SEC filings
The number that killed Silicon Valley Bank is in its last 10-K, under a tag the bank invented itself
In its FY2022 10-K, filed Feb 24, 2023, SVB FINANCIAL GROUP reports held-to-maturity securities of $91,321m and a fair value for the same book of $76,169m. The difference is -$15,152m, against $16,004m of stockholders’ equity in the same filing: -94.7% of equity. The bank was not hiding the number. It published it.
The fair-value leg carries the standard us-gaap element HeldToMaturitySecuritiesFairValue. The carrying amount does not. Its tag is DebtSecuritiesHeldToMaturityNetOfAllowanceForCreditLosses, and its versionfield equals the filing’s own accession number, which in these datasets is what a company extension looks like: an element the filer minted, not one it drew from the taxonomy. A screen written as version LIKE 'us-gaap%' returns the fair value and nothing to compare it against.
SVB FINANCIAL GROUP, FY2022 10-K, as filed
Consolidated balance-sheet instants only, at the filing’s own period end, in US dollars. The version column is the whole finding: an accession number where a taxonomy name should be.
| Tag | Version | Element | Value, $m |
|---|---|---|---|
| AccumulatedOtherComprehensiveIncomeLossNetOfTax | us-gaap/2022 | Taxonomy element | -$1,911m |
| Assets | us-gaap/2022 | Taxonomy element | $211,793m |
| DebtSecuritiesHeldToMaturityNetOfAllowanceForCreditLosses | 0000719739-23-000021 | Company extension | $91,321m |
| Deposits | us-gaap/2022 | Taxonomy element | $173,109m |
| HeldToMaturitySecuritiesFairValue | us-gaap/2022 | Taxonomy element | $76,169m |
| StockholdersEquity | us-gaap/2022 | Taxonomy element | $16,004m |
Source: SEC Financial Statement Data Sets, US Securities and Exchange Commission, Division of Economic and Risk Analysis. Facts as filed: uncleaned, unrestated, not audited by FinObservatory. Selection: fsds_num facts under accession 0000719739-23-000021, qtrs = 0, uom = USD, ddate = the submission's period, segments and coreg empty. The filing on EDGAR. Methodology
This is a panel, not an anecdote
The same construction, run across every bank and thrift holding-company 10-K in the estate (SIC 6020 to 6036, one filing per company per fiscal year). Each bar is the sum of that year’s held-to-maturity fair values minus the sum of their carrying amounts, divided by the sum of the same filers’ reported stockholders’ equity. Only filers that tag all three are in the bar, and the count of them is in each bar’s tooltip.
The gap runs from -22.5% of aggregate equity in FY2022 (180 filers) to -19.4% in FY2024 (175 filers). In FY2022, 55 of those 180 filers carried a gap worth more than 10% of their own equity and 12 carried one worth more than 25%. In FY2024 the counts are 39 and 7. Across the eleven fiscal years FY2011 to FY2021, the reading furthest from zero is +2.7% of aggregate equity, in FY2020.
Source: SEC Financial Statement Data Sets, US Securities and Exchange Commission, Division of Economic and Risk Analysis. Facts as filed: uncleaned, unrestated, not audited by FinObservatory. Selection: 10-K submissions with SIC 6020-6036 and prevrpt = 0, deduplicated to one filing per company-year by latest filed date; consolidated USD balance-sheet instants at the filing's period end. FY2025 is partial and is drawn faded: it holds 334 filings against 358 in FY2024, 24 fewer, because the dataset's filing window closes Mar 31, 2026. Methodology
FY2022, ranked by the gap against the filer’s own equity
The 8 largest shortfalls among the 180 FY2022 filers for whom all three legs are tagged. Names are as they appear in the SEC submission. These are as-filed values: FinObservatory has not restated them, and an extreme reading can be a filer’s own tagging error rather than a filer’s position.
| Filer | HTM carrying | HTM fair value | Gap | Equity | Gap, % of equity |
|---|---|---|---|---|---|
| SVB FINANCIAL GROUPcustom tag | $91.32bn | $76.17bn | -$15.15bn | $16.00bn | -94.7% |
| BANK OF HAWAII CORP | $5.41bn | $4.62bn | -$798.7m | $1.32bn | -60.6% |
| HORIZON BANCORP INC /IN/ | $2.02bn | $1.68bn | -$341.4m | $677.4m | -50.4% |
| TERRITORIAL BANCORP INC. | $717.8m | $591.1m | -$126.7m | $256.6m | -49.4% |
| BANK OF AMERICA CORP /DE/ | $632.83bn | $524.27bn | -$108.56bn | $273.20bn | -39.7% |
| CSB BANCORP, INC. | $247.4m | $212.0m | -$35.4m | $95.9m | -37.0% |
| FIRST GUARANTY BANCSHARES, INC. | $320.1m | $242.6m | -$77.5m | $235.0m | -33.0% |
| GREENE COUNTY BANCORP INC | $761.9m | $710.5m | -$51.4m | $157.7m | -32.6% |
The full table, every filer, FY2024.
