How effectively does a decentralized marketplace aggregate information that is dispersed throughout the economy? We study this question in a dynamic setting where sellers have private information that is correlated with an unobservable aggregate state. In any equilibrium, each seller's trading behavior provides an informative and conditionally independent signal about the aggregate state. We ask whether the state is revealed as the number of informed traders grows large. Surprisingly, the answer is no; we provide conditions under which information aggregation necessarily fails. In another region of the parameter space, aggregating and non-aggregating equilibria coexist. We solve for the optimal information policy of a social planner who observes trading behavior. We show that non-aggregating equilibria are always constrained inefficient. The optimal information policy Pareto improves upon the laissez-faire outcome by concealing information about trading volume when it is sufficiently high.