Economic Narratives and Private Lending DecisionsJOB MARKET PAPER
Dissertation Chair: Robert Davidson
Research indicates that economic actors are swayed by prevailing economic narratives (Shiller, 2019). Yet, while investors appear to be influenced by media narratives (e.g., Bybee, Kelly, Su, & Ramadorai, 2023), it is not obvious whether creditors should be similarly affected. Creditors have access to both firm-specific and aggregated private information across their various business lines (Boot, 2000), which gives them an information advantage over typical arms’ length investors. This project examines whether creditors, like equity investors, are influenced by prevailing economic narratives in the context of private lending decisions. Specifically, using the full-text historical archive of the Wall Street Journal, I take advantage of new capabilities in artificial intelligence and open-source language models (OSLMs) to examine whether alignment of firm-specific media narratives with popular narratives influences credit decision-making.
The Importance of Lender-Auditor Relationships in Private LendingWIP
with Andrew Acito
Prior literature has established that banks rely on lending relationships to minimize information asymmetry between lenders and borrowers. New research, relying on institutional-level analysis, finds that banks achieve similar benefits from sharing a common auditor with borrowers either because of familiarity with the auditor’s testing procedures or because the auditor’s relationship with the bank creates stronger incentives to perform well on the borrower’s audit (Francis & Wang, 2020). This work, however, raises broader questions about how relationships between individual bankers and auditors affect lending decisions. We examine whether prior personal relationships between lenders and borrowers’ auditors can influence loan syndicate formation and debt contract design.
Political Polarization, Executives’ Decision-making, and Firm-level Outcomes2ND YEAR PAPER
with Robert Davidson
Given the adverse effects of growing political polarization, we explore whether politically heterogeneous management teams can effectively work together and carry out firm objectives. Specifically, we examine whether political similarity among firms’ top executives impacts the probability of executive turnover and broad measures of firm performance. We find that politically similar management teams are less likely to experience executive turnover. We also find that politically similar management teams have less turnover during periods when the broader political environment is highly polarized.
ChatGPT and Other AI Assistants: A Conversational Guide for Accounting PhD StudentsNOT AI GENERATED
TThe purpose of this guide is to introduce accounting PhD students to the potential opportunities and challenges of using generative AI tools, such as ChatGPT. The first half provides an overview of how these tools work, highlighting limitations, such as potential bias in underlying training data, and risk of hallucination, AI's propensity for making things up. I also underscore data privacy and ethical considerations around proper AI usage. The second half explains how to use generative AI tools, including how to work with APIs, and offers practical strategies and suggestions for a range of use cases relevant to PhD students, including coursework, research, and teaching. While I include helpful links throughout, I conclude with a list of links to additional resources for the curious reader interested in learning more. Given the rapidly evolving AI landscape, my goal is not to be prescriptive, but rather to encourage independent thinking for applying AI tools. Therefore, this guide is intentionally written in terms of first principles so that the reader is ready to experiment with any new AI tools they might encounter. Ultimately, the key recommendation when using AI tools is simple, and merely requires the user do what AI cannot yet do on its own: think critically.
Chat BuilderPERSONAL PROJECT
CostGPT unexpectedly some buzz, so I created Chat Builder, a free tool that lets you set up your own chatbot with the OpenAI Assistants API. It's super easy: you get a Python script from Chat Builder, grab a couple of free online tools, follow some simple steps (all provided for you in the app), and voila! Your chatbot is ready to chat in no time! Best part? You can launch your custom AI chatbot today, for free, no coding needed!
Since OpenAI doesn’t use queries to its API to train its models, this is a great option for those concerned with privacy or copyright issues around lecture materials. For those considering a CustomGPT, while this option would require paying per query, I'm a PhD student and I didn't find it prohibitive to partly subsidize my students when I deployed it just before finals. Moreover, OpenAI recently reduced the pricing for its best performing GPT-4 model, and this option would not require students to spend money on a ChatGPT+ subscription.
For those of you who know your way around Python code, here is the GitHub repo with the template code I use for Chat Builder. The readme file has all the instructions for hosting the chatbot for free on the Streamlit cloud.
I hope you find this tool useful and please let me know how you use it!
