Forecasting for monetary policy making and communication at the Bank of England: a review

In July 2023 the Court of the Bank of England announced that Dr Ben Bernanke would lead an independent review into the Bank’s forecasting and related processes during times of significant uncertainty. That Review, published on 12 April, provides a thorough assessment of the Bank’s current forecasting approach, and the relationship between the forecast, monetary policy decisions, and their communication.
Published on 12 April 2024

Preface and acknowledgements

In July 2023 I accepted an invitation from the Court of Directors of the Bank of England to review economic forecasting at the Bank, with a particular emphasis on how forecasting can better support policymaking and communication during times of high uncertainty and structural change. This report is the product of that review. Earlier official reviews of the Bank’s forecasting include Stockton (2012) and Independent Evaluation Office (2015). I build to some extent on that earlier work. However, in light of the passage of time and, especially, the challenging economic environment of recent years, a fresh look at the construction and use of economic forecasts at the Bank seems timely.

In the process of conducting the review, I solicited a wide range of views. With the able assistance of Melissa Davey and her colleagues in the Bank’s Independent Evaluation Office (IEO), including Michael Lever and Sophie Stone,footnote [1] as well as of Sam Boocker, my research assistant at the Brookings Institution, I conducted some 60 interviews of individuals and small groups. Interviewees associated with the Bank included all current MPC members, selected past MPC members, senior Bank staff, the independent chairs of the Bank’s Citizens’ Forums, and Bank Agents (who serve as regional representatives and information gatherers for the Bank). Separately, the IEO team, Boocker, and I also hosted a series of working lunches with representative groups of Bank staff from different divisions and at all levels of seniority.

For the perspectives of people outside the Bank, we interviewed UK print and broadcast journalists, financial market participants, other UK forecasters (including representatives of the Office for Budget Responsibility and the National Institute of Economic and Social Research), academics, and economists of the Trade Union Congress. I also met with the chairs of the relevant Parliamentary committees in both the House of Commons and the House of Lords to update them on the project and to hear their views. I would like to express my gratitude to all the interviewees, who were uniformly generous with their time and forthcoming in their responses to our questions.

Some of the interviews and meetings were conducted online. However, Boocker and I spent three weeks at the Bank during the period running up to the 2 November 2023 policy decision and were consequently able to meet in person with many individuals and groups, both internal and external to the Bank. Boocker and I both attended staff meetings and meetings between staff and MPC members at which the forecast was developed and debated. With special permission I (without Boocker) attended the meetings at which the MPC discussed and finalised its policy decision. We reviewed relevant academic articles, official documents, and related materials. With the IEO taking the lead, we conducted case studies bearing on the Bank’s forecasting and use of forecasts in recent years.

With substantial help from the IEO team, we compared the forecasting procedures and recent forecasting records of the Bank with those of six peer central banks: the US Federal Reserve, the European Central Bank, the Reserve Bank of New Zealand, the Bank of Canada, the Norges Bank (Norway) and the Sveriges Riksbank (Sweden). We also compared the Bank’s forecast accuracy with that of external forecasters. For background information on the peer central banks, we held online meetings with the staff leading the forecast process at each of those banks (except for the Federal Reserve, with which I am already familiar) and reviewed publicly available materials. The central bank staff members with whom we spoke were eager to be helpful, going out of their way to help us better understand their forecasting procedures, the roles of staff and policymakers in forecast development, and the use of forecasts in policymaking and communication at their institutions.

The forecasting and policy challenges faced by the Bank of England in recent years were hardly unique, as peer central banks faced similar shocks and dealt with similar challenges. Still, the recent experience served as a stress test of forecasting at the Bank, including not only of the routine construction of forecasts but also of the use of forecasts in policymaking and public communication. The Bank, like other central banks and policy institutions, will be working to draw the appropriate lessons from this experience. The goal of this review is to assist in this effort.

Ben S. Bernanke
Brookings Institution
12 April 2024

Executive summary

This report reviews economic forecasting at the Bank of England. The report’s remit is broad (see Annex A for the published terms of reference). Specifically, the charge of the review is to ‘consider the appropriate approach to forecasting and analysis in support of decision-making and communications in times of high uncertainty from big shocks and structural change…’. To meet this charge, the report discusses and evaluates the current forecasting process at the Bank, including the adequacy of the forecasting infrastructure (data, software, and models); the utilisation of the staff; the process of constructing the quarterly forecast, including the interactions of the staff and the Monetary Policy Committee (MPC, or Committee); issues raised by high uncertainty and structural change; the use of the forecast in the MPC’s policy decisions; and the role of the forecast in the MPC’s communications with the public, the media, and financial market participants.

The structure of the report and its conclusions are summarised below. In brief, the recommendations made in this report have three broad objectives: first, to improve and maintain the Bank’s forecasting infrastructure, including data management, software, and economic models; second, to support an effective policy process by equipping the MPC and the staff to learn from past forecast errors, to identify and quantify risks to the outlook, and to deal with uncertainty and structural change; and third, to help the MPC better communicate its view of the economy, the risks and uncertainties surrounding its outlook, and the basis for and implications of the Committee’s policy choices.

Part I of the report sets the stage with some general observations, not specific to the Bank of England and optional for the informed reader, about the construction of economic forecasts and central banks’ use of forecasts in policymaking and in communications with the public. With that background, Part II describes the Bank’s current practices. Forecasts of the UK economy for the subsequent three-year period are constructed quarterly, in a process that begins in earnest some six weeks before a monetary policy meeting. Although Bank staff are responsible for producing a first draft of the forecast and providing supporting analysis, staff and MPC members work together to put together the final product, which is approved by the MPC. The forecast is subsequently published in the Bank’s Monetary Policy Report (MPR) and summarised in other official releases. Relative to other central banks, the forecast plays a particularly large role in the Bank’s public communications and accordingly receives considerable attention in the media and from financial market participants.

Part III compares the Bank’s processes and recent forecast accuracy with those of a set of peer central banks since 2015. We find that, while the accuracy of the Bank’s economic forecasts has deteriorated significantly in the past few years, forecasting performance has worsened to a comparable degree in other central banks and among other UK forecasters. The recent period was characterised by a series of large shocks that were, by their nature, difficult to forecast, and were generally not predicted by financial markets or external experts. Among these shocks were the pandemic itself, along with its economic and policy consequences; the sharp increases in oil, gas, and other commodity prices, especially following Russia’s invasion of Ukraine; and the sustained disruption of global supply chains during and following the pandemic. The large and correlated forecasting errors of central banks around the world during the past few years support the view that global shocks dominated local factors as sources of those errors at the Bank and elsewhere; and that, to the extent that deficiencies of forecasting methods or economic analysis account for the misses, the deficiencies were characteristic of the central banking community in general rather than the Bank alone. In short, given the unique circumstances of recent years, unusually large forecasting errors by the Bank during that period were probably inevitable. It is nevertheless important for the Bank to draw what lessons it can from the experience, including lessons regarding how it uses its forecasts, as other central banks will certainly do.

Part IV assesses the Bank’s construction and use of its forecasts and makes recommendations, listed below. In accordance with the objectives of our report, the recommendations are organised according to three major themes: building and maintaining a high-quality infrastructure for forecasting and analysis; providing a forecast process that better supports the MPC’s decision-making; and using the forecast to communicate the MPC’s outlook and policy rationale to the public.

Building and maintaining a high-quality infrastructure for forecasting and analysis

The most serious problems we found in our review are the deficiencies of the Bank’s forecasting infrastructure – the tools the staff uses to produce the quarterly forecast and supporting analyses. Some key software is out of date and lacks important functionality. With the staff fully engaged in the production of the current forecast, particularly during periods of extraordinary volatility, insufficient resources have been devoted to ensuring that the software and models underlying the forecast are adequately maintained (updated, stress tested, and periodically re-estimated). In particular, the baseline economic model, known as COMPASS, has significant shortcomings. These deficiencies in the framework, together with a variety of makeshift fixes over the years, have resulted in a complicated and unwieldy system that limits the capacity of the staff to undertake some useful analyses, including producing alternative forecast scenarios, using information gleaned from forecast errors to improve model specifications and forecasting methods, and considering alternative modeling frameworks. A positive development is that an effort to upgrade the data management system is under way. This report describes the issues with the forecasting infrastructure and makes four recommendations.

Recommendation 1. The ongoing updating and modernisation of software to manage and manipulate data should be continued with high priority and as rapidly as feasible. At completion, the modernisation project should ensure that:

(i) the economic and financial data available to the staff are comprehensive, covering all key sectors at the relevant frequencies; clearly defined, with sources provided; updated in a timely way; and easily searchable;

(ii) staff are able to export and transform data series as needed to construct figures, tables, and routine econometric analyses quickly and efficiently and with adequate source control;

(iii) large data sets, both time series and cross-sectional, can be ‘cleaned’ and used efficiently in substantive analysis and research; and

(iv) the inputting of data to the suite of economic and statistical models, especially for routine operations including forecasting and scenario analysis, is automated to the extent possible.

The Bank might consider whether adding a few data specialists to work with economists in accessing and working with data, especially larger and more complex data sets, would make the forecasting process work more smoothly.

