Treasury Bought Bonds.
The Market Wasn't Buying It.
A bigger buyback lifted yields for one session in August. Three weeks later a $6 billion operation landed at the low end of expectations, and the 10-year note went on to its highest close since November 2023.
Treasury doubled the maximum size of its long-end liquidity-support buybacks on August 19, and the bond market rallied hard: the 10-year yield fell 5.7 basis points to 4.647%, the 30-year roughly 9 basis points to 5.196%.1,2 The rally didn't survive the next session. By August 20 the 10-year had climbed back to 4.70% and the 30-year to 5.249%, with Reuters attributing the move to the same inflation and debt worries the buyback was supposed to quiet.3
Three weeks passed before Treasury named a number. On September 9 it disclosed an operation of up to $6 billion in 10-to-20-year notes for the following day, above the $4 billion floor but short of the $6 billion to $10 billion range dealers had penciled in. BNP Paribas had put the figure needed to surprise the market at around $7 billion.8,12 Treasury came in under that. The 10-year yield rose to 4.8528%, its highest close since November 2023.4,13
Set against $739 billion in privately held net marketable borrowing for the quarter, a $6 billion purchase is a rounding error: about 1/123rd of the number that actually matters to duration investors.5 That ratio is this note's central finding. Treasury's liquidity program can tighten spreads on specific off-the-run CUSIPs. It has not shown it can move the benchmark long end against fiscal supply, inflation uncertainty and a rising term premium working the other way.
The Reversal of the Rally
Treasury's August 18 release increased the maximum size of long-end liquidity-support operations in the 10-to-20-year and 20-to-30-year sectors from $2 billion to at least $4 billion, effective from September 9 through the November 4 quarterly refunding. Treasury said the change was meant to add liquidity support in sectors where it receives substantial volumes of high-quality offers.6
The market read it as a surprise, and it traded like one. Reuters had the 10-year down 5.1 basis points intraday to 4.655%, with CNBC recording a 4.647% close; the 30-year fell to a 5.196% close, down roughly 9 basis points on the day.1,2,11 It was the sharpest one-day move the long end had seen in weeks, and it happened without a Fed meeting or a data surprise behind it. Just a change to how Treasury buys back its own debt.
One session erased most of it. The 10-year rose 4.7 basis points to 4.70% on August 20; the 30-year rose 5.5 basis points to 5.249%, pulling it back toward the roughly 5.34% intraday high it had touched days earlier, a level not seen in 19 years.3,7 Reuters called it a renewed selloff, driven by the same inflation and debt concerns that had been building before Treasury's announcement.3
30-Year Treasury
Yields
Aug–Sep 2026
[Editor's note] Values are directly sourced from contemporaneous Reuters, CNBC and Yahoo Finance reporting cited in the text. The 30-year move on August 19 is reported by Reuters as "approximately 9 basis points" rather than an exact print; the September 9 30-year level is reported only as "moved toward 5.3%" and is plotted here as 5.30% — both are approximations, marked unverified to an exact tick. No daily or intraday series was available for the three weeks between August 20 and September 9 in the source material used for this note, so that segment of each line is shown dashed rather than solid, and no data point is plotted between the two confirmed sessions. The y-axis is one shared scale for both series, so the announcement-day move and the following session's retracement are directly comparable at their true relative size. Sources: Reuters (Aug 19, Aug 20, 2026), CNBC (Aug 20, 2026), Yahoo Finance (Sept 9, 2026).
The gap between those two sessions is the whole story. A demand shock from a policy surprise can move yields for a day. It takes something more durable to make that move stick: a change in expected supply, inflation, or short-rate paths. The August episode didn't have one.
