Least squares learning and business cycles

This paper investigates whether the neoclassical growth framework augmented with least squares estimated heuristic rules may reproduce U.S. business cycles. I consider various assumptions about the length of the information set, the influence of contemporaneous data on current forecasts, and the limit case in which learning is completed. Calibrated to the U.S. economy, this model may generate endogenous business cycles that do not exist under perfect foresight. If random productivity shocks are introduced, then the model is more volatile than under rational expectations or constant gain learning and reproduces some key U.S. business cycles stylized facts.