US Banking Before the Print

The US banking system goes into the third-quarter prints earning more than in any year of the FDIC's five-year ledger, on the widest annual margin in it, with normal credit, 47 problem banks and six small-bank failures this year. It is also growing faster than its capital, lending its fastest-growing dollars to other lenders, funded increasingly by uninsured money, and carrying $327 billion of paper losses on its bonds that the September rate rise has pushed back toward $500 billion. Twelve of the largest financial companies report between 13 and 15 October. This note is the ledger they report against and the story behind it: how two laws and two crises made the twelve; what the sector earns, lends and owes; the six businesses the twelve are and the one line the market prices them on; what their own managements fear and where they are taking the industry; whether the case from bitcoin and stablecoins that banking is ripe for disruption holds; and what would have to happen for the paper loss to become a real one.

TON618 Capital Research · As of 10 October 2026 · Download PDF

On 25 August the FDIC reported second-quarter net income of $90.1 billion for the 4,238 institutions it insures, a return on assets of 1.37% and a net interest margin of 3.32%; domestic deposits rose for an eighth straight quarter, loans grew 6.8% on the year, and unrealized losses on securities were $326.7 billion, flat on the quarter. Three weeks later the Federal Open Market Committee raised the funds rate a quarter point, to 3.75–4.00%, and the 5-year Treasury ended September 90 basis points above where it ended June. JPMorgan, Goldman Sachs, Citigroup and Wells Fargo report on Tuesday 13 October, Bank of America, Morgan Stanley, BlackRock and State Street on Wednesday, Schwab, PNC, U.S. Bancorp and BNY on Thursday. Part 2 of this note, on Friday 16 October, scores what they say against what is frozen here.

Terms used throughout, defined once. ROTCE: return on tangible common equity, net income to common shareholders over equity less preferred stock, goodwill and intangibles; the return a bank earns on the capital it actually holds. P/TBV: price to tangible book value, the market's multiple of that same capital. NIM: net interest margin, interest earned less interest paid, as a share of earning assets. Efficiency ratio: costs as a share of revenue; lower is better. Charge-offs: loans written off, net of recoveries, as a share of loans. CET1: common equity tier 1, the core regulatory capital ratio, capital over risk-weighted assets. Categories I–IV: the Fed's size and complexity tiers for large banks; the tier sets which rules apply, including whether the next term counts in capital. AOCI: accumulated other comprehensive income, where the paper gains and losses on available-for-sale (AFS) bonds sit; held-to-maturity (HTM) bonds are carried at cost and their paper loss is a footnote. The mark: that paper gain or loss. Call Report: the quarterly regulatory filing of each insured bank, which the FDIC publishes; the bank is the insured subsidiary, not the listed holding company, and both are shown here. NDFI: nondepository financial institutions — private-credit funds, mortgage companies, broker-dealers, insurers — the borrowers in the loan category that grew most in dollars.

The arc. The twelve are products of two laws and two crises. The Riegle-Neal Act of 1994 removed many of the restrictions on interstate branching that dated from 1927; Gramm-Leach-Bliley in 1999 repealed the parts of Glass-Steagall that had separated commercial and investment banking since 1933 and created the financial holding company, a year after Citicorp and Travelers had merged "in anticipation of a change in the law then under discussion in Congress", as the Federal Reserve's own history puts it. Between them they produced the 1998 wave that built today's Bank of America (NationsBank and BankAmerica), Citigroup, and, through 2000 and 2004, JPMorgan Chase. The year 2008 then sorted the survivors from the rest: in March JPMorgan took Bear Stearns with Federal Reserve financing, and in September Washington Mutual's banks from the FDIC for $1.9 billion; on Sunday 21 September the Board approved Goldman Sachs and Morgan Stanley as bank holding companies; Bank of America closed on Countrywide on 1 July and was approved for Merrill Lynch on 26 November, closing on 1 January 2009; Wells Fargo was approved for Wachovia on 12 October and closed on 31 December; Citigroup received a $306 billion asset guarantee and $20 billion of additional capital from Treasury, the Fed and the FDIC together on 23 November; and PNC was approved for National City on 15 December. The May 2009 stress test found ten of the nineteen largest firms short of capital by $74.6 billion, $33.9 billion of it at Bank of America, $13.7 billion at Wells Fargo, $5.5 billion at Citigroup; JPMorgan, Goldman, BNY, State Street, U.S. Bancorp and Capital One needed none. Dodd-Frank followed in 2010 with the annual stress test, orderly liquidation and the consumer bureau; the 2018 relief act and the 2019 tailoring rule loosened it eight and nine years later, sorting firms over $100 billion into the four categories that decide, among other things, whether paper losses on bonds count in capital.

The second crisis was March 2023: Silicon Valley Bank, $209 billion of assets, took $42 billion of withdrawal requests in a day and was closed on a Friday; the systemic-risk exception on the Sunday made every depositor at SVB and Signature whole; First Republic went to JPMorgan on 1 May; and the Fed's own review found supervisors who "did not fully appreciate" the vulnerabilities and a tailoring approach that had impeded oversight. The July 2023 Basel endgame proposal, a 16% increase in capital requirements in aggregate (19% for the largest firms) with paper losses counted at every firm over $100 billion, was cut to a 9% increase for the largest firms in September 2024 and replaced in March 2026 by a package the agencies say would modestly reduce capital overall, the leverage-ratio reform having taken effect on 1 April; the one element carried through every revision is that the mid-sized banks would count their bond marks in capital from 2028, on the proposal's schedule, if it is finalised as written.

The acquirers. JPMorgan Chase counts more than 1,200 predecessor institutions, the Manhattan Company of 1799 and Drexel, Morgan of 1871 among them; its modern shape is Chase Manhattan with J.P. Morgan in 2000, Bank One in 2004 (with its current chief executive), Bear Stearns and Washington Mutual in 2008 and First Republic in 2023, a $229 billion bank bought from the FDIC with a loss-share and a $2.8 billion bargain-purchase gain; assets $2.2 trillion at the end of 2008, $4.4 trillion at the end of 2025. Bank of America is the legal successor to NationsBank of Charlotte, which took BankAmerica of San Francisco in 1998; Countrywide came in July 2008 and Merrill Lynch on 1 January 2009, Merrill's name surviving on the wealth business; assets $1.8 trillion to $3.4 trillion. Wells Fargo was approved for Wachovia in October 2008, closed at year-end, and spent 2018 to June 2025 under a Federal Reserve order that froze its balance sheet at its end-2017 size for "pervasive and persistent misconduct" after the 2016 unauthorized-accounts scandal; its assets were $1.93 trillion in 2019 and $1.93 trillion in 2024, and $2.15 trillion seven months after the cap came off. PNC took National City in December 2008, BBVA's US bank in 2021 and FirstBank of Colorado in January 2026; U.S. Bancorp took Union Bank from MUFG in 2022 and with it a commitment to update its resolution plan and file implementation plans for heightened prudential standards; Truist is BB&T and SunTrust, merged in 2019; Capital One closed on Discover in May 2025 and with it a card network.

