The Frozen Balance Sheet

I — The Classification Decision

When a bank acquires a bond, it makes an accounting election with consequences that can persist for a decade or more. Under US GAAP, ASC 320 assigns debt securities to one of three categories: trading (marked to market through the income statement), available-for-sale (marked to market through other comprehensive income), and held-to-maturity (carried at amortised cost, with fair value disclosed only in footnotes). (FASB, ASC 320, 1993) The choice is formally governed by intent — an entity may classify bonds as HTM only if it has the "positive intent and ability" to hold them to maturity — but in practice it is a management discretion with limited external verifiability. (ASC 320-10-25-1(c))

The economic substance of the three categories is identical. A ten-year Treasury bond yielding 1.5% purchased in 2021 has the same market value on any given day regardless of which ledger it sits in. What differs is disclosure: under HTM classification, the $20 decline in that bond's price when yields rise to 4.5% appears nowhere on the income statement and nowhere in equity. It surfaces only as a parenthetical in the footnotes — visible to a careful analyst, invisible to a rushed regulator or a depositor reading a headline.

IFRS 9, which governs European and most non-US banks, takes a structurally similar approach but frames the test differently. Classification depends on two conditions: the entity's business model for managing the asset (hold to collect, hold to collect and sell, or other) and whether the instrument's contractual cash flows represent solely payments of principal and interest — the SPPI test. (IFRS 9, IASB, 2014) A bond held in a "hold to collect" business model with standard principal-and-interest terms qualifies for amortised cost measurement. The practical outcome mirrors US GAAP: interest-rate movements on such bonds pass through neither the income statement nor equity until realised.

Both frameworks acknowledge the tension. IFRS 9 restricts reclassification out of amortised cost to genuine changes in business model — rare in practice. ASC 320's taint mechanism operates differently: it permits transfer from AFS to HTM at any point (locking in the existing unrealised loss in accumulated other comprehensive income, to be amortised over the remaining life), but any subsequent sale from the HTM book, absent narrow exemptions, taints the entire portfolio and forces reclassification of all remaining HTM securities to AFS. (ASC 320-10-25-9; Crowe LLP, 2023) The asymmetry matters. The door to HTM is open; the door out is nearly sealed.

II — What Amortised Cost Conceals

The gap between amortised cost and fair value on a fixed-rate bond is a function of duration, the magnitude of rate movement, and time remaining to maturity. For a ten-year bond purchased at par with a 1.5% coupon, a 300 basis point rise in the risk-free rate produces a mark-to-market loss in the range of 20 to 25 percent of face value. Held at amortised cost, that loss is simply absent from the balance sheet.

This is not a rounding error. In the aggregate across the US banking system, unrealised losses on AFS securities reached $204 billion at end-2023, while losses on HTM portfolios — which appear nowhere in regulatory capital — reached $274 billion. (Congressional Research Service, 2024) Total system-wide unrealised losses thus exceeded $478 billion, against net income for the full year 2022 of approximately $263 billion. (Atlantic Council, 2023) The HTM component, the larger of the two, was structurally invisible to capital adequacy calculations.

The regulatory underpinning of this invisibility is the AOCI opt-out filter — a provision introduced in the US in 2013 that permitted all but the very largest banks to exclude accumulated other comprehensive income (which captures unrealised AFS movements) from Common Equity Tier 1 capital. (Federal Reserve, Basel III implementation, 2013) The result was a two-layer shield: HTM losses were excluded from the balance sheet entirely, and AFS losses, while recognised in equity, were excluded from the regulatory capital ratio that supervisors and investors use to assess solvency. A bank could carry bonds worth 80 cents on the dollar and appear well-capitalised on every metric that supervisors monitored in routine examinations.

The accounting does not reduce the loss. It determines who bears the information cost — and for how long.

