Knowledge that Transforms

Generative artificial intelligence (GAI) will likely alter many aspects of the financial reporting process and spawn a deep stream of academic research. We take an early step by examining the extent to which firms have begun using GAI in one important part of the reporting process: writing disclosures. We begin by evaluating a commercial tool's ability to detect GAI writing in disclosures, and we find that it reliably identifies even very small amounts of GAI usage in realistic samples. We then examine firms’ actual earnings press releases, conference call prepared remarks, risk factors, MD&As, and IPO filings through 2024 and find statistically significant GAI usage in all five disclosure types, with up to 4.5% of new text written by GAI in 2024. Usage is predictably higher in the cross‐section, and filings with higher GAI have systematically different linguistic properties. Our study provides early insights into the use and effects of GAI in financial reporting, and it motivates future research in this evolving area.

We study how foreign investors’ access to corporate information varies with pairwise geopolitical tensions between the investor's and investee's countries. Using a sample of 1,760 country‐pairs, we find that geopolitical tensions between a conference call host firm's country and a foreign country relate negatively with investor participation from that foreign country as well as the quality and tone of management responses in the call, suggesting reduced foreign investor information access with rising tensions. A single‐country analysis of Chinese AB‐share firms produces similar inferences. With higher China–foreign tensions, foreign attendance at private meetings with management declines, and English translations of local language disclosures become less frequent and shorter in length. Curtailment in English translations is more pronounced when local language disclosures contain politically sensitive words, suggesting lower information supply by firm choice. Reduced disclosures from geopolitical tensions are impounded in foreign investors’ asset pricing decisions, resulting in steeper foreign‐share price discounts.

We propose a new measure of firms' disclosure based on company websites, which are widely available and contain a wealth of information. For a sample of U.S. public firms, we construct our disclosure measure using historical website data, validate it by correlating it with extant measures of disclosure and information asymmetry, and explore its determinants. We then apply our measure to the study of U.S. private firms' disclosure and French firms' compliance with a nonfinancial disclosure mandate. Our applications illustrate that our website‐based measure provides a useful complement to extant measures of disclosure, which are more narrowly focused on investors in public capital markets.

We investigate the establishment of public audit oversight bodies (POBs) worldwide. We present descriptive evidence on POBs’ characteristics and factors influencing the timing of their establishment, finding that countries with stronger institutions, civil law traditions, and higher media attention to audit quality adopt POBs faster. While countries may choose similar POB design features, these choices do not align with the factors driving adoption speed. We also explore whether the finding of a positive impact on audit quality of the U.S. PCAOB generalizes to other countries. A difference‐in‐differences analysis over 20 years provides some evidence that POB establishment and their characteristics improve audit quality. However, the results appear sensitive to audit quality measures and research design. Our study offers the first broad‐based investigation of POB adoption and provides important nuance on the relation between POBs and audit quality.

This paper provides early evidence on the integration and impact of generative artificial intelligence (GenAI) in accounting at the accountant and task levels. Using survey data from 277 professional accountants, we document substantial heterogeneity in adoption patterns, perceived benefits, and concerns about GenAI. Using proprietary field data from an AI‐enabled accounting platform serving 79 small‐ and medium‐sized enterprises, we analyze over 200,000 transaction‐level records. We document that GenAI adoption is associated with significant productivity gains and systematic reallocation of effort away from routine data entry toward business communication and quality assurance tasks. GenAI use is also associated with improvements to financial reporting quality, evidenced by more granular ledgers and faster month‐end closing. Examining human–AI interaction, we find that accountants selectively intervene when AI confidence scores are low, consistent with complementarity between professional expertise and AI. A framed field experiment further shows that while AI assistance improves classification accuracy on average, reliance on non‐consensus AI recommendations can increase the risk of error. Overall, our findings highlight both the promise and the risks of GenAI in accounting and suggest that, in practice, AI is most effective as a tool that augments—rather than replaces—professional judgment.