Source: SEC Financial Statement Data Sets, US Securities and Exchange Commission, Division of Economic and Risk Analysis. Facts as filed: uncleaned, unrestated, not audited by FinObservatory. Selection: FY2022 10-K filer-years with a tagged HTM carrying amount, a tagged consolidated HTM fair value, and positive reported stockholders' equity, sorted ascending by gap as a percent of equity. Methodology
Accumulated other comprehensive income
Accumulated other comprehensive income sits inside equity. It is not the held-to-maturity gap, and the two must never be added: AOCI bundles available-for-sale marks with cash-flow-hedge marks, pension and foreign-currency translation, all net of tax. The bars below rest on a different filer sample from the gap bars above, and the count behind each bar is in its own tooltip.
At FY2022, aggregate AOCI is -14.0% of the aggregate reported equity of the 358 filers that tag both, against a held-to-maturity gap of -22.5% across the 180 filers that tag all three. Those are two different samples measuring two different things.
Source: SEC Financial Statement Data Sets, US Securities and Exchange Commission, Division of Economic and Risk Analysis. Facts as filed: uncleaned, unrestated, not audited by FinObservatory. Selection: AccumulatedOtherComprehensiveIncomeLossNetOfTax and StockholdersEquity (or the including-noncontrolling-interests element where a filer tags only that), consolidated USD instants at the filing's period end. Methodology
What fraction of filers can this be computed for
A gap needs three tagged facts. Absence of any one of them is reported here as absence, never as a zero and never as a claim about what the filer holds.
| Fiscal year | 10-K filers | Tag a carrying amount | Of those, no fair value | No carrying amount found | Gap computable | Carrying leg is a custom tag |
|---|---|---|---|---|---|---|
| FY2025partial | 334 | 216 | 56 | 118 | 159 | 5 |
| FY2024 | 358 | 232 | 57 | 126 | 175 | 6 |
| FY2023 | 361 | 235 | 55 | 126 | 180 | 8 |
| FY2022 | 370 | 241 | 60 | 129 | 180 | 12 |
| FY2021 | 389 | 205 | 48 | 184 | 156 | 15 |
| FY2020 | 395 | 189 | 44 | 206 | 145 | 24 |
In FY2024, 126 of the 358 bank 10-K filers tag no held-to-maturity carrying amount under any of the tag names this module looks for, and a further 57 tag a carrying amount but no consolidated fair value for it. These datasets cannot tell you which of those filers hold no held-to-maturity portfolio and which hold one they did not tag at this level, so this module makes no claim about either. 6 of the 232 FY2024 carrying amounts sit under a company-extension tag; in FY2022 it was 12 of 241.
Source: SEC Financial Statement Data Sets, US Securities and Exchange Commission, Division of Economic and Risk Analysis. Facts as filed: uncleaned, unrestated, not audited by FinObservatory. Selection: counts of bank and thrift 10-K filer-years by which of the three legs each filing tags on the consolidated face. Methodology
What this cannot tell you
- It is not a bank. A CIK is a holding company as consolidated in an SEC filing. The FDIC call reports behind /banks and /deposits are bank-subsidiary level, a different entity on a different consolidation. Nothing on this page should be joined to them or compared with their totals.
- A missing tag is not a zero. Every unfilled cell in this module is n/a. A filing whose note-level detail is absent (the SEC flags this per submission) can look like a non-discloser when it is only a tagging-depth artefact.
- The values are as filed. No restatement, no sign correction, no scale correction. A filer’s own error is reproduced here as the filer made it, which is the only honest way to show a dataset whose subject is tagging quality.
- An HTM gap is not a loss and not a forecast. Held to maturity, the securities pay par. The gap becomes real only if the book has to be sold, which is a funding question this module does not touch. It is a measure of how far the balance sheet’s stated equity sits from a mark-to-market equity, nothing more.
- The estate is financial filers only. All 102,023 submissions here carry an SIC code in 6000-6999, across 4,982 companies. There is no non-financial filer in this data, so no comparison to SEC filers at large can be made from it, and none is made.
The full HTM table
Every one of the 358 FY2024 bank and thrift 10-K filers, both legs where both are tagged and n/a where they are not, plus the year-by-year panel.
The comparability tax
Why the screen missed SVB: 18 us-gaap tags are reported by 95% or more of FY2024 bank 10-K filers and 4,713 company-extension names appear in exactly one filing.
Methodology
Every construction rule, the curated tag lists in full, and the traps in FSDS that make a naive query wrong rather than empty.
What this adds to /banks
/banks scores FDIC call reports and says on its own composite card that public ratios rated Silicon Valley Bank above average the quarter before it failed, because unrealised long-duration securities losses are invisible in those fields. That is right: the call-report table in this estate has 17 columns and neither an AOCI figure nor a held-to-maturity fair value is among them, the FDIC financials table has 38 and carries neither, and the holding-company financials table stops at 2021Q1. The cause is not invisible in XBRL. It is in the 10-K, in the filer’s own words, in a vocabulary the filer partly invented. That is the gap this module fills, and the only one it claims to.
Coverage check run against the estate’s own schemas: call_reports (17 columns), fdic_financials (38), bhc_financials (10 columns, latest report date 2021-03-31).
Estate vintage: 51,057,403 XBRL facts across 102,023 submissions, the last of which was filed Mar 31, 2026 (SEC quarterly dataset 2026q1). 9.9% of all facts in the estate are company extensions.