In the 'chicken' game, (stop,go) and (go, stop) are the two pure strategy Nash equilibrium profiles. If the game is played once, does it make more sense to use the rationalizability solution concept? The way I think of it is, that if we can't correctly predict the rival's action, then we might end up not coordinating. Thus, although Nash equilibrium gives a sharper prediction, rationalizability is more realistic to use, given that crashes do happen!
Any thoughts?
Rationalizability is a solution concept that is robust to small misspecifications about the information players have, as well as the payoffs of their actions. I do think it provides reasonable predictions in situations where Nash equilibrium requires coordination or precise beliefs about other players actions or beliefs, especially when players have not had enough opportunity to learn the game. This is a long way to say that I agree with you regarding the chicken game. Though I believe that the most common outcome is that all players decide to wait longer than necessary, that's the main coordination problem that signs, lights, and rules solve, I believe.
