REDRUM: Robust Estimation and Design for the Random Utility Model

Non-monotonicity of utility differences in CRRA-based Random Utility Models has been taken to suggest possible distortions in the estimation of risk preferences. We show that this concern is real only under restrictive assumptions on the stochastic component of the model. Under common scaling, changes in utility differences are absorbed by the precision parameter when this is estimated jointly with risk aversion, so that the likelihood in the risk aversion parameter is invariant. We then show that when different subsets of tasks are represented at different effective scales, distortions may arise if a common error scale is imposed, whereas allowing for parsimonious heteroskedasticity restores stable estimation. Simulations based on standard lottery tasks illustrate these results and clarify that the empirical relevance of non-monotonicity depends not on scaling per se, but on how the stochastic component is specified.