Bank Run

A bank run is a large number of a bank's depositors demanding repayment at or near the same time. It is a liquidity event rather than a verdict on the bank's balance sheet: a bank can be solvent, in the sense that its assets exceed its liabilities if held to maturity, and still be unable to produce enough cash on a given Thursday to satisfy everyone who asks.

The reason is structural and is the whole subject. A bank takes money that is repayable on demand and puts it into loans and securities that are not. That transformation is the service a bank performs, and it is the source of the fragility. Ordinary withdrawals are covered from cash and from incoming deposits, because on a normal day only a small fraction of depositors want their money. A run is what happens when that assumption breaks.

It is worth separating a run from a bank failure, because the two get used interchangeably and they are different events. A run is depositors leaving. A failure is the point at which a regulator closes the institution and appoints a receiver. A run can cause a failure, a failure can happen without a run, and the deposit insurance page covers what a depositor is owed once the second one occurs.

The first-mover incentive is what turns a rumor into a closure, and it is worth stating precisely. A bank pays withdrawals in the order they arrive, at face value, out of cash first. Once the cash is gone, the next withdrawal has to be funded by selling something, and a forced seller takes the price on offer. If the assets are worth less than their carrying value, each forced sale realizes a loss that reduces the pool available to everyone still in the queue. So a depositor who has no view whatsoever about the bank's solvency still has a reason to withdraw, namely that other people might. Belief about what others will do is sufficient. That is the sense in which a run is self-fulfilling, and it is why the federal response is built around changing the incentive rather than around reassuring people.

Interest rate risk is the modern shape of the problem, and it explains why the balance sheet can look fine. A bank that bought long-dated bonds when rates were low holds bonds worth less than face value once rates rise. Held to maturity they still pay par, so accounting rules do not necessarily require the loss to be recognized while the bank intends to hold them. Withdrawals destroy that intention. The paper loss becomes a realized one at the moment the bank has to sell, and it is realized precisely when the bank can least afford it. The Federal Reserve's review of the 2023 Silicon Valley Bank failure describes that sequence in the bank's own case, finding that the firm "invested those deposits in longer-term securities and did not effectively manage the interest-rate risk, including actively removing hedges as rates were rising."

Two federal backstops stand between the mechanism and the outcome, and each has a limit worth knowing.

The first is deposit insurance. Its function here is not to compensate after the fact but to remove the reason to queue in the first place: a depositor who will be made whole up to a statutory limit gains nothing by being early. The insurance page carries the limit and, more importantly, the three-part qualifier that decides how far it actually stretches. The consequence for run risk is direct, and the 2023 failures illustrated it: runs concentrate in uninsured deposits, because those are the balances where arriving early still pays. The Fed's review records that it was uninsured depositors who read the bank's March 8 announcements as distress and began withdrawing.

The second is the discount window, the Federal Reserve's lending to depository institutions. The Board describes it as playing "an important role in supporting the liquidity and stability of the banking system", and says that providing liquidity this way "is one of the original purposes of the Federal Reserve System and other central banks around the world." Its statutory framework is section 10B of the Federal Reserve Act. The relevant point for a run is that the window lets a bank borrow against assets rather than sell them, which converts a fire sale into a loan. Two conditions constrain it. Primary credit, which the Board calls "the principal safety valve for ensuring adequate liquidity in the banking system", is available only to institutions "in generally sound financial condition". And "all discount window loans must be collateralized to the satisfaction of the lending Reserve Bank." Collateral is the binding constraint in a fast run, and the Fed's review says so of Silicon Valley Bank in one sentence: it "did not have enough cash or collateral to meet the extraordinary and rapid outflows."

What changed in 2023 was speed, and the central bank said so itself. In the transmittal letter to the review, the Vice Chair for Supervision wrote that "the combination of social media, a highly networked and concentrated depositor base, and technology may have fundamentally changed the speed of bank runs. Social media enabled depositors to instantly spread concerns about a bank run, and technology enabled immediate withdrawals of funding." The underlying figures are the argument for that sentence. On March 9, 2023, Silicon Valley Bank experienced a total deposit outflow of over $40 billion. Overnight and into the following morning, it told supervisors it expected over $100 billion more during the day on March 10. The California Department of Financial Protection and Innovation closed the bank on the morning of March 10 and appointed the FDIC as receiver. The Fed characterizes the episode as a run "fueled by social media and SVB's concentrated network of venture capital investors and technology firms that withdrew their deposits in a coordinated manner with unprecedented speed."

