The Liquid Bond That a Bank May Not Be Able to Use
U.S. banks’ investment securities portfolios are the subject here, and the question is specific: why did banks not simply sell liquid bonds when interest rates surged in 2022–23? The answer matters because a bond can be easy to trade in a market and still be difficult for a bank to use as a shock absorber. Understanding that difference helps readers judge whether a bank has actually reduced its exposure, merely reduced the exposure visible in its accounts, or preserved a short-term accounting position at the cost of future flexibility.
The central problem is not that banks forgot what interest-rate risk was. It is that several reasonable forms of adjustment carried different costs at the same time. A bank could stop buying long-duration bonds, sell bonds it already owned, use derivatives to hedge the exposure, or classify some securities as held-to-maturity. Those choices are not interchangeable. One changes the flow of new risk; another realizes a loss immediately; a third requires an accounting and operational infrastructure; a fourth can make risk less visible without changing the bond’s market value. The 2022–23 episode shows what happens when those distinctions become consequential all at once.
A bond can be liquid for the market and costly for the owner
A marketable bond has a price and can, in principle, be sold quickly. That is the ordinary meaning of liquidity. But a bank does not experience a sale only as an exchange of an asset for cash. It also experiences the sale through its income statement, equity, regulatory capital, funding position, and relationship with investors and depositors. If the bond is trading below the value at which the bank has been carrying it, selling turns an unrealized loss into a realized one. The market has already recognized the loss; the bank’s accounts and capital measures may not have done so in the same way.
The Philadelphia Fed working paper uses Silicon Valley Bank to make this pressure concrete, while warning against treating one bank as the whole explanation. A quotation reproduced in the paper estimated that by March 31, 2022, SVB had about $7 billion in market-value losses against roughly $16 billion of equity. Selling underwater securities could have shortened the maturity of the portfolio, but it would also have recognized a loss large enough to consume almost half of that equity. The example illustrates why “sell the liquid bonds” can be a dangerous instruction when the transaction itself becomes a public capital event. It does not establish that every bank faced SVB’s exact balance-sheet or funding path, nor does it prove that selling would have been the best economic decision in every case.
The distinction can be summarized as three different questions:
- Economic exposure: How much can the value of the bank’s assets and liabilities change when rates move?
- Adjustment capacity: Which actions can the bank actually take without unacceptable accounting, regulatory, operational, or funding costs?
- Accounting visibility: How much of that exposure appears immediately in the measures that investors, regulators, and managers watch?
Confusing these questions produces the wrong diagnosis. A bank can have a highly liquid bond portfolio but limited adjustment capacity. It can reduce accounting visibility while leaving economic exposure high. And it can lower future risk by buying less without repairing the risk already embedded in its existing holdings.
The important implication is not that a bank should never sell. It is that the usefulness of a sale depends on which objective the bank is protecting. Selling can reduce duration and raise cash, but it can also expose a capital shortfall immediately. Holding the bond can avoid that recognition for a time, but leaves the market-value risk in place. The decision is therefore a constrained choice among imperfect margins, not a simple test of whether a security can find a buyer.
The 2022–23 rate surge made the trade-off visible
The episode studied by the Philadelphia Fed is unusually revealing because the shock created pressure in several directions. The Federal Reserve raised the federal funds rate from near zero in March 2022 to a range of 5.25–5.5% by July 2023. Mostly fixed-rate securities lost market value. By mid-2022, nearly all securities in the study’s sample were trading below amortized cost, and about one-third of the value-weighted holdings had losses greater than 10%.
At the same time, the most important securities did not behave like simple fixed-coupon bonds. Agency mortgage-backed securities are backed largely by fixed-rate mortgages whose borrowers can prepay when refinancing is attractive. When rates rise, refinancing becomes less attractive, prepayment slows, and the mortgages remain outstanding longer. The MBS therefore becomes longer-lived precisely when higher rates are already reducing its price. Its duration rises rather than falling, and its convexity can be negative. Callable bonds represented about 60% of the securities holdings in the study’s sample, while agency MBS made up more than half of the portfolio.