Recommendation 2. Model maintenance and development should be an ongoing priority, supported by a significant increase in dedicated staff time and adequate resources, including specialised software as needed. To be most effective, the dedicated staff should have ample opportunities to interact with ‘front-line’ forecasting staff, MPC members, and external experts. The maintenance and development staff should ensure that forecasting models are regularly evaluated, re-estimated when new data become available, stress tested against alternative scenarios, and modified as needed to reflect new perspectives on the economy.

Recommendation 3. Over the longer term, the Bank should undertake a thorough review and updating of its forecasting framework, including replacing or, at a minimum, thoroughly revamping COMPASS. The specific framework and models to employ should be decided over a period of time by the staff with MPC input. However, so that staff can respond to policymakers’ requests for new analyses in a timely way, flexibility, transparency, and ease of use (including automation of processes now carried out manually) should be important criteria for a restructured system.

Recommendation 4. Based on the lessons of recent years, a revamped forecasting framework should include at least the following key elements:

(a) rich and institutionally realistic representations of the monetary transmission mechanism, allowing for alternative channels of transmission;

(b) empirically based modelling of inflation expectations, with a distinction between short-term (eg, one-year) and longer-term (eg, five to ten years) expectations, and without the assumption that longer-term inflation expectations are always well-anchored;

(c) models of wage-price determination that allow gradual adjustment and causation from prices to wages as well as from wages to prices;

(d) detailed models of the financial sector, the housing sector, the energy sector, and other key components of the UK economy;

(e) greater attention to, and ongoing review of, supply-side elements and their role in the determination of inflation and growth. Important supply-side factors include changes in productivity, labour supply, the efficiency of job-worker matching, supply-chain disruptions, and trade policy. Notably, analyses of inflation should consider supply-side factors as well as the state of aggregate demand.

Recommendation 4 is not intended to imply that the Bank’s current framework lacks all these features by any means. Rather, it is a checklist of key elements that a revamped framework should be sure to include.

Providing a forecast process that better supports the MPC’s decision-making

The goal of the forecasting process, of course, is to help the MPC make better policy decisions and to effectively communicate those decisions to the public. This report reviews how the MPC uses the forecast today, noting the evident strengths of the process but also suggesting possible improvements.

To deliver a forecast process that better supports the MPC’s decision-making, the report makes three related recommendations. We argue that the current bias toward making incremental changes in successive forecasts, together with the use of human judgements that paper over problems with the models, may slow recognition of important structural changes in the economy. Building on the joint analytical work of the staff and MPC during each forecast round, a systematic effort should be made to address these issues. In addition, policymaking could be made more systematic and coherent by supplementing the central forecast with additional information and analysis, notably insights drawn from alternative scenarios (including forecasts made based on alternative paths for the standard conditioning assumptions). Currently, the Bank regularly publishes a scenario that assumes constant interest rates, and in recent years it has occasionally used scenarios to explore the consequences of energy price shocks and other risks. Expanded use of alternative scenarios would facilitate comparisons of possible policy choices, more accurately quantify the risks to the forecast, and help the Committee learn from past forecast errors. The report also suggests possible changes in the use of personnel, including incentivising the accumulation of experience in key substantive areas and making better use of the share of PhD researchers’ time devoted to policy work.

Recommendation 5. Incrementalism (the practice of basing new forecasts on previous forecasts, with marginal adjustments) and the use of ad hoc judgements may obscure deeper problems with the underlying forecasting framework or unrecognised changes in the structure of the economy. The staff should be charged with highlighting significant forecast errors and their sources, particularly errors that are not due to unanticipated shocks to the standard conditioning variables. Models and model components that may have contributed to forecast misses should be regularly evaluated and discussed, as well as the determinants of variables whose forecasts are consistently dominated by extra-model judgements. Staff should routinely meet with MPC members to consider whether structural change, misspecification of models, or faulty judgements warrant discrete changes to the key assumptions or modeling approaches used in forecasting. Willingness to modify existing frameworks and to consider new data or other information is particularly important during periods of high uncertainty. The Bank should also build on existing vehicles for external engagement to capture a broad range of views.

Recommendation 6. The Bank should review its personnel policies to determine if existing staff could be deployed in ways that improve the forecasting infrastructure and forecast quality. In general, employees should be more strongly encouraged and incentivised to accumulate experience and expertise in specific substantive areas (eg, through in-role promotions), rather than being forced to change fields or responsibilities to get promotions and raises. Researchers with doctorates should continue to spend part of their time in undirected or loosely directed research, with the best researchers afforded the opportunity to continue working on their individual agendas throughout their careers. However, during the portion of their time devoted to current forecasting and analysis, employees with more advanced degrees should be rewarded for taking leading roles, especially in longer-term model maintenance and the development of new and existing models. PhD researchers can also contribute by undertaking substantive analyses directly related to current issues and, as appropriate, by being given the chance to lead in technical areas that make good use of their training and research experience.

Recommendation 7. To improve the MPC’s policy discussion, the central forecast should be regularly augmented by alternative scenarios, with the specific scenarios ideally decided upon at an early stage of each forecast round by the MPC and staff. Among the types of scenarios that could be considered are those that: (1) allow for direct comparisons of the likely effects of alternative policy paths on the outlook; (2) help to assess the effects and costs of possible risks to the outlook arising from unexpected changes in exogenous variables; (3) can be used to evaluate the effects of the Committee’s policy choices on the economy if one or more of its key assumptions about the structure of the economy are wrong; and (4) can be used to decompose historical forecast errors into portions due to judgements, conditioning assumptions, and other factors.

Using the forecast to communicate the MPC’s outlook and policy rationale to the public

Effective communication is essential for effective monetary policy. Good communication helps the public understand the rationale and implications of policy choices and can make policy work better by helping to anchor inflation expectations and by influencing asset prices. Relative to other central banks, the Bank of England relies heavily on its central economic forecast (which, it should be emphasised, involves human judgement and diverse information sources as well as the output of econometric models) as a communications device. This report argues that the publication of selected alternative scenarios along with the central forecast would improve the Bank’s communications, providing the public with additional useful information about the rationales for policy choices, the risks to the forecast, and the robustness of the MPC’s policy plans in the face of uncertainty about key aspects of the economy’s state and structure.

Recommendation 8. The publication of selected alternative scenarios in the MPR, along with the central forecast, would help the public better understand the reasons for the policy choice, including risk management considerations. The publication of selected alternative scenarios could also provide the public with information about the Committee’s policy reaction function and its views of the monetary transmission mechanism. The MPC should determine which scenarios are published, choosing those that members deem to be most informative about the policy decision at a particular time. There should be no presumption that the same scenarios will be published in each MPR.

The Bank’s forecast is conditioned on a set of standard, externally determined assumptions about the future course of policy rates, fiscal policy, exchange rates, and commodity prices. Unfortunately, these standard conditioning assumptions – for example, the assumption that future policy rates will follow the path revealed in futures markets – may not always accurately represent the views of the MPC, with the result that the central forecast may not fully reflect the Committee’s outlook for the economy. This report makes two related recommendations regarding the standard conditioning assumptions.

Recommendation 9. Because the standard conditioning assumptions do not necessarily reflect the MPC’s views but can have potentially significant effects on the forecast, and because the central forecast by itself does not provide a clear rationale for policy decisions, the MPC should de-emphasise the central forecast based on the market rate path in its communications and be exceptionally clear in warning about situations in which it judges the standard conditioning assumptions to be inconsistent with its view of the outlook. Methods for doing this include giving more attention to published alternative scenarios in discussions of the outlook and policy; emphasising to an even greater degree the conditionality of the forecast on exogenous assumptions not chosen by the MPC; and, when appropriate, using the MPC’s limited discretion to modify the standard conditioning assumptions. Judgemental adjustments might also be used to offset the effects on the forecast of conditioning assumptions with which the Committee disagrees, but that approach has the significant disadvantage of sending inaccurate signals to market participants about the MPC’s assessment of the rate path consistent with the Committee’s objectives.

Recommendation 10. To put less emphasis on the central forecast, to simplify its policy statement, and to reduce repetitiveness in its communications, the MPC should replace or cut back the detailed quantitative discussion of economic conditions in the Monetary Policy Summary in favour of a shorter and more qualitative description, following the practice of most peer central banks.

A more aggressive approach to addressing the problem of potentially inappropriate conditioning assumptions, following the practice of several peer central banks, would be to replace the market-based path for the policy rate with the MPC’s own forecasts of Bank Rate, based either on a collective judgement or by aggregation of individual member judgements. However, that change would be highly consequential and this report recommends leaving decisions on this issue to future deliberations.

Communicating to the public the high degree of uncertainty associated with any economic forecast is important. Currently, the MPC uses fan charts to convey the range of uncertainty in forecasts of key economic variables at varying horizons. The report argues that fan charts suffer from significant analytical weaknesses and have outlived their usefulness.

Recommendation 11. Despite their distinguished history, the fan charts as published in the MPR have weak conceptual foundations, convey little useful information over and above what could be communicated in other, more direct ways, and receive little attention from the public. They should be eliminated. However, it remains important to communicate the degree of forecast uncertainty and the balance of risks. A section in the MPR should be devoted to uncertainty and the balance of risks in the forecast. Beyond verbal discussion that describes uncertainty and risk in qualitative terms (terms that should be echoed in other Bank releases), this section could include the record of forecasting errors by the Bank, perhaps including new time series figures and discussion; an analysis of recent forecast errors, together with steps taken (if any) to correct the factors that contributed to those errors; and an overview of the risks to the outlook, possibly with reference to alternative scenarios published in the MPR. Mean forecasts as currently constructed do not provide additional useful information and should also be dropped from publications in favor of more qualitative descriptions of risks and uncertainty surrounding the outlook.

