Market-Cap Bias in LLM-Derived Textual Signals

Test the conjecture that textual signals extracted by large language models from financial news are of higher quality for large-cap companies than for small-cap companies due to news coverage bias.

Background

In comparing portfolio weighting schemes, the survey observes that market-capitalization-weighted portfolios showed slightly higher returns than equally weighted portfolios in multiple studies. The authors hypothesize a potential explanation: LLM-extracted textual signals might be more reliable for larger firms due to greater and higher-quality news coverage.

Validating or refuting this conjecture requires empirical analysis linking market capitalization, news coverage characteristics, and the accuracy or predictive value of LLM-derived signals.