A representation theorem for guilt aversion

• Provides axiomatic foundations for two utility models of guilt aversion: an existing linear utility and a novel logarithmic utility.
• Argues from psychological grounds that the assumptions implied by the logarithmic utility are more realistic.
• Applies both utility models to well-known laboratory games and shows that the logarithmic utility achieves sharp and realistic predictions.

Guilt aversion has been shown to play an important role in economic decision-making. In this paper, we take an axiomatic approach to guilt by deducing a utility representation from a list of easily interpretable assumptions on an agent's preferences. It turns out that our logarithmic representation can mitigate the problem of multiplicity of equilibria to which psychological games are prone. We apply the model in three well-known games and show that its predictions are consistent with experimental observations.