Leveraging AI to Identify Small Business Regulations at the Federal, State, and Local Levels

We introduce a novel database of small business exemptions at the 3-digit NAICS industry level for 48 US States in 2023. Our dataset is derived using Large Language Model (LLM) technology, which is validated using more traditional Natural Language Processing (NLP) methods. We demonstrate that this data is correlated with other measures of small business friendliness at the state level. Furthermore, we apply our LLM classifier to the US Code of Federal Regulations and show that the industry pattern of small business regulation at the state level is similar to industry patterns at the federal level. We find that industries with higher levels of small business exemptions exhibit greater economic dynamism (i.e., more establishment and job formation). Finally, we extend the same LLM-based screening approach to municipal ordinances, analyzing 3,241 jurisdictions. This municipal extension shows that local "small business" definitions are substantially more heterogeneous than those used at the state and federal levels, varying widely across jurisdictions and policy domains.