No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
When Is a Liability NOT a Liability? Textual Analysis, Dictionaries, and 10-Ks
Abstract
Previous research uses negative word counts to measure the tone of a text. We show that word lists developed for other disciplines misclassify common words in financial text. In a large sample of 10-Ks during 1994 to 2008, almost three-fourths of the words identified as negative by the widely used Harvard Dictionary are words typically not considered negative in financial contexts. We develop an alternative negative word list, along with five other word lists, that better reflect tone in financial text. We link the word lists to 10-K filing returns, trading volume, return volatility, fraud, material weakness, and unexpected earnings.
... Financial sentiment refers to the collective attitude, emotions, and opinions expressed by investors, analysts, and the public toward financial markets. The concept of financial sentiment has been widely studied, particularly in the context of stock market prediction (Tetlock, 2007;Schumaker and Chen, 2009;Bollen et al., 2011;Loughran and McDonald, 2011;Heston and Sinha, 2017;Kirtac and Germano, 2024a,b). ...
... Sentiment analysis is now commonly integrated into algorithmic trading models, where real-time news feeds and investor sentiment indices inform buy and sell decisions (Rao and Srivastava, 2014;Karoui, 2017). Given this trend, there has been a large influx of financial sentiment-related papers, with a focus on leveraging deep learning and LLMs to enhance accuracy in market predictions (Loughran and McDonald, 2011;Araci, 2019). ...
... Schumaker and Chen (2009) expanded this research by analyzing breaking news headlines, showing that sentiment-derived features could enhance stock return forecasting. These approaches were further refined by Li (2010) and Loughran and McDonald (2011), who developed domain-specific sentiment dictionaries tailored for financial text analysis, improving sentiment classification accuracy. ...
Financial sentiment has become a crucial yet complex concept in finance, increasingly used in market forecasting and investment strategies. Despite its growing importance, there remains a need to define and understand what financial sentiment truly represents and how it can be effectively measured. We explore the nature of financial sentiment and investigate how large language models (LLMs) contribute to its estimation. We trace the evolution of sentiment measurement in finance, from market-based and lexicon-based methods to advanced natural language processing techniques. The emergence of LLMs has significantly enhanced sentiment analysis, providing deeper contextual understanding and greater accuracy in extracting sentiment from financial text. We examine how BERT-based models, such as RoBERTa and FinBERT, are optimized for structured sentiment classification, while GPT-based models, including GPT-4, OPT, and LLaMA, excel in financial text generation and real-time sentiment interpretation. A comparative analysis of bidirectional and autoregressive transformer architectures highlights their respective roles in investor sentiment analysis, algorithmic trading, and financial decision-making. By exploring what financial sentiment is and how it is estimated within LLMs, we provide insights into the growing role of AI-driven sentiment analysis in finance.
... Tetlock (2007) uses the Harvard-IV (HIV) dictionary to measure the sentiment of the Wall Street Journal articles and finds that pessimism in news articles is negatively associated with the Dow Jones Industrial Average (DJIA) on the next day. Instead of using the HIV dictionary, Loughran and McDonald (2011) use a novel sentiment dictionary designed specifically for finance (hereafter referred to as "LM dictionary") [4]. They measure the sentiment of 10-K filings and show that their dictionary is better at predicting filing returns than the HIV dictionary. ...
... Sentiment analysis studies are proliferating in various fields. Conventional textual analyses in finance do not seem to provide strong forecasting powers because existing models compute sentiment scores based on predefined dictionaries that may contain less contents related to finance and accounting (Loughran and McDonald, 2011). Moreover, Loughran and McDonald (2016) raise concerns about readability measures such as the Fog Index, which is popularly used in finance and accounting. ...
... 4. Loughran and McDonald (2011) criticize the use of the HIV dictionary, arguing that sentiment analysis in business communications should avoid "classification schemes derived outside the domain of business usage" (p. 62). ...
We study the effectiveness of textual information in predicting the returns of crude oil futures and understanding the behavior of market participants. Using a machine learning method to extract oil market sentiment from news articles, we find that the computed sentiment is significantly effective in explaining the crude oil futures returns, while existing textual analyses based on pre-defined dictionaries may mislead the contexts in the oil market. Consistent with previous findings that returns help explain the change in traders’ positions, the sentiment scores based on the machine learning method are also useful in explaining the behavior of different types of traders. Our empirical findings underscore the fact that accurately identifying textual information can increase the accuracy of oil price predictions and explain divergent behaviors of oil traders.
... First, previous studies found limited incremental value for other word categories, such as positive words (Gandhi et al. 2019a, b). Research indicates that investors are likely to react more strongly to negative language in company news compared to positive language (Tetlock 2007;Tetlock et al. 2008;Loughran and McDonald 2011). Moreover, negative tones contain more informational content than positive tones (Henry and Leone 2016;Kothari et al. 2009). ...
... To address these questions, our study examines a sample of banks operating in the MENA region over the period from 2012 to 2022. We employ a content analysis approach along with the context-specific text dictionary developed by Loughran and McDonald (2011) to identify the negative reporting tone. Subsequently, we analyze the impact of this negative reporting tone on bank insolvency risk and investigate the moderating role of family and state ownership. ...
... This research makes several significant contributions to the existing literature. First, we extend studies on the effect of disclosure tone on risk, which have primarily focused on non-financial firms (e.g., Loughran and McDonald 2011;Iatridis 2016;Yekini et al. 2016;Bassyouny et al. 2020;Campbell et al. 2020;Oliveira et al. 2021;Elshandidy and Zeng 2022), by examining the unique context of banks, which operate under distinct regulatory frameworks and risk-taking paradigms. Second, we enhance the research of Del Gaudio et al. (2020) by exploring the previously understudied MENA context. ...
This study investigates the dual objectives of examining the effect of negative reporting tone on bank insolvency risk in the Middle Eastern and North African (MENA) region and analyzing the moderating role of ownership structure in this relationship. The empirical analysis is based on banks operating in the MENA region from 2012 to 2022. We employ a content analysis approach using the Loughran and McDonald dictionary to assess negative reporting tone. Our methodology then tests the impact of this tone on insolvency risk and explores the moderating influence of family and state ownership. The empirical results reveal that a more negative tone in annual reports increases bank insolvency risk. These findings align with appraisal theory and social identity theory, highlighting the significance of language in financial disclosures. Regarding ownership structure, state ownership emerges as a significant moderator, attenuating the negative impact of tone on insolvency risk. This research represents one of the first studies to examine the impact of disclosure tone on bank risk in the MENA region while specifically considering the distinctive characteristics of ownership structure.
... Consequently, if the word lists are not context-specific, there is a risk of introducing bias. In contrast, LSS requires a word list as a starting point, in our case an uncertainty word list from Loughran & McDonald (2011), and calculates the corpus-specific context words based on their proximity to the starting words, the uncertainty word list. Additionally, a sentiment score is applied to these context words to ascertain whether they have a positive or negative connotation. ...
... Furthermore, in each year on average 60 speeches were given with a standard deviation of 17. Pre-processing steps as for the removal of numbers, English stopwords, punctuation, symbols, URLs and separators were conducted to build a document-feature matrix, as a ready-to-use input for scaling estimation. Loughran-McDonald sentiment dictionary (Loughran & McDonald, 2011) gathers 297 different words related to uncertainty. In Figure 5.1, the frequency of Loughran-McDonald uncertainty words over time can be seen with evidence of a downward trend regarding the frequencies in governor speeches indicating avoidance of uncertainty terms. ...
... But they fail to determine sources of uncertainty and tie them to specific topics of interest that better explain the context used for such assertions. Moreover, the availability of sentiment dictionary or wordlists for uncertainty (Loughran & McDonald, 2011) cannot guarantee an effective application as context words, being corpus-specific, are often ignored when building such indices, despite being able to confirm hypotheses and meet economic developments. ...
