Distributional Approach to Risk Preferences

We propose a distributional framework for eliciting risk preferences that treats an individual’s attitude toward risk as a full probability distribution rather than a point estimate. By parameterising preferences with the flexible beta family, our approach encompasses the entire spectrum from extreme risk aversion to risk neutrality and even risk-seeking behaviour, while simultaneously allowing for heterogeneous stability of those attitudes across contexts. Our agent-based simulations show that (i) the true underlying preference distribution is recoverable with negligible bias and (ii) the precision of recovery is a systematic function of elicitation design richness, providing clear guidance for experimental design. Benchmarking on the comprehensive laboratory dataset of Holzmeister Schmidt (2021) confirms two central results: (1) out-of-sample predictive accuracy is at least on par with canonical point-estimate methods, and (2) our method delivers a second, policy-relevant moment—the subject-specific variance of risk taking—without sacrificing parsimony.