Central reservations − speech by Alan Taylor

Introduction

Good afternoon to everyone, and my congratulations to CEPR and Barclays for yet another successful iteration of the Monetary Policy Forum here in London.

It is certainly a great honour to be asked to speak to you today and my thanks to Beatrice Weder di Mauro, and to Silvia Ardagna and Jack Meaning, and their teams, for this kind invitation and all the careful planning and arrangements.

Now, many of us here work in or near Threadneedle Street. Like me, you might take the steps down to the tube outside the Bank and, if your luck is down, be confronted with the announcement that “the Central line is partially suspended”. But I never really thought we would be making the same kind of announcement inside the Bank.

Of course, the keen followers of the Bank of England, of which there are many (if not most) in this room, will have noticed that the latest April 2026 Monetary Policy Report did not contain anything that could be described as a central forecast. Instead, it contained only ‘scenarios’ (Chart 1) – forecasts that are conditioned on a variety of non-standard judgements, but which excluded the usual central projections that are produced by Bank staff and that are always conditioned on market curves (for rates and energy) and a standard set of modelling assumptions.

Perhaps for me this is fortuitous timing. I had been planning a speech about the role of the central forecast, and its place in the past, present and future life of the MPC. But I was worried that it would be of limited interest. Usually, the thought of listening to a central banker talk about forecasting might seem like a dull prospect. But now, I said to myself, perhaps recent events might make it a hotter topic – and maybe I can, just about, detect a rising frisson of excitement out there in the audience?

So, to state what might seem obvious, to not publish a central forecast is an unusual decision for the Bank of England. And it generated significant attention and reaction at the time (for the most part positive). In fact, the last and only time this has happened in recent memory was in May 2020, the first forecast round after the outbreak of the Covid pandemic, where the central projection (and fan chart) was replaced by just a single ‘illustrative scenario’.^{footnote [1]} That same report stated that “[w]hile the scenario is highly conditional, it helps to illustrate the nature and potential scale of the impact of the pandemic on the economy”.

Now, strictly speaking, this type of characterisation applies to any conditional forecast. Conditional forecasts, like the MPC’s traditional central forecast, are always ‘what-if’ exercises, and if the conditioning paths do not come to pass, then the forecasts will justifiably not match observed reality after the fact.

A critically important example of this ‘what-if’ character can be seen in the Bank’s recent inaugural Forecast Evaluation Report which found that, while forecast errors were indeed historically large over the post-Covid period, much of the peak error (Chart 2) can be explained by ex-post changes in the conditioning path for energy prices following the Russian invasion of Ukraine and by other global factors (Forecast Evaluation Report – January 2026 | Bank of England).^{footnote [2]} Similar findings have been uncovered by other central banks.

Thus, using its crystal ball, had the Bank been able to know before the invasion what the ex-post actual outcomes for oil and gas prices and global factors would be, and had ex-ante fed them into the forecast machinery, then it would have successfully predicted much of the downstream effect on aggregate inflation in 2022-24. But the Bank, like the market, and like everyone else, did not know (and has no crystal ball either).^{footnote [3]}

It’s a strange definition of failure: as a professor, I know my students would not be very happy if I graded this week’s homework using last week’s answer key. But I digress.

My actual point is that, while the previous reasoning might seem almost tautological, it underscores the idea that all conditional forecasts have something of a scenario-like flavour, even in the best of times: they are always ‘what-ifs’ drawn from some part of the unconditional outcome probability distribution. It’s just that with a central forecast we are paring it down to some concept that conveys the centrally weighted mass of that distribution.

However, all that said, back at the time of the May 2020 report, the unusual illustrative scenario that was presented was yet still ‘central’ in one obvious way – in that it was singular, and was conditional on the MPC’s collective best guess of how the economy might evolve given the noisy information available at the time. This contrasts with the scenarios that you saw in our latest April 2026 report. Here again, radical uncertainty prevailed once more, but none of the published paths (A, B, C) were produced in a way consistent with what either the staff, or the MPC as a whole – or both – might judge their traditional ‘best guess’ of how the economy might evolve given available information.

So perhaps that unusual outcome is as good a prompt as any for me to embark on some reflection and discussion on the central forecast – its rise and fall, the triumphs and tragedies, and the important role it still has to play.

In the end, we get will back to the granular details of the April scenarios, a discussion of the outlook, and along the way I will discuss what a traditional baseline path would have looked like at the time and what we can learn from that.

But first let me step back a bit and talk about the evolutionary path of the MPC’s central forecast, past, present and future. I offer some thoughts, very much my own, on the pros and cons of this crucial piece of our analytical machinery and policy toolkit – as it was and as it might be.

The central forecast in the past

The Bank of England’s central forecast has played a defining role, as a key device or vehicle in the delivery of the MPC’s policy and communication strategy. But, like any vehicle, our forecast machine also has its functional limits, or performance envelope.

