Term premium has risen, and may not have peaked

The importance of “term premium” in discussions about yield curves has been on the rise this year. As a reminder, term premium is an estimate (and it is only an estimate) of the additional yield investors require for locking away their money in, say, 10-year bonds instead of buying a short-term risk-free instrument (think 3-month T-Bills) and reinvesting it over that same 10 years. All else equal, higher term premium leads to higher yields for government bonds in longer maturities.

There is a long list of variables that can influence term premium, but broadly speaking the higher the uncertainty about the predicted path of short-term interest rates in the next 10 years, the higher the premium should be. It follows that this premium should rise when uncertainty about future inflation rises (as this increases uncertainty about the future of short-term rates). Also, if budget deficits are elevated and there is an elevated supply of 10-year bonds, then yields required by investors should increase even if expectations for short-term rates remain unchanged. Importantly, a steepening of the yield curve does not necessarily mean term premium has increased. For example, higher yields at the long end of the curve may simply reflect higher expectations of future short-term rates rather than any particular uncertainty about rates, and in that case the term premium might remain unchanged.

Given that the world is experiencing more volatile and higher inflation because of the US-Iran war, and the supply of US Treasuries (USTs) shows no sign of slowing down, it would make perfect sense to see term premium moving higher this year.

The reality, however, is that this depends on how we estimate the future path of short-term rates, which as you can imagine is probably more art than science. The two most widely cited term premium models are the Adrian, Crump and Moench (ACM) and the Kim-Wright (KW), both of which are produced by the Federal Reserve (Fed).

While the term premium estimates are in the same ballpark, there has been a notable divergence in recent weeks (see Exhibit 1). The ACM model shows a reduction in the term premium year-to-date, from 75bp to 61bp, while the KW model suggests a significant increase from 57bp to 96bp. It is also possible the measures have diverged even more in recent days – a slight lag in the publication of the KW index means that data only runs to September 11, while the data for the ACM model runs to September 17.

While getting different answers from different econometric models may seem trivial, for fixed income investors the point is that they feed into the market narrative around rates and can also influence policymakers’ thinking. As an example, Stephen Miran – a former member of the Fed and former chairman of the Council of Economic Advisers – wrote an article in the Financial Times in late August arguing that the term premium had not increased in 2026 and was in fact slightly lower than at the beginning of the year, concluding there were no fiscal fears being priced into yield curves. Our view would be that the KW data series is more consistent with the current narrative in markets, and we would be cautious in concluding that term premium has not increased simply because the ACM model says so.

The main difference between the ACM and KW term premium models is that the former uses exclusively information contained in the yield curve, whereas the latter also accounts for a survey of future short-term interest rates from Blue Chip Financial Forecasts. In other words, KW incorporates explicit forecasts of future short-term rates by economists. The aim of doing so is to provide some sort of anchor for long-run average short-term rates which helps with assessing how quickly short-term rates return to that average. In practice, this results in lower volatility in the KW term premium model than that of the ACM.

Looking at past data, we find it hard to believe that market participants shifted their expectations for the 10-year term premium by some 150bp in the space of a couple of months, as the ACM model suggests occurred back in 2023. In addition, the ACM model shows a low of -150bp term premium back in 2020, while the KW bottomed out at a more moderate -60bp in the same period.

We conclude that the term premium has most likely moved significantly higher this year (and at a minimum has not decreased), confirming a trend that has been in place since mid-2020. The big question of course is how much further it can go, which obviously has implications for the direction of longer dated government bond yields.

The KW series starts in 1990 and a couple of years later it reached its all-time high of 251bp. Averages can vary quite significantly depending on the starting and ending point of the sample, but for what it's worth, the average between the beginning of the year 2000 and September 2007 (when the Fed begun cutting rates in response to signs of trouble ahead) is almost exactly 100bp. While from that point of view we could say that the term premium has normalised to pre-2008 levels, we would point out that the 100bp average hides a peak of 180bp in 2002 and a couple of other lower peaks of 165bp and 140bp in the period. With underlying forces of inflation volatility and increased UST issuance, it would be a brave call to suggest term premium has peaked at this stage.

While this does not necessarily mean that USTs cannot rally (we think they would if oil prices declined, for example), it does mean that all else equal we see an ongoing headwind for UST performance.