This dissertation consists of six papers that focus on the measurement of uncertainty in economic contexts and offer different methods for quantifying uncertainty. The main objective is to contribute to the body of research on uncertainty in economics. The work largely uses text processing techniques, with the premise that uncertainty is a latent variable that cannot be measured directly, but can be approximated through communication. The first paper measures uncertainty in terms of future forecast errors, which are useful to economic decision makers and show the ability to predict future growth of industrial production. The research then moves to text-based measures of uncertainty, with two papers examining how online users' uncertainty can predict future economic developments. The final three papers examine central bank communication and find that uncertainty sentiment fluctuates over time but remains consistent across speakers. An uncertainty index derived from central bank speeches is shown to be an effective predictor of economic outcomes. The final chapter empirically tests the global central bank uncertainty index and confirms its predictive power for economic developments, at times outperforming previous indicators. Ultimately, this dissertation provides valuable tools for measuring and forecasting economic uncertainty, primarily through text-based indicators.
... However, while sustainability reporting has advanced in addressing stakeholder needs, audit reports often fail to provide meaningful insights into audit risks. Loughran and McDonald (2011) emphasized that narrative disclosures in financial reports, including audit reports, contain valuable information but require advanced analytical tools, such as sentiment analysis, to extract actionable insights [6]. The complexity of qualitative details in audit reports further emphasizes the need for accessible and reliable audit risk measures. ...
... However, while sustainability reporting has advanced in addressing stakeholder needs, audit reports often fail to provide meaningful insights into audit risks. Loughran and McDonald (2011) emphasized that narrative disclosures in financial reports, including audit reports, contain valuable information but require advanced analytical tools, such as sentiment analysis, to extract actionable insights [6]. The complexity of qualitative details in audit reports further emphasizes the need for accessible and reliable audit risk measures. ...
... Advanced analytical techniques, particularly sentiment analysis, have been recognized as powerful tools for uncovering insights from textual data. Loughran and McDonald (2011) pioneered sentiment analysis in financial disclosures, developing a framework for extracting sentiment from textual data and introducing the concept of domain-specific lexicons to enhance the relevance of sentiment-based metrics [6]. Their study demonstrated that sentiment analysis could reveal critical insights into audit risk that are often overlooked by traditional quantitative proxies. ...
This study introduces the Audit Risk Sentiment Value (ARSV), a novel audit risk proxy that leverages sentiment analysis to address limitations in traditional audit risk measures such as audit fees (LNFEE), audit hours (LNHOUR), and discretionary accruals (|MJDA|). Traditional proxies primarily capture quantitative dimensions, overlooking qualitative insights embedded in audit report narratives. By systematically analyzing sentiment and tone, ARSV captures nuanced audit risk dimensions that reflect the auditor’s risk perception. The study validates ARSV using a dataset of South Korean firms listed on the KOSPI from 2018 to 2023. The results demonstrate the ARSV’s superior explanatory power, as confirmed through the Vuong test, showing consistent performance across binary and continuous measures of explanatory language. ARSV bridges the gap between qualitative and quantitative audit risk assessments, offering significant benefits to auditors, regulators, and investors. Its ability to enhance the interpretability of audit reports improves transparency and trust in financial reporting, addressing stakeholder demands for actionable, forward-looking information. Furthermore, ARSV aligns with global trends emphasizing sustainability and accountability by integrating qualitative insights into audit practices. While this study provides robust evidence supporting ARSV effectiveness, its focus on South Korean firms may limit its generalizability. Future research should explore ARSV application in diverse regulatory and cultural contexts and refine the sentiment analysis tools using advanced machine learning techniques. Expanding ARSV to include other unstructured data, such as management commentary, could further enhance its applicability. This study marks a significant step toward modernizing audit methodologies, aligning them with evolving demands for comprehensive and transparent financial reporting. The empirical analysis reveals that ARSV outperforms traditional audit risk proxies with significantly higher explanatory power. Specifically, ARSV achieved a pseudo R2 of 0.786, compared to 0.608 for LNFEE, 0.604 for LNHOUR, and 0.578 for |MJDA|. The Vuong test results further validate ARSV superiority, with Z-statistics of −12.168, −12.492, and −9.775 when compared against LNFEE, LNHOUR, and |MJDA|, respectively. The model incorporating ARSV demonstrated a 62.454 F-value and an Adjusted R2 of 0.599, highlighting its robustness and reliability in audit risk assessment. These quantitative metrics underscore ARSV’s effectiveness in capturing qualitative audit risk dimensions, offering a more precise and informative measure for stakeholders.
... Financial sentiment refers to the collective attitude, emotions, and opinions expressed by investors, analysts, and the public toward financial markets. The concept of financial sentiment has been widely studied, particularly in the context of stock market prediction (Tetlock, 2007;Schumaker and Chen, 2009;Bollen et al., 2011;Loughran and McDonald, 2011;Heston and Sinha, 2017;Kirtac and Germano, 2024a,b). ...
... Sentiment analysis is now commonly integrated into algorithmic trading models, where real-time news feeds and investor sentiment indices inform buy and sell decisions (Rao and Srivastava, 2014;Karoui, 2017). Given this trend, there has been a large influx of financial sentiment-related papers, with a focus on leveraging deep learning and LLMs to enhance accuracy in market predictions (Loughran and McDonald, 2011;Araci, 2019). ...
... Schumaker and Chen (2009) expanded this research by analyzing breaking news headlines, showing that sentiment-derived features could enhance stock return forecasting. These approaches were further refined by Li (2010) and Loughran and McDonald (2011), who developed domain-specific sentiment dictionaries tailored for financial text analysis, improving sentiment classification accuracy. ...
Financial sentiment analysis has become a central tool in market forecasting, with an increasing number of academic studies incorporating sentiment measures into financial prediction models. I investigate the origins and use of sentiment measures in finance, tracing their evolution from market-based and lexicon-based approaches to advanced natural language processing techniques. The emergence of large language models has significantly improved the accuracy and depth of sentiment estimation. I examine how BERT-based models, such as RoBERTa and FinBERT, are optimized for structured sentiment classification, while GPT-based models, including GPT-4, OPT, and LLaMA, are more effective for financial text generation and real-time sentiment interpretation. A comparative analysis of bidirectional and autoregressive transformer architectures highlights their respective advantages in algorithmic trading, investor sentiment analysis, and financial decision-making. Hybrid approaches that combine classification and generative capabilities enhance predictive performance in sentiment-driven trading strategies. Findings underscore the increasing role of LLMs in financial sentiment analysis, enabling more nuanced, context-aware sentiment extraction from financial news, earnings reports, and social media data.
... develop a new ECB-specific dictionary by tailoring the Loughran and McDonald (2011) finance-specific dictionary. Tailoring involves several modifications: updating the dictionary to include British spellings, reclassifying certain words (e.g., 'stability,' 'efficiency,' 'lag') from positive or negative to neutral based on their context in ECB texts, and including common two-and three-word phrases (bigrams and trigrams) involving sentiment words. ...
... Tailoring involves several modifications: updating the dictionary to include British spellings, reclassifying certain words (e.g., 'stability,' 'efficiency,' 'lag') from positive or negative to neutral based on their context in ECB texts, and including common two-and three-word phrases (bigrams and trigrams) involving sentiment words. The Loughran and McDonald (2011) dictionary has been used in a number of studies that analyse central bank communication (see e.g. Tillmann and Walter (2019), Baranowski et al. (2021), Schmeling and Wagner (2024)). ...
... We adopt a textual analysis method to assess managerial integrity. Using the dictionary approach (Loughran and McDonald, 2011), we extract the frequency of key terms relevant to integrity from the Management Discussion and Analysis (MD&A) section of annual reports [7]. Based on Guiso et al. (2015) and Li et al. (2021), we identify key terms related to integrity as: "Integrity"; "Honesty"; "Truthfulness"; "Sincerity"; "Loyalty"; "Morality"; "Trust", "Reliance"; "Fairness"; "Justice"; "Accountability"; "Ethic"; "Responsibility"; "Transparency"; "Accountable"; "Governance"; "Ethical"; "Transparent"; "Responsible"; "Oversight"; "Independence"; "Objectivity"; "Moral"; "Trustworthy"; "Fairness"; "Hold accountable"; "Assure"; "Fiduciary responsibility"; "Credibility"; "Privacy"; "Fiduciary duty"; "Rigor". ...
