Measuring climate-related financial risks using scenario analysis

Exploring how financial institutions can use scenario analysis to quantify climate change risks
Published on 17 April 2024

By Lewis Holden, Jordan King, Harriet Richards, Caspar Siegert and Lukasz Krebel.

Climate change – through both physical impacts and the effects of economies transitioning to net zero – poses financial risks in both the short and long term. These risks are relevant to a wide range of institutions across the financial system. This includes the Bank of England, given its significant financial operations. However, these risks are challenging to quantify, which could limit financial institutions’ abilities to mitigate against these risks into the future. This article explores how central banks and financial institutions can use scenario analysis to quantify these risks. It focuses on how financial institutions can ‘extend’ macro-climate scenarios to undertake granular asset-level analysis of financial risks, drawing on examples across sovereign bonds, corporate bonds and residential mortgages. It also discusses how scenario analysis outputs can be applied to financial institutions’ existing financial modelling toolkits. While scenario analysis applied in this way is at the cutting edge of climate financial risk measurement, it is still subject to a number of limitations.

1: Introduction

Climate and financial risk


Financial institutions are increasingly considering climate-related factors as part of their investment and/or lending decisions. Broadly, there are two – often complementary – reasons for this:

Reducing risk and/or maximising returns: Climate change will have macroeconomic and financial impacts, and therefore potentially impact financial asset values (eg Bank of England (2022), Network for Greening the Financial System (NGFS) (2023), Bank of England (2022)). Financial risks are generally larger for longer maturity assets. The value of these assets could re-price in an orderly way over time in response to crystallising physical and transition risks. More extreme losses could still materialise in the near term if investors sharply repriced assets in a ‘climate Minsky moment’ (eg CEPR (2022)). It is therefore valuable for investors, central banks and the broader financial system to focus on climate-related financial risks.

Net-zero alignment: There has been growing recognition of the role that the financial system adjusting financial flows can play in supporting the transition to net zero (eg Glasgow Financial Alliance for net zero). Investors that want to support the transition to net zero have therefore refocused investments towards assets more closely aligned with net zero, irrespective of risk-adjusted returns.

The Bank’s consideration of climate risks flows directly from its primary monetary and financial stability objectives (see Bank of England Act (1998)). To achieve its financial and monetary stability objectives the Bank undertakes financial operations, including collateralised credit operations, repurchase agreements and asset purchases. The full range of the Bank’s policy and balance sheet tools are set out in the Bank of England Market Operations Guide. The Bank also undertakes market operations as part of funding its activities and managing its own foreign currency reserves. As part of its financial risk management framework the Bank considers a range of risks to ensure it has the resources available to achieve its primary objectives. Climate risks are one of the risks the Bank considers, as set out in the Bank’s 2023 Climate Disclosure.

The rest of this article focuses only on climate as a financial risk given the Bank’s interest from a financial system-wide risk perspective, and the risks it is exposed to through its own financial operations.

The scenario analysis toolkit

The initial challenge that financial institutions face in looking to mitigate climate-related financial risks is measuring those risks. In the context of disclosures of climate-related risks and opportunities (eg Task Force on Climate Related Financial Disclosures (TCFD), International Sustainability Standards Board (ISSB)) various metrics for climate financial risks have been developed that can help with this (Box A).

As climate risk is increasingly considered a material financial risk (eg Financial Stability Board), it is important to quantify the impact it could have on assets and/or profit and loss. As described in Box A, backward and forward-looking metrics may provide a sense of climate-related financial risks, but they do not directly quantify those risks. This is the challenge that scenario analysis seeks to solve.

For this reason, this article focusses on the use of scenario analysis to measure climate financial risks.

Box A: Measuring climate financial risks

Backward-looking ‘proxy’ metrics

Backward-looking ‘proxy’ metrics include the core metrics recommended by the TCFD and ISSB disclosure frameworks, such as carbon intensity, carbon footprint and financed emissions. They are so-called ‘proxies’ because they only provide an indication of exposures to climate-related financial risks, but do not directly quantify financial exposures. They benefit from being simple to understand, with widespread availability of underlying data. However, they are subject to limitations:

  • they do not consider any actions that an issuer is planning to take to mitigate their exposure to climate related financial risks;
  • they do not consider the cost of these planned emissions reductions, or the costs of any additional actions that an issuer might take in response to future climate policies; and
  • it is challenging to translate qualitative proxy metrics into decision useful information that can be incorporated into risk management frameworks.

