Network for Greening the Financial System
Technical document
Guide to climate
scenario analysis
for central banks
and supervisors
June 2020
NGFS REPORT
1
C
limate change, and our response to it, will have a signicant impact on economic and nancial systems. The impacts
will be far-reaching in breadth and in magnitude; subject to tipping points and irreversible changes; and are uncertain
yet at the same time totally foreseeable. In particular, while we do not know now exactly what physical and transition
risks will materialise, we do know for sure that we will face some combination of those risks. And, crucially, we also know that
the size and balance of these future nancial risks and economic costs will depend on the actions we take today.
If we act now, then we maximise our chances of achieving an orderly transition to a carbon neutral economy. By acting early
we minimise transition risks, and by limiting global warming to a range of 1.5˚C to 2.0˚C relative to pre-industrial levels, we
simultaneously minimise the extent to which the physical risks from climate change materialise. If instead meaningful adjustment
is delayed, then the greater will be its disruption – whether from higher physical risks, or from a more disorderly transition, with
markets potentially repricing sharply, and the provision of nancial services perhaps disrupted. And of course, if we fail to act
at all, that puts us on a path to global warming of 3.0˚C or more, leaving us all exposed to the potentially catastrophic physical
risks that arise with an ever hotter planet.
We do not know what state of the world will materialise. But as central banks and supervisors we have a responsibility to prepare
for the potential impacts from climate change in a variety of possible future states of the world. Scenario analysis is key to us
doing that. It lets us explore impacts and exposures under a range of dierent potential pathways.
To date, central banks and supervisors that have wanted to do climate scenario analysis have faced a number of obstacles. There
is an abundance of climate models to choose from, and it is not immediately clear which ones are most relevant. In addition,
the eld of climate modelling is technical and dicult to penetrate for non-experts. It is complicated further by the lack of a
clear methodological framework for translating climate scenarios into macro-nancial analysis.
That is why the NGFS has developed a set of Reference Scenarios, along with this Guide on how to conduct scenario analysis.
The NGFS Reference Scenarios provide, for the rst time, a harmonised set of high-level climate scenarios, available in a publicly
accessible database, in which both transition and physical climate change impacts are included in a consistent way. To allow
central banks and supervisors to get the most use from these scenarios, the Guide provides practical advice on using scenario
analysis to assess climate risks to the economy and nancial system. The NGFS scenarios provide a foundation for decision-
useful nancial and economic analysis. And they will be useful not only to central banks and supervisors, but also to nancial
rms and to corporates as they too seek to manage their exposure to these risks.
Challenges and shortcomings remain. Indeed we are close to the start of this intellectual journey not at its end. That is why we
will work towards an updated set of scenarios that will be published later in the year. To ensure that those scenarios will be as
complete, coherent and useful as possible, we would like to invite everyone, not just central banks and supervisors, to engage
with us on this important topic.
We simply cannot aord to be unprepared.
Joint foreword by Frank Elderson and Sarah Breeden
Sarah Breeden
Chair of the workstream “Macronancial”
Frank Elderson
Chair of the NGFS
NGFS REPORT
2
Table of Contents
Foreword
1
Executive summary
4
Origin of the NGFS
6
1. Introduction
7
2. Identifying objectives, material risks and stakeholders
9
2.1 Objectives 9
2.2 Assessing material risks 9
2.3 Stakeholders 11
3. Scenario design
12
3.1 Climate scenario assumptions 12
3.2 Further scenario design choices 13
3.3 Overview of the NGFS Scenarios 15
4. Assessing economic impacts
21
4.1 Economic impacts assessed 21
4.2 Transmission channels 21
4.3 Methods 22
4.4 Key assumptions and sensitivities 24
4.5 Rening the results 25
5. Assessing nancial risks
26
5.1 Financial risks assessed 26
5.2 Transmission channels 26
5.3 Methods 27
5.4 Key assumptions and sensitivities 31
5.5 Rening the results 32
NGFS REPORT
3
6. Communicating and using the results
33
6.1 Communication of the results 33
6.2 Uses of the results 34
Bibliography
35
Acknowledgements
38
Annex – Examples of scenario analysis
39
NGFS REPORT
4
Executive summary
The members of the Network for Greening the Financial
System (NGFS) acknowledge that nancial systems and
nancial institutions are exposed to signicant impacts
from climate change. They encourage central banks and
supervisors to lead by example and integrate climate risks
into nancial stability monitoring and supervision. Climate
risks include physical risks, related to the physical impacts
from climate change, and transition risks, related to the
adjustment to a net-zero emission economy.
To this end, the NGFS committed to publishing the
rst-of-its-kind Guide on climate scenario analysis for
central banks and supervisors. The forward-looking
nature of climate risks and the inherent uncertainty about
future events make it dicult to assess them using standard
risk modelling methodologies. Scenario analysis oers a
exible ‘what-if’ methodological framework that is better
suited to exploring the risks that could crystallise in dierent
possible futures.
This Guide provides practical advice on using scenario
analysis to assess climate risks to the economy and
nancial system. It is based on the initial experiences of
NGFS members and observers, and also aims to progress
discussion on the methodologies used. While mainly
aimed at central banks and supervisors, many aspects
of the Guide might also prove informative to the wider
community.
The Guide provides a four-step process. It recognises
that this eld is still relatively in its infancy and that there
is no universally agreed approach.
Four-step process
Step 1
Identify objectives and exposures. Scenario analysis
is relevant to many objectives of central banks and
supervisors. It can be used to stress test nancial rms
and the nancial system, explore structural changes to
the economy and/or assess risks to central banks’ own
portfolios.
A materiality assessment can be useful at the outset to
help determine the risk drivers that will be in or out of
scope. Atargeted exercise would focus on the impact
of these risks on a small number of economic indicators,
sectors, nancial asset classes and/or nancial rms, while
a system-wide risk assessment would be more expansive.
Step 2
Choose climate scenarios. Most publicly available climate
scenarios were originally designed for policy evaluation
and research, and are therefore not entirely appropriate for
central banks and supervisors’ purposes. The NGFS has been
working with the academic community to publish a set of
high-level reference scenarios that can be used for scenario
analysis in a comparable way across dierent jurisdictions.
Each central bank and supervisor will need to make a
number of additional design choices to tailor the scenarios
to the specic exercise. This includes choices related to
the risks covered, the number of scenarios, time horizon
and the specic outputs that will be needed (the ‘scenario
variables’). Early consideration should also be given to
how detailed the analysis will need to be. This will have
an important bearing on the scenario design.
Step 3
Assess economic and nancial impacts: Central banks
are interested in assessing the impact of climate risks on a
wide-ranging set of economic and nancial variables (e.g.
GDP, ination, equity and bond prices, loan valuations)
etc. This includes risks that arise from dierent physical
and transition outcomes across a wide range of sectors
and geographies.
A range of methods is used to model these economic impacts.
This includes several types of bespoke climate-economy
models such as Integrated Assessment Models (IAMs) and
Computable General Equilibrium (CGE) models. Central
banks are considering how to combine these approaches
with the more traditional economic modelling tools they
use with the aim of providing a wider range of outputs and
greater detail about individual economic sectors.
NGFS REPORT
5
A major challenge remains that many macroeconomic
models are used to assess divergences from long-run
equilibria rather than fundamental shifts in the economy.
However, conversely, climate-economy models tend to
have much more simplistic macroeconomic modelling
and it is more dicult to calibrate them accurately. The
NGFS scenarios (as well as other scenarios) are working
to bridge this gap but in the interim it is likely that a suite
of models will be required.
Methodologies for nancial assessment of climate risks
are also developing. Several central banks are considering
how best to integrate climate scenarios into stress testing
exercises. These range from shorter-term, top-down
modelling exercises undertaken by the central bank, to
exercises with a longer time horizon, in some cases with
bottom-up participation by nancial rms. A key challenge
is obtaining granular enough information on how the
scenario would aect economic activity to assess the
nancial risks.
Step 4
Communicating and using results. Communicating the
results, and the key assumptions underpinning them,
will help increase awareness. This may provide a basis
for follow-up actions from central banks and supervisors
and encourage financial institutions to improve their
risk-management practices. The scenario analysis exercise
may lead to further analyses of specic pockets of risk and
monitoring of key risk indicators. It can also inform whether
existing regulatory policies (e.g. capital treatment) and
approaches (e.g. economic forecasting) are t for purpose.
Next steps
This Guide is intended to evolve over time as experience
using scenarios to assess climate risks grows. For the
next phase of the Guide, the NGFS will leverage further
insights from the practical experiences of central banks and
supervisors as an increasing number undertake scenario
analysis.
NGFS REPORT
6
O
N
E
P
L
A
N
E
T
S
U
M
M
I
T
DECEMBER
2017
central banks
and supervisors
established the Network
of Central Banks and Supervisors
for Greening the Financial System.
representing 5 continents.
As of end-June 2020, the NGFS consists of
The NGFS
is a coalition
of the willing.
It is a voluntary, consensus-based forum
whose purpose is to share best practices,
contribute to the development of climate
–and environment– related risk
management in the financial sector
and mobilise mainstream finance
to support the transition towards
a sustainable economy.
The NGFS issues
recommendations
which are not binding
but are aimed at inspiring
all central banks and supervisors
and relevant stakeholders
to take the necessary
measures to foster
a greener financial system.
66
Members
13
Observers
Origin of the NGFS
NGFS REPORT
7
1. Introduction
The NGFS’s goal is to share best practices and equip
central banks and supervisors with the tools to identify,
assess and mitigate climate risks in the nancial system.
In its rst comprehensive report, published in April 2019,
the NGFS recommended that central banks and supervisors
integrate climate factors into nancial stability monitoring
and supervision. This guide is a direct follow-up to that
recommendation.
The distinct nature of climate risks poses a challenge
to standard risk assessment approaches. Climate risks
have long time horizons with high uncertainty about how
policy and socio-economic factors might evolve; they are
global and economy-wide in nature; and they are complex,
varying from region to region and sector to sector. These
distinct characteristics are not captured by risk assessment
approaches that rely on top down modelling and historical
trends, are narrowly focused, and assume the structure
of the economy and nancial system remain unchanged.
Scenario analysis is an essential tool to overcome these
challenges. It provides a exible ‘what-if’ framework for
exploring how the risks may evolve in the future. These
scenarios can help a wide range of players better understand
how climate factors will drive changes in the economy and
nancial system, including central banks and supervisors,
nancial rms, companies and policy makers.
However, the use of climate-related scenario analysis is
relatively new and methodologies are still developing.
Some of the main issues include the lack of integration of
physical risk, transition risk and macro-nancial transmission
channels; lack of available data and research to calibrate
the scenarios and assess impacts; and lack of technical
expertise on climate science and environmental economics
within the nancial sector.
The NGFS has been working with the academic
community to publish a set of standardised scenarios that
can be used for macro-nancial analysis in an open-source
platform. This includes a standardised set of transition risk,
physical risk, and macroeconomic variables and the key
assumptions that they rely on. The scenarios draw primarily
on existing mitigation and adaptation pathways assessed
by the Intergovernmental Panel on Climate Change (IPCC)
reports. Over time the aim is to work with the academic
research community to make the scenarios more directly
relevant for macro-nancial analysis.
The guide sets out some practical considerations for how
to use these climate scenarios to assess macroeconomic
and nancial risks. The rst of its kind, it is based on the
initial experiences of NGFS members that have implemented
or plan to implement climate-related scenario analysis and
will be enhanced over time.
Scenario analysis involves four broad steps: identifying
objectives and exposures, choosing scenarios, assessing
impacts and communicating results. The guide is set
out as follows:
• Chapter 2: Identifying objectives, material risks and
stakeholders;
• Chapter 3: Choosing relevant scenarios;
• Chapter 4: Using the scenarios to assess economic impacts;
• Chapter 5: Using the scenarios to assess nancial risks;
• Chapter 6: Communicating the results and next steps.