... Criteria for this classification include consecutive losses over two years, net assets dropping below the stock's face value, or the possibility of ceasing operations due to substantial disasters or accidents. 8. Following the methodology outlined by Loughran and McDonald (2011), we categorize and consolidate closely related terms to enhance the efficiency of our analysis. 9. Following Ghosh and Olsen (2009), a firm's unadjusted environmental uncertainty, when divided by the industry environmental uncertainty, results in the industry-adjusted environmental uncertainty. ...
Purpose-The objective of this study is to investigate how the implementation of an Emission Trading Scheme (ETS) influences an ETS-regulated firm's level of earnings smoothness. Design/methodology/approach-Using a staggered difference-indifferences model based on China's ETS pilots commencing in 2013, this study investigates how the implementation of ETS pilots affects regulated firms' earnings smoothing relative to non-regulated firms. The sample period spans from 2008 to 2019. This model incorporates time-invariant firm-specific heterogeneity, time-specific heterogeneity, and a series of firm characteristics to establish causality. Robustness tests justify findings. Findings-The results show that after implementing an ETS pilot, regulated firms increase their earnings smoothness relative to non-regulated firms. Regulated firms strategically smooth their earnings to obtain additional financial resources and meet compliance costs arising from an ETS. Further analysis reveals that regulated firms' earnings smoothing activity is a function of environmental regulations, managerial integrity, and capital market incentives. Originality/value-This study deviates from past research focusing on the environmental consequences of ETS by indicating that an ETS affects regulated firms' financial reporting decisions. Specifically, regulated firms resort to earnings smoothing as a short-term exit strategy from financing concerns arising from environmental regulations. This finding expands prior literature primarily focusing on the effect of tax and financial reporting regulations on earnings smoothness. This study also indicates that firms utilize earning smoothing to lower their short-term cost of capital, which enables them to access additional financing at a lower cost and reconfigure their operations to meet stakeholder environmental demands.
... The digitalisation of enterprise. Loughran and McDonald (2011) showed that quantifying and analyzing key information through word frequency can reflect business characteristics. With the advancement of natural language processing technology, using text analysis to construct the enterprise digitalisation metric is increasingly becoming a major approach (Chen and Srinivasan, 2024;Wu and Li, 2024;Wu et al., 2021;Yuan et al. 2021;Zhao et al. 2021). ...
... In addition, the Word2vce technique requires specific text to train the model, and according to the research needs, we choose the content of managerial discussion and analysis (MD&A) disclosed in the annual reports of listed companies as the research text. This text reflects the review of business conditions and the outlook for future development of the enterprise during the reporting period, and it has been widely recognized as the research object (Loughran and McDonald, 2011). ...
Employing a machine learning measure, we find that mixed ownership reform in state-owned enterprises (SOEs) significantly advances digitalisation. This effect is primarily achieved through the pay-performance sensitivities of management and corporate risk-taking; it is particularly pronounced in competitive industries and among corporate decision-makers who did not experience the Great Chinese Famine during childhood. Our study not only explores digitalisation measures with the help of cutting-edge natural language processing techniques but also expands the literature on digitalisation motivation and the impact of mixed ownership reform on business decisions. The findings have important implications for promoting digitalisation strategies in SOEs.
... Building on Zhou [14] and Loughran & McDonald's [15] text analysis methodology, we develop a multidimensional green transition index (GTR) using annual report disclosures. Conventional approaches relying on green patents or environmental investments inadequately capture strategic and managerial dimensions of transition, particularly in service industries. ...
Amidst the deepening implementation of China’s “Dual Carbon” national strategy, this study integrates the Natural Resource-Based View with financing constraint theory to investigate the pollution mitigation effects of corporate green transition. Utilizing textual analysis on panel data from heavily polluting A-share listed companies (2013-2022), we develop a multidimensional green transition index and employ fixed-effects regression models. Key findings reveal: (1) Corporate green transition demonstrates significant marginal improvement effects on pollution abatement, with enhanced efficacy observed in firms facing stringent environmental regulations; (2) Financing constraints exhibit substantial negative moderating effects on the green transition-pollution reduction nexus. Theoretically, this research advances micro-level understanding of environmental governance economics. Practically, it provides empirical support for optimizing transition finance instruments and calibrating environmental regulation intensity gradients.
... 9 We collect about 907,000 analyst reports in our sample period from the East Money webpage and the Wind database. For each report, we use the Loughran and McDonald (2011) tone dictionary and compute analyst tone as the number of positive (i.e., optimistic) sentences scaled by the total number of sentences. ...
We explore the impact of political factors on local analysts’ earnings forecasts on the basis of a scenario of financial misconduct in Chinese state-owned enterprises (SOEs). The results show that local analysts’ earnings forecasts for SOEs involved in financial misconduct are less accurate and more optimistically biased. Heterogeneity analysis reveals that forecast bias and optimism by local analysts are greater when officials have stronger promotion incentives, when regions are less market-oriented, and when violating SOEs contribute more to taxation and employment. In a further analysis, we find that local analysts who rely heavily on political information will issue more biased and optimistic forecasts on violating SOEs. Although investors perceive the optimistic forecasts of local analysts as less credible, analysts’ catering behaviors can mitigate the negative impact of misconduct on the stock prices of SOEs in the year of occurrence. As information asymmetry is alleviated in longer windows, investors’ views will reverse. Moreover, forecast bias will reduce the likelihood of local analysts attaining star status. Finally, as a reward for catering to government needs, local brokerages whose affiliated analysts have provided optimistic forecasts are more likely to become underwriters in the seasoned equity offerings of SOEs and local government bond issuances. Our study contributes to the literature on factors affecting analyst forecast bias and analysts’ catering behaviors by highlighting the reciprocal relationships between the government and information intermediaries.
... ABFSA Early financial sentiment analysis employed dictionary-based and statistical methods [15][16][17]. Dictionary-based methods rely on predefined sentiment lexicons, such as the Harvard Inquirer or the Loughran-McDonald financial sentiment dictionary [18]. Popoola et al. [13] introduced Term Frequency and Inverse Document Frequency (TF-IDF) to solve the sentiment classification task. ...
Aspect-based financial sentiment analysis (ABFSA) is a challenging task at the intersection of finance and natural language processing, which aims to infer the trend of a specific entity (e.g., a company or stock) by identifying the sentiment reflected in financial-related texts, while there are a great deal of outstanding works dealing with aspect-level sentiment analysis task. However, there remains a deficiency in research explicitly targeting the ABFSA. In particular, due to the expressive differences between texts in the financial domain and social media texts, existing models lose sentiment details when mining the semantics of finance-related texts. To tackle this issue, the paper conducts a thorough analysis of the expressive features of financial texts and proposes a dual-enhanced GCN network (DEGCN) for financial sentiment analysis. DEGCN is composed of two main components: The Sentic-enhanced GCN focuses on fine-grained sentiment connections between words to overcome the loss of sentiment details caused by unified modeling, and the Domain-enhanced GCN is designed based on the characteristics of financial texts, dividing each sentence into finer categories for classification, thereby significantly reducing noise introduced by domain quantification inconsistencies. Extensive experiments and ablation studies on two public benchmark datasets demonstrate that the proposed model achieves significant improvements over previous baselines.
