How Much Do Deposits Run Off in Thirty Days? Four-Week Outflows from US Commercial Banks, 1973–2026, Against Basel III Liquidity Coverage Assumptions

The Basel III liquidity coverage ratio (LCR) requires banks to hold liquid assets against assumed 30-day deposit run-offs of 3% to 10% for retail deposits and 25% to 40% for operational and other corporate deposits. I measure actual four-week changes in deposits at US commercial banks from the Federal Reserve's weekly H.8 release, seasonally adjusted, 1973 to October 2026. Since 1990 the worst four-week fall for all banks, net of trend, was 6.45% (ending 10 October 2001, the unwinding of a post-September 11 surge); the 0.1% quantile is 2.79%. For small domestic banks the worst was 4.32%, ending 29 March 2023, when small banks lost 4.20% of deposits in four weeks while large banks lost 1.30%. Weekly deposit growth is negatively autocorrelated (minus 0.37), so the square-root-of-time rule overstates four-week volatility by 22% to 37%; I derive the exact variance of a k-week sum and confirm it by simulation. Weekly growth has excess kurtosis of up to 20. Gaussian quantiles understate the 0.1% tail for all banks and for small banks, and Cornish-Fisher quantiles overstate it by a factor of two to five. I prove that in an equicorrelated system the standard deviation of the aggregate outflow is the single-bank value times the square root of rho plus (1 minus rho) over N, so for assumed correlations of 0.1 to 0.25 aggregate data understate single-bank run risk by a factor of about 2 to 3. The worst aggregate four-week outflows are of the same order as the LCR's retail run-off rates and well below its corporate rates; single-bank runs such as Silicon Valley Bank's, which lost about a quarter of its deposits in one day, exceed all of them.