When correlation does not matter : Robust Bayesian estimates of risk preferences

Georgalos, Konstantinos (2026) When correlation does not matter : Robust Bayesian estimates of risk preferences. Economics Letters, 264: 112957. ISSN 0165-1765

Abstract

We use experimental data on risk preferences to estimate structural models of choice with Hierarchical Bayesian methods. We assess whether ignoring ex-ante parameter interdependence affects individual-level inference. Using both experimental and simulated data, we find that assuming prior independence yields posterior distributions virtually identical to those from a correlated hierarchical structure. This result holds across elicitation methods and degrees of correlation, suggesting that modelling cross-parameter covariance may often be unnecessary. Our findings support the view that, in well-identified designs, data dominate the prior, simplifying the practical implementation of Bayesian structural estimation.