Method & limits

Who publishes this

Bankfailures.fyi is an independent, non-commercial project run by one person. It sells nothing, takes no advertising, accepts no payment from any financial institution, and has no relationship with the FDIC or any bank named on it. It is not a rating agency and its output is not a rating.

It is not financial advice. It is a published calculation from public filings, of the kind any reader could repeat — and this page exists so that they can.

The formula, field by field

The Texas Ratio compares the assets a bank is having trouble collecting on against the capital it holds to absorb losses. Bank analysts developed it during the 1980s Texas banking crisis. We calculate it as:

(NALNLS + P9LNLS + ORE)
÷ ((EQ − INTAN) + LNATRES)

Every term is a field on the FDIC's public BankFind financials endpoint, taken from the bank's most recent quarterly Call Report. All are reported in thousands of dollars and multiplied by 1,000 for display.

On preferred stock. Because EQ includes preferred, the cushion is slightly overstated for any bank that has some, and its ratio reads a little low. Among the 500 largest banks only 24 hold preferred at charter level — it is nearly always issued at the holding company instead — and excluding it does not move any of them into a different band. The effect is real but small, and we would rather state it than imply a precision we do not have.

If any component is missing from a bank's filing, no ratio is calculated and the bank shows as N/A. It is never treated as zero — a missing numerator would otherwise render an unknown bank as perfectly healthy. In practice this is almost always a US branch of a foreign bank, which files a different schedule with no equity line of its own.

There is no ratio either when the denominator is zero or negative — when tangible equity plus reserves has been wiped out. Arithmetically that looks the same as missing data, and it means the opposite, so we show it differently: as No cushion, not N/A. Citizens Bank of Sac City looked exactly like this on its final filing before the FDIC closed it in 2023.

The still-accruing check

This is the one place where we deliberately depart from simply printing the standard number, so it is worth explaining in full.

The middle term of the numerator, P9LNLS, counts loans that are more than 90 days late but on which the bank is still recognizing interest income. A bank only does that when it still expects to be repaid. For large mortgage servicers the reason is usually a government guarantee: FHA and VA loans repurchased out of Ginnie Mae pools stay on the balance sheet, delinquent, accruing, and carrying almost no risk of loss because the guarantee stands behind them.

Counted as distress, those balances can overwhelm the ratio. MidFirst Bank's Q1 2026 filing reported $6.32bn of them — 96.6% of its numerator, and 99.1% of that 1–4 family residential — against a loan loss allowance of 0.67% of loans and non-accrual loans of 0.64%. The standard ratio reads 162.4%. Drop the accruing term and the same bank reads 5.6%. The bank is not in distress; the metric is being fed paper it was never designed for.

The FDIC's financials endpoint publishes no government-guaranteed balance, so the guarantee cannot be netted out directly. What we do instead:

• Calculate a core ratio alongside the standard one, using only non-accrual loans and repossessed property — the part of the numerator a guarantee cannot be hiding in.
• Where the accruing term is at least 40% of the numerator and removing it would put the bank in a different band, we mark the bank Flagged and publish no band at all.
• Both figures stay on screen. The headline number remains the standard Texas Ratio, so it still reconciles with anything else you compare it against.

Both conditions matter. Flagging on the 40% share alone would strip the label from 41 of the largest 500 banks, most of them small ratios sitting comfortably in the same band either way — that buys no honesty and loses information. Requiring that the band actually change narrows it to the handful of banks where the answer genuinely depends on the questionable term. At the most recent regeneration that was 5 banks out of 500.

This is a heuristic, not a measurement. It identifies a pattern strongly associated with guaranteed servicing paper; it does not prove a guarantee exists. The authoritative figure is Call Report Schedule RC-N, Memorandum item 1 — past due loans wholly or partially guaranteed by the U.S. Government — which is not exposed by this API. A bank may also be flagged for reasons that have nothing to do with mortgages.

Where the risk bands come from

Only the 100% line has any standing. Above it, problem assets exceed the cushion held against them, and that is the threshold the literature actually cites. It is a reason to look more closely. It is not a prediction of failure, and plenty of banks have carried a ratio above it and recovered.

Everything else — Strong below 20%, Good 20–40%, OK 40–70%, Poor 70–100% — is our own invention, chosen to make the number readable at a glance. No regulator or standards body recognizes them. A different site could draw the lines elsewhere and be equally justified.

