Can the Fed and Treasury Keep Bond Yields Down? A Historical Analysis
- Brett Owens
- Aug 25
- 9 min read
Updated: 3 days ago
The Finding
Can central banks and Treasuries actually hold long-term interest rates down? I realize there are narratives out there. But what does the data actually say?
I measured 14 episodes across the U.S., Japan, Australia, and Spain, using actual government data.
Surprisingly, these interventions are not successful at all. They succeed at keeping yields lower for mere days, not weeks, not months. Start with the United States. Across nine episodes, we measured the median time it took for the 10-year yield to "round trip" back to its pre-announcement level. We typically see yields relax after these announcements, the bond market calm down. But the median relief is just two trading days. Four of the nine bought no relief at all. The 10-year closed at or above its starting level on announcement day itself.
Narrow it to the operations where the Treasury buys back its own bonds (recycling debt, selling more at the short end so it can buy back at the long end), and the median relief is exactly zero days.
The only interventions that bought more than a month came during genuine crises: 2008 and 2020, both financial collapses. Those were quantitative easing, where the Fed printed the money to do it, more than a recycling effort.
And the most durable intervention in the entire study? The ECB's 2012 OMT backstop. Believe it or not, the ECB never purchased a single bond. Spanish 10-year yields stayed down for the full two years we measured, on nothing but Mario Draghi's threat to do whatever it takes. Draghi's threat outperformed actual buybacks by two orders of magnitude.
Why I Ran This Study
The U.S. Treasury resumes its buyback operations on September 9. They'll be selling additional short-term debt and using the proceeds to buy back long-term debt.
Every time an intervention like this is announced, we hear the same narrative: the government is "supporting the bond market," doing it to bring long-term yields down and support the prices of long-duration bonds. (Bonds, remember, tend to move inverse to interest rates.) But is that actually what happens?
Instead of getting pulled in two directions by the talking heads, why don't we look at every major intervention in long-duration paper since 2008? Let's see how this actually works. And yes, as I mentioned, the relief tends to last only days.
The Method
Definitions first. The relief window is the primary measure: we take the 10-year yield the trading day before the announcement, then measure how long the relief lasts, how long the 10-year trades below that level. In theory, the relief could be long-lasting, even permanent. In practice, it tends to take only a handful of trading days before the yield goes right back to where it was.
Second, sensitivity. Here we require three consecutive closes above that baseline, in other words a definitive retrace.
Day count: day zero is announcement day, so the day after is day one.
The series: U.S. daily Treasury yields and the 10-year term premium, daily JGB yields from Japan's Ministry of Finance, the daily Australian 10-year from the RBA, and the daily Spanish 10-year from the Banco de España.
The U.S. Record: Nine Episodes
# | Date | Event | Day-1 move (10y) | Max relief | Days to round-trip | Sensitivity check |
1 | Nov 2008 | QE1 announcement | −24bp | 127bp | 121 | 122 |
2 | Mar 2009 | QE1 Treasury expansion | −51bp | 51bp | 26 | 28 |
3 | Nov 2010 | QE2 | +4bp | none | 0 | 4 |
4 | Sep 2011 | Operation Twist | −7bp | 23bp | 4 | 4 |
5 | Dec 2012 | QE3 Treasury extension | +6bp | none | 0 | 0 |
6 | Oct 2019 | Bills + repo response | +9bp | none | 0 | 0 |
7 | Mar 2020 | Unlimited QE | −16bp | 40bp | 160 | 186 |
8 | May 2024 | Buyback, first operation | +7bp | none | 0 | 142 |
8b | Jun 2024 | Buyback, first long-end operation | −4bp | 5bp | 2 | 2 |
The sensitivity column applies the looser round-trip definition (three consecutive closes above baseline). Where the two columns diverge sharply, which happens on event 8, we discuss it below rather than citing either number alone.
The median across all nine events was just two trading days: 0, 0, 0, 0, 2, 4, 26, 121, 160. Only three lasted more than twenty days: the initial QE in 2008, its 2009 expansion, and the March 2020 effort.
Why did those last? They shared two features the rest of the sample lacks. They were unanticipated, and they arrived during serious market dysfunction rather than as scheduled policy. Every well-telegraphed program round-tripped, went back to where it started, in zero to four days. That's an observation about a limited sample, not a tested hypothesis. But given the data we have to work with, it's a reasonable one.
