Bank health / Fragility / Methodology

Fragility map methodology

Every number on the fragility map is computed from the FFIEC Central Data Repository call-report bulk archive (2001Q1-2026Q1, 101 quarters, 666,701 bank-quarters), at build time, from the fields below. Nothing is hand-entered and nothing is modeled: unrealized losses are the difference between two figures every bank itself files each quarter.

What this page is NOT

• Not SRISK and not LRMES. Those are market-data constructs (expected capital shortfall conditional on a market crash, from equity prices and correlations). This page uses no market prices at all, only accounting figures banks file themselves. Calling it SRISK would be wrong in both directions.
• Not a market-value stress test. No scenario is applied. The losses shown have already happened at filing-date market prices; the question the map asks is only who holds them, against how much capital, with how much runnable funding.
• No loan-book marking. Jiang, Matvos, Piskorski and Seru mark loan books to market too, and for good reason: First Republic's duration losses were mostly in loans, and this page's securities-only lens shows it at 29.9% of tier 1 at 2022Q4, far less alarming than SVB's 104.1%. The securities leg is shown alone because it is exactly observable (both cost and fair value are filed); marking loans requires a model, and this page publishes no modeled values.

Field-by-field MDRM mapping

Codes verified against the item-description row inside the bulk files themselves, in both the 2005Q1 and 2023Q1 archives (descriptions shown verbatim as filed). RCFD is the consolidated value, RCON domestic offices; the consolidated value is preferred per bank where both exist.

Derived values: unrealized HTM = fair value minus amortized cost (1771 - 1754); unrealized AFS likewise (1773 - 1772); the headline ratio divides their sum by tier 1 capital, falling back to total equity for the small share of bank-quarters without a tier 1 figure. Negative means loss. All amounts are thousands of USD as filed.

The uninsured-deposit coverage caveat

RC-O Memorandum 2 (RCON5597, estimated uninsured deposits) is completed only by banks with $1 billion or more in assets. In 2026Q1 that is 1,011 banks, which nonetheless hold 95.4% of system deposits, so the reported figure covers almost all deposit dollars while covering a minority of banks. The remaining 3,270 banks carry a proxy, flagged as such everywhere it appears: amount in deposit accounts above the insurance limit minus the number of such accounts times the limit (RC-O Memorandum 1; $250,000-basis items RCONF051/F052 in the modern archive, $100,000-basis items RCON2710/2722 in the pre-2010 era, the era detected per quarter from which items the file carries). The proxy is a lower bound: it removes one insured slice per account and cannot see ownership categories, so it understates uninsured balances relative to a bank's own estimate.

Foreign-office deposits (RCFN2200) are not FDIC-insured and RC-O covers domestic offices only, so the headline uninsured share adds foreign-office deposits to the numerator: (RCON5597 + RCFN2200) / (RCON2200 + RCFN2200). This is the definition under which Silicon Valley Bank's 2022Q4 uninsured share is 94.4%, consistent with the figures reported after its failure. A domestic-only variant (RCON5597 / RCON2200) ships in the dataset alongside.

Anchors against published figures

• System aggregate: this panel sums to $688.2B of unrealized losses in 2022Q3 (HTM $368.0B, AFS $320.3B) against the $689.9B the FDIC QBP press release, 2022Q3 reports (HTM $368.5B, AFS $321.5B), a gap of 0.2%. For 2023Q1 this panel sums to $515.5B against the $515.5B in the FDIC's 2023Q1 release. The residual gap is a universe difference: the QBP covers FDIC-insured institutions, while the call-report archive also includes a small number of non-insured filers and the two counts differ by a few dozen institutions in any quarter.
• Single bank: Silicon Valley Bank's 2022Q4 call report yields $15.2B of unrealized HTM losses here, against the $15.2B net unrealized loss on HTM securities disclosed in its FY2022 Form 10-K (amortized cost $91.3B). First Republic and Signature reconcile the same way; the exact figures are in the build receipts.

Thresholds and universe

• The danger quadrant (loss beyond half, or all, of tier 1 while uninsured funding exceeds half of deposits) is a reporting convention chosen for legibility, not an estimated run threshold. Sensitivity to the choice is visible directly: both loss thresholds are drawn on the chart.
• The universe is every bank filing a call report (FFIEC 031/041/051), so pure holding-company figures never enter; SVB here is the bank subsidiary, not SVB Financial Group, which is why its 10-K figures match to a tenth of a billion but not exactly.
• The uninsured share can marginally exceed 100% for custodial banks (State Street files an uninsured-deposit estimate slightly above its domestic deposits, because RC-O Memorandum 2 includes accrued unpaid interest that the total-deposit line excludes). Values are reported as filed; the scatter clamps its axis at 100% for drawing only.
• AFS unrealized losses already sit in equity through AOCI for most banks (and flow to regulatory capital only for the largest); HTM losses sit in neither. The ratio shown deliberately puts both against tier 1 to answer a single question: what happens to capital if the whole securities book had to be sold at filing-date prices.

Source data: FFIEC Central Data Repository bulk download, all schedules, 2001Q1-2026Q1. cdr.ffiec.gov. Back to the fragility map.