NGFS REPORT
8
Figure 1. Overview of the scenario analysis process
Identify objectives, material risks and stakeholders
Identify most material risks
Assess materiality
Map key stakeholders
Identify target audience
Dene appropriate number of scenarios
1
2
Design scenarios
Based on
Relevance to local context
Severity of scenario
Time horizon of scenario
Assumptions on:
- Socioeconomic context
- Climate policies
- Technological evolution
determined by
ease of
implementation
Scope of results
more
scenarios
fewer
scenarios
ease of
communication
Depending on scenario analysis objective
A. Assessing nancial rm-specic risks
B. Assessing nancial system-wide risks
C. Assessing macroeconomic impacts
D. Assessing risks to central bank’s own
balance sheet
Policy actions
E.g. micro/macro-prudential, monetary, economic, scal
Communication
E.g. awareness, stimulate research and discussions
Expected scenario outcomes
uncertainty
capture of
long-term eects
longer time
horizon
shorter time
horizon
Depending on scenario analysis objective
and nancial instruments being analysed
Dene appropriate time horizon
Central banks and supervisors should ensure the scope of the exercise is focused on key exposures. This involves identifying the institution’s
objectives, the most material risks and the key stakeholders to involve.
Possible objectives:
Central banks and s
hould choose scenarios that are relevant to the risks
they want to explore.
4
Communicate and use results
Central banks and supervisors should consider the information to disclose in
order to improve rms’ risk-mitigation practices and foster further research.
Assess impacts
3
• Assess nancial risks
Specics:
- What key variables?
- What level of granularity?
- What tools and how to use them?
- How to rene outputs?
- What are the key assumptions and sensitivities?
Specics:
- Which type of exercise?
- Which method?
- Which data?
- How to recalibrate scenarios?
- What are the key assumptions and sensitivities?
Type of exercise
Bottom-up
exercise
Top-down
exercise
ease of
implementation
granularity of
outputs
Depending on granularity of analysis
Level of granularity
Temporal
resolution
Low
Macro to
rms &
households
Medium
High
Spatial
resolution
Global to
sub-national
Decade,
year,
quarter, etc.
Central banks and supervisors should select the methods and tools
needed to assess the potential macroeconomic and nancial impacts.
• Assess economic impacts
Economic
resolution
Type of risk to be analysed
(i.e. physical vs transition)
NGFS REPORT
9
2. Identifying objectives,
material risks
and stakeholders
This chapter sets out the preparatory work that
institutions should do to ensure the scope of the exercise
is focused on key exposures. This involves determining
how the exercise relates to the institution’s objectives,
assessing the materiality of climate risks to these objectives
and identifying the key stakeholders.
2.1 Objectives
Central banks and supervisors should rst consider
how the exercise will relate to their objectives. This
will help determine the breadth of analysis undertaken.
There is a trade-o between obtaining a holistic view
of the risks and the amount of resources needed.
Scenario analysis can be relevant for:
A. Assessing specic risks to nancial rms, including the
impact on rm balance sheets, protability, capital
and / or business models. See also the NGFS Guide
for Supervisors on Integrating climate-related and
environmental risks in prudential supervision;
B. Assessing nancial system-wide risks, including their
aggregate size, distribution and systemic nature;
C. Assessing macroeconomic impacts, including the short
and long-run eects on growth, employment, ination
and terms-of-trade;
D. Assessing risks to a central bank’s own balance sheet,
including arising from their market operations and other
portfolios they manage (e.g. on behalf of government).
Central banks and supervisors should also consider
how to integrate scenario analysis into existing risk
assessment processes. For example, by incorporating
climate scenarios into a nancial system stress test or a
macroeconomic forecast. Table 1 below sets out some further
examples. These exercises can be quantitative or qualitative.
1 https://www.ngfs.net/sites/default/les/medias/documents/ngfs_rst_comprehensive_report_-_17042019_0.pdf
2 https://www.ngfs.net/sites/default/les/medias/documents/ngfs-report-technical-supplement_nal_v2.pdf
Table 1. Examples of how central banks and supervisors assess dierent risks
Objective Types of risk assessment Useful for
A Assess nancial rm-specic risks Stress testing, challenging rm
capital adequacy assessments
Microprudential policy
Identifying risks related to safety and soundness
B Assess nancial system-wide risks Stress testing, research on individual
transmission channels
Macroprudential policy
Identifying systemic risks and macroeconomic
impacts
C Assess macroeconomic impacts Macroeconomic forecasting,
research on structural changes
Understanding macroeconomic outlook
Monetary policy
D Assess risks to own balance sheet Credit and market risk analysis, stress testing Managing risks to own operations
TCFD disclosures
2.2 Assessing material risks
Scenario analysis should aim to assess the most material
risks to the institution’s objectives. A materiality assessment
can help identify the climate drivers that are likely to have
the most signicant impacts. This will help identify relevant
scenarios and prioritise analysis, on the basis that it would
be impractical to determine all potential risks at the outset.
It is very important to be clear about the risk drivers that
are in or out of scope. These judgments should be revisited
after the conclusion of the exercise to ensure the scenario
analysis is focused on the most relevant risks.
Central banks and supervisors should rst gather all
relevant information that is available, bearing in mind
that there will likely be information gaps. Good starting
points include the First NGFS Comprehensive Report
1
and
Technical Supplement.
2
These set out climate risk drivers
and their possible impacts on the financial system and
economy. This will most likely need to be supplemented
NGFS REPORT
10
with jurisdiction-specic research on climate risks from
the nancial sector, government, industry and academia,
including climate scientists.
Central banks and supervisors should then identify the
types of risks that will be included in the assessment.
Transition risks relate to action taken to reduce emissions
to reach net zero greenhouse gas emissions. Physical risks
relate to the eects of global warming on physical capital,
human health and productivity and agriculture. Macro-
nancial risks refer to the standard nancial risk categories
(e.g. credit, market, operational) and economic indicators
(e.g. output, unemployment, ination).
Climate risks are complex and there are many dimensions
to consider. These include: the extent to which the risks vary
depending on the time horizon (e.g. short-term, medium-
term, long-term risks), the risk distribution (e.g. average
losses, losses from worst-case low-probability events) and
how much is known about the potential impacts from
events where we have little historical experience.
Table 2. Research questions to identify potential risks and assess materiality
Type of risk Research question Source of information
Climate
Physical risk
• What are the most material domestic physical hazards from
extreme events (e.g. ooding, extreme temperature changes,
windstorms) and from gradual changes in climate
(e.g. changes in agricultural yields or water availability,
sea-level rise, heating and cooling requirements)?
• What eects could there be on real estate and infrastructure,
business continuity, people and food systems?
• Are there any signicant international transmission channels
(e.g. import/export of food, supply chains)?
• What kind of adaptation measures are being implemented (e.g. shift
in crop types, water regulations, coastal protection measures)?
•
NGFS publications
•
Government reports
•
Academic research (including IPCC reports)
•
Financial industry reports on climate risks
•
Public data sets (e.g. physical hazards,
energy eciency, emissions)
Transition risk
•
What type of government policies are being considered /
implemented (e.g. carbon tax, direct regulation, subsidies)?
•
Which technological trends could play a key role in the coming
decades (e.g. renewable energy, carbon capture and storage,
electrication of motor vehicles)?
•
Are there any signicant changes in consumer preference
(e.g. transport demand, diets, energy-ecient housing,
energy-ecient appliances)?
•
Which sectors of the economy are particularly at risk of policy
or technological disruption (e.g. energy sector, agriculture,
construction, industry, mobility and freight transport)?
Macro-nancial
Financial
•
What are the largest exposures of banks by type of asset
(e.g. retail credit, wholesale credit, trading book)?
•
What are the largest insurance underwriting exposures?
•
What are the largest exposures for capital markets (equities,
corporate bonds, derivatives, structured products)?
•
What is the geographical distribution of these exposures?
For corporate exposures, this should take into account both
jurisdiction and operating locations.
•
What is the distribution across economic sectors for these
exposures?
•
Financial regulatory data
•
Central bank statistical information
•
Review of relevant variables in internal
nancial and macroeconomic models
•
Academic research
Macroeconomic
•
What are the most material drivers of changes to macroeconomic
conditions (e.g. GDP and potential growth, unemployment,
interest rates, ination)?
•
What is the current sectoral composition of the economy
and how is this changing?
NGFS REPORT
11
2.3 Stakeholders
Central banks and supervisors should consider how their
stakeholders will be involved in the scenario analysis.
These stakeholders could be included explicitly, as part
of the exercise (e.g. in a rm-based stress test); and/or
as part of the target audience for the results (refer to
Chapter 6 for more details on communication). There are
ve main groups:
• Financial institutions (including banks, insurers, asset
owners and asset managers) are developing their own
scenario analysis expertise. Credit rating agencies are
also looking at scenarios to rene and develop their
ratings methodologies. These eorts can both inform,
and learn from scenario analysis undertaken by central
banks and supervisors. Supervised entities may also
participate directly in the exercise.
• Financial standard setters may find the results of
scenario analysis useful in developing domestic and
international standards for nancial institutions.
• The general public is an important stakeholder given the
role of central banks and supervisors as public institutions.
Scenario analysis may inform, and be informed by, the public
discourse around risks and responses to climate change.
• Governments and international bodies. National
mitigation and adaptation plans, and international
coordination on these issues, will be a key input into
the scenario analysis. Information on the transmission
channels and macro-nancial impacts of the exercise
may in turn inform and inuence government policy.
• The academic community engages in research on the
impacts of climate change. Central banks and supervisors
have a role to play in fostering and learning from research
on the role played by the economy and nancial system
NGFS REPORT
12
3. Scenario design
Climate scenarios explore dierent possible climate
change futures and pathways towards achieving
long-term climate goals. This chapter sets out the main
assumptions underpinning climate scenarios and some
further scenario design choices to be made. It nishes by
providing an overview of the NGFS scenarios.
3.1 Climate scenario assumptions
Climate scenarios are the core input into assessing
the macro-nancial impacts from climate change. It is
important that central banks and supervisors consider the
assumptions being made, and choose scenarios that are
relevant to the risks they want to explore. The key model
assumptions and design choices relate to emissions and
climate outcomes, the socioeconomic context, climate
policy, technology and consumer preferences.
Atmospheric concentration
of greenhouse gases
The Intergovernmental Panel on Climate Change (IPCC)
collates and assesses physical and transition scenarios
that are continuously developed by the climate research
community. The IPCC is the main body responsible for
globally coordinating and publishing assessments on
climate change for policymakers. These scenarios set out
pathways for the emissions of greenhouse gases, their
future atmospheric concentrations, and projections for
consequent climate impacts. The research community
has collectively chosen four Representative Concentration
Pathways (RCPs
3
) to help standardise and improve
comparability of climate change analysis. These RCPs have
now been updated for the IPCC’s ongoing 6
th
Assessment
cycle (2015-2022).
4
The NGFS is working with partners from the academic
community to make these scenarios more relevant
for macro-nancial analysis. This includes enhancing
macroeconomic modelling and improving the coherence
between physical and transition risk modelling.
A wider range of scenarios than the RCPs are also being
considered such as the emissions pathways consistent with
governments’ current policies and more abrupt emissions
reduction scenarios.
Socioeconomic context
The socioeconomic backdrop of the scenarios helps to
contextualise the setting in which the climate scenario
occurs. A world in which consumption patterns become
more sustainable could have a marked reduction in
emissions, whereas a world in which fossil-fuel development
continues will either increase emissions or reinforce the
pathway we are currently on.
The Shared Socioeconomic Pathways (SSPs) have also
been standardised by the research community to help
coordinate climate scenario modelling. They can be used
to estimate how dierent levels of climate change mitigation
(under the RCPs) could be achieved under a possible
socio-economic pathway. They are based on quantitative
projections of three variables – GDP, population, and
urbanisation rate – as well as detailed narratives describing
technological advancement, international cooperation or
resource use, foreseen for a wide range of countries and
regions, up to 2100.
