How it works
Every day, an enormous volume of cash changes hands overnight in the : one party hands over a Treasury bond and takes cash, promising to buy the bond back tomorrow at a fractionally higher price. That fraction is the interest rate. SOFR is what you get when you take a day's worth of those trades — tri-party repo, and cleared bilateral and general-collateral trades — sort them by rate, and take the median weighted by how much money moved. The New York Fed publishes the number the following morning, along with the percentiles of the underlying distribution, so you can see not just the middle of the market but its tails.
Two consequences follow from that construction. First, SOFR is essentially risk-free: the lender is holding a Treasury bond as security, so the rate reflects the price of cash, not the creditworthiness of whoever is borrowing it. Second, it is entirely backward-looking. There is no forecast in it and no judgement. It is a record of what happened.
For a borrower, the mechanics are dull and constant. Illustratively: a fund posts $100m of Treasuries, receives $98m of cash after a 2% haircut, pays a rate a hair above SOFR for one night, and does the whole thing again tomorrow. Multiply that across the system and you have the machinery that finances most of the leverage in the . Loans and swaps that reference SOFR usually do not use a single day's print; they compound the daily rate over an interest period, or use a term rate derived from SOFR futures.
The Federal Reserve does not set SOFR. It fences it. Interest on reserve balances gives banks a reason not to lend cash below that rate; the overnight reverse puts a soft floor under money funds; the Standing Repo Facility is meant to act as a ceiling by lending against Treasuries on demand. When reserves are plentiful, SOFR sits quietly inside the fence. When reserves get scarce, it starts pressing against the top of it, and month-ends and quarter-ends produce spikes as balance sheet capacity disappears for a day. In September 2019 the fence broke outright and overnight repo rates more than doubled intraday, which is how a plumbing problem announces itself.
Why it matters to this crash
Our central claim about the Treasury market is that one of its largest marginal buyers is leveraged and funded overnight. When we wrote on 18 August that nobody in the USC survey named the as the thing holding the market together, the point was that a few basis points of spread only becomes a business if you borrow the bond position at close to the overnight rate and repeat it many times — SOFR was 3.62% that day. The trade's vulnerability is not the direction of yields. It is the cost and availability of that overnight cash.
Which makes SOFR the single most useful number for testing whether the current arrangement holds. On 19 August we described the Treasury doubling its buybacks of 10- to 30-year bonds while funding them with more bill issuance, at the same time as the New York Fed cut its reserve-management purchases to zero for 14 August to 14 September. That combination pushes more short paper into the front end while withholding the bank reserves that would ordinarily absorb it. Our conclusion then was that money funds and banks have to keep taking the bills without repo tightening, and that SOFR — 3.65% at the time — was the place to watch for the failure. It had not moved.
That is the honest state of the evidence. The setup is unusual; the gauge is calm.
What would make this dangerous
A SOFR print that settles persistently above interest on reserve balances rather than touching it at quarter-end and falling back. Turn-of-month spikes that fail to reverse within a day or two. A widening gap between the median SOFR and the 99th percentile of the published distribution, which means the marginal borrower is paying far more than the typical one — is where a squeeze shows up first, because the weakest-funded participants are priced out before the average moves at all.
Sustained use of the Standing Repo Facility would be the loudest signal. It exists precisely so that nobody has to bid the market up in a panic, so meaningful take-up means somebody could not find cash at a sane price. Rising on Treasury collateral would do the same damage through a different door: force leveraged holders to shrink, and the buyer becomes a seller into a falling market. Watch, too, for restarting reserve purchases earlier than announced. That would be the Desk conceding that reserves are no longer ample, and it would be visible in SOFR before it was visible in any statement.