Forward-looking metrics

TCFD recommends the use of forward-looking metrics and scenario analysis as a tool to overcome some of the limitations described above. Forward-looking transition risk metrics include for example the decarbonisation targets of issuers, as well as the associated ‘Implied Temperature Rise’ (ITR) of an asset.footnote [1] Although forward-looking metrics address the limitation of backward-looking metrics not factoring in future decarbonisation plans, they are still limited by not directly measuring the financial risks an asset will be exposed to. However, metrics such as ITR can be useful for values-based investors to measure portfolio alignment with net zero. Forward looking physical risk metrics are particularly important given historical backward-looking data will underestimate future physical risks.

Scenario analysis metrics

Scenarios are hypothetical pathways – they are not forecasts or predictions. Instead, scenario pathways explore a range of plausible outcomes (TCFD 2023). Scenario analysis models the overall impact that a given top-down climate scenario has on the value of a financial asset by modelling the various transmission mechanisms between the projected scenario variables and the value of the asset in question.

The use of scenario analysis is particularly pertinent for climate-related financial risks given that standard risk modelling approaches (that may eg rely on historical default rates and/or asset price volatility) do not capture the unprecedented nature and uncertainty of climate-related risks.

2: Scenario analysis to measure climate-related financial risks


The Bank considers scenario analysis as a key tool in assessing the financial risks posed by climate change and uses scenario analysis to measure the climate-related financial risks it is exposed to through its own financial operations. To date, the Bank has focused scenario analysis on the three asset classes which it has historically been most exposed to through its financial operations – sovereign bonds, corporate bonds and residential mortgages. For example, the Bank of England Climate Disclosure 2023 incorporated scenario analysis to measure financial risks across all of these three asset classes, as well as detailing the Bank’s risk frameworks for managing the climate-financial risks associated with them.

The Bank supports the development of climate scenarios for central banks, supervisors and the financial sector through participation in the NGFS and international bodies. To support the Financial Policy Committee (FPC) and Prudential Regulation Committee (PRC) in the pursuit of their statutory objectives the Bank conducted the Climate Biennial Exploratory Scenario (CBES) in 2021–22, exploring climate-related financial risks for the largest UK banks and insurers, as well as the financial system more broadly. The Bank considers climate scenario analysis an important part of the risk assessment toolkit for supervised firms. As noted in the 2024 priorities letters to banksfootnote [2] and insurers, the Prudential Regulation Authority (PRA) expects firms to make further progress on climate scenario analysis and banks to consider the use of tailored and ambitious stress scenarios in relation to climate change impacts.footnote [3]

The primary challenge the article seeks to address is that macro-climate scenarios ‘off the shelf’ provide a starting point for analysis but generally do not provide the level of detail end users need to undertake asset-level financial risk analysis. Therefore, end users of scenarios generally need to ‘extend’ scenarios to apply them to the specific assets they are considering. This article firstly sets out further details on scenario analysis and this specific challenge, before explaining how the Bank is approaching this challenge with respect to sovereign bonds, corporate bonds and residential mortgages. Although examples relate primarily to the Bank, the learnings are relevant for other central banks, private sector financial institutions and the broader financial system.

Common climate scenarios

Climate scenarios can take different forms. Institutions exploring scenario analysis for the first time may start with qualitative narratives, rather than quantitative scenarios.

There are four quantitative macro scenarios which dominate the climate risk market: International Energy Agency (IEA) scenarios, Intergovernmental Panel on Climate Change scenarios, NGFS scenarios and the Principles for Responsible Investment scenario. Financial institutions can also develop their own, in-house quantitative scenarios, or combine approaches – ie develop their own scenarios within the broader ‘universe’ of a pre-existing scenario. This may help financial institutions develop scenarios relevant to their specific exposures and is important since all scenarios have limitations (see below).

