This paper analyzes the learning behavior of a risk-averse agent. We f
ind two conflicting effects in the experimental behavior: a stronger p
reference for the ex post reduction in uncertainty, but ex ante the re
turns to information are more uncertain.