Bank for International Settlements – BIS’ Post

Large language models (LLMs) offer powerful tools for analysing unstructured text data, but their full potential requires careful planning, good research design and awareness of the tools’ limitations. In the latest BIS Quarterly Review, Byeungchun Kwon, Taejin Park, Fernando Perez-Cruz, and Phurichai Rungcharoenkitkul provide an accessible introduction to LLMs through a practical primer tailored to economists. Mirroring econometric practices, the primer presents a step-by-step workflow including data preparation, signal extraction, quantitative analysis and outcome evaluation. To illustrate, it applies the workflow using more than 60,000 news articles to study perceived US stock market movers. Throughout, the primer highlights best practices as well as common pitfalls to help researchers make the most of LLMs. Sample code is available on GitHub to facilitate adoption. An online glossary additionally provides succinct technical background on relevant technologies underlying LLMs. Read the full article: https://bit.ly/49pribs #BISQuarterly #MachineLearning #AI