The practical reading for a depositor is narrower than the drama suggests. Concentration is the risk factor a run needs. A bank funded by a few thousand similar customers who all talk to each other is exposed in a way that a bank funded by millions of unrelated insured households is not, and the second description fits most of the banking system. For a household, the actionable question is not whether a bank might be run on but whether its own balances sit inside the insurance limit, and that is an arithmetic question rather than a forecasting one.

Something makes depositors doubt the bank, or doubt what other depositors will do. Withdrawals begin. Cash goes out first. Once cash is exhausted the bank must either borrow against its assets or sell them, and selling at a discount turns an unrealized loss into a realized one. If the realized losses exhaust the bank's capital, the regulator closes it and a receiver takes over paying insured depositors and liquidating what is left.

A hypothetical example, with round numbers chosen so the arithmetic can be checked by hand. All figures are in millions.

A bank holds $100 of cash and bonds with a face value of $900, bought at par, so its books show $1,000 of assets. It owes depositors $920. Book equity is $1,000 minus $920, which is $80.

Rates have risen, so those bonds would fetch only 80 cents on the dollar today: $720 rather than $900. Nothing in the accounts says so, because the bank intends to hold them to maturity and be repaid at par. Marked to market, though, the bank's assets are $100 plus $720, which is $820, against $920 of deposits. On that measure it is already $100 short. Left entirely alone, it collects par at maturity and the shortfall never materializes.

Now $300 of deposits leave. The first $100 comes from cash. The remaining $200 has to come from bond sales at 80 cents, so the bank sells bonds with a face value of $250 to raise $250 × 0.80 = $200, and realizes a loss of $250 minus $200, which is $50. Book equity falls from $80 to $80 minus $50, which is $30. On paper the bank is still solvent, and it now holds $650 of face-value bonds against $620 of deposits.

Another $300 leaves. There is no cash left, so the whole amount comes from sales: face value $375 raises $375 × 0.80 = $300, realizing a loss of $75. Book equity is now $30 minus $75, which is negative $45. The bank holds $275 of bonds against $320 of deposits, and the insolvency is no longer hidden. A regulator closes it.

Three things are worth reading off that sequence. The bank that failed was the same bank in every respect except the withdrawals, which is the point: the run did not reveal the loss, it caused the loss to be recognized. Second, the depositors who left first got 100 cents on the dollar and the deficit is carried by those who stayed, which is exactly the incentive the queue creates. Third, borrowing against those bonds instead of selling them would have avoided both loss-realizing sales, which is what the discount window exists to do and why the availability of acceptable collateral decides how a fast run ends.

Pros (of the arrangements that make runs rare)

• Deposit insurance addresses the cause rather than the symptom. A depositor inside the coverage limit gains nothing by withdrawing early, so the queue never forms.
• Central bank lending against collateral lets a solvent bank meet withdrawals without selling assets into a falling market, which stops a liquidity problem from manufacturing a solvency problem.
• The Federal Reserve's own post-mortem on the 2023 failure was published in full, including its criticism of its own supervision, so the mechanism is documented at source rather than inferred.
• The structural risk factor is measurable in advance. Concentrated, uninsured, fast-moving funding is visible in a bank's own disclosures.

Cons

• Insurance protects the depositor, not the bank. An institution funded mainly by uninsured balances is still runnable, and the 2023 failures were of exactly that kind.
• The discount window requires collateral acceptable to the lending Reserve Bank, and primary credit is limited to institutions in generally sound condition, so the backstop is weakest in the situations that need it most.
• Withdrawals can now be initiated instantly and coordinated publicly, which compresses the time available for any official response.
• A run is self-fulfilling, so accurate information about a bank's solvency does not by itself stop one.
• Uninsured balances, whether a business payroll account or a household holding above the limit at one institution, are the exposure a depositor can actually do something about, and checking is arithmetic rather than forecasting.