... In this section, we investigate the channel for the impact of mandatory disclosure of private in-house meetings on firms' greenwashing behaviors, specifically focusing on the corporate culture of integrity. Following the dictionary-based approach outlined by Loughran and McDonald (2011), we analyze the key terms frequency associated with corporate culture of integrity within the Management Discussion and Analysis (MD&A) section of firms' annual reports. Drawing on the frameworks established by Guisoet al. (2015) and Li et al. (2021), we identify essential terms related to the corporate culture of integrity, including "Integrity", "Honesty", "Truthfulness", "Sincerity", "Loyalty", "Morality", "Trust", "Reliance", "Fairness", "Justice", "Accountability", "Ethic", "Responsibility", "Transparency", "Accountable", "Governance", "Ethical", "Transparent", "Responsible", "Oversight", "Independence", "Objectivity", "Moral", "Trustworthy", "Fairness", "Hold accountable", "Assure", "Fiduciary responsibility", "Credibility", "Privacy", "Fiduciary duty", and "Rigor". ...
This study explores whether and how mandatory disclosure of private in-house meetings affects corporate environmental behaviors. Employing a difference-in-differences model, we find that regulated firms significantly reduce their greenwashing behaviors in response to the implementation of mandatory disclosure of private in-house meetings. We identify that mandatory disclosure of private in-house meetings decreases greenwashing behaviors by fostering a corporate culture of integrity. Firms with higher information asymmetry before the mandate exhibit a greater reduction in greenwashing behaviors after mandatory disclosure of private in-house meetings. Furthermore, we document that firms reducing greenwashing as a result of this mandatory disclosure policy witness an improvement in their social reputations. Our study offers insightful implications to policymakers and practitioners by shedding light on the significant role of investor-focused communications in mitigating greenwashing behaviors, thereby bolstering firms' environmental integrity and accountability.
... Furthermore, our exploration with randomly selected in-context samples highlights the potential for further improvements through more effective in-context sample retrieval strategies. 1931& McDonald, 2011. However, they struggle with automation, scalability, and interpretation accuracy. ...
Recently, large language models (LLMs) with hundreds of billions of parameters have demonstrated the emergent ability, surpassing traditional methods in various domains even without fine-tuning over domain-specific data. However, when it comes to financial sentiment analysis (FSA)\unicode{x2013}a fundamental task in financial AI\unicode{x2013}these models often encounter various challenges, such as complex financial terminology, subjective human emotions, and ambiguous inclination expressions. In this paper, we aim to answer the fundamental question: whether LLMs are good in-context learners for FSA? Unveiling this question can yield informative insights on whether LLMs can learn to address the challenges by generalizing in-context demonstrations of financial document-sentiment pairs to the sentiment analysis of new documents, given that finetuning these models on finance-specific data is difficult, if not impossible at all. To the best of our knowledge, this is the first paper exploring in-context learning for FSA that covers most modern LLMs (recently released DeepSeek V3 included) and multiple in-context sample selection methods. Comprehensive experiments validate the in-context learning capability of LLMs for FSA.
... Siering [46] recently examined the impact of both topical and linguistic features on loan default prediction. To extract topics, the author employed a financial text analysis method [29] to construct a domain-specific dictionary. The identified topics captured elements such as the loan purpose, the borrower's requests for assistance, expressions of reliability, and appreciation. ...
Peer-to-peer (P2P) lending connects borrowers and lenders through online platforms but suffers from significant information asymmetry, as lenders often lack sufficient data to assess borrowers’ creditworthiness. This paper addresses this challenge by leveraging BERT, a Large Language Model (LLM) known for its ability to capture contextual nuances in text, to generate a risk score based on borrowers’ loan descriptions using a dataset from the Lending Club platform. We fine-tune BERT to distinguish between defaulted and non-defaulted loans using the loan descriptions provided by the borrowers. The resulting BERT-generated risk score is then integrated as an additional feature into an XGBoost classifier used at the loan granting stage, where decision-makers have limited information available to guide their decisions. This integration enhances predictive performance, with improvements in balanced accuracy and AUC, highlighting the value of textual features in complementing traditional inputs. Moreover, we find that the incorporation of the BERT score alters how classification models utilize traditional input variables, with these changes varying by loan purpose. These findings suggest that BERT discerns meaningful patterns in loan descriptions, encompassing borrower-specific features, specific purposes, and linguistic characteristics. However, the inherent opacity of LLMs and their potential biases underscore the need for transparent frameworks to ensure regulatory compliance and foster trust. Overall, this study demonstrates how LLM-derived insights interact with traditional features in credit risk modeling, opening new avenues to enhance the explainability and fairness of these models.
... A second strand of literature pertains to the application and development of natural language processing (NLP) tools for financial economics research (Hoberg and Manela, 2025). Methodological advancements in this area have evolved from early dictionarybased approaches (Tetlock, 2007;Loughran and McDonald, 2011), to text regressions (Manela and Moreira, 2017;Kelly et al., 2021), to topic modeling (Bybee et al., 2024), and most recently, to the integration of LLMs (Jha et al., 2025;Chen et al., 2023;Lv, 2024). These developments underscore the growing demand for more advanced and scalable tools to answer research questions in finance and economics. ...
Large language models are increasingly used in social sciences, but their training data can introduce lookahead bias and training leakage. A good chronologically consistent language model requires efficient use of training data to maintain accuracy despite time-restricted data. Here, we overcome this challenge by training chronologically consistent large language models timestamped with the availability date of their training data, yet accurate enough that their performance is comparable to state-of-the-art open-weight models. Lookahead bias is model and application-specific because even if a chronologically consistent language model has poorer language comprehension, a regression or prediction model applied on top of the language model can compensate. In an asset pricing application, we compare the performance of news-based portfolio strategies that rely on chronologically consistent versus biased language models and estimate a modest lookahead bias.
... The study measures the corporate competitive strategy preference type by the proportion of the total word count of competitive strategy vocabulary to the total word count of the annual report. Compared to measuring word frequency with total word count, this approach effectively avoids the influence of document size on the test results (Loughran and McDonald, 2011). The specific steps are as follows: ...
Purpose-This study investigates how corporations navigate the increasingly prominent field of environmental, social and governance (ESG) through the lens of resource dependence theory (RDT). It aims to elucidate the strategic responses of companies to media-driven public sentiment on ESG, examining the alignment of their operations and competitive strategies-specifically differentiation and cost leadership-to the external resource of media ESG sentiment.
Design/methodology/approach-Employing Python software, this research extracted over two million ESG-related news articles from Baidu News. Using machine learning and text analysis, the study assesses the media ESG sentiment and its correlation with the competitive strategies of China's A-share listed companies over a period from 2007 to 2022. The approach leverages RDT to understand how firms adjust their strategies in response to media-driven public sentiment on ESG.
Findings-The findings indicate that positive media ESG sentiment acts as a crucial external resource, significantly influencing firms' strategic alignment toward minimizing ESG public sentiment risks and enhancing competitive positioning, especially in the social (S) and governance (G) domains. This alignment is evident in firms' adoption of differentiation and cost leadership strategies, affirming the study's theoretical prediction within the RDT framework.
Originality/value-This paper provides a novel contribution by integrating RDT with the analysis of media-driven ESG sentiment to explore corporate strategic adjustments. It offers empirical evidence on the theory's applicability in contemporary strategic corporate management, particularly in the context of ESG challenges. The research deepens the understanding of the interplay between media ESG sentiment and corporate strategy, highlighting the strategic importance of positive media sentiment in the ESG landscape.
The World Economic Forum (WEF) Global Risk Reports (GRRs) are published annually with the aim to uncover the most pressing challenges facing the world. However, the GRR have been criticized for presenting an overly simplistic and potentially biased portrayal of interconnected global risks and crises. Despite their influence, no in-depth, interannual analysis of the GRRs has been conducted to date. To address this gap, we analyze GRRs from 2006 to 2024 using textual analysis, systematic screening, and back- and forecasting methodologies. Our findings reveal a linguistic shift toward a technical, expert-driven narrative that frames global risks as regulatory challenges rather than opportunities for systemic transformation. Comparing text versus survey responses, the text of GRRs overemphasize economic considerations, marginalize environmental and social dimensions, and underrepresent ecological impacts. A comparison of GRR risk likelihoods with historical shocks shows consistent misalignment across most risk categories. By perpetuating an anthropocentric, business-centered, and fragmented representation of global risks, non-critical interpreations of the GRR can themselvself amplify risks to global sustainability and equity at a time of multiple interacting criss. We propose practical recommendations for use of the GRR and how they can be recalibrated to better represent multiple interacting global risks.