This article focuses on the use of NGFS scenarios as they are specifically designed for the financial sector and are therefore well suited to assessing climate-related financial risks.

The NGFS scenarios consist of climate-related (eg temperatures, emissions) and macroeconomic (eg GDP and inflation) variables across different geographies. Transition risks are modelled using projections from Integrated Assessment Models (IAMs) of impacts of different policy ambitions on energy, emissions and land use, while physical risk models project acute and chronic physical risk impacts. Macroeconomic modelling of the consequences of both transition and physical risk on macro-financial variables is undertaken using the NiGEM model (NGFS (2023)).

Table A provides a full summary of the various NGFS scenarios which are referred to throughout this article.

Table A: Overview of NGFS scenarios



Net Zero 2050 (‘Orderly’)

Net Zero 2050 gives at least a 50% chance that global warming is limited to 1.5°C through stringent climate policies and innovation, reaching global net zero CO₂ emissions around 2050. Some sovereign issuers such as the USA, EU and Japan reach net zero for all greenhouse gases by this point. Therefore, physical risks are relatively low, but transition risks are relatively high.

Below 2oC (‘Orderly’)

Below 2°C gradually increases the stringency of climate policies, providing a 67% chance of limiting global warming to below 2°C. Global net zero CO2 emissions are achieved around 2070. Therefore, both physical and transition risks are relatively low.

Divergent Net Zero (‘Disorderly’)

Divergent Net Zero also gives at least a 50% chance that global warming is limited to 1.5oC, and also reaches net zero CO2 emissions globally by 2050. However, the transition comes with higher costs due to divergent policies introduced across sectors and a quicker phase out of fossil fuels. This leads to very large transition risks but limited physical risks.

Delayed Transition (‘Disorderly’)

Delayed Transition also provides a 67% chance of limiting global warming to below 2oC. However, it assumes global annual emissions do not decrease until 2030. Strong policies are then needed to limit warming to below 2°C. This leads to both large transition risks, and moderate physical risks.

Nationally Determined Contributions (NDCs) (‘Hothouse world’)

NDCs incorporates all policies governments have committed to, even if not yet implemented. The result is global warming of 2.6oC, associated with high physical risks but low transition risks.

Current Policies (‘Hothouse world’)

Current Policies assumes only currently implemented policies persist, leading to around 3oC of warming. Transition risks are therefore very limited, while physical risks are very high.


  • Source: NGFS (2022).

Scenario analysis – limitations

Climate scenario analysis remains a relatively novel tool, and scenario providers such as the NGFS continuously work to improve them by incorporating the latest state of climate knowledge, computational capabilities and user needs. Despite the significant progress made, climate scenarios continue to have a range of limitations including:

  • Scenarios do not capture the full suite of potential climate risks. For example, NGFS scenarios capture the effects of a subset of potential chronic and acute physical risks (NGFS (2023), NGFS (2024)). This limitation is being addressed as scenarios are refined over time to incorporate a broader range of climate risks.
  • Calibration of estimated chronic damages from temperature rises remains an area of ongoing research.footnote [4] For example, damage functions that describe the relationship between temperature increases and GDP may be predicated on historical relationships. This may underestimate the impacts of severe climate change, given that the increases in temperatures we have seen so far have been limited, and that the relationship may be highly non-linear (Institute and Faculty of Actuaries (2023)).
  • scenarios do not account for the effects of ‘tipping points’ which can significantly increase the effects of climate change (NGFS (2024)). They therefore should be seen as providing a robust but non-exhaustive estimate of potential damages from climate change.

While scenario narratives illustrate different levels of impacts under given assumptions and limitations outlined above, they do not necessarily represent the most likely or most severe potential outcomes for a given narrative. Given the above considerations, conducting scenario analysis requires being clear on the objectives of the exercise, and adjusting scenarios to suit these purposes. This should be clearly communicated alongside the assumptions and limitations of the analysis.

