Verifiable Uncertainty (Revision 2 )

Real-world uncertainties can be difficult to classify into the neat boxes of risk and ambiguity. The authors suggest verifiability as a practical test for risk, and verifiable uncertainty — often in the form of mechanically implemented equally likely prizes — as a real-world representation of risk. The authors propose a set of rules by which an agent contrasts a general (unverifiable) event against the verifiable m out of n chances of winning a prize. Collectively the rules advocate a decision procedure in the spirit of the smooth model of ambiguity, where the agent evaluates verifiably equally likely prizes — an object they call classical lottery — by average (verifiable) utility and, for general prospect, combines various scenarios of how unverifiable uncertainty might unfold. The agent can hold distinct attitudes toward verifiable uncertainty and unverifiable uncertainty. In particular, combining scenarios in a conservative manner reflects aversion to unverifiable uncertainty. The authors illustrate verifiability as a useful concept for both normative and descriptive decision making.