Quantitative tightening does its damage in overnight funding first, then in the market for dollars abroad, and in bond yields only last. That is the asymmetry the policy headline hides. QE pushes money into a calm system on a schedule everyone has read. QT pulls money out quietly, and nobody learns where the floor is until they hit it.
September 17, 2019: the night repo broke
In the second week of September 2019, nothing on the Federal Reserve's calendar looked dangerous. The Fed had already stopped shrinking its balance sheet at the start of August, earlier than its original plan. Bank reserves, the deposits commercial banks hold at the Fed, had fallen from a peak of roughly $2.8 trillion in 2014 to around $1.4 trillion. Officials believed $1.4 trillion was still comfortably "ample."
Then two ordinary events landed on the same day. Corporate tax payments pulled cash out of bank deposits and into the Treasury's account at the Fed, and a batch of Treasury coupon securities settled, which needed buyers to pay cash for them. Reserves dropped at the exact moment demand for overnight cash rose.
The Secured Overnight Financing Rate (SOFR), the rate for borrowing cash overnight against Treasury collateral, went from 2.43% on September 16 to 5.25% on September 17. Some general collateral repo traded near 10% intraday. The effective federal funds rate printed 2.30%, above the top of the Fed's own 2.00% to 2.25% target range. The Fed injected $53 billion through emergency repo operations that morning. By October it was buying Treasury bills at $60 billion a month and insisting this was not QE.
Here is the episode in one number. The SOFR jump was:
Overnight spread shock = 5.25% − 2.43% = 2.82 percentage points
A dealer financing a $10 billion Treasury inventory overnight paid an extra $10,000,000,000 × 2.82% ÷ 360 ≈ $783,000 for a single night. Multiply that across every leveraged holder of Treasuries and you have a market where the cost of carrying government bonds has nearly doubled in a day, with no change in the Fed's policy rate.
What this means for you: the 2019 break started in a market most investors never look at, and it reached Treasuries through the cost of financing them, not through any change in their credit quality.
Why ample turns to scarce without warning
Why did $1.4 trillion stop being enough? Because the demand for reserves is not a straight line. It is flat across a wide range and then turns almost vertical.
When reserves are abundant, an extra $100 billion barely moves overnight rates, because no bank needs it. Banks hold reserves to meet the Liquidity Coverage Ratio, a regulation requiring enough high-quality liquid assets to survive a 30-day stress, plus intraday payment needs and resolution planning buffers. Each large bank also keeps an internal "lowest comfortable level" of reserves, a number set by its treasury department and never published.
That last point produces the asymmetry. QE adds reserves everywhere at once, so every bank ends up above its comfort level and rates sit calmly on the floor the Fed sets. QT removes reserves from the system as a whole, but the scarcity shows up at the margin: at the one or two banks that would normally lend cash into repo when rates rise. Once those marginal lenders hit their private floors, they stop lending, and nobody else steps in at the old price.
In 2019, reserves were concentrated in a handful of the largest banks. Total reserves looked fine. The banks that held them had no spare capacity to deploy on September 17, because their internal floors sat higher than the Fed had assumed.
You cannot observe the scarcity point in advance, because it is the sum of private risk limits that change with regulation, market volatility, and each bank's own stress testing. The Fed finds the edge of ample the same way a driver finds black ice.
What this means for you: treat any official estimate of "ample" as a guess with wide error bars, and expect QT to end on the day that guess is proven wrong, not on a pre-announced date.
What the drain forces elsewhere
The policy narrative tracks the size of the balance sheet. The transmission runs through three markets that sit downstream of reserves.
The chain runs like this. Reserves fall toward scarcity, so overnight repo rates rise above the rate the Fed pays on reserves, known as interest on reserve balances (IORB). Higher repo costs squeeze leveraged Treasury holders, especially funds running the cash-futures basis trade, which buys Treasury bonds, sells futures against them, and finances the bonds in repo. When repo costs jump, that trade's carry turns negative, and forced unwinds put cash Treasuries on dealer balance sheets that are already full. Treasury market liquidity thins out at the very moment yields are most volatile.
The third link is the dollar. Foreign banks without access to the Fed's balance sheet borrow dollars through FX swaps, exchanging their own currency for dollars over a set period. When US banks hoard reserves, they charge more to lend dollars into those swaps, and the cross-currency basis, the premium foreign borrowers pay above the interest rate differential, widens. That premium shows up as dollar strength at quarter-ends and as pressure on any foreign institution funding dollar assets with short-term borrowing.
So the second-order move of QT is a tightening of dollar funding conditions worldwide, delivered through a repo market most rate forecasters do not model. A 25 basis point rise in the policy rate is announced and priced. A 50 basis point blowout in the cost of dollar swaps around a quarter-end is neither.
What this means for you: if you hold leveraged duration or unhedged dollar funding, your QT exposure is a funding-cost exposure first and a yield exposure second.
