Between the Lines: Textual Features in Financial Reports and Expected Stock Returns
The input text analyzes predictive performance of traditional and textual stock characteristics using machine learning for portfolio return forecasts.
What it examines
This study employs machine learning on 40 textual features from corporate reports, comparing them with 137 traditional stock characteristics. It examines if text analysis can predict stock returns through portfolio sorting and various forecasting models. The goal is to understand whether textual data adds economic value in asset pricing.
What it concludes
The results indicate that textual analysis offers little additional return predictability beyond traditional methods. Benefits appear only in small, hard-to-trade stocks and dissipate over time. Applications include refining asset pricing models, though further work on advanced text methods and alternative data sources is needed.
Evidence objects
The study reveals financial report text analysis offers insights yet delivers impact, as advanced machine learning utilizing similarity, readability, and sentiment features shows weak predictive power versus traditional market metrics.
key_findings bullet 1 · key_findings · validation V0
Researchers compare standard stock characteristics and textual features across various models, including OLS, PLS, LASSO, and ensemble methods, using nearly three decades of U.S. market data and robust cross-validation techniques.
key_findings bullet 2 · key_findings · validation V0
The analysis uncovers that textual signals lose significance, particularly in hard-to-trade small stocks, with modest returns from similarity measures in microcap stocks constrained by arbitrage, low investor attention, and overpricing.
key_findings bullet 3 · key_findings · validation V0
The paper reassesses textual features in financial reports using machine learning by integrating traditional asset pricing models with modern NLP measures. Although it finds limited predictive power beyond conventional methods, it introduces a refreshing perspective challenging prevailing assumptions and offering a moderately original discussion that motivates exploration in financial analytics.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … First, we adopt a classical asset pricing lens to evaluate … for structural variation in security pricing and systematic … stocks, we exclude securities with a closing stock price …
Source row: 301 · abstract type: snippet