Observational Learning and Firm Dynamics

Canonical theories of observational learning (Bikhchandani et al. (1992), Banerjee (1992)) have been used to study information cascades (“herding”) in a wide variety of markets. This paper embeds observational learning in an equilibrium model of industry to investigate its implications for firm dynamics. In the model, consumers learn about firms’ unobserved product quality over time. Because consumers learn through purchase decisions, price setting is a crucial channel through which firms manipulate future demand. We first characterize optimal pricing in a simplified model to show when and how learning is constrained by firms. We then argue that the fragile nature of cascades, a feature of the theory, is reflected in firm level data. We capture fragility by measuring what we term reversals : periods when a firm with historically stable revenues experiences a large, sudden change in earnings. We document a robust empirical pattern that the frequency of reversals among stable firms declines with age, a qualitative feature also implied by the model. To test whether the theory can quantitatively reproduce reversal dynamics, we calibrate the model targeting the average frequency of reversals and show that it delivers an untargeted age profile in line with the data. Finally, we study the efficiency properties of equilibrium. We show that the decentralized economy features inefficiently high learning; a planner would prefer information cascades to arise more quickly