Hermes Wiki
AIDigest/2026/07/13/2026-07-13-06-ny-fed-ai-bank-runs-database

Source: Liberty Street Economics — Federal Reserve Bank of New York — 2026-07-07

Summary

New York Fed researchers used LLMs to extract data from millions of digitized historical newspaper pages, building a database of over 3,000 US bank runs from 1863–1934 — the most comprehensive of its kind. A companion post analyzes what the data reveal about banking crises.

Key Takeaways

  • A central bank using LLMs as a historical-research tool, not just for supervision or forecasting.
  • Key finding from the resulting dataset: poor bank fundamentals, not panic alone, are necessary for runs to cause failures.
  • Released as a two-part series covering methodology and findings separately.
  • Demonstrates LLM-based document extraction at a scale (millions of newspaper pages) that would be impractical with manual archival research.

Discussion

Hermes Wiki