They also flatten a real distinction the ratio cannot make: a 15% reading means different things for a credit card bank and a residential mortgage lender, and this site does not adjust for business model or peer group. Read the band as a rough reading aid and the number as the fact.

What this ratio cannot tell you

The Texas Ratio is a backward-looking measure of credit quality. It answers one question: has this bank's loan book gone bad faster than its capital can absorb? Where a bank fails for some other reason, it has nothing to say.

Five FDIC-insured banks failed in 2023. Computed the way this site computes it, from each one's final Call Report before failing:

Four of the five would have shown a reassuring green. First Republic's figure is from the quarter that ended five weeks before it was seized. Silicon Valley Bank, Signature and First Republic failed in runs driven by uninsured deposits and interest-rate losses, none of which this ratio measures. Heartland Tri-State failed through fraud, which no ratio measures.

The fifth is the exception that shows what the ratio is for. Citizens Bank of Sac City failed from concentrated losses in its loan book — an ordinary credit failure — and the Texas Ratio saw it coming a long way off: 14.0% in Q1 2022, 48.7% by Q2, then 141.8% in Q3 2022 and above 140% every quarter after. That was thirteen months before it was closed. By its final filing the bank's tangible equity and reserves were already negative, so the ratio is undefined and this site would show N/A — which is why an N/A here is never a clean bill of health.

That is the honest summary: this measure catches deteriorating loan books, sometimes early. It does not catch runs, rate risk or fraud, and in 2023 those caused four failures out of five.

Things this ratio does not measure, all of which mattered more in 2023:

• Uninsured deposit share — over 90% at SVB, and the reason the run moved so fast.
• Unrealized securities losses — the gap between what a bond portfolio is worth on the books and what it would fetch today.
• Regulatory capital ratios — leverage and CET1, which have published thresholds set by regulators rather than by us.
• Funding mix and deposit flows — reliance on brokered or wholesale money, and whether deposits are leaving.
• Concentration — in commercial real estate, or in a single volatile industry.

This page used to end here, telling you those were the figures to go and read and leaving you to find them. All five are now on the site, from the same filing, beneath every bank's ratio and as columns on the watchlist. How each is built is below. They do not turn one ratio into a prediction — six measures no more predict a failure than one does — but they are the figures that were actually moving in 2023.

The five measures it cannot see

Every one comes from the same quarterly Call Report as the Texas Ratio, through the same public endpoint, and every one is null rather than zero where a bank does not report what it needs.

One rule governs this whole section. Where a regulator has published a threshold, we name it and name who set it. Where none exists, we show the figure and stop. That is why the capital ratios below carry a category and the uninsured deposit share does not — not because one matters more, but because someone else drew that line and nobody has drawn the other. The Texas Ratio's own bands are our invention, and five more sets of invented bands would have cost this section the only thing that makes it worth reading.

1. Uninsured deposit share

DEPUNINS ÷ (DEPUNINS + DEPINS)

DEPUNINS is the bank's own estimate of deposits above the $250,000 insurance limit and DEPINS the estimate of those below it, both from Call Report Schedule RC-O. The instructions permit estimation where account-level data is not readily available, so these are estimates, not counts.

Why that denominator and not total deposits. Both terms cover deposits in domestic offices plus insured branches in US territories. DEPDOM, the obvious-looking denominator, covers domestic offices alone — so dividing by it gives a bank with a territory branch a numerator scoped wider than its denominator. State Street reads 100.7% uninsured that way, which is not a fact about State Street. Silicon Valley Bank's final filing measures 93.8% on the construction above and 93.9% on the other, so the figure the March 2023 reporting used is preserved either way.

Deposits in foreign offices are in neither term and carry no FDIC insurance at all. They are shown separately where they are material rather than folded in, because mixing a reported estimate with a residual would make both less reliable than they look.

No regulator publishes a threshold for this, so the only comparison offered is the rest of the industry: as of Q1 2026, across the 4,286 banks reporting it, the median was 24.4% and the industry as a whole 43.2%. The gap between those two is the largest banks, which hold most of the uninsured money. Those figures are stamped with their quarter in the code and are recomputed on every watchlist regeneration.