In isolation, on the test-bench, its analytical machinery performs its precise and
well-defined function. But out of the shop, the road testing has not always been perfectly smooth. There have sometimes been conceptual tensions, mechanical quirks, and unintended messages, which, all told, provide lessons about what a forecast can – and cannot—do. Looking back at how the central projection was constructed and used over time reveals both its enduring value and its sporadic pitfalls. It may also help, in part, explain the nascent desire to de-emphasise the central forecast.

Most notably, the central forecast has always sat uneasily between being an input and an output of the policy process. Conceptually, it was meant to represent the MPC’s best collective judgment about the most likely path for inflation and activity, among other things, conditional on a particular assumption about interest rates (and other asset prices as well as other conditioning paths). But such assumptions, to condition on the market curve (or, in olden days, on a constant rate path), created a structural ambiguity.

If the forecast was an input only, the rate path should have been endogenous; if it was an output, the MPC should have been comfortable with whatever the market curve implied. In practice, the Bank often found itself trying to straddle both positions, an exercise in mental gymnastics that created recurring problems.^{footnote [4]}

The audience was left to iterate back and forth between where the curve sent the economy away from target, and how the policy path might therefore need to be different from the curve. This was a convoluted fixed-point problem at the best of times, and led to an awkward dynamic where, by backward induction, the MPC often found itself exercising de-facto forward guidance based on its published inflation forecast at year 2 to 3. This dynamic implied – or indeed, required – at times a quite material ‘judgemental’ overlay to the forecast when inflation conditional on the market-implied path ventured too far away from target, so as to signal appropriate policy guidance.

Still, in purely mechanical terms, the construction of the forecast had strengths: it was at least an attempt, against sometimes significant odds, at a systematic, model‑based projection that tried to be somewhat transparent in its broad architecture. Indeed, as originally conceived, the combination of a core DSGE model with a suite of auxiliary models, set in a fairly simple user interface as a way to produce forecasts (Burgess et al., 2013), was both innovative and elegant. The forecast’s weaknesses were not primarily in the technical underpinnings it was born with, but rather in the complex institutional framework it had to survive in.

These weaknesses tended to be most exposed when the curve was dislocated, given the choice of conditioning. And, as I argued in my speech at Sintra last year, we should not be surprised that this keeps on happening (Taylor, 2025). In the last 20 years, at least, the market curve (in the UK and the US) has actually been a very poor predictor of future interest rates (Chart 4). And better model-based alternatives do exist (Taylor, Brandt and Dotta, 2025), an approach taken by other central banks – which warrants consideration, a topic to which I will return.

Though this problem has recurred for many years, the example that inevitably sticks in my mind is the recent path of the UK economy since 2024, when I joined the MPC. Since then, a ‘swoosh’ in the market curve has become, in some circles, a noteworthy phenomenon: that is to say, the tendency for the ‘risk-free’ market curve to imply a medium‑term rise in rates, sometimes to levels well above any plausible definition of the nominal natural short rate.

For example, COMPASS, the main DSGE model used by Bank staff for medium-term analysis, now embeds the staff’s assessment of a central estimate of the UK nominal neutral rate of 3% (see Annex 1 of the November 2025 Monetary Policy Report). This level equates with my own central estimate of neutral, and that of many other models, which I have by now referred to several times (e.g., Taylor, 2025).

Inevitably, then, when the MPC conditions on a market curve distorted by a wedge between the OIS curve and market expectations – a wedge often attributed to risk premia – the forecast inherits those distortions. You can see this wedge in the difference between the market curve and median expectations for the ‘most likely’ Bank Rate path in the Bank’s Market Participants’ Survey (Chart 5). These expectations are consistent with a new term structure model recently developed by Bank staff, as shown in the aqua bars, which reveals a large risk premium component in the OIS curve and a close alignment with MaPS on the expected rate component. Thus, in recent times as well as in general, if we see an elevation of the OIS curve due to risk premia, then this represents a tightening of financial conditions. Now, for sure, this is important to include in any good forecast, which should condition on all asset prices; but it is not at all an unpolluted prediction of future Bank Rate, and therefore should not be presented or interpreted as such.

This risk premium disconnect in the OIS curve has, in turn, created a subtle but at times powerful distortion in the wider forecast structure since our forecast approach has traditionally internalised changes in the market curve regardless of whether they are driven by risk premia or expectations for Bank Rate. In recent years, that has meant that the forecast was conditioned on (what is often interpreted or described as) an ‘expected path for Bank Rate’ which was structurally tighter than the actual true underlying expectation. Reiterating: I would describe that as appropriate for conditioning, but distorted as an interpretation of Bank Rate.

