How do Multi-player Beauty Contest Games affect the Level of Reasoning in Subsequent Two Player games?

Behavioural economists have employed beauty contest games in order to test for agents’ depth of reasoning. In multi-player beauty contest games, fully rational players should all bid zero. In experiments, it is observed that average bids are consistently much greater than zero, which may reflect a population of players consisting of a combination of full and bounded rationality. In contrast, in two-player beauty contest games, bids above zero unambiguously reflect irrational play. In this paper, we conduct an experiment in which agents play several rounds of a multiplayer beauty contest game, but then switch to a two-player beauty contest game for the remaining rounds. We demonstrate that bids are above zero for the multi-player game and remain above zero for the switching rounds (we postulate that players seem initially to anchor on their behaviour in the multi-player game). However, players learn rapidly, with the average bid rapidly approaching zero towards the final rounds of the twoplayer game.