Testing Ambiguity Theories: Revisiting Ellsberg's Paradox in a New Experimental Design

To solve the Ellsberg (1961) paradox, prominent models like MEU and KMM start with viewing ambiguity as representing second-order risk, due to the range of admissible first-order lotteries associated with an uncertain act, and explain ambiguity aversion as paying some sort of second-order risk premium. In our design, by playing Ellsberg’s two-color problem twice with replacement, where a different color wins each time, the admissible lotteries are restricted into a class of mean-preserving spreads where the 50-50 risky urn exhibits the highest variance. We show that MEU and KMM share with Savage’s SEU the same predictions, that risk-averse (-seeking) DMs shall choose the ambiguous (50-50 risk) prospect. Yet, we observe that substantial numbers of subjects violate these predictions. It appears that the ambiguity premium is partially paid to avoid the ambiguity issue per se, which seems to be the implicit driving force behind the experimental studies on source dependence in consistency with the CEU model.