Screen

Fragility screen
hidden securities losses, next to reliance on flight-prone deposits.

Two figures, computed independently for every covered institution: the securities portfolio's own hidden mark-to-market loss (Schedule RC-B fair value below amortized cost, held-to-maturity plus available-for-sale) as a share of book equity, and the already-validated uninsured-deposit ratio. Neither is combined into a single score — this codebase's own rule against an invented weighting between two named mechanisms applies here exactly as it does everywhere else. The list below shows institutions currently at or above their own asset cohort's 75th percentile on both figures at once, as of their latest reported period.

Why these two dimensions, together: this follows Jiang, Matvos, Piskorski & Seru, "Monetary Tightening and U.S. Bank Fragility in 2023" (NBER Working Paper 31782 / Journal of Financial Economics 2024) — the paper explaining the March/May 2023 failures. Backtested live against this database's own data (the three post-2016 failures with the needed schedules — SVB, Signature Bank and First Republic — the same coverage limit this product's CECL backtest already found): all three ran a securities loss equal to a large share of, or in SVB's case exceeding, their entire book equity, alongside an uninsured-deposit ratio near or above 50%, in the quarters before failure. A comparison bank held over the identical 2022–2023 rate-hike window stayed well below both thresholds throughout.

A limitation worth reading before this list: this measures only the securities portfolio. The full published methodology also marks the loan book to market using its implied repricing duration — a real modeling step this product has not built, since doing so credibly would mean introducing an estimation method this codebase has not validated against reported figures. This screen is therefore a lower bound on true balance-sheet fragility, not the complete measure, for any bank whose loan book (not just its securities) lost value to rising rates. There is also no persistence requirement here, unlike the reserve-adequacy screen's four-quarter rule: a securities mark-to-market loss is a precisely reported fact each quarter, not a flow figure with reporting-timing noise to filter, and this product's live backtest covers too few real cases to responsibly calibrate a specific consecutive-quarter threshold.

Running the fragility screen…