What causes a bank run and why are they self-fulfilling?

The short answer

Banks operate on maturity transformation: deposits payable on demand fund loans payable over years. At any moment, only a fraction of total deposits is held as cash or liquid reserves. If enough depositors try to withdraw simultaneously, the bank cannot meet the claims, regardless of whether its loan book is healthy.

This is why the question of solvency is often secondary in a run. The trigger can be a rumor, a downgrade, or a peer institution failing — what matters is whether each depositor expects others to withdraw. If the answer is yes, individually rational behavior becomes withdrawing first.

Diamond and Dybvig formalized this in 1983: a banking system has two equilibria, one stable and one runs-driven. Both are self-consistent. The choice between them is essentially a coordination problem.

→ New to financial fragility? Systemic fragilities pillar

What the data shows

The empirical pattern across centuries is consistent — bank runs cluster around shocks, but their speed depends on the technology of coordination. Selected episodes (FDIC and BIS records, 1907-2023):

• Pre-FDIC era (US, 1873-1933): more than 9,000 bank failures during 1929-1933 alone, many triggered by runs in geographically concentrated waves
• Northern Rock (UK, September 2007): first British bank run since 1866, depositor queues over 5 days, ~£1bn withdrawn before government guarantee
• IndyMac (US, July 2008): $1.3bn withdrawn over 11 business days before FDIC takeover
• Silicon Valley Bank (US, March 2023): $42 billion withdrawn in a single day (March 9), with $100 billion in pending withdrawals queued for March 10 — 81% of total deposits in 36 hours

The exception that proves the rule: post-FDIC US banking saw few uninsured retail runs from 1934 to 2007. The fragility never disappeared — it was suppressed by guarantees that made the coordination problem irrelevant for insured depositors.

→ Dataset: Credit spreads and bank stress

Why it happens — the macro mechanism

The structure of bank funding makes runs possible by design. Three channels operate simultaneously.

The maturity transformation channel. A bank borrows short (deposits) and lends long (mortgages, corporate loans, securities). Reserves cover only a small fraction of total liabilities — typically 10-15% in liquid form. The bank is solvent on a balance-sheet basis but illiquid against simultaneous redemption. Funding vs market liquidity distinguishes these states.

The coordination channel. Each depositor’s optimal action depends on what other depositors do. If others stay, leaving costs nothing extra and risks transaction friction. If others withdraw, leaving means losing access. Contrary to the textbook view that runs require fundamental weakness, the coordination logic alone produces them — this is the angle most retail commentary misses, and it is precisely why a solvent bank can fail.

The information shortcut speeds this up: most depositors cannot read a balance sheet, so they substitute observable signals — peer behavior, headlines, social media — for fundamental analysis.

The institutional channel. Central banks, deposit insurance, and discount windows exist precisely to break the coordination loop. When credible, they shift the equilibrium toward staying. When their credibility is questioned, the loop reopens. FDIC insurance mechanics details the US framework.

Synthesis by regime: in the pre-1933 US banking system, runs occurred routinely and clustered geographically — bank failures averaged hundreds per year through the 1920s. Under the FDIC regime (1934-2007), insured retail runs effectively disappeared while uninsured wholesale runs persisted (Continental Illinois 1984, IndyMac 2008). In the post-2007 digital regime, runs returned in a new form: instantaneous, coordinated through social platforms, and concentrated on uninsured wholesale deposits — SVB’s 36-hour collapse marks the transition.

A bank run does not require a sick bank — it requires a coordinated belief, and modern coordination is essentially free.

→ Framework: Systemic fragilities

What it means for different economic actors

Savers. Insured depositors are protected by the FDIC up to $250,000 per ownership category. The coordination problem does not apply to them in practice — their incentive to run is muted by the guarantee.

Corporate treasurers. Cash balances above the insurance limit face the full coordination logic. Diversifying across banks, using ICS sweeps, or holding Treasury bills directly are the standard responses to concentration risk.

Investors. Bank equity and subordinated debt absorb losses before depositors are touched. Stress events repeatedly show that uninsured deposit concentration is a leading equity-price signal — share prices move first when a run becomes plausible.

A common error is treating run risk as a function of solvency only. The historical evidence — Northern Rock was solvent on paper, SVB was within capital ratios — shows that the coordination dimension can dominate. The specifics are documented in the assumptions that mislead investors on bank runs and banking crises.

Practical observation

Go deeper

Related questions

• How did the 2023 regional bank crisis differ from 2008?
• How does social media accelerate bank runs?
• Why did SVB fail so quickly in March 2023?
• How does FDIC insurance actually work?

Frequently asked questions

Is a bank run possible at a profitable, well-capitalized bank?

The historical record indicates yes. Northern Rock met regulatory capital requirements when its run began in September 2007. SVB exceeded minimum CET1 ratios on the day before its collapse. The coordination logic of runs operates on liquidity and depositor expectations — not on long-term solvency. A bank that is fundamentally solvent on a hold-to-maturity basis can still be insolvent on a forced-sale basis if all depositors withdraw at once.

How does deposit insurance change the run equilibrium?

Credible insurance removes the incentive for the insured tier of depositors to participate in a run, which often suffices to break the coordination problem at the retail level. The 2023 US episodes showed the limit of this mechanism: when the bulk of deposits are above the insurance ceiling, the insured tier becomes informationally irrelevant and the run is driven by uninsured concentration. This is why deposit insurance design — coverage levels, ownership categories, sectoral exposure — directly conditions banking-system stability.

Why are runs harder to model than other financial events?

Standard models price assets based on expected cash flows. A run is fundamentally different because the outcome depends on the equilibrium chosen by depositors, not on a single best estimate of bank value. Diamond-Dybvig multiple equilibria, Morris-Shin global games, and modern coordination-friction models all attempt to address this, but each requires assumptions about how depositors form beliefs about other depositors. There is no single risk-neutral price for run probability.

Last updated — 14 June 2026