How it actually works
The premium is defined against a theory. If investors were indifferent to time, a ten-year yield would be nothing more than the average short rate expected over the next ten years. Rolling one-year bills ten times would earn the same as buying the ten-year, and any gap would be arbitraged away.
It does not work that way, because the person buying the ten-year is taking a risk the roller is not: they are locked in. Inflation could run hotter than anyone expects. The government could issue far more paper than anyone planned. The bond could have to be sold in a bad month at a loss. So the long bond has to pay more, and that extra is the term premium.
An illustration, with round invented numbers. Suppose the market expects short rates to average 4% over the next ten years, and the ten-year note yields 4.8%. The term premium is the 0.8 percentage points left over — eighty basis points of pure compensation for holding , unrelated to any view about .
Only one of those three numbers is observable. The yield is a market price. The expected path of short rates is not; it has to be estimated from surveys, forwards and statistical models, of which the New York Fed's is the most quoted. So the term premium is a — what is left after you subtract an estimate from a fact. Different models give different answers, and everybody using the number knows this and uses it anyway, because the alternative is having no way to distinguish "the market thinks rates will be higher" from "the market wants paying more to take the risk".
That distinction is the entire point. Those two things look identical on a yield chart and demand completely different responses. A rate-expectations move is a forecast, and a central bank can argue with a forecast. A term-premium move is a price, and the only way to change a price is to make the risk smaller or find someone willing to bear it cheaply.
The cleanest tell is what happens to rate volatility at the same time. If yields are rising because people are frightened of the next policy meeting, expected volatility rises with them. If yields rise while volatility falls, the market is calm about the near term and has simply up the cost of the far term.
Why it matters to this crash
That second pattern is exactly what we have been watching. On 20 August we wrote it up directly: the New York Fed's term premium estimate around 80 basis points, close to its highest in twelve years, while the of rate volatility sat at 9.83, down 26.9% over a month. Nobody expects a violent surprise from the Fed. They are declining to lend for thirty years at the old price.
The repricing showed up in the auctions. A 30-year sale on 11 August cleared at 5.216%, the highest for that maturity since 2001, and the 10-year the next day at the highest financing cost since 2007. By the following Tuesday the 30-year yield had reached 5.33%, a level last seen in June 2007, with the 30-year TIPS yield at 3.09% — real yields, not an inflation scare alone (our 19 August dispatch).
It is not a US story either, which is what makes it structural rather than a Washington problem. The 30-year Bund reached 3.78%, the 30-year gilt 5.86%, Japan's 10-year 2.93%, its highest since 1996. Robin Brooks noted rising almost everywhere in a week when US inflation data came in soft — France +14bp, UK +13bp, Japan +11bp. That is a risk premium moving, not a growth forecast.
Which is why the Treasury's response looked to us like the wrong tool. Doubling long-end liquidity-support buybacks from $2bn to at least $4bn per operation knocked almost 10bp off the 30-year on the Wednesday; by Thursday it had climbed as much as seven basis points and given the move back. Buybacks address a liquidity problem. The term premium is not a liquidity problem, and the surprise announcement arguably added to it: Jefferies' Thomas Simons said the break in communication "reduces the overall credibility of their guidance".
And it reaches well past the . Every AI data-centre valuation and every is discounted against a long rate that has now moved. Worse, the is competing for the same duration buyers: Morgan Stanley expects $250–300bn of AI-related debt issuance this year, into a pool where gross federal debt has reached $40tn, up $3tn in a year.
What would make this dangerous
The specific thing to watch is the divergence continuing: the term premium estimate grinding higher while the MOVE index stays near single digits. That combination says the repricing is deliberate and unhurried, which is precisely why it will not stop when policy asks it to.
After that, in rough order of seriousness. Persistent at , meaning the stated demand is not there at the screen price. Forward yields in Europe and Japan continuing to rise alongside US ones on days when US data is soft, which rules out the cycle as an explanation. Gold rising while long yields rise — we called that a credibility trade rather than a rate trade on 18 August, and it is the cleanest signal available. And a debt-ceiling episode interrupting bill issuance while Treasury is funding long-bond buybacks with bills, which we flagged on 20 August: the maturity profile has been shortened by hand, and shortening works only while the short end rolls effortlessly.