2. Unrealized securities losses

available for sale: SCAF − SCAA
held to maturity:  SCHF − SCHA

Fair value less amortized cost, for each of the two portfolios. The difference between them is the entire point:

• Available for sale is carried at fair value, so its mark is already inside the equity figure this site uses, through accumulated other comprehensive income. It is visible.
• Held to maturity is carried at amortized cost, so its mark appears nowhere on the balance sheet, nowhere in equity, and nowhere in the Texas Ratio. It is disclosed in a footnote.

We therefore apply only the held-to-maturity mark to tangible equity — adding the available-for-sale one would double-count what AOCI has already taken. Silicon Valley Bank's final filing shows a $15.16bn unrealized held-to-maturity loss against $15.17bn of tangible equity: marking that book to market would have left $12m. Its Texas Ratio that quarter was 0.88%.

No regulator publishes a threshold for this either, and an unrealized loss only becomes a real one if the bonds have to be sold — which is precisely what a deposit outflow forces, and why this measure and the one above it are the same story told twice. Most banks also elect to exclude the available-for-sale mark from regulatory capital, so neither figure here sits inside the capital ratios below. Where a bank classifies nothing as held to maturity we say none held, not zero.

3. Regulatory capital ratios

Reported by the bank, not computed by us: RBC1AAJ (leverage), RBCT1CER (common equity tier 1), RBC1RWAJ (tier 1 risk-based) and RBCRWAJ (total risk-based). These are read against the Prompt Corrective Action categories at 12 CFR 324.403:

This is a reading of published thresholds, not a supervisory determination. A bank's actual category can also be lowered by a written capital directive or an unsatisfactory examination rating, and neither is public — so a bank shown here as well capitalized may in fact have been placed a category lower. All three banks that failed in the March 2023 run were well capitalized on this test on their final filings, which is the most useful thing this measure has to say about itself.

Community Bank Leverage Ratio banks. A qualifying smaller bank may elect the framework at 12 CFR 324.12, hold above 9% leverage, and be deemed well capitalized without reporting risk-weighted ratios at all. Those banks show one ratio rather than four. That is a lawful choice of rulebook, and the site says so rather than showing the missing three as gaps in a filing.

One artifact we correct. Where a bank does not report tier 1 capital — a US branch of a foreign bank files no equity line at all — the FDIC still publishes the ratios derived from it, as a literal 0. Taken at face value that puts those branches at the top of any list sorted by weakest capital, reading 0.00%. We treat a derived ratio without the capital behind it as unreported. A bank that genuinely held no tier 1 capital is caught by the tangible-equity test in the last row above, which is the rule's own backstop for that case.

4. Funding mix and deposit flows

Brokered deposits (BRO) against both deposits and assets; borrowings, being fed funds purchased and repo (FREPP) plus other borrowed money (OTHBOR), against assets; and the change in total deposits (DEP) against the filing exactly one and four quarters earlier.

Those comparisons are located by date, never by position in a list. A bank that skipped a quarter would otherwise have a five-quarter-old figure labeled year on year. Where the matching quarter is not on file, no comparison is made.

Brokered money is placed by a third party chasing rate, so it leaves when the rate does. Two published thresholds bear on it: under 12 CFR 337.6 a bank that is not well capitalized cannot accept brokered deposits without a waiver, which ties this measure to the one above it; and under 12 CFR 327.16(e) the deposit insurance assessment applies a brokered deposit adjustment above 10% of assets. We report which of those a bank crosses, against assets, because that is the denominator the rule uses.

A deposit fall is not automatically bad. Banks shed expensive brokered money deliberately, and a shrinking bank is not a failing one. It is reported without a verdict — but it is the number that moved fastest at every bank that failed in 2023: First Republic's last filing showed deposits down 40.8% in a single quarter, and Signature's down 13.8%.

5. Concentration

CRE = LNRECONS + LNREMULT + LNRENROT
÷ (RBCT1J + RBCT2)

Construction and land development, multifamily, and non-owner-occupied nonfarm nonresidential property, over total risk-based capital. Owner-occupied property (LNRENROW) is deliberately excluded, as the guidance excludes it: repayment there depends on the occupying business rather than on the property market.

Read against the two screening criteria in the December 2006 Interagency Guidance on Concentrations in Commercial Real Estate Lending (FDIC FIL-104-2006):

• Construction and land development at or above 100% of total risk-based capital; or
• Total commercial real estate at or above 300% of total risk-based capital and grown 50% or more over the prior 36 months.