This study examines the association between pessimistic tones in earnings announcements and firm value, as well as the role of CEOs’ financial experience during the Covid-19 pandemic. Fixed effects regression was employed to analyze 1,380 firm-year observations from Indonesia Stock Exchange-listed non-financial enterprises during the pandemic. The analysis results indicate that a pessimistic tone in earnings announcements negatively impacts firm value, while the CEO's financial experience reduces this negative effect. The study added to the literature by revealing that CEOs' financial experience acts as a credibility signal for investors, reducing the association between pessimistic tone and firm valuation during pandemic.
Stock price prediction is an extensively researched topic as the precise prophecy of stock trends is decisive in the investment marketing sphere. With increasing opinions by many market giants on the internet about given stocks, it surges the necessity to study these sentiments in detail for forthcoming predictions. From these articles on the internet, natural text is generated by examining factors that affect the values of stocks and therefore these texts are reliable features to go ahead with this study. The idea behind tackling such work is that conglomerates and businesses are able to tangibly understand the aftermath of articles that usually mobilize public opinion and gear them in a certain direction. The aim of this study is to utilize named entity recognition (NER) on a neural network framework for stock trend prediction through latent Dirichlet allocation using these natural texts generated from internet articles. This method is used to understand the words that occur at the highest frequency and add the most information to the corpus depending on the topic’s importance. With this, the model adopts K × K words that have the most decisive impact on the target class that has been created with which it alters the sparse density matrix that has been generated. The proposed model of the NER-based neural network was fitted on a real-world dataset, and its performance was good in comparison with state-of-the-art models developed by fellow researchers. However, since the model does not use the BERT tokenizers, it cannot be adjudged on the FinBERT model, and therefore, the preprocessed data is fed to a pruned recurrent neural network which is robustly stopped with a simple callback function. The final result was a strong 0.81 tetrachoric correlation between the testing target class and the predicted target class. With this, the model provides a different approach to natural language processing, especially those with high sparse density for stock prediction.
Bu çalışma, Türkiye Cumhuriyet Merkez Bankası'nın (TCMB) 2006-2024 yılları arasında yayımladığı para politikası basın duyurularını metin analizi yöntemleriyle incelemektedir. Çalışmada doğal dil işleme (NLP) teknikleri kullanılarak kosinüs benzerliği, Gizli Dirichlet Tahsisi (LDA) konu modellemesi ve duygu analizleri yapılmıştır. Kosinüs benzerliği analizi, duyurular arasındaki içerik tutarlılığını ölçerken, LDA yöntemi ile duyuruların tematik yapısı ve temel konuları belirlenmiştir. Duygu analizi ise metinlerin pozitif, negatif ve nötr tonlarını ortaya koymuştur. Sonuçlar, TCMB'nin iletişim stratejisinin ekonomik koşullara bağlı olarak değiştiğini ve piyasa beklentilerini yönlendirme amacı taşıdığını göstermektedir. Çalışma, merkez bankası iletişiminde kullanılan dilin ekonomik etkilerini değerlendirmesi bakımından literatüre katkı sunmaktadır.
This paper investigates the bidirectional relation between the Brazilian Central Bank communication and the yield curve. Using latent factors, observable macroeconomic variables, and observable variables representing Central Bank communication, we estimate a model that summarizes the yield curve. We find evidence of the effects of Brazilian Central Bank communication on the movements of the yield curve and the impact of the yield curve components in Brazilian Central Bank communication. In particular, Central Bank communication can shape yield curve curvature and slope. Additionally, we find a strong relation between Central Bank communication and the curvature of the yield curve. These results show that Central Bank communication impacts market players, making it a valuable instrument for monetary policy.
This paper explores the information content of the untargeted narratives of the Bank of England (BoE), the European Central Bank (ECB), and the Federal Reserve (Fed) and whether it has the potential to improve forecasting performance. We apply the Latent Dirichlet Allocation (LDA) method to extract topics from the corpus of text data. We then evaluate the impact of these central bank officials' speeches on macroeconomic and financial variables forecasting from 1997 to 2018. Our results suggest that the forecasting model, incorporating information from speeches, produces estimates with a lower forecasting error for several variables in the UK, the EU, and the US. For certain variables, the forecast improvement is more pronounced after the global financial crisis of 2008.
The work herein reviews the scientific literature on Machine Learning approaches for financial risk assessment using financial reports. We identify two prominent use cases that constitute fundamental risk factors for a company, namely misstatement detection and financial distress prediction. We further categorize the related work along four dimensions that can help highlight the peculiarities and challenges of the domain. Specifically, we group the related work based on (a) the input features used by each method, (b) the sources providing the labels of the data, (c) the evaluation approaches used to confirm the validity of the methods, and (d) the machine learning methods themselves. This categorization facilitates a technical overview of risk detection methods, revealing common patterns, methodologies, significant challenges, and opportunities for further research in the field.
This paper examines how local/global CSR activity induced local/remote ownership affects firm performance. The results show that local CSR activity generates positive externality to local owners, who in turn increase their ownership, and hence, also their monitoring role in local firms. Global CSR activity builds public image, and thus attract remote investors to local firms, which increases diversification and hence, liquidity of firm stock. The shift in local ownership increases firm value more for global CSR activities that target a broader group of stakeholders. Thus, some monitoring by local owners is necessary to prevent overinvestment in CSR that might lead to the manifestation of agency problems. In the short-run stock returns increase more for positive news on global CSR activities, partly, because of overreaction. In the long-run there is a return reversal, however, stock returns are still positive, indicating that CSR activities not only generate overreaction but also influence firm fundamentals.
Environmental, social, and governance (ESG) principles have gained prominence in the capital markets. While ESG ratings are widely used to assess corporate sustainability, their disagreement and time lag limit their effectiveness for investors. This study proposes ESG news sentiment as an alternative measure of corporate ESG activities and investigates the investor responses to it. Using sustainability reports, we developed lexicons to classify news into ESG and non‐ESG categories and further categorized ESG news into environmental, social, and governance dimensions. Through hierarchical regression analysis, we examine the impact of media‐derived ESG factors on market value and stock returns. Our findings reveal that ESG news sentiment is more positively associated with firm performance compared to non‐ESG news. This is particularly evident for social and governance dimensions, with effects varying across industries depending on their ESG concerns. These results demonstrate the value of ESG news sentiment analysis in understanding market responses to corporate sustainability practices.
Emerging climate risk perception (CRP) has drawn significant attention to its critical role in driving firms' ESG performance. We construct CRP development at the firm level employing the text analysis method. We explore the causal relationship between CRP and ESG performance using a data set covering listed firms from 2011 to 2022 in China. Our results demonstrate that CRP promotes firm ESG performance, and it is more evident in non-high-tech, non-heavy polluting, and labor-intensive firms. In addition, promoting sustainable green innovation, environmental protection investment, and alleviating information asymmetry are three important channels through which CRP affects ESG performance. Further analysis indicates that CRP strengthens firms' green value by improving total green factor productivity. Our findings offer actionable insights for firms to achieve green transformation in practice.
Synopsis
The research problem
Our study explores the association between tax enforcement and tone management in the Management’s Discussion and Analysis (MD&A).
Motivation
Prior studies have shown that tax enforcement influences financial reporting due to the tax authority’s scrutiny of financial statements to detect tax evasion, which deters financial reporting manipulation. However, the spillover effect of tax enforcement on qualitative disclosures remains unexplored. Motivated by the relevance of MD&A as a comprehensive discussion of a firm’s conditions and a target for manipulation, we investigate the association between tax enforcement and tone management.