The converted. Goldman Sachs and Morgan Stanley were approved as bank holding companies on 21 September 2008 to reach the Fed's window and have spent the years since deciding what kind of bank to be. Goldman built a consumer deposit base under the Marcus name and has, since 2023, sold the GreenSky and Marcus loan books, the General Motors card and, over the next two years, the Apple card, which "will substantially complete the narrowing of such focus"; its assets went from $1.1 trillion at the end of 2008 (as later recast to a December year-end; $0.88 trillion at its own November 2008 year-end) to $0.85 trillion in 2009 and $1.8 trillion in 2025. Morgan Stanley bought ETRADE in 2020 and Eaton Vance in 2021 and describes "the transformation of Morgan Stanley over the past decade and a half" into a firm whose wealth business had revenues of $31.8 billion at a 29% margin in 2025 across advisers, workplace plans and ETRADE; assets $0.7 trillion to $1.4 trillion. Schwab, a savings-and-loan holding company rather than a bank holding company, took TD Ameritrade in October 2020 and finished converting its clients in May 2024; in 2023 its clients moved cash out of sweep to higher yields, bank deposits fell from $367 billion to $290 billion in a year, home-loan-bank borrowings doubled to $26 billion and total assets fell by $59 billion; by the end of 2025 the borrowings were $1.9 billion. It had $51.7 billion of assets in 2008 and $491 billion in 2025, about half of it the bank.

The custodians and the manager. BNY has been in business since 1784, Alexander Hamilton's bank, and became BNY Mellon on 1 July 2007; it holds $59 trillion of client assets in custody or administration on a $472 billion balance sheet. State Street's oldest ancestor is the Union Bank of 1792; it is a trust bank by charter, $54 trillion in custody and administration on $366 billion of assets, and the one transaction its latest 10-K narrates is the one it abandoned, Brown Brothers Harriman's investor-services business, in November 2022. BlackRock was founded in 1988, merged with PNC's asset manager in 1995, listed in 1999, bought Merrill Lynch Investment Managers in 2006 and Barclays Global Investors, with iShares, for about $15 billion in cash and stock on 1 December 2009; in 2024 and 2025 it added Global Infrastructure Partners, Preqin and HPS Investment Partners, "our largest integrations in more than 15 years", to reach $14 trillion of assets under management on a $170 billion balance sheet.

The rescued one. Citigroup dates itself to the City Bank of New York in 1812 and to Travelers in 1998; it is the one of the twelve rescued outright in 2008 and the one that has spent 2021 to 2026 shrinking by design: consumer banking exited in nine of the fourteen markets across Asia, Europe, the Middle East and Mexico it set out to leave, Banamex a quarter sold in December 2025 ahead of a listing, Poland signed, Russia sold in February 2026, five businesses left. Its assets were $1.94 trillion in 2008 and $1.95 trillion in 2019.

What the histories have in common. Each crisis produced consolidation and the acquirers were the survivors; the largest balance sheet was the one the state chose to work through — Bear Stearns with Federal Reserve financing and Washington Mutual's banks from the FDIC in 2008, First Republic from the FDIC in 2023, all three to JPMorgan — while Wachovia and Merrill Lynch went to Wells Fargo and Bank of America as private bids approved by the Board nine days (Wachovia) and ten weeks (Merrill) after signing; and the rule cycle has tightened within two years of each failure and loosened eight to nine years later, with the direction changing within three years of 2023. The two names whose assets stood still for years, Citigroup from 2008 to 2019 and Wells Fargo from 2019 to 2024, are the two whose constraint was their own rather than the cycle's.

Returns and margin. The FDIC's selected indicators for all insured institutions run as follows: return on assets 1.32% year to date in 2026, against 1.20% for 2025, 1.12% for 2024, 1.09% for 2023, 1.11% for 2022 and 1.23% for 2021; return on equity 13.1%, against 11.8%, 11.4%, 11.4%, 11.8% and 12.2%; net interest margin 3.32%, against 3.30%, 3.22%, 3.30%, 2.95% and 2.54%. Second-quarter net income was $90.1 billion, up 12% on the quarter and 29% on the year, on higher noninterest income (trading, fees, one-time gains on equity securities) and a margin a basis point wider. The sector is earning more, on more assets, at a wider annual spread, than in any year of the five.

Credit. Net charge-offs 0.58% of loans year to date (0.62% in 2025, 0.68% in 2024, 0.52% in 2023, 0.27% in 2022); noncurrent assets plus foreclosed property 0.52% of assets (0.54%, 0.55%, 0.47%, 0.39%). Both have been flat for two years after normalizing from the 2021–22 lows. The problem-bank list is 47 (59 a year ago, 66 at the end of 2024), the lowest since 2023 and 1.1% of institutions, "in the normal range of 1 to 2 percent for non-crisis periods"; two banks failed in the first half with $517 million of assets between them, and four more community banks have closed since 30 June, six for the year on the FDIC's list at 9 October; 4.2% of institutions are unprofitable, the lowest share since 2022.

Capital and growth. The core leverage ratio is 8.98%, down 17 basis points in the quarter and 39 on the year, and the tier 1 risk-based ratio 13.75%, down 17 basis points, because, in the FDIC's words, "asset growth outpaced capital accretion": total assets grew 5.9% on the year to $26.5 trillion, loans 6.8% to $13.9 trillion. The largest dollar increase in loans over the year was to nondepository financial institutions, $279 billion or 22%, to a level of about $1.5 trillion on the FDIC's count (the Fed's weekly series for commercial banks puts it at $2.07 trillion on 23 September); the next were commercial and industrial loans and loans to purchase or carry securities, the last of them the fastest in percentage terms (+30%). Domestic deposits grew $143 billion, an eighth consecutive quarter; estimated uninsured deposits grew $317 billion, so insured deposits fell.

What changed. Nothing in the health measures, which are all at or near their five-year best. Three things in the shape: the sector's leverage ratio fell 39 basis points on the year, to its 2022 level, after rising in each year of the ledger from 2021 to 2024; its largest loan growth is lending to other lenders; and its new funding is uninsured. And one thing in the footnotes: unrealized losses on securities, which had fallen from $684 billion in September 2023 to $306 billion at the end of 2025, stopped falling ($325 billion in March, $327 billion in June) and, on the September curve, are back near $500 billion (§8).

The tiers. The Fed sorts large banks into four categories that set which rules apply. Category I is the eight US global systemically important banks: JPMorgan, Bank of America, Citigroup, Wells Fargo, Goldman Sachs, Morgan Stanley, BNY and State Street. Category II is $700 billion or more of assets or $75 billion of cross-jurisdictional activity (Northern Trust). Category III is $250 billion or more, or $100 billion with heavy wholesale funding, nonbank assets or off-balance-sheet exposure: U.S. Bancorp, PNC, Truist, Capital One, Schwab, and the US arms of TD and BMO. Category IV is everything else over $100 billion: Key, Citizens, M&T, Regions, Fifth Third, Huntington, Ally, American Express, Synchrony, First Citizens, Santander, HSBC USA, USAA. The tier decides, among other things, whether paper losses on available-for-sale bonds count in regulatory capital: they must at Categories I and II, and may be elected out at III and IV, which most have done. Of the twelve, eight are Category I, three are Category III on the Fed's June list (U.S. Bancorp, at $726 billion of assets, describes an upcoming Category II transition, which would end its election), and BlackRock sits outside the bank framework altogether.

The models. Within that frame the twelve are six businesses, with a seventh in the peer set. The universal banks (JPMorgan, Bank of America, Wells Fargo, Citigroup) hold $0.8–1.5 trillion of loans each at the holding company and earn 46–56% of revenue from fees at JPMorgan, Bank of America and Wells Fargo (cards, markets, investment banking, wealth, payments) and 31% at Citigroup, whose $17.1 billion of quarterly net interest income is the largest share of revenue among the four.

