Skip to content
FinObservatory

Bank health / Fragility

The rate-shock fragility map

When rates rise, the bonds banks bought at low yields lose value. Call reports record that loss exactly: Schedule RC-B files both amortized cost and fair value for every bank's securities, held-to-maturity and available-for-sale. This page computes it for every US bank in every quarter since 2001 (666,701 bank-quarters) and puts it against the share of each bank's deposits that are uninsured, the funding that can leave overnight. The FDIC publishes only the aggregate; the joint distribution below is what decided which banks were fragile in March 2023.

$325B
Unrealized securities losses, 2026Q1
$688B
Peak, 2022Q3
0.2%
Assets at banks with loss > half of tier 1, 2026Q1
3.9%
Danger-quadrant peak (2022Q3)

Source: FFIEC Central Data Repository, call report bulk data (Schedules RC, RC-B, RC-O, RC-R) Panel 2001Q1-2026Q1; all dollar amounts as filed, thousands of USD; losses shown in billions. Methodology

Where every large bank sits, 2026Q1

The 150 largest banks by assets. Right means more uninsured funding; down means deeper unrealized securities losses relative to tier 1 capital; point area scales with assets. The amber points are Silicon Valley Bank, Signature Bank and First Republic Bank at 2022Q4, the last call report each filed before failing, computed from the same panel. SVB sat at -104.1% of tier 1 with 94.4% uninsured funding.

Source: FFIEC Central Data Repository, call report bulk data (Schedules RC, RC-B, RC-O, RC-R) Uninsured share is RC-O Memorandum 2 plus foreign-office deposits where reported (1011 banks holding 95.4% of system deposits in 2026Q1); smaller banks carry an account-size proxy, marked in the tooltip. See the methodology for both definitions. Methodology

The danger quadrant through time

Share of banking-system assets held by banks whose unrealized securities loss exceeds half (blue) or all (red) of tier 1 capital while more than half of their deposits are uninsured. The 2022-2023 tightening put 3.9% of system assets in the quadrant at its 2022Q3 peak, against 2.9% at the 2008Q4 peak of the global financial crisis, a credit crisis in which securities portfolios were not the problem. By 2026Q1 the quadrant holds 0.01% of system assets.

Loss > half of tier 1Loss > all of tier 1

Source: FFIEC Central Data Repository, call report bulk data (Schedules RC, RC-B, RC-O, RC-R) Thresholds are reporting conventions for this page, stated in full on the methodology page, not model output. Tier 1 capital from Schedule RC-R; total equity substitutes where tier 1 is not reported. Methodology

Why the interaction is the story

An unrealized loss alone is not distress. Drechsler, Savov, Schnabl and Wang show why: the deposit franchise, the ability to pay depositors less than market rates, is a real asset that gains value exactly when rates rise, hedging the duration loss on the bond book. But that hedge exists only while depositors stay. Uninsured depositors are the ones who leave, so a bank funded by them holds the duration loss without the offset, and expecting others to run makes running rational. Jiang, Matvos, Piskorski and Seru mark bank assets to market and show the 2023 failures were exactly the banks deep in this page's danger quadrant. First Republic filed 2022Q4 with unrealized securities losses of 29.9% of tier 1, modest beside SVB's 104.1%, and still failed: its duration losses sat in the loan book, which this page deliberately does not mark. DeMarzo, Krishnamurthy and Nagel supply the counterpoint: sticky deposits do not imply negative duration, because running the franchise carries fixed operating costs and the franchise is worthless if depositors leave, so its value can fall exactly when rates rise. On their accounting the hedge is weaker even for banks whose depositors stay, and a map like this one understates fragility rather than overstating it.

What this page is and is not (accounting-based fragility, not SRISK, no loan-book marking), the field-by-field MDRM mapping, the $1 billion RC-O coverage caveat and the proxy construction: fragility methodology.