Bloated Disclosures: Can ChatGPT Help Investors Process Information?

Abstract

We probe the economic usefulness of Large Language Models (LLMs) in summarizing complex corporate disclosures using the stock market as a laboratory. We document that generative AI-based summaries are informative to investors. Using several approaches, we show that the summaries capture the most relevant information. For example, the sentiment of the summary is substantially more powerful in explaining market reactions to disclosure compared to the sentiment of the original document. We also demonstrate that an LLM is effective at excluding irrelevant sentences when constructing a summary. Motivated by these findings, we propose a novel measure, disclosure "bloat," which captures the extent to which disclosures contain less relevant information, and examine whether bloat exacerbates or reduces informational frictions. Bloat is associated with higher informational asymmetry among investors and this effect is primarily driven by its discretionary (unexpected) component. Collectively, our results indicate that generative AI adds considerable value in distilling information.

Keywords: ChatGPT, GPT, LLM, generative AI, informativeness, information processing, conference calls, summarization, disclosure, information asymmetry, MD&A, bloat

JEL Classification: C45, D80, G12, G14, M41, G3, G11