The markets houses (Goldman Sachs, Morgan Stanley) earn 81% and 87% from fees and trading: Goldman's $20.3 billion of net revenue is $4.0 billion net interest and $16.4 billion the rest ($7.6 billion of it market making), Morgan Stanley's $21.3 billion is $2.8 billion and $18.6 billion ($6.9 billion asset management, $6.7 billion trading). Morgan Stanley is the hybrid: wealth management is $8.9 billion of the $21.3 billion and carries $196 billion of loans and $436 billion of deposits, so it is half markets house, half broker-bank. Their banks hold loans at 28–33% of assets, mostly "other" loans to funds and financial counterparties, and are funded by sweep and brokered deposits with almost no noninterest-bearing money.

The custody banks (State Street, BNY) earn 75–79% of revenue outside net interest, from servicing and management fees on $57.9 trillion (State Street) and $62.6 trillion (BNY) of assets under custody or administration, with net interest income a $0.9–1.4 billion second engine on institutional operating deposits; loans are 11–13% of their banks' assets, securities 26–37%, and 68–76% of their deposits are uninsured operating balances. The broker-bank (Schwab) is the mirror: 63% of its bank's assets are securities, bought with sweep cash that costs 0.19% (0.34% all-in cost of funds at the bank), beside $67 billion of mortgages and pledged-asset lines, against $13.1 trillion of client assets at the broker; the holding company's $7.1 billion of net revenue is $3.4 billion net interest and $3.7 billion (53%) outside it, most of that asset-management, trading and bank-deposit-account fees.

The regionals (PNC, U.S. Bancorp, Truist) are spread lenders: loans 59–61% of assets, commercial and industrial 22–32% of the book, residential mortgages 21–32%, commercial real estate 12–15%, fees 31–43% of revenue (PNC $4.1 billion net interest against $2.8 billion of fees; U.S. Bancorp $4.4 billion against $3.3 billion, a third of the fees from payments). The card bank (Capital One, in the peer set, not the week) runs cards at 55% and auto at 20% of loans, an 8.5% margin and 3.3% charge-offs. And the asset manager (BlackRock) has no bank book: $176 billion of assets against $15.3 trillion of assets under management, revenue 81% base fees priced on that AUM ($5.7 billion of $7.1 billion), performance fees ($0.3 billion), Aladdin technology subscriptions ($0.6 billion), and a 25× earnings multiple to match; its only rate exposure runs through client flows and market levels.

The loan books, and the bucket that grew. At the bank level, one bucket stands out across the twelve: "other loans", the Call Report line that holds loans to nondepository financial institutions, loans to purchase or carry securities and the rest, is 30–34% of the universal banks' books, 66% of Goldman's, 82% of Morgan Stanley's, 81–93% of the custody banks', 51% of Schwab's (its pledged-asset lines), 20–28% of the regionals' and 10% of Capital One's. Commercial real estate, the category the 2023–24 regional worry was about, is 0–15% everywhere. Cards are 23% of Citibank's book and 55% of Capital One's; everywhere else under 15%. The "other loans" bucket, which holds the sector's nonbank lending alongside margin and pledged-asset lines, is already the largest single bucket at eight of the eleven with bank books.

Thirteen lead insured banks (the eleven of the twelve that have one, plus Truist and Capital One from the peer set) on six Call Report measures, first half of 2026 annualized (return on equity for the second quarter alone), against the same period a year earlier and against all insured banks where the FDIC publishes a figure:

Margin. The sector's margin widened 9 basis points over the year; among the thirteen it widened most at Schwab (+71, as sweep cash replaced bank borrowings that had cost about 4.5%), Capital One (+87, the Discover book), State Street (+17) and PNC (+16), and narrowed at Morgan Stanley (−47, its wealth-management segment's deposits costing 2.6% at period end), Goldman (−21) and Wells Fargo (−16). The cost of funds fell at twelve of the thirteen (BNY −151 basis points, State Street −72, Schwab −62, Goldman −41) as 2025's rate cuts passed through, and rose only at Morgan Stanley. The September hike reverses that direction; the third-quarter cost-of-funds line is the first place it shows, and §9 freezes it.

Returns. Bank-level return on equity runs from 10.0% (Truist) to 20.7% (BNY) and 20.5% (JPMorgan Chase Bank), with the sector at 13.8% for the quarter; it rose over the year at eleven of the thirteen, was flat at Wells Fargo, and rose most at Capital One (+34 points, from a loss), Schwab (+8.4), State Street (+5.6), BNY (+5.0), JPMorgan (+4.1) and Citibank (+3.6), and fell at Morgan Stanley's bank (−5.5). At the holding companies, where the market looks, the ranking changes on tangible equity (Morgan Stanley, Goldman and Schwab rise), because the base is tangible common equity, not bank equity (§5).

Credit. Charge-offs are between zero and 0.6% at eleven of the thirteen (BNY net recoveries, JPMorgan 0.61%), 1.06% at Citibank and 3.30% at Capital One, both card books, against 0.58% for the sector; allowances 1.1–2.3% of loans at the universal and regional banks, 5.0% at Capital One, 0.3–0.6% at the markets and custody banks whose loans are collateralized lending to funds. Charge-offs are lower on the year at nine of the thirteen, flat at Schwab and Truist, and higher at Capital One and State Street.

Efficiency. Schwab's bank runs at a 17.6% efficiency ratio because the cost base of the franchise sits in the broker, not the bank; the markets banks at 37–39% for the same reason; the universal and regional banks at 51–59%; the custody banks at 57–66%, where the servicing cost base sits inside the bank.

Capital and funding. CET1 ranges from 11.2% (PNC Bank) to 18.2% (Morgan Stanley Bank), with Schwab's bank at 30.1% on a small risk-weighted base and with its paper losses excluded; every bank is above its requirement, and the gaps are in §8. Noninterest-bearing deposits, the cheapest funding, are 24–29% of deposits at JPMorgan, Bank of America, Wells Fargo, PNC, Truist and BNY, 12–16% at Citibank, U.S. Bank and State Street, and effectively zero at Goldman, Morgan Stanley and Schwab, whose deposits are sweep and brokered balances. Uninsured deposits are 42–52% of the money centers' and regionals' books, 68–76% of the custody banks', and 19–21% at Schwab and Capital One.

One line. Take each company's price to tangible book value (market capitalization over common equity less preferred, goodwill and intangibles at 30 June) and its return on tangible common equity (first-half 2026 net income to common, annualized, over average tangible common equity), for the eleven of the twelve with a bank book and ten large regionals. The twenty-one points fit a single line: P/TBV = −0.76 + 0.165 × ROTCE, R² 0.88. The market pays about 0.16× of tangible book for each point of return on it, and values a bank at exactly tangible book when it earns about 11% on tangible equity. On full-year 2025 returns, which strip out the half-year's one-off items, the line is −0.96 + 0.209 × ROTCE, R² 0.91, book at 9.4% (nineteen points). Either way the group is priced on one variable, tightly.

Who sits where. The money centers and the markets houses are on the line or a little below it: JPMorgan and Goldman about 0.35× of tangible book below their fitted values, Bank of America 0.14× below, Wells Fargo 0.06× and U.S. Bancorp a tenth below, Citigroup a tenth below at 1.3× book on a 13% return. The market is not paying a premium for the highest-return scale franchises. The premiums are at the names with the least balance sheet per dollar of equity: Schwab at 6.9× book on a 42% return and BNY at 4.7× on 32% sit above the line, State Street a quarter-turn above it; their returns are high because their tangible equity is small against their balance sheets (which is also why their paper losses are large against it, §8). The regional peers run 0.8–2.4× book and 10–21% returns (Northern Trust, a custody bank in the peer set, sets both ceilings), with the eight others' median at 1.8× and 14.6%; Ally (0.8× book, 8.9× trailing earnings) and Zions (1.2×, 8.0×) are the cheapest in the set, Truist third at 1.2× and 10.6×. On earnings the group is 11–18× trailing except BlackRock, an asset manager at 25×. Shareholder yield, dividends plus buybacks, runs among the thirteen in the table that report one from 2.2% (Morgan Stanley) to 9.2% (Truist). Fifth Third and Zions are the two points far from the line on the half-year fit; Fifth Third is near it on the full-year fit, Zions stays far from it.

