LEAST MEAN SQUARES LEARNING IN SELF-REFERENTIAL LINEAR STOCHASTIC-MODELS

We analyze Self-Referential Linear Stochastic models under bounded rat ionality assuming that agents update their beliefs by means of the Lea st Mean Squares algorithm. This learning mechanism is less complex tha n Recursive Ordinary Least Squares learning and appears to be more pla usible as a learning device for economic agents. We prove convergence of the learning mechanism, the convergence conditions are different fr om those required by Recursive Ordinary Least Squares learning. (C) 19 97 Elsevier Science S.A.