Contextualizing Profitability

Abstract

We study the role of context in asset pricing by focusing on the narratives surrounding profitability. Using a large language model, we incorporate narrative context into the measurement of profitability. Contextualized profitability outperforms conventional profitability measures both in statistical and economic terms. Its predictive power stems from the model's ability to learn transitory vs. persistent variation in profits. Furthermore, the factor based on contextualized profitability is superior in pricing portfolios of assets and eliminates alpha in small extreme growth portfolios - the biggest challenge facing the five-factor model (Fama and French, 2015). Our results imply that incorporating narrative context not only improves investment strategies but also enhances the asset pricing tests.

Keywords: Contextual information, asset pricing, machine learning, neural networks, large language models, BERT, Transformer, operating profitability, factor models

JEL Classification: C13, C45, C55, C58, G11, G12, M41