In progress · A book and a measurement record
The Supervisory Compass
How examiners watched banks, markets, and the spaces between, 1863–2024.
For one hundred and sixty years, the Office of the Comptroller of the Currency, the Federal Reserve, and the FDIC have written down what they thought banking risk was. We read all of it. The result is a panel that lets you watch supervisory attention rise, decay, and migrate across ten risk domains and three agencies, year by year, from the National Banking Acts to the present.
Where to start
Read Chapter 1
Edward Green walks into a bank in Malden in December 1863 and shoots the cashier. Hugh McCulloch's examiners take notes. The cycle begins.
read →Open the Explorer
Pick an agency, a risk domain, and a year range. Watch the supervisory attention curve. The data runs in your browser.
explore →Check a narrative
Type a year, an event, or an actor. See what supervisors at the OCC, Fed, and FDIC actually weighted at the time.
stress-test →Browse findings
Each forensic finding from the book has a permanent page with the live chart, the inference details, and the source documents.
browse →Download the panel
The full SPI panel as parquet, CSV, or Stata. Codebook, source-document index, replication code. CC BY.
download →Teach with it
Slide decks, problem sets, sample syllabi, and standalone classroom cases — one kit per chapter.
teach →What this site is
Most histories of bank supervision are narrative. They name regulators, describe crises, and tell you what the participants believed they were doing. Those histories are necessary, but they cannot be checked against the supervisory record at scale, because nobody had read the supervisory record at scale. We did.
This site is the place where the record can be queried. When a narrative history emphasizes a particular episode — the founding of the FDIC, the S&L crisis, the 2008 collapse, Operation Choke Point — you can come here and see how much supervisory text the agencies of the day actually devoted to it. Sometimes the narrative weighting matches the data. Sometimes it does not. Both outcomes are informative.