In macroeconomic models with dispersed information, agents have an incentive to learn from endogenous variables, requiring them to forecast the forecasts of others. This paper revisits the model of Townsend (1983) to characterize how this mechanism affects the equilibrium dynamics. The first part of the paper simplifies, revises, and extends past results about situations when prices are fully revealing. The second part explains that full revelation does not occur in the original model and proves that the equilibrium state vector is infinite-dimensional. It also provides a new numerical solution procedure for such cases, which operates entirely in the frequency domain.