Proponents of amortised cost accounting argue, not unreasonably, that for a genuine hold-to-maturity investor the day-to-day market price of a bond is economically irrelevant: principal is repaid at par, coupons arrive on schedule, and interim fluctuations cancel to zero over the holding period. The argument is coherent for an insurance company or a pension fund with long-dated, relatively stable liabilities. It is considerably less coherent for a commercial bank funded primarily by demand deposits — liabilities that are contractually short regardless of how long the bank intends to hold its assets.

III — The Mechanism of Amplification

A bank run does not require that depositors discover the bank is insolvent. It requires only that they become sufficiently uncertain — that the expected cost of staying exceeds the transaction cost of leaving. Accounting opacity affects both sides of that equation. It prevents depositors from forming accurate assessments of the bank's condition, and it prevents the bank from taking corrective action before the threshold is reached.

The amplification mechanism operates in three stages. First, opacity delays. HTM classification means that a deteriorating portfolio generates no income-statement signal, no capital-ratio deterioration, and no supervisory escalation trigger for as long as the bank does not sell. Management has both the accounting latitude and the regulatory incentive to defer any recognition. During this phase, the true economic condition of the bank diverges from its reported condition, and the gap widens with each rate hike.

Second, forced realisation concentrates the signal. When a liquidity event — accelerating deposit outflows, margin calls, or a credit rating downgrade — compels the bank to sell securities, it must sell at market price. That realised loss is then visible simultaneously to regulators, analysts, rating agencies, and depositors. The accounting switch from concealment to revelation is instantaneous. There is no gradual disclosure.

Third, the revelation triggers rational flight. A depositor who held stable expectations suddenly receives information that the bank's capital position is materially worse than previously disclosed. If a significant proportion of deposits are uninsured — above the statutory guarantee threshold — the rational response is to withdraw. This is not panic in the behavioural sense; it is a Bayesian update on new information that the accounting framework had suppressed.

The taint mechanism intensifies the second stage. Once any HTM security is sold, the entire portfolio must be reclassified to AFS, forcing every unrealised loss into equity at once. A bank that sells $1 billion of HTM bonds to meet deposit withdrawals may trigger the recognition of $15 billion in losses across the reclassified portfolio. The very act of seeking liquidity produces a capital shock that accelerates the liquidity crisis it was meant to address.

IV — Silicon Valley Bank: The Mechanics in Practice

SVB's balance sheet in the years before its March 2023 failure illustrates the mechanism with unusual clarity because the numbers were large, the timeline was compressed, and the public record — assembled by the Federal Reserve's Office of Inspector General — is detailed. (Federal Reserve OIG, 2023)

Between 2019 and 2021, SVB grew total assets from roughly $71 billion to $211 billion as pandemic-era stimulus flooded the technology sector with deposits. Management invested the proceeds predominantly in long-duration mortgage-backed securities and Treasuries, seeking yield in a near-zero rate environment. In March 2022, SVB reclassified approximately $91 billion of these securities from AFS to HTM. The reclassification locked in the then-existing unrealised loss in AOCI but shielded the portfolio from any further mark-to-market deterioration as the Federal Reserve began its tightening cycle. (Federal Reserve OIG, 2023)

SVB portfolio summary — year-end 2022

The HTM book stood at $91.3 billion at amortised cost against a disclosed fair value of $76.2 billion: an unrealised loss of $15.1 billion, unrecognised in capital.

The AFS book held $26 billion at fair value, carrying $2.5 billion of unrealised losses — recognised in AOCI, but excluded from CET1 through the opt-out filter.

Roughly $17.6 billion of losses therefore sat outside the capital calculation, against reported regulatory capital of approximately $15 billion.

