We consider the dynamics of learning under ambiguity when learning is costly and is chosen optimally. The setting is Ellsberg's two-urn thought experiment modified by allowing the agent to postpone her choice between bets so that she can learn about the composition of the ambiguous urn. Signals are modeled by a diffusion process whose drift is equal to the true bias of the ambiguous urn and they are observed at a constant cost per unit time. The resulting optimal stopping problem is solved and the effect of ambiguity on the extent of learning is determined. It is shown that rejection of learning opportunities can be optimal for an ambiguity averse agent even given a small cost.
↧