Do investor emotions shape beliefs about stock market crashes?
Core argument: Using more than two decades of Shiller's Stock Market Confidence Survey and option prices, the authors decompose investor crash beliefs into fundamental and subjective components, finding the subjective component highly volatile, typically right-skewed, and nearly always positive.
Using more than two decades of Shiller’s Stock Market Confidence Survey we decompose crash beliefs into subjective and fundamental components based upon option prices, [which] reflect the time-varying beliefs and preferences of the marginal investor in these markets [who] take[s] into account fundamental factors and on average, rationally assess[es] crash probabilities. The subjective crash probability measure is highly volatile, typically right-skew, and nearly always positive. To better understand the subjective component in crash risk forecasts, we apply a large language model (LLM) to [survey response] narratives, to construct high-frequency measures of sentiment/emotion. The model is fine-tuned using rich data on human-derived emotion annotations to narratives. We find that investor sentiment is correlated with the subjective crash probability component and find strong evidence that it is related to non-fundamental variation. [Finally] we find that exposure and attention to news of rare, extreme events: i.e. nearby earthquakes, large lottery winnings, and excess COVID deaths is significantly associated with a higher or lower rare crash probability, depending on emotional valence.




