Increasingly, theoretical papers explore the implications of belief distortions, while empir-
ical papers use belief distortions to explain systematic patterns observed in the data. Gaining
a deeper descriptive understanding of exactly how people process information and form beliefs
is clearly important. In this paper we conduct a laboratory experiment which allows us to
explore how subjects update their beliefs in relation to Bayes rule. In particular, the design
provides a robust test for the asymmetric updating hypothesis, where individuals overweight
positive information relative to negative information, in the realm of monetary outcomes. In
contrast to earlier papers in this literature, which examine updating over attributes relating
to one's self-image, we find no evidence of this asymmetry in updating over states associated
with monetary outcomes. On aggregate we find that participants in our experiment are sur-
prisingly close to Bayesian, however using a finite mixture model analysis, we show that this
aggregate behaviour masks the existence of three distinct types of updating behaviour - each
of which is very distinct from Bayesian, none of which exhibits asymmetric updating.