Current Research.
My current projects include the following:
“Sticky Charters? The Surprisingly Tepid Embrace of Officer-Protecting Waivers in Delaware”
With Eric Talley
This article investigates the reaction to a much-heralded 2022 legal reform in Delaware that permitted a corporation’s charter to exculpate its officers from monetary exposure for breaching their fiduciary duty of care. To isolate reactions to this statutory reform, we make extensive use of generative AI tools to identify and interpret charter amendments that introduce officer-facing waivers. We find a surprisingly tepid rate of uptake among Delaware corporations through the end of the first post-reform year, notwithstanding widespread predictions that corporate entities would quickly storm the exculpation exits once permitted to do so.
Our study makes two contributions to the empirical study of law—one methodological and the other substantive. Methodologically, we develop a novel and powerful use case for deploying large language models as a tool for distilling and extracting technical provisions from legal texts (in this case corporate charters), allowing us to accelerate and streamline an endeavor that would have consumed substantial time and resources using traditional human-labeling protocols. Notably, and in a significant departure from previous machine learning tools, ChatGPT accomplishes this set of tasks without the need for training data specifically tailored for this purpose. Perhaps most impressive is the accuracy with which ChatGPT can operate: we perform several validation exercises, which generally indicate that our proposed method yields highly accurate results.
Substantively, we demonstrate that Delaware’s statutory invitation attracted few takers in its first year of effectiveness: specifically, we show that only a modest minority of eligible corporations amended their charters to include officer-facing waivers. This tepid rate of uptake, moreover, persists even in corporations that went public after the reform’s effective date, suggesting that transaction costs are unlikely to be the culprit for the listless response. Furthermore, we show that stock market investors also exhibited a muted response to the reform, raising doubts about whether firms feared amendments would trigger an adverse market reception. Our results seem more consistent with alternative explanations, ranging from the plausible irrelevance of Delaware’s reform, to a risk-averse reticence among corporate managers who rationally adopt a “wait and see” approach to gauge how such waivers are received by both courts and corporate stakeholders while keeping their options open.
Contractual Analogies and the Realities of Bylaw Changes
Corporate bylaws are contracts. Or are they? Recently, there has been a significant push by both courts and commentators to equate corporate bylaws to contracts. This perspective is applied not only in interpreting bylaws, where it is well-established that Delaware courts use principles guiding contract interpretation to resolve ambiguities about the meaning and scope of bylaws. It extends to justifying the binding nature of bylaws on shareholders as well. The argument is that bylaws, to which shareholders are assumed to have assented, should bind them just as any contract would. However, this analogy is complicated by the fact that bylaws are routinely updated through unilateral actions by the board of directors. These amendments can introduce novel provisions that significantly impact the mechanisms by which shareholders can hold directors accountable, such as through lawsuits and director elections. Against this background, an important justification for treating bylaws as contracts is what can be described as their dual amendability—both the board of directors and the shareholders have the power to change them. Consequently, it is argued that if shareholders are dissatisfied with the bylaws established by the board, they are free to amend them.
While several commentators have criticized the bylaws-as-contracts analogy, this article is the first to specifically argue against treating the dual amendability of bylaws as an argument to treat bylaws as contracts. In the reality of today’s publicly-traded corporations, it is almost impossible for shareholders in many corporations to effect lasting changes to the corporation’s bylaws. This is not only because of the standard collective action problem that bedevils shareholder action more generally. It is also because, in many corporations, there are provisions in the charter and in the bylaws themselves that substantially increase the requirements for shareholder action beyond what is the norm for other decisions. These provisions, such as supermajority voting requirements or other stringent conditions, create significant barriers that prevent shareholders from effectively counteracting unilateral changes made by the board. To support its argument, the article presents new empirical evidence that illuminates the current realities of bylaw amendments and the rules that govern them.
