By Andras Lengyel and Danny Walker.
Bank staff estimated a term structure model designed to separate near-term risk premia from central expectations for Bank Rate, using data on OIS rates and incorporating survey-based expectations from market participants.
The model estimates suggested that the upward slope at short maturities following the war largely reflected risk premia. After stripping out these premia, the model-implied expected path for Bank Rate was broadly flat over the next year.
The level of risk premia at the short end was unusual by historical standards, albeit not unprecedented, with higher levels seen during the very high inflation period of 2022 to 2023.
Short-end risk premia reflect the compensation investors require for uncertainty around the policy path, which we argue in turn reflected uncertainty about the war and its macroeconomic effects.
The UK overnight interest rate swap forward curve is widely used as a measure of expectations for future Bank Rate – but that assumes risk premia are small
The UK OIS forward curve – ‘the forward curve’ – is the primary financial market instrument used to read off market expectations for Bank Rate.footnote [1] The fixed rate on an OIS contract covering a given future window equals the market's probability-weighted expectation of the average Bank Rate over that window plus any compensation required for bearing any associated risks, known as the risk premium (Piazzesi and Swanson (2008)).
This distinction has rarely mattered in the recent past. When the near-term macroeconomic and policy outlook is clear and well understood, there is little uncertainty to price at the short end of the curve – by which we mean maturities under two years – and therefore little risk premium to be earned. As a result, the short end of the forward curve has historically tracked survey-based measures of Bank Rate expectations fairly closely. For example, the gap between the three-month three-month OIS rate and the three-month forecast for three-month rates in the Consensus Economics survey has a mean of -17 basis points and a standard deviation of +27 basis points.
The UK forward curve began to slope up materially following the Iran war in February 2026, but market intelligence and surveys pointed to central expectations that Bank Rate would be on hold
During the highly uncertain period following the outbreak of war in Iran in February 2026, the short end of the forward curve steepened. If read naively as a measure of Bank Rate expectations, this would have implied higher rates over the coming months. But the Market Participants Survey results (MaPS) modal expectation for Bank Rate was broadly flat over the next year, and this profile was corroborated by contemporaneous market intelligence. As shown in Chart 1, this meant a sizeable gap had opened between the forward curve and survey-based expectations of the most likely path for Bank Rate at short maturities, which was not there before the war broke out.
This post uses the outbreak of the Iran war in February 2026 as a case study. It asks a simple question. How do we explain the gap between the forward curve and market intelligence about Bank Rate expectations?
Chart 1: OIS one-month forward curve and MaPS median expected Bank Rate path: current round versus pre-war benchmark
Footnotes
- Notes: Panel A reports the same comparison for a pre-war reference round to illustrate the historically close comovement of the two series. Panel B compares the OIS one-month forward curve and the MaPS modal Bank Rate expectation in the latest policy round.
- Sources: Bank of England, Bloomberg Finance L.P. and Bank calculations.
A model tailored to estimate risk premia at the short end of the curve
Affine term structure models (ATSMs) are standard tools for decomposing yield curves into expectations and risk premia – they are used regularly at the Bank of England and at other central banks.footnote [2] Seminal contributions in this class include those of Joslin et al (2011), who show how Gaussian ATSMs can be estimated efficiently using principal components of yields as the starting point, and Adrian et al (2013), who propose a computationally convenient three-step regression estimator.
In standard ATSMs, models are estimated to fit the full cross-section of maturities as well as possible, which is useful for average fit but not always ideal for policy inference, where the relevance of the short end of the curve outweighs the rest. This is because when fit is evaluated across the full maturity spectrum, errors at the short end of the curve are traded-off against errors elsewhere. In environments where risk premia are concentrated at the short end – as in the current episode – a model calibrated in the conventional way may incorrectly attribute some of the short-end risk premia into its estimate of expected Bank Rate. In practice, standard ATSMs often fit best in the middle of the curve, beyond the short end (Joyce et al (2010) and Malik and Meldrum (2014)). In episodes like the current one, that can overstate how much markets genuinely expect Bank Rate to rise.
