AI Academic Insights
Discover how S&P Global data is powering AI research at top universities.
Artificial Intelligence is rapidly transforming financial analysis, providing innovative ways to interpret complex data and assess risks with precision. On this page, explore pioneering research from institutions like Booth School of Business and the University of Manchester, that uses S&P Global’s datasets to explore the potential of generative AI models. These videos delve into how AI tackles real financial challenges: from summarizing complex disclosures to pinpointing nuanced risks and forecasting market trends.
Can ChatGPT Help Investors Process Information?
Alex Kim from the University of Chicago Booth School of Business explores whether large language models like ChatGPT can distill complex narrative financial disclosures to assist investors. With a focus on identifying “disclosure bloat,” the study analyzes how summarizing lengthy content impacts market behavior. Using GPT-3.5 to generate concise summaries of SEC filings and conference call transcripts, the research shows that these summaries outperform original documents in explaining stock return movements. The findings highlight ChatGPT’s potential to amplify relevant sentiments, making information clearer and more impactful for market participants.
Read the research paper on SSRN
S&P Global data used to conduct this research:
Machine-readable transcripts
Uncovering Corporate Risk Using Generative AI
Maximilian Muhn from Booth School of Business discusses how generative AI models, like ChatGPT, assess corporate risks using conference call transcripts. This research focuses on political, climate, and AI-related risks, highlighting the advantages of AI over traditional dictionary-based approaches. By analyzing 70,000 transcripts from S&P Global’s database, the study finds that AI measures outperform previous methods in correlating with market volatility. Additionally, firms exposed to higher risks show behavior changes, such as investment cutbacks and increased lobbying. The study underscores how generative AI can provide deeper, more nuanced risk assessments, paving the way for enhanced financial decision-making.
Read the research paper on SSRN
S&P Global data used to conduct this research
Machine-readable transcripts
Financial Statement Analysis Using Large Language Models
Professor Valeri Nikolaev from the University of Chicago Booth School of Business examines GPT-4’s ability to conduct financial statement analysis. The research tests GPT-4 against professional analysts in predicting a company’s future performance using financial data. While GPT-4 alone shows modest accuracy, using a “chain of thought” reasoning significantly boosts results, even surpassing human analysts. This study emphasizes the potential for AI and humans to complement each other, showing that combined approaches yield the best outcomes. GPT-4’s performance against specialized AI benchmarks also reveals that its analytical reasoning closely mimics human expertise in tackling complex financial tasks.
Read the research paper on SSRN
S&P Global data used to conduct this research:
Compustat® Financials
Re-Visiting Large Language Models in Finance
Eghbal Rahimikia from the University of Manchester presents research on a suite of year-specific large language models (LLMs) developed for accounting and finance. Leveraging a range of data including S&P Global’s Key Developments dataset, these models are trained on historical data from 2007 to 2023 to eliminate look-ahead bias, a common limitation in general-purpose LLMs. The study demonstrates that these specialized models consistently outperform larger models, including LLaMA versions 1 through 3, across trading scenarios. Rigorous testing further validates their superior performance, highlighting the impact of tailored AI in advancing financial analysis.
Read the research paper on SSRN
S&P Global data used to conduct this research:
Key Developments