Comparing QE with Liquidity Operations
We have nine U.S. episodes, but really two different animals: Treasury cash-management operations, and crisis-level QE. Let's break them out as separate cohorts.
Cohort | What it is | Median days of relief (10y) | On the term premium |
QE-duration (n=6) | Central bank buying duration outright | 15 | 16.5 |
Plumbing / liquidity (n=3) | Short-end and money-market operations (the buyback's category) | 0 | 0 |
The final column, term premium, matters most. It strips out the policy-rate expectation component, so it speaks to whether an operation bought any real duration relief. For Treasury buybacks, the answer is zero days.
We do have one disparate number. The first buyback operation, May 2024, scored zero days on our primary definition and 142 on our sensitivity definition, the largest divergence in the dataset, driven by a single day's close seven basis points above baseline. That's why we cite the cohort median rather than any single episode. The cohort-level zero doesn't lean on that one operation.
The closest measured analogue to what starts September 9 is the first long-end buyback operation of June 2024. It round-tripped in two trading days, the longest relief any operation in this cohort produced, against a cohort median of zero.
Japan and Australia: The Mirror Image
Japan's now-famous yield-curve-control exits and Australia's 2021 target abandonment run in the opposite direction: withdraw the suppression, and yields jump by design. So here we measure the opposite effect: how far yields rose, and how long before they returned to baseline.
Date | Event | Day-1 move | Max rise | Days back to baseline |
Sep 2016 | BOJ introduces yield-curve control (10y) | +3bp | +22bp | 2 |
Dec 2022 | BOJ widens band to ±0.50% (10y) | +15bp | +90bp | 60 |
Jul 2023 | BOJ "greater flexibility" (10y) | +11bp | +116bp | never (within 2 years) |
Oct 2023 | BOJ sets 1.0% reference (10y) | +6bp | +93bp | 3 |
Oct 2021 | RBA declines to defend target, market break (AU 10y) | +3bp | — | 7 |
July 2023 is the most interesting. Japanese 10-year yields never returned to their pre-announcement level. The later, larger-looking October announcement reverted in three days, because by then the market had already repriced. The regime break happens when the market stops believing, not when the announcement lands. Australia is the same lesson: the formal discontinuation was a non-event, because the market had already broken the target the week before.
We publish original income research like this, running the numbers before we trust the narrative. No hype, just the data. Have it sent straight to your inbox:
The Most Durable Intervention Never Bought a Bond
Here's the irony of the whole thing.
In July 2012, ECB President Mario Draghi promised to do "whatever it takes." In September, the ECB published the OMT program's technical details. Spanish 10-year yields fell 39 basis points on Draghi's speech, and never bounced back to their pre-announcement level within the next two years. At the end of that window, yields were still below where they started, and still falling. It's the only time we see that in the study. The longest U.S. episode was 160 days; OMT never round-tripped within the 504 trading days we measured.
And here's the kicker: not one bond was ever purchased under OMT. The longest-lasting relief in all fourteen episodes came from a threat that was never executed, while actual purchases of actual bonds bought a median of 15 days for crisis QE, and zero for Treasury buyback operations. Draghi's credibility outperformed actual cash operations by two orders of magnitude.
Some caveats. The Italian half of the OMT story is absent: there's no free daily Italian 10-year data, and Italy's 2012 case was the more acute one. And the euro crisis was resolving on several fronts through 2012 and 2013, so the easing in Spanish yields could owe something to that broader abatement as well.
What Income Assets Returned Afterward
As investors, we want to know what happened to income investments after these interventions. Here are the median changes across all nine U.S. episodes, measured from the pre-announcement baseline.