The critical observation is that SVB's disclosed fair-value footnotes contained the entire story. The $15.1 billion HTM gap was reported every quarter as it accumulated through 2022. (SMU Cox School of Business, 2023) The information was public. But it did not appear in the capital ratios that supervisors monitored, it did not trigger supervisory escalation, and it did not prompt the bank's board to act with urgency. The AOCI opt-out — which the Federal Reserve estimated inflated SVB's CET1 ratio by nearly two percentage points — ensured that the reported number remained comfortably above minimum thresholds throughout. (Congressional Research Service, 2024)

The collapse followed a textbook three-stage sequence. Venture capital funding began contracting in 2022, forcing technology companies to draw down SVB deposits to fund operations. Deposit outflows accelerated. To raise cash, SVB liquidated its entire AFS portfolio in early March 2023 at a $1.8 billion after-tax loss and simultaneously announced a $2.25 billion capital raise. (Federal Reserve OIG, 2023) The announcement — made on the same afternoon that Silvergate Bank announced voluntary liquidation — communicated distress rather than stability. Prominent venture capital firms advised portfolio companies to withdraw funds. Within 24 hours, customers had requested $42 billion in withdrawals, nearly 25 percent of total deposits. The California Department of Financial Protection and Innovation closed the bank on March 10, 2023. It was the second-largest bank failure in US history by assets.

Two subsidiary points deserve note. First, SVB's bonds were not exotic instruments. The HTM portfolio was 75 percent agency residential mortgage-backed securities and 16 percent agency commercial mortgage-backed securities — assets with no material credit risk, carrying the implicit backing of the US government. (GAAP Dynamics, 2023) The problem was duration, not credit quality: government bonds in a rising-rate environment. Second, the speed of the run — $42 billion in 24 hours — reflected the concentration of SVB's depositor base. Over 90 percent of deposits were uninsured, held predominantly by technology companies with treasury functions sophisticated enough to move funds within hours. Digital banking removed all friction from the exit decision.

V — Contagion and the System-Wide Dimension

SVB's failure did not remain contained. On March 12, 2023, New York regulators closed Signature Bank — the third-largest US bank failure in history — after it experienced over $10 billion in single-day withdrawals driven by depositor panic amplified by SVB's collapse. (Federal Reserve, April 2023) First Republic Bank, perceived as similarly exposed, suffered deposit outflows exceeding $100 billion in the weeks that followed and was eventually acquired by JPMorgan Chase in May 2023 under FDIC facilitation.

The contagion mechanism was informational rather than directly financial. Investors and depositors, newly alert to the existence of large unrealised bond losses across the banking system, began scanning balance sheets for institutions with elevated HTM portfolios, high concentrations of uninsured deposits, and limited hedging. The KBW Bank Index fell sharply; equity markets in effect performed the risk-screening that the accounting framework had failed to provide. (Chicago Fed Working Paper, 2025)

The system-wide fragility had been building since 2020. The Federal Reserve estimates that US banks added approximately $2.3 trillion in securities between the start of 2020 and end-2021, almost entirely at long duration and near-zero yields. The subsequent rate cycle — 525 basis points of tightening between March 2022 and July 2023 — generated unrealised losses at a scale that had no modern precedent in the banking system. That the accounting framework permitted the vast majority of those losses to remain invisible to capital calculations was, in retrospect, a structural failure of disclosure rather than a natural feature of sound banking practice.

VI — Mitigation Options

Responses to the SVB episode fall into three broad categories, each operating at a different layer of the problem.

Accounting reform

The most direct intervention is to require fair-value measurement for all bank securities, eliminating the HTM category or at minimum prohibiting mid-life AFS-to-HTM transfers. The case for this is straightforward: if commercial banks are funded by short-term deposits, they do not in any economically meaningful sense hold assets to maturity — the deposit base can and does disappear before the bonds mature. Forcing fair-value recognition would align reported capital with economic reality and restore the income-statement signal that the HTM classification suppresses. (GAAP Dynamics, 2023)

The objection — that mandatory mark-to-market creates pro-cyclical income-statement volatility and penalises genuine long-term holders — is not without merit. A bank that actually does hold a thirty-year bond to term suffers no cash loss from interim rate movements; requiring it to report paper losses in income is arguably misleading in the opposite direction. The practical resolution would be to prohibit reclassification from AFS to HTM after initial acquisition, which would allow intention-based classification while eliminating the strategic use of HTM as a loss-quarantine mechanism.

