Keynote on scenarios and risk
The event opened with Domenico Giannone’s keynote speech, summarising his paper: Scenario Synthesis and Macroeconomic Risk. Against the backdrop of the Bernanke Review, which called for greater use of scenarios in policy formulation and communication, this could hardly be more relevant.
The authors develop a novel methodology bridging two risk-analysis traditions: statistical forecasting of the full distribution of possible outcomes – akin to the fan charts published by the Bank since 1996 – and scenario analysis. Professor Giannone highlighted the core challenge: how to reconcile statistical forecasts with judgmental narrative approaches like scenarios when considering risk and uncertainty in policy settings. The proposed Bayesian methodology provides a practical way to quantify the likelihood of a set of chosen scenarios, while also assessing how well those span the risks implied by a formal statistical forecast distribution.
As an illustration, had this tool been applied to the Fed’s ‘Tealbook’ scenarios ahead of the global financial crisis, the underlying reference distribution – from an outlook-at-risk model – would have signalled value in exploring a more extreme downside case to better encompass the looming recession risks. During periods of elevated uncertainty, such a tool can prove invaluable to policymakers.
MTP discussion
The remainder of the event featured presentations on four MTPs, covering dynamic stochastic general equilibrium (DSGE) modelling, nowcasting, structural VAR modelling, and quantitative risk assessment. Each was followed by active discussion from an audience of experts from academia, public policy and the private sector, with contributions from invited discussants.
An estimated DSGE model for the UK economy
The inaugural MTP, ‘Decompositions, forecasts and scenarios from an estimated DSGE model for the UK economy’, documents the latest evolution of the Bank’s medium-scale, open-economy DSGE model of the UK economy – known as 'COMPASS’. Alongside forecasts and shock decompositions, it can be used to construct scenarios, either by simulating a range of economic shocks or altering underlying structural parameters.
The presentation highlighted innovations since the original 2013 Staff Working Paper, most notably the introduction of an energy sector – building on earlier Bank research – to better capture the role of energy prices in inflation dynamics. As the MTP describes, this improves the model’s ability to reflect the channels through which energy can affect demand and supply in general equilibrium.
Chart 1 illustrates results from a ‘counterfactual’ exercise asking: ‘Had we known in advance how key exogenous variables including energy prices evolved, what would the model have predicted for inflation and GDP growth?’. Grey lines show data available as of November 2021 and aqua lines show data as of November 2024. Orange lines show the counterfactual forecast starting from November 2021 and conditioned on the realised path for energy prices over the 2022−24 period, alongside 68% and 90% confidence intervals.
Such counterfactuals can help to assess how well the model is able to capture the dynamics generated by certain shocks – in this case energy – even if the shocks themselves are impossible to predict in real time and so can lead to forecast errors. Chart 1 shows the resulting counterfactual forecast for UK CPI inflation is relatively close to the observed profile. While data outturns were still somewhat higher, they broadly fall within the model’s confidence bands. The remaining gap may reflect second-round effects and non-linearities that are the subject of ongoing research. For real GDP growth, the counterfactual is largely in line with the observed outturns.
Discussion included a suggestion to model energy, and other key mechanisms like expectations formation, as regime-dependent – salient during crises, more muted during calmer periods. There was agreement that such non-linearities would be valuable, particularly for exploring tail risks via scenarios.
Chart 1: Counterfactual DSGE model forecasts
Inflation (year on year)
Real GDP growth (year on year)
Energy contribution to CPI inflation (quarter on quarter)
Bank Rate
Footnotes
- Notes: Figure 12 in Albuquerque et al (2025). Counterfactual projections for year-on-year inflation and GDP growth based on data available as of 2021, but conditional on realised values of key conditioning paths as of November 2024. Grey lines show data as of November 2021, aqua lines represent data as of November 2024, and orange lines are model’s counterfactual forecasts with 68% and 90% confidence intervals (orange dotted lines).
Nowcasting UK GDP
MTP No.2, Nowcasting GDP at the Bank of England: a Staggered-Combination MIDAS Approach, outlines one of the Bank’s preferred approaches to ‘nowcasting’ UK GDP. Statistical estimates of current GDP growth, or ‘nowcasts’, are a critical input to policymaking. Nowcasting remains an active area for applied model development, with a range of approaches for UK GDP emerging since the Office for National Statistics began to publish monthly GDP data in 2018 (eg NIESR work).
The Staggered-Combination MIDAS (SC-MIDAS) approach is designed to address specific challenges of nowcasting a lower-frequency variable – in this case quarterly GDP growth, which remains the more important concept to policymakers – when a higher-frequency measure – monthly GDP – is also available. The method combines mixed-frequency ‘MIDAS’ regressions with forecast combination techniques, drawing on both ‘hard’ monthly GDP data and ‘soft’ surveys like S&P Global’s PMIs to forecast the ‘first estimate’ of quarterly GDP.