We addressed these limitations by adapting the workhorse Joslin et al (2011) framework in three ways that made it better suited to the February 2026 episode. First, we estimated the model on OIS rates rather than government bond yields. Gilt yields can embed liquidity premia, gilt demand and supply imbalances, and market distortions that are particularly variable at short maturities and unrelated to Bank Rate expectations (Joyce et al (2010) and Duffie (1996)). OIS rates therefore provided a cleaner input for inferring the expected policy path. Second, we allowed for a richer factor structure than the commonly used three-factor specification, enabling the model to capture more flexible yield curve dynamics. Third, we incorporated survey information – from the MaPS and Consensus Economics – on near-term Bank Rate expectations directly into the estimation.
The core challenge in any term structure model is to identify real world (physical-measure) expectations – what investors genuinely expect Bank Rate to be – separately from risk-neutral pricing, which also reflects compensation for bearing risk. Box A explains this distinction in more detail. The risk premium links these two objects, but neither is directly observable in isolation. Using yields alone to identify physical-measure dynamics is imprecise in short samples, but adding survey forecasts as noisy observations of expected rates can materially improve identification of the expectations component (Kim and Orphanides (2012) and Guimarães (2014)). We therefore used near-term MaPS and Consensus Economics expectations as additional information to help estimate physical-measure dynamics.
Note that the decompositions produced by term structure models are subject to significant uncertainty. They might be sensitive to choices about model structure and estimation periods. They are only one input into the Bank’s assessment of financial conditions and Bank Rate expectations and are routinely considered alongside a number of other sources of information and analysis by the Monetary Policy Committee.
Our term structure model pointed to a large increase in risk premia at the short end of the UK curve following the outbreak of the war in Iran
Chart 2 shows the model’s estimated risk premia for the UK spot OIS curve at the six-month, 18-month and three-year maturities. Consistent with the common prior that near-term policy uncertainty is generally low, six-month premia are typically close to zero. The sharp increase at this maturity in February 2026 was therefore a notable departure from recent historical norms before 2022. Over the previous decade, estimated premia at this horizon had only been higher during the very high inflation period of 2022 and 2023.
Chart 2: Model-implied risk premia at selected OIS maturities
Footnotes
- Notes: Estimated risk premia (basis points) at six-month, 18-month and three-year OIS maturities.
- Sources: Bloomberg Finance L.P. and Bank calculations.
What did the model tell us about Bank Rate expectations?
After cleaning the curve from the risk premium component, the model provides an estimate of the expected path for Bank Rate. Chart 3 compares the forward curve, the MaPS modal Bank Rate expectation, the model-implied expected path for Bank Rate and the model-implied risk premium. Although the forward curve was upward sloping following the Iran war, the model-implied expectation for Bank Rate was broadly flat over the following year and declined thereafter.
The upward slope in the forward curve was therefore driven mainly by risk premia rather than by a central expectation that Bank Rate would be raised during the year. This conclusion was reinforced by MaPS, where the median profile implied a central expectation for Bank Rate to remain flat for a year before being cut. The model and survey evidence were therefore closely aligned, even though the model was not calibrated to match the MaPS modal path exactly. While noting some model and estimation uncertainty around point estimates, the evidence overall suggested that the forward curve contained a sizeable short-end risk premium, which meant financial conditions were effectively tighter than the expected Bank Rate path would imply on its own.
Chart 3: UK OIS forward curve, MaPS Bank Rate expectations and model-implied Bank Rate path
Footnotes
- Notes: Chart shows the UK OIS one-month forward rate (solid dark line), the MaPS modal Bank Rate expectation (markers), and the model-implied physical-measure expected path for Bank Rate (solid coloured line) as of the latest policy round.
