Adaptive learning of rational expectations using neural networks (Q1978600)

This paper investigates how adaptive learning of rational expectations may be modeled with the help of neural networks. Necessary conditions for the convergence of the learning process towards (approximate) rational expectations are derived using a simple nonlinear cobweb model. The results obtained are similar to results obtained within the framework of linear models using recursive least squares learning procedures. In general, however, convergence of a learning process based on a neural network may imply that the resulting expectations are not even local minimizers of the mean-squared prediction error.