We study the asset pricing implications of learning in an environment in which the true model of the world is a multivariate one, but agents update only over the class of simple univariate models. Thus, if a particular simple model does a poor job of forecasting over a period of time, it is...