Silicon Valley Bank Collapse: Anatomy of a 2023 Bank Run

A small bank that grew very fast

Silicon Valley Bank, or SVB, served technology companies and the start-ups that venture capital firms backed. In 2017 it held about $45 billion of deposits and about $50 billion of assets.² By March 31, 2022 its deposits exceeded $200 billion and its assets stood at $217 billion.²

The money arrived because its customers were raising money. SVB Financial Group told investors that its clients were getting cash from public offerings, SPAC fundraising and venture capital rounds, which were at notably high levels in 2021 and early 2022.¹ Deposits rose from $103.2 billion in 2020 to $191.4 billion in 2021.²

The Federal Reserve's own review (https://www.federalreserve.gov/publications/files/svb-review-20230428.pdf) counted a 271 percent rise in assets from the end of 2018 to the end of 2021¹, against 29 percent for the banking industry as a whole.¹ The Government Accountability Office (https://www.gao.gov/products/gao-23-106736) (GAO) measured a shorter window, 2019 to 2021, and found 198 percent for SVB, 134 percent for Signature Bank and a median of 33 percent for 19 peer banks.⁴ The two figures differ because they start in different years, not because the sources disagree.

Where the money went

A bank has to put deposits somewhere. SVB made fewer loans than the average large bank, 35 percent of assets against 58 percent, and bought securities instead. Its securities holdings rose from $23 billion in 2018 to $125 billion in 2021, an increase of 443 percent, and its held-to-maturity (HTM) holdings grew from $15 billion to $98 billion.² Most of the HTM portfolio was agency mortgage-backed securities with maturities of 10 years or more, and the portfolio had a weighted-average duration of 6.2 years at the end of 2022.¹

Two accounting terms matter here. A security classed as held to maturity is carried at its purchase cost, not at today's market price, on the basis that the bank means to keep it until it is repaid. The Fed review adds that selling a portion of an HTM portfolio would generally force the whole portfolio to be reclassified and marked to market.¹ That made it costly for SVB to sell some of these bonds to raise cash.

The price of a long, fixed-rate bond falls when interest rates rise, because new bonds pay more. As interest rates rose from 0.25 percent in March 2022 to 4.5 percent in December 2022, SVB's unrealized losses on HTM securities grew from about $1.3 billion at the end of 2021 to about $15.2 billion a year later, and losses on securities held for sale grew from $313 million to $2.5 billion.² The word unrealized means the bank had not sold the bonds, so the loss was on paper. Our comparison: $15.2 billion is close to the $15.5 billion of equity capital SVB reported in its 2022 year-end filing.²

Deposits that could leave in a day

The other half of the problem was who held the deposits. FDIC insurance covers deposits up to $250,000 per depositor, per bank, per ownership category, and the part above the cap is uninsured.² At the end of 2022 about 94 percent of SVB's total deposits were uninsured¹, against 41 percent at the average large bank, and venture-backed companies held more than half of the deposits.¹

The bank also looked weak on the inside. The Fed review says SVB failed its own internal liquidity stress tests and removed interest-rate hedges rather than managing the risk of rising rates.¹ The Fed's inspector general (https://oig.federalreserve.gov/reports/board-material-loss-review-silicon-valley-bank-sep2023.pdf) called the complete removal of the hedges a significant error by management.² A director told supervisors in 2022 that controls always lag growth.¹

Supervisors were slow too. The San Francisco Fed downgraded SVB in June 2022 and began an enforcement action in August 2022, but did not finish it before the bank failed.⁴ SVB had crossed $100 billion of average assets in June 2021 and become what the Fed calls a Category IV firm, but some of that category's requirements, including stress testing, had not yet started because of transition periods; its first stress test would have come in 2024.¹

Three days that changed the rules

On March 8, 2023 the bank announced a restructuring. By the Fed's account it had sold $21 billion of securities held for sale, booked a $1.8 billion after-tax loss, planned to raise $2.25 billion of capital and to increase its term borrowing by $15 billion to $30 billion.¹ The inspector general's report gives the planned capital raise as $2 billion, so we show the Fed's figure and note the difference.² Silvergate Capital said the same day that it would wind down its bank, which, the Fed review says, further affected depositor sentiment.¹

Depositors read the announcement as a sign of distress. On March 9 customers asked to withdraw about $42 billion, nearly 25 percent of the bank's $166 billion of deposits² and almost 300 percent of its capital.² Management expected more than $100 billion of further outflows on March 10, and the bank did not have enough cash or collateral to pay.¹

The California regulator closed the bank on the morning of March 10 and named the FDIC receiver.¹ The FDIC moved insured deposits to a newly created institution, the Deposit Insurance National Bank of Santa Clara, and told customers that insured deposits would be available by Monday, March 13.⁵ At the end of 2022 the bank had about $209.0 billion of assets and $175.4 billion of deposits.⁵

Then, on Sunday March 12, the authorities went further. According to the Fed's announcement (https://www.federalreserve.gov/newsevents/pressreleases/monetary20230312a.htm), after consulting the President and acting on recommendations from the FDIC and Fed boards, the Treasury Secretary approved actions that, in the Fed's words, fully protect all depositors, both insured and uninsured.⁶ GAO describes this as the systemic risk exception, and reports that the FDIC and the Fed judged that not guaranteeing uninsured deposits would likely have caused more bank runs (https://therealigned.com/story/bank-runs-explained) and harmed the wider economy.⁴

The Fed also opened the Bank Term Funding Program (BTFP). It offered loans of up to one year to banks and other eligible institutions, took Treasuries, agency debt and mortgage-backed securities as collateral, and valued that collateral at par, meaning face value.⁶ The Treasury made up to $25 billion available from its Exchange Stabilization Fund as a backstop.⁶ In plain terms, a bank sitting on bonds that had lost market value could borrow against them at face value without selling.