Regulatory capital treatment

In July 2023, US federal banking regulators proposed extending the AOCI inclusion requirement — currently applicable only to Category I and II banks — to all institutions with over $100 billion in assets. (Federal Reserve, FDIC, OCC, Basel III Endgame Proposal, 2023) This would increase the number of banks required to reflect AFS unrealised losses in their CET1 capital from nine to 37. The AOCI opt-out, introduced as a compromise in 2013 and further narrowed under the 2019 regulatory tailoring rule, allowed SVB to report a capital ratio roughly two percentage points above what full AOCI inclusion would have shown. Removing that filter means that a rising-rate environment reduces reported capital ratios automatically, triggering supervisory attention before the bank is forced into asset sales.

The limitation of this approach is that it applies only to AFS securities. HTM unrealised losses — which exceeded AFS losses in aggregate at end-2023 — remain entirely outside the capital framework under both the existing rules and the proposed Basel III Endgame amendments. The rationale offered is that if a bank genuinely holds bonds to maturity, it will never realise those losses. This is theoretically correct. It is also the argument that SVB's management made until March 8, 2023.

Liquidity backstop design

The Federal Reserve's Bank Term Funding Program (BTFP), announced on March 12, 2023, addressed the immediate systemic threat by offering one-year loans to banks and depository institutions against eligible collateral — US Treasuries, agency securities, and mortgage-backed securities — valued at par rather than market price. (Atlantic Council, 2023) The BTFP effectively neutralised the forced-realisation mechanism: a bank facing deposit outflows could pledge underwater HTM bonds at face value to obtain liquidity without selling, avoiding the loss recognition that triggers taint and further capital shock.

The BTFP is an emergency instrument rather than a structural solution. It removes the liquidity pressure that forces realisation, but it does not reduce the underlying interest-rate risk; it merely socialises the carrying cost of that risk through the central bank's balance sheet. If rates remain elevated and the bonds' below-market coupons persist for years, the economic loss accrues somewhere — to the bank's net interest margin, to the Federal Reserve, or ultimately to the taxpayer.

A more durable variant would be the permanent extension of discount-window access on terms that reflect collateral quality rather than market price, combined with a framework that incentivises hedging of duration risk at origination. Hedge accounting under both ASC 815 and IFRS 9 is available for AFS securities; it is not available for HTM securities, since GAAP reasons that a hold-to-maturity investor has no market-price exposure to hedge. This exclusion is a further consequence of the original classification choice: once a bank declares it will never sell, it forfeits both the accounting signal and the hedging permission that might have managed the risk.

— The Fundamental Mismatch

The accounting debate around HTM classification is in some respects secondary to a more foundational problem: the banking business model pairs long assets with short liabilities, and no accounting framework can alter that underlying structural exposure. A bank that funds a ten-year bond portfolio with overnight deposits is exposed to both interest-rate risk and liquidity risk regardless of where it draws the HTM classification line. The classification determines visibility, not vulnerability.

What accounting standards do determine is the speed and legibility of the feedback loop. Under full fair-value accounting, a bank that makes a poor duration bet sees the consequence immediately in its capital ratio, its stock price, and its supervisory ratings. Under HTM accounting, the consequence is deferred until a liquidity event forces realisation — at which point the information arrives in concentrated form, simultaneously visible to all counterparties, without the gradual disclosure that might have allowed orderly adjustment.

There is a reasonable case that the HTM category serves genuine purposes for institutions with genuinely stable, long-term funding bases. The problem is that commercial banks with deposit-funded balance sheets have successfully argued for its application to themselves, exploiting a distinction — between intent and economic reality — that the accounting standard cannot adequately police. The result is a periodic pattern: rates rise, losses accumulate invisibly, a forced-realisation event concentrates the signal, depositors update rapidly, and the run follows. SVB was not the first iteration of this sequence, and the proposed regulatory responses, valuable as they are, do not address the foundational asymmetry between how assets and liabilities are measured.

The question is not whether a bank intends to hold its bonds to maturity. The question is whether its depositors intend to stay that long.