Asset | 1 month | 3 months | 6 months | 1 year |
S&P 500 (SPY) | +3.3% | +9.9% | +15.1% | +24.0% |
Long Treasuries (TLT) | +0.4% | +1.5% | −0.8% | −1.6% |
REITs (VNQ) | +4.3% | +9.1% | +14.2% | +14.5% |
Utilities (XLU) | −1.6% | +6.3% | +9.0% | +13.9% |
Preferreds (PFF) | +0.8% | +3.4% | +6.9% | +5.1% |
Gold (GLD) | −1.5% | +3.6% | +11.4% | +27.6% |
A caution on the REIT line: we don't necessarily want to buy REITs when the Fed intervenes. Central banks tend to intervene at moments of maximum distress, so some portion of these forward returns comes from measuring off market bottoms. The ranges are enormous. REITs run from negative 7% to plus 107%. Much of this table is rebound effect on equity prices.
The point that survives the caveat: long-duration Treasuries (ironically, the actual target of these interventions) were the only group with a negative median a year out. The relief was gone within days, and the asset the intervention was supposed to help performed the worst.
What This Means for Us as Income Investors
When these intervention headlines hit, the narrative writes itself, and we've heard it already with this one. The government has our back, so load up on long bonds.
The data says otherwise. Across fourteen episodes in four countries, the scheduled, telegraphed operations (the category the current Treasury buyback belongs to) bought a median of zero days of rate relief. Not zero months. Not zero years. Zero days.
The only things that held yields down, when anything did, were crisis-level responses where the Fed was kitchen-sinking it (2008 and 2020), or Draghi's case, where the promise was credible enough that he never had to follow through.
Those aren't conditions we can schedule off a headline. So we don't build the portfolio around them. The portfolio that got us here doesn't need the 10-year to cooperate. We want businesses and funds that keep paying while yields do whatever they're going to do. We'll leave the traders to bet on the Fed's next move, and we, as always, will take the dividends either way.
This is the kind of research we send our subscribers, free. Contrarian income ideas, backed by data, no hype. Sign up below:
Limitations
Joint effects. Most U.S. dates are FOMC statement days bundling rate decisions and guidance; March 2020 sits inside a fortnight of near-daily crisis announcements. These episodes cannot be cleanly isolated, which is why we describe "episodes like this" rather than attributing precise effects to single programs.
Small samples. Cohort medians rest on six and three episodes. The sensitivity definition is reported per episode and never as a cohort median. Three numbers do not have a meaningful central tendency.
The asset table is "afterward," not "during." Relief lasted a median of 2 to 4 days; the shortest asset horizon is one month. By then, the relief was already gone in eight of nine episodes.
Asset prices are single-source (exchange-traded fund adjusted closes from one vendor, unverified against a second) and carry that caveat.
Absent by design: the United Kingdom's 2022 gilt intervention was not measured because the Bank of England's data license does not clearly permit commercial republication of derived figures. Italy's leg of OMT was not measured because no freely licensed daily Italian 10-year series exists. We measure only what we can publish lawfully.
Data sources and required attributions
U.S. series: daily Treasury constant-maturity yields and the ACM 10-year term premium. Source: FRED, Federal Reserve Bank of St. Louis (public domain); calculations by Brett Owens.
Japan: daily JGB yields. Source: Ministry of Finance, Japan; calculations by Brett Owens. Charts and computed figures created by editing Ministry of Finance data. The underlying data is available free from the Ministry of Finance.
Australia: Source: RBA 2026; calculations by Brett Owens. Computed figures are not endorsed by the Reserve Bank of Australia. The underlying data is published free of charge on the RBA website.
Spain: Elaboración propia con datos extraídos del sitio web del Banco de España (www.bde.es). Daily secondary-market 10-year government bond yields; data as last updated August 19, 2026. Calculations by Brett Owens; computed figures are not endorsed by the Banco de España. The underlying data is available free of charge at www.bde.es.
Asset prices: exchange-traded fund adjusted closing prices, single vendor, unverified against a second source; calculations by Brett Owens.
Disclosure
Pre-commitment: the round-trip definitions, measurement windows, and cohort rules in this study were fixed in our internal research log (study TD-109) on August 22, 2026, before the analysis touched data, and the number designated for citation, the plumbing-cohort primary median, was ruled the same day, before publication. Where a definition materially changes a result, both readings are shown above.
This is research, not individualized investment advice. Nothing here recommends buying or selling any security. Yields, prices, and relationships shown are historical; past interventions do not predict future ones. Figures are as of the dates stated and will not be updated in place without a changelog note.
© 2026 How to Retire on Dividends
Comments