The test hypotheses
We test two competing hypotheses: increased tax enforcement is associated with more tone management, and it is associated with lesser tone management.
Target population
Our study should be of interest to corporate stakeholders, including managers, investors, regulatory authorities, and policymakers.
Adopted methodology
Ordinary least squares regressions, staggered difference-in-differences model, and path analysis.
Analysis
Our study explores a sample of Chinese A-share listed firms from 2008 to 2020. We measure tone management using Huang et al.’s [(2014). Tone management. The Accounting Review, 89(3), 1083–1113. https://doi.org/10.2308/accr-50684 ] model, which regresses the tone level in MD&A on firm fundamentals, and use the discretionary tone portion as a proxy. Tone level is calculated as the difference between the number of positive and negative sentiment words divided by their sum.
Findings
We find that increased tax enforcement is associated with lesser tone management. Path analysis shows that this effect is driven by constraints on earnings management and enhanced accounting conservatism, suggesting that tax enforcement improves financial reporting quality, which in turn disciplines disclosure tone. Cross-sectional tests further indicate that the effect of increased tax enforcement varies with tax compliance levels, managerial incentives to influence investor perceptions, and external monitoring.
We study rare events data, binary dependent variables with dozens to thousands of times fewer ones (events, such as wars, vetoes, cases of political activism, or epidemiological infections) than zeros ("nonevents"). In many literatures, these variables have proven difficult to explain and predict, a problem that seems to have at least two sources. First, popular statistical procedures, such as logistic regression, can shar ply underestimate the probability of rare events. We recommend corrections that outperform existing methods and change the estimates of absolute and relative risks by as much as some estimated effects repor ted in the literature. Second, commonly used data collection strategies are grossly inefficient for rare events data. The fear of collecting data with too few events has led to data collections with huge numbers of obser vations but relatively few, and poorly measured, explanator y variables, such as in international conflict data with more than a quarter-million dyads, only a few of which are at war. As it turns out, more efficient sampling designs exist for making valid inferences, such as sampling all available events (e.g., wars) and a tiny fraction of nonevents (peace). This enables scholars to save as much as 99% of their (nonfixed) data collection costs or to collect much more meaningful explanator y variables. We provide methods that link these two results, enabling both types of corrections to work simultaneously, and software that implements the methods developed.
Using word content analysis, we decompose information in the initial public offering prospectus into its standard and informative
components. Greater informative content, as a proxy for premarket due diligence, results in more accurate offer prices and
less underpricing, because it decreases the issuing firm’s reliance on bookbuilding to price the issue. The opposite is true
for standard content. Greater content from high reputation underwriters and issuing firm managers, through Management’s Discussion
and Analysis, contribute to the informativeness of the prospectus. Our results suggest that premarket due diligence and disclosure
by underwriters and issuers can serve as a substitute for costly bookbuilding.
Extracting sentiment from text is a hard semantic problem. We develop a methodology for extracting small investor sentiment from stock message boards. The algorithm comprises dierent classifier algorithms coupled together by a voting scheme. Accuracy levels are similar to widely used Bayes classifiers, but false positives are lower and sentiment accuracy higher. Time series and cross-sectional aggregation of message information improves the quality of the resultant sentiment index, particularly in the presence of slang and ambiguity. Empirical applications evidence a relationship with stock values - aggregate tech sector sentiment is found to predict stock index levels, but not at the individual stock level. The algorithms may be used to assess the impact on investor opinion of management announcements, press releases, third-party news, and regulatory changes.
We study rare events data, binary dependent variables with dozens to thousands of times fewer ones (events, such as wars, vetoes, cases of political activism, or epidemiological infections) than zeros ("nonevents"). In many literatures, these variables have proven difficult to explain and predict, a problem that seems to have at least two sources. First, popular statistical procedures, such as logistic regression, can shar ply underestimate the probability of rare events. We recommend corrections that outperform existing methods and change the estimates of absolute and relative risks by as much as some estimated effects repor ted in the literature. Second, commonly used data collection strategies are grossly inefficient for rare events data. The fear of collecting data with too few events has led to data collections with huge numbers of obser vations but relatively few, and poorly measured, explanator y variables, such as in international conflict data with more than a quarter-million dyads, only a few of which are at war. As it turns out, more efficient sampling designs exist for making valid inferences, such as sampling all available events (e.g., wars) and a tiny fraction of nonevents (peace). This enables scholars to save as much as 99% of their (nonfixed) data collection costs or to collect much more meaningful explanator y variables. We provide methods that link these two results, enabling both types of corrections to work simultaneously, and software that implements the methods developed.
Clarify is a program that uses Monte Carlo simulation to convert the raw output of statistical procedures into results that are of direct interest to researchers, without changing statistical assumptions or requiring new statistical models. The program, designed for use with the Stata statistics package, offers a convenient way to implement the techniques described in: Gary King, Michael Tomz, and Jason Wittenberg (2000). "Making the Most of Statistical Analyses: Improving Interpretation and Presentation." American Journal of Political Science 44, no. 2 (April 2000): 347-61. We recommend that you read this article before using the software. Clarify simulates quantities of interest for the most commonly used statistical models, including linear regression, binary logit, binary probit, ordered logit, ordered probit, multinomial logit, Poisson regression, negative binomial regression, weibull regression, seemingly unrelated regression equations, and the additive logistic normal model for compositional data. Clarify Version 2.1 is forthcoming (2003) in Journal of Statistical Software.
This paper examines whether the “soft” information contained in the text of management’s quarterly earnings press releases is incrementally informative over the company’s reported “hard” earnings news. We use Diction, a textual-analysis program, to extract various dimensions of managerial net optimism from more than 20,000 corporate earnings announcements over the period 1998 to 2006 and document that unanticipated net optimism in managers’ language affects announcement period abnormal returns and predicts post earnings announcement drift. We find that it takes longer for the market to understand the implications of soft information than those of hard information. We also find that the market response varies by firm size, turnover, media and analyst coverage, and the extent to which the standard accounting model captures the underlying economics of the firm. We also show that the second moment of soft information, the level of certainty in the text, is an important determinant of contemporaneous idiosyncratic volatility, and it predicts future idiosyncratic volatility.
This book is an introduction to the statistical analysis of word frequency distributions, intended for linguists, psycholinguistics, and researchers work ing in the field of quantitative stylistics and anyone interested in quantitative aspects of lexical structure. Word frequency distributions are characterized by very large numbers of rare words. This property leads to strange statisti cal phenomena such as mean frequencies that systematically keep changing as the number of observations is increased, relative frequencies that even in large samples are not fully reliable estimators ofpopulationprobabilities, and model parameters that emerge as functions of the text size. Special statistical techniques for the analysis of distributions with large numbers of rare events can be found in various technical journals. The aim of this book is to make these techniques more accessible for non-specialists. Chapter 1 introduces some basic concepts and notation. Chapter 2 describes non-parametricmethods for the analysis ofword frequency distributions. The next chapterdescribes in detail three parametricmodels, the lognormal model, the Yule-Simon Zipfian model, and the generalized inverse Gauss-Poisson model. Chapter 4 introduces the concept of mixture distributions. Chapter 5 explores the effectofnon-randomness inword use on the accuracy of the non parametric and parametric models, all of which are based on the assumption that words occur independently and randomly in texts. Chapter 6 presents examples of applications.
Even with hindsight, the ability to explain stock price changes is modest. R 2s were calculated for the returns of large stocks as explained by systematic economic influences, by the returns on other stocks in the same industry, and by public firm‐specific news events. The average adjusted R 2 is only about .35 with monthly data and .20 with daily data. There is little relation between explanatory power and either the firm's size or its industry. There is little improvement in R 2 from eliminating all dates surrounding news reports in the financial press. However, the sample kurtosis is quite different when such news events are eliminated, thereby revealing a mixture of return distributions. Non‐news dates also indicate the presence of a distributional mixture, perhaps due to traders acting on private information.