The second criterion needs both legs. A bank at 400% of capital with a flat portfolio does not meet it, and marking it as though it did would invent a finding the guidance does not make. Where a bank has no filing from 36 months back the growth leg is unknown, and the site says so rather than defaulting either way.

Meeting a criterion is not a finding of anything. The guidance says in terms that a bank exceeding these levels may be identified for further supervisory analysis, and that the criteria are neither limits nor evidence of unsafe practice. Signature Bank sat at exactly 300% on its final filing; so do a number of perfectly ordinary community banks in the list today.

This figure is a floor. The guidance also counts loans that finance commercial real estate without being secured by it, and that balance is not exposed by this API. A bank's true concentration can only be higher than what is shown, never lower.

Concentration in a single industry is shown a second way, without a threshold, because none exists: the loan book split into the categories the FDIC breaks out — commercial real estate, 1–4 family residential, commercial and industrial, consumer and agricultural — as shares of gross loans, alongside the FDIC's own asset concentration hierarchy label for the bank. Those shares do not add to 100%; they are the categories the FDIC publishes, not a complete split of the book.

Charters, not holding companies

Call Reports are filed by each banking charter, not by the group that owns it. Most people think in brands; the FDIC thinks in charters, and the two do not line up.

Large groups often run several charters of very different sizes. Wells Fargo & Company, for instance, has Wells Fargo Bank, N.A. at roughly $1.85 trillion in assets alongside much smaller separately chartered banks. Searching “Wells Fargo” returns all of them, and they do not carry the same figures.

We list banks under their legal charter name, so the brand on your branch or debit card may not match. If you are unsure which charter holds your account, your deposit agreement or statement will name it, and the FDIC certificate number is the unambiguous identifier. Credit unions are insured by the NCUA rather than the FDIC and do not appear on this site at all.

Data, vintage and reproducing it

Every figure comes from the FDIC's public BankFind API at api.fdic.gov/banks, sourced from quarterly Call Reports. Nothing here is real-time or proprietary.

There is an unavoidable lag. Banks file roughly 30 days after a quarter closes and the FDIC publishes some weeks after that, so the most current figure available is generally between two and five months old. Every result shows the quarter it came from. Banking conditions can change faster than that, and in 2023 they did.

Two paths, one calculation. A single bank lookup queries the FDIC live. The 500-bank list is a snapshot file regenerated each filing cycle by scripts/fetch_watchlist_data.py, which uses the same arithmetic for the Texas Ratio and for all five measures above — the two implementations are deliberate duplicates, because a browser and a build script cannot share code, and each carries a comment pointing at the other. The list header states the reporting period; the snapshot also carries each bank's own filing date, and any bank whose quarter differs from the rest is marked on its row rather than being quietly averaged in. If the snapshot falls a full cycle behind, the page says so itself.

You can check any of this. Take a certificate number, request the fields listed above from the FDIC endpoint, and do the division. If your answer differs from ours, we would like to know — see below.

Every bank result carries a link to the exact request the page made, built from the same function that made it, so the data behind a figure cannot drift from the figure. The one thing that can go stale without anyone noticing is the industry comparison quoted beside the uninsured deposit share: it is a fact about a single quarter, so it is stamped with that quarter wherever it appears, and the regeneration script recomputes and prints it every time it runs.

Corrections

This site publishes calculated figures about named, regulated institutions. That obliges us to be reachable and to fix what is wrong.

If you are a bank, a journalist or a reader who believes a figure here is wrong — whether the arithmetic, the underlying filing, or the framing around it — write to [email protected] with the bank's name or certificate number and what you believe the correct figure to be. We aim to reply within two business days.

Where a number is wrong we will correct it and note what changed. Where a number is right but misleading, we would still like to hear about it — the still-accruing check described above exists precisely because a technically correct figure was telling readers something untrue.

Where the FDIC's own published data is the source of an error, we will say so and link the filing rather than quietly adjust it.

If you are a bank named on this site and you believe a figure or the framing around it misrepresents you, write to the same address. We will look at it on receipt, correct anything that is wrong, and publish a note of what changed. A bank that disputes a figure we still believe is right may send a statement of its position, and we will carry it alongside the figure.