SOFR and the Death of LIBOR

For four decades, the floating leg of nearly every swap, the reset on your mortgage, and the coupon on hundreds of trillions of dollars of loans all pointed at a single number: LIBOR. It was the heartbeat of global finance. It was also, it turned out, a number that a handful of banks essentially made up over the phone — and that, after the financial crisis, described a market that had largely stopped existing. This lesson is the autopsy and the rebirth: why LIBOR died, what SOFR is, and the one genuinely awkward consequence of the switch — that you now have to compound an overnight rate in arrears to get the period rate you used to be quoted up front.

What LIBOR was: a benchmark built on banks’ word

Analogy. Imagine setting the national price of milk by phoning eighteen dairy farmers each morning and asking, “What do you reckon you’d pay for milk today?” — then averaging their guesses. Nobody has to actually buy any milk. That guess-based average was LIBOR.

The definition. LIBOR — the London Interbank Offered Rate — answered one question: at what rate could a leading bank borrow unsecured funds from another bank, for a given term? Every business day, a panel of major banks each submitted an estimate for several tenors (overnight, 1-month, 3-month, 6-month, 12-month) in several currencies. The top and bottom submissions were trimmed and the rest averaged. That average, published, became the reference rate that floating legs, loans, and adjustable mortgages reset against.

Three features matter, because SOFR will overturn every one of them:

• Survey-based. It came from banks’ submissions — what they said — not from a ledger of actual trades.
• Unsecured. It priced lending with no collateral, so it embedded the credit risk of banks: the chance the borrowing bank goes bust before repaying.
• Term. It was quoted for a period in advance — a 3-month LIBOR fixed today told you the rate for the next three months, known up front.

Why LIBOR died: a scandal on top of structural rot

Analogy. LIBOR didn’t die of one disease. It had a heart attack — a public scandal — sitting on top of a chronic, terminal organ failure nobody had treated.

Failure one: the rigging scandal (2012). Because submissions were opinions, they could be nudged. Investigators found traders at multiple banks colluding to shade their firm’s LIBOR submissions up or down — sometimes to flatter the bank’s apparent health, more cynically to profit on their own swap and futures positions that paid off if LIBOR fixed a hair higher or lower. A few basis points, multiplied across trillions in notional, is real money. Billions in fines followed; criminal cases were brought. A global benchmark had been treated as a tradeable knob.

Failure two: the rate had no market underneath it. This was the deeper, quieter killer. After 2008, banks stopped lending to each other unsecured for a term — it was too risky, and regulation made it expensive. So the very transactions LIBOR was meant to summarise had largely vanished. Panel banks were left submitting expert judgement about a price that almost no real trade was setting. Regulators called it “a rate in search of a market.” A benchmark resting on opinion isn’t just manipulable — it’s unanchored.

So the UK’s Financial Conduct Authority (FCA), LIBOR’s regulator, announced it would stop compelling banks to submit, setting LIBOR’s end date in motion. You can’t fine your way out of a rate that no longer measures anything.

SOFR, the replacement: secured, overnight, real

Analogy. If LIBOR was “what do you reckon milk costs?”, SOFR is a till-roll: it tots up every actual milk sale that happened yesterday and reports the real average price. No opinions, no phone calls — just the tape.

The definition. SOFR — the Secured Overnight Financing Rate — is the rate at which firms borrow cash overnight, secured by US Treasury collateral, in the repo (repurchase agreement) market. The New York Fed publishes it daily as a volume-weighted median of those actual trades — roughly a trillion dollars-plus of real transactions every day. Its four properties are the exact mirror image of LIBOR’s:

• Secured. Loans are collateralised by Treasuries. If the borrower fails, you keep the collateral — so there’s essentially no bank credit risk baked in.
• Overnight. It’s a one-day rate, not a term rate. (We’ll deal with that wrinkle next.)
• Transaction-based. Computed from a deep tape of real trades, not a survey — nothing to “submit,” and nothing to nudge.
• Nearly risk-free. With collateral and an overnight horizon, SOFR is about as close to a true risk-free rate as a real market produces.

The chart below shows the same idea swaps people care about: lock a fixed rate, or float on an overnight benchmark that can drift up, down, or sideways. The flat line is the certainty LIBOR-style term setting felt like; the moving line is the overnight reality you now have to compound.