What the Treasury Is Actually Buying
The program runs through reverse auctions in older, less liquid nominal coupon securities: not new issuance, and not a Federal Reserve balance-sheet operation. It is a secondary-market tool aimed at the distribution of outstanding CUSIPs, and Treasury has been explicit that it doesn't touch the government's aggregate financing requirement. The August 2 borrowing release said as much directly: buybacks aren't expected to significantly affect privately held net marketable borrowing, because new issuance replaces whatever gets purchased.5
Treasury's August 4 refunding materials had penciled in up to $38 billion of off-the-run liquidity-support purchases across maturity buckets for the quarter, alongside up to $25 billion of cash-management purchases in the one-month-to-two-year bucket. The TBAC presentation that followed reported roughly $20.1 billion of liquidity-support purchases and $24.6 billion of cash-management purchases actually completed through late July.9,10
| Measure | Amount | Relative to a $4–6B operation |
|---|---|---|
| Long-end operation announced for Sept. 10 | Up to $6B | Reference amount |
| Previously indicated long-end maximum | ≥$4B / operation | $6B is 1.5× the $4B floor |
| Privately held net marketable borrowing, Jul–Sep 2026 | $739B | ≈123× a $6B operation |
| Privately held net marketable borrowing, Oct–Dec 2026 | $628B | ≈105× a $6B operation |
| Liquidity-support buybacks planned, Aug refunding quarter | Up to $38B | ≈6.3× a single $6B operation |
| Liquidity-support buybacks completed through late July | ≈$20.1B | ≈3.4× a $6B operation |
Borrowing
Scale
Structure · 2026
[Editor's note] The x-axis is logarithmic because the three figures span more than two orders of magnitude; a linear scale would render the $6B and $38B bars invisible next to $739B. Exact values are printed on each bar. This chart measures scale, not expected price impact — a $6 billion purchase can move a thinly traded CUSIP more than its face value implies if it lands when dealer inventory is constrained. Sources: U.S. Treasury marketable-borrowing release (Aug. 2, 2026); Treasury refunding statement (Aug. 4, 2026).
None of this says the ratio is a forecast of impact. A buyback can matter more than its size suggests if it hits an illiquid bucket at the right moment. But it fixes the program's ceiling: liquidity support can move the margin, not the primary-market supply behind a borrowing number two orders of magnitude larger.
Why the Operation Failed to Move the Long End
Fiscal supply, inflation uncertainty and the term premium were all working against the buyback that week, and none of them are things a reverse auction in off-the-run CUSIPs can fix.
Fiscal supply
CBO's February 10, 2026 baseline put the fiscal 2026 deficit at $1.9 trillion, or 5.8% of GDP, rising toward $3.1 trillion, or 6.7% of GDP, by 2036. Debt held by the public goes from 101% of GDP in 2026 to 120% by 2036 on the same projection.14,15 Those numbers don't set Tuesday's yield. They set the supply-risk floor under the term premium that does.
Inflation and rate uncertainty
Reuters' August 19 report put the 5-year TIPS breakeven at roughly 2.286% and the 10-year at 2.304%, above 2% and not consistent with an inflation scare, though not nothing either.1 The long-end selloff reads as a combination of inflation compensation, real-rate expectations and term-premium pressure. It isn't a single clean inflation shock, which is part of why a liquidity announcement couldn't offset it.
The term premium itself
Estimates diverge by model. The San Francisco Fed's yield-decomposition put the 10-year term premium at approximately 1.29% on September 2, against a 4.85% yield. The FRED ACM-style series had it at roughly 0.8892% on September 4.16,17 The gap is methodology, not a data contradiction. The direction both series point to is the same: long-end yields carry real compensation for risk beyond the expected path of short rates, and that compensation had been rising.
Foreign demand complicates any simple "buyers' strike" story. Treasury's June TIC data, published August 16, showed $207.1 billion of foreign purchases of long-term U.S. securities ($169.8 billion from private foreign investors, $37.3 billion from official institutions), even as foreign holders cut Treasury-bill positions by $29.0 billion.18,19 Foreign buyers didn't walk away. They just weren't buying enough long duration to hold the August rally, and a $6 billion buyback wasn't going to change their calculus either.
