Utility elicitation and maximization from uncertain certainty equivalents

We extend the literature on utility elicitation using an optimization approach taking as input certainty equivalents of the lotteries offering random payoffs. Unlike other known elicitation methods, ours is non-parametric and is always feasible even if reported certainty equivalents are erroneous. Suitable for all risk-averse individuals, the output of our elicitation method is an (approximate) utility function that is increasing, concave and piece-wise linear. We then demonstrate how our approximate utility functions lend themselves to subsequent data-driven expected utility maximization in both ambiguity-neutral and ambiguity-averse setups. Our findings are not only theoretical, but they are also supported by various experiments based on both synthetic and real-world data.