Overnight repo lending
A Socratic walk-through of overnight repo lending — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why will a lender hand a near stranger a fortune overnight without ever examining the borrower's books?
A money-market fund lends a securities dealer a very large sum before lunch and expects it back the next morning. It has not read the dealer's accounts, does not know what the money is for, and would struggle to name the firm's chief executive. On any ordinary account of lending this is reckless.
Yet this is among the largest and historically among the safest lending markets there is. Which suggests the lender has found a way to make the borrower's soundness beside the point. How would you build a loan that does not require you to know anything about who is borrowing?
Reasoning it through
REASONING #Ask first what makes lending risky. Not that the borrower might fail — that you might fail to be paid when they do. Guarantee the second and the first stops concerning you.
So try removing the borrower from the transaction. Instead of a loan, do a sale: the dealer sells you a government bond today for 98 and agrees to buy it back tomorrow for a fraction more. Economically that is a secured overnight loan at an interest rate. Legally it is two purchases, and the difference matters enormously. You are not a creditor with a claim on a bond — you own the bond. If tomorrow's repurchase never happens, you do not join a queue behind other creditors or wait years for an administrator's ruling. You already have the asset and can sell it.
That is the first pillar and it is a legal one: in the major jurisdictions repurchase agreements are carved out of the usual insolvency stay, precisely so this can be true.
Now the second pillar, which is arithmetic. The lender advances 98 against a bond worth 100. That two-point gap is the haircut, and it defines exactly how wrong the lender may be. If the bond falls one per cent overnight, they hold 99 against a 98 claim and are whole. They lose only if the collateral falls more than two per cent before it can be sold. For a short-dated government bond, an overnight move of that size is very rare — which is why haircuts on such collateral are small and haircuts on volatile or illiquid collateral are large. The haircut is not a fee. It is the lender's stated tolerance for being wrong about the price.
Notice that the same gap disciplines the other side. The borrower has pledged 100 to obtain 98. Walking away means forfeiting more than was borrowed, so default is unattractive quite apart from any court. The over-collateralisation is a hostage, working in both directions at once.
Third, the tenor. Overnight is not a quirk of convention; it is the lender's exit. Every morning the loan matures and the lender chooses whether to roll it. Any doubt about the borrower or the collateral can be acted on within a day, without negotiation, without a covenant, without needing anyone's agreement. The right to walk away tomorrow substitutes for the right to know anything today.
Put those together and the design is clear: the credit analysis has moved off the borrower and onto the collateral, and the collateral is chosen to be the sort of thing nobody needs to investigate. Government bonds are attractive here not because they yield well but because any two parties will agree what they are worth without research. That is the property being purchased.
Which gives the test. If the mechanism is really collateral rather than counterparty, the price of repo money should vary far more by what is pledged than by who is borrowing. That is what is observed: rates against standard government collateral sit in a narrow band across borrowers of quite different standing, while haircuts differ substantially across collateral classes. If we instead saw repo rates tracking each borrower's own credit spread while haircuts stayed flat, this whole account would be wrong.
And the mechanism names its own failure. The market is safe only while the collateral remains something nobody needs to think about. Let doubt enter its value — as it did with mortgage-backed collateral from 2007 — and lenders must suddenly form an opinion about the asset. They respond by raising haircuts, so the borrower must find more collateral or shrink, and by declining to roll. That is a run, but not the classic one: what fails first is not confidence in the borrower but the assumption that the collateral needed no examination.
The analogy
THE ANALOGY #A pawnbroker. He does not ask your name, your employer or what you want the money for. He looks at the watch, decides what he could get for it in a hurry, and lends less than that. If you never return, he has lost nothing, because he is holding the watch and can sell it.
the pawnbroker's loan runs for months against an item whose value he assesses once and then locks in a drawer, whereas a repo lender's collateral is repriced continuously in a market and the loan matures every morning — so his real protection is not the haircut alone but the ability to stop lending tomorrow, an option the pawnbroker does not have.
Clarifying the model
THE MODEL #Three refinements.
The collateral is not idle in a vault. Lenders commonly re-use securities received as collateral to secure their own borrowing, so one bond can support several transactions at once — efficient, and also a channel by which a disruption anywhere in the chain propagates further than any single loan's size suggests.
Second, "does not examine the books" is a claim about what the structure requires, not about what prudent firms do. Serious lenders do set counterparty limits. The point is that the market's daily volume could not exist if every loan required real diligence — the design is what makes not-knowing tolerable.
Third, the uncomfortable half. The insolvency carve-out that makes repo safe does not make the risk disappear. It moves repo lenders to the front of the queue, so when a firm fails, more of the loss falls on unsecured creditors, and in a large enough failure on the public. Some of repo's celebrated safety is genuine engineering, and some is priority granted at somebody else's expense. It is worth being honest about which part is which.
A picture of it
THE PICTURE #How to readRead downward as time, across a single night. The first two exchanges are the whole loan, dressed as a sale and a promised repurchase — the note beneath is the point of that dressing, because owning the bond is what removes the insolvency queue. The upper block is the ordinary morning, and the interest is tiny: a night on 98 at four per cent a year is 98 times 0.04 divided by 360, about a penny. The lower block is the only case the lender has to survive, and it involves the market for the bond rather than the dealer at all — which is why the dealer's own soundness never had to be assessed.
What became clearer
WHAT CLEARED #Repo works by making the borrower irrelevant. Legal title replaces a creditor's claim, so the lender need not wait on anyone's insolvency; a haircut states exactly how far the collateral may fall before the lender is hurt; and an overnight term means the decision is remade every morning, so the right to leave stands in for the right to know. The market's fragility is the same fact seen from the other side: it depends on collateral that nobody needs to investigate, and the moment investigation becomes necessary, haircuts rise and the funding disappears in days.
Where to go next
ONWARD #- Why a rising haircut and a falling asset price can drive each other, and what stops the spiral.
Key terms
TERMS #| Term | What it means |
|---|---|
| Repurchase agreement | a sale of securities combined with an agreement to buy them back at a set price on a set date, economically a secured loan. |
| Haircut | the margin by which pledged collateral exceeds the cash advanced, expressed as a percentage of the collateral's value. |
| Rehypothecation | a lender's re-use of collateral it has received to secure its own borrowing. |
Every term the collection defines is gathered in the glossary.