As a consequence, in the last year or so, the implied stance of monetary policy in the forecast has been consistently and, in my view excessively, restrictive – materially dragging on inflation in the 2^nd and 3^rd years of the forecast (Chart 6). As such, when the MPR forecasts are rationalised through the lens of the COMPASS model, it implies the presence of positive demand shocks for the duration of the forecast, in order to justify inflation ending up at or close to – rather than below – 2 percent by the 2-3-year mark under such restrictive rates (Chart 6). This has been our lot in recent times.

Arguably, these implied shocks, necessary to rationalise the forecast, can be somewhat occluded, sitting there not always as genuine explicit model-based judgments coming to the fore, but as deeper artefacts of the interaction between the conditioning path and subjective judgements.

One source of demand shocks that failed to appear was an assumed rebound in consumption and investment as the economy grew and rates normalised. But this did not happen, and savings rates have stayed higher than predicted, possibly because, in the uncertain economy post-Covid, post-tariff, and maybe now post-Iran conflict, households and firms have had other ideas.

The result (Chart 7) has been a persistent one-sided bias in our forecast errors from February 2025 up to February 2026. As I noted in recent speeches in Oslo and New York, we cumulatively revised our inflation path down and our slack measures up (higher unemployment, or a more negative output gap). Relative to the central forecast, outturns were coming in cumulatively wrong to the downside – not necessarily because the MPC misread the economy, but, in part, because the framework and conditioning assumptions we had employed boxed in our analysis and forced the model to generate shocks that never materialised.^{footnote [5]}

Chart 7: Bank of England forecast updates (vintages 2024-25)

Percent

• Sources: The chart shows the cumulative forecast errors from Bank of England central forecasts (2024-2025) for CPI inflation (top left), unemployment rate (top right), private sector pay growth (bottom left), and the savings rate (bottom right). 2024 forecasts are depicted in grey, while 2025 forecasts are depicted in green (February), yellow (May), purple (August), and orange (November).

To me, the lesson is straightforward: as I have argued before the TSC and elsewhere, a central bank should generally not be in the business of implicitly having to ‘predict’ shocks, in order to rationalise its forecast. In fact, conceived correctly, such a thing is impossible. If the only way to reconcile the conditioning path with the inflation target is to invent future disturbances, explicitly or implicitly via conditioning assumptions, the framework – not the economy is – the problem.

I think this experience shows one danger of outsourcing our conditioning assumptions when market curves are noisy and volatile, and when risk premia may be distorting them.

This is especially the case in a framework of marginal forecast updating. At the Bank, every forecast has been built on top of the previous one, by layering on ‘news’, as these innovations are called, that is, inter-round data releases and changes in conditioning assumptions. In that world, the cumulative artefacts of these distortions may start to quietly dominate the forecast, making it harder to convey policymakers’ judgements.

One inevitability, however good one’s forecast is, and ours may become, is that ‘errors’ will be made. Perfectly predicting the future is impossible. And, in the face of that impossibility, one aspect of the past framework that deserves praise in my view – despite often being maligned as vague or uninformative – is the fan chart.

At the very least, the fan chart could acknowledge uncertainty, could display the asymmetry of risks, and could remind everyone that point forecasts are fragile – as recent evidence suggests (McMahon, Naylor, Rholes, Rickards, 2026). The fan chart did not pretend to know the future; it showed the range of plausible outcomes. Its reputation suffered mainly not because the construct was inherently flawed, in my view; but because its ingredients – the central forecast and skew – were sometimes mis‑specified, and its implementation – conflating statistical probability distributions based on past performance with ad hoc judgments, as well as the absence of a complementary tool for tail event discussion – leaving room for improvement (Aikman et al., forthcoming).

While the implementation can be fine-tuned, the mis-specification component needs to be accepted, more generally, as an occupational hazard. How bad is ‘bad’ forecasting? Shocks happen. Monetary policy operates in a world of structural breaks, geopolitical surprises, and behavioural shifts. No forecast can eliminate that uncertainty, especially in times like these.

For me, the real issue is whether the forecast framework amplifies errors or dampens them. The reliance on conditioning paths that embedded excessive noise and implied convenient future adjustments meant that errors accumulated rather than cancelled out. When the economy deviated from the assumed path – as it inevitably did – the central forecast was left exposed.

So, looking back over three decades, I think the central forecast has been valuable as a communication tool and as a disciplined way of – among a range of other inputs and considerations – organising the MPC’s thinking. But in practical use, some of its weaknesses were structural, not incidental. The tension between input and output, the mechanical influence of a poorly-predicting and sometimes-distorted market curve, the implied artificial shocks often required to hit the target, and the surrounding uncertainty – all hinted at vulnerabilities.

The experience of the past decade suggests to me a simple principle: a central forecast should reflect more weight on genuine judgments informed by a broad ensemble of clean and simple models, as well as MPC expertise, and less weight on mechanical conditioning assumptions. In other words, it should perhaps look more like a scenario.