5
Technological evolution
Climate scenarios dene the technology pathways that
lead to a reduction in emissions. This varies from model
to model but typically includes increasing energy eciency,
decarbonisation of power sources (via the phase-out of
fossil generation and increasing low-carbon technologies
like renewables), increasing electrication, more ecient
land use, and some direct carbon dioxide removal from
the atmosphere through bioenergy with carbon capture
and storage and/or land-related sequestration (e.g.
aorestation). Scenarios make assumptions about how
these technologies progress over time, to project how levels
of investment and deployment rates develop in the future.
3 For further discussion of RCPs see van Vuuren et al. (2011).
4 O’Neill et al., 2016.
5 Riahi et al., 2017.
NGFS REPORT
13
Climate policies
Climate scenarios also make either implicit or explicit
assumptions about how climate policies may evolve.
The key policy assumptions relate to:
• Timing: whether action is taken sooner or later, which
has a signicant impact on the rate of required emissions
reductions;
• Policy mechanism: including the policy mix (e.g.
taxation, cap-and-trade carbon pricing, subsidies,
emissions restrictions, industry regulations), who pays
(governments, companies and/or households) and how
government revenues (if any) are redistributed;
• Policy certainty: whether policy implementation is
relatively gradual and predictable, or unanticipated and
abrupt. This could for example take place in the context of
a delayed policy response, with a sudden implementation
of new regulations (e.g. ban on coal, imposition of carbon
taxes) rather than a smooth phase-in period;
• Policy coordination: the degree of coordination across
countries in tackling climate change.
Climate-economy models set out the types of technology
changes needed to transition, but are not always explicit
on the policy mechanisms to get there. Global climate
transition pathways are derived from integrated assessment
models (IAMs) that model the interaction between energy,
land, economy and climate systems.
Models vary by how explicitly and granularly they
take dierent policy mechanisms into account. Some
climate-economy models have been developed to explore
the impact of specic types of policies (e.g. carbon tax) on
the economy. This includes computable general equilibrium
models (CGEs) and other macro-econometric models.
Other models focus more on the nature and costs of
transformations in the energy system, and are less explicit
on policy mechanisms to get there. More detail on these
approaches is provided in Chapter 4.
Consumer preferences
Climate scenarios make a number of assumptions about
how consumer preferences evolve. In their simplest form
climate-economy models assume that the transition is
primarily led by the supply side of the economy (i.e. new
technologies allow for the provision of existing goods and
services at a lower emissions intensity). However, there is
increasing academic research and policy attention on how
much shifts in consumer preferences for certain goods
and services could contribute to achieving climate goals.
Examples include demand for dierent forms of transport,
agricultural land and dietary preferences.
Climate impacts
Climate scenarios provide information on how
temperature and other biophysical processes are
changing. The underlying climate and hazard models
provide a range of projections depending on dierences
in the input assumptions used and methodology. It is
therefore important to understand how summary statistics
(e.g. temperature outcome in 2100) have been derived and
compare to the wider distribution of results.
Climate scenarios also provide information on how
these changes in climate will aect people’s health and
productivity, physical capital and food systems. This requires
making an additional number of assumptions about the level
of adaptation and how economic activity will be aected.
These assumptions are further explored in Chapter 4.
3.2 Further scenario design choices
There are a number of further choices related to how
the underlying climate scenarios are integrated into
the exercise. These relate to the types of risks explored, the
number of scenarios, granularity, time horizon and calibration.
Climate risks explored
Central banks should consider the types of climate
risks they want to explore. Physical and transition risk
scenarios are often modelled separately. If the scenario
is intended to assess the macro-nancial impacts of both
risks, the models should be as coherent as possible. At a
high level the scenario narratives should be aligned to the
same emissions pathway and temperature outcome as far
as possible. The scenario models should also use consistent
input assumptions (e.g. on policy, technology and the
socioeconomic context). The NGFS scenarios cover both
risks. Care needs to be taken to avoid double-counting of
macro-economic impacts. A full integration would require
simultaneously considering physical impacts and transition
policies in the scenario development.
NGFS REPORT
14
Number of scenarios
Multiple scenarios should be used to explore dierent
plausible scenarios and trade-offs that may exist
between them. For example, scenarios with high global
emissions can be used to explore physical risks. Scenarios
with a reduction in emissions can be used to explore
transition risks. This transition can be assumed to occur
with coordinated policy, investment in new technologies
and gradual capital replacement, or in a disorderly way
with late, sudden and/or unanticipated shifts in policy, the
economy and nancial system.
The number of scenarios that central banks and supervisors
choose to analyse will depend on the objective of the
exercise, the materiality of the macro-nancial risks, and
resources available. Analysing more scenarios will lead to a
more comprehensive and holistic view of the risks. However,
the broader scope can constrain how deeply particular details
can be explored for a given level of resourcing.
Figure 2. Trade-os of analysing more versus fewer
scenarios
Scope of
results +
Fewer
scenarios
+ Fewer resources
required
+ Ease of
communication
More
scenarios
Scenario granularity
Central banks and supervisors should determine the
level of granularity at which they want to assess the
risks. This will have an important bearing on the design of
the exercise, choice of scenarios and data required. Possible
levels of resolution are set out in Table 3 below.
Dierent climate-economy models oer dierent levels
of sectoral and geographic coverage. Historically climate-
economy models tended to focus more on the energy and
land systems, and model world regions at an aggregate
level. There are also domestic models run by individual
countries that provide a greater level of national resolution.
In practice, the scenario will almost always need to be
supplemented with additional modelling and data. This
is further explained in Chapters 4 and 5.
Table 3. Possible levels of granularity
Economic
resolution
Geographical
resolution
Level of granularity
Low Macroeconomy Global
Medium Sectoral level Country to regional
High Firm / Household level Postcode down to
individual property
location
Scenario analysis focused on individual nancial rms
and their portfolios will typically need to be undertaken
at a high level of granularity. For example, ood risk
may impact households on one end of the street and not
the other. Similarly, the risks to a fossil fuel company will
substantially depend on costs of production and whether
the company has plans to broaden out its strategy. In order
to fully assess nancial stability risks, aggregation of these
granular risks is typically required.
Scenario analysis at a medium or low level of granularity
will typically be sucient for assessing the impact on
the macroeconomy. It may also be sucient to understand
the aggregate risks faced by the nancial sector, particularly
in high-risk sectors. This type of top-down analysis can
also be useful to understand the potential feedback loops
between the nancial sectors and the real economy.
Time horizon
The appropriate time horizon for the chosen scenarios
will depend on the objective of the specic exercise.
Shorter time horizons are useful to analyse the types of
nancial risks that could crystallise within business planning
horizons and to assess the impact on regulatory capital
more precisely. EIOPA (2019), Norges Bank (2019) and De
Nederlandsche Bank (2018)
6
used a 5-year scenario length
in their analyses of climate-related risk.
Longer time horizons are useful to gauge exposures to
structural changes in the economy and nancial system,
and to consider how the strategic decisions of nancial
rms could aect the risks. Banque de France/ACPR (2020),
Bank of England (2019) and Danmarks Nationalbank (2019)
consider timelines of up to 2050, 2080 and 2100, respectively.
7
6 Vermeulen et al., 2018.2 O’Neill et al., 2016.
7 Note that in the case of a ‘no additional policy action scenario’, the Bank of England proposes to assume that the more material risks anticipated in
the period from 2050 to 2080 occur by 2050.
NGFS REPORT
15
8 No existing climate change scenarios in the literature consistently matched the narrative of a “Too little, too late” scenario including policy disruptions
at the same time. This fourth class of scenarios might be covered in the next NGFS scenario release.
9 The climate model MAGICC was used to estimate the temperature outcomes of emissions pathways in the IPCC Special Report on 1.5 °C Warming
(SR15). It emulates historic warming, climate sensitivity and the warming projections of Earth System Models.
10 Huppman et al., 2018.
Short-term scenarios can help convey a greater sense of
urgency, and are perhaps easier to conceptualise, but
provide a relatively limited view of how the risks unfold
relative to long-term scenarios. This, however, comes with
an important caveat that the longer the scenario, the greater
the uncertainty band around the results. This increases the
importance of choosing an initial set of starting assumptions
that reect the risks that will be explored.
Even when a short time horizon is chosen for analysis, it
will often still be useful to have long-term scenario outputs
available, for example, where nancial markets are assumed
to price in future expectations.
Frequency
Central banks and supervisors should consider the
desired frequency of analysis. For example, risks could
be assessed at an interval of 1 year, 5 years, 10 years, etc.
over the duration of the scenario. This is important to
consider because climate scenarios often cover long
time horizons (out to 2100) with model time steps of
5 years or more.
Annual changes can be derived from longer-term
periods if more frequent scenario outputs are not
available. However, it will often be necessary to reconsider
the scenario assumptions and consider other short-term
eects that could arise. For example, this could include
assumptions around the extent to which the economy
diverges from equilibria and whether there is market
volatility or credit tightening within the nancial system.
Calibration
Central banks and supervisors may approach scenario
analysis with dierent questions in mind, and should
calibrate the scenarios accordingly. For example, they
may be interested in mapping out a required adjustment
path for the nancial sector under plausible climate change
scenarios, or they may be interested in exploring potential
losses under worst-case scenarios.
At a high level, the scenario calibration can be conducted
in at least two ways. First, one can select climate scenarios
that are more or less severe in terms of physical and transition
risks. Second, for variables for which a probability distribution
is available (e.g. probability of reaching a particular climate
outcome, probability of a physical hazard occurring), one can
decide to focus more on mean or median ranges, or on tail risk.
3.3 Overview of the NGFS Scenarios
The NGFS published in June 2020 a set of reference
scenarios that can be used to explore the economic
impacts and financial risks from climate change.
This included three representative scenarios aligned with
the categories of the NGFS Scenarios Matrix – Orderly,
Disorderly and Hot house world.
8
They are accompanied
by ve alternate scenarios to provide further context and
facilitate a more robust analysis.
The NGFS Scenarios are not forecasts, but rather explore
risks in a range of future states of the world. In line with
the NGFS Scenario Matrix, scenarios were selected to explore
moderate (1.5-2°C) and high (3+°C) levels of warming by
the end of the century. They were also selected to show
a variety of dierent transition pathways for reaching a
given warming outcome.
Multiple models were used to produce the scenarios
to capture a range of uncertainty in the results. The
transition pathways were generated by three dierent
integrated assessment models (GCAM, REMIND-MAgPIE,
and MESSAGEix-GLOBIOM) to provide dierent views
of how the economy responds to mitigation policy.
The climate model MAGICC
9
was also used to simulate the
temperature response to the NGFS scenarios and provide
an uncertainty band to a change in emissions.
The scenarios were produced jointly with a consortium
of leading research institutions building on the existing
transition scenario database for the IPCC Special Report
on 1.5°C Warming
10
and relevant physical risk impact
data. The rst iteration focussed on bringing together
NGFS REPORT
16
relevant physical and transition pathways from the existing
literature in a coherent way. An update will be released in
the last quarter of 2020 with renements and improvements
to the scenario assumptions, macroeconomic modelling,
,and regional and sectoral granularity.
Below is a brief description of the main assumptions and
characteristics of the NGFS reference scenarios. Further
detail can be found in the NGFS Scenario Presentation, the
NGFS Scenario Database and accompanying NGFS Scenario
Technical Documentation available here.
Scenario assumptions
All selected scenarios build on the same background
socio-economic assumptions, namely the SSP 2 “Middle
of the road”, where the world follows a path in which social,
economic and technological trends do not shift markedly
from historical patterns.
They do, however, vary according to how policy action
is assumed to evolve in the future. Scenarios that assume
currently implemented policies (NPi) or planned policies as
stated in the Nationally Determined Contributions (NDC)
under the Paris Agreement result in high levels of warming
by the end of the century.
11
Orderly scenarios assume that an
optimal emissions price is introduced immediately to limit the
rise in temperatures to ‘well-below’ 2 degrees (66% likelihood)
by the end of the century. Most of the disorderly scenarios
assume that such an emission price is only introduced after
2030. In any case, emission price trajectories are provided
for all scenarios so that the marginal costs of mitigating
emissions in each one can be compared.
