Robustness of Level-k Reasoning in Modifled Beauty Contest Games

This paper studies how the predictive power of level-k models varies as we perturb the standard setting of beauty contest along two dimensions: the strength of the coordination motive and the information symmetry. We use Morris and Shin (2002) model as the unifled framework for our study, and show that the predictive power of the level-k models varies considerably along these two dimensions. Speciflcally, level-k models are successful in predicting the subject behavior when the information is symmetric and the coordination motive is the strongest. When we decrease the strength of the coordination motive or introduce private information, the predictive power of level-k models signiflcantly decreases.