In choosing the scenarios to use for scenario analysis presented in this article, we have sought to select some of the more severe scenarios available to us. This is in line with the objective of prudent financial risk management – namely to measure potential financial losses in severe scenarios as currently modelled. For example:

  • Assessing transition risks to sovereign bonds using the most disorderly NGFS scenario (Divergent Net Zero).
  • Assessing physical risks to residential mortgages using the IPCC’s highest warming scenario, in which physical risks are larger than any of the NGFS scenarios.

Extending macro-climate scenarios to measure climate-related financial risks

Macro-climate scenarios typically provide broad overviews of climate developments. For example, NGFS scenarios project variables such as GDP, carbon prices and interest rates at the country or regional level. However, financial portfolios contain heterogenous assets distributed across various sectors, geographies, or industries. Each of these assets will be very differentially affected by changes in these macro variables. So, any analysis of financial risks must consider asset-specific risks to achieve a reasonable level of accuracy, even if financial institutions want to measure portfolio-level risks.

To use macro climate scenarios to measure asset-level climate-related financial risks, end users of scenarios must commonly develop toolkits to ‘extend’ between macro scenario variables and measurement of asset-level impacts. Key considerations in the development of these toolkits include:

Spatial granularity and resolution: Assessment of asset-level financial risks requires finer granularity than macro scenarios typically provide. This is a particular issue for consideration of physical climate risks. For example, assets within the same region may face different levels of climate risk based on circumstances such as their proximity to the sources of physical climate risks, such as waterways. End users may therefore need to develop toolkits to identify these granular asset-level impacts.

Related variables: While climate scenarios provide the paths for a large number of key variables (such as GDP), they will not provide a projection for every variable of interest (eg sovereign debt/GDP ratios). Users may need to derive related variables by relying on existing theoretical or empirical relationships between related variables and scenario projected variables.

Temporal misalignments: Macro climate scenarios generally project impacts over i) limited and ii) long time horizons. However, financial assets will have a broad range of maturities from months to years. End users may need to estimate effects at time horizons relevant to them and determine to what extent longer-term effects are relevant for the asset under consideration.

Intra-sectoral variabilities: Sectors and sub-sectors of the economy face different climate risks based on their differing exposure to transition and physical risks. But even within a sub-sector, the impacts will vary significantly across firms. For example, the industrials sector may be exposed to substantial transition risks, but specific companies within the sector may be well positioned to benefit from net-zero opportunities. Macro scenarios cannot reasonably be expected to distinguish between the impact that climate factors have on individual companies. Hence, end-users may need to develop toolkits to apportion sectoral impacts to the asset-level.

The following three sections of this article cover examples of how we have approached the challenge of ‘extending’ macro climate scenarios to undertake asset-level scenario analysis, which might also be useful for other financial institutions.

3: Applying macro scenario analysis: sovereign bonds


Outstanding sovereign debt as of end-2022 was more than $60 trillion and made up just over half of all outstanding debt securities globally (BIS, (2022)). Sovereign debt – particularly that issued by advanced economies – is seen as lower risk than other forms of debt. Because of this, and because sovereign debt markets are highly liquid, sovereign debt plays a particularly important role in the global financial system. For example, the pricing of other assets is often relative to a ‘risk-free’ sovereign asset, and sovereign bonds are widely accepted as collateral.

The largest proportion of the Bank’s financial assets are held in a separate legal vehicle known as the Bank of England Asset Purchase Facility Fund, indemnified by HM Treasury, to implement the MPC’s asset purchase programme. All of the portfolio is now Sterling UK Government bonds (gilts). Most of the Bank’s assets are therefore sovereign bonds. A full breakdown of the Bank’s sovereign asset portfolios can be found in Section 3 of the Bank’s 2023 Climate Disclosure.

Determinants of sovereign bond yields and prices

Sovereign bond yields (and therefore the price of sovereign bonds, which move inversely to yields) are a function of a complex interplay of determinants (eg Baker et al (2016), Afonso et al (2015), D’Agostino and Ehrmann (2014), Afonso et al (2012) and Poghosyan (2012)). They include both ‘fundamental’ determinants of yields (eg long run macroeconomic determinants), as well as more temporary drivers, such as changes in yields due to forced selling pressures in parts of the financial system (see Breeden (2022)). There is consensus of the following fundamental determinants influencing sovereign bond yields.