For the valuations. JPMorgan (universal, 2.7× book on 23%), Goldman Sachs (markets house, 2.7× on 23%) and BlackRock (asset manager, 25× earnings, no book) are the three archetypes the house values after their prints. The line gives each of the first two a residual to explain, −0.34× and −0.36×; BlackRock is priced on fees, not book.

Where they put the risk. Read together — the July calls, the 2026 shareholder letters, the risk-factor sections of the FY2025 10-Ks and the one substantive September appearance, Bank of America's at the Barclays conference — we rank the twelve managements' risks by how many raise each in a letter or in prepared remarks, unasked, then, where the counts tie, by the weight of the language. Quotations from the Goldman, Morgan Stanley, BNY, PNC, Schwab and BlackRock calls rest on a single transcription, the State Street call on two independent transcriptions; the rest are from the companies' own documents. First, geopolitics: nine of the twelve raise it unprompted and most list it first; the war in Iran and its effect on energy prices, Russia's war and sabotage in Europe, the US–China relationship and the tariff realignment. JPMorgan's chairman puts it "first and foremost" and writes that the outcome "may very well be the defining factor in how the future global economic order unfolds — then again, it may not"; Morgan Stanley's chief executive calls it one of the two "known unknowns" of 2026; Citigroup's chief executive gives it as the reason not to thin her capital buffer. Second, AI, in both directions, also raised unprompted by nine: the durability of a capital-expenditure cycle that Morgan Stanley's chief executive sizes at $850 billion this year against a $575 billion forecast and $1.3 trillion next, "10%–15% of the way through", with "periods when technology or capital investment is ahead of adoption"; and AI itself, which BNY's 10-K says "has the potential to transform economic and societal structures in ways that are difficult to predict or control, potentially threatening the viability of our company and the broader financial system". Goldman's chief executive wrote that "concerns about private credit, including underwriting quality or exposure to software companies that may be adversely affected by AI, are a reminder that the credit cycle has not been repealed".

Third, asset prices, in three letters (JPMorgan's, Wells Fargo's, Morgan Stanley's) and in the heaviest language: JPMorgan's "getting close to as good as it gets", a household net worth of 560% of GDP against 460% at the 2006 housing peak, and bank stocks "fully valued"; Wells Fargo's "credit spreads are the narrowest they've been in 30 years … When this kind of environment persists, bubbles are created"; Morgan Stanley's letter: "Asset valuations continue to reflect several years of substantial gains." Fourth, private credit and lending to nonbanks, which is where the sector's loan growth is, in three letters and placed ahead of deposits on the weight of the language (JPMorgan's five paragraphs): it appears in the letters of the three largest wholesale franchises (JPMorgan's runs to five paragraphs — "losses on all leveraged lending in general will be higher than expected … credit standards have been modestly weakening pretty much across the board"), in U.S. Bancorp's 10-K as the only dedicated nonbank-loan-book risk factor among the twelve, and on one call, BlackRock's, as "idiosyncratic risk … the last quarter of last year. We've seen actually a stabilization". As a risk it does not appear on the three money-center calls; of the eight other transcripts on file only BlackRock's raises it (Wells Fargo posts none); and Bank of America mentions it only in forward-looking-statement boilerplate, its 10-K naming asset managers and funds and finance companies as its largest commercial concentrations — its chief executive, asked in September whether credit quality, higher rates, the AI build-out or private credit kept him up at night, answered "No".

Fifth, deposit repricing under a higher-for-longer path, present at eight of the twelve, most of it when analysts asked, and placed last on the weight of the unprompted language: JPMorgan's chief financial officer named "yield-seeking flows" as "probably a little bit of a risk" and declined to extrapolate the bank's rate tables to a 7% world; PNC's chief executive said that a commercial bank "without a retail franchise" finds funding "really tight" and that "retail shares [are] moving aggressively to the larger players"; BNY expects an 80% beta on dollar deposits, as last cycle.

What they do not say. Commercial real estate is discussed by everyone who mentions it in the past tense, and PNC is growing the book again. Consumer credit is "as good as it's been in a long, long time" at Bank of America and stable or improving everywhere; the watch item each names is the labour market, JPMorgan adding a cohort with negative real wages. Cyber is in every 10-K and in JPMorgan's letter as "one of our biggest risks", and in JPMorgan's and Wells Fargo's letters, and on a call or appearance only once, Bank of America's in September. The 2026 capital rewrite got more call minutes than any other regulatory topic — the G-SIB surcharge "is still broken … Frankly, it's not right, and it's un-American" (JPMorgan's letter), the double count of operating and market risk (Citigroup, JPMorgan), the leverage ratio as the binding constraint (Goldman, Morgan Stanley, State Street) — but the managements frame it as a grievance or a tailwind, not a risk; no management on any call discussed the proposal that would count their bond marks in capital, though U.S. Bancorp's gave its AOCI-inclusive ratio in prepared remarks (§8). And nobody on the four money-center calls was asked about the securities book at all.

Where they say it is going. Five bets recur. One, revenue grows while headcount falls, with AI as the mechanism: ten of the twelve say AI is raising their productivity (BlackRock frames it as an investment engine; JPMorgan's dissent is over who keeps the gain), five report headcount lower than a year earlier (Bank of America, Citigroup, Wells Fargo, State Street, BNY) and Goldman's is lower since the start of the year; the numbers are Bank of America's (130–140 implemented use cases "at a cost of $400 million, generating a benefit of $800 million", a budget to double next year, headcount "march[ing] down" by attrition from 210,000–212,000 at the start of the year), State Street's ($1 billion of run-rate benefit by 2029, three-quarters of it expense, about $500 million of one-time costs, mostly severance, to get it), Citigroup's (219,000 employees, over $800 million of severance year to date, and "we may take more severance in the second half"), Wells Fargo's (headcount down 7% on efficiency initiatives generally; on AI, "a 35% increase in code written" and, for now, "we haven't reduced resources because of the efficiencies gained") and BNY's (headcount down 7%, "134 digital employees"). JPMorgan is the dissent on the premise, not the practice: "you don't uniquely benefit from AI. The ultimate beneficiary of AI will be our customers … If that were true, our margins would be 80% today because of computerization over the last 20 years", said while reporting 30–40% job cuts in discrete areas and a compute bill in which its chief financial officer expects "some meaningful acceleration" in the second half, of a number he called trivial for the year.

Two, advice and wealth as the growth engine, fed by the consumer franchise or the workplace: Morgan Stanley's $10 trillion of client assets and 20 million relationships, Goldman's "very, very fragmented" ultra-high-net-worth market with "no big leader", Bank of America's Merrill Edge growing accounts 10% a year, Schwab's "bull market for advice" with 5% of its households on a fee-based solution against 31% willing to pay, Citigroup's retail bank folded into Wealth.