The data presented in this article are obtained as part of an ongoing data gathering effort that collects information on the complete chartering and bylaw histories of all publicly traded corporations incorporated in Delaware operating anytime since 1995. Information on relevant provisions is obtained from these documents with the help of OpenAI’s GPT-4o model. Due to the prohibitive costs of evaluating the entirety of all documents using this or similar large language models, I implement a “cascading pipeline” reminiscent of the one proposed by Zehua Li, Neel Guha, and Julian Nyarko in a recent contribution: Initially, candidate parts of a charter or bylaw that might contain a relevant provision are identified using a more cost-effective BERT model, significantly reducing the volume of text fed into GPT-4o. GPT-4o is then used for a fine-grained evaluation of the contents of the relevant provisions. The effectiveness of this data gathering approach is validated with the help of human research assistants. As such, this project also contributes to the literature on how novel artificial intelligence technologies can assist in the extraction of legally relevant information from legal texts, an enterprise that has the potential to significantly change the way empirical legal research is conducted.
“DECODEM: Data Extraction from Corporate Organizational Documents via Enhanced Methods”
With Eric Talley and others
This project seeks to revolutionize empirical research in corporate law and finance by laying the groundwork for a new generation of open-source corporate governance datasets, thereby enhancing the power and reach of this type of research. To achieve this goal, the project develops advanced methods for the automated extraction of legally relevant information from corporate charters and bylaws. In particular, it leverages state-of-the-art natural language processing technologies, including large language models, to overcome the limitations of traditional data extraction methods, which often rely on manual coding and are hampered by restricted scope and questionable reliability.
“Measuring Dark Patterns in CCPA Opt-Out Mechanisms”
With Marshini Chetty, Nick Feamster, Lior Strahilevitz and Van Tran
To safeguard consumer privacy, the California Consumer Privacy Act (CCPA) requires businesses to provide consumers with an option to opt out of the sale and sharing of their personal information. However, businesses often implement overly burdensome procedures that make the opt-out process effectively unusable. The California Privacy Rights Act (CPRA) was introduced to strengthen the CCPA and address these shortcomings. Given the recent enactment of the CPRA and its distinction as the first U.S. law targeting the use of dark patterns, the effectiveness of this legislation remains an open question.
In this study, we develop a pipeline to record and analyze how businesses implement their opt-out processes. We completed the entire opt-out process for a number of websites likely subject to the CPRA, which includes submitting an opt-out request via the opt-out link and completing any required verification processes. We characterize observed implementation patterns of the opt-out process and examine dark patterns within these processes. This analysis provides valuable insights into the practical challenges and effectiveness of the CPRA in mitigating manipulative practices that undermine consumer rights.
“Lawma: The Power of Specialization for Legal Tasks”
With Ricardo Dominguez-Olmedo, Vedant Nanda, Rediet Abebe, Stefan Bechtold, Christoph Engel, Krishna Gummadi, Moritz Hardt, and Michael Livermore
Annotation and classification of legal text are central components of empirical legal research. Traditionally, these tasks are often delegated to trained research assistants. Motivated by the advances in language modeling, empirical legal scholars are increasingly turning to prompting commercial models, hoping that it will alleviate the significant cost of human annotation. Despite growing use, our understanding of how to best utilize large language models for legal tasks remains limited. We conduct a comprehensive study of 260 legal text classification tasks, nearly all new to the machine learning community. Starting from GPT-4 as a baseline, we show that it has non-trivial but highly varied zero-shot accuracy, often exhibiting performance that may be insufficient for legal work. We then demonstrate that a lightly fine-tuned Llama-3 model vastly outperforms GPT-4 on almost all tasks, typically by double-digit percentage points. We find that larger models respond better to fine-tuning than smaller models. A few tens to hundreds of examples suffice to achieve high classification accuracy. Notably, we can fine-tune a single model on all 260 tasks simultaneously at a small loss in accuracy relative to having a separate model for each task. Our work points to a viable alternative to the predominant practice of prompting commercial models. For concrete legal tasks with some available labeled data, researchers are better off using a fine-tuned open-source model.