Expected policy rates: Investors generally target real returns. Therefore, if short-term interest rates that investors can receive in money markets (and that are driven by monetary policy) are higher, sovereign bonds become less attractive and their prices fall (eg Baker et al (2016), Poghosyan (2012)).

Sovereign credit risk premia: If investors assess that the risk of a sovereign issuer defaulting has increased, they will demand an additional risk premium on bond yields (Ghosh et al (2013), D’Agostino and Ehrmann (2014)). Investors’ assessment of a sovereign’s default risk depends on a range of factors, including governance, economic fundamentals and fiscal position.

Other risk premia: In addition to credit risk premia, investors will demand additional risk premia associated with other types of risk, such as liquidity premia and term premia (eg BIS (2018); Chaumont (2018)).

Climate-related factors can affect the first two determinants. The crystallisation of physical and transition climate risks could have macroeconomic and macrofinancial impacts – for example on inflation (eg Bank of England (2022)) and government spending and revenues (eg OBR (2023)). Climate risk transmission mechanisms on the third are less clear, and not considered further here.

Scenario analysis – Changes in ‘risk-free’ component of interest rates

What do macro-climate scenarios provide?

NGFS scenarios project inflation as well as both a short-term policy rate and a long-term (10-year) interest rate at the sovereign-issuer level for each year out to 2050. These projections are available for all the NGFS scenariosfootnote [5] (Table 2.A) and a baseline hypothetical scenario.footnote [6]

In more ‘disorderly’ transition scenarios, carbon prices are higher, which along with other macro effects, increases inflation. Changes in interest rates in these scenarios are a function of these inflationary effects – with central banks increasing policy rates to combat inflation. This means we see larger interest rate shocks in the most disorderly scenarios. In business-as-usual scenarios (eg Current Policies), there are no interest rate shocks (as carbon prices are equivalent to the baseline scenario).

‘Extending’ scenarios to undertake asset-level analysis

Macro scenarios provide projections for all major sovereigns, but only a short-term interest rate and a rate at the 10-year tenor. To provide a full analysis, it is necessary to extrapolate macro scenario projections across the whole yield curve to measure financial risks across a whole portfolio (a case of dealing with ‘temporal misalignment’).

This is because portfolios do not only consist of short-term debt and 10-year debt. They typically also include bonds with intermediate or much longer maturities. It is therefore necessary to interpolate and extrapolate within the NGFS scenarios to estimate how yields on bonds of other maturities change.

Interpolating between short-term and 10-year interest rates to obtain the interest rates for eg 5-year bonds is fairly straightforward. Extrapolating the yield curve to obtain eg the interest rate for a 30-year bond is slightly more complex. One practical way of doing so is to assume that investors have perfect foresight, and that the yield of a 30-year bond today is simply determined by the pathway of short-term interest rates in a given climate scenario over the next 30 years.footnote [7]

Scenario analysis – Changes in ‘credit risk’ component of interest rates

What do macro climate scenarios provide?

Debt/GDP ratios and sovereign credit ratings are widely used as measures of sovereign credit risk (eg Baker et al (2016), Afonso et al (2015)). The NGFS scenarios do not directly project these variables for different climate scenarios. However, the NGFS project some variables which can be used as constituent ‘inputs’ to construct how these metrics evolve in different climate scenarios. This includes GDP growth, government carbon tax revenues and interest rates (which determine the cost of servicing existing debt).

‘Extending’ scenarios to undertake asset-level analysis

End-users wishing to measure how debt/GDP ratios or sovereign credit ratings change over climate scenarios primarily need to overcome the ‘Related variables’ challenge to derive desired variables from macro scenarios.

When considering this for our portfolios, we have looked at a bottom-up approach to project changes in debt/GDP ratios based on economic relationships between debt/GDP ratios and variables for which macro scenarios do provide projections. This involves estimating models for the impact of climate change on the numerator (ie debt, through changes in government expenditure and tax receipts), while taking the denominator (ie GDP) from the macro scenarios.

