An experimental test of the predictive power of dynamic ambiguity models

In this paper we report results from an economic experiment where we investigate the predictive performance of dynamic ambiguity models in the gains domain. Representing ambiguity with the aid of a transparent and non-manipulable device (a Bingo Blower) and using two-stage allocation questions, we gather data that allow us to estimate particular parametric forms of the various functionals and compare their relative performance in terms of out-of-sample fit. Our data show that a dynamic specification of Prospect Theory has the best predictive capacity, closely followed by Choquet Expected Utility, while multiple-prior theories can predict choice only for a very restricted subset of our subjects.