Three, private markets as product and as origination partner: Goldman's $75–100 billion a year of fundraising and $750 billion of fee-paying alternatives by 2030, BlackRock's $400 billion by 2030 with HPS and insurers earning "150–350 basis points over Treasuries", Wells Fargo's Overland Advantage with Centerbridge ($3.6 billion of direct loans beside $3.3 billion of bank loans since mid-2024), Morgan Stanley's EquityZen, Schwab's Forge.

Four, bank-controlled tokenisation and always-on dollar rails: Bank of America's chief executive, in September, on the stablecoin question — "the real question is, would it attack and take non-interest bearing accounts … it will shrink the banking system and the lending capacity … We've minted our own stablecoins … the tokenized deposit may let us close down the need to move amounts of money on the weekend"; Citigroup's "ensure that as markets evolve, Citi remains the platform through which liquidity moves"; Wells Fargo's stablecoin with Early Warning Services and Zelle volume of $336 billion; BNY's "always-on operating model", USDC custody and mint-and-burn, and 24×7 dollar transfers; State Street's tokenised fund servicing by year-end and a stablecoin-reserves fund it calls "table stakes"; BlackRock's $500 million digital-assets revenue target on the $60 billion of Circle's reserves it cited in July ($65 billion of stablecoin reserves in all at year-end 2025) and a plan to be "a digital wallet native asset manager"; Schwab's spot crypto and a view that 24-hour trading is of uncertain value since "1%–2% of trading actually happens outside of market hours". FedNow, the Fed's own rail, is named by one of the twelve in any document on file, PNC, as a treasury-management feature in its annual report.

Five, organic growth first and bolt-ons only, with the consolidation they expect being their clients', in what Morgan Stanley's chief executive calls "a normalization of regulation cycle" and Goldman's says has turned "absolutely no way" into "maybe"; capital returned when not deployed; and the US branch re-expanded as the deposit and wealth on-ramp — JPMorgan's 900 branches in five years, U.S. Bancorp's $300 million a year in ten markets, Wells Fargo's 700 refurbishments in 2025.

The one thing the managements and the challenger agree on. That the dollar is going on-chain and that the question is who issues it. The managements' answer is a deposit token, a consortium coin, a reserves fund and custody; the challenger's is a token issued by someone else that pays its holder. Whether a bank-issued token pays anything is one of the dated tests in §9; §7 tests which reading of the record holds.

The case, in its proponents' terms. The critique of American banking that comes from bitcoin, stablecoins and fintech is eight arguments usually run as one. The rails are old and closed: Fedwire, through which bank-to-bank dollars finally move, runs about 22 hours on business days, and the Fed's own plan takes it to six days a week in 2028 or 2029 and to seven "no sooner than two years" after that; FedNow, the instant rail the Fed built in 2023, had more than 1,800 participants at its third anniversary against about 8,600 banks and credit unions. The system is expensive and the cost falls on the people with the least: merchants paid $187 billion to accept $11.9 trillion of card payments in 2024, insured banks booked $9.3 billion of deposit-account service charges in the second quarter alone, and sending $200 across a border cost 6.5% on the World Bank's count for the first quarter of 2025. Savers are paid almost nothing while the Fed pays banks the policy rate: the funds target is 3.75–4.00% and the FDIC's September national rates are 0.37% on savings and 0.07% on interest checking; Coinbase pays its subscription members 3.5% on a USDC balance. The funding model is run-prone by construction and insured by the public: Silicon Valley Bank took $42 billion of withdrawal requests in eight hours, with $100 billion queued for the next morning, and the systemic-risk exception made every depositor whole; a fully reserved token cannot be run in that sense, which is why the White House's own economists write that, with full-reserve backing, stablecoins "could serve as" "an instantiation of the old idea of 'narrow banking'". Access to central-bank money is gatekept: Custodia waited 27 months for a master account, was refused, and lost at the Tenth Circuit in October 2025; and the FDIC's 2022–23 "pause letters" to 24 banks and the loss of banking access by at least 30 digital-asset firms between 2022 and 2024, documented in the FDIC's own release and a House committee report, showed that regulators and incumbents could cut a competitor off from the dollar without a rule, a vote or a court. The successor product, on this view, is already here: $312 billion of dollar stablecoins, $11.6 trillion of adjusted transfer volume in 2025, and a Treasury advisory committee relaying Standard Chartered's projection of about $2 trillion by 2028.

The scale the case has to live with. US commercial banks held $19.7 trillion of deposits on 30 September. The $312 billion of stablecoins is 1.6% of that, and a stock of tokens is not a count of households that closed a checking account; deposits grew by more than the stablecoin stock's eighteen-month increase in each of the last several quarters. The "$6.6 trillion" that appears in the bank associations' letters is the Treasury committee's count of the transactional-deposit tier most exposed if stablecoins pay yield, not an estimate of money that has moved. A Coinbase-commissioned study using USDC as the proxy finds no statistically significant relationship between stablecoin growth and community-bank deposits; both sides' trillion- and billion-dollar lending effects are models.

The banks' case, in their own terms. Deposits fund loans and a full-reserve token does not; the $4.8 trillion of community-bank deposits fund about $4 trillion of local lending, and a product that pays deposit-like interest, is used like a deposit and is run-prone like a deposit, but sits outside insurance, capital, liquidity, supervision and the Community Reinvestment Act, is arbitrage rather than innovation. The run risk runs the other way too: USDC traded at $0.87 on 11 March 2023 with $3.3 billion of its reserves stranded at SVB, and it was the banking system's rescue that re-pegged it; Moody's counted more than two thousand stablecoin de-pegs in 2022. On the rails and the cost, the banks' answer is that the always-on dollar already exists inside the perimeter — The Clearing House's real-time network moves up to $10 million "24 by seven", in Bank of America's chief executive's words, and Zelle moved $1.2 trillion in 2025 at no charge to the consumer — and that a depositor who wants the policy rate is put into a money fund inside the same franchise ("we put the people in the market"). And the issuers' reserves are Treasuries and bank deposits — stablecoin growth is a bid for T-bills (USDC's reserve fund is custodied at BNY and managed by BlackRock), not an exit from the system.

What they are building. Within eighteen months of the GENIUS Act, JPMorgan put a deposit token on a public blockchain and ran more than $5 billion a day through its Kinexys network in April and, by its own count, more than $7 billion by late June; Citi settled about $1 billion a day of tokenised deposits in May and has been live in seven markets since 28 September; BNY launched tokenised deposits in January with Circle among the first clients and is the named cash custodian in Morgan Stanley's proposed bitcoin trust; twenty-one financial institutions including Bank of America, Citigroup, Goldman Sachs, Wells Fargo and PNC announced in a joint release on 1 September, carried on the Wells Fargo, Santander and BBVA newsrooms, a new company to be formed in the second half of 2026 and a jointly issued dollar stablecoin for the first half of 2027, with JPMorgan not among them; Schwab began spot bitcoin and ether trading in April and Morgan Stanley's E*Trade in July; BlackRock's IBIT holds $66 billion. Zelle, owned by the banks, moved $1.2 trillion in 2025 free to the consumer; it does not touch the merchant side of the cost claim.

