Reference points as information

We develop a model of reference dependence based on the idea that reference points provide comparative information to people who are uncertain how to value objects or actions. When people are uncertain how much they value an option, they may still be able to tell whether it is better or worse than a reference point on each dimension, thus providing information. In a Bayesian framework, value uncertainty and comparative information generate regularities such as thinking in gain-loss categories; income targeting and bunching; a preference for advantages over tradeoffs; avoiding losses; coherent arbitrariness; behavioral attenuation; contrast, assimilation, decoy, and range effects; joint-vs.-separate evaluation differences; simultaneous overreaction to small and underreaction to large changes; and - when allowing for caution regarding value uncertainty - the endowment effect. The model predicts that reference effects are most pronounced when value uncertainty is high and the reference point well-known. We then develop a theory of endogenous reference points based on which option is most informative for choice by reducing value uncertainty. This yields several novel predictions, including when the reference point is determined by the status quo, expected outcomes, similarity-based memory, or other options.