It is desirable to design auctions in such a way that participants have a financial incentive to provide true information about their interests. At first glance, an ideal description of this truthfulness requirement is that, in all possible situations, a person reporting the true valuation benefits gets the largest possible value of the difference v − p between the value v of the objects gained in the auction minus the corresponding payment p. However, such a universal truthfulness requirement is too strong: such auction schemes exist, but the resulting solutions are often of low quality.

A much better quality is attained by truthful-in-expectation solutions, in which reporting the true valuation maximizes the expected monetary gain m = v − p. However, this does not necessarily mean that in the real-life implementations of the corresponding auction designs, participants will report their true valuations. The reason is that while it may sound reasonable to maximize the expected value of monetary gain, this is not, in general, how people make decisions. The decisions of a rational person correspond to maximizing the expected utility, where utility u(m) is an increasing (and usually nonlinear) function of the monetary gain.

Different participants may make decisions based on different utility functions. From this viewpoint, it is desirable to look for auction schemes in which the truthful valuation is beneficial for all possible utility functions. This condition is difficult to check directly, since it requires considering all possible monotonic functions; however, it is known that such a condition can be equivalently reformulated as the requirement that F(a) is smaller than or equal to G(a) for all a, where F(a) is the truthful-valuation-case probability that the monetary m does not exceed a and G(a) is the similar probability for the case of a false valuation. This property is known as first-order stochastic dominance.

The authors describe all social-choice functions that are truthful in first-order stochastic dominance. Interestingly, it turns out that these are the same functions that allow truthful-in-expectation solutions--but to gain first-order stochastic dominance, one needs to somewhat change the payment rules. Another interesting result is that the situation does not improve if we only consider, for example, risk-averse participants: if we can guarantee truthfulness for them, then we can also guarantee truthfulness for all participants.

These interesting new results ensure that the corresponding auction designs encourage truthfulness and are, thus, ready for practical implementation.