Bank Runs: Why Solvent Banks Can Fail Overnight
Bank Runs: Why Solvent Banks Can Fail Overnight
A bank with $92 million of deposits and $8 million of equity, holding securities worth less today than at maturity, repays every depositor if nobody runs. If withdrawals pass about $65 million, forced sales wipe out its equity. Whether it survives depends less on its loans than on what each depositor expects the others to do. In March 2023, Silicon Valley Bank lost about $42 billion of deposits in a day.
What you'll learn
Why a solvent bank can fail
A bank promises depositors their money on demand while lending it out for years. That mismatch is the service: households and businesses get liquidity, and borrowers get long-term funding. It also creates a fragility that Diamond and Dybvig formalized in 1983. If depositors expect others to withdraw, withdrawing first is the sensible response, because the bank can pay early withdrawers in full only by selling assets at a loss that falls on whoever is left. The expectation of a run can cause the run.
A run is an equilibrium, not a misjudgment. In their model a bank can sit in either of two states with identical assets: a good one in which depositors trust each other to stay, and a bad one in which each fears the others will leave. Nothing about the loans needs to change for the bank to move from one to the other.
A run, worked
Take a bank with $92 million of deposits and $8 million of equity. It holds $20 million in cash and $80 million of securities it intends to hold to maturity, when they will repay $80 million. Interest rates have risen, so those securities would fetch only $68 million if sold today.
| Withdrawals | Securities sold, at book value | Loss realized | Equity remaining |
|---|---|---|---|
| $20 million | $0 | $0 | $8.0 million |
| $30 million | $11.8 million | $1.8 million | $6.2 million |
| $40 million | $23.5 million | $3.5 million | $4.5 million |
| $60 million | $47.1 million | $7.1 million | $0.9 million |
| All $92 million | $80 million | $12 million | −$4 million |
The bank's solvency depends on its depositors' behavior. At book value it is solvent; marked to market it is $4 million short. Each uninsured depositor who expects others to run is correct to run first, since early withdrawers are paid in full from cash and the losses fall on those who stay. Per our correlated failure analysis, the run is a coordination problem: individually rational, collectively destructive.
Silicon Valley Bank, 2023
The worked example is close to the structure of Silicon Valley Bank. It had invested heavily in long-dated securities when rates were low, and as rates rose its unrealized losses grew roughly as large as its equity. Its deposits were concentrated among venture-backed companies, and the great majority were above the insurance limit. On March 8, 2023, it announced a loss on securities sales and a plan to raise capital. On March 9, depositors withdrew about $42 billion, and the bank faced requests to withdraw roughly $100 billion more the next day. It was closed on March 10.
- Speed was new. Withdrawal requests moved by mobile banking and coordinated through group chats among founders and investors, compressing a run that once took days or weeks into hours.
- Concentration mattered. A depositor base of similar firms advised by a small number of investors behaved as one.
- The losses were not hidden. They were disclosed in public filings. Jiang, Matvos, Piskorski and Seru estimated that many U.S. banks carried mark-to-market losses large enough that a run by uninsured depositors could make them insolvent, which is why regulators treated the episode as a systemic threat.
Per our deposit franchise analysis, a bank's deposits are valuable because they are usually stable and cheap. That value rests on the assumption that depositors stay, and SVB showed how quickly the assumption can fail.
What stops runs
| Tool | How it breaks the run equilibrium | Cost or limit |
|---|---|---|
| Deposit insurance | Insured depositors are paid regardless of others' actions, so they have no reason to run | Covers $250,000 per depositor per bank per ownership category; larger balances remain exposed |
| Lender of last resort | The central bank lends against good collateral, so the bank need not sell assets at a loss | Requires collateral and willingness to borrow; stigma can deter use |
| Liquidity requirements | More cash and liquid assets cover a larger share of withdrawals without forced sales | Lower returns; does not protect against a full run |
| Capital requirements | More equity absorbs larger realized losses before depositors are affected | Raises the cost of lending |
| Suspension of convertibility | Historically, banks halted withdrawals so that running gained nothing | Freezes the liquidity that deposits are supposed to provide |
Deposit insurance is the classic answer, and it carries the classic cost. Diamond and Dybvig showed it eliminates the bad equilibrium at no cost if it is credible, since it is never used. But insured depositors have no reason to monitor the bank, which weakens discipline. Calomiris and Kahn argued that the right to withdraw on demand was itself a disciplining device, letting informed depositors exit a badly run bank; insurance removes that discipline along with the panic.
For a business, the practical question is how much of its cash sits above the insured limit. Per our business cash management analysis, spreading balances across banks, using insured sweep programs or holding Treasury bills for reserves keeps operating cash from depending on a single bank's depositor base. Per our intermediation analysis, each layer between a depositor and an insured account is another place where access can fail even when the money is safe.
