The term premium is the compensation investors require for bearing the risks associated with holding a long-term bond instead of repeatedly rolling over short-term securities. It is not directly observable but can be estimated using term structure models.
A long-term yield can be decomposed into two parts:
The Adrian, Crump, and Moench (ACM) model, published by the Federal Reserve Bank of New York, is the most widely used term premium estimate. It decomposes Treasury yields into expectations and risk premium components using a no-arbitrage affine term structure model.
Term premiums can be negative, as they were for much of the 2010s during quantitative easing, when central bank purchases compressed the compensation for holding duration risk. Positive term premiums tend to emerge when inflation uncertainty rises, fiscal deficits expand, or the Fed is reducing its balance sheet.
Changes in the term premium drive long-term yields independently of expectations about short-term rates. This distinction matters for interpreting yield curve movements: a rising 10-year yield could reflect either higher expected policy rates or a rising term premium, with very different economic implications.
The market shows one number where the theory needs two.
A 10 Yr yield of 4.00% is a fact. Whether that number is a 4.00% expected average policy rate with no premium, or a 3.50% expected path plus 50 bps of premium, is not a fact. Both decompositions produce the same observable yield. Only the sum trades.
This is why the term premium is always an estimate and never a quote. Anyone reporting a term premium has assumed something about the other half of the split.
Two ways of assuming it are in common use.
An affine term structure model works from the shape of the entire curve rather than from any single yield.
It starts by extracting a few factors that explain nearly all the variation across tenors, usually a level factor, a slope factor, and a curvature factor. It then estimates how those factors have historically evolved, which produces the expected path of short rates. Applying no arbitrage gives the yield that path alone would justify. The gap between that yield and the observed yield is the term premium.
Two cautions follow from the method, and both are worth carrying.
The level is model dependent. Different models applied to the same curve on the same day can differ by 50 bps or more on the level of the premium. Treat any single estimate as one reading rather than a measurement.
The direction is more reliable than the level. Models tend to agree on whether the premium is rising or falling even when they disagree on where it sits. That makes the change in the estimate more useful than its value.
The estimate also depends on history. A model fitted through a long period of falling rates carries that experience into its expected path, which is one reason estimates were revised after the inflation of the early 2020s.
The decomposition earns its keep when a long yield moves and you need to know why.
If the 10 Yr rises because the expected policy path rose, the market is saying the economy is stronger or inflation is higher than it thought. Policy is expected to respond. The front of the curve usually moves with the long end, so the curve flattens or shifts.
If the 10 Yr rises because the term premium rose, the market is saying it now wants more compensation to hold duration. The expected path has not changed. This can follow heavier Treasury supply, a central bank reducing its holdings, or rising uncertainty about inflation. The front end often stays anchored, so the curve steepens.
The two cases have different consequences beyond the bond market. A premium driven rise tightens financial conditions without any improvement in growth, which is the less comfortable of the two. It also weakens the case for holding long duration as a hedge, because the yield rose for reasons unrelated to the economic cycle.
The forward rate curve carries the same ambiguity. A steep forward curve can reflect expected tightening or a demand for premium, and only a decomposition separates them.
Yes, and it has been for extended periods. A negative premium means investors accept less yield on a long bond than the expected path of short rates alone would justify. Large scale central bank purchases can produce this by removing duration from the market. Strong demand for a safe long dated asset can do the same.
Not by itself. It means investors are paying for something other than yield, such as the hedging value of a bond that rallies when equities fall, or a regulatory requirement to hold long dated government debt. Those are real reasons to own an asset, and they can persist for years.
No. It is a model output rather than an instrument, so no security has the premium as its price. The closest expression is a long duration or curve position, which takes on the premium together with everything else moving that yield. A trade sized from a premium estimate therefore carries model risk on top of market risk. The ACM model page covers how the estimates themselves behave.