Ties de Kok

Business complexity involves important tradeoffs for managers and investors, but empirical evidence is limited by measurement issues. We construct and validate a measure of business complexity using a GPT model fine-tuned on narrative disclosures and inline XBRL tags. We first show that our measure is associated with slower price formation in capital markets, consistent with complexity increasing processing costs. Next, we apply our measure to study the complexity of debt, an economically important topic that encompasses a wide range of features. The results show that non-standard debt features such as call and convertibility provisions underlie debt complexity. We also find that debt complexity correlates with more persistent interest expense and better performance when lending conditions worsen, suggesting it is in part an adaptive response to manage financial risk. Overall, our study underscores the tradeoffs of business complexity and provides a flexible measure of complexity for future research.

We use direct measures of information acquisition from a business intelligence service to provide descriptive evidence on small business managers’ information choices. We show that managers are heavy consumers of email pushes and dashboards but exhibit substantial time-series and cross-sectional variation in use. The information managers acquire largely facilitates general performance monitoring; product, inventory, and staff management tools are also widely used. Both economic and psychological factors seem to motivate managers’ information choices. Cross-sectional analyses provide support for economic motivators such as business complexity and within-store analyses raise the possibility that hedonic motivations matter—i.e., managers appear to consume information, at least in part, based on their expectation of how the information will make them feel. Still, on net, information acquisition appears productive, as greater acquisition predicts higher future sales and fewer future stockouts. Overall, our study provides micro-level evidence on managerial information acquisition, which is important in part to inform the design and use of increasingly pervasive data products such as dashboards.

Generative Large Language Models (GLLMs), such as ChatGPT and GPT-4 by OpenAI, are emerging as powerful tools for textual analysis tasks in accounting research. GLLMs can solve any textual analysis task solvable using non-generative methods, as well as tasks previously only solvable using human coding. While new and powerful, GLLMs also come with limitations and present new challenges that require care and due diligence. This paper highlights the applications of GLLMs for accounting research and compares them to existing methods. It also provides a framework on how to effectively use GLLMs by addressing key considerations such as model selection, prompt engineering, and ensuring construct validity. In a case study, I demonstrate the capabilities of GLLMs by detecting non-answers in earnings conference calls, a traditionally challenging task to automate. The new GPT method achieves an accuracy of 96% and reduces the non-answer error rate by 70% relative to the existing Gow et al. (2021) method. Finally, I discuss the importance of addressing bias, replicability, and data sharing concerns when using GLLMs. Taken together, this paper provides researchers, reviewers, and editors with the knowledge and tools to effectively use and evaluate GLLMs for academic research.

The timely flow of financial information is critical for efficient capital market functioning, yet we have little understanding of firms' and auditors' collective abilities to maintain timely financial reporting while under duress. We use COVID as a stress test case to examine whether reporting systems can withstand systemic increases in complex economic events and coordination challenges. Despite COVID-related challenges persisting through 2020 and beyond, we document surprisingly modest average delays in financial reports during COVID, and only in Q1-2020. Reporting timeliness reverts to pre-COVID levels no later than Q2-2020. We find no evidence of meaningful declines in actual reporting quality during COVID, but we do find some evidence consistent with declines in perceived reporting quality. Overall, our findings indicate that current financial reporting processes are remarkably robust, and provide insights about financial reporting more broadly. In particular, given that nearly all firms were able to weather the unprecedented disruptions caused by COVID, our findings imply that most material reporting delays observed outside of COVID are likely due to either a firm's strategic choices or to exceptionally fragile reporting processes.

This study evaluates EBITDA as a financial performance measure and investigates the use of EBITDA in financial reporting. First, we take issue with recent comments that both the SEC and the IASB have levied against non-GAAP earnings numbers, and in particular EBITDA. While EBITDA allegedly provides an accurate reflection of the operations and abstracts from how assets are financed, we argue that (net) operating profit already provides this information without the necessity of making subjective adjustments. Also, our evidence suggests that EBITDA paints a rosy picture of the firm's profitability and cash-generating ability. Next, using textual analysis, we investigate the prevalence of EBITDA in financial disclosures based on a large sample of 15,895 annual reports and 51,758 earnings releases from S&P 1500 firms between 2005 and 2016. We find that 14.8% of sample firms disclose and emphasize EBITDA numbers. EBITDA disclosures modestly increase over time and tend to be rather sticky in nature. In our cross-sectional analyses, we find that, consistent with our hypotheses, EBITDA-reporting firms are smaller, more leveraged, more capital-intensive, less profitable and have longer operating cycles than non-EBITDA reporting firms. They also exhibit higher forecast errors and a higher likelihood of missing the analyst forecast benchmark. Additional tests further underscore the opportunistic nature of EBITDA disclosures as we find that these firm characteristics are more strongly associated with the disclosure of adjusted EBITDA measures, and less strongly associated with the disclosure of EBITA and EBIT.

With social media platforms such as Twitter, firms can easily engage with public discourse, i.e., public conversations involving many stakeholders and topics ranging from #TacoTuesday to #StopAsianHate. We ask whether and how investors respond to corporate participation in public discourse. Using various measures of investor attention (Twitter cashtag mentions (e.g., $CRM), Reddit mentions, retail trading, and total trading volume), we find greater investor attention when firms engage with active public discourse with messages that garner high online user impressions. We also find some evidence that when firms engage, they are more likely to face online scrutiny and more extreme market outcomes, especially when engaging with polarizing topics. Finally, when we consider investor preference alignment, we find that ESG fund ownership increases after firms engage meaningfully with environmental trending topics, even controlling for general environmental tweets. These findings broaden our understanding of firm communication and investor response to firms speaking out.

I study the role of product development updates and funder involvement in improving reward crowdfunding outcomes. Reward crowdfunding is a growing source of funding for entrepreneurs with limited access to traditional financing. The limited regulation in reward crowdfunding markets, however, leaves it up to the entrepreneur and the funders to solve their agency problems. Using data from one of the largest reward crowdfunding platforms, I find that entrepreneurs who provide less frequent voluntary updates experience more involvement by funders in the product development process. The entrepreneur responds to this heightened involvement by adjusting their announcement length but not their update frequency. Finally, I find that funders who are responsive in their product development involvement improve project outcomes. Taken together, my results support that voluntary updating and funder involvement can work together to resolve agency frictions in reward crowdfunding markets.

This paper studies whether increasing supervision impairs a loan officer's motivation to produce and share information. We first exploit an organizational change at a large European bank and find that centralization improves the use of soft information in evaluating the creditworthiness of small-to-medium size loan applicants. Next, we conduct a lab experiment and find that participants are more likely to accept an increase in supervision when it is motivated by changes in the external environment of the organization. Taken together, our study informs financial institutions, regulators, and academics on the potential benefits of centralization and provides insights on how to achieve employee acceptance when introducing a new supervision policy.