The oil beneath it is worth roughly the same to any of them. None of them can measure it exactly, so each produces an estimate, and those estimates scatter around the true value. Some come in high, some low. The auction then selects, mechanically, the company whose estimate was highest. Not the most accurate estimate. The highest one. If the estimates were honest and simply imprecise, the highest is very likely to be an overestimate, and the company that submits it pays accordingly. Capen and his colleagues called the result the winner's curse.

The mechanism needs three conditions to operate, and they are worth stating because they are what limits it. The thing being auctioned must be worth about the same to everybody, which is what economists call a common value. The bidders must be forming their estimates independently rather than copying each other. And they must not be able to see what the others think before they commit. Where those conditions fail, the effect fails with them. In an auction where the item is genuinely worth more to one bidder than another, a high bid is a preference rather than an error.

What is not settled is how often this happens outside a controlled setting. In laboratory auctions the effect is easy to produce: inexperienced bidders reliably overpay, and John Kagel and Dan Levin spent decades documenting it. The same work found that experienced bidders learn to shade their numbers down and largely stop falling for it. Field evidence is much weaker. Kagel and Levin say directly that claims of a winner's curse are exceedingly difficult to support with field data, because the records are unreliable and other explanations for overbidding are usually available. Studies of the same offshore lease sales disagree with each other depending on which discount rate and oil price assumptions they use. Hendricks, Porter and Boudreau found average profits went negative once seven or more companies bid on a block, which supports the idea, while other assumptions applied to the same data show the companies profiting anyway.

Richard Thaler drew a distinction that survives the argument. In the strong version, the winner pays more than the thing is worth and loses money. In the milder version, the winner still makes a profit, just less than they expected when they bid. The milder version is much harder to disprove and much easier to recognise from the inside.

Most people meet this from the other side of the table, and the shape is the same. You ask five suppliers to quote on a job. The quotes scatter, because each firm is guessing at scope, at how long the work will take, at what will go wrong. You take the lowest one. That number came from whichever firm was most optimistic about the work, and optimism about work is how projects run over. The overrun, the quality drop, the corner cut halfway through to survive the price they gave you: none of that is bad luck arriving afterwards. It was selected for at the moment you chose.

The same structure sits under acquisitions, under bidding for contracts, and under any hiring process that ends with several offers on the table. Wherever several people independently value the same thing and the extreme value wins, the winner is standing at the end of a distribution rather than at the answer.

The defence is one number, written down before anyone else's arrives. What is this worth if nobody else is interested? That figure is the only estimate you will produce that is not contaminated by the competition. After that, knowing other people are bidding is itself information, and it points one way: your number is more likely to be high than low, so it should come down before you submit it. On the buying side the same knowledge points at the cheapest quote, which is the one most likely to have missed something.

Where this stops applying is worth holding onto, because it is the difference between discipline and timidity. If you genuinely know something the other bidders do not, your high number is your advantage rather than your error. The company with better survey data should bid more and should win. So the question to ask of your own number is whether you are seeing something they cannot, or whether you simply want it more.

Source: Capen, Clapp and Campbell, Competitive Bidding in High-Risk Situations, Journal of Petroleum Technology, volume 23, 1971, pages 641 to 653, DOI 10.2118/2993-PA. The laboratory evidence and the limits of the field evidence: Kagel and Levin, The Winner's Curse and Public Information in Common Value Auctions, American Economic Review, volume 76, 1986, pages 894 to 920, and their book Common Value Auctions and the Winner's Curse, Princeton University Press, 2002. Thaler's two versions of the curse are in Anomalies: The Winner's Curse, Journal of Economic Perspectives, 1988.

The Winner's Curse

Richard Thaler, 1992

Thirteen anomalies, each one a place where people reliably do something economic theory says they will not. The auction chapter is the origin of the phrase, and the rest of the book is the same move applied to lotteries, sports tickets and savings decisions.