Runs outside banks
The logic applies wherever short-term claims fund illiquid or uncertain assets.
- Money market funds faced runs in 2008 and 2020 when investors feared that funds would break their stable share price.
- Intermediaries holding customer funds. Per our bank partnership analysis, consumer apps that hold balances through partner banks add a layer between the customer and the insured account. The 2024 collapse of Synapse, a middleware provider, left end users unable to reach their funds for months while records were reconciled, a reminder that insurance protects against a bank's failure, not an intermediary's.
- Businesses holding customer prepayments. Per our customer deposit analysis, a company using deposits to fund operations faces a run if customers lose confidence and demand refunds at once.
- Private credit funds with periodic redemption rights can face the same dynamic when many investors ask to leave together. Per our private credit analysis, redemption gates are a modern form of suspension.
The counter-case
Are runs panics or information?
The Diamond and Dybvig model treats runs as self-fulfilling and unrelated to fundamentals. An alternative view, associated with Gorton's study of U.S. banking panics, holds that runs typically follow real news about bank losses and occur when depositors cannot tell which banks are sound. On that view, runs are a crude but rational response to genuine risk, and the policy goal is better information and stronger balance sheets rather than guarantees. SVB fits both readings: its losses were real and public, but healthy banks with similar depositor bases also faced outflows that subsided only when authorities guaranteed all of SVB's deposits.
Both views point to the same practical lessons for anyone holding other people's money. Match the liquidity of assets to the liquidity promised to funders, hold enough equity to absorb forced sales, and know how concentrated the funders are. Per our financial accelerator analysis, fire sales by one institution lower asset values for every other holder, which is how a run at one bank becomes pressure on many.
Testable implications
- Banks with larger shares of uninsured deposits should suffer larger outflows after a common shock, holding asset losses constant.
- Outflows should be faster where depositors are concentrated and connected, such as industry-specific or digitally active depositor bases.
- Unrealized losses should predict outflows only in combination with uninsured funding, since insured depositors have no reason to run on losses alone.
- Guarantee announcements should halt outflows at banks with similar depositor bases even when their own fundamentals are unchanged, the signature of a coordination component.
Per our rate transmission analysis and information timing analysis, the third implication matters for any lender funded by deposits: rising rates create losses on fixed-rate assets that are harmless while funding is stable and dangerous the moment it is not.
Frequently asked questions
Banks fund illiquid assets with on-demand deposits, so expecting others to withdraw makes withdrawing first rational.
Forced sales turn unrealized losses into realized ones. In the example, withdrawals above about $65 million exhaust $8 million of equity.
Rate-driven securities losses plus concentrated uninsured deposits. About $42 billion left in one day, and it was closed on March 10, 2023.
For insured balances, yes, but larger balances remain exposed and insurance weakens depositor discipline.
Key takeaways
- Banks fund illiquid assets with on-demand deposits, so a run can be self-fulfilling.
- In the example, the bank repays everyone if no one runs and becomes insolvent once withdrawals pass about $65 million.
- Forced sales turn unrealized losses into realized ones that fall on remaining depositors.
- Silicon Valley Bank combined large unrealized losses with concentrated uninsured deposits and lost about $42 billion in a day.
- Deposit insurance, central bank lending, liquidity and capital each break the run equilibrium at a cost.
- Insurance weakens depositor discipline, which Calomiris and Kahn argued was a function of demandable deposits.
- Runs occur wherever short-term claims fund illiquid assets, including funds and intermediaries.
- Whether runs are panics or information, the lesson is to match asset liquidity to funding liquidity.
References
- Friedman, M., & Schwartz, A. J. (1963). A Monetary History of the United States, 1867–1960. Princeton University Press.
- Diamond, D. W., & Dybvig, P. H. (1983). Bank Runs, Deposit Insurance, and Liquidity. Journal of Political Economy, 91(3), 401–419.
- Gorton, G. (1988). Banking Panics and Business Cycles. Oxford Economic Papers, 40(4), 751–781.
- Calomiris, C. W., & Kahn, C. M. (1991). The Role of Demandable Debt in Structuring Optimal Banking Arrangements. American Economic Review, 81(3), 497–513.
- Board of Governors of the Federal Reserve System (2023). Review of the Federal Reserve's Supervision and Regulation of Silicon Valley Bank.
- Jiang, E. X., Matvos, G., Piskorski, T., & Seru, A. (2024). Monetary Tightening and U.S. Bank Fragility in 2023: Mark-to-Market Losses and Uninsured Depositor Runs? Journal of Financial Economics, 159, 103899.
This analysis is educational and does not constitute investment, banking or legal advice. The worked bank balance sheet is a stylized illustration and does not describe any particular institution. Figures on Silicon Valley Bank are approximate and drawn from public regulatory reviews. Deposit insurance coverage depends on ownership categories and FDIC rules.