Importantly, this report’s proposed changes to the use of the forecast in policymaking and communication are dependent on improving the capabilities and flexibility of the forecasting infrastructure. Accordingly, the last recommendation is about sequencing and resources.

Recommendation 12. A phased approach to implementing changes proposed in this report, focused first on improving the forecasting infrastructure, while moving cautiously in adopting changes to policymaking and communications, is likely to be necessary. To facilitate infrastructure improvements and address existing deficits, the commitment of additional resources will be required, at least for a time.

Part I: Why and how do central banks forecast?

Economic forecasting is difficult even under the best of circumstances. Modern economies are complex and ever-changing, and they are subject to unpredictable shocks, including non-economic shocks such as pandemics or wars. Even the current state of the economy is difficult to observe (‘nowcasting’ the economy is a specialised skill) as most economic data are available only with a lag and provide at best a rough, statistically noisy, and often subject to revision snapshot of current economic developments. Recessions – periods of economic contraction – are particularly difficult to anticipate. Many economists expected a recession to occur in the United States in 2023, for example, but economic growth and job creation remained strong. This is not to say that economic forecasting is impossible – both experience and formal studies confirm that forecasts made in real time do contain useful information about the future courses of key economic variables – but it is inevitably subject to a high degree of uncertainty, uncertainty which increases rapidly for forecasts of the more distant future and during periods of large shocks or rapid structural change.

So why do central banks and other policymakers continue to devote so much time and resources to making economic forecasts? For central banks, forecasts are important for two broad reasons: they aid in the formulation of policy. And they are a tool for communicating policy plans and rationales to the public and financial markets.footnote [2]

Policy formulation

Monetary policy makers understand that their actions affect the economy only over time – with ‘long and variable lags’, as Milton Friedman famously put it. Because of these inevitable lags, policymakers must take a view of how the economy is likely to evolve, at least in broad terms, which in turn provides a basis for choosing policies that promote the achievement of policymakers’ objectives over time. Developing that view requires policymakers to do the hard work of analysing the forces affecting the economy and ensuring that their proposed policies are consistent with that view. At the same time, uncertainty about the structure of the economy and the inevitability of unanticipated shocks imply that forecasts can never be set in stone. Effective policymakers recognise that their view of the outlook and the associated policy strategy must be continually updated as new information arrives, and they communicate this point repeatedly to the public.

Astute policymakers also recognise that economic forecasts can be viewed as tests of their current understanding of the economy and the effects of policy. When a forecast proves to be significantly off target, in a way that is not easily explained by unanticipated shocks, analysis of the sources of the error can help improve that understanding. A tendency for forecasts to miss in the same direction over an extended period is a particularly strong signal that the forecasters’ implicit or explicit model of the economy should be re-thought and future forecasts and policy strategies modified accordingly.

A crucial distinction is that between unconditional and conditional forecasts. An unconditional forecast is an unqualified prediction of what is likely to happen in the future, given all the information currently available to the forecaster. A conditional forecast is a prediction that is contingent on specific, possibly counterfactual, assumptions. For example, policymakers might construct conditional forecasts to gain insight into how the economy would likely evolve if energy prices were 20% higher than in the baseline forecast or if the world economy were to slow more than currently expected. Forecasts can also be constructed conditional on alternative assumptions about how the economy works, eg, conditional forecasts could be used to estimate how the outlook would be affected if the sustainable rate of unemployment were lower than currently believed, or if the sensitivity of nominal wage growth to inflation were higher. Again, conditional forecasts do not necessarily reflect the forecasters’ beliefs about what will actually happen in the future. Instead, they are attempts to provide insight into how the world might look conditional on alternative assumptions about the structure of the economy, or about forces influencing the economy – assumptions that may or may not reflect what the forecasters actually expect.

The comparison of alternative conditional forecasts, also called scenario analysis, can aid policy formulation in several ways. First, choosing the best policy strategy requires comparing the likely economic outcomes under alternative policies – that is, constructing a set of scenario analyses with differing assumptions about policy – and choosing the strategy that is forecast to best meet policymakers’ objectives over time.

Second, scenario analysis can help policymakers adjust their strategy to account for risks to the outlook, eg, higher and more persistent inflation than implied by the baseline forecast. Under a policy approach known as risk management, policymakers might choose to take out some ‘insurance’ against bad outcomes. For example, in the case in which the most concerning risk is that inflation will be higher than in the central forecast, a risk management approach might involve running a tighter policy than would be chosen based on the main economic scenario alone. Scenario analyses can help quantify the economic consequences of various risks to the outlook, providing guidance about the extent to which risk management considerations should affect policy choices.

Third, similar considerations apply when policymakers are uncertain about a key parameter or some other structural element of their model of the economy. Scenario analyses can show how uncertainty about a structural feature translates into uncertainty about key aspects of the economic outlook. If the linkage is strong, policymakers may wish to invest more effort in improving their understanding of the underlying issue, and they may be more cautious in policy decisions whose effects are dependent on that feature.

Finally, scenario analysis can be used to decompose the sources of past forecast errors. For example, policymakers could reconstruct past forecasts under the counterfactual premise that the forecast’s conditioning assumptions were accurately known in advance. The residual forecast errors in this scenario could then be attributed to factors other than unexpected changes in the conditioning variables, such as model misspecification or faulty judgements. Exercises of this type can be useful diagnostic tools for evaluating and improving the forecasting framework.

Communication with the public and financial markets

The regular publication of an economic forecast by the central bank has several communication functions. Perhaps most obviously, the forecast provides the public with a broad rationale for the policy decision, eg, a forecast that inflation will remain too high justifies a tighter policy than average, all else equal. In general, the relationship between the central bank’s economic outlook and the policy strategy it chooses is called the policy reaction function. Analysis of the responses of policy to monetary policy makers’ outlook over time helps identify the bank’s reaction function and improves the ability of outsiders to anticipate how policymakers will respond to various contingencies.

Better understanding of the reaction function by the public and financial market participants in turn helps to align private economic decisions and general financial conditions more broadly with the central bank’s view, which can make policy more effective in moving the economy in the desired direction. Policy is also made more effective if communications, together with the central bank’s policymaking record, help to tie down (‘anchor’) longer-run inflation expectations. If the public are confident that the central bank is committed to achieving its inflation target in the medium term, the risks of a self-fulfilling, expectations-driven wage-price spiral are much reduced.

Although publication of a forecast along with the policy decision can help the public learn about the policy reaction function over time, once again scenario analyses can make the process more precise and effective. For example, if (along with the central forecast) the central bank publishes alternative scenarios that indicate how policy would likely be set if the economy were to evolve in a manner different than expected, the public will be able to draw sharper inferences about the reaction function and thus better anticipate future policy actions. In addition, the publication of alternative scenarios in which the expected future path of interest rates is allowed to vary could provide useful information to the public about how policymakers expect monetary actions to affect the economy (the monetary transmission mechanism), providing further insight into why policymakers made the decisions they did.

In sum, publishing a forecast can be an important tool for making the central bank more transparent and accountable. Monetary policy affects the lives of almost everyone, and policymakers owe the public clarity about the factors driving their decisions. Published forecasts force policymakers to explain (1) the analysis underlying the forecast, including the factors that policymakers see as most important for the outlook; (2) the reasons for significant changes over time in the outlook, and how policymakers are responding to those changes; (3) retrospectively, the factors responsible for significant past forecast errors and how policymakers are adjusting to and learning from those errors; and (4) the consistency of the policy plan with policymakers’ mandated objectives. Again, publication of alternative scenarios along with the main forecast can potentially provide even greater transparency, in that those scenarios may reveal information about the factors that policymakers considered in their decision-making and illustrate risks about which policymakers are particularly concerned.

Finally, the publication of forecasts invites two-way communication with the public. For example, sufficiently detailed explanations of the central bank’s outlook should make it possible for outside analysts to think along with policymakers, probing the forecast’s underlying assumptions and providing commentary and feedback that conceivably could improve policy decisions.

The construction of economic forecasts

Even inaccurate forecasts, if constructed by consistent methods using the best available information, can help improve the coherence and predictability of policy. But of course, all else equal, better forecasts imply better policymaking and communication. The construction of economic forecasts, at most central banks, is done primarily by professional staff, in many cases with some input from policymakers. Effective forecasting makes use of many kinds of information besides official economic data and remains as much an art as a science.

For forecasting and economic analysis, most central banks make use of econometrically estimated macroeconomic models, which are sets of equations that provide quantitative representations of key behavioural relationships (such as the responsiveness of aggregate consumption to changes in labour income or asset values, for example), while ensuring that necessary relationships (such as national income accounting identities) are respected in the forecast. When simulated on a computer, and conditional on underlying assumptions about the behaviour of variables outside the model (exogenous variables), macroeconomic models produce forecasts of key economic series, such as inflation and unemployment. Economic quantities whose values are determined by the model are called endogenous variables.

