Text‐Based Investor Sentiment in Asset Pricing: A Survey of Belief Formation Mechanisms

ABSTRACT

Text‐based investor sentiment has become a widely used explanatory variable in asset pricing, yet its economic interpretation remains unsettled. A large empirical literature documents that sentiment measures derived from news, corporate disclosures, and social media predict short‐horizon returns, affect volatility, and explain trading activity. These findings are often interpreted as evidence of sentiment‐driven mispricing, but such conclusions depend on assumptions about how investors form, process, and revise beliefs. This survey re‐evaluates the literature through a mechanism‐based framework that distinguishes between competing channels of belief formation, including non‐fundamental demand, biased beliefs, rational information aggregation, attention and salience, and feedback effects. We show that similar empirical regularities can arise from fundamentally different economic mechanisms and therefore require careful theoretical interpretation. The survey clarifies the role of machine‐learning‐based sentiment measures, highlights the limits of interpreting predictability as market inefficiency, and provides a framework for linking sentiment evidence to coherent asset‐pricing models.