- Sources: Bank of England, Bloomberg Finance L.P., Consensus Economics and Bank calculations.
Note that technically market prices and surveys summarise different objects: prices are tied to the full risk-neutral distribution, while MaPS responses summarise what market participants see as the most likely outcome,footnote [3] which is best interpreted as the mode of the P distribution.
Why would there be risk premia at the short end of the curve?
Risk premia at the short end of the curve reflect the compensation investors require for uncertainty around the policy rate path. The uncertainty around the policy rate path is likely to reflect, in large part, uncertainty about the outlook for the economy. In February 2026, uncertainty about the economic outlook rose materially as investors dealt with uncertainty around the scale and duration of the war and assessed its potential macroeconomic effects, for example via a large initial increase in global energy prices. There may also have been uncertainty around how central bank policy rates would respond to the outlook.
As well as uncertainty, risk premia can reflect changes in the skew or tail risks around the expected policy path. For example, if investors place greater weight on upside risks to Bank Rate, the mean of the distribution would rise even if the most likely outcome changes little. Risk pricing can amplify this, shifting the mean of the market-implied distribution further out. These upside risks may have been present following the outbreak of the war. Box A – a technical annex – explains these distinctions in more detail.
Conclusion
The UK forward curve is a rich source of information about the expected path of Bank Rate, but it needs to be interpreted carefully because risk premia can obscure the signal it conveys. The outbreak of war in Iran in February 2026 presents a case study. A term structure model designed to measure near-term risk premia – anchored with survey data on Bank Rate expectations – suggested that market participants did not expect Bank Rate to rise in their central case even though the curve began to slope materially upwards. Instead, the upward slope in the short end of the curve primarily reflected unusually large risk premia during a period of heightened uncertainty, as well as an associated upside balance of risks to the rates outlook.
Box A: Technical annex explaining distinctions in more detail
Physical (P) and risk-neutral (Q) distributions and upside skew in expectations.
Market prices should not be read mechanically as expectations, and they summarise different objects than surveys. Under the physical measure (P), probabilities represent investors’ real-world beliefs about future outcomes, so the P distribution reflects investors’ real-world expectations of Bank Rate. P is not directly traded and is therefore not directly observable.
The risk-neutral measure (Q) is a reweighted probability measure, transformed by risk aversion. Future states that investors particularly dislike receive greater weight, and risk-neutral probabilities embed both beliefs and compensation for bearing risk. The Q distribution is the resulting ‘risk-adjusted’ distribution used to price assets. In this sense, OIS forwards, caps/floors, and swaptions are pricing moments of the same Q distribution. Option prices can therefore be used to recover the option-implied density of future rates (Breeden and Litzenberger (1978) and Bahra (1997)).
Chart A provides an illustrative schematic depiction of how the P and Q distributions typically relate to each other, including differences in central mass, skewness and tails.
Chart A: Illustrative chart of true (P) and market-implied (Q) distributions for Bank Rate, when expectations are skewed to the upside
Footnotes
- Source: Bank of England.
Risk premia are the first-moment component of the P – Q wedge, the wedge between the mean of the two distributions: EQ[it+h] = EP[it+h] + λt,h. But, OIS rates also reflect higher moments of the Q distribution. Differences in variance, skewness and tail mass can move option prices and the OIS-implied path even when survey modal expectations are stable.
While the P distribution is not traded, probabilistic surveys, including MaPS, can be practical proxies for this distribution (Wright (2017)). Chart B shows an empirical counterpart to Chart A, about Bank Rate expectations at the end of December 2026. We use the most recent MaPS survey distribution for Bank Rate expectations under P. These are the aqua bars. The Q distribution is constructed using option prices on SONIA futures
Chart B shows that the upside skew observed in investors real-world expectations is amplified in the observed empirical Q distribution through trading in financial markets. The orange distribution displays a larger positive skew, as the mean of the distribution is further from the mode than in the aqua distribution P.