The scenarios also make a range of assumptions about
how technology evolves. The availability of Carbon Dioxide
Removal (CDR) technologies is a key driver in particular.
If the availability of these CDR technologies is assumed to
be limited, much sharper increases in emissions prices are
required. In addition, a diverse set of technology assumptions
is embedded in each scenario, related to the costs and
quantities of fossil resources, the availability of solar, wind
and geothermal resources, land, geological storage, etc.
See the Technical documentation for more details.
11 National Policies implemented (NPi) scenarios describe energy, climate and economic projections based on currently implemented national
policies. Nationally Determined Contribution (NDC) scenarios consider policies additional to those represented in the NPi scenarios, assuming that
all countries fully implement their pledged contributions.
Figure 3. Key aspects of the Representative scenarios
- 10
0
10
20
30
40
50
60
70
80
2020 2030 2040 2050 2060 2070
Gt emissions / year
Orderly (all GHGs) Orderly (CO
2
)
Disorderly (all GHGs) Disorderly (CO
2
)
Hot house world (all GHGs) Hot house world (CO
2
)
3°C+
1.5–2°C
Emissions across scenarios
Orderly Disorderly Hot house world
-
100
200
300
400
500
600
700
800
2020 2030 2040 2050
USD (2010)/t CO
2
Emission price development across scenarios
NGFS REPORT
17
0
25
50
75
Per cent
100
2020 2030
Orderly
2050 2020 2030
Disorderly
2050 2020 2030
Hot house world
2050
Evolution of the primary energy mix
by scenario
Renewables
and biomass
CoalGasGas OilNuclear
0
1
2
3
4
5
2005 2035 2065 2080 2095
°C above pre-industrial levels
2020 2050
Current policiesOrderly
RCP2.6 median
RCP2.6 range (90%)
RCP6.0 median
RCP6.0 range (90%)
Source: “IIASA NGFS Climate Scenarios Database”. See “NGFS Climate Scenarios for Central Banks and Supervisors” for further details.
NGFS REPORT
18
Description of the NGFS Reference Scenarios
1. An orderly transition
The representative scenario for an orderly transition
assumes immediate action is taken to reduce emissions
consistent with the Paris Agreement. It assumes the
introduction of an emissions price in 2020 which increases
by $10/tonne CO
2
per year
12
and is calibrated to keep global
warming well-below 2 °C. It also assumes the full availability
of CDR technologies. This corresponds to reaching net
zero CO
2
emissions between 2050 and 2070. Since policy
measures are introduced early and increasing progressively,
physical as well as transition risks are assumed to remain
low over the period. Note that the availability of CDR
technologies at scale is still uncertain as there has not
been much deployment yet.
Two other alternate scenarios have also been selected.
The rst shows how Paris targets could be reached with
limited use of CDR technologies. The second alternate
scenario shows a pathway to limiting global warming to
1.5 °C by the end of the century. Both are more ambitious
than the representative scenario and require an even
higher emissions price to reduce emissions. They would be
suitable to be used for orderly but more stressful scenarios
in nancial risk analysis.
2. A disorderly transition
The representative scenario for a disorderly transition
shows a much more challenging pathway to meeting
the Paris Agreement targets. In this scenario, climate
policy follows NDCs until 2030. Acknowledging that these
eorts will not be enough to meet commitments, the
emissions price is revised substantially upward after 2030.
The scenario further assumes that there will be only limited
CDR technologies available. The period of delay means that
net zero CO
2
emissions must be reached more quickly, by
around 2050. Correspondingly the increase in emissions
prices is much more rapid at $35/tonne CO
2
per year
13
.
Two other alternate scenarios have been also selected.
The rst one is the “Delayed 2 °C scenario” that is similar
in its assumptions to the representative scenario, but
assumes full CDR technology and is therefore less adverse.
Emissions prices are more than three times less than in the
representative scenario, with mild transition risks. Net zero
CO
2
emissions will be reached between 2050 and 2070.
The second alternate scenario is a 1.5 °C scenario with
limited use of CDR technologies. This scenario is the most
disruptive scenario of the set. It assumes that an immediate
global emissions price is introduced to rapidly reduce
emissions in line with the 1.5 °C target while available CDR
technology is limited.
3. A “Hot house world” scenario
The representative scenario for a “Hot house world”
assumes that only current policies are implemented.
As a result, the climate goals set out in the Paris
Agreement are not met, leading to substantial physical
risks over the medium and long term. It is an extrapolation
of what would happen if no additional measures were taken.
The change in emissions price is therefore assumed to be
negligible. This scenario would result in severe physical
risks, with an estimated median temperature rise of over
2 °C by 2050 and close to 4 °C by 2100.
An alternate scenario, labelled Nationally Determined
Contributions (NDCs), is included, taking into
consideration all pledged but not yet implemented
policy measures. The estimated physical risks would be
slightly lower than in the Current Policies case, but still well
above the Paris target, with a median temperature rise of
over 2 °C by 2050 and over 3 °C by 2100. The estimated
transition risks would still be quite limited.
12 Esmaon using results from the REMIND-MAgPIE model.
13 Idem.
NGFS REPORT
19
Box 1
Short-term scenarios used by other authorities DNB and the ESRB/ECB
In its transition risk stress test, DNB explores four short-term
disruptive energy transition scenarios (see Vermeulen et
al., 2018). In line with common stress testing practice
(cf. BCBS, 2018), the four stress scenarios were chosen
to be ‘severe but plausible,’ thus capturing tail risks. To
determine scenario narratives that would qualify as severe
but plausible for the short term, DNB tested the scenario
assumptions with external experts.
Unlike the NGFS scenarios, the DNB scenarios are not
explicitly tied to a temperature outcome and focus
on transition risk. This approach has the advantage of
creating scenarios that are more or less independent of
climate science. The guiding assumption is that the energy
transition is ultimately a socio-political and technological
phenomenon, which can occur under varying conditions
of climate change. A drawback of this approach, however,
is that there is no direct link between the DNB scenarios
and the well-known IPCC scenarios. In addition, the DNB
approach is mainly eective for short time horizons where
it is safe to ignore the interplay between the energy
transition and climate change, but may be less suitable
for exercises that focus on a longer time horizon. The DNB
scenarios cover a ve-year time period.
DNB calibrated its scenarios along two axes, which each
reect a key driver of the energy transition: policy and
technology (gure 3). This resulted in one scenario with a
delayed policy response (“policy shock”), one scenario with
an asymmetric “technology shock” and one scenario in which
both disruptions occur simultaneously (“double shock”). In
the case where neither disruption occurs, it is assumed that
the lack of an energy transition triggers a drop in condence
for consumers, businesses and investors (“condence shock”).
The rst two scenarios, i.e. a delayed policy response and
an asymmetric technology shock, are also considered in the
pilot stress test that is currently being developed by the joint
ESRB/ATC and ECB/FSC project team. …/…
Figure 4. DNB scenarios
1
1 In the “Policy shock” scenario, policies aiming at achieving the goals set by the Paris Agreement are initially deferred. As a result, policies reducing
CO
2
emissions and limiting the increase in global temperature to below 2 degrees Celsius above pre-industrial levels are ultimately introduced
in a disorderly manner. The late implementation of policies necessitates abrupt adjustments leaving the private sector, and subsequently the
nancial sector, with little time to accommodate changes.
The asymmetric “Technology shock” scenario considers a positive breakthrough in energy storage technology. Because the breakthrough is
unforeseen, it becomes a source of disruption for the economy and the nancial sector. This results in a precipitous redistribution of resources
across sectors, defaults and write-os of carbon-intensive assets.
In the “condence shock” scenario, it is assumed that policy uncertainty triggers a sudden drop in condence, such that consumers delay their
purchases, producers invest more cautiously and investors demand higher risk premiums. As a result, there is a setback in GDP, stock prices fall
and lower ination leads to lower interest rates.
NGFS REPORT
20
The scenarios are derived within the multi-country model
NiGEM that provides detailed information about the
evolution of macro-nancial variables at a country level.
In the delayed policy response it is assumed that an
abrupt policy change aiming at mitigating climate change
translates into a sudden and sharp increase in the carbon
price by US$ 100 per tonne at the global level. An abrupt
increase in energy prices leads to sharp devaluation of
trading assets, reected in the drop of stock and bond
prices, and the deterioration of economic conditions
for the entire 5-year horizon. In case of a technological
innovation shock, the technological breakthrough would
allow the share of renewable energy to double over a
five-year period. The asymmetric technology shock
leads to a temporary economic slowdown because of
frictions associated with the switch from the old to the
new technology, but the new technology ultimately
supports economic growth. The double shock resembles
the technology shock pattern, but with a steeper initial
setback of economic growth due to the increase in the
carbon price. The condence shock scenario is modelled
as a drop in consumption and an increase in the cost of
capital of rms and the risk premia of investors, which
together lead to a broad economic slowdown.
In the ESRB/ATC-ECB/FSC exercise, both scenarios (policy
and asymmetric technology shocks) are considered
against the baseline accommodating current policies.
2
In the DNB exercise a baseline was not explicitly dened,
given that it is unclear what a short-term “business-as-
usual” scenario might look like in the context of climate
change. Indeed, if business-as-usual is interpreted as a
scenario in which no additional climate change mitigation
policies are implemented, this would be a scenario in
which physical risks will likely increase signicantly in
the long run. In the short run, this may well result in a
condence shock as depicted in the bottom-left corner
of gure (Figure 4).
2 Beyond the three-year horizon of the ECB forecast, the European economies are assumed to gradually converge to their long-run average growth
and ination rates.
NGFS REPORT
21
4. Assessing economic
impacts
This Chapter sets out information on the process of
using scenarios to assess economic impacts. This includes
identifying the macroeconomic impacts assessed, relevant
transmission channels, the method of assessment, any
key assumptions and sensitivities, and rening the results.
4.1 Economic impacts assessed
For many types of climate scenario analysis, a key aspect
of the climate scenario will be the types of economic
impacts from the climate risks to be assessed. In the
short to medium term this could include impacts on the
level of GDP, unemployment and ination. Over long-term
horizons this could also include the cumulative impact on
the long-run determinants of growth (e.g. capital, labour
and total factor productivity) and changes in demand (e.g.
consumption, investment, government expenditure and
terms of trade). Climate scenarios may also provide some
insight on structural questions such as:
• Economic structure: What are the structural shifts between
sectors (e.g. from energy-intensive to less energy-
intensive)? Are there lasting changes from lowering
energy intensity, for example a shift in the share of GDP
from goods to services? This may also have a knock-on
impact on international trade and policy settings.
• International competitiveness and trade ows: How is
international trade affected by the materialising of
physical or transition risks? For example, the shift in
preferences away from carbon-intensive products
can have a signicant impact on terms of trade for oil
producers. Physical risks may similarly have an impact
on terms of trade, for example on food production. What
is the relative impact between regions and countries?
What is the eect on exchange rates?
• Policy settings: How will monetary policy adapt to
climate change? What would be the impact on natural
interest rates? Related to the scal stance, what would
be the impact on borrowing and debt; what impact
does this have on nancial variables like sovereign
bond yields?
4.2 Transmission channels
Transition risk
Macroeconomic impacts from transition risks arise
from a fundamental shift in energy and land use that
will aect every sector of the economy. At a high level
this could lead to some of the existing capital stock being
‘stranded’ and labour market frictions as the economy shifts
towards lower, and ultimately, net-zero emissions activities.
The size of the impacts will depend on how gradually and
predictably, or abruptly and disorderly, this transition takes
place, and how investment in new technologies aects
productivity.
These impacts are likely to aect economies in dierent
ways depending on economic structure, institutional
settings and the specific climate policies pursued.
These policies could include scal policy (e.g. carbon
pricing; public investment or subsidies), structural policy
(competition policy or labour market policy to help facilitate
the transition, impacting wage and price dynamics) and
regulation and standards (e.g. setting emissions standards
or targets for certain sectors).