Economists use several different types of macroeconomic models for forecasting, either alone or in combination. So-called semi-structural models are (typically quite large) economic models that loosely combine equations describing key sectors of the economy and that flexibly model expectations formation and economic dynamics. The Federal Reserve’s FRBUS is an example of a semi-structural model. Another type of macroeconomic model, of which the Bank of England’s COMPASS is an example, are dynamic stochastic general equilibrium (DSGE) models. DSGE models are built up from microeconomic representations of the behaviour of individual households and firms. These models are dynamic in that they focus on intertemporal choices, such as saving and investment decisions; stochastic, in that decision-makers are assumed to consider randomness in the environment; and general equilibrium, in that the models require that all economic choices be determined simultaneously in a mutually consistent way. The roots of semi-structural models are the large Keynesian models first developed in the 1960s. DSGE models had their beginnings in the new classical revolution of the 1980s but, unlike the original new classical models, most DSGE models in use today include slow adjustment of wages and prices and other Keynesian features.

While a macroeconomic model may help the forecaster construct a baseline outlook for the economy, in almost all cases it will be supplemented by other types of models and sources of information. Supplementary models used by central banks include both so-called sectoral models and statistical models. Sectoral models come in different forms, but – as the name implies – they typically provide more detailed representations of a particular sector of interest, such as the energy sector, the housing sector, or the financial sector. The predictions of such models can be used to fill in areas where the coverage of the overarching macroeconomic model is thin. This extra detail can be especially important when unusual developments in a particular sector – say, problems in banking and credit markets – have the potential to influence the overall economy.

Statistical models use little or no economic theory but instead rely on ‘black box’ mathematical models that are estimated from the historically observed behaviour of certain variables or sets of variables. For example, vector autoregression models are built on estimates of the responses over time of a set of variables (say, inflation, output, and unemployment) to changes in the past values of the same variables. Because they are typically unconstrained by economic theory, the predictions of vector autoregressions and other statistical models provide useful checks on the forecasts of macroeconomic models, which incorporate many (possibly incorrect) assumptions about economic behaviour.

Importantly, notwithstanding the role of formal models, no central bank relies entirely on models for its forecast. Human judgement remains a critical, sometimes even dominant, element of most real-world forecasts. People (including both staff and policymakers) can identify and correct for factors not adequately captured by econometric or statistical models, including possible structural changes to the economy or historically unusual developments. The people overseeing the forecast can do this in part because they have access to sources of information not available to the models, such as business and community contacts, journalistic accounts, and personal experience in the private sector, government, or academia. People who work extensively with particular models or methods also become sensitive to the ‘blind spots’ of the models – factors excluded from the models that can be important at times – and they can adjust model output, formally or informally, to compensate for systematic undershoots or overshoots in previous forecasts. In short, the predictions of formal econometric models are only one of several inputs, and not necessarily the most important one, to the forecasting process.

Finally, traditional forecasting methods are increasingly being supplemented by methods based on new technologies or data sources. Many central banks already make use of large data sets (‘big data’), such as (anonymised) credit card or mortgage records, to get more timely and granular information about the state of the economy. During the pandemic, many central bank staffers (including at the Bank of England) consulted closely with epidemiologists and other public health professionals to better understand Covid-19’s economic consequences. Artificial intelligence tools, which can extract information from immense bodies of qualitative and quantitative data, seem certain to be increasingly important for monitoring the economy and forecasting in the future.footnote [3] Central banks are already preparing for that eventuality.

Part II: The construction and use of economic forecasts at the Bank of England

Background

By statute (the Bank of England Act of 1998), monetary policy at the Bank of England is the responsibility of the Monetary Policy Committee, or MPC. Except when there is a vacancy, the MPC has nine members. A non-voting representative of the Treasury also attends policy meetings. The five internal (to the Bank) members of the MPC include the Governor, three Deputy Governors (with special responsibilities for, respectively, monetary policy, financial stability, and markets and banking) and the Bank’s Chief Economist. Four external members are appointed by the Chancellor of the Exchequer for renewable three-year terms. External members are chosen through an open application process and need not be (and often are not) UK citizens. The rationale for including external members is to promote greater diversity of thought and to bring skills and experience to monetary policy making that may differ from those of the internal members.

The Committee is responsible for maintaining price stability, which is defined in an annual remit letter from the Chancellor. Currently the MPC is charged with keeping the CPI measure of inflation close to a 2% target and, subject to that, supporting the Government’s economic policy, including its objectives for growth and employment. Each member has one vote on policy, and is responsible for that vote (for example, individual members are called to testify before Parliamentary oversight committees). Following a recommendation made by Kevin Warsh in his review of the Bank’s communication practices (Warsh (2014)), policy meetings are now held eight times per year instead of monthly as in the past.

The Bank of England Act also requires the MPC to produce a quarterly report that contains:

(a) a review of the monetary policy decisions published by the Bank in the period to which the report relates,

(b) an assessment of the developments in inflation in the economy of the United Kingdom in the period to which the report relates, and

(c) an indication of the expected approach to meeting the Bank’s objectives of price stability and, subject to that, supporting the Government’s economic policies for growth and employment.

Although the Act does not explicitly specify production of an economic forecast, the MPC, like policymakers at most central banks, regularly include forecasts in their reports, in greatest detail in the quarterly Monetary Policy Report (MPR, formerly the Inflation Report). Among the key variables forecast are consumer price inflation (CPI), the growth rate of real gross domestic product (GDP), the unemployment rate, and the excess or shortfall of aggregate demand relative to aggregate supply, which is intended to indicate the degree of pressure on prices. Forecasts are made for the subsequent three-year period. The forecasts are modal, meaning that they predict the most likely outcomes of the forecasted variables, although, as we will discuss, attention is also given to the subjective distribution of less likely outcomes around the modal forecast. The MPC’s forecasts are typically described as the best collective judgement of the Committee, a phrase that is undefined but suggests that all MPC members (or perhaps a majority) are comfortable with the forecast, or at least its broad outlines.

Staffing the forecast

The Bank staff involved in the forecast process are mostly members of the Monetary Analysis (MA), International and Markets Directorates (Figure 1). A little over half of these staff have master’s-level qualifications, and around a third have doctorates. Of the staff with doctorates, many combine work on modelling, current economic analysis, and forecasting with a broader, typically more academic research agenda. Within MA, staff are divided into groups that specialise in the analysis of current economic conditions and short-term developments; the medium-term outlook; model development and monetary strategy; longer-term or structural changes in the economy; money, credit, and financial market developments; MPC communications; and the Agency network. Staff across two Divisions in the International Directorate provide projections and analysis of the rest of the world. And MA staff co-ordinate with staff outside the Directorate working on financial markets and financial stability issues.

In addition, each external MPC member is supported by two Bank staffers, who in turn can connect the external member with other staff as needed. The external members (past and present) that we interviewed expressed strong satisfaction with their staff support. External members in some cases can call on additional research support from associates in their home institutions, subject to appropriate insulation from sensitive materials.

Figure 1: Organogram of the areas that are involved in the forecast process

Shows that the majority of areas involved in the forecast process are in the Monetary Analysis Directorate, with other involved areas including the External MPC Unit and parts of International and Markets Directorates.

The Bank’s modelling and forecasting tools

The Bank’s forecasts incorporate diverse information. The staff use a suite of models, both economic and purely statistical, to analyse different aspects of the economy and the outlook. However, the staff and the MPC do not mechanically use the outputs of these models to produce the forecast but instead combine their outputs with considerable judgement and diverse information from outside sources.

As at most central banks, a benchmark macroeconomic model is used to provide a baseline forecast to help organise and interpret input from other models and human judgements. Currently, the Bank’s benchmark model is a DSGE model called COMPASS, which in turn replaced earlier benchmark models.footnote [4] COMPASS is a medium-sized macroeconomic model, including 18 observable variables.footnote [5] It reflects both new classical and new Keynesian influences. Following the new classical approach, it assumes optimising behaviour and rational expectations by households and firms.footnote [6] In the new Keynesian tradition, the model includes the assumption that wages and prices adjust only slowly; absent this assumption or a similar one, monetary policy would not affect the real economy in model simulations. COMPASS also contains the assumption that long-run inflation expectations are fixed at 2% (‘well anchored’ expectations), which creates a presumption that, in simulations of the model, inflation will eventually return to target.footnote [7]

Although COMPASS retains its official status as the Bank’s benchmark model, its role in constructing the forecast has diminished considerably in recent years. This has reflected both various shortcomings of the model that have become apparent and a lack of investment in supplementary models intended to fill in missing details of the (relatively small) COMPASS model (eg, COMPASS itself does not include detailed representations of the financial sector or the energy sector). Other identified shortcomings include COMPASS’s inability to capture fully some key channels of monetary transmission and its tendency to predict over-rapid returns of the economy to its steady-state equilibrium (including to 2% inflation). Reflecting the de-emphasis of COMPASS, the model is no longer used to predict the effects of changes in interest rates or asset prices on the economy, a fundamental element of the forecast. More generally, the shape of the forecast is not significantly constrained by the a priori theoretical properties of this model. At this point, perhaps the most important role of COMPASS is to provide a framework for aggregating the output of other models and human judgements and to ensure that key accounting relationships among variables are maintained.