In this study, we measure managerial affective states during earnings conference calls by analyzing conference call audio files using vocal emotion analysis software. We hypothesize and find that when managers are scrutinized by analysts during conference calls, positive and negative affect displayed by managers are informative about the firm's financial future. In particular, we find that managers exhibiting positive (negative) affect are positively (negatively) related to contemporaneous stock returns and future unexpected earnings. However, analysts do not incorporate the information when determining short term earnings forecasts. When making stock recommendation changes, however, analysts incorporate positive affect but not negative affect. We observe market underreaction to negative affect as if market participants follow analyst recommendation changes. Together, this study presents new evidence that managerial vocal cues contain useful information about firms' fundamentals, incremental to both quantitative earnings information and qualitative "soft" information conveyed by the linguistic content. We appreciate the assistance of Amir Liberman and Albert De Vries of Nemesysco for helpful discussions and for providing the LVA software for our academic use. We acknowledge helpful comments and suggestions from for excellent research assistance.
We examine IPO price formation and strategic disclosure by analyzing the word content of 9,818 IPO filings including the initial prospectus as well as each amendment. We find three primary results that motivate extensions to IPO theory. First, disclosure is an important component in price formation as the relative size of the document sections predicts the magnitude of the partial price adjustment, first day IPO returns, and long-run post-offer performance. Second, the writing of the prospectus is collaborative effort involving under-writers, legal counsel, auditors and the issuing firm with different authors per-forming separate functions in the disclosure of information. A key conclusion is that issuing firm managers, through MD&A, play a surprisingly integral role in the bookbuilding process. Third, information generated during bookbuilding is asymmetrically disclosed. Positive information is withheld for strategic or proprietary reasons while negative information is disclosed as a hedge against litigation risk.
We examine the relation between accruals quality and internal controls using 705 firms that disclosed at least one material weakness from August 2002 to November 2005 and find that weaknesses are generally associated with poorly estimated accruals that are not realized as cash flows. Further, we find that this relation between weak internal controls and lower accruals quality is driven by weakness disclosures that relate to overall company-level controls, which may be more difficult to audit around. We find no such relation for more auditable, account-specific weaknesses. We find similar results using four additional measures of accruals quality: discretionary accruals, average accruals quality, historical accounting restatements, and earnings persistence. Our results are robust to the inclusion of firm characteristics that proxy for difficulty in accrual estimation, known determinants of material weaknesses, and corrections for self-selection bias.
We document systematic evidence of risk effects of disclosures culled from a virtually exhaustive set of sources from the print medium. We content analyze more than 100,000 disclosure reports by management, analysts, and news reporters (i.e., financial press) in constructing firm-specific disclosure measures that are quantitative and amenable to replication in future. We expect credibility and timeliness differences in the disclosures by source, which would translate into differential cost of capital effects. We find that when content analysis indicates favorable disclosures, the firm's risk as proxied for by the cost of capital, stock return volatility, and analyst forecast dispersion decline significantly. In contrast, unfavorable disclosures are accompanied by significant increases in risk measures. Analysis of disclosures by source - corporations, analysts, and the financial press - reveals that negative disclosures from financial press sources result in increased cost of capital and return volatility, and favorable reports from financial press reduce the cost of capital and return volatility.
This paper examines the tone and content of the forward-looking statements (FLS) in corporate 10-K and 10-Q filings using a Naive Bayesian machine learning algorithm. I first manually categorize 30,000 sentences of randomly selected FLS extracted from the MD&As along two dimensions: (1) tone (i.e., positive versus negative tone); and (2) content (i.e., profitability, operations, and liquidity etc.). These manually coded sentences are then used as training data in a Naive Bayesian machine learning algorithm to classify the tone and content of about 13 million forward-looking statements from more than 140,000 corporate 10-K and 10-Q MD&As between 1994 and 2007. I find that firms with better current performance, lower accruals, smaller size, lower market-to-book ratio, and less return volatility tend to have more positive forward-looking statements in MD&As. The average tone of the forward-looking statements in a firm's MD&A is positively associated with future earnings and liquidity, even after controlling for other determinants of future performance and there is no systematic change in the information content of MD&As over time. Finally, the evidence indicates that financial analysts do not fully understand the information content of the MD&As in making their forecasts.
Earnings press releases are an important means by which many firms communicate to investors. This study examines whether investors are influenced by how earnings press releases are written - the tone and other stylistic attributes - using actual earnings press releases and archival capital markets data in a standard short-window event study. To measure the tone and other stylistic aspects of press releases, I use elementary computer-based content analysis. Tone is measured using a frequency count of positive or negative words. Other stylistic aspects are the overall length of the press release, the overall percentage of numbers versus words, and the complexity of the words used. Results suggest that tone of earnings press releases influences investors' reactions to earnings. An explanation for this result is provided by prospect theory (Tversky and Kahneman, 1981, 1986), which predicts that framing financial performance in positive terms, will cause investors to think about the results in terms of increases relative to reference points. Limited evidence is presented that other stylistic attributes of earnings press releases affect investors' reactions to earnings.
I examine the role of information processing costs on post earnings announcement drift. I distinguish between hard information - quantitative information that is more easily processed - and soft information which has higher processing costs. I find that qualitative earnings information has additional predictability for asset prices beyond the predictability in quantitative information. I also find that qualitative information has greater predictability for returns at longer horizons, suggesting that frictions in information processing generate price drift. Using a tool from natural language processing called typed dependency parsing, I demonstrate that qualitative information relating to positive fundamentals and future performance is the most difficult information to process.
This study explores whether the Management Discussion and Analysis (MD&A) section of Form 10-Q and 10-K has incremental information content beyond financial measures such as earnings surprises, accruals and operating cash flows (OCF). It uses a well-established classification scheme of words into positive and negative categories to measure the tone change in a specific MD&A section as compared to those of the prior four filings. Our results indicate that short window market reactions around the SEC filing are significantly associated with the tone of the MD&A section, even after controlling for accruals, OCF and earnings surprises. We also show that the tone of the MD&A section adds significantly to portfolio drift returns in the window of two days after the SEC filing date through one day after the subsequent quarter s preliminary earnings announcement, beyond financial information conveyed by accruals, OCF and earnings surprises. The incremental information of tone change is larger the weaker is the firm's information environment.
We measure managerial affective states during earnings conference calls by analyzing conference call audio files using vocal emotion analysis software. We hypothesize and find that, when managers are scrutinized by analysts during conference calls, positive and negative affects displayed by managers are informative about the firm's financial future. Analysts do not incorporate this information when forecasting near-term earnings. When making stock recommendation changes, however, analysts incorporate positive but not negative affect. This study presents new evidence that managerial vocal cues contain useful information about a firm's fundamentals, incremental to both quantitative earnings information and qualitative “soft” information conveyed by linguistic content.
We analyze the information content of the ambient noise level in the Chicago Board of Trade's 30-year Treasury Bond futures trading pit. Controlling for a variety of other variables, including lagged price changes, trading volumes, and news announcements, we find that the sound level conveys information which is highly economically and statistically significant. Specifically, changes in the sound level forecast changes in the cost of transacting. Following a rise in the sound level, prices become more volatile, depth declines, and information asymmetry increases. Our results offer important implications for the future of open outcry and floor-based trading mechanisms.
The proportion of U.S. firms paying dividends drops sharply during the 1980s and 1990s. Among NYSE, AMEX, and Nasdaq firms, the proportion of dividend payers falls from 66.5% in 1978 to only 20.8% in 1999. The decline is due in part to an avalanche of new listings that tilts the population of publicly traded firms toward small firms with low profitability and strong growth opportunities—the timeworn characteristics of firms that typically do not pay dividends. But this is not the whole story. The authors' more striking finding is that, no matter what their characteristics, firms in general have become less likely to pay dividends.
