Learning to predict rationally when beliefs are heterogeneous

This paper develops an adaptive learning scheme for an asset market with heterogeneous traders who are characterized by linear mean-variance preferences. We introduce the concept of a perfect forecasting rule which generates rational expectations for fundamentalists in the presence of investors with possibly non-rational beliefs. When these non-rational investors base their decisions on simple technical trading rules, perfect forecasting rules are approximated by successively estimating the excess demand function from historical data. Conditions are given under which trajectories generated by this adaptive learning scheme converge to trajectories with rational expectations for fundamentalists. (C) 2003 Elsevier B.V. All rights reserved.