What Is a Bank Run? A Belief That Makes Itself True

Banks take deposits that can be withdrawn on demand and lend them out in loans that cannot be recalled quickly, which is maturity transformation and is the core function they perform. That arrangement is useful, since it channels short-term savings into long-term investment, and it means no bank holds enough cash to satisfy all its depositors simultaneously. In normal conditions that is fine, since withdrawals and deposits roughly offset. If depositors come to believe others will withdraw, each has a reason to withdraw first regardless of what they think about the bank's actual soundness, since the last in the queue gets nothing. That gives the situation two possible outcomes, one where everyone stays and the bank is fine and one where everyone runs and it fails, and which occurs depends on expectations rather than on the loans.

The measures adopted address different parts of the problem:

• •Deposit insurance, which guarantees balances up to a limit and removes the incentive to run for covered depositors
• •A central bank lending against good collateral to solvent institutions facing a liquidity shortage
• •Reserve and liquidity requirements, forcing banks to hold assets that can be converted quickly
• •Suspension of withdrawals, which stops a run and destroys confidence, used historically and now rare
• •Supervision intended to prevent a bank becoming genuinely unsound in the first place
• •Resolution arrangements allowing a failing bank to be taken over without depositors losing access

Queues outside branches are now the least important form. Wholesale funding, where banks borrow short-term from other institutions, can evaporate within days if lenders lose confidence, and several failures during the 2008 crisis took that form with no retail depositor involved. Money market funds experienced the same dynamic. The 2023 failure of a substantial American bank was notable for speed, with tens of billions withdrawn within hours as customers coordinated through messaging and moved money electronically, which compressed a process that historically took days into an afternoon and prompted a reassessment of whether liquidity rules designed around slower runs remain adequate. Concentrated deposits from a single industry, most of them above the insured limit, made that institution unusually vulnerable, and the episode illustrated that the classic dynamic operates faster rather than having gone away.

The pattern has recurred often enough to shape institutions. Nineteenth-century banking panics in Britain and the United States produced repeated failures and prompted the development of central bank lending as a response, with the guidance for handling them written during that period. The failures of the early 1930s in the United States were catastrophic, with thousands of banks closing and the resulting contraction of credit deepening the depression substantially, and deposit insurance was introduced in response and largely ended retail runs there. A British bank experienced the first run on a high street institution there in over a century in 2007, with queues outside branches, which was resolved by a government guarantee and led to substantial reform of the deposit protection arrangements. Each episode produced institutional change, which is why the arrangements now in place look designed rather than evolved.

The distinction between the two is central to policy and is difficult to apply in practice. A liquidity problem means a bank has good assets and cannot convert them to cash quickly enough, and the appropriate response is lending against those assets, which resolves the situation at no ultimate cost. A solvency problem means the assets are worth less than the liabilities, and lending merely postpones the failure and transfers losses to whoever lent. The classic guidance is to lend freely against good collateral at a penalty rate to solvent institutions, which is nearly two centuries old and still quoted. The difficulty is that during a crisis nobody can tell which situation applies, since asset values are uncertain precisely when everyone is selling, and a liquidity problem becomes a solvency problem if forced sales push prices down far enough.

Deposits withdrawable on demand fund loans that cannot be recalled, so no bank can pay everyone at once and a belief that others will withdraw is reason enough to withdraw first. Insurance and central bank lending address the expectation rather than the loans. The 2023 failures showed the same dynamic compressed into hours by electronic transfer and coordination.