The authors use two different methods to disentangle the effects of changing firm characteristics and changing propensity to pay on the percent of dividend payers. They find that, of the total decline in the proportion of dividend payers since 1978, roughly one-third is due to the changing characteristics of publicly traded firms and two-thirds is due to a reduced propensity to pay dividends. This lower propensity to pay is quite general—dividends have become less common among even large, profitable firms.
Share repurchases jump in the 1980s, and the authors investigate whether repurchases contribute to the declining incidence of dividend payments. It turns out that repurchases are mainly the province of dividend payers, thus leaving the decline in the percent of payers largely unexplained. Instead, the primary effect of repurchases is to increase the already high payouts of cash dividend payers.
This study examines the investor response to Form 10-K and 10-Q reports filed between 1996 and 2001. The samples comprise essentially the entire body of EDGAR filings, including the small business (SB) versions of each filing type. The study documents that the absolute value of excess return is reliably greater on the day of and on the one or two days immediately following the filing date. The response is stronger around a 10-K date than a 10-Q date, more elevated for delayed filers, and increases significantly over the study period for both filing types. A regression analysis indicates that differences in response due to filing delay and year of filing are not subsumed by other attributes of the information environment, such as changes in industry composition, day of week, market capitalization, and shares held by institutions.
This paper examines the relation between annual report readability and firm performance and earnings persistence. I measure the readability of public company annual reports using the Fog index from the computational linguistics literature and the length of the document. I find that: (1) the annual reports of firms with lower earnings are harder to read (i.e., they have a higher Fog index and are longer); and (2) firms with annual reports that are easier to read have more persistent positive earnings.
This paper identifies five common risk factors in the returns on stocks and bonds. There are three stock-market factors: an overall market factor and factors related to firm size and book-to-market equity. There are two bond-market factors, related to maturity and default risks. Stock returns have shared variation due to the stock-market factors, and they are linked to bond returns through shared variation in the bond-market factors. Except for low-grade corporates, the bond-market factors capture the common variation in bond returns. Most important, the five factors seem to explain average returns on stocks and bonds.
Estimates of the cost of equity for industries are imprecise. Standard errors of more than 3.0% per year are typical for both the CAPM and the three-factor model of Fama and French (1993). These large standard errors are the result of(i) uncertainty about true factor risk premiums and (ii) imp ecise estimates of the loadings of industries on the risk factors. Estimates of the cost of equity for firms and projects are surely even less precise.
This paper examines whether the "soft" information contained in the text of management's quarterly earnings press releases is incrementally informative over the company's reported "hard" earnings news. We use Diction, a textual-analysis program, to extract various dimensions of managerial net optimism from more than 20,000 corporate earnings announcements over the period 1998 to 2006 and document that unanticipated net optimism in managers' language affects announcement period abnormal returns and predicts post-earnings announcement drift. We find that it takes longer for the market to understand the implications of soft information than those of hard information. We also find that the market response varies by firm size, turnover, media and analyst coverage, and the extent to which the standard accounting model captures the underlying economics of the firm. We also show that the second moment of soft information, the level of certainty in the text, is an important determinant of contemporaneous idiosyncratic volatility, and it predicts future idiosyncratic volatility.
We examine whether a simple quantitative measure of language can be used to predict individual firms' accounting earnings and stock returns. Our three main findings are: (1) the fraction of negative words in firm-specific news stories forecasts low firm earnings; (2) firms' stock prices briefly underreact to the information embedded in negative words; and (3) the earnings and return predictability from negative words is largest for the stories that focus on fundamentals. Together these findings suggest that linguistic media content captures otherwise hard-to-quantify aspects of firms' fundamentals, which investors quickly incorporate into stock prices. Copyright (c) 2008 by The American Finance Association.
I quantitatively measure the interactions between the media and the stock market using daily content from a popular "Wall Street Journal" column. I find that high media pessimism predicts downward pressure on market prices followed by a reversion to fundamentals, and unusually high or low pessimism predicts high market trading volume. These and similar results are consistent with theoretical models of noise and liquidity traders, and are inconsistent with theories of media content as a proxy for new information about fundamental asset values, as a proxy for market volatility, or as a sideshow with no relationship to asset markets. Copyright 2007 by The American Finance Association.
Financial press reports claim that Internet stock message boards can move markets. We study the effect of more than 1.5 million messages posted on Yahoo! Finance and Raging Bull about the 45 companies in the Dow Jones Industrial Average and the Dow Jones Internet Index. Bullishness is measured using computational linguistics methods. "Wall Street Journal" news stories are used as controls. We find that stock messages help predict market volatility. Their effect on stock returns is statistically significant but economically small. Consistent with Harris and Raviv (1993) , disagreement among the posted messages is associated with increased trading volume. Copyright 2004 by The American Finance Association.
This paper describes a simple method of calculating a heteroskedasticity and autocorrelation consistent covariance matrix that is positive semi-definite by construction. It also establishes consistency of the estimated covariance matrix under fairly general conditions.
Using a sample free of survivor bias, the author demonstrates that common factors in stock returns and investment expenses almost completely explain persistence in equity mutual funds' mean and risk-adjusted returns. Darryll Hendricks, Jayendu Patel, and Richard Zeckhauser's (1993) 'hot hands' result is mostly driven by the one-year momentum effect of Narasimham Jegadeesh and Sheridan Titman (1993), but individual funds do not earn higher returns from following the momentum strategy in stocks. The only significant persistence not explained is concentrated in strong underperformance by the worst-return mutual funds. The results do not support the existence of skilled or informed mutual fund portfolio managers. Copyright 1997 by American Finance Association.
The goal in information retrieval is to enable users to automatically and accurately find data relevant to their queries. One possible approach to this problem is to use the vector space model, which models documents and queries as vectors in the term space. The components of the vectors are determined by the term weighting scheme, a function of the frequencies of the terms in the document or query as well as throughout the collection. We discuss popular term weighting schemes and present several new schemes that offer improved performance. 1. Introduction Automatic information retrieval is needed because of the volume of information available today --- there is too much information to be indexed manually. Most people have used some type of information retrieval system in the form of Internet search engines. Search engines are based on information retrieval models such as the Boolean system, the probabilistic model, or the vector space model [7]. We focus on the vector space model, de...
Modern information retrieval: A brief overview, Working paper, Google, Inc. When Is a Liability Not a Liability? 65 Stone The General Inquirer: A Computer Approach to Content Analysis
- Singhal
- Amit
Singhal, Amit, 2009, Modern information retrieval: A brief overview, Working paper, Google, Inc. When Is a Liability Not a Liability? 65 Stone, Philip J., Dexter C. Dunphy, Marshall S. Smith, and Daniel M. Ogilvie, 1966, The General Inquirer: A Computer Approach to Content Analysis (MIT Press, Cambridge, MA).
Strategic disclosure and the pricing of initial public offerings, Working paper
- Kathleen Hanley
- Gerard Weiss
- Hoberg
Hanley, Kathleen Weiss, and Gerard Hoberg, 2008, Strategic disclosure and the pricing of initial
public offerings, Working paper, University of Maryland.
The Stuff of Thought: Language as a Window into Human Nature, Penguin Group
- Steven Pinker
Pinker, Steven, 2007, The Stuff of Thought: Language as a Window into Human Nature, Penguin
Group, New York, NY.
Yahoo! for Amazon: Opinion extraction from small talk on the web, Working paper
- Sanjiv Das
- Mike Chen
Das, Sanjiv, and Mike Chen, 2001, Yahoo! for Amazon: Opinion extraction from small talk on the
web, Working paper, Santa Clara University.
The Netherlands )CrossRef
- R Baayen
- Harald
- Bernard R Berelson
Berelson, Bernard R., 1952, Content Analysis in Communication Research (The Free Press, Glencoe,
IL).
- See Manning
See Manning and Schütze (2003), Jurafsky and Martin (2009), or Singhal (2009).