Working paper · May 2026
160 Years of Examination Priorities
What Bank Supervisors Focus On and Why It Changes
Abstract
We construct the first continuous measure of what bank supervisors focus on from 50.7 million words of examination documents spanning 1863 to 2025. Institutional architecture explains virtually all variation in supervisory emphasis; financial crises and political transitions contribute almost none. Formal regulation and supervisory emphasis are complements for codifiable risks but substitutes for judgment-intensive domains, consistent with incomplete contracts. The index tracks lending standards, congressional testimony, charter switching, and cross-regulator acquisition patterns.
Key findings at a glance
The visualizations below present the paper's main results interactively. All time-series charts are powered by the same SPI panel available for download. Regression results are drawn from the paper's tables.
Total supervisory emphasis over 160 years
Total SPI (sum of all domain-specific keyword frequencies per 1,000 words) for each agency by year. The OCC series begins in 1863, the Fed in 1914, and the FDIC in 1934.
What drives supervisory priorities?
Incremental R² from nested models. Agency identity alone explains 42% of variation in domain-specific emphasis. Financial crises and political party each contribute less than 2%.
The codifiability gradient: complements vs. substitutes
Domain-level coefficients from the regulation-supervision nexus regression. When Congress writes rules for codifiable risks (capital, liquidity, credit), supervisors increase emphasis (complements). For judgment-intensive domains (compliance, governance), supervisors pull back (substitutes).
Within-county deposit validation
Same county, same year, different regulator. When a supervisor increases credit emphasis, its banks lose deposits relative to differently regulated banks in the same market. Capital and liquidity emphasis attract deposits. 925,450 bank-county-year observations, 1981–2024.
Explore further
SPI Explorer
Query the full panel interactively. Pick any agency, domain, and year range.
explore →Download the data
The full SPI panel as parquet, CSV, or Stata. Codebook and replication code. CC BY.
download →Cite this paper
BibTeX, APA, and Chicago citation formats for the paper and the dataset.
cite →