Physical risk
Macroeconomic impacts from physical risk could arise
from both an increase in the frequency and severity of
severe weather events, and gradual climate change. These
risks may have wide-ranging direct economic impacts on:
• People: including labour productivity, mortality and
morbidity (e.g. from changes in temperature extremes)
and leisure;
• Physical capital: due to destruction of property and
infrastructure (e.g. from oods, windstorms) and diversion
of resources and investment into reconstruction and
replacement;
• Natural capital: due to disruption to agriculture (e.g. from
crop failure) and other ecosystem services (e.g. from
shifts in the productivity and distribution of sh stocks).
This could lead to signicant knock-on impacts on the
economy depending on the nature of the threat, the level
of resilience and level of local adaptation.
NGFS REPORT
22
4.3 Methods
Central banks have a range of models for making
economic forecasts. These models provide a central
organising framework, which can be deployed to study a
wide range of economic mechanisms and eects. These
macroeconomic models can be easily modied to assess
even some channels of climate risk, such as a change in
commodity prices or weather shocks that aect supply.
These physical risk impacts could be much larger, and
occur much sooner, than anticipated. The distribution of
events is shifting such that our historical analysis of both
the climate and economic impacts underestimate the size
of the risks. The earth is currently on a trajectory towards a
‘Hothouse Earth’ state with potentially irreversible impacts
(shown in Figure 5 below). This could be further accelerated
by tipping points such as loss of ice sheets, rainforest cover
and permafrost.
8 https://blogs.worldbank.org/ppps/embracing-uncertainty-better-decision-making
These factors make it very difficult to accurately
assess macro-nancial impacts once global warming
passes a certain threshold, such as 2°C of warming
compared to pre-industrial levels. For this reason, in
addition to the methods set out below, central banks
should consider decision-making frameworks for dealing
with deep uncertainty, such as those produced by the
World Bank.
8
Figure 5. Planetary thresholds and risks of a hot house earth pathway
Source: Steffen et al. (2018). Figure 5 shows the pathway of the Earth system out of the previous glacial-interglacial limit cycle to its present position in the hotter
anthropecene. Currently, the Earth System is on a Hothouse Earth pathway driven by human emissions of greenhouse gases and biosphere degradation toward a
planetary threshold at ~2 °C, beyond which the system follows an essentially irreversible pathway driven by intrinsic biogeophysical feedbacks. The other pathway
leads to Stabilized Earth, a pathway of Earth System stewardship guided by human-created feedbacks to a quasi-stable, human-maintained basin of attraction.
“Stability” (vertical axis) is defined here as the inverse of the potential energy of the system. Systems in a highly stable state (deep valley) have low potential energy,
and considerable energy is required to move them out of this stable state. Systems in an unstable state (top of a hill) have high potential energy, and they require
only a little additional energy to push them off the hill and down toward a valley of lower potential energy.
However, economic models typically used by central
banks have a number of limitations that make them
ill-suited to studying climate risks. Typically, they are
used to assess short-run divergences from long run
equilibria rather than investigate structural changes in the
economy. Also, they usually have a limited representation of
energy and agricultural systems, lack of economic sectoral
granularity and their modelling horizons often do not
extend much further than the business cycle.
NGFS REPORT
23
There are bespoke models that have been developed
to study interactions between physical and transition
risks and the economy. These models have primarily
been developed for academic research and/or advice for
policymakers. However, while broad in scope, they also have
a number of limitations. At the less complex end, only a
simple growth model is used or the costs (associated with
mitigation policies and/or climate damages) are estimated
in non-economic terms. While more complex models have
now also been developed, they still tend to focus on a
limited number of transmission channels and produce
a narrow scope of macroeconomic indicators. The NGFS
Scenarios are working to address some of these challenges.
In the interim it is likely that central banks will have to
deploy a combination of approaches to understand the
macroeconomic impacts. For example, climate-economy
models can be used to develop coherent scenarios, and
traditional macro models can be used to expand the number
of economic variables for assessing risks. Table 4 below sets
out the main types of models that exist and how they can
be used. They have been split by their lineage as either
climate-economy models or adapted macroeconomic
models. Also, see Box 2 for more information on the work
of Bank of Canada to estimate macroeconomic eects
using a CGE model.
Table 4. Types of economic models to assess climate risks
Lineage Model type Description Example
Integrated climate-economy
models
1
Cost-benet IAMs Highly aggregated model that optimises
welfare by determining emissions abatement
at each step
DICE, DSICE (Cai et al., 2012, Barrage, 2020)
IAMs with detailed energy system
and land use
Detailed partial (PE) or general equilibrium
(GE) models of the energy system and land use.
General equilibrium types are linked to a simple
growth model
PE: GCAM, IMAGE GE: MESSAGE,
REMIND-MAgPIE, WITCH
2
Computable General Equilibrium
(CGE) IAMs
Multi-sector and region equilibrium models
based on optimising behaviour assumptions
G-CUBED, AIM, MIT-EPPA, GTAP, GEM-E3
Macro-econometric IAMs Multi-sector and region model similar to CGE
but econometrically calibrated
E3ME, Mercure et al., 2018
Stock-ow consistent IAMs Highly aggregated model of climate change
and the monetary economy that is stock-ow
consistent
Bovari et al., 2018
Other climate-economy
models
Input-output (IO) models Model that tracks interdependencies between
dierent sectors to more fully assess impacts
Ju and Chen, 2010
Koks and Thissen, 2016
Econometric studies Studies assessing impact of physical risks
on macroeconomic variables (e.g. GDP, labour
productivity) based on historical relationships
Khan et al., 2019
Burke et al., 2015
Dell et al., 2012
Natural catastrophe models
and micro-empirical studies
Spatially granular models and studies assessing
bottom-up damages from physical risks
SEAGLASS (e.g. Hsiang et al., 2017)
Modied standard
macroeconomic models
DSGE models Dynamic equilibrium models based on optimal
decision rules of rational economic agents
Golosov et al., 2014
Cantelmo et al. 2019
E-DSGE Slightly modied standard frameworks (that
allow for negative production externalities)
Heutel, 2012
Large-scale econometric models Models with dynamic equations to represent
demand and supply, coecients based
on regressions
NiGEM (e.g. Vermeulen et al., 2018)
1 IAM taxonomy adapted from Nikas et al., 2019.
2 Model documentation available at www.iamcdocumentation.eu/index.php/IAMC_wiki
NGFS REPORT
24
Box 2
Using CGE models to estimate macroeconomic eects:
Lessons from the Bank of Canada
1 Ens and Johnston, 2020.
The Bank of Canada released a study that adapted
climate-economy models to better understand potential
sources of economic and nancial risks.
1
In it, the authors
set out examples of the types of scenarios that could
generate economic and nancial risks; they do this by
varying assumptions on key variables, like climate policy,
in plausible ways. They assess the risks around these
scenarios using a computable general equilibrium (CGE)
model that provides extensive sector-level detail on the
potentail impacts of each scenario. An IAM model is used
to inform a discussion of the economic costs and risks
associated with higher temperatures. The scenarios have
a long horizon, focussing on eects until 2050, but show
that the impacts could be material much sooner and over
a short period of time.
The results provide insights on the distribution of risks for
the global economy and nancial system, highlighting
signicant economic risks surrounding climate change and
the transition to a low-carbon economy. The timing and
magnitude of global and sectoral GDP impacts, among other
outcomes, look considerably dierent across the mix of
scenarios. The results also suggest that while transition risks
can be avoided through inaction, this comes at a signicant
economic cost through higher physical damages and risks.
Action that comes late (as proxied by the introduction of
carbon taxes) must be more abrupt to keep temperature
increases in check, which raises transition risks. Earlier
action also allows more time for new technologies to enter
the market in response to price signals, leading to a larger
green energy sector and lower transition costs.
4.4 Key assumptions and sensitivities
Transition risk
Climate scenarios are not projections. The scenario design
will have a signicant bearing on the nature and size of
the economic impacts. Some of the key transition pathway
assumptions include the speed and timing of policy action,
the type of policy implemented (taxes, regulations), the
progress in technology (both in carbon emission reduction
and in carbon capture and storage technology), and shifts
in behaviour from companies and consumers.
Models are also sensitive to assumptions made about
how the economy and nancial sector respond to shocks.
This includes assumptions related to:
• Market clearing: how much consumer demand will be
matched by the supply of goods in the short and/or
long run;
• Investment: whether the level of investment in the
economy is constrained by savings (possibly leading to
crowding out eects) or can grow;
• Role of the nancial sector: whether the nancial sector
eciently allocates capital and provides the investment
required or not;
• Monetary policy responses: how monetary policy responds
to shocks to the economy.
These assumptions help to explain why some models
suggest the transition will result in decreased growth while
others report a positive green growth eect. Equilibrium
models, such as CGEs, are generally characterised by market
clearing assumptions and are mostly without frictions. In such
models, investments are typically constrained by the level
of savings and economies always operate at full potential.
In non-equilibrium models, investment is not necessarily
required to match savings and money may be available to
fund investments and innovation. Non-equilibrium models
also tend to assume economies operating sub-optimally
and hence away from productivity frontiers. When these
imperfections (in the baseline) are resolved by climate policy,
the result can be improved eciency and higher growth
impacts. The eects of introducing market frictions have
also been replicated in some CGE studies.
9
9 Chateau and Saint Martin (2013) introduce labour market imperfections (restrictions to worker mobility and wage rigidity) into the baseline of a CGE
and implemented climate policy that addressed these imperfections by recycling carbon revenues to reduce labour taxes and maintain real wages.
NGFS REPORT
25
Physical risk
There is a great level of uncertainty around the current
estimates of economic damages that result from climate
change. Early approaches in cost-benet IAMs (e.g. in
DICE) were estimated using ‘enumerative’ approaches,
using impact assessment and expert judgment to quantify
dierent types of physical risk damages. Variations of these
functions are still being used widely but lack a proper
empirical foundation, and there is wide agreement that
they underestimate economic damages.
More recently, CGE approaches have focused on
developing an empirical, bottom-up assessment of
physical risks within an equilibrium framework. These
models have sought to quantify an increasing number
of channels (health, tourism, agriculture, sea-level rise)
based on updated impact estimates. The size of impacts
substantially depends on the channels covered and whether
the eects are considered to be temporary (e.g. drop in
agricultural production aects short-run GDP growth) or
permanent (e.g. increased temperatures reducing labour
productivity). Despite a recent growth in empirical studies,
high-level approximations of the economic impacts must
still often be made. This is due to lack of granular data,
uncertainty related to the underlying bio-physical processes
and uncertainty related to the future level of adaptation.
There is an increasing amount of macro-econometric
research that aims to empirically calibrate top-down
damage functions. By linking climate variables such as
temperature to aggregate macro outcomes they may
capture a wider range of damages than micro-founded
approaches. However they are still subject to a number
of limitations. These include:
• Non-linearities: studies using historical data make
implicit assumptions about future impacts. However
these historical trends may not hold in the future due to
socio-economic changes (e.g. migration), or because a
particular threshold has been reached (e.g. agricultural
yields or labour productivity drop o sharply above a
given level of climate change). Some studies have used
innovative approaches to account for these potential
non-linearities but are still subject to uncertainty about
future responses and adaptation.
• Channel coverage: macro-econometric approaches
may still not capture all relevant transmission channels.
This may be due to the spatial and temporal aggregation
of data (e.g. average yearly temperatures), or because
of a narrow focus on a single macroeconomic indicator
(e.g. labour productivity).
• Feedback effects: temperature change is typically
considered to be exogenous in models, however in
practice this will be aected by economic growth.
Further sources of uncertainty are the discount rate applied
to future damages and the uncertainties stemming from
the modelling of the climate impacts themselves.
