Nicolás Magner , Pablo A. Henríquez , Aliro Sanhueza
{"title":"Decoding risk sentiment in 10-K filings: Predictability for U.S. stock indices","authors":"Nicolás Magner , Pablo A. Henríquez , Aliro Sanhueza","doi":"10.1016/j.frl.2025.107472","DOIUrl":null,"url":null,"abstract":"<div><div>This study demonstrates that the tone of the risk factors section in the 10-K reports of U.S. public companies predicts returns on major U.S. stock indices. We created five tone indicators using text mining, the Loughran-McDonald dictionary, and AI-calibrated alternatives (GPT-3.5-turbo-0125, GPT-4, GPT-4o, and GPT-4o-mini). These indicators showed significant predictive power for weekly returns, with optimism correlated with higher returns. Tone measurements based on GPT-4 outperformed the others in terms of predictive accuracy. We analyzed the Loughran-McDonald dictionary’s utility and highlighted the underexplored risk factors section, offering novel insights into sentiment analysis and financial forecasting.</div></div>","PeriodicalId":12167,"journal":{"name":"Finance Research Letters","volume":"81 ","pages":"Article 107472"},"PeriodicalIF":7.4000,"publicationDate":"2025-05-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Finance Research Letters","FirstCategoryId":"96","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1544612325007317","RegionNum":2,"RegionCategory":"经济学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"BUSINESS, FINANCE","Score":null,"Total":0}
引用次数: 0
Abstract
This study demonstrates that the tone of the risk factors section in the 10-K reports of U.S. public companies predicts returns on major U.S. stock indices. We created five tone indicators using text mining, the Loughran-McDonald dictionary, and AI-calibrated alternatives (GPT-3.5-turbo-0125, GPT-4, GPT-4o, and GPT-4o-mini). These indicators showed significant predictive power for weekly returns, with optimism correlated with higher returns. Tone measurements based on GPT-4 outperformed the others in terms of predictive accuracy. We analyzed the Loughran-McDonald dictionary’s utility and highlighted the underexplored risk factors section, offering novel insights into sentiment analysis and financial forecasting.
期刊介绍:
Finance Research Letters welcomes submissions across all areas of finance, aiming for rapid publication of significant new findings. The journal particularly encourages papers that provide insight into the replicability of established results, examine the cross-national applicability of previous findings, challenge existing methodologies, or demonstrate methodological contingencies.
Papers are invited in the following areas:
Actuarial studies
Alternative investments
Asset Pricing
Bankruptcy and liquidation
Banks and other Depository Institutions
Behavioral and experimental finance
Bibliometric and Scientometric studies of finance
Capital budgeting and corporate investment
Capital markets and accounting
Capital structure and payout policy
Commodities
Contagion, crises and interdependence
Corporate governance
Credit and fixed income markets and instruments
Derivatives
Emerging markets
Energy Finance and Energy Markets
Financial Econometrics
Financial History
Financial intermediation and money markets
Financial markets and marketplaces
Financial Mathematics and Econophysics
Financial Regulation and Law
Forecasting
Frontier market studies
International Finance
Market efficiency, event studies
Mergers, acquisitions and the market for corporate control
Micro Finance Institutions
Microstructure
Non-bank Financial Institutions
Personal Finance
Portfolio choice and investing
Real estate finance and investing
Risk
SME, Family and Entrepreneurial Finance