Silicon Valley Bank's insolvency was disclosed in its own annual report, filed weeks before the run. The headline numbers looked fine. The footnotes did not.
The bank that told you, in writing
At year-end 2022, SVB reported roughly $212 billion in total assets and about $16 billion in shareholder equity. Analysts anchored on that equity figure, on a capital ratio comfortably above regulatory minimums, and on management's guidance about deposit trends. None of it was false. All of it was beside the point.
Two other disclosures sat in the same filing. The first was the securities note. SVB carried about $91 billion of held-to-maturity (HTM) securities, bonds the bank has declared it will keep until they mature. Accounting rules let it record them at amortized cost, which is roughly what it paid, rather than at market value. The note also disclosed their fair value, the price they would fetch if sold that day. That figure was about $15 billion lower. One footnote had nearly erased the $16 billion of equity on the front page.
The second disclosure was the deposit base. The filing reported that the large majority of SVB's deposits were uninsured, meaning above the $250,000 federal deposit insurance limit per depositor. These accounts belonged mostly to venture-backed startups and their investors, a small, tightly connected group that shared the same group chats.
Put the two notes together and you get a bank whose real equity was close to zero, funded by depositors with every reason to leave and the means to leave at once. On March 8, 2023, SVB announced it had sold about $21 billion of available-for-sale securities at a loss of roughly $1.8 billion and needed to raise capital. On March 9, depositors tried to withdraw about $42 billion in a single day. Regulators closed the bank on March 10.
The market treated this as a sudden shock. The filings show it was a slow arithmetic fact that became a fast liquidity event.
Why a bank balance sheet hides its own risk
A bank is a leveraged bet on two things: duration, the sensitivity of an asset's price to interest rates, and credit, the chance that borrowers fail to repay. The equity is the thin difference between two enormous numbers, assets on one side and liabilities on the other.
That structure is the root of the reading problem. A small percentage error in a large number becomes a large percentage error in a small one.
Here is the arithmetic with clean numbers. Take a bank with $100 of assets, funded by $92 of deposits and $8 of equity. Suppose $40 of those assets are bonds with a duration of 6 years, so their price falls about 6% for every 1 percentage point rise in rates.
Price change ≈ -Duration × Change in yield
If rates rise 2 points, the bonds lose about 6 × 2 = 12% of their value. On $40 of bonds, that is a $4.80 loss. The bank's $8 of equity is now about $3.20 in economic terms, a 60% drop from a move in rates that would cost a bond fund holder just 12%.
That leverage ratio, roughly 12.5 times in this example, is ordinary for a bank. It is why banks earn decent returns on thin margins, and it is why small asset marks matter so much.
What this means for you: never read a bank's equity as a stand-alone number. Always ask how much of it survives a realistic mark on the asset side.
The funding side comes first
Most readers start a bank analysis with the asset side: loan growth, credit quality, net interest margin. The better order is the reverse, because the funding side determines whether the asset-side losses ever become real.
An HTM loss is a paper loss as long as the bank never has to sell. Bonds held to maturity pay back face value. The loss only crystallizes when depositors withdraw faster than the bank can meet withdrawals from cash, maturing assets, and borrowing capacity. So the question that decides everything is: who can leave, and how fast?
Three disclosures answer it.
Uninsured share of deposits. Insured depositors have little reason to run because the government guarantees their money. Uninsured depositors have every reason to run at the first sign of trouble, since being early is free and being late is costly. Banks report estimated uninsured deposits in their annual filings and in regulatory call reports.
Concentration. A million households with $20,000 each behave differently from 30,000 companies with $5 million each. Granular depositors are slow, distracted, and mostly insured. Concentrated depositors are informed and coordinated. SVB's depositors were concentrated by industry and by social network, which is the worst combination.
Available liquidity. Look at cash, unencumbered securities that can be pledged as collateral, and committed borrowing lines such as Federal Home Loan Bank advances. Compare that total to uninsured deposits. If uninsured deposits are several times larger than accessible liquidity, the bank depends on depositor patience.
What this means for you: compute uninsured deposits divided by readily available liquidity before you look at a single loan figure. That ratio tells you whether the asset-side marks are hypothetical or live.
Reading the held-to-maturity marks
Banks classify securities in two main buckets. Available-for-sale (AFS) securities are marked to market, and unrealized losses flow through accumulated other comprehensive income (AOCI), a component of equity. HTM securities are not marked at all on the balance sheet; their fair value appears only in the footnote.
This creates a quiet gap. The same bond, held in the HTM bucket, can carry a large market loss that never touches reported equity. During 2022, as rates rose rapidly, many banks reclassified securities from AFS into HTM, which froze the reported value and stopped further marks from hitting equity.
