Abstract: The presentation focuses on research using text analysis in accounting and finance and involve material from three papers and one research idea using text from quarterly Earnings Conference Calls (ECC)1. These transcripts consist of an introduction, and a Q&A session, where analysts ask questions to the management. They are currently available for download in WRDS via the library. The first paper: Investors and financial analysts have limited processing power and in certain situations, this causes disagreement among investors resulting in high trading volumes (when many investors buy and sell at the same time, there is a disagreement on future firm prospects and value). In this study, we examine how investors' limited information processing capabilities can exacerbate disagreements when faced with divergent inquiries from financial analysts. We argue that the more varied the questions among analysts—in terms of themes, novelty, and tone—the higher the processing costs for investors to understand the information presented. This increase in processing costs can lead to greater disagreement among investors, as evidenced by higher trading volumes. We use text-based proxies such as number of questions, novel content brought up by financial analysts, and theme divergence from the company introduction significantly increases trading volumes (i.e. increased disagreement among investor). The second paper: As noted above, investors and financial analysts have limited processing power and may initially underreact to information releases, resulting in a delayed market reaction. This delay is called Post-Earnings Announcement Drift (PEAD). Traditionally, PEAD is measured as abnormal return in relation to expected return even after the new earnings information has been publicly released. Recent research using computerised text analysis to identify ‘good’ and ‘bad’ news using bigrams to create a measure based on ECC text. The study at hand builds on and extend this study by, for introductions, questions, and answers, identifying themes2 (e.g. market outlook, industry specific issues, financial statement analysis and business outlook) and characteristics of the questions and answers (e.g. topic divergence among introduction and questions and answers , number of questions, specificity, novelty, uncertainty, complexity, and tone). Shapely Additive exPlanations (SHAP) is used to assess the extent to which each text measure influences and predicts PEAD. Thus, the study contributes by identifying specific text content that drives PEAD. The third paper: We examine whether financial analysts request information of sustainable materiality. That is, questions about the environmental impact companies’ actions have made on other stakeholders by affecting ecosystems, climate change, resource depletion, human rights, and social well-being. The reason for investigating this is that one of the major global standard setters suggested that such disclosures should not be required by companies as it is deemed less important for main stakeholders such as investors and creditors. Using generative AI on a sample of US ECCs between 2007 and 2019, we find that only a small fraction of financial analysts’ questions concerns sustainable materiality, but they do exist and appear to increase in the latter part of the sample period. 1 ECCs are transcripts of companies’ meetings with financial analysts after quarterly earnings releases. 2 Themes are created using Latent Dirichlet Allocation (LDA) for topic modelling, validated by researchers. Research idea/future research project: Half-truths are statements that omit important information which changes the interpretation of it. We will use the ECC transcripts to analyse companies’ use of half-truths. Specifically, we will analyse in which situations companies resort to half-truths when communicating with financial analysts. We a priori predict that companies in financial distress will be more likely to resort to half-truths to avoid being penalised by investors. Using LLM, we will identify and classify half-truths using concepts from dialogue semantics and pragmatics, such as implicature, presupposition, enthymeme and topos. There are several studies examining capital market effects of ECC features such as tone, conversational engagement, and linguistic style (Haag et al., 2022; Rennekamp et al., 2022). However, to our knowledge, there is no research on capital market effects of half-truths. We intend to use the above dialogue semantics to create measures of half-truths to detect in which circumstances and when companies resort to using half-truths and their capital market implications.
Location: Attend in person at room J411 or via Zoom, https://gu-se.zoom.us/j/66299274809?pwd=Yjc2ejc2VVhraXVJMmhWeWtOQ2NuUT09
Time: 13:15-15:00