Asset Pricing Program Meeting, Spring 2026

Why It Matters

The approach links textual market narratives to measurable valuation dynamics, offering a novel channel for explaining return variation, identifying potential mispricing, and informing both asset-pricing theory and systematic trading strategies.

Summary

Researchers presented a framework that measures how markets ‘represent’ firms using language embeddings from financial news and targeted foundation models, producing firm-by-year vector representations. They decompose annual cross-sectional stock-return variation and find these representation-related terms explain about 20% of variation—roughly two-thirds from changes in how feature bundles are valued and one-third from shifts in the representations themselves. The paper shows representations often mean-revert, that trading strategies betting against recent high-valued representation pivots earn returns, and that attention dynamics (e.g., AI pivots) help drive representation volatility. The authors have released their specialized models to the research community for replication and further study.