There is a second gap in regulatory capital. Under U.S. rules, banks below the largest size tier were permitted to exclude AOCI from their regulatory capital calculation. So even AFS losses, visible in reported equity, could be absent from the capital ratio a bank advertises. A healthy-looking capital ratio and a nearly wiped-out economic equity position can coexist in the same filing.
The adjustment is simple. Take reported equity, subtract the after-tax HTM unrealized loss from the securities footnote, and compare the result to tangible assets. Then do the same for the AFS losses if the regulatory capital figure excludes them. For SVB, this adjustment alone moved equity from about $16 billion to roughly $1 billion before tax effects.
A reclassification from AFS to HTM during a rate rise deserves particular attention. Read it as management choosing not to show you a number.
What this means for you: every bank analysis you do should include an "adjusted tangible equity" line that marks HTM securities to fair value. If that line is near zero, the reported capital ratio is not describing the bank you are looking at.
The economics underneath: why banks build this risk on purpose
None of this is accidental. Banks exist to perform maturity transformation: borrowing short through deposits that can be withdrawn on demand and lending or investing long through mortgages, business loans, and bonds. The spread between the low rate paid on deposits and the higher rate earned on long assets is the bank's core income.
The arrangement works because deposits are stickier than their legal terms suggest. A checking account can leave today, but most do not, because moving operating accounts is a hassle and households rarely shop for rates. That stickiness is a real economic asset, sometimes called the deposit franchise. When rates rise, a bank with sticky deposits keeps paying little on them while earning more on new assets, so the franchise becomes more valuable precisely as the bond portfolio loses value. The two partly offset.
The offset is the whole reason HTM losses do not automatically mean insolvency. For a bank whose deposits are granular, insured, and slow-moving, a bond mark is closer to a timing issue than a hole.
The mechanism breaks when the deposit franchise turns out to be worth less than assumed. That happens in three situations. Depositors may be concentrated and uninsured, as at SVB, so the stickiness evaporates under stress. Depositors may start chasing yield in money market funds when rates rise, raising the bank's deposit beta, the share of a rate increase it must pass to depositors. And communication may become fast enough that a run that once took weeks now takes hours, which shrinks the time a bank has to borrow or raise capital.
SVB failed all three tests at once. Its deposits were concentrated, its startup clients were burning cash as venture funding dried up, and its run was coordinated digitally in a single day. The bond loss was the fuel; the funding structure was the spark.
What this means for you: an HTM loss is only as dangerous as the deposits standing behind it. Judge the two together, never apart.
Where this reading goes wrong
The opposite error is just as real. Marking every bank's HTM book to market and declaring the sector insolvent ignores the deposit franchise entirely. A large retail bank with tens of millions of small, insured accounts can carry substantial unrealized losses for years and simply let the bonds mature. Its deposits are not going anywhere in a hurry, and the losses unwind as the securities approach face value.
A second pitfall is treating the funding analysis as static. Uninsured share and concentration are reported at quarter-end. A bank can look granular in December and lose a block of large commercial depositors by February. Rising use of brokered deposits or wholesale borrowing between reports is often the first sign that core deposits are leaving.
A third is ignoring credit. Duration was the 2023 story, but a bank is also a credit bet. Commercial real estate loans, particularly office loans at smaller banks, can produce losses that hit equity directly through loan loss provisions, the reserves a bank sets aside for expected defaults. A bank with a manageable bond mark and a concentrated office loan book has simply moved the problem to a different footnote.
The historical pattern is that bank failures often combine a hidden asset loss with a fragile funding base; either alone is frequently survivable. That pattern is worth watching rather than treated as a law, since some banks fail on credit alone.
What this means for you: the adjusted-equity calculation is a starting point. Pair it with a judgment about deposit behavior and a separate look at loan concentrations before drawing any conclusion.
Applying the reading to bank stocks now
Bank equity behaves like the leveraged instrument it is. Because the equity sliver is thin, modest changes in perceived asset values or funding stability produce outsized moves in the share price, and those moves cluster in stressed, volatile periods rather than spreading evenly over time.
That clustering is the market's own version of the balance sheet arithmetic. Regional banks hold more of their funding in commercial and uninsured accounts than the largest money-center banks, so they tend to sit closer to the SVB end of the spectrum. Calm periods reward the carry. Stress periods reprice the funding risk all at once.
The practical routine is short. Read the deposit footnote first and note the uninsured share. Compare it to cash and borrowing capacity. Then open the securities note, mark HTM to fair value, and recompute tangible equity. Check whether AOCI is excluded from the advertised capital ratio. Finally, scan loan concentrations for any single category large relative to equity.
A bank balance sheet tells you the truth only if you read it in the order that risk actually travels: from depositors who can leave, to assets that must be sold, to equity that absorbs the loss. SVB's filings described that path in full, and the headline equity number described almost none of it.
