Memorandum, Department of Economics, University of Oslo 2000,38
The concepts of sequential and quasi-perfect rationalizability are deﬁned in an epistemic model by means of lexicographic probabilities. These are non-equilibrium analogs to sequential and quasi-perfect equilibrium, for which epistemic characterizations are provided. The deﬁned rationalizability concepts are shown to imply backward induction in generic perfect information games, but they do not yield forward induction. The relationship between various concepts are shown and illustrated.