There are a broad range of contributors to debt which could be impacted by climate change. For example Barrage (2023) focuses on physical climate risks and the impact on government healthcare costs, the IMF (2023) focus on carbon tax revenues and the cost of financing the transition, and the OBR (2023) considers a range of impacts on taxation, the cost of financing a green transition and secondary effects on debt interest. Each of these therefore consider a sub-set of transmission mechanisms or countries. In choosing the contributors to debt to consider, we have sought to balance the broad range of potential contributors across both physical and transition risks with the aim of producing comparable results across sovereign issuers. Box B describes the contributors used in developing debt/GDP projections for our illustrative analysis presented in this article. A full breakdown of data sources can be found in the Annex.

In this analysis, sovereign credit ratings have been projected to 2050 using a random forest machine learning model. Random forest machine learning models are widely used to predict credit ratings (eg De Moor et al (2018)) as they are able to capture inherent non-linearities in sovereign credit ratings (eg Klusak et al (2021)). The model has seven features, described below. Data sources are described in the Annex.

  • Debt/GDP ratio.
  • GDP per capita.
  • GDP growth.
  • Current account balance/GDP.
  • GDP growth volatility.
  • Whether the issuer is an EU member state or Japan.

The model is trained on sovereign ratings for advanced economy sovereign issuers between 1995 – 2022. Historical data for model features is taken from the World Bank for GDP growth and volatility, GDP per capita and current account balance/GDP.

Financial modelling

To derive shocked yield curves, we start with the ‘risk-free’ yield curves we have derived for each climate scenario above. We then additionally estimate the impact that changes in sovereign credit risk (through debt/GDP ratios and sovereign credit ratings) have on the credit risk component of sovereign yields.

This is an example of where historical empirical data can be used as proxies to enhance scenario analysis. As a starting point we assume that:

  • A 1 percentage point increase in debt/GDP increases sovereign bond yields by 1 basis point. There is significant heterogeneity in estimates of this effect historically and our estimate is at the lower end of the range (eg Ardagna et al (2007), Conway and Orr (2002), Baldacci & Kumar (2010)).
  • A 1 notch rating downgrade increases sovereign bond yields by 10 basis points. This is a mean of Afonso et al (2011) findings that a 1 notch ratings downgrade increases yields by 8 basis points and Gande and Parsley (2005) findings that it leads to an increase by 12 basis points.

Where a risk – for example an increase in debt/GDP – crystallises 18 years after the start of the scenario, the shock to the yield curve is applied to the 18-year tenor of the curve.

A total yield curve shock, which is the sum of the shock to the risk-free curve and the additional credit component, can then be computed. The scenario mean shocked yield curves for G7 issuers are shown in Chart 1.

These curves demonstrate the illustrative impacts of a ‘climate Minsky moment’ in the immediate future. That is, investors wake up to climate risks and factor the associated impacts on risk-free and credit risk components into sovereign bonds yields.footnote [8] This is a simplified assumption – investors would need to know with certainty which climate scenario we are in, and they would need to have perfect foresight on how this specific scenario will unfold. But it is useful to explore the potential impact of a sudden repricing.

The Net Zero 2050 and Divergent Net Zero transitions are associated with immediate, large shocks to the yield curve at shorter tenors. This reflects instantaneous shocks to the risk-free rate due to monetary policy responses to high carbon prices. Current Policies and NDCs scenarios are initially associated with limited shocks at shorter tenors, but larger shocks accrue at longer tenors as credit risks build due to increasing physical impacts. Across all scenarios, climate financial risks grow progressively larger at longer tenors.

Chart 1: Mean yield curve shocks

Losses are larger for portfolios with longer-dated maturities which lose greater than 9% of their value.


  • Source: Bank of England analysis. See the Annex for full breakdown of data sources.

These shocked yield curves can be combined with the existing financial risk toolkit to estimate impacts on the value of any given fixed income portfolio. This can be undertaken using industry-standard interest rate sensitivity analysis to reprice bonds as a function of their duration and convexity (eg Fixed Income Mathematics (2003)).