The central forecast today

Where does that leave us now? The central forecast may appear to be less visible today – increasingly de‑emphasised in speeches and other communications, pushed backstage while scenario analysis takes the spotlight, and stripped of some of its old visual prominence, to the point that sometimes it might even appear to be invisible.

But it has not and should not disappear, far from it. As I will argue, it is bound to sit at the gravitational centre of the entire forecasting and communication framework. Understanding its pervasive force is essential: seeing what scenarios can and cannot do, and why the central projection still matters, is essential for making sense of the Bank’s current approach.

Scenarios have increasingly moved to stake their place as a key tool of modern monetary‑policy communication. They allow policymakers to explore alternative states of the world, highlight risks, and show how different shocks would affect inflation, activity, and how policymakers’ reaction functions map those in different states of the world. They are flexible, intuitive, and politically attractive because they avoid the false precision of a single path.

But like other specialized pieces of machinery, scenarios too have limits. They cannot provide a baseline probability distribution; they cannot anchor expectations; and they cannot substitute for a coherent view of the most likely trajectory of the economy. A scenario is, by definition, also a conditional narrative – but conditional on something that is not strongly expected to materialise or is less likely than the baseline outcome. The most likely outcome remains, at least weakly, more informative.

In an increasingly uncertain world, a key analytical challenge will be to ensure we have the creative and technical capability to identify and model the set of (policy-relevant) scenarios to inform our internal deliberations and decision-making. To this end, Bank staff have taken great strides in the past year as part of the Monetary Policy Transformation (MPT) project.

With the proliferation of scenarios, however, as well as divergent views in our new paragraphs on both their materiality and likelihood, and preferred strategies for managing the set of risks they encompass, the collective aspect of our messaging also becomes more challenging. I think we are all still adjusting and calibrating our way in this new world.

This, I would argue, is why the central forecast and its probability distribution – as represented by the fan – remains, deep down, invaluable and unavoidable.

As I often put it, even if one takes out the fan chart, the fan is still there.

The uncertainty around the baseline projection does not vanish simply because it is no longer drawn on a chart. The distribution of risks still exists in the model, and in the MPC’s judgment – and should still remain in the public’s expectations. Removing the visual does not remove the underlying structure, and it may even convey a misguided signal of precision that risks credibility when forecast errors, even those within normal confidence intervals, subsequently materialize, as they inevitably will.

Similarly: even if one takes out the central forecast, it is still there.

Why? Because mechanically, every scenario is implicitly or explicitly a deviation or perturbation of the central case – in either conditioning-data-space or in model-space.^{footnote [6]} One cannot construct a coherent upside or downside scenario, nor the accompanying narrative, without first knowing what you are deviating or perturbing from, and in which directions and by what amounts.

This leads to a key point that there is no tension between the old role of the central forecast and the new role of scenarios. The central projection is the baseline, the reference point, the anchor, from which everything else is derived. If that baseline is wrong, everything built on top of or around it is wrong. A mis‑specified central forecast contaminates some or all parts of every scenario, and therefore every risk assessment, and every policy narrative.

So for me, as we move forward, de‑emphasising the central forecast should not be confused with ignoring it.

I believe this should not, and I trust will not, happen, as staff continue their work to improve how it is built, while the MPC continues to improve how it is used. Taken together, for me, the shift toward scenario‑heavy communication does not decrease but rather increases the importance of investing heavily in improving the central forecast – even if, on occasion, that central forecast is not much talked about or even taken out.

And as I noted in the previous section, recent experience shows how problematic it can be when the baseline is neglected in a way that leaves it systematically biased. When the central projection repeatedly over- or underestimates inflation pressures, the entire communication strategy can be swayed as other the lines may also move off kilter, not just the central line. Scenarios may thus also become skewed, risk assessments become asymmetric, and – as the process unwinds – policy explanations can become reactive rather than anticipatory.

There are a number of aspects in which the central forecast could be improved, in turn helping to improve the derivative scenarios too. And indeed Bank staff are working very hard and impressively, as I speak, to do just that. I hope, and am optimistic, that the result will be a better central projection that reflects genuine technical skill and judgment rather than artefacts of past decisions and outsourced conditioning choices, and I am convinced that it has to remain a core analytical output.

In sum, seen properly, scenarios are a complement not a substitute for the central forecast.

They enrich the narrative, illuminate risks, and help the MPC communicate, especially in times of great uncertainty. But they cannot carry the weight of the entire forecasting system.

Without a sound central path based on given conditioning data (in whatever model or suite of models) from which to build off, the perturbed scenarios may also get unmoored. Our challenge is to integrate scenarios into a coherent framework where the central projection still provides the main pathway and the scenarios chart possible turns along a wider road.