As reliance on COMPASS has diminished, Bank staff have increasingly depended on a suite of sectoral and purely statistical models, modified by human judgements, for constructing the baseline forecast. For example, a disaggregated semi-structural model is used to assess the domestic economic effects of changes in interest rates, asset prices and credit spreads.footnote [8] This model predicts the responses of households, corporations, and banks to changes in asset prices and yields, allowing for a more disaggregated and granular analysis than could be done using COMPASS alone. This supplementary sectoral model can be used for analysing the effects of unconventional monetary policies (such as asset purchases) as well as of more standard policies.

Although incorporating diverse sources of information and analysis should in principle improve forecast accuracy, the process of putting the various inputs to the forecast together has become increasingly complex and consumes a high fraction of staff energy and attention. Steps that ideally would be executed automatically are instead done manually. The high priority assigned to producing the current forecast in time for the next policy meeting reduces the staff time available for longer-term projects, including improving the data and software infrastructure and the maintenance and development of forecasting models and methods.

Conditioning assumptions

Importantly, the MPC’s forecast does not necessarily represent the Committee’s best guess of what will actually happen to the economy. It is instead a conditional forecast, as defined in Part I: the MPC’s outlook conditional on a set of variables following exogenously given paths (not necessarily the paths that the Committee thinks most likely). The key conditioning variables in MPC forecasts, which are regularly set out and explained in the MPR, are: (i) the future path of short-term interest rates (as implied by the market curve, or, in an alternative simulation, by the assumption that short-term interest rates will remain constant at the current level); (ii) government spending, taxation, and other fiscal policies as announced in the most recent official statements; (iii) the exchange rate between the pound and other currencies (weighted by their shares of trade with the UK), which is assumed to follow a path halfway between that implied by international interest rate differentials and the current exchange rate; and (iv) energy prices, as implied by futures prices of oil and natural gas.footnote [9] Some consequences of using externally determined conditioning assumptions rather than assumptions freely chosen by the MPC are discussed in Part IV.

Besides the forecasts of headline variables (inflation, growth, unemployment), the MPR also provides three-year forecasts for a list of subsidiary variables, including variables treated as exogenous to the UK forecast (such as world GDP) and endogenous variables of interest derived as part of the forecast (such as average weekly earnings and consumer spending). In practice, in its communications with the public the MPC focus on the headline variables, although other variables (eg, wages) also receive attention.

The quarterly forecast process

The forecasts of the Bank of England are often described as MPC-owned but staff-led. The professional staff do most of the preparatory work, in formal and informal consultation with MPC members, and they are responsible for drafting the descriptions of the economic situation and the outlook that appear in the MPR. MPC members provide feedback, add their own judgements, and, when they are satisfied, approve the forecast and the staff’s description of it in the MPR. As the forecast is supposed to represent the best collective judgement of the Committee, its approval involves discussion and negotiation among MPC members. In practice, MPC members may differ from the collective forecast at least to some degree, but members appreciate that differences of view are normal and even desirable. In any case, there is no formal mechanism for dissent from the forecast per se.footnote [10] Members with differing outlooks have ample opportunity to present their views in public fora including speeches and testimony at the Treasury Select Committee.

Broadly speaking, the development of the quarterly forecast follows a regular sequence (see Figure 2 for the sequence from the November 2023 round). Although informal discussions among staff and between staff and MPC members occur on an ongoing basis, the release of the Quarterly National Accounts data by the Office for National Statistics, approximately five to six weeks before the Bank’s policy meeting and the subsequent publication of the forecast, marks the official beginning of the new forecast round. The staff begin the process by updating the previous forecast to reflect new information, primarily economic and financial data that have become available over the previous quarter, including changes in the paths of the conditioning variables. Information derived by periodic staff ‘stock-takes’, for example about productivity and the supply side of the economy, may also be incorporated at this stage.

Figure 2: Timeline for the November 2023 forecast round

Shows the 5-week forecast process beginning with the ONS release of quarterly national accounts data, with the majority of staff forecast meetings and milestones taking place while new ONS data is being released in weeks 1-3. The majority of MPC forecast meetings and milestones take place during a particularly busy period from the middle of week 3 to the end of week 4. Formal policy meetings start in week 4, with discussion meetings following the closure of the 15 day asset pricing window, and conclude with MPC sign off in the middle of week 5. The process ends with MPR publication towards the end of week 5.

About five weeks before publication, staff convene for the ‘constraints meeting’. At this meeting the staff discuss updates to the forecast in progress. Much of the discussion at this meeting concerns the likely outcomes for the current (not yet completed) quarter and the following quarter. Staff estimate the values of key variables for those quarters using, for the most part, either data already in hand (including data that are preliminary and/or ‘high-frequency’, eg, released daily, weekly, or monthly) or short-term forecasts, combined with judgement. For example, the near-term inflation rate is constructed by aggregating individual forecasts for the prices of various goods and services, taking into account additional information about wages and input costs based on current official and survey data. The current and subsequent quarter are called the ‘constraint quarters’ (hence the name of the meeting) because the final forecast is constrained to match the output of the short-term forecasting exercise for those quarters.

The UK is a small open economy which is sensitive to developments abroad. Accordingly, forecasts for economic activity and inflation in the global economy and in major economic regions (the euro area, the United States, China, other emerging markets) are important inputs to the UK forecast. The forecasts for foreign activity and inflation are consequently constructed early in the process (with updates as needed) by a separate staff group (the International Directorate) responsible for monitoring international developments. The regional forecasts draw from a range of sources, including official economic and financial data, information and forecasts provided by international agencies such as the International Monetary Fund, small economic models, and the judgements of staff, which include specialists in the economies of each major region. The outputs of this process most relevant to the UK forecast are the expected global demand for UK exports, world GDP weighted by UK trade, and a measure of world export prices.

The constraints meeting is followed by staff meetings at which a provisional medium-term forecast is fleshed out and possible changes to staff and MPC judgements that might be incorporated into the forecast are discussed. Staff characterised this process to us as being top-down, that is, new information about medium-term developments is layered incrementally on top of the forecast produced in the prior iteration. In the second or third week of the process (three to four weeks before the policy meeting), members of the MPC join the staff for a Benchmark Meeting, during which the staff present the provisional forecast, discuss key assumptions incorporated in the forecast, and suggest issues that may deserve further discussion. The meeting includes a presentation by the staff responsible for the international forecast. MPC members provide feedback, which may result in supplementary analyses and/or tweaks to the forecast by the staff.

Again, judgemental modifications to model outputs are important throughout the forecast process. Staff make judgements when constructing the benchmark forecast, discussing those that are more controversial or consequential with the MPC. MPC members add their own judgements, which have played an important role in forecasts in recent years. For example, based on their observation of the economy and analysis of previous forecast errors, MPC members have come to believe that the second-round effects of inflation on wage growth are currently larger and more persistent than those captured by the models, and they have accordingly modified the forecasted profiles for inflation and nominal wage growth. Judgements are cumulative, that is, quantitative judgements made in each round are typically added to judgements from earlier rounds. Staff keep records of judgements made or modified from round to round, and they may suggest modifications or additions to MPC judgements based on their own analyses.

The MPC’s judgemental adjustments to forecasts are related to but not identical to so-called Key Judgements. Key Judgements are assessments made collectively by MPC members, with staff input, that lay out the Committee’s preferred narratives or interpretations of the outlook and are thus potentially useful as communications devices. Qualitative descriptions of the MPC’s Key Judgements are featured prominently in the MPR.

Following the Benchmark Meeting, the forecast is refined in a series of meetings between staff and MPC members, known as Key Issues Meetings. The joint MPC-staff meetings include presentations and discussions of staff analyses of issues bearing on the forecast, including discussions of monetary policy strategy. The forecast must also be updated to reflect changes in the conditioning assumptions (eg, changes in the market path of short-term rates, from which the forecast’s assumptions about the course of the policy rate are derived). By the fifth week of the forecast round, about one week before the policy meeting, the MPC and staff are ready to hold official draft meetings, at which staff present and MPC members discuss the near-final forecast. During the week prior to the policy decision the MPC signs off on the forecast and on the accompanying text of the MPR.

Although most of the meeting time running up to the policy decision is devoted to the central forecast and supporting analyses, the staff and MPC also work together to develop fan charts, which supplement the main forecast in the MPR (see Box 1). The purpose of the fan charts is to provide the public with information about the range of uncertainty around the forecasts of inflation, GDP growth, and unemployment over the next three years. A separate fan chart is constructed for each of these variables.

Except when the schedule is affected by international meetings and the like, the formal policy decision is deliberated in MPC meetings on a Friday and the following Tuesday, with the official vote on policy taken on Wednesday. On Thursday the decision and the MPR (including the forecast) are released, along with a statement and the minutes of the policy meeting. The Governor and the Deputy Governors for Monetary Policy and Markets and Banking hold a press conference and sit for additional media interviews to provide additional information and context to the decision.

Box 1: Constructing fan charts

The MPR contains fan charts that summarise the uncertainty surrounding the forecasts of inflation, GDP growth, and unemployment. The fanlike shape of these charts reflects the fact that the range of uncertainty inherent in the forecasts broadens with the forecast horizon. The modal forecast for each variable lies in a band roughly in the centre of the respective fan chart. Coloured regions around the central line show the range of possible outcomes and the subjective probabilities that the actual outcomes will fall within each range. For example, a relatively narrow band around the modal forecast is expected to include the realised path of the forecasted variable, in the judgement of the staff and MPC, about 30% of the time. A band corresponding to the full width of the fan chart is expected to include the realised path of the forecasted variable 90 out of 100 times.

