Herding through learning in an asset pricing model

In this paper we show how uncertainty and learning can lead to a disconnection between fundamental values and prices in a simple asset pricing model. Agents use prices, besides an idiosyncratic exogenous signal, to infer fundamental values: as agents accumulate information, they put increasing weight on the public signal and in the limit they ignore completely their private information. The Bayesian equilibrium implies that agents end up relying only on prices in their signal extraction problem, an outcome that reminds the rational herding result in sequential decision making. We also consider two extensions that should mitigate this e¤ect, namely constant gain adaptive learning and Bayesian learning with an explicit probability of change in the fundamental. In both cases the problem persists, though somewhat mitigated. As a by-product, we also establish a connection between the constant gain parameter in adaptive learning and the subjective probability of exogenous changes in Bayesian learning.