Why does this matter? Because monetary policy is ultimately about expectations. Households, firms, and investors – if not right now, but in most states of the world – still need a sense of the most likely path of inflation and rates. They need a baseline against which to interpret news. If central banks don’t articulate a central view, the risk is that others will fill the vacuum – markets, commentators, or other policymakers.

For a central bank trying to convey its reaction function, collective communication depends on the ability to articulate a coherent central narrative, even if that narrative is shrouded in uncertainty and policymaker disagreement.

The central forecast in the future

The central forecast today may be becoming less visible, but in my view it is more important than ever. It now shapes the scenarios, even as it anchors expectations, and it still structures the MPC’s internal debate.

The Bank may choose not to publish it prominently – or even at all – but we cannot operate without it. The real challenge will be to continue our efforts to keep on improving the forecast – to make it more transparent and robust, less volatile and outsourced.

Where might that take us? Let me summarise a few points of personal opinion, including highlighting important areas where Bank staff have already made impressive progress.

First, the central forecast still functions as something between an input and output of the policy process, even if the MPC has moved to reduce its prominence. It is not in the full sense a ‘staff forecast’ – even though it has moved in that direction. More responsibility has shifted toward staff, which I welcome. However, a clean separation of staff responsibility and policymaker involvement is still some way off, so it seems that it will remain a balancing act for now.

Second, a priority is improving the models that underpin the central forecast itself, a complex structure that is still under renewal. This is a dimension that staff have invested a significant amount of work in, as documented in the novel Macro Technical Paper series. Crucially, the approach has moved away from previous incremental adjustments on top of legacy assumptions, which had tided us over for many years, to one that starts from first principles. I very much welcome this work, and encourage it to continue. As the work progresses, I would in particular encourage development of a new version of COMPASS with a better model of the labour market and also development of a new semi-structural model that will refresh the core toolkit, even as we continue to incorporate insights from a wider ensemble of add-on models, cross-checks, and sensitivity analyses.

Third, we can continue to develop the process of scenario production. This is another area where work is underway as part of MPT. As part of that, I believe we can look at the ECB as a useful model here. It maintains a scenario library under continuous review, allowing policymakers to activate alternative cases quickly when new risks crystallise. This avoids the scramble to design scenarios at the last minute and ensures that risk analysis is systematic rather than reactive.

Fourth, the Bank could also be more flexible in the number of scenarios it uses. The traditional trio – central, upside, downside – might be too limiting when uncertainty is high or skewed; and then too bracket-y when normality resumes. Sometimes, as my MPC predecessor Charles Goodhart has pointed out, an odd number of scenarios can be unfortunate if, wrongly, people automatically tend to think the middle one is central. We could also support scenarios with probabilistic guidance that helps the public understand how risks are distributed across the bins. This recognises that uncertainty is, at times like now, clearly not always uniform or symmetric and that different magnitudes of risks can coexist.

Fifth, another area where staff have impressively expanded our analytical capabilities is in the production of monetary policy rules, including the optimal policy projections (‘OPP’). However, their use as both inputs and outputs requires care. In my view, these can be useful for tentative validation or sense-checking, but they can also generate unstable or misleading outputs when the underlying data are noisy or when structural relationships shift. And as the number of rules proliferate their massages can start to blur. They should inform judgement, not replace it, and need judicious pruning: backward-looking and contemporaneous rules sit poorly next to our forward-looking mandate.

Sixth, I believe we may need to rethink our conditioning on the market rate curve. As I have often argued, this has not been a stable and reliable basis for the forecast, and it also can be distorted by risk premia. But it is also obviously not the only option. Other central banks have shown different ways of handling the problem. The Riksbank publishes its own model-based policy rate path, produced by staff, with a range of outcomes, making explicit the conditional trajectory behind its forecast. The Federal Reserve goes one step further in communicating policymaker’s own individual ‘preferred’ path for rates, using the dot plot. These approaches acknowledge uncertainty but avoid outsourcing the entire conditioning assumption to the swaps market.

Now it may be objected that we, the MPC members, can’t agree on a path, so why go there? But that is also doubtless true of the Fed and the Riksbank, and yet it doesn’t stop them in their ways. Contrariwise, we each have our various disagreements even about the OIS market curve on a recurring basis, and yet we have been able to move beyond those dispersed views and use it anyway, with suitable caveats from time to time, for its specific purpose in our existing forecast process.

For me, the advantage of using a more stable, model-driven path, such as in the Riksbank approach, is twofold. It ensures that we are not adding unwanted noise, time-varying risk premia, or volatile market conditions into the medium-term forecast process; and reciprocally this may in turn temper that noise by better anchoring the curve. And it frees us from the mental gymnastics and circular communications games of yore, whose effectiveness and clarity have been rightly cast in doubt on many occasions after decisions land askew. Can this be done without our own manufactured curve being seen as forward guidance? Of course – that is the task of good communications. And it is not as if every time, in the gymnastics game, when we did hit the target at year 2 or 3 of the forecast, it has managed to avoid being treated as curve validation, i.e., forward guidance, anyway.