Funding conditions added a second pressure. Deposit growth weakened and, for many banks, turned negative by mid-2023. A bank facing outflows may need to raise wholesale funding, pay more for deposits, or sell assets. Securities are more liquid than loans, so they are an obvious place to look. But a portfolio containing underwater, duration-sensitive MBS is not a neutral cash reserve: selling can crystallize a loss, while not selling can leave the bank exposed to additional rate moves and continued funding pressure.
The observed bank response was correspondingly uneven. Purchases fell from a peak of about 20% of portfolio value per quarter in 2021 to roughly 3–4% in most of 2022–23, and the duration of new purchases also declined. That is a meaningful response: banks were willing to stop adding to the problem and to choose shorter exposures. Yet outright sales did not increase to the same degree; in fact, sales were unusually low. Qualified hedging did not expand materially either. Instead, banks classified a larger share of securities as HTM, with higher-risk bonds more likely to receive that classification.
This pattern is the subject of the series. It does not say that banks took no action. It says that they favored some actions over others, and that those actions affected different layers of the problem. Buying less changes the future path of the portfolio. Selling changes the current asset mix but realizes the loss. Hedging can reduce net exposure if the bank can qualify and maintain the hedge. HTM classification changes how rate movements appear in selected accounting and capital measures, while leaving the market value of the bond and many of the constraints on future use intact.
What the evidence can establish
The main source is a revised July 2026 Philadelphia Fed working paper that uses position-level FR Y-14Q data for roughly 30–35 U.S. banking organizations with at least $100 billion in assets. Those filers represented about 77% of commercial-bank assets and 78% of securities in 2021:Q4. The study links individual positions to duration, convexity, callable status, accounting classification, and qualified hedges. It also develops a method for inferring sales from quarter-end portfolio snapshots and checks that method against aggregate realized gains and losses. That makes the evidence unusually close to the actual security-level choices, while still leaving limits: a position that disappears must be distinguished from a maturity, call, paydown, exchange, or resecuritization, and the sample is not every U.S. bank.
The OCC’s Spring 2026 Interest Rate Risk Statistics Report provides a useful but non-identical comparison. It covers 810 OCC-supervised midsize and community banks and federal savings associations, not the largest banks included in the Philadelphia Fed’s main sample. Its tables report modeled changes in net interest income and economic value of equity under parallel rate shocks, policy limits, and assumptions about non-maturity deposits. The data are drawn from observations dated March 31, 2024, through December 31, 2025, and 74% of observations fall in the final year of that range. The OCC explicitly describes the statistics as informational, not as suggested limits or forecasts of realized losses.
That distinction between sources matters. The Philadelphia Fed study is best suited to asking what large banks did with particular securities during the 2022–23 surge and which characteristics predicted their choices. The OCC report is better suited to showing that modeled interest-rate sensitivity remains heterogeneous across a broader supervisory population. Neither source alone proves that one accounting rule or one risk-management policy would have prevented a bank failure. Neither gives a complete view of every nonqualified derivative, every liability’s market value, or every manager’s expectation about future rates.
What this series can establish is more precise and more useful. Banks’ securities portfolios were not passive, but their adjustment capacity was segmented. The market shock interacted with callable-asset mechanics, deposit behavior, loss recognition, hedge-accounting requirements, and classification rules. Once those interactions are visible, “Why didn’t they just sell?” becomes a better question: which cost made the sale unattractive, which alternative was feasible, and did the chosen alternative reduce economic risk or only move it outside the most visible measure?
The first dependency is the market mechanism itself. Before accounting categories or managerial incentives can be assessed, we need to see why rising rates made some of the bonds banks already held both less valuable and longer in duration. That is the starting point for understanding why a liquid portfolio could become such a difficult instrument of adjustment.