Recent work within accounting and finance has highlighted that modern AI systems exhibit superhuman performance on a variety of foundational activities within these fields. However, the literature often does not provide economic rationale for why AI models seem to outperform, largely because these models are a black box. Through a series of experiments, I set out to open the black box and provide direct evidence on how and why AI models appear to perform so well on accounting and finance‐related tasks. I show that much of the superior performance of AI models can be attributed to artifacts of the modeling itself, rather than to mechanisms grounded in economics. Focusing on two key components of AI models, which may bias inferences in papers that rely on them, I first show that Large Language Model (LLMs) exhibit extremely poor numerical reasoning and thus application in these settings should proceed with caution. Second, I highlight that commercial LLMs suffer from significant look‐ahead bias, which may explain a large portion of their predictive ability in various settings.

We examine whether and how common ownership affects Environmental, Social, and Governance (ESG) ratings—an important research question given the increasing use of these ratings in investment decisions and corporate evaluations. We find that companies with major shareholders in common with the rating agency (“sister firms”) tend to receive higher ESG ratings. When a company becomes a sister firm through a change in the rating agency's ownership structure, its rating from that agency is subsequently upgraded, whereas its ESG ratings from other agencies remain unchanged. Sister firms exhibit greater rating disagreements across agencies than other firms. The higher ESG ratings for sister firms are partly attributable to the transfer of immaterial positive ESG information through common owners. The common ownership effect is more pronounced when the owner can exert a greater influence on the rating agency. Moreover, sister firms with initially elevated ratings demonstrate poorer future ESG performance. Overall, our findings suggest that owners can affect ESG ratings of their portfolio companies in a way consistent with their influence and interest.

This paper examines how market participants trade on private information about firm fundamentals using the largest known case of informed trade of earnings announcements. From 2011 to 2015, a cartel of sophisticated traders illegally obtained early access to and traded on over 1,000 firm earnings announcements. Using this setting, I identify the information in earnings announcements that these market participants found most price relevant. The informed traders preferred announcements with larger earnings and sales surprises relative to forecasts, quantitative managerial guidance, and more extreme news sentiment. Despite their perfect foresight, the traders performed, perhaps surprisingly, poorly relative to hypothetical trading strategies based on comparable foresight. Frictions that limited their performance include price impact, risk aversion, and information processing costs. The trading performance of these informed traders implies that information about firm fundamentals explains little of the cross‐sectional variation in earnings announcement returns, even for sophisticated market participants.

We conduct an experiment to investigate the use of peer evaluations for compensation purposes. Although organizations often rely on peer evaluations for incentive compensation, it is not well understood how peer feedback should be used and designed to ensure non‐distorted evaluations and motivate effort provision. We study peer evaluations in form of bonus allocation proposals, thereby enabling a quantifiable test of our hypothesis. We distinguish between discretionary use (i.e., allocation by the manager) and formulaic use (i.e., allocation by the team via the average) of self‐including and self‐excluding proposals. We find that, relative to self‐including proposals, self‐excluding proposals are less distorted, irrespective of use, but lead to more effort provision only under formulaic use. Under discretionary use, the benefits of self‐excluding proposals are offset, as managerial biases enter bonus allocations. In sum, our findings illustrate benefits of delegating bonus allocations to teams through formulaic use of self‐excluding peer evaluations and extend the understanding of how organizations can effectively incorporate peer evaluations into incentive compensation.

We examine the effect of internal employee coordination on customer trust, focusing specifically on employees’ responsiveness to each other as an important, quantifiable, and objective aspect of internal coordination. Using proprietary data from a company with exogenous assignment of employees to teams that serve individual customers, we study how inter‐employee responsiveness influences customer trust. Each customer is served via an app‐based group chat by a randomly assigned team of employees. Our data include more than 2 million group chat messages with over 16 thousand customers. We find that inter‐employee responsiveness serves as a credible signal that helps build customer trust, as evidenced by their subsequent contracting choices. The effect is more pronounced when the signal is (1) more frequent and (2) more intense. Our findings highlight the novel value of internal employee responsiveness as a credible signal that helps build trust with external stakeholders.