Taking stock here, let’s see where we are after so much evolution and progress in the last 18 months. Here, it is worth recognising the difference between the highly visible communications changes and the slower, less glamorous work of rebuilding the analytical machinery. But after the Bernanke review, we have to complete both.

The most visible stuff is widely talked about and may seem to some like low-hanging fruit – adjusting charts, streamlining policy statements, adding individual paragraphs, or introducing presentation of scenarios – but this is only a small part of the progress that’s needed, and that Ben Bernanke indeed advised us to undertake.

The fullest gains will come after we finish tackling the harder, behind‑the‑scenes tasks of intellectual and conceptual refresh: rethinking conditioning assumptions, getting a better grip on neutral, refining model structure, improving data inputs, re‑estimating behavioural relationships, and testing a new ensemble of alternative specifications. These changes, that staff are working hard to complete, do take time and, importantly, they require sustained investment in skills and technology. They rarely attract public attention, but they are the low‑visibility reforms that ultimately determine whether the forecasting system can become more reliable and the policy process more robust.

Overall, my unifying theme is that scenarios can only build from, and not replace, a strong central forecast. Internal work on improving the central projection has already taken great strides and is essential to bolster the entire framework.

Conclusion: where are we now?

Let me now conclude with the current outlook, while presenting my own views, as seen through the lens of the April 2026 MPR forecast, and using a few illustrative charts.

Here is the same chart we saw at the start, but now in Chart 8 with what I think of as the ‘traditional’ baseline path shown in purple and labelled ‘Scenario T’ (T for traditional). This scenario encompasses standard conditioning on rate and energy curves (based on market futures), and standard judgements about the size of second-round effects that are consistent with those in recent central forecasts, as we have normally presented them.

I think what this makes clear is the tremendous skew in the nature of the risks we face from the ongoing energy shock. Scenarios like A, B, and also, in between them, T, sit near the middle of the probability distribution and are reasonably central, at least based on information at the end of April. Scenario C captures a more extreme scenario, in the right tail of energy price dynamics, calibrated by Bank staff at roughly the 95th percentile of outcomes as judged by options pricing in energy and gas curves at that time.

And future energy price uncertainty is not the only dynamic we have to wrestle with. The standard policy prescription calls for us to look through temporary energy shocks, and to only respond forcefully if there is a risk of a shift in inflationary dynamics or de-anchoring, in the medium term. This depends on judgement, and on how we calibrate the somewhat ill-defined concept of so-called ‘second-round effects’ which I would understand as excessive medium-term responses of non-energy prices and wages, which could become self-reinforcing. As shown in Chart 9, from my MPC colleague Megan Greene’s recent speech, layering on a stronger degree of wage and price persistence (proportional to that calibrated in Scenario C) to the moderately adverse energy shock in Scenario B, we would end up with a more worrisome trajectory, dubbed ‘Scenario M’.

Now, just in case you have not seen enough scenarios already, here is another exercise. We have seen the three main scenarios A, B, C in which the two key dimensions of the calibration are changed at once; and we have also layered in a couple of extra alternatives, T and M (Table A). That gets us up to five.

Why go off-diagonal? In the absence of a published baseline, the ABC journey which stays on-diagonal also makes it difficult to disentangle the marginal contribution of each effect: the variation in the magnitude and persistence in the energy shock (low, medium, high) along one dimension in the rows, and the variation in the degree of wage and price persistence on the other dimension (low, medium, high) in the columns.

For greater transparency, if that’s the right word, and albeit perhaps at the cost of simplicity, we might need to plot all of the nice cells in the 3-by-3 matrix. And here we see that result, in Chart 10, showing all six off-diagonal trajectories as grey solid lines around the three published scenarios A, B, and C.^{footnote [7]}

What do I take away from these exercises? I will emphasise two features.

First, the size of energy shock drives much of what is happening to a first degree of approximation, at least over the first year of the forecast, and perhaps everything across the full forecast if it happens to be the trigger that then determines how wage and price dynamics endogenously shift as well. The diagonal may not be a choice but an inevitability.

Second, even if those wage and price dynamics do shift, we will not see very much evidence of it in the first 12 months. Looking at the complex off-diagonal chart, one thing that is clear (maybe the only thing) is that there is not much variation in the off-diagonal versus on-diagonal paths in the short run, but more so in years 2 and 3.

So, with that said about those longer-term trajectories, let me turn to the short-run trajectory, where less uncertainty lives. And I will focus on the short-term inflation forecast, or STIF, as we call it at the Bank (Chart 11).