4.5 Rening the results
Assessing macroeconomic impacts and vulnerabilities
is an iterative process. It may be useful to consider how
sensitive the results are to changing some of the key
assumptions in either the underlying climate scenario
(discussed in Chapter 3) or the macroeconomic modelling
(discussed in this Chapter). It may also be useful to consider
how much the scenario would have to change (e.g.
temperatures) to produce a given result (50% reduction in
GDP). This process of iterating on the scenario and exploring
dierent outcomes can be just as insightful as the size of
the impacts themselves.
NGFS REPORT
26
5. Assessing nancial
risks
This chapter provides information on using climate
scenarios to assess nancial risks. This includes identifying
the scope of nancial risks assessed, relevant transmission
channels, methods of assessment, key assumptions and
sensitivities and rening the results. Often it builds on
macroeconomic analysis done as part of the exercise (see
Chapter 4).
5.1 Financial risks assessed
Central banks and supervisors should rst consider the
nancial impacts they wish to measure and the metrics
that will be assessed. A targeted exercise may focus on
a small number of nancial rms, nancial asset classes
and types of risks – for example, using scenarios to assess
the agriculture-related credit risks for a few nancial rms.
Onthe other end of the scale, a system-wide stress test
could involve both nancial and macro channels, multiple
sectors and dierent types of nancial rms.
There are at least three dimensions to consider, informed
by the results from the initial materiality assessment
(see Chapter 2):
• Firm coverage: banks, insurers, asset managers, asset
owners, CCPs and other nancial market infrastructure;
• Financial risks: credit, market, operational, liquidity,
underwriting;
• Financial products: credits (e.g. mortgages, consumer
credit, corporate loans and bonds, sovereign bonds),
equities, derivatives, insured liabilities.
The depth of the analysis can dier depending on
the materiality of the climate risks in the scenario. For
example, while some risks (e.g. market risk on listed equities
in the energy sector) may require in-depth analysis, it may
be sucient to analyse less material risks (e.g. credit risk on
loans to IT services companies) using sectoral or macro-
level proxies of risk.
5.2 Transmission channels
Transition risk
Financial impacts could arise from direct exposure
to aected companies or households. The scenario
therefore needs to be suciently granular to assess the
costs and opportunities at the required sectoral and regional
granularity. Direct impacts could include:
• Corporate protability: companies could face higher
operating costs (e.g. on carbon-intensive inputs) or
changing demand for certain goods or services.
• Asset stranding: companies may have to write o capital
assets that are no longer economically viable and / or
permissible to use. For example, companies may lose
value because of changing market expectations on their
ability to generate income in the future (e.g. fossil fuel
companies with reserves that cannot be utilised).
• Corporate legal liability: to shareholders or investors due
to mismanagement of the transition. This could lead to
higher legal liabilities or directors and ocers (D&O)
insurance claims.
• Household income: due to households bearing some of
the costs of the transition, for example from higher taxes
(e.g. on fuel) or higher energy or food prices.
• Property values: where residential or commercial buildings
require signicant improvements in energy eciency or
other upgrades to be let or sold under building regulations.
These impacts could be further amplied by changes in
the broader macroeconomic environment discussed in
Chapter 4. Relevant factors will likely include the level of
output, employment, relative prices, interest rates, sovereign
debt and exchange rates.
Physical risk
Physical risks could also result in various nancial
impacts on households and companies. Some of the
direct impacts could include:
•
Corporate protability: revenues due to direct damage
or supply chain disruption. Companies could also face
higher costs from investing in adaptation.
• Household income: due to climate-related disruption of
economic activity or impacts on health.
NGFS REPORT
27
• Property values: where real estate or other infrastructure is
particularly exposed to a particular hazard (e.g. ooding
to coastal real estate). The nature of the nancial risks will
also depend on the price and availability of insurance.
While only a relatively small number of households or
companies may be exposed to the hazard, there may also
be knock-on impacts on the broader economy. These indirect
eects should be captured by the macroeconomic modelling.
5.3 Methods
Top-down and bottom-up exercises
Central banks and supervisors should dene the extent
to which they will perform the analysis themselves,
or invite nancial rms to participate in the exercise.
• Top-down exercise: central banks and supervisors apply
their own calculations to nancial institutions’ reported
data. A uniform framework permits greater consistency
and comparability of the results. However, granular data
as well as qualitative information is required to assess
climate risks in a meaningful way.
• Bottom-up exercise: the regulator chooses a scenario and
calibrates the scenario variables, but then asks nancial
institutions to perform quantitative and qualitative
analysis of how the scenario would aect their balance
sheets. Providing more granular scenario variables can
help limit the risk of inconsistent interpretations of
the scenario. Alternatively, the regulator may in a less
structured way encourage an industry-wide initiative by a
group of nancial institutions or industry associations to
proactively choose representative scenarios and share the
result of the analysis with central banks and supervisors.
There are advantages to both top-down and bottom-up
exercises. Top-down exercises are easier to plan and quicker
to execute as they do not require brieng and coordinating
with regulated rms. However, bottom-up exercises can
permit greater depth of analysis as nancial rms often
have more data than supervisors, thus allowing for a closer
analysis of the underlying risks.
In practice, nancial impact assessment often combines
both approaches to obtain multiple perspectives on the
impact of the scenario. For example, bottom-up exercises
will benet from in-house desk-based analysis to gain some
initial insights on the scenario and develop benchmarks
that can be used to conrm or challenge rms’ individual
results. On the other side, top-down exercises may benet
from review and/or some independent analysis from rms
or other subject matter experts to cross-check the results.
Modelling approaches
Financial risks can be modelled at varying levels of
granularity. At an aggregate level, macroeconomic
indicators from the climate scenarios (e.g. GDP,
unemployment) can be used to estimate the implied
impact on nancial markets (e.g. yields and equity indices).
However, for the reasons explained in Chapter 2 this will
not typically be granular enough to meaningfully assess
climate risks to a given portfolio.
Sectoral-level modelling approaches have been
developed to overcome some of these challenges. This
involves downscaling a macroeconomic indicator (e.g.
GDP) to sectoral-level (e.g. sector gross value added) using
relevant proxies for the underlying climate risks. See Box 3
for examples of how the Banque de France/ACPR and De
Nederlandsche Bank increased the sectoral resolution of
their climate risk analyses.
Given the complexity of the transmission channels, it will
often be insightful to model the risks at an even more
granular level, for example on individual companies
and households. This requires obtaining data on the
location and characteristics (e.g. emissions, energy use)
of the underlying borrower or issuer. Micro models (e.g.
cash ow models, natural catastrophe models) can then be
used to estimate impacts to the relevant indicator such as
property values, corporate protability and/or household
wealth. This analysis can also take account of the strategy
of the counterparty to respond to the risks where this
information is accessible.
Bottom-up quantication can inform, and be informed
by, top-down modelling of aggregate eects. See the
Bank of England’s 2021 Biennial Exploratory Scenario as
an example of how top-down and bottom-up approaches
can be combined.
10
10 Bank of England, 2019.
NGFS REPORT
28
Box 3
How Banque de France/ACPR and De Nederlandsche Bank
increased the sectoral resolution of their climate risk analyses
1 Allen et al., 2020.
2 Devulder and Lisack, 2020.
3 Vermeulen et al., 2018.
Banque de France/ACPR
The Banque de France and ACPR have developed a
climate stress test framework focused on transition risks.
1
The economic modelling in the framework consists of
several components, including the National Institute
Global Econometric Model (NiGEM) model. Since NiGEM
produces only aggregate economic outputs, the model is
coupled with a static general equilibrium sectoral model
developed by Banque de France, which is designed
to propagate a tax shock and/or a productivity shock
across sectors.
2
The model relies on input-output data to represent the
production in each sector in each country, as a process
involving non-energy and energy intermediate inputs
from all countries, and domestic labour. All these inputs
are substitutable to various degrees, and rms optimise
their intermediate demands given the relative prices
of inputs in a perfectly competitive environment. The
model is then closed to form a general equilibrium set-up
by adding a representative household in each country,
supplying labour inelastically and consuming goods
from all countries.
The shares of inputs in production, the relative sizes of
the sectors and the consumption shares are calibrated
to match sectoral input-output and nal consumption
data from the World Input Output Database (WIOD). The
substitution elasticities are obtained from the literature.
The baseline version of this model assumes perfect
risk sharing across countries, imposing that relative
consumption responds positively to changes in real
exchange rates. For simplicity, the model ignores physical
capital, such that production requires only labour and
intermediate inputs. The demand side amounts to nal
consumption from households.
De Nederlandsche Bank
In De Nederlandsche Bank’s transition risk stress test
3
,
four disruptive transition scenarios are simulated with
an adapted version of NiGEM to create a set of mutually
consistent paths for a set of macroeconomic (e.g. GDP,
unemployment, price level) and macro-nancial (e.g.
interest rates, stock price indices) variables. Since
NiGEM produces economic outputs at an aggregate
level by geographical region, and not at sector level,
DeNederlandsche Bank developed sector-specific
“transition vulnerability factors” (TVFs). The TVFs allowed
the NiGEM outputs to be translated to a sectoral level.
The approach can be summarised as follows:
• The TVFs are dened such that the average TVF of
the economy (weighted by the value added of each
economic sector) is equal to 1. In the stress test scenarios,
sectors with a TVF smaller than 1 are aected less than
the economy as a whole, while sectors with a TVF larger
than 1 are aected more than the economy as a whole.
• The TVF of each sector is dened as the embodied
emissions of a sector relative to its value added.
Embodied emissions include all emissions created in
the production process for a rm’s nal goods, including
all upstream emissions created in other sectors. The
emissions and value added data were sourced from
EXIOBASE, a global and detailed input-output database
that covers a wide range of countries and industries.
• The TVFs are then adjusted in some of the scenarios
to more accurately reect the risk drivers that were in
play in each scenario, since embodied emissions alone
do not always best capture the risks.
• The TVFs are multiplied by the stock price indices
simulated with NiGEM to produce sectoral stock price
impacts. The sectoral stock price impacts can be used
to calculate losses on equity exposures and also served
as input to calculate losses on loans and bonds.
NGFS REPORT
29
Expanding the scenario by modelling
additional variables
Additional variables may be needed due to the limited
number of macro-nancial outputs available from the
climate model underpinning the scenario. For example,
the scenario model may provide some detail on the impact
on output and interest rates, but nothing on the yield curve.
If, for practical reasons, these additional features cannot
be embedded in the underlying model, they may have to
be estimated in other ways. Options include:
• Simulate the missing variables in a separate model.
One could take a model that can produce the desired
outputs and then calibrate the model, as closely as
possible, on the basis of the chosen scenario. An advantage
of this method is that it is model-based, thus providing an
economic rationale for the outputs. Another advantage
of using a suite of models is that it can capture a broader
range of relevant transmission channels, and thereby
provide a more comprehensive view of the impacts.
A disadvantage of this strategy is that it becomes more
dicult to ensure the internal consistency of the scenario
since the model used to produce the scenario diers
from the models used to produce additional outputs.
This can be managed by ensuring a consistent set of
assumptions where possible and using the results to
recalibrate the scenario.
• Wider estimates from academic literature. If the scenario
does not provide the required variable it is possible that
it has been estimated in other studies (e.g. the potential
impact of ooding on supply chain risk).
• Past trends. By observing how the variables of interest
moved during historical periods, one may form an
educated guess about what would happen in the
scenario. For example, one could analyse the eects
of a previous oil price drop (e.g. following the great
nancial crisis) or extreme temperature changes (e.g. 2003
European heatwave) on a particular exposure. However,
this option should be approached with caution given the
likelihood of climate risks resulting in unprecedented,
structural changes.
Time and discounting
Given a scenario and type of exercise, one may face
some further methodological questions. Some typical
dimensions for which further assumptions may be required
include:
•
Balance sheet evolution. If the scenario plays out over
time, as opposed to a point-in-time ‘snapshot’ scenario,
the behaviour of nancial institutions might evolve as the
scenario unfolds (Table 5). In many stress testing exercises,
for example, the simplifying assumption of static balance
sheets is made, requiring nancial institutions to hold
their portfolios constant over time and replace maturing
assets with new, similar assets. In contrast, dynamic
balance sheets allow for the inclusion of management
actions, so that institutions can react to climate regulatory
changes, news or different customers’ preferences.
