SEMINAR 7 May 2012: Guillaume Hollard, Centre d'Économie de la Sorbonne, Paris

Do Level-0 Players Really Exist? Evidence from Experimental Games Involving Step Reasoning

Abstract

Nash equilibrium often fails to predict what players actually do in one-shot laboratory games. Various models were designed to account for observed behavior, known as level-k or cognitive hierarchy models. A common feature to these models is that they are iterative. Level-0 players are defined has playing randomly and higher levels play a best-response against lower levels. While it is recognized that the way of modeling level-0 players is the key of these models, these players have not been studied per se. This paper thus focuses on level-0 players. We first use an individual criteria to identify a set of players whose strategy in a series of beauty contest game can hardly be rationalized. We thus provide several empirical facts regarding that subset of players. We find that these players indeed behave as level-0 players. The next step of our analysis is to provide a full set of controls aiming at understanding what drives the proportion of level-0 in experimental games (e.g. subjects did not pay attention, stakes were too low, etc). We conclude by a discussion on what drives the proportion of level-0 in games and suggest ways to enhance the explanatory power of current models.