We consider an environment in which traders with finite memory update their forecast rules at random intervals by OLS. In this context, overparameterization of the forecast rules can destabilize the learning dynamics. This instability tends to be attenuated by greater memory and less frequent...