Hypothesis Testing and Ambiguity Aversion

We study a model of non-Bayesian updating for ambiguity averse agents, based on the Hypothesis Testing model of Ortoleva (2012). Agents have a set of priors that they update as follows. If all priors assign to new information a probability above a threshold, they update every prior using Bayes rule. Otherwise: they look at a prior over sets of priors, update it, and choose the set to which the prior over sets of priors assigns the highest likelihood. When the threshold is zero this coincides with Bayesian updating when defined, but it also prescribes behavior when it is not defined.