Over long time horizons, management actions will be the
primary driver of impacts but are very dicult to predict.
Box 4 sets out how the Bank of England approached the
challenges of long-time horizons in its 2021 Biennial
Exploratory Scenario.
•
The discount rate. If the scenario has a relatively long time
horizon, the question of whether and how to discount
nancial values in later periods should be considered.
This is relevant in bottom-up exercises where the balance
sheets of participating rms are allowed to change
over time.
Table 5. Assumptions of balance sheet evolution
Focus Time dimension
Static analysis Risk on current balance sheet Understand current exposure
Less dependent on assumptions
Dynamic analysis Risks associated with potential changes in
balance sheets, also as a consequence of
changes in behaviour
Dependent on assumptions about behaviour
NGFS REPORT
30
Data collection
Limited available data and research are a signicant
challenge. Central banks and supervisors should consider
the data they need to assess the risks themselves and in
bottom-up exercises the data needed by rms. A typical
data collection process could be broken down into the
following questions:
• Which data is readily available? To minimise administrative
burden, the risk assessment should be based as much
as possible on readily available data. At the same time,
the unique nature of climate-related risks can imply that
a proper risk assessment requires data that has not yet
been collected. For instance, available climate-related
datasets often cover only public companies and very
rarely privately-owned companies, to which nancial
institutions are exposed. This can pose a signicant
obstacle to the analysis.
• Can the required data be constructed on the basis of
available datasets? Often, some data is available but
distributed over various datasets, which would need to
be combined to create one coherent set. See Box 5 for
more information on the DNB’s approach to this in their
transition stress test. Such sectoral data may still not
be granular enough to assess rm-level risks. This may
require combining top-down (sectoral statistics) and
bottom-up (rm level operating activity) data.
• Can the required data be requested from nancial institutions
and/or their counterparties? This option will be most viable
in bottom-up stress testing exercises. This may have an
ancillary benet of promoting more engagement on
climate risk management between nancial institutions
and the real economy. Before conducting such a survey,
however, it is useful to check: (a) whether institutions
themselves or their clients have access to the desired
data, and (b) which format of delivery would be both
manageable for the nancial institutions and workable
for the institution carrying out the analysis.
Box 4
The Bank of England’s modelling framework for the 2021
Biennial Exploratory Scenario on the nancial risks from climate change
1 Bank of England, 2019.
2 Physical risks in the No Additional Policy Action Scenario are proposed to be assessed from 2050-2080.
In its 2021 Biennial Exploratory Scenario (BES), the Bank
of England has proposed a detailed exploration of the
impact of climate scenarios on banks’ and insurers’ balance
sheets with a time horizon of 2020 to 2050.
1
To make this
approach feasible, participating banks and insurers would
have to calculate impacts over time in the following way:
• Participating institutions model the impact of the
scenarios up to 2050 (i.e. a 30-year time horizon with
impacts assessed at 5 yearly intervals).
2
However,
the physical risk variables would be calibrated in a
conservative way to capture the physical impacts in
the second half of the century.
• To make the modelling feasible, in the rst part of the
exercise participants assume the nominal size and
composition of their balance sheets to be xed over
the time horizon of the scenario. In practice this means
rms assess climate-related risks at each point of the
time horizon against their current balance sheet.
• In the second part of the exercise, this constraint would
be relaxed, and rms identify the management actions
they would take to reduce their risks. These would
be reported as a mix of qualitative responses and
quantitative metrics.
NGFS REPORT
31
Outputs
Central banks and supervisors may wish to translate the
nancial risks into relevant metrics to inform decision-
making. Relevant outputs could include: asset impairment,
mark-to-market valuation, risk weighted asset ratios, capital
buer depletion, return on equity, and change in business
model (portfolio allocation, lending paths, insurance
coverage and pricing). The metrics chosen should align
with the objectives of the exercise (see also Chapter 3 of
the NGFS Guide for Supervisors).
5.4 Key assumptions and sensitivities
Transition risk
The types of transition risks that are considered, and
the way in which these risks are modelled, can have a
strong bearing on the results of the exercise. Capturing
transition risks is challenging both because it can materialise
in complex and varying ways, and because data and
model aggregation might make it dicult to accurately
pinpoint where the risks materialise. Some examples of
the sensitivities in modelling transition risks are:
• Multiple transmission channels: there may be revenue
drivers (a decline in sales), cost drivers (carbon prices)
and asset devaluation (e.g. stranded fossil fuel assets,
real estate), with varying degrees of impact.
Box 5
Data collected by De Nederlandsche Bank for its transition risk stress test
1 Vermeulen et al., 2018.
The transition risk stress test conducted by De
Nederlandsche Bank
1
considered the equity, bond and
corporate loan portfolios (for banks), of more than 80 banks,
insurers and pension funds located in the Netherlands.
The data collection process included the following steps:
1) The equity and bond exposures of Dutch nancial
institutions were gathered from the ECB’s Securities
Holdings Statistics (SHS) database. Using International
Security Identier Numbers (ISINs) at security-level, the
exposures were matched with NACE codes from the
Centralized Securities Database (CSDB) to determine
to which sector each exposure belonged. If no NACE
code could be identied using the CSDB, the ISINs were
cross-checked against the Thomson Datastream ratings
database. In this way, industry classications in NACE,
NAICS, GICS, TRBC and SP format were collected for all
stocks and bonds in the banks’, insurers’ and pension
funds’ portfolios. The alternative classication formats
were converted to the NACE format through publicly
available mappings.
2) An issue that emerged as a result of using NACE codes
was that a large amount of securities were classied as
NACE code K.64 (nance) while, upon closer inspection,
many of these K.64-classied securities were issued
by nancing vehicles of rms that are active outside
of the nancial sector. For example, a bond issued by
BMW Finance is marked as K.64 (nance), while for the
purposes of the stress test it was more appropriate to
assign it to C.29 (manufacturing of motor vehicles), i.e.
the industry of the parent company. To correct this,
the stock and bond holdings were again checked
against the alternative classications in the Thomson
Datastream ratings database; the K.64 code was
replaced if one of the other databases listed a dierent
industry classication.
3) To obtain NACE-codes for the corporate loan portfolios
of banks, De Nederlandsche Bank conducted a targeted
survey among the largest Dutch banks. To ensure
simplicity for banks and usability in the stress test
models, the survey resembled standard regulatory
reporting (i.e. COREP) templates. The survey provided
the exposure amounts of banks’ IRB- and SA-portfolios
disaggregated by sector. For IRB-modelled loans the
exposures were further disaggregated by internal
risk bucket, probability of default, loss given default
and maturity.
NGFS REPORT
32
• Business model changes: there could be shifts in corporate
business models in response to climate shocks (e.g. a rm
could rebalance away from fuel-intensive production
following a new, stringent energy law).
• Pass through costs: the reaction of rms and consumers
will depend on technological constraints and preferences,
respectively, which will aect supply/demand elasticities
along the value chains and at nal consumption level.
• Classifying counterparties: where a standard industrial
sector taxonomy is used often balance sheets assets
cannot be neatly categorised, particularly for large
companies that span multiple activities.
Physical risk
Financial impacts from physical risks should be
understood as having a wide band of uncertainty,
particularly further out in the time horizon. The size
of the nancial risks depends on assumptions about how
the economy and nancial system will respond to events
that have no precedent. While some micro impacts may
be based in part on existing channels that are regularly
assessed (e.g. impact from ood damage on insurance
claims) the probability and / or impact of many other
channels has not been robustly estimated (e.g. costs from
supply chain disruption). Even where case studies exist,
it may not be easy to readily identify the locations of the
economic activity and supply chain from the data.
5.5 Rening the results
Given the novel nature of climate risks, both central banks
and supervisors and (where relevant) participating rms
will likely learn a lot about the underlying transmission
channels and key sensitivities in the rst round.
At the end of the exercise, central banks and supervisors
should consider revisiting the scenario assumptions and
performing a second round of the exercise. This can be
useful to explore systemic risks (e.g. participating rms all
indicate they will exit from a particular sector at the same
time) or any other channels that were not identied during
the initial materiality assessment.
NGFS REPORT
33
6. Communicating
and using the results
This chapter sets out the nal stage of communicating and
using the results of scenario analysis.
6.1 Communication of the results
Communicating the results of scenario analysis improves
awareness of climate risks. This may encourage rms to
improve their risk-management practices and foster further
research – particularly where new pockets of risk have
been identied.
Information disclosed
Central banks and supervisors should consider the
information disclosed, given that climate scenario
analysis methodologies are still evolving, and that
the lack of data remains a signicant barrier. This is
particularly relevant for scenario-based stress testing
exercises where the supervisor has a choice of publishing
individual and/or system-wide results (e.g. means and
ranges). Details disclosed could include qualitative and
quantitative information on the scenario, impact on
nancial variables (e.g. asset quality, stock prices), regulatory
numbers (e.g., capital, leverage and liquidity) as well as
impact on macroeconomic variables (e.g. GDP, changes
in the capital stock, sectoral shifts).
Since climate-related scenario analysis is a relatively
novel activity, there is also significant value in
sharing details on the methods, assumptions and key
sensitivities. This includes the objectives, the specic
scenarios, risk coverage, the rationale for the selections,
as well as any limitations of the analysis and how these
might aect the results. This communication can help
establish market conventions and practices on disclosure.
Eective internal communication is also critical for building
organisational capacity and integrating the results into
supervisory approaches.
Dene target audiences
The target audience will be closely tied to the
stakeholders identified as part of scoping out the
exercise (see Chapter 2). The audience may include
nancial rms, standard setters, general public, government
including international bodies, other central banks and
supervisors and the academic community.
Select communication methods
Numerous communication options are available to
central banks and supervisors looking to share the
results of their scenario analysis (see Table 6). Public
disclosure can take place on websites, periodic publications
(analytical notes, nancial stability reports), via speeches by
senior ocials and on social media. Conferences are also
an eective way to have direct discussions with specialists
and related parties on the analysis. For rm-specic results,
bilateral meetings may be more appropriate.
Table 6. Communication methods
Communication methods Target audience Objectives
Disclosure Public
Government
Raise awareness
Provide detailed information
Encourage initiatives such as the TCFD
Inform government policy action
Conferences Public
Specialists
Related parties
Raise awareness
Eective and timely communication
Two-way dialogue
Bilateral meeting Government
Institutions
Raise awareness
Two-way dialogue
Share the result of comparative analyses and
range of practices
Provide feedback to encourage advancement
of institution’s risk management practices
Inform government policy action
Internal communication Central bankers Supervisors Raise awareness
Receive valuable inputs
Consider nancial regulatory initiatives
Training
NGFS REPORT
34
6.2 Uses of the results
Scenario analysis should be an ongoing iterative
process. The initial results will identify new pockets of risks
and key sensitivities of the scenario that were not initially
included. These aspects can form the basis of follow-up
analysis and research. Possible follow-up actions include:
• Using the insights as part of supervisory decision-making.
Forinstance, requesting more detailed information on
climate risks from rms, such as exposures, plans for
enhancing its risk management framework, and its
strategy for climate-related risks. (See NGFS Guide for
Supervisors).
Box 6
Communication Examples
Some central banks and supervisors have communicated
and/or are considering communicating the results of
scenario analysis or stress testing in various ways.
• Banco de España will share the result of stress testing
with institutions and publish it in their nancial stability
report.
• Bank of England will disclose aggregate system-wide
results of their climate stress tests including means
and ranges, and provide feedback to individual rms.
• Banque de France/ACPR will disclose only aggregate
system-wide results and provide feedback on an
individual basis to specic rms to ensure the coherence
of the overall exercise.
•
De Nederlandsche Bank has published a rst estimate
of the possible impact of a disruptive energy transition
on the Dutch nancial sector.