The empirical basis for the fan charts, in the first instance, is the historical record of the Bank’s forecast errors (since 2004, with 2020 errors downweighted) for each variable. A wider fan chart corresponds to a history of larger forecast errors on average and, accordingly, a presumption that the future course of the variable is more uncertain.

As with the central forecast, the MPC is given the opportunity to add judgements to the fan charts. (MPC members may decline to add judgements to the charts or may simply retain judgements made in earlier forecast rounds.) These possible judgements are of two types. First, MPC members may feel that the current level of uncertainty about the outlook is greater or less than implied by historical forecast errors, leading them to broaden or narrow one or more fan charts accordingly. This type of adjustment appears to occur only rarely. Second, more frequently, members may feel that the probability of a particular variable being (for example) higher than the modal forecast is greater than the probability of its being lower than forecast. That is, there is ‘upside risk’ to that particular variable. To capture unbalanced risks, the MPC may (in this example) judgementally reshape the distribution to add more weight above the modal path and subtract weight from below the modal path. Technically, the MPC adds skew to the distribution of outcomes represented by the fan. In general, a large skew (a significantly ‘lopsided’ fan chart) suggests that the Committee puts considerable weight on the possibility that the realised path of the forecasted variable will be above (rightward skew) or below (leftward skew) the modal forecast.

For each of the variables with associated fan charts, the assumed skew permits the calculation of mean (average or expected) forecast values, defined as the sum of possible outcomes weighted by the probability of their occurrence.footnote [11] For a variable with (say) upside risks, and thus a rightward skew, the mean forecast exceeds the modal forecast, and vice versa. The MPR and its supplementary materials currently include both estimated mean and modal forecasts for inflation, growth, and unemployment.

The forecast and public communications

The day after the policy decision during a monetary policy round where a Monetary Policy Report is published has become known as ‘super Thursday’, because of the volume of materials released to the public along with the announcement of the policy decision.footnote [12] The key materials released on super Thursday include the Monetary Policy Summary, the Monetary Policy Report, and the minutes of the policy meeting. The MPC’s economic forecast typically plays a central role in all of these materials, as well as in the press conference, interviews, and media coverage that follow the announcement.

Monetary Policy Summary

The Monetary Policy Summary has evolved somewhat over time, but in its current form it begins by reporting the decision, the vote, and the policy preferences (eg, for a rate increase instead of no change) of the members who voted against the Committee action. It notes that the updated economic projections are published in the MPR and states the conditioning assumptions underlying the forecast.

Following this brief introduction comes a review of recent developments in the UK economy, typically including the evolution of GDP, labour markets, pay growth, and, of course, the Bank’s target, CPI inflation. Some quantitative and qualitative discussion of the outlook comes next. The modal forecast for inflation is given, as is the rough date at which inflation is expected to return to target. The statement summarises the MPC’s views of the risks to the outlook and may provide a mean inflation forecast (which, as discussed in Box 1, is greater than the modal inflation forecast when the risks to inflation are to the upside, and vice versa). The statement may also note how the forecast changes in the alternative scenario in which Bank Rate is assumed to remain constant instead of following the path implied by market prices.

Qualitative forward guidance about future policy, if any, is given at the end of the policy summary. For example, the summary of the November 2023 meeting, which I attended, ends by noting that ‘policy is likely to need to be restrictive for an extended period of time’ and that ‘[f]urther tightening in monetary policy would be required if there were evidence of more persistent inflationary pressures.’ The goal of this guidance is presumably to shape market expectations for the policy rate over the next few quarters in a way that minimises market volatility and aligns market pricing and financial conditions more generally with the achievement of the MPC’s economic objectives.

In issuing a summary statement after its policy decision, the Bank is following standard practice for monetary policy makers. However, the Bank’s Monetary Policy Summary offers significantly more, and more quantitative, detail than those of most other central banks on both recent economic developments and the economic forecast. This additional detail results in a policy statement that is longer and denser than those of peer central banks. For example, the Bank’s Monetary Policy Summary for the November 2023 meeting and the three previous meetings in which summaries were released had an average word count of 920, which is about double that of the other central banks we considered.footnote [13]

Monetary Policy Report

Of the Bank’s communication vehicles, the quarterly Monetary Policy Report (MPR) provides the most comprehensive overview of the state of the economy. The MPR is similar to publications released by most but not all of the peer central banks, which are typically also published on a quarterly schedule.footnote [14] The MPR begins by reprinting the Monetary Policy Summary, discussed above. Following the Summary, the two principal sections of the Report provide further information on (1) the outlook for the UK economy and the risks to that outlook; and (2) current economic conditions.

The section on the outlook displays a table (Table 1.A) showing the MPC’s economic forecasts for the next quarter and subsequent three years of GDP, inflation (both modal and mean values), the unemployment rate, and the degree of excess supply or excess demand (that is, the aggregate output gap). The forecast’s conditioning assumptions, including the market-implied path for Bank Rate, the policy rate, are again clearly stated, discussed, and summarised in a table (Table 1.B). Forecasts for subsidiary economic variables (eg, household consumption, business investment, and housing investment), prepared by the staff to be consistent with the forecasts for the headline variables, are also presented in a table and discussed in the text. The alternative forecast conditional on the assumption that the policy rate will remain constant is also presented. Supplementary tables that include detailed forecasts for a longer list of variables and other quantitative information are available for download.

The outlook section of the Report includes qualitative discussions of the MPC’s Key Judgements, which, as discussed earlier, are narratives that help shape the Committee’s outlook and the official forecast. To illustrate the degree of risk and uncertainty, the fan charts for GDP, unemployment, and inflation are displayed. The risks to each major variable are also described qualitatively and discussed.

The second major section of the MPR reviews current economic conditions, although it also contains some forward-looking material (eg, the implications of wage growth or current rates of producer price inflation for consumer price inflation in the next few quarters). This section begins with an overview of the global economy and financial markets, reflecting their important influence on the UK economy. Developments in market interest rates and credit conditions and the impact of interest rates on activity are also discussed. The coverage of inflation includes, among other factors, an update on measures of inflation expectations. Central bankers generally agree that keeping medium-term inflation expectations near target, which tends to moderate the aggressiveness of firms’ pricing and workers’ wage demands, is important for gaining control of realised inflation.

Boxes and annexes, either within or following the two main sections, contain other information and analyses. Regular features of the MPR include a review of monetary policy developments since the last meeting, an update of the reports of the Bank’s Agents on business conditions, and a summary of the outlooks of external forecasters drawn from a regular survey. About once a year the MPR also includes an annex on how the economy has evolved recently relative to the MPC’s projections. The discussion of recent forecast misses is supplemented by figures comparing earlier projections of key variables with the outturns. Usefully, it typically distinguishes recent forecast errors caused by unexpected movements in conditioning variables from errors arising from other factors, eg, a misreading by the MPC of the links between unemployment and wage growth.

The Report also routinely includes chapters and boxes on special topics. ‘In focus’ chapters take a deeper look at aspects of current economic developments or the outlook, eg, the outlook for inflation (May 2023), factors affecting aggregate supply (February 2023), developments in the labour market (August 2022), and the effects on the economy of the rise in energy prices (May 2022). Other, occasional boxes provide short reviews of particular issues of interest, eg, quantitative tightening (August 2023), the cash flow channel of monetary policy (May 2023), and international comparisons of the behaviour of inflation (August 2022).

Minutes

Following another recommendation of Warsh (2014), the minutes of the policy meeting are published together with the Monetary Policy Summary and the MPR (in relevant months) on the day of the policy announcement. Most central banks publish minutes only with some weeks’ delay. The Bank appears to have accepted the argument that the logistical difficulties of producing minutes within a day after the multi-day meeting concludes are compensated for by avoiding an additional market-moving announcement some weeks after the policy decision, and perhaps by eliminating the risk of leaks in the interim period. Quick publication of the minutes also provides useful context for the Governor’s press conference.

The minutes usually begin with a summary of the MPC’s discussion of current developments and the outlook, including the global economic outlook, financial market developments, and the outlook for UK growth, unemployment, and inflation. The minutes also typically report both the future rate path implied by market pricing and that given by the median respondent in the Bank’s Market Participants Survey.

This is followed by a further review of economic developments and the outlook. In the spirit of risk management, the minutes report the discussion of the risks to the outlook, and the mean (as opposed to modal) projections for inflation may be repeated. Much of the material to this point (except possibly the results of the Market Participants Survey) repeats or summarises material available in other documents.

The minutes then delve into the debate around the policy decision. The MPC, where is it not uncommon for several members to cast public votes against the Committee’s policy action, is less consensus-oriented than policy committees at, say, the Federal Reserve or the European Central Bank. MPC members who vote in the minority are asked to state the action they would have preferred, and a precis of the majority and minority rationales for their preferences regarding the policy decision (without specific attribution to individuals) is presented.

