The Gilt market one year on: evaluating our fixed income expected return estimates

In July 2025 we published Why are UK Gilt yields trading at high levels?, where we argued that the rise in UK yields between 2023 and 2025 was not the 2022 mini-budget episode repeating itself. In 2022, the spike was driven almost entirely by term premiums, and was therefore a tactical opportunity. In 2025, however, our decomposition of UK yields showed the opposite: the term premium was negative while the increase in equilibrium real rates drove the increase in yields. This meant that the compensation for duration risk in the Gilt market, and consequently the expected return on Gilts, was low, at least at short horizons. The table below shows our expected return estimates at the one-year horizon for major government bond markets on 10 July 2025, the day we published the Insights piece.

The UK is the only market in the table with a negative Value component. Carry on Gilts was among the highest on offer at 3.9%, but a negative term premium contribution subtracted 1.6%, leaving a total expected return of 2.3%, roughly half the prevailing yield. On yield alone, Gilts looked like one of the more attractive markets in the set. The realised return over the following year was 2.5%, close to the estimate, demonstrating the value of models that identify granular drivers of the yield curve in a volatile macro environment.

The chart below shows that, one year later, the market-perceived equilibrium real rate is still elevated, and the term premium is still negative, once again indicating relatively low expected returns on Gilts going forward. There are several candidate explanations for why markets anticipate persistently elevated real rates, ranging from the Bank of England being forced to run permanently tighter monetary policy to a lack of structural demand for Gilts from pension funds.

The focus for the UK is relatively narrow, of course, and a natural question is how our expected return estimates performed across other markets. The chart on the left below shows the expected and realised one-year returns for major markets and segments, starting on 10 July 2025. The forecast period coincided with a large negative shock to global duration: realised returns came in below expectations almost everywhere, with Japanese segments furthest from the line. What did not happen is an inversion of the ordering; the cross-sectional relationship remained positive at 0.31. A large part of the increase in JGB yields is due to an increase in term premium, as we have discussed previously. For comparison, using yields instead of our estimates of expected returns would have led to worse results, achieving a correlation of 0.23.

To benchmark the correlation over the past year against the history, the right panel in the chart above shows the scatterplot of expected and realised one-year returns on the same set of markets and segments for the entire sample period, using non-overlapping observations. The correlation is 0.48, somewhat higher but well within expected variation.

Note that the evaluation over the last 12-month period is a genuine out-of-sample exercise. With very few exceptions, CMA providers do not report one-year expected returns, so it is hard to systematically benchmark our estimates against others at that horizon. One provider reporting one-year estimates achieved a correlation of 0.22, which is materially lower than our 0.48 over multiple periods or 0.31 for the most recent period.

It is possible, however, to benchmark longer-horizon 5- and 10-year estimates against other providers. One thing to note is that the fixed income asset class we are evaluating has a reasonable degree of demonstrated predictability for its returns. Using a directly observable yield as an expected return gives investors a relatively good proxy, and realised returns tend to line up better with expected returns compared to other asset classes such as equities. This is mostly because the cash flows do not vary and risk premiums are relatively volatile. In one of our previous Insights pieces, we have shown that major CMA providers publish estimates of expected equity returns that are negatively correlated with subsequently realised returns. This is not the case in fixed income, where it is mostly a matter of how positive the correlation is.

The charts below plot our expected returns against subsequently realised returns at the 5-year (left panel, correlation 0.80) and 10-year (right panel, 0.82) horizons.

For comparison, below are the same charts from large investment houses and consultancies, whose estimates are available. The correlation at the 5-year horizon across all providers is 0.45 (left panel) and at the 10-year horizon is 0.41 (right panel). The breakdown by provider is shown directly in the charts.

Our estimates of expected returns are more reliable than those of the major vendors. The main reason is how the yield curve is handled. A government bond yield is frequently used directly as a return forecast. This treats the yield as a single object when it is not. Decomposed, it contains components with materially different degrees of persistence. The equilibrium real rate and long-horizon inflation expectations move on a secular timescale and set the level of the curve. Policy rate expectations and the term premium are cyclical, and they account for the bulk of the variation that resolves over the horizons an allocator can act on.

This distinction determines the expected return. Two markets trading at an identical yield can have different expected returns if the composition of yields diverges. The UK Gilt market is a good example. Our 2025 decomposition showed a negative term premium alongside a high and rising equilibrium real rate: a market whose yields had reset structurally rather than dislocated temporarily.

Decomposing yields into economically meaningful drivers is also what helps produce unbiasedness rather than mere ordering. A predictor can rank markets correctly while being systematically too high or too low in level. Our estimates sit close to the 45-degree line with the magnitude approximately right, not just the ranking. The peer panels sit well below it.

The single Gilt call matters less than what sits behind it: the same decomposition produces these relationships across two decades, eleven markets, and horizons from one to 30 years.