The balance on a banking app looks like cash in a vault. Most of it is out in the world as mortgages and business loans. That is fractional reserve banking, and it means no bank can pay all its depositors on the same day. This episode explains why a bank run follows from that fact. The Diamond-Dybvig model shows that joining a run can be perfectly rational: if you expect others to withdraw, your best move is to get in line first, even when you know the rumor behind the panic is false. Sociologist Robert K. Merton used bank runs as his prime example of a self-fulfilling prophecy.

The history runs from 16th century English goldsmiths who issued more notes than they had gold, to the panics that began in Tennessee and Kentucky in November 1930 and spread through the Great Depression. Along the way come the defenses: imposing stone architecture, deposit insurance, weekend takeovers by the FDIC, and central banks as lenders of last resort. Then comes the modern problem. Silicon Valley Bank lost $42 billion in deposits in a single day in March 2023, a speed that rules built around a 30-day panic were never designed to handle.

• In 1832 British reformers set off a deliberate run under the slogan Stop the Duke, Go for Gold, pressuring the Duke of Wellington aside and clearing the way for the Reform Act.
• Canada avoided the banking panics that wrecked the United States in the 1930s because its banks had nationwide branch networks, while American unit banks depended on a single local economy.
• Old banks built cavernous lobbies so that lines stayed out of public view, staged visible cash deliveries, and paid employees’ relatives to stall at the teller windows.
• In June 2025 a hacktivist group called Predatory Sparrow disabled Iran’s Bank Sepah without stealing money. The outage alone triggered a run that spread to other banks.
• Proposed defenses for digital panics include dormant emergency payment nodes and algorithmic withdrawal fees that rise as more people try to pull money at once.