Here, the April STIF shows where we are now, and where we expect to be in the next six months, for inflation, and it provides a way to decompose the shift in the inflation trajectory relative to the prior February 2026 forecast. That February forecast, as we all now painfully recall, had inflation hitting the 2% target in April and remain there or thereabouts over the rest of the horizon, alongside a modest amount of slack.

But that February 2026 forecast did not see the energy shock coming. And now that trajectory is gone. But was it wrong? This is a good time to do a forecast error decomposition, or more precisely, a forecast revision decomposition. If we look at the bars here on the right, we can see that the April CPI inflation was 2.8% actual versus 2.0% in the February forecast. What does the accounting say? It is almost fully explained by direct energy (like petrol directly consumed) and a little bit by indirect energy (input costs in other sectors); and over the rest of the year this will remain mostly true, even as the indirect piece builds due to slower pass through.

This is therefore a deeply disappointing chart to stare at. We were so close to achieving the desired target outcome, and we know what could have been. Arithmetically, the subsequent dislocation of inflation from that former path is verifiable, mechanical, and exogenous. Alas, we just have to accept that, as it is what it is.

But in closing let me also be clear that our starting point in this shock is also very different. So we must guard against recency bias: the temptation to fight the last war.

Now looks very different to 2022, or as compared to other previous energy shocks in 2011 or even going back to the 1970s (Chart 12). We enter this shock with a very weak economy. In purely domestic terms, there is excess supply or slack in labour market, and a meaningful output gap.

The weak state of the labour market is critical and will act to restrain second round effects through wage-price spirals. Unemployment is rising and wage growth has slowed.

Although sticky as always, wage growth has now converged down to the 3%-3.5% level (Chart 13), after a time when real wages were naturally and unavoidably catching up with a lag to the 2022 inflation shock. And consequently, as is typical, falling services inflation is following wage indicators down, also with a lag. The normalization of wages at these levels is something that was already clear from high-frequency wage settlements data as long as 6 to 9 months ago, as I have pointed out (e.g., Taylor, 2026). And Bank staff see these levels as roughly a target-consistent rate of wage growth, under a productivity growth assumption of 1% per annum, similar to what we last saw in a fairly stable and healthy economy, close to equilibrium potential, in 2018 and 2019.

Of course, we cannot be complacent. Could a new cycle of wage and price inflation be reignited? If the energy price path does not deviate markedly from the current benign market curves, I am doubtful. Our own models at the Bank suggest that the weak state of the economy leaves little room for aggressive wage bargaining by workers, and limited pricing power in output markets for firms.

For me these impressions have been corroborated in discussions with businesses during the four Agency visits I have made this year, to the East Midlands, Wales, the West Midlands, and the South West. In contrast to 2024, when I joined the MPC, wage pressures are not a very prevalent concern right now; instead, what I mostly hear about is the soft labour market, weakness in confidence, stalling consumer demand, export challenges amid rising trade frictions, global uncertainty, and (of course) the new spike in energy costs. These latter issues have all risen to the top of the worry list when it comes to prospective economic conditions for the rest of 2026 and into 2027.

Is that kind of shift in the labour market and wage dynamics material and likely to persist? Or am I just leaning on a small sample of noisy data and anecdotes? As I see it, some important leading indicators from our recent DMP survey results confirm my state-dependent interpretation and outlook (Chart 14). In recent years, firms have reported a close alignment between their price inflation expectations and their wage growth expectations, both for the economy as a whole and for their own firm. When prices moved up fast, so did wages. But in the last six months that correlation has broken down: firms now report that they expect an uptick in inflation in the coming year, but they also say that expect that wages will not follow.

Chart 14: Firms’ expectations of wage and price growth

Wages and CPI expectations (left) and wages and own price expectations (right)

• Source: Wage growth and own price expectations from the Decision Maker Panel (DMP) and CPI from the ONS.

Let me now discuss what I think all this means for rates, both my view and the market view (Chart 15). And of course, given the uncertainty, this can only be a rough approximation that has to cover a very wide range of possibilities.

Developments continue apace. After the MoU, a benign outcome for energy prices by next year is now more likely.

For example, an update to Scenario B, last week, sat close to the updated Scenario A. And looking to the second half of this year, the more favourable energy curves combined with a recent inflation print 0.4 points below expectations, lead us to expect inflation to peak a 3¼ % later this year, ½ point lower than the 3¾% projection in the April MPR.

Still, nothing is guaranteed, uncertainty remains.

One can still imagine that we might end up in a fairly benign scenario, say, close to Scenario T: a result that is to be hoped for under a rational process, where a resolution to the conflict would soon be found. We might then follow something close to the current energy curves.

That is not to say there still wouldn’t be bumps in the road: shipping in the Strait might resume in fits and starts, with risks unclear and ships out of position; oil and gas production in the Gulf might struggle to start until storage starts to clear, which happens only after outbound flows look reliable; shortages of some products could emerge and be highly disruptive for some sectors and regions of the global economy.

