Held to maturity

A bond does not lie about its value: the market sets it every day. A bank, though, has a choice the market does not, the choice of whether to look. US accounting offers it, for that, a reassuringly named category, held-to-maturity, in which a bond stays recorded at its original cost, indifferent to its real value, as long as the bank promises not to sell it. At the end of March 2026, US banks lodged $214 billion of unrealized losses there, invisible on the balance sheet by the sole grace of a promise. The promise holds as long as nothing forces the sale. In March 2023, Silicon Valley Bank discovered, in a few hours, the fragility of that conditional.

Three drawers for one bond

The same bond changes nature depending on the accounting drawer the bank files it in, and the choice of drawer governs the value the world will grant it. The first drawer, trading, marks the bond to market and runs the swings through profit and loss: the loss shows at once. The second, available-for-sale, also marks the bond to market, but lodges the loss in equity, without touching earnings: the loss shows, discreetly, on the balance sheet. The third, held-to-maturity, marks nothing at all. The bond stays at amortized cost, as if its value had not moved, and the loss appears nowhere, neither in earnings nor in equity. It exists economically, it is absent in the accounts.

The scale of the silence can be measured. The FDIC’s quarterly profile puts banks’ total unrealized losses at $325.1 billion in the first quarter of 2026, of which $214.5 billion in held-to-maturity and $110.6 billion in available-for-sale. Two thirds of the loss, in other words, sleep in the one drawer where the rule permits not counting it. The rate rise of 2022 gouged the price of the long bonds bought when money was free; held-to-maturity made it possible not to keep the ledger of it.

The move of 2022

The rule offers more than shelter, it offers an exit. A bank whose available-for-sale securities were bleeding into equity as rates rose could reclassify them into held-to-maturity, freezing the loss at the transfer date and, from then on, ceasing to remark it. In January 2024, Fifth Third thus transferred $12.6 billion of securities from available-for-sale to held-to-maturity, carrying $994 million of unrealized losses out of the reach of remeasurement. The operation erases not a cent of loss; it moves it to the blind spot.

Academics watched that move closely, and their conclusion is uncomfortable. A study by the Becker Friedman Institute in Chicago on bank fragility and reclassification into held-to-maturity shows that the banks reclassifying the most were also, on average, the most fragile: the accounting move was not neutral, it signalled a strain it served precisely to mask. The gesture that renders the loss invisible is also the one that betrays its weight.

The filter that neutralizes even the visible losses

The visible part of the loss, that of available-for-sale securities, nonetheless does not dent the banks’ stated strength, and the reason is a second device, less known than the first. Regulatory capital, the famous ratio that measures solvency, excludes these losses thanks to an option, the AOCI filter, which about 99% of US banks have used. In plain terms, a bank can display a comfortable capital ratio while carrying, in its equity, losses that genuinely impoverish it. Held-to-maturity removes the loss from the balance sheet; the AOCI filter removes from regulatory capital the part of the loss the balance sheet does acknowledge. Between the two sieves, the number that governs confidence, the solvency ratio, ignores most of the gap between the bonds’ book value and their real value.

March 2023, the promise that breaks

The defensive reasoning is solid, and it must be laid out before it is tested. A government bond held to maturity repays its principal in full; its market loss is only a bump along the road, bound to fade as the bond nears its term. Held to the end, it costs nothing. In that light, held-to-maturity does not mask a loss, it avoids recording a loss that will never be one. The argument is right, on one condition, a single one: that the bank can actually hold to maturity, hence that it is never forced to sell.

Silicon Valley Bank lived the fall of that condition in real time. Sitting on a base of volatile deposits, it had piled up long bonds classified held-to-maturity, whose unrealized losses did not appear in its ratios. When its depositors moved to withdraw their funds, it had to sell those securities to find cash, and the sale turned, in a few hours, an accounting non-loss into a real and fatal one. The “held-to-maturity” category is a bet on calm; a run is exactly the stress that voids the bet, at the precise moment the hidden loss turns lethal. The accounting had given no warning, not by error, but by design. To postpone the fatal instant, SVB had, moreover, funded itself at the window of the Federal Home Loan Banks, that lender of next-to-last resort which lets a bank push back the sale, and thus the recognition, a little longer.

The remedy in reverse

The lesson of 2023 had been drawn, on paper. The major prudential reform known as the Basel III endgame planned to remove the AOCI filter for banks above one hundred billion dollars in assets, forcing them to make their available-for-sale losses finally weigh on their capital, with a phase-in beginning 1 July 2025. The correction aimed straight at SVB’s flaw. It had two limits, though: it touched only available-for-sale, never held-to-maturity where the largest share of the loss sleeps, and it covered only the biggest banks.

Those limits are now the least of it, because the reform itself is ebbing. Signals from the Federal Reserve point to a capital-neutral Basel III endgame in 2026, a polite formula for saying the tightening will be gutted. The regulatory unwinding we documented elsewhere, that rollback on bank capital, carries away with it the one measure that answered directly the cause of the most spectacular failure since 2008. Three years after SVB, the flaw is wider than it was then in held-to-maturity terms, and the warning light someone had set out to switch back on is being unscrewed.

One must guard against catastrophism, and the qualification is the same as the defence: most of these $214 billion will fade if the banks are never forced to sell, and many will not be. The danger is not a $214 billion hole opening tomorrow. It is narrower, and more insidious: the accounting certifies that a bank is fine right up to the instant it is not, and the condition that turns the invisible loss into a fatal one, a deposit flight, is precisely the one no balance sheet displays. Held to maturity is not a lie, it is a promise, and a promise is worth only the stability that lets it stand. The market, for its part, does not file its prices in drawers. It knows the value of a bank’s bonds before the bank does, and it does not wait for the run to remind it.

---

Sources

• FDIC, “Quarterly Banking Profile, First Quarter 2026” (total unrealized losses $325.1bn, of which $214.5bn held-to-maturity and $110.6bn available-for-sale)
• Becker Friedman Institute, University of Chicago, “Bank Fragility and Reclassification of Securities into HTM” (the most fragile banks reclassified the most; 99% of banks use the AOCI filter)
• Securities and Exchange Commission, Fifth Third Bancorp, Form 10-K 2023 (transfer of $12.6bn from AFS to HTM in January 2024, $994m of losses at transfer)
• Congressional Research Service, “Banks’ Unrealized Losses, Part 1: New Treatment in the Basel III Endgame Proposal”, IN12231 (removal of the AOCI filter for banks above $100bn, phase-in from 1 July 2025)
• Bloomberg, “Fed remarks point to capital-neutral Basel III Endgame in 2026”
• Office of Financial Research, “The State of Banks’ Unrealized Securities Losses”, May 2025 (scale and dynamics of unrealized losses)