This paper compares the most significant expectational stability criteria that have been used to assess the plausibility of perfect foresight trajectories in forward-looking dynamical systems: determinacy of trajectories, absence of neighbour sunspot trajectories, and convergence of " evolutive " and " eductive " learning processes. It examines, within a set of increasingly complex dynamical models, the robustness of an equivalence principle suggested by the analysis of the simplest classes of those models.