The compounding-in-arrears challenge

The catch. SOFR is overnight. But your swap’s floating leg pays over a period — say three months. So how do you turn 90-odd one-day rates into one period rate? You compound them: roll the money forward day by day, each day earning that day’s SOFR. For a period made of days $t$, the compounded period rate is

where $d_t$ is the number of calendar days that day’s rate applies (a weekend uses Friday’s rate for all three days) and we divide by 360 for the money-market day count. It’s the same “interest on interest” you’ve seen since the very first lesson — just applied one day at a time.

The genuinely new problem. LIBOR was set in advance: on day one you knew the rate for the whole next period. Compounded SOFR is only known in arrears — you can’t finish the product until the last day of the period has printed. You learn the period’s rate roughly when the payment is due, not at the start. That’s not a bug; it’s the unavoidable cost of using a real overnight rate instead of a forecast.

Worked example. Take a tiny five-business-day stub, Monday through Friday, on a notional of $10,000,000. Friday’s rate covers the weekend (Saturday and Sunday too), so it applies for 3 days. Each daily factor takes the form $\left(1 + \text{SOFR}_t \cdot \frac{d_t}{360}\right)$:

Multiply the daily factors together and you get $F = 1.00103456$. Subtract 1 and the compounded period rate is 0.10346% over those 7 calendar days. On $10,000,000 that’s a floating-leg accrual of $10{,}000{,}000 \times 0.0010346 \approx 10{,}346$ — i.e. $10,346.

Annualise it ($0.0010346 \times \frac{360}{7}$) and you get about 5.32% — sensibly between the daily rates, exactly as it should be. Notice you couldn’t have written that $10,346 cheque on Monday: Friday’s print is part of the product, so the number only finalises at the end.

For products that truly need the rate up front. Some cash products — a small-business loan, a consumer mortgage — can’t tell the borrower their rate only at the end. For those, the CME publishes Term SOFR: a forward-looking SOFR term rate (1m / 3m / 6m) derived from SOFR futures, so it’s known in advance like LIBOR was, while still being rooted in the SOFR market. Most derivatives use compounded SOFR; many loans use Term SOFR.

Converting legacy LIBOR contracts. A swap written against 3-month LIBOR can’t just be repointed at SOFR — SOFR is lower, because it lacks LIBOR’s credit component. Repapering it cleanly would hand value from one counterparty to the other. The fix is the Credit Spread Adjustment (CSA): a fixed add-on, defined in the ISDA fallback as the median LIBOR-minus-SOFR difference over the five years before LIBOR’s cessation. Bolt that historical spread onto compounded SOFR and the converted contract lands economically close to the old LIBOR one — no windfall, no robbery.

The transition: how the world actually switched

The timeline. This wasn’t a flip of a switch; it was a multi-year migration with hard deadlines:

• End of 2021: banks stopped writing new contracts referencing USD LIBOR.
• ~June 2023: the remaining USD LIBOR tenors (including the heavily used 1m, 3m, 6m) ceased publication.
• Legacy trades still tied to LIBOR fell back automatically via the ISDA fallback protocol: counterparties who adhered agreed in advance that, on cessation, their LIBOR references would switch to compounded SOFR plus the fixed credit spread adjustment. “Tough legacy” contracts that couldn’t be repapered got legislative backstops naming a SOFR-based replacement.

The result: trillions of dollars of swaps, loans, and securities were re-pointed from a phoned-in survey rate to a transaction-based overnight rate, mostly without litigation, because the fallback math was agreed before the lights went out on LIBOR.

Putting it together

LIBOR was the London Interbank Offered Rate: a survey-based, unsecured, term benchmark — banks’ guesses about what they’d pay to borrow for a month or three, set in advance. It died of two diseases at once: a rigging scandal (an opinion you can nudge for profit) and a structural rot (the unsecured term market beneath it evaporated, leaving “a rate in search of a market”). Its replacement, SOFR, flips every property: secured by Treasuries, overnight, transaction-based ($1T+ a day), and nearly risk-free — with no built-in credit spread, which is why it stays calm in crises that once sent LIBOR soaring. The price of using a real overnight rate is compounding in arrears: you multiply daily $(1 + \text{SOFR}_t \cdot d_t/360)$ factors and only know the period rate at the end — softened by lookbacks and payment delays, sidestepped by Term SOFR for products needing the rate up front, and reconciled with old contracts via the credit spread adjustment in the ISDA fallback. New USD LIBOR stopped end-2021, the main tenors ceased mid-2023, and legacy trades fell back to compounded SOFR plus that spread. SOFR is not LIBOR renamed — it’s a different machine that happens to do the same job.

Next — swap spreads and asset swaps — we measure the gap between the swap rate and the government yield, and use an asset swap to strip a bond down to a clean spread over SOFR.