Regulatory Exposure and Corporate Heterogeneous Adaptation: Evidence from LLM-Based Analysis of Federal Regulations and SEC Filing
Abstract
We develop a novel framework for measuring regulatory exposure by leveraging large language models (LLMs) and retrieval-augmented generation (RAG). We construct two complementary exposure metrics: perceived exposure, derived from firms' explicit references to federal regulations in their public filings, and objective exposure, measured as the semantic similarity between firms' business descriptions and the full text of contemporaneous Federal Register rules. Critically, we employ RAG to reconstruct the historical lineage of each regulatory strand, linking new rules to their statutory and administrative antecedents. Thereby our measurement captures the cumulative context that determines firm relevance to the regulation. Applied to U.S. public firms (2000-2025), the framework reveals substantial within-industry heterogeneity and time-varying exposure patterns. We then show that firms adapt heterogeneously: some reconfigure core business strategies in response to objective exposure, while others intensify issue-specific lobbying aligned with their perceived regulatory threats. This LLM- and RAG-driven approach offers a scalable, theory-grounded methodology for studying how firms navigate, respond to, and shape the evolving regulatory state.
