Epistemic foundation of the backward induction paradox

Can we learn about the future from observing past decisions? The present thesis revolves around this question by addressing the backward induction paradox. This paradox concerns the phenomenon that past deviations from the backward induction strategy profile by one opponent might make a player deviate from his backward induction strategy if he believes the opponent will deviate from his backward induction strategy profile in the future again. In particular, this thesis aims at giving an epistemic foundation for the described paradox. In order to do so, the Dekel-Fudenberg procedure is adapted by imposing stochastical independence between the beliefs that one player holds about the behavior of his opponents. This novel procedure, named the Independent Dekel-Fudenberg procedure, and the epistemic conditions which lead to it, provide a foundation for the backward induction paradox, as it captures that in games where some player(s) play repeatedly, other outcomes than the backward induction one can be supported, while in games where every player only plays once, the only outcome which results from the Independent Dekel-Fudenberg procedure is the backward induction outcome