What happened in the September 2019 repo episode?
On 17 September 2019 the US overnight repo rate, which had been trading around 2.2%, printed as high as roughly 10% intraday. SOFR fixed at 5.25% that day. More striking still, the effective fed funds rate rose to 2.30% — above the top of the FOMC's 2.00–2.25% target range. The central bank had temporarily lost control of its own policy rate. It is the best available teaching case for everything in this unit.
What actually drained the cash
Nothing exotic. Two ordinary calendar events landed on the same day, on top of a system that had quietly become tighter than anyone realised.
- 16 September was the quarterly corporate tax date — the statutory 15th fell on a Sunday that year, so the payments settled on the Monday. Companies paid the government, which moved cash out of bank deposits and into the Treasury's account at the Fed. As Unit 1 showed, that drains reserves directly.
- The same day brought roughly $54 billion of Treasury coupon settlements, moving still more cash from dealers to the Treasury while loading dealers with securities to finance.
- In the background, reserves had already fallen from a peak near $2.8 trillion to about $1.4 trillion through several years of balance sheet runoff and a growing Treasury account.
The puzzle worth sitting with
The banking system held roughly $1.4 trillion of reserves. Repo was printing near 10%, secured by Treasury collateral. Why did nobody lend?
Because reserves are not the same thing as willingness to part with them. They were concentrated at a handful of very large banks; a large share was committed to intraday payment obligations and internal liquidity requirements that are not optional; and dealer balance sheets were already full of Treasuries from heavy issuance, leaving little room to intermediate. "Reserves exist somewhere in the system" and "reserves are available to the bank that needs them at 4pm" are different statements.
The response
The New York Fed conducted overnight repo operations from 17 September — the first such operations since 2008 — then added term repos, and from mid-October began buying Treasury bills at around $60 billion a month to rebuild reserves. The purchases were explicitly framed as reserve management rather than quantitative easing, a distinction that generated a great deal of commentary at the time.
The durable fix came later. In July 2021 the Fed established the Standing Repo Facility: a permanent ceiling that lets eligible counterparties borrow cash against Treasury and agency collateral at a set minimum bid rate whenever they choose. That is the same corridor logic from Unit 1 lesson 3, retrofitted after an episode demonstrated it was missing.
What it teaches
The demand curve for reserves has a long flat stretch and then a steep wall. On the flat stretch, draining reserves changes nothing at all, day after day, which is genuinely reassuring right up until it isn't. Nobody — not the market, not the dealers, not the central bank — knew where the wall was. "Ample" turned out to be a level you discover by hitting it, not a level you calculate.
The second lesson is that the trigger was mundane. No institution failed, no asset defaulted, no scandal broke. Two routine calendar events on a system with slightly less slack than assumed produced a 10% overnight rate. Fragility in money markets does not require a villain.
This is a factual account of a documented past episode. It is not a forecast, and nothing here should be read as a view on any current or future policy setting.
In the data
The episode fits in three days of one spread, SOFR minus the bottom of the Fed's range:
On 17 September 2019 SOFR printed 325 basis points above the bottom of the range, 5.25% against 2.00%, after 43 the day before and 55 the day after. Because both sides of the subtraction are shown, the spike can be pinned on the side that moved: the range did not change that day, the price of secured cash did.
Try it now
- The SOFR series holds 2,119 fixings, from the first on 3 April 2018 to 25 September 2026 (counted on 28 September 2026). The highest is 5.40, printed on 28 December 2023, 2 January 2024 and 1 July 2024, when the Fed's range was 5.25–5.50. So the 2019 spike is not the top of the series. The week around it is below: measure 17 September against its neighbours instead, and against that day's range of 2.00–2.25. How many basis points above the days either side, and above the ceiling, did it print?
- The top of the Fed's range minus SOFR for the same week is below. A negative value means the secured overnight rate printed above the top of the target range. Over the whole series, 2,119 business days, that happened on 61 days (counted on 28 September 2026), most recently on 31 December 2025 at −12; apart from 17 September 2019, the breaches ran from −1 to −65 basis points. Place the week below against that record.
- List, from memory, the three things that drained reserves in September 2019 — and then name which of them recurs every quarter.