Chart B: Empirical distributions: June MaPS and SONIA future option implied PDF on 5 June
Footnotes
- Notes: MaPS survey question asks participants to indicate the percentage probability that they attach to Bank Rate being at the indicated (discrete) levels after the December 2026 meeting. Buckets at the ends are open ended (<2.5% and >5.00% respectively), we use 2.25 and 5.25 to back out the probability weighted means. Solid aqua line connects the individual bins for illustrative purpose, but it does not represent the true underlying distribution. SONIA PDF is the fitted probability density function constructed from December SONIA future option prices on 5 June.
- Sources: Bank of England, Bloomberg Finance L.P. and Bank calculations.
Our Gaussian model captures the mean component of the P – Q wedge well, but it is limited in representing pronounced asymmetry. As a result, in periods with strong skew, some distributional effects can be absorbed into the estimated risk premium. This does not overturn the core result of the decomposition, but it is important when interpreting the level of short-end premia.
The key practical implication for the March 2026 case study is that the steep forward curve reflected a combination of three factors: conventional risk premia, distributional differences between physical and risk-neutral expectations (including skew and tails), and any residual measurement error in the survey itself. Disentangling these contributions precisely would require moving beyond a Gaussian framework, which we leave for future work. But the central finding is robust to this ambiguity: once premia and distributional effects are taken into account, the central expectation for Bank Rate during the initial months after the Iran war was considerably more subdued than the forward curve suggested at face value.
Tobias Adrian, Richard K. Crump and Emanuel Moench (2013), Pricing the term structure with linear regressions. Journal of Financial Economics, 110(1), pages 110–38.
Bhupinder Bahra (1997), Implied risk-neutral probability density functions from option prices: theory and application. Bank of England Staff Working Paper No. 66.
Douglas T. Breeden and Robert H. Litzenberger (1978), Prices of state-contingent claims implicit in option prices. Journal of Business, 51(4), pages 621–51.
Benjamin H. Cohen and Peter Hördahl and Dora Xia (2018), Term Premia: Models and Some Stylised Facts. BIS Quarterly Review, September.
Gregory R. Duffee (2002), Term premia and interest rate forecasts in affine models. Journal of Finance, 57(1), pages 405–43.
Darrell Duffie (1996), Special repo rates. Journal of Finance, 51(2), pages 493–526.
Rodrigo Guimarães (2014), Expectations, risk premia and information spanning in dynamic term structure model estimation. Bank of England Staff Working Paper No. 489.
Scott Joslin, Kenneth J. Singleton and Haoxiang Zhu (2011), A new perspective on Gaussian dynamic term structure models. Review of Financial Studies, 24(3), pages 926–70.
Michael Joyce, Peter Lildholdt and Steffen Sorensen (2010), Extracting inflation expectations and inflation risk premia from the term structure: A joint model of the UK nominal and real yield curves. Journal of Banking & Finance, 34(2), pages 281–94.
Don H. Kim and Athanasios Orphanides (2012), Term structure estimation with survey data on interest rate forecasts. Journal of Financial and Quantitative Analysis, 47(1), pages 241–72.
Shesheryar Malik and Andrew Meldrum (2014), Evaluating the robustness of UK term structure decompositions using linear regression methods. Bank of England Staff Working Paper No. 518.
Monika Piazzesi and Eric T. Swanson (2008), Futures prices as risk-adjusted forecasts of monetary policy. Journal of Monetary Economics, 55(4), pages 677–91.
Alan M. Taylor (2026), Central reservations. Speech, Bank of England, June.
Jonathan H. Wright (2017), Forward-Looking Estimates of Interest-Rate Distributions. Annual Review of Financial Economics, 9, pages 333–51.
Bank Insights articles do not necessarily represent the views of the Bank of England’s policy committee members.
Guimarães (2014) or Malik and Meldrum (2014), or Cohen et al (2018) for a survey.