The test: resisting the technology, or defending the perimeter. The record on the claim that the banks are using political and regulatory channels to resist change is a dozen documents, dated here, and it is precise. The trade groups asked Congress in August 2025 to extend the GENIUS Act's ban on issuer-paid yield to exchanges and affiliates; asked Treasury in November to read "interest or yield" as "any economic benefit… regardless of what it is called"; told the Senate in May 2026 that the compromise rewards clause "will enable evasion"; and on 13 July 2026 asked the Senate leadership to delete outright the clause allowing rewards "calculated by reference to a balance, duration, tenure" — the clause that would let a stablecoin pay what a savings account pays. They asked the OCC on 17 July 2025 to delay the trust charters of Circle, Ripple and Fidelity Digital Assets; the OCC conditionally approved five such charters on 12 December 2025, and on 2 October 2026 the community-bank association sued to vacate the OCC's March 2026 trust-charter rule, its interpretive letter and a later approval. The Bank Policy Institute said it was "deeply concerned" by the Kansas City Fed's grant of a limited master account to Kraken on 4 March 2026, and the ABA and the Consumer Bankers Association said in their 24 July 2026 release that "legal eligibility alone should not guarantee access" to the Fed's proposed payment account. They sued the day the open-banking data rule was finalised, 22 October 2024, on grounds of statutory authority, third-party liability and cost recovery, and the rule is still not in force; the regulator itself has since reopened it on the same points. They oppose the card-routing bill, a merchant-versus-bank fight rather than a crypto one. Banks with more than $50 billion of assets and their trade groups spent $87 million on federal lobbying in 2025 against the digital-asset sector's $41 million, and the American Bankers Association's own disclosures list the GENIUS, STABLE and CLARITY Acts, the Credit Card Competition Act, debit interchange and open banking as the issues lobbied.

Two readings fit the same documents. In the first, every ask targets the one feature that makes an on-chain dollar a better product than a checking account — a return on the balance — and the escalation from comment letter to lawsuit against the OCC's chartering framework is what resistance looks like. In the second, every letter accepts stablecoins as payment instruments, the asks are three textual edits rather than a ban, the ABA and the Consumer Bankers Association called the Fed's proposed non-bank payment account, in their release, "a prudent approach to responsible innovation if additional safeguards are adopted", the suit against the OCC turns on whether a trust charter without fiduciary business is lawful, and the same banks are live on-chain with deposit tokens and a consortium stablecoin. What the record does not show: that the banks caused the CLARITY Act's failed cloture vote on 15 September (contemporaneous reporting attributes it to the bill's ethics title), that any measurable deposit flight has occurred, or that the banks oppose blockchain settlement as such. On electoral money the visible ratio runs the other way: the crypto industry's Fairshake committee disbursed $196 million in the 2024 cycle and $94 million through August 2026; the bank side's electoral giving runs through corporate and association committees the house has not totalled.

The honest formulation is narrower than either side's: the banks face a real technological change, are adopting it on their own terms, and are using the political and regulatory channels available to them to keep a deposit-like return on an uninsured, unsupervised on-chain balance from being offered without a bank's obligations — the competitor's edge on one reading, the perimeter on the other. Whether that is resisting the technology or defending the perimeter is the reading each side will take of the same twelve documents, each of which is tabulated with both readings in the note's data. The dispositive documents are ahead: Treasury's and the OCC's final rules on affiliate yield (the statutory deadline passed on 18 July), the ICBA suit, and the consortium token's terms in the first half of 2027 — whether it pays anything at all.

The number. At 30 June the FDIC counted $326.7 billion of unrealized losses on bank securities, 5.5% of their amortized cost: $109.8 billion on available-for-sale bonds and $216.9 billion on held-to-maturity bonds. The loss was $690 billion at its September 2022 peak and $684 billion in September 2023, and had fallen to $306 billion by the end of 2025 before stopping. Over the last 19 quarters every 100 basis points on the 5-year Treasury has moved it by about $219 billion (a realized duration of 3.7 years, R² 0.93), and the books pull back toward par by about $21 billion a quarter when yields are flat. The 5-year rose 90 basis points between 30 June and 30 September. On that, the loss at the quarter-end is $497–517 billion: 8.4–8.7% of the book, 19% of the system's $2.64 trillion of equity against 32% at the 2022 peak, and three-quarters of the two peaks in dollars. The two nearest precedent moves on the 5-year (Q4 2024, +80 basis points; Q3 2022, +105) put it at $466–498 billion; the Q3 2023 quarter, a smaller move on a longer book, at $596 billion. Nobody has published a third-quarter figure; the only number in circulation is a syndicated summary of a Barron's piece on one bank's held-to-maturity book, Bank of America's, which puts it beyond $90 billion if rates rise another half point from its $82 billion at June.

What it is, and is not. A paper loss on a bond that is held to maturity is a locked-in below-market yield until the bond pays off: no margin call, no forced sale, just the difference between the 2.7% the book earns (Bank of America's average on $881 billion of debt securities) and the 5% new paper pays, for as long as the paper lasts. The $500 billion is the present value of that shortfall and the $21 billion a quarter is the shortfall being earned back. It becomes something else in two cases. The first is a forced sale, which needs deposits to leave; the quarter's deposits grew, and §9 leaves the question to two dated tells. The second is the rule: the available-for-sale part of the loss flows through AOCI into CET1 at the eight Category I firms among the twelve without any sale, and the agencies' March 2026 proposal, which they are reported to be finalizing at a December meeting, would phase it into the Category III and IV banks' capital from 2028 to 2032.

Why it heals slowly. Of the system's $5.84 trillion of securities, $1.70 trillion is mortgage pass-throughs with more than fifteen years of contractual maturity left: 29% of everything the banks hold, and $1.77 trillion at the end of 2022, so three and a half years of a 6–7% mortgage rate have taken 4% off it. The rest of the book has shortened; that part has not. At the pace of the last two years, the held-to-maturity book takes 20 years to run off at U.S. Bank, 17 at PNC, 16 at Truist, 14 at Bank of America, 11 at Schwab, 10 at JPMorgan Chase Bank, 8 at Wells Fargo Bank and four to five at Citibank and State Street. The line in Reuters' preview of the quarter, from a fund manager, that most banks have "reduced the duration of their portfolio" is true of the available-for-sale books, which are 40–64% hedged with derivatives at the six largest firms that disclose the amount and fit as 0.7–3.2-year books; it is not true of the held-to-maturity books, which fit as 3.4–6.8-year books, cannot be hedged for accounting purposes, and hold three-fifths of the projected loss. Wells Fargo's own expected maturities are 6.9 years on its available-for-sale book and 9.7 on held-to-maturity; U.S. Bancorp's 5.8 and 8.1. Only Citigroup, of the twelve, publishes a sensitivity of its capital to rates: $3.0 billion of AOCI and 18 basis points of CET1 per 100 basis points.

Who carries it. As a share of the bank's own equity at June: Charles Schwab Bank 67% (about 98% on the September curve), Bank of America NA 35% (46%), Wells Fargo Bank and U.S. Bank about 20% (about 30%), JPMorgan, Citibank, Morgan Stanley, PNC and BNY 6–12% (10–24%), Goldman under 1%. At the holding companies the same order holds at lower levels (Schwab 26% → 42%, Bank of America 28% → 37%, Wells and U.S. Bancorp 20% → 30%); Bank of America's $82 billion held-to-maturity loss, $68 billion of it on agency mortgage bonds, is the largest in dollars. The full table by company is in the note's data.