Reading the Signal Correctly
Separate three levels of pricing here, because conflating them is how a reasonable program gets mistaken for a failed one, or a disappointing print gets mistaken for evidence the whole strategy doesn't work.
Tactical. The August announcement produced a genuine one-day rally. Most of it was gone within 24 hours.2,3 Expectations. The September $6 billion figure beat the stated $4 billion floor but missed the range dealers had priced in, and yields rose on the gap between what was promised and what was delivered.4,8 Fundamental. Fiscal borrowing, inflation uncertainty, policy-rate expectations and the term premium kept doing what they were doing before Treasury said anything, largely untouched by the operation.5,14,17
None of that means buybacks are pointless. It means their honest job is liquidity and relative value in specific securities, with a weaker and less durable pull on the benchmark yield everyone actually quotes.
What would change this
Dealer capacity, auction demand and a credible deficit path would each give the program more room to work. Greater dealer balance-sheet capacity would let Treasury's reverse auctions pass through further into off-the-run liquidity. Stronger underlying auction demand (domestic or foreign investors already adding duration) would let purchases reinforce a real buying impulse instead of trying to manufacture one. A credible multi-year deficit path would chip away at the fiscal component of the term premium, giving liquidity support a friendlier backdrop to operate in. None of those three conditions were in place in August or September.
Through the November Refunding
The larger operation sizes hold through the November 4 quarterly refunding, when Treasury will set the next round.6 Judge the remaining operations on market functioning, not headlines: bid-to-cover and tail behavior at relevant auctions, bid-ask spreads and off-the-run-to-on-the-run pricing gaps, dealer inventories, and whether benchmark yield moves after each operation actually persist.
A program working as intended looks like tighter off-the-run spreads and steady auction demand, with no deterioration in market functioning. A program falling short looks like what September 9 produced: repeated announcement-day yield increases, spreads that don't tighten, and a term premium that keeps climbing regardless of operation size. Those are different questions from whether Treasury can lower outright borrowing costs, and this note's evidence says it can't do that on its own.
Liquidity trades, not duration calls
Buyback announcements can open tactical opportunities in eligible off-the-run securities and relative-value positions. Treating one as a bullish signal for outright 10-year or 30-year duration is a different, weaker trade than the data support.
Expect event volatility, not trend
Buyback calendars, CUSIP disclosures, operation results and quarterly refundings are likely to keep generating curve and liquidity volatility around each announcement. The market has shown it prices realized size against expectations, not the stated existence of the program.
What would move our assessment
A credible multi-year deficit path, sustained dealer capacity, stronger auction demand, or a clear catalyst for term-premium compression. Absent those, larger operations without improved market-functioning metrics carry diminishing marginal effect on benchmark yields.
Risk framing
Treat any interaction between debt management, fiscal policy and monetary policy as a market-pricing risk, not a change in institutional mandate. Nothing in the evidence reviewed here establishes the latter.
Prepared by: Icarus Asia Research, Credit & Macro desk.
Methodology: Event-window comparison of Treasury long-end liquidity-support announcements (Aug. 18–20 and Sept. 9–10, 2026) against contemporaneous 10-year and 30-year Treasury yield moves reported by Reuters and CNBC; scale comparison of announced operation sizes against Treasury's own published borrowing and buyback estimates; term-premium context drawn from the San Francisco Fed and FRED ACM-style series.
Key assumptions: Yield figures use Reuters/CNBC market reporting for the announcement and follow-through sessions rather than a single proprietary vendor feed (Bloomberg or Tradeweb closes were not available in the source material for this note); a production version intended for trading use should be rebuilt on one consistent vendor series before circulation. The August 18 pre-announcement reference level is not directly reported and has not been included as a data point; only confirmed prints are charted. The 30-year move on August 19 is sourced as an approximate figure ("approximately 9 basis points") rather than an exact print and is labeled as such in Figure 1.
Data provenance: All figures are drawn from the primary sources linked inline throughout this note. None are Icarus Asia estimates or models unless explicitly labeled “Icarus Asia calculation” in the text.