If something closer to that benign scenario plays out, we must be ready to act. In my view, interest rates can and should resume their downward path to neutral in Scenario T. The UK and world economies will have been materially damaged in the meantime. The higher energy costs and lingering uncertainty will have lowered incomes, dented confidence, and subdued demand even further.

But the pre-war weak domestic economic backdrop remains. If even more slack then opens up, as in Scenario A, and if inflation looks to undershoot, we may end up having to cut quickly and could even see Bank Rate below neutral for a while.

The caveat is that we do not yet have that clarity. And a substantial and opposing tail risk remains. The geopolitical noise far exceeds the signal, at least for now. Indeed, the presence of this deep source of uncertainty was a key reason for moving away from a central forecast in April. Scenario B is not out of the question, even if Scenario C remains out in the tail of the distribution. We cannot be sure that the conflict will not drag on, and that upward pressure on the path of energy prices will not take us away, possibly even far away, from the current curve. In the worst-case scenario, nothing is off the table, and to defend the nominal anchor the Bank would face tough decisions and even harder trade-offs. But, given our mandate, we must do what we have to do.

Markets understand this, and they report rate path expectations (Chart 16) in our MaPS survey that clearly reflect an understanding of how our reaction function works as they judge where rates would go in different states of the world, under the three main ABC scenarios: reactions that are not out of line with the rate paths I just sketched out. Markets also see no medium-term threat to our inflation target (Chart 17), and see inflation being back to target after this shock has passed.

Chart 16: MaPS expectations for end-26 Bank Rate under different scenarios

Share of respondents

• Source: Bank of England’s Market Participants Survey (MaPS). Dotted lines represent median responses.

So right now, we are hit by yet another exogenous energy shock, but this time we find ourselves in a weak economy and in an already restrictive policy stance, 75 bps above my estimate of neutral, and likewise above where we might soon have been (or what I would have voted for) in the no-war counterfactual rate path.

Until we have greater certainty, then, an extended hold at this level is, to me, very much the correct and appropriately measured policy response we need, given the balance of risks.

To conclude, geopolitics is now driving economic outcomes, and hence monetary policy. It is a very uncomfortable and unwelcome place to be, but the reality we must face. So we must watch both economic and geopolitical developments carefully and be ready to act in either direction, as necessary.

Thank you.

Thanks to Lennart Brandt, Vitor Dotta, and Matthew Naylor for help preparing this speech, and to Harry Austin, Andrew Bailey, Matthew Bird, Lydia Hennings, Bob Hills, Tom Key, Simon Lloyd, Clare Lombardelli, Becky Maule, Louisa Noble, Amar Radia, Martin Seneca, Sumer Singh, Bradley Speigner, Meryem Torun, Gellert Turkevi-Nagy, and Jan Zacek for their comments and help with data and analysis.

References

Aikman, D., R. Bidder, S. Lloyd, G. Mantoan, S. Maso, A. Mori and M. Tong (forthcoming), ‘Forecasting macroeconomic risks in the UK’.

Alati, A., M. Arazi, J. Barrdear, A. Gimber, E. Hughes, L. Laureys, S. Lloyd, O. Ozturk, J. Page, K. Reinold, E. Tong, M. Tong and M. Waldron (2025), ‘Tools for endogenous monetary policy analysis: optimal projections and instrument rules’, Bank of England Macro Technical Paper No. 4.

Albuquerque, D., J. Chan, D. Kanngiesser, D. Latto, S. Lloyd, S. Singh and J. Žáček (2025), ‘Decompositions, forecasts and scenarios from an estimated DSGE model for the UK economy’, Bank of England Macro Technical Paper No. 1.

Brignone, D., S. Goel, S. Lloyd, G. Mantoan, N. Raviraj and A. Renzetti (forthcoming), ‘Making scenarios add up: spanning risks with scenario synthesis’, Bank Insights.

Burgess, S., E. Fernandez-Corugedo, C. Groth, R. Harrison, F. Monti, K. Theodoridis, and M. Waldron (2013), ‘The Bank of England's forecasting platform: COMPASS, MAPS, EASE and the suite of models’, Bank of England Staff Working Paper No. 471.

Taylor, A. (2025), ‘The end of the road’, speech given at the London School of Economics and Political Science, 4 July, Bank of England.

Taylor, A. (2026), ‘Getting the right directions’, speech given at the Monetary Policy Mandate Conference at Norges Bank, Oslo, 2 March, Bank of England.

Taylor, A. M., L. Brandt, and V. Dotta (2025), ‘Does r* predict policy rates?’, ECB Forum on Central Banking.

McMahon, M., M. A. Naylor, R. Rholes, P. Rickards (2026), ‘Anchors aweigh? The effect of communicating forecast uncertainty’, CEPR Discussion Paper (and forthcoming BoE SWP).