New supervised approach extracts sentiment from 10-K filings

Researchers have developed a supervised lexicon-learning approach to extract sentiment from 10-K filings, specifically focusing on the Item 1A risk-factor sections. This method was trained against both return and volatility labels at sector, portfolio, and individual firm aggregation levels. The study found that while full-filing text yields more accurate sentiment at broader aggregation levels, the narrower Item 1A section performs better for individual firms. A baseline using the Loughran-McDonald dictionary showed a consistent negative correlation with price, highlighting the value of supervised approaches for regulatory disclosures. AI

IMPACT This research could improve automated analysis of financial disclosures, potentially aiding investors and regulators in identifying risks and market trends.

RANK_REASON Academic paper detailing a new methodology for sentiment analysis on financial documents. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.LG →

• Form 10-K
• Item 1A
• Loughran-McDonald
• Nasdaq-100

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →