Researchers have identified more than 3,000 bank run episodes between 1863 and 1934 using artificial intelligence to scan millions of historical newspaper articles, revealing that the majority of bank runs didn't actually result in bank failure. The new study, published by economists at the Federal Reserve Bank of New York, Federal Reserve Bank of Richmond, and MIT Sloan School of Management on July 7, 2026, creates the most comprehensive database of bank runs in U.S. history by extracting information from over 374 million digitized newspaper pages. The findings challenge conventional assumptions about bank runs, showing that many distressed banks survived without closing their doors.
The data reveals a striking split: researchers found 1,906 bank runs that didn't involve failure compared to 1,515 runs that did end in failure. Overall, the database contains 13,069 episodes involving a suspension and 10,330 episodes involving a failure during the 1863-1934 period. The run rate spiked dramatically during major crisis years—1873, 1884, 1893, 1907, and the Great Depression—with runs without failure especially pronounced during these systemic panics. The Panic of 1893 featured a particularly large number of runs without failure but with temporary suspensions, while runs with failure peaked during the Panic of 1893 and the Great Depression. The geographic center of banking distress shifted over time, starting concentrated along the Eastern Seaboard in cities like New York and Philadelphia, then spreading westward by the 1890s to the Great Plains, Mountain West, and Pacific Coast as the banking system expanded into agricultural and mining frontiers.
The researchers built their dataset by feeding roughly one million "candidate articles" mentioning banks and financial distress keywords into large language models, which then extracted structured information about what happened, when, and why. The study explains that the LLMs can understand nineteenth-century newspaper prose in context—recognizing when an article describes depositors "clamoring for their money" without using the word "run," and filtering out false positives like articles about ships running into riverbanks. The authors note that historical regulatory filings documented when banks closed but "do not indicate whether a run occurred before failure," and there are no systematic records of banks that survived runs without closing. Every identified episode is documented at www.finhist.com/bank-runs, where users can browse by state, city, year, or bank type and read the original newspaper articles.
The database reveals patterns that were previously hidden in the historical record. Systemic crises appeared as nationwide waves—the Panic of 1873 clustered first in New York City, then radiated to Chicago, St. Louis, New Orleans, and smaller cities, while the 1893 crisis originated in the interior with more runs in the West and South tied to distress in railroads, silver mines, and agriculture. The research also uncovers regional episodes that never reached national panic status: the 1920s, remembered as prosperous, showed a steady stream of bank failures in the agricultural Midwest and South, foreshadowing the Great Depression's surge. The authors say this new database "opens the door to answering a range of questions that were previously difficult to study," particularly about when and why runs historically led to failure versus survival—insights that can inform modern policy discussions about financial stability.