Learning, non-equilibrium beliefs, and non-pecuniary payoffs in an experimental game

We present a parametric learning model of players' dynamic and possibly out-of-equilibrium beliefs about other players' preferences that also incorporates random utility (noise). We estimate the model using the data from the four-country ultimatum game experiments of Roth et al. (1991). We find evidence that in the US and in Israel, the estimated beliefs of proposers are stationary and out-of-equilibrium, that in Slovenia, they are in equilibrium, and that in Japan, they are out-of-equilibrium, change from period to period and move away from equilibrium over time. In Japan and in the US, the estimated proposers' beliefs are further away from the uniform prior than the estimated equilibrium beliefs. The results seem to provide support for a non-pecuniary payoff explanation in all countries.