Ghostly Preferences: When Coherence Requires a Strict Preference Among Equivalent Random Variables

Presenter: Teddy Seidenfeld Abstract: When utilities for outcomes are bounded, or when all variables are simple, it is required by (canoncial) expected utility to have preferences defined over probability distributions or lotteries over outcomes. That is, under such circumstances two random variables with a common probability distribution over outcomes – equivalent random variables – must occupy the same place in an agent’s preference ordering over lotteries. For example, with outcomes determined by a flip of a fair coin (i.e, where the probability of landing heads = the probability of landing tails = 1/2), winning $1 if heads and losing $1 if tails is equally preferred to losing $1 if heads and winning $1 if tails. Also, then one is indifferent between trading one of these lotteries for the other.