As Chart 2 illustrates, this structure allows the model to put greater weight on timelier and less volatile soft data at longer horizons (first chart) but increasingly exploit the mechanical link between monthly and quarterly GDP as monthly GDP outturns become available. Accuracy increases markedly through the quarter as a result (second chart). The paper further shows that SC-MIDAS consistently outperforms a range of competitor approaches.
Discussion touched on alternative nowcasting approaches and how they might be used to complement this model. These included dynamic factor models to target GDP revisions beyond the first estimate, quantile-MIDAS models to estimate near-term risks, and more sophisticated Bayesian techniques.
Chart 2: SC-MIDAS combination weights and RMSE evolution
Footnotes
- Notes: Figures 9 and 11 in Moreira (2025). First chart shows evolution of combination weights for ‘hard’ and ’soft’ signals over a 180-day nowcast window. Second chart shows corresponding RMSEs (2005−19) for ‘full’ SC-MIDAS nowcast and ‘hard’ and ‘soft’ components.
A structural VAR for the UK economy
MTP No.3, A Structural VAR for the UK economy, introduces a versatile tool that can be applied to decomposing the structural drivers of forecasts and their revisions, alongside more traditional structural VAR applications like impulse response, historical and variance decompositions. It can also be applied to a range of alternative VAR specifications, making it a flexible tool for policy analysis.
The presentation focused on one policy-relevant use case: this model’s ability to decompose the drivers of successive forecast revisions. As the paper explains, forecast revisions can reflect either information from new data or revisions to existing data. The authors’ approach provides an intuitive interpretation of forecast changes, using ‘SVAR’ techniques to identify the structural shocks responsible and the channels through which they operate.
Chart 3 demonstrates this. The first two charts shows overall revisions to inflation and real-GDP growth forecasts between May and August 2024. The bottom two charts decompose those into structural shocks: with the largely unrevised inflation forecast resulting from offsetting shocks to world and UK demand at the time, and the upward revision to GDP growth emanating from a range of sources, including stronger domestic demand and lower energy prices. Such decompositions can help policymakers understand forces shaping the outlook, as well as uncertainties around them.
Among other avenues to explore, the discussion suggested potential enhancements to the sign-restriction identification approach, for example by complementing them with prior information on the short-run elasticity of activity to oil-price changes.
Chart 3: Decomposing effects of newly identified shocks from period T data
Inflation (year on year)
Real GDP growth (year on year)
Inflation (year on year) decomposition (difference)
Real GDP growth (year on year) decomposition (difference)
Footnotes
- Notes: Figure 8 in Brignone and Piffer (2025). The first two charts show joint effect that the shocks estimated in 2024 Q2 have on year-on-year inflation and real GDP growth. Pointwise median and 68%/90% credible intervals reported. The last two charts show decomposition of the median response.
Forecasting macroeconomic risks in the UK
The fourth MTP presented, Forecasting Macroeconomic Risks in the UK, is forthcoming. It applies quantile-regression methods to build density forecasts, building on earlier Bank work on both inflation- and GDP-at-risk. Such tools can help policymakers better understand the scale and direction of risks, and their evolution.
From a forecasting perspective, these tools can illustrate how the probability distribution over potential future outturns changes over time. Chart 4 provides an example, plotting one-year-ahead forecast distributions for UK CPI inflation produced by the model as of January 2024 (aqua) and January 2025 (orange). While the modal forecast shifted closer to 2%, the distribution of risks became more skewed to the upside.
They can also help shed light on the drivers of movements in the tails of predictive distributions. For inflation, the most pronounced variation occurs in the right tail, where inflation expectations, economic slack, global oil prices and domestic financial conditions can all be drivers of upside risks in one to two-year ahead predictive distributions. For GDP growth, variation is more concentrated in the left tail, where financial conditions (in the near term) and credit growth measures (over the medium term) are key indicators of downside risks.
Helpful advice on the adoption of these tools in a policy setting included using a plurality of specifications and density combination as a way of guarding against the illusion of a comprehensive uncertainty assessment stemming from any one model.
Chart 4: Predictive distributions for CPI inflation forecasts
Footnotes
- Notes: Aikman et al (forthcoming). One-year-ahead predictive density for UK CPIH inflation from quantile regressions, estimated in pseudo-real-time in January 2024 (orange) and 2025 (aqua).
Only the beginning
This ’forum’ was the first of many to be held around the UK, as the MTP series continues to document Bank model development – echoing themes highlighted in the collection of reactions to the Bernanke Review published by Kings College London.
For this event, Bank staff would like to thank the National Institute for Economic and Social Research for hospitality, King’s College London’s Impact Acceleration Fund for financial support, Professor Giannone for his path-breaking keynote, discussants for thoughtful suggestions, and attendees for active participation.
This post was prepared by Simon Lloyd and Andre Moreira, with the help of David Aikman, Neil Lakeland and Margherita Servente of the National Institute of Economic and Social Research.
Share your thoughts with us at BankOverground@bankofengland.co.uk.
