Dynamic Information Aggregation: Learning from the Past

In an environment with dispersed information, how much can agents learn from past endogenous aggregate outcomes such as prices or output? We show that, in a rational expectations equilibrium, two possible regimes can arise endogenously: a perfect revealing regime and a confounding regime. The economic fundamental can be perfectly inferred in the former but is only partially revealed in the latter. The confounding regime arises only when general equilibrium feedback effects are strong enough. Then, even when the past aggregate outcomes are perfectly observed, the effects of informational frictions are persistent over time. Furthermore, in the confounding regime, the aggregate outcomes do not permit a finite-state representation, and they display an initial under-reaction relative to their perfect information counterparts followed by a delayed overreaction. In a standard New Keynesian model, we show that the confounding regime is more likely to arise under a dovish monetary policy rule.