We have applied the stressed yield curves to synthetic portfolios of G7 debt (Chart 2). To construct these synthetic portfolios, the allocation to different issuers is weighted by each issuer’s outstanding debt. To demonstrate the impact of duration, we construct two different portfolios. The first assumes the average maturity of bonds in the portfolio is equal to the average maturity of each issuer’s actual outstanding debt. The majority of bonds in this portfolio have maturities less than eight years. The second contains longer average maturities with a mean maturity of circa 20 years.

For the actual maturity portfolio, most of the losses are driven by the shock to the ‘risk-free’ component, rather than sovereign credit risks. This is because shocks to risk-free rates are front-loaded in early years of the scenario and are only a function of transition risks, whilst credit risks accrue more gradually over the course of the scenario. Therefore, losses in a Current Policies scenario would be larger for longer maturity bonds and continue to build beyond the end of our scenario time horizon. The most stressful scenario for the actual maturities’ portfolio is Divergent Net Zero, where the portfolio loses circa10% of its value. This is because interest rates increase substantially in the near term for G7 issuers. While this is not necessarily problematic from a macroeconomic perspective, it has a meaningful impact on the value of a fixed-income portfolio.

Chart 2: Sovereign climate risk stress test of an indicative G7 portfolio (a)

The Below 2oC and Net Zero 2050 scenarios see the lowest mean increase in debt to GDP while Current Policies and Nationally Determined Contributions see the highest.


  • Sources: Bank of England analysis. Bloomberg Finance L.P. and Eikon by Refinitiv. See the Annex for full breakdown of data sources in underlying model.
  • (a) Portfolio weighted using each issuer’s outstanding sovereign debt and with maturity profiles similar to actual outstanding debt.

Losses are significantly larger for the portfolio with longer-dated maturities (Chart 3). This is because sovereign credit risk shocks are substantially larger at longer tenors, while shocks to the risk-free rate also persist. The longer-dated portfolio loses >9% of its value in all scenarios, and in Divergent Net Zero loses >20% of its value. The proportional increases for the Delayed Transition and Current Policies scenarios are particularly large. This is because sovereign credit risk shocks associated with the physical impacts of a Current Policies scenario take longer to start crystalising and will continue to build beyond the end of the scenario.

Chart 3: Sovereign climate risk stress test of a long-duration G7 portfolio (a)

Credit rating changes are negative across all scenarios with Current Policies and Nationally Determined Contributions seeing the largest downgrades.


  • Sources: Bank of England analysis. Bloomberg Finance L.P. and Eikon by Refinitiv. See the Annex for full breakdown of data sources in underlying model.
  • (a) Portfolio weighted using each issuer’s outstanding sovereign debt. Maturities randomly created from a uniform distribution where minimum maturities are circa 15 years and maximum circa 27 years.

Box B: Debt/GDP components modelled for scenario analysis

Gross Domestic Product (GDP)

We take annual GDP projections from the NGFS for each issuer between 2022 and 2050, which incorporate the effect of both transition and physical risks.

Debt – changes in tax revenues

As a starting point, we assume government expenditure and tax/GDP ratios remain the same as they are today across all scenarios. Therefore, in scenarios with poorer GDP performance, such as those where output is lower due to materialisation of physical impacts, foregone tax revenues are replaced by additional government borrowing.

Debt – carbon tax revenues

Governments generate carbon tax revenues through carbon pricing regimes. The higher the carbon price and emissions, the higher a government’s carbon tax revenues. We take projections of carbon tax revenues for each issuer from the NGFS. We apply the NGFS’s assumptions about recycling of those carbon tax revenues. In Below 2oC and Net Zero 2050 scenarios, 50% of revenues are recycled into public investment, and 50% to reduce debt. In other scenarios, all revenues are used to reduce income taxes.

Debt – fuel taxation

Governments generate tax revenues from various forms of taxation, including that of transportation fuels (eg fuel duties). As transportation decarbonises, revenues will decline. We use NGFS emissions projections for the road transport sector as a proxy for the rate of decarbonisation and assume 50% of lost revenue is replaced through new taxation by 2050 and 50% is replaced by additional borrowing.

Debt – green investment

Public spending may be required to support economy-wide decarbonisation by 2050. To estimate costs, we use carbon prices from the NGFS scenarios as a proxy for the marginal cost of investment required to abate a tonne of CO2. By calculating the CO2 abated to 2050, we can estimate the spending required to achieve this. Final costs are calculated net of any recycling of carbon tax revenues.