1. June 2021SVB crosses $100 billion of average assets and becomes a Category IV firmSource 1 (opens the source in a new tab)
2. Mar to Dec 2022Interest rates rise from 0.25 percent to 4.5 percentSource 2 (opens the source in a new tab)
3. June 2022San Francisco Fed downgrades SVB; an enforcement action begins in AugustSource 4 (opens the source in a new tab)
4. Mar 8, 2023SVB announces the securities sale and capital raise; Silvergate announces its wind-downSource 1 (opens the source in a new tab)
5. Mar 9, 2023About $42 billion of withdrawals requestedSource 2 (opens the source in a new tab)
6. Mar 10, 2023California regulator closes SVB; FDIC appointed receiverSource 5 (opens the source in a new tab)
7. Mar 12, 2023All depositors protected; Bank Term Funding Program announcedSource 6 (opens the source in a new tab)
8. Mar 13, 2023Insured depositors have access to fundsSource 5 (opens the source in a new tab)
9. Sep 25, 2023Fed inspector general publishes its material loss reviewSource 2 (opens the source in a new tab)
10. Mar 11, 2024BTFP stops making new loansSource 7 (opens the source in a new tab)
11. Mar 30, 2026Final scheduled payment date of the FDIC special assessmentSource 3 (opens the source in a new tab)

The strongest case against the rescue

Moral hazard is the incentive to take on greater risk as a result of being protected from the consequences of risk-taking.

— FDIC, Options for Deposit Insurance Reform, May 2023

The same FDIC report is careful to say that deposit insurance also has benefits, because it protects small savers and reduces run risk, and it sets out three options: a finite limit (kept or raised), unlimited coverage, or higher coverage targeted at business payment accounts.⁹ It did not conclude that the 2023 guarantee was wrong.

The Bank Policy Institute authors made a narrower criticism of the loan programme. They estimate that because loans were made against collateral at par, the aggregate shortfall between what was lent and what the collateral was worth reached more than $20 billion⁸, and they point out that the loans cost one-year OIS plus 10 basis points, a low rate that could be repaid without penalty.⁸ They also note that borrowing surged again in late November 2023 when the loan rate fell below the rate the Fed pays on bank reserves, creating a profit from borrowing, until the Fed changed the rate in January 2024.⁸

appears to have operated largely as intended, but at considerable risk and cost

— Bank Policy Institute authors on the Bank Term Funding Program, April 2025

There is a distributional point as well. Accounts above $250,000 made up less than 1 percent of deposit accounts by number at the end of 2022, and the top 1 percent of banks by assets held about 77 percent of uninsured deposits.⁹ The uninsured balances that the guarantee covered therefore sat in a small share of accounts and were concentrated in the largest banks. Those facts do not settle whether the guarantee was right. They describe who held the money it protected.

What the Silicon Valley Bank collapse cost

Within weeks the FDIC estimated that the failure would cost its Deposit Insurance Fund about $20 billion; in May 2023 it lowered the estimate to about $16.1 billion.² To recover the loss from protecting uninsured depositors, the FDIC set up a special assessment on large banks.³

The Fed's loan facility had a different cost profile. The BPI authors count nearly 9,000 loans to about 1,327 borrowers at a peak above $165 billion.⁸ The Fed stopped making new loans on March 11, 2024. Its report of November 12, 2024 shows $57.56 billion outstanding on October 31, 2024, with collateral of $71.94 billion pledged and $8.86 billion of interest and fees received.⁷ The BPI authors date the last repayment to March 11, 2025.⁸ We could not confirm that the Fed and BPI value the collateral on the same basis or on the same dates, so we do not net their figures against each other.

What changed in the rules and the numbers

The Fed's own review, released on April 28, 2023, listed four findings: the board and management failed to manage risks, supervisors did not fully appreciate the vulnerabilities, supervisors did not make the bank fix problems quickly enough, and the Fed's tailoring of rules after a 2018 law, together with a shift in supervisory stance, reduced standards and promoted a less assertive approach.¹ Michael Barr, who oversaw the review, described the review as a first step and a self-assessment that takes an unflinching look at the conditions that led to the failure.¹⁰

The review also tested one counterfactual. SVB had opted out of counting unrealized gains and losses on securities held for sale in its regulatory capital. Had it counted them, the review calculates, capital would have been $1.9 billion lower and its common equity ratio 10.4 percent instead of 12.1 percent.¹ The review adds that the lower capital may have led the bank to raise more capital or make different business decisions.¹

The FDIC's May 2023 report set the failure in a wider trend. The uninsured share of all deposits peaked in 2021 at its highest level since 1949, and the 2022 total value of uninsured deposits was more than five times its real value in 1990.⁹ About 28 percent of FDIC-insured institutions had more than a third of their domestic deposits uninsured at the end of 2022.⁹