Whose capital it reaches. At the eight Category I firms the quarter's available-for-sale move after tax is worth about 29 basis points of CET1 at JPMorgan Chase Bank, 36 at Wells Fargo Bank, 40 at Morgan Stanley Bank, 26 at Goldman Sachs Bank USA, 10–11 at Bank of America NA and Citibank (short, hedged books), 29 at State Street Bank and about 100 at BNY (a large book on small risk-weighted assets, before hedges) — 4% to 16% of the gap between each bank's ratio and its parent's requirement. At Schwab, PNC and U.S. Bancorp it is worth nothing today (89, 29 and 29 basis points if it counted), by election; under the proposal's schedule, a fifth of it counts in 2028 and all by 2032. At full recognition the available-for-sale marks on the September curve are about 395 basis points of CET1 at Schwab's bank and 110–200 at a group of Category III and IV banks (Key, Ally, HSBC USA, Capital One, Fifth Third, Huntington, Citizens, Regions); Schwab's whole AOCI, most of it on paper moved to held-to-maturity, is about 680 basis points of the parent's risk-weighted assets.

The two ranges the move breaches on paper. Schwab manages to an adjusted Tier 1 leverage ratio that puts AOCI back in: 6.8% at June against a stated "long-term operating objective … of 6.75% - 7.00%", with about $160 million of headroom after a $1 billion buyback in the quarter. The quarter's available-for-sale move at the parent is $1.4 billion after tax on the fitted duration, which takes the ratio to 6.5% static, and $0.9–1.1 billion on Schwab's own stated durations of 1.9 years (with derivatives) and 2.3 years, which takes it to 6.6%; both are below the objective before third-quarter earnings, which at the second quarter's rate are worth 50–60 basis points of the ratio before buybacks. State Street's Tier 1 leverage ratio was 5.3% against a medium-term assumption of 5.25–5.75% in its own deck; the quarter's move is worth 13 basis points of it, to about 5.2%, before a quarter's earnings worth about 30. Both are self-set ranges, both already count the mark (State Street's as a GSIB, Schwab's by its own measure), and what the prints will test is whether earnings carried them or buybacks paused. Nobody on either July call, management or analyst, mentioned AOCI, the election or the proposal.

What the calls did not say. Roughly fifty analyst questions were put on the four money-center calls of 14 July; none asked about the securities book, the marks, AOCI or hedging. On the five other calls read for this note (Goldman, Morgan Stanley, Schwab, PNC, U.S. Bancorp) analysts asked about securities yields and repricing at Schwab and U.S. Bancorp, where management also discussed its 9.4% AOCI-inclusive CET1 ratio, and about nothing in the book at the other three; on none of the nine were the marks, the election or the proposal raised as a question. JPMorgan took $395 million of net securities losses in the quarter, mostly on sales of agency mortgage bonds, to reposition; PNC $139 million; across the eighteen bank repositioning trades filed since July, about $16 billion of bonds went for about $1 billion of losses, nine of them paired with one-off gains (six from the Visa share exchange), two of them transfers out of held-to-maturity, none citing the rate rise; JPMorgan's $395 million, taken from its 10-Q, sits outside that tally.

The tail, gamed: how a paper loss becomes a real one. The house does not expect what follows; it is the chain that would have to run for §8's number to matter, written so the steps can be watched. One. A bond bought at 100 that now trades at 85 still pays 100 at maturity. The bank's loss is the better yield it could have earned on that money: Bank of America's book yields 2.7% while the 5-year Treasury pays about 5%. Nothing forces that loss into the open as long as the bank can wait. Two. What takes the waiting away is a need for cash faster than the bonds pay it back — deposits leaving. The money that leaves is the money that is not insured and not tied to a payroll or a business account; at the lead banks it is 42–52% of deposits at the universal banks and regionals, 68–76% at the custody banks, 19% at Schwab — a low share that did not stop about 45% of Schwab's 2023 outflow being uninsured money. Three. The bank meets the first outflows without selling anything: cash on hand, then borrowing from the Federal Home Loan Banks and the Fed against the bonds (at market value, unless the Fed chooses to lend at face value as it did in March 2023), then repo. Schwab's 2023 is the template for this stage: $77 billion of deposits left in a year, it borrowed $26 billion from the Home Loan Banks at about 5% against bonds earning about 2%, sold only $8.5 billion of bonds (at a $62 million loss), and paid for it in earnings. Four. The point of no return is the held-to-maturity box. Under the accounting rules, outside narrow exceptions, a bank that sells one bond out of it must mark the whole box to market: the entire loss, deferred for years, arrives on one day and in public. Five. That announcement is what starts the run rather than ending it — Silicon Valley Bank disclosed a $1.8 billion after-tax loss on the sale of its available-for-sale bonds on a Wednesday and received withdrawal requests for $42 billion on the Thursday, before any held-to-maturity bond was touched: the market treats a disclosed loss on any bond sale as the box being opened. The acute sequence is disclosed loss → run; the slow outflow of steps two and three precedes both. Six. The size of the day-one hit, at the 30 September curve, as a share of the lead bank's own equity: Schwab Bank 98%, Bank of America NA 46%, Wells Fargo Bank and U.S. Bank about 30%, BNY 24%, State Street 20%, PNC 16%, Morgan Stanley Bank 15%, JPMorgan Chase Bank and Citibank about 10%, Goldman 6%. If a sovereign-debt shock added 300 basis points to the 5-year, the system's mark would go from about $509 billion to about $1.17 trillion — 20% of the book's cost and 44% of the system's equity, against 32% at the 2022 peak — and the holding-company marks to 92% of equity at Schwab, 68% at Bank of America, 64% at U.S. Bancorp, 59% at Wells Fargo, 45% at State Street and PNC, 46% at BNY, 19–26% at JPMorgan, Citigroup, Morgan Stanley and Goldman. Seven. Who could meet a 2023-style run without opening the box (Exhibit 9): with cash, reverse repo and available-for-sale bonds at market against every uninsured dollar, Morgan Stanley Bank 1.56×, Schwab Bank 1.54× (at least 0.55× once pledged securities are excluded), Goldman 1.18×, BNY 1.13×, Citibank 0.98×, State Street 0.93×, JPMorgan 0.90×; then Bank of America NA 0.68×, Wells Fargo Bank 0.66×, U.S. Bank 0.55×, PNC 0.50× (at least 0.16× ex-pledged). The long-bond banks are the low-cover banks; what protects them is the quality of the deposits, consumer and operating money that did not move in 2023.

Which crisis, and why the two compound. A credit bust — private credit, AI infrastructure — is a loan-loss event, and on the house's reasoning a rates-down event, as in 2008 and 2020: a flight to quality shrinks the paper loss by about $219 billion per 100 basis points on the 5-year, and the bond book becomes the liquidity rather than the problem; the damage runs through provisions on the $1.5–2 trillion of lending to nonbanks, where a 5% loss is $75–100 billion, a fifth to more than a quarter of a year of sector earnings, and the first loss sits with the funds and insurers; a loan-loss event in which rates do not fall is the compounding case below. A sovereign-debt shock is the opposite case: rates up, so the mark grows while the asset everyone is selling is the collateral the banks would pledge, and the 2023 fix — the Fed lending at face value against underwater Treasuries and mortgage bonds — becomes the central bank financing a stressed sovereign, the circuit breaker weakest exactly when it is needed. The two compound through liquidity, not by adding the losses: credit losses consume the capital and earnings that would absorb a crystallized mark; the nonbank borrowers draw their committed lines as the deposits leave, so cash goes out of the loan door and the front door at once; the cover ratio falls on both sides as the bonds that are the cover lose value; and the same balance sheets carry both legs — Wells Fargo Bank and U.S. Bank, two of the four with sub-0.7× cover (Bank of America NA and PNC are the others), hold 31% and 20% of their loans in the "other loans" bucket that carries the lending to nonbanks, beside held-to-maturity books that take eight and twenty years to run off. From 2028, if the March proposal is finalised, the Category III and IV banks would be counting bond marks in capital on a five-year schedule, so a stress that lands mid-phase-in would meet a rule eating capital during the run. What stands between that chain and a failure: hedges on 40–64% of the available-for-sale books at the largest firms, deposit quality at the consumer banks, sector earnings of about $360 billion a year against a $1.2 trillion tail mark (three years of earnings, not a write-off), and the fact that in a crisis the largest banks receive deposits. The shape of the tail is therefore not a story about small banks: two of the four thinnest covers are GSIBs, Bank of America and Wells Fargo, whose deposit quality and, outside the sovereign case, access to the Fed are the mitigants, and two are Category III banks, U.S. Bank and PNC, which §1 records as acquirers (Union Bank, National City), not casualties. The distressed firms §1 records that changed hands — Washington Mutual, Wachovia and First Republic — went to JPMorgan and Wells Fargo, and Countrywide and Merrill Lynch to Bank of America; Citigroup, the largest of them, was kept standing by a Treasury, Fed and FDIC guarantee; Silicon Valley Bank and Signature were closed.