Both the summary of the debate and a subsequent paragraph reflecting the views of the MPC as a whole may contain forward-looking views or guidance about the future evolution of policy, guidance that is typically repeated in the Monetary Policy Summary and at the post-meeting press conference. In contrast to the Monetary Policy Summary, supporters and opponents of the policy action are named, along with the preferred policy action (but not the individual rationales) of opponents, reflecting the requirement of individual responsibility. Interviewees from the media and financial markets told us that laying out the pros and cons of the action and providing forward-looking guidance were highly useful.

A section of the minutes on ‘operational considerations’ has in recent years included information about changes in the total stock of assets held by the Bank for monetary purposes (quantitative easing or tightening).

Box 2: Comparisons to forecasting procedures at peer central banks

To gain perspective on forecasting at the Bank of England, we compared forecasting processes at the Bank with those of six other central banks, selected because of their global importance (the Federal Reserve, the European Central Bank) or because, like the Bank of England, they are inflation-targeting central banks making policy for advanced but comparatively small open economies: the Swedish Riksbank, the Norwegian Norges Bank, the Bank of Canada, and the Reserve Bank of New Zealand. (We excluded the Bank of Japan because both the recent experience and the institutional structure of the Japanese economy are quite different from the other central banks we consider.) Our comparisons are based on interviews with staff at the various banks and on our reading of public documents.footnote [15]

Forecasting procedures at the peer banks share many features with those of the Bank of England. Most forecast headline economic variables, such as growth, inflation, and labour market indicators, and many forecast global economic developments as well. Notably, however, published forecasts by peer banks are typically much less detailed and are less prominently featured in public communication than those made by the Bank.

At the majority of the peer banks, including the Riksbank, the Norges Bank, the Reserve Bank of Canada, and the Reserve Bank of New Zealand, policymakers formally sign off on the forecast and provide input to staff on its development. However, our sense from interviews was that, at these banks, the involvement of policymakers in forecast construction is generally less extensive than at the Bank of England. For example, at the Bank of Canada the staff present a complete forecast to the Governing Council; based on the staff forecast (which is not published) and additional information received prior to the meeting, the Council publishes its own consensus forecast in its Monetary Policy Report. Forecasts at the peer central banks also differ in the conditioning assumptions used (see Box 3 below). In particular, several central banks publish their own forecasts of the policy rate, rather than taking market-based forecasts of rates as a conditioning assumption.

At the European Central Bank and the US Federal Reserve, the staff produce forecasts with little or no policymaker input. Probably this practice at the ECB and the Fed is accounted for by the fact that the policy committees at both of these central banks are large and geographically dispersed, making meaningful policymaker involvement in forecast development difficult. The ECB publishes its staff forecast, and it is an important part of the bank’s communication; however, ECB policymakers do not formally sign off on the forecast and make their own risk assessments. The Fed does not publish the staff forecast (except when policy meeting transcripts are released, with a five-year lag), using it only internally in policy deliberations.footnote [16] Instead, the Fed releases the (anonymous) individual projections of key macroeconomic variables by participants of the Federal Open Market Committee (FOMC) in a quarterly publication called the Summary of Economic Projections. These projections do not reflect the official views of the FOMC as a whole, a fact that is sometimes a source of public confusion, but the policymaker projections are typically strongly influenced by the staff forecast (circulated in advance of the policy meeting) and the work of staff economists at the regional Federal Reserve Banks.

Like the Bank of England, in their forecast development all of the six peer central banks combine formal modelling techniques, including economic models and statistical tools, with substantial judgement.footnote [17] Some peer banks use semi-structural models, like the Federal Reserve’s FRBUS or the Bank of Canada’s LENS model, as their benchmark models. However, the Norges Bank and the Reserve Bank of New Zealand use DSGE models, analogous to the Bank’s COMPASS model, as their central models, and other banks use DSGE models to cross-check the forecasts of their central models and, in some cases, for policy analysis. Most peer banks also use a suite of supplementary or sectoral models, as well as purely statistical models, in constructing their forecasts. Most also use atheoretical techniques, based on currently available data, to ‘nowcast’ current and near-term conditions.

Most of the peer banks do not produce fan charts. The European Central Bank’s staff forecast includes fan charts for GDP growth and inflation (both headline and core). The ECB’s fan charts are symmetrical and are based on past projection errors. The Norges Bank has recently used a statistical method to estimate the uncertainty around projections of output, inflation, and house prices. The FOMC’s survey of participants asks them for directional assessments of the risks to headline macro variables, as well as of general uncertainty, but the resulting charts do not purport to show the distribution of outcomes of those variables. The Riksbank, the Bank of Canada, and the Reserve Bank of New Zealand do not currently publish fan charts. However, all central banks discuss risk and uncertainty in qualitative terms, often in their post-meeting monetary policy statements.

Part III: Comparisons of forecast accuracy

As we have seen, the forecasting process at central banks has many purposes: if well done, it can help policymakers make good decisions by providing an opportunity for systematic review of economic developments and issues and for assessing the likely impacts of alternative policy choices. The forecast can likewise be a useful tool for clarifying policymakers’ current views of the economy and explaining them to the public. Even inaccurate forecasts, if they reflect the best available economic information and are thoughtfully adjusted when new information arrives, can help increase the coherence and predictability of policy. Nevertheless, accurate forecasts are obviously important, both as a guide to policy and because forecast accuracy can be an indicator of how well staff and policymakers understand developments in the economy.

This review is prospective, aimed at strengthening the Bank of England’s forecasting process and its use of forecasts going forward. Nevertheless, a look at the historical record is useful. In that spirit, this section briefly compares the Bank’s forecast accuracy since 2015 with that of the other six central banks in our comparison set (the Bank of Canada, the US Federal Reserve, Norway’s Norges Bank, the Reserve Bank of New Zealand, the Swedish Riksbank, and the European Central Bank) as well as with that of external forecasters.footnote [18] We focus primarily on one year ahead forecasts of inflation, GDP growth, and unemployment. We also consider what we call here ‘one quarter ahead’ forecasts, which are estimates covering the year extending from three quarters prior to the date of the forecast to one quarter beyond, and thus should perhaps be more properly called ‘nowcasts’. We end this part with a brief examination of the responses of monetary policy at the seven central banks to the recent inflation.

To anticipate the results, we confirm that the forecasting performance of all the central banks in our study, as well as that of external forecasters, deteriorated significantly with the onset of the pandemic and the subsequent inflation. The Bank of England suffered the common deterioration in forecast accuracy, but we find that, overall, its record is generally in the middle of the pack, and its policy response to recent developments, as indicated by changes in its policy rate, was also qualitatively similar to that of the other central banks. There appears therefore to be little basis for singling out the Bank from its peers for criticism.footnote [19] At the same time, the marked decline in forecasting performance by all central banks (and other forecasters) provides strong motivation for reviewing the forecasting processes and the use of forecasts at all these entities, including the Bank.

Before proceeding, we should acknowledge that an entirely objective comparison of central banks’ forecasting records since 2020 or so is probably not possible. As already noted, in recent years the global economy has faced a series of large shocks. The pandemic shut down businesses and schools for extended periods, even in the absence of government-ordered lockdowns. Supply chains were disrupted during and following the pandemic, as indicated by a 4.3 standard deviation increase (relative to the pre-pandemic average) in the New York Fed’s index of global supply-chain pressures. Oil and natural gas prices rose sharply, especially following Russia’s invasion of Ukraine, with global oil prices rising by about 150% between the start of 2021 and June 2022, and UK gas prices rising by 520% between the beginning of 2021 and August 2022. Food prices also spiked along with the prices of grain and other critical commodities.

Importantly for our purposes, these shocks had different effects on different economies (eg, increases in natural gas prices had more severe effects on the UK and the euro area than on the US), and individual economies faced idiosyncratic shocks (and different government responses) as well. Differences in the mix and size of shocks hitting each economy would have affected the difficulty of making accurate forecasts for that economy and consequently the validity of international comparisons. Moreover, national definitions of the key forecasted variables differ in important details (eg unlike the Fed’s measure, the ECB’s inflation measure does not include the imputed rents of owner-occupied housing).

Another consideration in making comparisons is that the conventions governing the timing of central bank forecasts are also not uniform. In particular, the timing conventions at the ECB and the central banks of Norway, Sweden, and New Zealand are broadly consistent with the Bank of England’s approach of providing quarterly forecasts for the three-year period following the date of the forecast. We can thus compare the forecasts of these central banks to those of the Bank without additional adjustment for timing differences. In contrast, the Bank of Canada and the Federal Reserve publish quarterly forecasts for growth and inflation (measured from the fourth quarter of one year to the fourth quarter of the next year) for both the current year and the next two calendar years (eg, Fed forecasts made in March, June, September, and December 2020 all apply to the years ending in 2020 Q4, 2021 Q4, and 2022 Q4).

The Bank of Canada kindly provided us one year ahead forecasts for inflation that use a timing convention similar to that of the Bank of England, and we use that data in the relevant figure and table below.footnote [20] Otherwise, to roughly align timing between the Bank of England’s medium-term forecasts and those of the Federal Reserve and the Bank of Canada, we used year ahead forecasts for Q4 published in December (for the Federal Reserve) and January (for the Bank of Canada). Application of this procedure produces a roughly apples to apples comparison but allows for only one forecast comparison per year, not enough to draw strong conclusions. In what follows, we include forecast series for which we have only partial data in the relevant figures but exclude them from the statistical comparisons. As the Federal Reserve does not publish one quarter ahead forecasts, we also exclude the Fed from all comparisons of one quarter ahead forecast accuracy.

