The seller of a used car knows how it has been driven and the buyer does not. The owners of a firm know more about its prospects than an acquirer does, and an applicant for insurance knows more about the risk than the insurer. In each case the party with less information must ask what the willingness of the other side to trade reveals.
The game tests whether buyers take account of adverse selection, that is, of the fact that an offer is accepted only by sellers for whom it is a good deal. It shows how hidden information can shrink a market or close it, although every trade would benefit both sides. It also shows a failure of reasoning that is common outside the classroom, in which a buyer values what is for sale at its average quality and ignores what acceptance implies.
Akerlof (1970) showed that when sellers know quality and buyers do not, the price reflects average quality, owners of good cars withdraw, average quality falls, and the process may continue until only the worst cars are traded or none at all. The default setting is the form of the problem introduced by Samuelson and Bazerman (1985) as the acquisition of a company. Quality is equally likely to be any number from 0 to 100, the car is worth its quality to the seller and one and a half times its quality to the buyer, and the seller accepts any price that is at least the quality.
A price is then accepted only for cars of lower quality than the price, whose average quality is half the price and whose average worth to the buyer is three quarters of the price. Every positive price loses money on average, and the best offer is zero. Trade collapses although the buyer values every car more than the seller does. If the buyer's valuation is more than twice the quality, or if the lowest quality is well above zero, some trade survives, and both cases can be set on the dashboard. Eyster and Rabin (2005) model the error as a general tendency to neglect the link between what other people do and what they know.
Most buyers nevertheless make substantial offers. In the experiments of Samuelson and Bazerman (1985), more than nine in ten participants made a positive offer, and a majority offered between 50 and 75. The reasoning commonly given is that the average car is worth 50 to the seller and 75 to the buyer, so that a price between the two is fair to both.
Experience helps little, as Ball, Bazerman and Carroll (1991) found when they gave participants twenty rounds with feedback and payment and only five of 69 learned to offer a dollar or less. Charness and Levin (2009) show that the error persists when the seller is replaced by a computer and the problem becomes an individual decision. Their result locates the difficulty in reasoning about a contingency and not in beliefs about other people. For a market with sellers who choose quality, Holt and Sherman (1999) describe a classroom exercise in which quality and price fall once buyers can no longer see quality before they buy.