Debt – debt interest

Debt interest projections account for three effects. Firstly, that the real terms cost of servicing non-inflation linked debt reduces in scenarios with higher inflation. Secondly, that debt servicing costs are higher in scenarios with higher levels of new debt issuance. Thirdly, that debt servicing costs are higher for new debt issuances in scenarios with higher interest rates.

Debt – acute physical spending

This component captures that higher physical risks will lead to higher government expenditure, either because governments provide support to households in response to, or to mitigate against, physical risks. We assume that government spending on acute physical impacts is 20% of acute physical GDP losses – which is calibrated to historical datafootnote [9] – and that this spending is funded via borrowing.

Box C: Results of projections of changes in debt/GDP and sovereign credit ratings using approach in this article

The largest increases in credit risks are observed in business-as-usual scenarios with large physical risks. More orderly transitions are associated with lower levels of credit risk.

Across advanced economy sovereign issuers, the mean increase in debt/GDP compared to a baseline scenario is lowest in the Below 2oC and Net Zero 2050 scenarios (circa +50 percentage points). The mean increase in debt/GDP is highest in the Current Policies and Nationally Determined Contributions (NDCs) scenarios (circa +105 percentage points) (Chart 4). Estimates in the Below 2oC and Net-Zero 2050 scenarios are a similar order of magnitude to previous estimates (eg OBR 2023 and International Monetary Fund 2023).

Chart 4: Distribution of projected debt/GDP ratios in 2050 (a) (b)

Net Zero 2050 and Divergent Net Zero see immediate shocks while Current Policies and Nationally Determined Contributions have limited shocks at shorter tenors but larger shocks at longer tenors. All scenarios see risks grow at longer tenors.


  • Source: Bank of England analysis. See the Annex for full breakdown of data sources.
  • (a) Boxplots shown for advanced-economy issuers.
  • (b) Boxes indicate the 25th and 75th percentile issuer. Line indicates median and cross indicates mean. Outliers indicated by error bars, with outliers outside 1.5 times the interquartile range away from the mean indicated by data points.

Intra-scenario heterogeneity in issuer debt/GDP projections can be explained by:

Physical risk vulnerability: issuers most exposed to physical risks suffer larger GDP losses and indirect effects on debt.

Tax burden: in scenarios where GDP is lower, tax receipts are proportionally lower – and especially so for issuers with high tax burdens – leading to higher borrowing.

Fuel tax burden: the more reliant on fuel taxation an issuer is, the more their debt/GDP increases as fuel tax receipts decline.

Interest rate projections and primary deficits: when interest rates are stressed, there are larger debt servicing costs for issuers – especially if they have large primary deficits.

Transition: countries which must decarbonise more suffer larger reductions in fuel tax income and larger transition GDP losses, with subsequent feedback effects increasing debt too.

Inter-scenario heterogeneity in debt/GDP projections can be explained by:

Physical risks: in scenarios with larger physical risks, GDP loss is greater – reflecting the impact of chronic physical GDP impacts on productivity, and the effects of acute physical risks crystallising. This also feeds into higher debt, for example through lower tax revenues and higher acute physical spending.

Transition risks: scenarios aligned with lower temperatures see larger reductions in fuel tax revenues. More disorderly transitions are associated with larger GDP losses, as well as higher carbon prices. This increases carbon tax revenues, but also increases debt due to higher interest rates and debt interest costs.

Across all scenarios, the mean change in sovereign credit rating is negative – ie sovereign credit ratings are lower in the stressed scenario compared to the hypothetical baseline scenario (Chart 5). In the ‘Orderly’ and ‘Disorderly’ scenarios, median downgrades are small (<circa 1 notch) and some issuers benefit from marginally higher ratings. The largest downgrades are found in the Current Policies and NDCs scenarios, where median downgrades are circa -1 notch, with the largest downgrades for a minority of issuers exceeding circa -6 to -7 notches.

Chart 5: Distribution of projected changes in sovereign credit rating in 2050 (a) (b)