In this paper we analyze forward-looking economic models under bounded rati
onality. We consider a learning mechanism characterized by exponentially fa
ding memory with a learning step not vanishing in the limit. The dynamics o
f the model under bounded rationality can be parametrized with respect to t
he memory of the learning mechanism. We show that memory plays a stabilizin
g role in a local sense: it induces local convergence towards a stationary
rational expectations equilibrium and in some cases it does not allow for n
onperfect foresight attractors obtained through local bifurcations. We disc
uss this learning mechanism in pure exchange overlapping generations models
. We also analyze models with predetermined state variables, in this settin
g the effect of memory on the learning dynamics is controversial. (C) 2000
Elsevier Science B.V. All rights reserved. JEL classification. D83; D84; E2
1; E32.