The economics behind the cliff
Why does the pattern exist at all? Because the actors who keep repo rates near the policy rate are balance-sheet constrained, and QT eats exactly the slack they use to do that job.
In a calm market, arbitrage holds repo in line. If repo rises above IORB, a bank with excess reserves earns more by lending cash into repo than by leaving it at the Fed, so it lends, and that lending pulls repo back down. The arbitrage is costless only while the bank has reserves above its internal floor and room under its leverage limits. The supplementary leverage ratio, a capital requirement on total assets regardless of their risk, charges the bank for every repo loan it makes, even one backed by Treasuries. Near quarter-ends, when banks report the balance sheets used to set their capital surcharges, the willingness to expand shrinks further.
QE never stresses this mechanism, because it creates reserves that land at those same large banks and push them deep above their floors. QT reverses the quantity but not the distribution. Runoff drains reserves wherever Treasury cash flows happen to pull them, including from the marginal lenders. The arbitrage that anchors repo then weakens in steps, invisible for months, until a tax date or settlement day removes the last slack.
The mechanism also has a release valve that changes the failure mode. In 2021 the Fed created the Standing Repo Facility, which lends cash against Treasuries at a set rate to eligible counterparties. In principle it caps how high repo can spike. In practice, banks avoid using it because they fear it signals weakness, so the cap is softer than it looks on paper. A spike can still happen; it just has a ceiling that someone eventually decides to use.
The mechanism breaks in the other direction too. When the Fed holds a large buffer of excess liquidity outside reserves, QT drains that buffer first. That is what happened after 2022. The Fed's overnight reverse repo facility, where money market funds park cash at the Fed, peaked above $2 trillion. For most of the second QT cycle, runoff mainly emptied that facility while bank reserves barely moved. The scarcity cliff only came back into view once reverse repo balances neared zero. The Fed slowed Treasury runoff in mid-2024, cut it again in 2025, and ended runoff on December 1, 2025, as repo rates began printing above IORB more often.
What this means for you: the size of the buffer sitting in reverse repo tells you how far QT is from the cliff; once that buffer is gone, every Treasury cash drain hits reserves directly.
When the asymmetry argument misleads
The asymmetry is real, but it gets misapplied in three common ways.
The first mistake is treating QT as a mirror image of QE for bond yields. Studies of QE found that announcements compressed the term premium, the extra yield investors demand for holding long bonds instead of rolling short ones, because the announcements told markets the Fed would keep rates low. QT is announced months in advance and runs on autopilot, so it carries much less signal. Its effect on long yields comes mainly from returning duration supply to private hands, and that effect is smaller and slower. Pinning a sharp bond selloff on QT usually overstates its role.
The second mistake is confusing the size of the balance sheet with the level of reserves. The Fed's liabilities include currency, reverse repo balances, and the Treasury General Account (TGA), the government's checking account at the Fed. A TGA rebuild after a debt-ceiling resolution, like the one in mid-2023, can drain several hundred billion in reserves within weeks with no QT at all. If you track only runoff, you miss the larger swings.
The third mistake is assuming the 2019 cliff sits at the same dollar level today. Reserve demand grows with bank balance sheets, deposit volatility, and regulation. The comfort floor in 2026 is not $1.4 trillion; it is some unknown higher figure, and the number cannot be reverse-engineered from history.
What this means for you: use the asymmetry to understand funding risk, not to forecast the 10-year yield.
The current read on rates and the curve
The 10-year Treasury yield stands at 5.28% and the 2-year at 4.83%, as of October 2, 2026, giving a 2s10s spread of +0.45%. A positively sloped curve with long yields above 5% fits a story of term premium rebuilding as private investors absorb more duration. How much of that rebuild traces to the Fed's earlier runoff, and how much to heavy Treasury issuance, cannot be separated from these figures alone. Our read is that fiscal supply is doing more of the work, with QT having set the stage by removing the Fed as a price-insensitive buyer.
Runoff has ended, but the asymmetry still governs what comes next. With reverse repo depleted, every large Treasury settlement, tax date, or TGA rebuild now pulls directly on reserves. The funding market sits closer to the steep part of the demand curve than it did for most of the last three years.
What this means for you: in the current setup, a quarter-end funding spike is more likely than a QT-driven yield shock.
What to watch instead of the balance sheet
The balance sheet tells you where the Fed has been. Four spreads tell you where the funding system is.
Watch SOFR minus IORB. When SOFR prints persistently above the reserve rate, banks are no longer arbitraging repo down, and reserves have stopped being ample in practice. Watch Standing Repo Facility usage at month-ends and quarter-ends; regular use means the release valve has replaced the arbitrage. Watch the cross-currency basis for euro and yen against the dollar, since its widening is how US funding stress exports itself abroad. And watch the Treasury's cash balance around tax dates and debt issuance, because those flows now hit reserves one for one.