For each of the central banks in our comparison set, and with considerable help from their respective staffs, we gathered each bank’s real-time forecasts and the corresponding outturns for the three headline variables noted above: inflation, real GDP growth, and the unemployment rate (where available).footnote [21] Revisions of the data as of 23 January 2024 were used where applicable. We chose the inflation rate corresponding to the official target rate of each central bank. A more complete description of the data and their sources can be found in Annex B.

Inflation

After years of relative stability, inflation around the world began what would prove to be a sustained rise in 2021, peaking in mid-2022. Figure 3 below shows actual inflation rates since 2015 faced by the seven central banks in our comparison set. As is clear from the figure, the recent inflation experience was broadly similar across countries. The UK suffered the highest peak inflation (with Sweden and the euro area not far behind), but, like the other economies, has seen rapid disinflation since mid-2022.

Figure 3: Annual inflation rates, 2015–23

The chart shows inflation rates for the UK and the 6 peer central banks fluctuating between 0% and 3% in the period before 2020, before spiking up across all countries from the beginning of 2021 with the UK experiencing inflation around 10%, similar to Sweden and euro area, but above the other peers.

To what extent was the surge in global inflation anticipated? Figure 4 shows one year ahead inflation forecast errors for all seven central banks through 2023 Q3. The figure is followed by a table showing, for various subperiods, the root mean squared forecast errors (RMSEs) for the Bank of England and the four other central banks that share the Bank’s forecast timing conventions, plus the Bank of Canada (which provided comparable data). A higher RMSE corresponds to larger errors and less accurate forecasts. Available (annual) observations from the Federal Reserve are included in Figure 4, but because of the differences in forecast timing conventions noted earlier, the Fed is excluded from the RMSE table.

Figure 4: Inflation, one year ahead forecast errors, 2015–23

The chart shows all central banks sampled experiencing inflation forecast errors between -2 and +2 percentage points from 2015 until 2020. After 2020 all see spikes in forecast errors, all above 4 percentage points. Errors reduce across the sampled central banks from the backend of 2022.

Table 1: RMSEs, one year ahead inflation forecasts

Period

BOE

CPI inflation

ECB

HICP inflation

Riksbank

CPIF inflation

Bank of Canada

CPI inflation

Norges Bank

CPI inflation

RBNZ

CPI inflation

2015–19

0.64

0.67

0.38

0.41

1.02

0.65

2020 Q1–2021 Q1

1.02

0.96

1.12

1.30

0.71

0.46

2021 Q2–2023 Q3

4.60

4.99

5.01

3.07

3.63

4.22

Figure 4 and Table 1 both show that all the central banks whose records we were able to compare did reasonably well in forecasting one year ahead inflation in 2015 through 2019, a period of relatively stable inflation, and even during 2020, the first year of the pandemic. However, the surge in inflation that began in mid-2021 was largely, though not entirely, unanticipated by all the central banks. Based on RMSEs, the Bank of England’s inflation forecasts were neither the worst nor the best of the central banks shown in Table 1. During the critical 2021 Q2–2023 Q3 period, when inflation was most extreme, the Bank’s inflation forecasts, as quantified by the RMSE metric and with the caveats given earlier, were better than those of the ECB and the Riksbank but worse than those of the Bank of Canada, the Norges Bank, and the Reserve Bank of New Zealand.footnote [22]

Figure 5 shows one quarter ahead inflation forecast errors (forecasts are for annual, not quarterly inflation) for the comparison group of central banks (excluding the Federal Reserve), and Table 2 below shows the corresponding RMSEs.

Figure 5: Inflation, one quarter ahead forecast errors, 2015–23

The charts shows inflation forecast errors range between -1 and +1 percentage points from 2015 until 2020. Errors are much more volatile thereafter, with central banks generally missing above until late 2020. The UK stands out for a large forecast miss in late 2022. Errors stabilise in 2023.

Table 2: RMSEs, one quarter ahead inflation forecasts

Period

BOE

CPI inflation

ECB

HICP inflation

Riksbank

CPIF inflation

Bank of Canada

CPI inflation

Norges Bank

CPI inflation

RBNZ

CPI inflation

2015–19

0.20

0.31

0.25

0.23

0.37

0.40

2020 Q1–2021 Q1

0.38

0.46

0.81

0.42

0.71

0.56

2021 Q2–2023 Q3

1.23

1.38

1.76

0.70

1.12

1.27

Recall that at the Bank of England, as at most central banks, very short-term forecasts are constructed mostly by extrapolation of available data and various statistical models, to which the longer-term forecast is forced to conform. Due to their backward-looking component, these forecasts also require anticipation of revisions to past data. One quarter ahead forecasts are thus not particularly representative of central banks’ forecasting processes in general. That said, by the RMSE criterion the Bank’s one quarter ahead inflation forecasts were the most accurate of the six central banks in 2015–19 and in 2020 Q1–2021 Q1, and, despite the large inflation miss in 2022 Q4 (see Figure 5), they were similar to those of the other banks during the most recent period.footnote [23]

GDP growth

Like inflation, GDP growth became much more variable during the pandemic period. Figure 6 shows the wide swings in output since 2020 in all the represented economies, with the fluctuations in the UK being especially large. Figure 7 depicts the one year ahead forecast errors of the seven central banks in our comparison set, and Table 3 shows the comparative accuracy of five banks (excluding the Fed and the Bank of Canada, for whom we do not have complete data) by the RMSE criterion.

Figure 6: Four-quarter GDP growth rates, 2015–23

The chart shows positive GDP growth between 0 and 5% for all 7 countries sampled until a decline in late 2019. There is a marked trough in early 2020, before a recovery with growth peaking in 2021. Growth rates then gradually decline to low positive levels in late 2022.

Figure 7: Four-quarter GDP growth, one year ahead forecast errors, 2015–23

The chart shows very low forecast errors for GDP growth across all sampled countries until early 2020, when all have major misses to the downside. Errors continue to look volatile thereafter, until stabilising around late 2022.

Table 3: RMSEs, one year ahead forecasts of GDP growth

Period

BOE

GDP

ECB

GDP

Riksbank

GDP

Norges Bank

GDP

RBNZ

GDP

2015–19

0.85

1.68

1.01

0.76

0.82

2020 Q1–2021 Q1

12.95

7.23

4.44

5.25

5.74

2021 Q2–2023 Q3

3.75

4.99

1.96

0.99

4.05

We do not present a figure showing one quarter ahead forecast errors for GDP, but Table 4 presents the relevant statistical comparison. The Bank of Canada is now included.

Table 4: RMSEs, one quarter ahead forecasts of GDP growth

Period

BOE

GDP

ECB

GDP

Riksbank

GDP

Bank of Canada

GDP

Norges Bank

GDP

RBNZ

GDP

2015–19

0.66

1.67

1.28

0.75

0.53

1.14

2020 Q1–2021 Q1

10.97

9.11

6.18

2.91

3.89

7.18

2021 Q2–2023 Q3

2.76

4.84

1.92

0.99

0.82

1.93

At both the one-year and one-quarter horizons, the Bank unsurprisingly failed to forecast the extraordinary decline in activity in early 2020, following the arrival of the pandemic. The large miss elevated the Bank’s RMSE during 2020. Before and after 2020, however, the Bank’s one year ahead GDP forecasts are again in the middle of the pack (third of five in both periods), by the RMSE criterion. At the quarterly horizon, the Bank’s comparative record was second best of six in 2015–19 but fifth of six in the most recent period. In November 2022, the Bank forecast a mild but extended recession that did not in fact occur, a miss due in part to the forecast’s externally given conditioning assumptions (see Part IV).

Unemployment

Unemployment forecasts may be the most difficult to compare across countries as labour market institutions, government policies, and the methods of defining unemployment and collecting the relevant data are not uniform.footnote [24] Differences in policies were particularly stark during the pandemic. For example, the UK responded to the possibility of mass layoffs by instituting a furlough scheme, while the US mostly supported workers directly, through unemployment insurance and stimulus checks. The Swedish government largely avoided lockdowns although it provided labour market subsidies. With those caveats, Figure 8, which shows one year ahead forecasting errors, and the two accompanying tables below describe the performance of our comparison set of central banks in forecasting unemployment (the ECB and the Bank of Canada do not publish forecasts of unemployment).

Figure 8: Unemployment, one year ahead forecast errors, 2015–23

The chart shows low forecast errors for unemployment until late 2019. Norway, the US and Sweden see a marked spike, with large positive forecast errors peaking in 2020, before stabilising in mid to late 2021. The UK and New Zealand see a small increase in errors later in 2020, but then large negative misses in early 2021, before stabilising in 2022.

Table 5: RMSEs, one year ahead forecasts of unemployment

Period

BOE

Unemployment

Riksbank

Unemployment

Norges Bank

Unemployment

RBNZ

Unemployment

2015–19

0.55

0.56

0.32

0.41

2020 Q1–2021 Q1

0.84

2.12

2.85

0.59

2021 Q2–2023 Q3

1.55 (a)

0.83

0.42

2.03