Bank runs
A Socratic walk-through of bank runs — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why can everyone rushing to take their money out of a sound bank bring it down?
We assume a business fails because something inside it was rotten — bad loans, bad bets, bad books. But a bank whose loans are all performing, whose assets comfortably exceed what it owes, can be dead within a week for no reason other than that its depositors queued. Nothing about the loans changed. So what exactly is it that fails, if not the assets?
Reasoning it through
REASONING #Start with what a bank is actually promising, because the whole puzzle lives in the promise. On one side it takes deposits repayable on demand: your money, whenever you ask. On the other it makes loans that run for years and cannot be recalled at will. That mismatch is not a flaw someone forgot to fix — it is the service. Savers want their money available, borrowers want it committed, and the bank gives both by holding only a fraction in cash, because on any ordinary day only a fraction of depositors ask.
Notice what that rests on: that withdrawals stay uncorrelated. So what happens when they correlate?
Ask what you should do if you believe everyone else is about to withdraw. The bank pays in the order people arrive, out of a limited stock of ready cash. Those at the front are paid in full; those at the back are paid from whatever can be raised by selling long assets in a hurry, and possibly not at all. So if I think you are going, my best move is to go first, even if I am certain the loans are sound. And if I think you will sit tight, my best move is to leave my money where it is.
That is the uncomfortable result Diamond and Dybvig formalised in 1983: the same contract supports two equilibria. In one, nobody runs, and nobody should. In the other, everybody runs, and everybody should. Both are internally consistent. Nothing in the bank's balance sheet picks between them — what picks between them is what each depositor expects the others to do. A run can therefore be self-fulfilling: the belief that the bank will fail is the thing that makes it fail.
Why does it actually fail, though, rather than merely inconvenience everyone? Because raising cash fast destroys value. A loan book sold in an afternoon fetches less than the same loans held to maturity; a bond portfolio dumped into a thin market prints a loss that would never have been realised otherwise. The run manufactures the insolvency it was afraid of.
This gives us the distinction that does most of the work here. A bank is insolvent if its assets are worth less than its liabilities even given all the time in the world. It is merely illiquid if the assets are worth more but cannot be turned into cash today at that value. A run kills both kinds equally — and telling them apart in the moment is genuinely hard, which is why the post-mortem argument about whether a failed bank was "only illiquid" is so rarely settled. If assets can only be sold at fire-sale prices, is the fire-sale price the true price?
Now the intervention follows from the diagnosis. If the problem is beliefs, change the beliefs. Deposit insurance guarantees the deposit, so I have no reason to join the queue even when I am sure you will. That deletes the second equilibrium rather than paying for its consequences — which is why insurance funds are small relative to the deposits they cover, and why that is not the scandal it sounds like. The older tools do the same job less elegantly: suspending convertibility, and a lender of last resort on Bagehot's rule of lending freely against good collateral at a penalty rate, which is precisely a bet that the bank is illiquid rather than insolvent.
The analogy
THE ANALOGY #Think of a crowded theatre with one narrow exit. Walking out calmly is better for everyone than a stampede, and nobody wants a stampede. But if you believe the person behind you is about to bolt, bolting first is the sensible thing to do — and the belief alone is enough to produce the crush.
In a theatre the danger and the exit are fixed, whereas here the crush itself narrows the door — every hurried sale lowers the value of what is left for those behind. And an announcement that the exits are guaranteed would calm nobody in a real fire, while it is exactly what calms a bank run.
Clarifying the model
THE MODEL #Three refinements, because this model is easy to over-read.
First, a run is not evidence that the assets were bad — but neither is it evidence that they were good. Most historical runs began with genuinely bad news, and the empirical weight of pure panic against real deterioration is still argued over. The claim is not that fundamentals never matter; it is that they are not sufficient to determine the outcome once a second equilibrium exists.
Second, deposit insurance buys the stability with a real cost. If my deposit is guaranteed, I have no reason to check what the bank is doing with it, so the discipline depositors once supplied disappears — and the bank can take more risk knowing its funding will not flee. That is why insurance never travels alone: capital requirements, supervision and resolution rules exist to do the monitoring that insured depositors no longer bother with.
Third, the run has moved. Guaranteed retail depositors mostly do not queue any more; large uninsured balances and wholesale funders do, and they can leave in minutes rather than days, which shortens every response time the older tools assumed.
A picture of it
THE PICTURE #How to readStart at "Deposits stable" and follow the two arrows out of "Doubt spreads" — they are the heart of it, because both are rational and the only thing separating them is what each depositor expects the others to do. The right-hand path is the run: correlated withdrawals drain the cash, forced sales turn a timing problem into a real loss, and only then is the bank genuinely insolvent. The arrow to "Backstop credible" is where insurance or a lender of last resort acts, and note where it returns to — it removes the reason to queue rather than paying the queue off.
What became clearer
WHAT CLEARED #A bank does not fail here because its assets were bad; it fails because a promise of instant repayment on slow assets has two consistent outcomes, and expectations alone choose between them. The run then makes itself true by forcing sales that destroy value. Which is why the cure is aimed at belief rather than at cash — a guarantee that nobody needs to use is worth more than a vault that would eventually run dry.
Where to go next
ONWARD #- Why the same two-equilibrium logic reappears in sovereign debt and in currency pegs.
- What a run on a money market fund or a stablecoin shares with a run on a bank, and what it does not.
Key terms
TERMS #| Term | What it means |
|---|---|
| Maturity transformation | funding long-dated assets with short-dated liabilities, the core service a bank provides. |
| Illiquid vs insolvent | unable to raise cash today at fair value, versus owing more than the assets are worth at any horizon. |
| Fire sale | forced selling into a market too thin to absorb it, realising a loss that holding would not have caused. |
| Lender of last resort | a central bank lending against sound collateral at a penalty rate to a bank judged illiquid rather than insolvent. |
Every term the collection defines is gathered in the glossary.