What we will be watching, in plain terms. Six things, in the order the week will show them. The cost of money. Twelve of the thirteen lead banks paid less for their funding in June than a year earlier; the September hike pushes the other way, and the first number to read in each release is the cost of funds (the rate on interest-bearing deposits in the release), with the deposit beta each management states beside it. The guide. No July outlook assumed more than one hike, late in the year — Bank of America's chief financial officer assumed September and called it "net-net-net, it's a positive for us", Schwab's assumed December and said it "doesn't impact the financials for 2026", PNC's chief executive called the effect "very small in 26", U.S. Bancorp's "not consequential"; the net interest income guide against a hike that has now happened says whether those assumptions held. Who kept the deposits. The noninterest-bearing share and the uninsured share after the hike, and Schwab's September cash report on the 14th or 15th: the second quarter's deposit growth was all uninsured, and the question is whether that money stayed when rates rose. Credit. Card charge-offs at Citigroup and Capital One, and any sentence about lending to nonbanks — a balance, a reserve, a markdown — from JPMorgan, Wells Fargo or Morgan Stanley, the three banks tripwire five tracks, and if balances are broken out, whether the allowance against them is in line with the bank's commercial book, which would close the question rather than open it. The bond marks. The change in AOCI at each of the eleven with books against the projections in §8, and whether anyone sold at a loss. Capital return. The buyback pace at Schwab and State Street, whose own ranges the quarter's move reaches. Part 2 scores all six on Friday 16 October; where a watch item has a number that can be frozen, a tripwire below freezes it.

Tripwires we will track. Thirteen, frozen here and scored in Part 2 on Friday 16 October; each has a silence branch. One, the marks. Each of the eleven with books reports its third-quarter AOCI change and, where the release carries it, its available-for-sale and held-to-maturity marks; scored "larger" if the securities component of the AOCI change exceeds 1.4 times the available-for-sale projection in the note's data, "smaller" under 0.6 times, "inside" between; held-to-maturity marks not disclosed until the 10-Q are scored "pending". Two, the reach. At the eight Category I firms, the AOCI line of the CET1 roll against the basis points in §8, scaled to the parent where the roll is reported there; fires if the sign is wrong or the size outside half to one-and-a-half times; not broken out on release day, scored on the 10-Q. Three, crystallization. Any of the twelve reports a securities sale at a loss, a repositioning, or a held-to-maturity transfer or sale; fires on any, silence scored as none. Four, the election. Schwab, PNC or U.S. Bancorp states how it would treat AOCI under the proposal or changes its election; silence scored as none. Five, nonbank lending. JPMorgan, Wells Fargo or Morgan Stanley discloses balances, reserves or markdowns on lending to nondepository financial institutions, or fields a question on it; fires on any markdown or reserve build; a first-time breakout of balances without either is scored "disclosed", favorable where the allowance is in line with the bank's commercial book, since it closes the visibility gap the steelman's first point rests on; silence scored as such, on the banks' own documents only. Six, the system. The FDIC's third-quarter total against $497–517 billion, due on the FDIC's usual cadence in late November. Seven, the two ranges. Schwab's preliminary adjusted Tier 1 leverage ratio (15 October) and State Street's Tier 1 leverage ratio (14 October): below 6.75% and 5.25% respectively fires; inside the range is scored "held, by what"; not in the release, scored on the 10-Q. Eight, JPMorgan's line. The Corporate "investment securities gains/(losses)" line negative again and the 10-Q repeating the repositioning sentence; positive or zero scored as stopped. Nine, Citigroup's gauge. The 10-Q's AOCI at risk per 100 basis points against $3.0 billion and 18 basis points at June; outside $2.5–3.6 billion fires; not updated, scored on the 10-K. Ten, the December rule. The agencies finalize the capital package at a December meeting with the AOCI phase-in at the proposed schedule from the 1 January 2027 date the proposal uses for illustration; a faster or slower schedule or a different date fires; not finalized by 31 December, scored as slipped. Eleven, Schwab's sweep. The September monthly activity report: transactional sweep cash against $485.7 billion at June and $483.3 billion at August; below $473 billion fires; flat or up, the deposit question stays closed this quarter. Twelve, PNC's swaps. The third-quarter AOCI change split between the securities and the $78 billion receive-fixed swap book at 3.75%; a swap-driven change larger than the securities' fires; the split not given on release day, scored on the 10-Q. Thirteen, the tail. Federal Home Loan Bank advances or other borrowed money (where Federal Reserve borrowing sits in the Call Report) up more than 50% and more than $5 billion quarter on quarter at any of the eleven lead banks whose deposits fell more than 3% in the same quarter, in the third-quarter Call Reports (published by the FDIC in late November) — on the 2023–26 record that pairing fired only in the quarter of the 2023 run, at Schwab; borrowing that grows while deposits hold is balance-sheet management and does not count — or any of the twelve reporting a quarter-on-quarter fall in deposits of more than 3% together with a securities sale at a loss on release day; fires on either; neither on release day is scored as none, and the Call Report leg is scored "pending" until the FDIC's late-November data; with tripwire three it is the pair whose firing would move the house's reading of §8 from an earnings drag to a liquidity question. Where a watch item has a frozen number it is scored by a tripwire — the marks by one to three, nonbank lending by five, the two ranges by seven, Schwab's cash by eleven; the rest, the cost of money, the guides, the noninterest-bearing and uninsured shares of deposits, and card credit, are scored in Part 2 as a narrative, because none of the thirteen tripwires freezes a number for them (the cost of money and the deposit shares carry thresholds in the reversal block above, which Part 2 will also read).

The dates. Consensus EPS as carried by Robinhood on 8 October, archived with the note's data. Tuesday 13 October before the open: JPMorgan ($5.91), Goldman Sachs ($13.79), Citigroup ($2.63), Wells Fargo ($1.85). Wednesday 14: Bank of America ($1.12), Morgan Stanley ($2.95), BlackRock ($14.15), State Street ($3.66). Thursday 15: Schwab ($1.66), PNC ($4.91), U.S. Bancorp ($1.32), BNY ($2.27). All twelve dates are company-confirmed. After their prints, JPMorgan, Goldman Sachs and BlackRock are valued as the three archetypes.

What the note cannot settle. Whether the collateral behind the nonbank lending is good. Whether a held-to-maturity loss that is never counted is ever paid. What the parents bought, sold or hedged between July and September. Whether the December meeting happens and what schedule it sets. And whether a growing uninsured deposit base is a strength or a fuse.