• Monetary Authority of Singapore (MAS) intends
to share the results of any climate stress tests with all
participants to encourage the adoption of best practices,
and spur development in terms of modelling techniques
and climate data gathering.
• Identify whether the risks are being suciently mitigated
by existing processes. For example, scenario-based stress
testing may help identify risks that are under / over
capitalised. Inaddition, the macroeconomic assessment
could provide insights on channels that are not yet
captured as part of regular economic forecasting.
• Scenario analysis on own operations may identify how
climate change could aect the risk in and eectiveness
of central banks’ operational policies, such as its balance
sheet investments. Central banks may also include the
results in thematic and impact investing considerations,
screening criteria for asset purchases, and voting and
engagement. Applying climate change scenarios when
assessing the value of the central bank’s own portfolio
can provide an opportunity "to lead by example".
11
11 Battiston, 2019.
NGFS REPORT
35
Allen, T., Dees, S., Boissinot, J., Caicedi Graciano, M.,
Chouard, V., Clerc, L., … & Vernet, L. (2020)
Climate-related scenarios for nancial stability assessment:
An application to France. Banque de France Working paper,
forthcoming.
Bank of England (2019)
2021 Biennial Exploratory Scenario on the nancial risks
from climate change. Bank of England, Discussion paper,
33 pp.
Banque de France/ACPR (2020)
Présentation des hypothèses provisoires pour l’exercice pilote
climatique. Autorité de Contrôle Prudentiel et de Résolution,
Consultation paper, 36 pp.
Barrage, L. (2020)
Optimal Dynamic Carbon Taxes in a Climate–Economy
Model with Distortionary Fiscal Policy. The Review of
Economic Studies, 87(1), 1-39.
Battiston, S. (2019)
The importance of being forward-looking: managing
nancial stability in the face of climate risk. Financial Stability
Review, Banque de France, 23.
Bovari, E., Giraud, G., & Mc Isaac, F. (2018)
Coping with collapse: a stock-ow consistent monetary
macrodynamics of global warming. Ecological Economics,
147, 383-398.
Burke, M., Hsiang, S. M., & Miguel, E. (2015)
Global non-linear eect of temperature on economic
production. Nature, 527(7577), 235-239.
Cai, Y., Judd, K. L., & Lontzek, T. S. (2012)
DSICE: a dynamic stochastic integrated model of climate and
economy. RDCEP Working Paper 12-02.
Cantelmo, M. A., Melina, M. G., & Papageorgiou, M. C.
(2019)
Macroeconomic Outcomes in Disaster-Prone Countries.
International Monetary Fund, Working Paper No. 19/217,
October 2019.
Chateau, J., & Saint-Martin, A. (2013)
Economic and employment impacts of climate change
mitigation policies in OECD: A general-equilibrium
perspective. International Economics, 135, 79-103.
Christiano, L. J., Eichenbaum, M. S., & Trabandt, M. (2018)
On DSGE models. Journal of Economic Perspectives, 32(3),
113-40.
Danmarks Nationalbank (2019)
Climate change can have a spillover eect on nancial stability.
Danmarks Nationalbank Analysis, N° 26.
Dell, M., Jones, B. F., & Olken, B. A. (2012)
Temperature shocks and economic growth: Evidence
from the last half century. American Economic Journal:
Macroeconomics, 4(3), 66-95.
Devulder, A., & Lisack, N. (2020)
Carbon Tax in a Production Network: Propagation and Sectoral
Incidence. Banque de France Working paper n° 760, April 2020.
EIOPA (2019)
Insurance sector climate-related transition risks. EIOPA
Discussion paper, EIOPA-BoS-19-571.
Ens, E. & Johnston, C. 2020
Scenario Analysis and the Economic and Financial Risks from
Climate Change. Bank of Canada Sta Discussion Paper/
Document d’analyse du personnel – 2020-3.
Golosov, M., Hassler, J., Krusell, P., & Tsyvinski, A. (2014)
Optimal taxes on fossil fuel in general equilibrium.
Econometrica, 82(1), 41-88.
Heutel, G. (2012)
How should environmental policy respond to business
cycles? Optimal policy under persistent productivity shocks.
Review of Economic Dynamics, 15(2), 244-264.
Hsiang, S., Kopp, R., Jina, A., Rising, J.,
Delgado, M., Mohan, S., ... & Larsen, K. (2017)
Estimating economic damage from climate change in the
United States. Science, 356(6345), 1362-1369.
Bibliography
NGFS REPORT
36
Huppmann, D., Rogelj, J., Kriegler, E., Krey, V.,
& Riahi, K. (2018).
A new scenario resource for integrated 1.5° C research.
Nature climate change, 8(12), 1027-1030.
IPCC (2014)
Climate Change 2014: Synthesis Report. Contribution of
Working Groups I, II and III to the Fifth Assessment Report
of the Intergovernmental Panel on Climate Change [Core
Writing Team, R.K. Pachauri and L.A. Meyer (eds.)]. IPCC,
Geneva, Switzerland, 151 pp.
Ju, L. & Chen, B. (2010)
An input-output model to analyse sector linkages and
CO
2
emissions. Procedia Environmental Sciences, 2,
1841-1845.
Kahn, M. E., Mohaddes, K., Ng, R. N., Pesaran, M. H.,
Raissi, M., & Yang, J. C. (2019)
Long-term macroeconomic eects of climate change: A cross-
country analysis (No. w26167). National Bureau of Economic
Research.
Koks, E. E., & Thissen, M. (2016)
A multiregional impact assessment model for disaster
analysis. Economic Systems Research, 28(4), 429-449.
Mercure, J. F., Pollitt, H., Viñuales, J. E.,
Edwards, N. R., Holden, P. B., Chewpreecha, U., ...
& Knobloch, F. (2018)
Macroeconomic impact of stranded fossil fuel assets. Nature
Climate Change, 8(7), 588-593.
NGFS (2019a)
A call for action: Climate change as a source of nancial risk.
First Comprehensive Report, Network for Greening the
Financial System, Paris, France 42 pp.
https://www.ngfs.net/sites/default/les/medias/documents/
ngfs_rst_comprehensive_report_-_17042019_0.pdf
NGFS (2019b)
Macroeconomic and financial stability: Implications of
climate change. Technical supplement to the First NGFS
Comprehensive Report, Network for Greening the Financial
System, Paris, France, 51 pp.
https://www.ngfs.net/sites/default/les/medias/documents/
ngfs-report-technical-supplement_nal_v2.pdf
NGFS (2020)
Guide for Supervisors – Integrating climate-related and
environmental risks in prudential supervision. Network for
Greening the Financial System, Paris, France, 62 pp.
https://www.ngfs.net/sites/default/les/medias/documents/
ngfs_guide_for_supervisors.pdf
Nikas, A., Doukas, H., & Papandreou, A. (2019)
A detailed overview and consistent classication of climate-
economy models. In Understanding Risks and Uncertainties
in Energy and Climate Policy (pp. 1-54). Springer, Cham.
Norges Bank (2019)
Technological advances and climate measures can inuence
banks’ credit risk. Norges Bank Sta Memo, NR 6.
O’Neill, B. C., Kriegler, E., Riahi, K., Ebi, K. L.,
Hallegatte, S., Carter, T. R., ... & van Vuuren, D. P.
(2014)
A new scenario framework for climate change research:
the concept of shared socioeconomic pathways. Climatic
change, 122(3), 387-400.
O’Neill, B. C., Tebaldi, C., Van Vuuren, D. P., Eyring, V.,
& Friedlingstein, P. Hurtt, 440 G., ... Sanderson, BM
(2016)
The Scenario Model Intercomparison Project 441
(ScenarioMIP) for CMIP6. Geoscientic Model Development,
9 (9), 3461–3482. 442.
Riahi, K., Van Vuuren, D. P., Kriegler, E., Edmonds, J.,
O’neill, B. C., Fujimori, S., ... & Lutz, W. (2017)
The shared socioeconomic pathways and their energy,
land use, and greenhouse gas emissions implications: an
overview. Global Environmental Change, 42, 153-168.
Steen, W., Rockström, J., Richardson, K., Lenton, T. M.,
Folke, C., Liverman, D., ... & Donges, J. F. (2018)
Trajectories of the Earth System in the Anthropocene.
Proceedings of the National Academy of Sciences Aug 2018,
115 (33) 8252-8259.
TCFD (2017)
Final Report: Recommendations of the Task Force on Climate-
related Financial Disclosures. Task force on Climate-related
Financial Disclosures, 74 pp.
NGFS REPORT
37
Van Vuuren, D. P., Edmonds, J., Kainuma, M., Riahi,
K., Thomson, A., Hibbard, K., ... & Masui, T. (2011)
The representative concentration pathways: an overview.
Climatic change, 109(1-2), 5.
Vermeulen, R., Schets, E., Lohuis, M., Kolbl, B.,
Jansen, D. J., & Heeringa, W. (2018)
An energy transition risk stress test for the nancial system of
the Netherlands. DNB Occasional Studies, 160-7.
NGFS REPORT
38
Acknowledgements
This Guide on Scenario analysis is a collaborative eort of the members of the macronancial workstream 2 of the NGFS.
The Guide was coordinated jointly by the NGFS Secretariat at the Banque de France (Thomas Allen and Antoine Boirard)
and the Chair’s team at the Bank of England (Ryan Barrett and Theresa Löber). Tanguy Séné (Bank of England) provided
research and editing support.
The Chair of the Macronancial workstream is grateful to the four sub-group leads for their signicant contributions:
Rie Asakura (Japan Financial Services Agency), Stéphane Dees (Banque de France), Edo Schets (De Nederlandsche Bank)
and Haakon Solheim (Norges Bank). Members of the sub-group were: Valérie Chouard (Banque de France), Erik Ens
(Bank of Canada), Dejan Krusec (European Central Bank), Giacomo Manzelli (Banca d’Italia) and Laura Parisi (European
Central Bank). We also extend our acknowledgements to Christoph Bertram, Elmar Kriegler, Franziska Piontek (Potsdam
Institute for Climate Impact Research) and Alex Koberle (Imperial College London) who provided external comments.
The Chair is also grateful to all other NGFS Members and Observers who provided comments and feedback. This included
BaFin, Banca d’Italia, Banco de España, Banco de Portugal, Banco de la República (Colombia), Bank Al-Maghrib, Bank
Negara Malaysia, Bank of Canada, Bank of Japan, Banque de France/ACPR, Bundesbank, Central Bank of Hungary, Comision
Nacional Bancaria y de Valores (Mexico), Danmarks Nationalbank, De Nederlandsche Bank, European Banking Authority,
European Central Bank, Hong Kong Monetary Authority, International Monetary Fund, Japan FSA, Monetary Authority
of Singapore, National Bank of Slovakia, Norges Bank, OECD, Reserve Bank of Australia, Reserve Bank of New Zealand,
Sveriges Riksbank, Swiss National Bank and the World Bank.
NGFS REPORT
39
Annex – Examples of scenario analysis
This annex sets out more information on how individual central banks and supervisors have used climate scenarios to
assess macroeconomic and nancial risks. These are summarised in Table 7 below.
Table 7. Examples of published climate-related risk assessments by central banks and supervisors
Authority Publication Type of method Description
Bank of Canada Link Scenario analysis to better
understand macroeconomic
and nancial system risks
(desk-based)
Explore illustrative scenarios to assess transition and physical risks
related to climate change. Examine macroeconomic, sectoral and
technological shifts using a CGE model (supported by results
from an IAM.)
Bank of England (i) Link Stress test (rm-based) Participating institutions (large UK banks and insurers) are required
to calculate the impact on their exposures for three detailed
climate scenarios provided by the Bank of England.
Bank of England (ii) Link Stress test (rm-based) Insurers analysed impact of physical and transition risk on both
their assets and liabilities in three policy scenarios.
Banque de France/
ACPR (i)
Link Financial system exposure
analysis (desk-based)
Analysis of exposures of French banks and insurers to sectors with
high GHG emissions.
Banque de France/
ACPR (ii)
Link Stress test (rm-based) Pilot exercise to assess the resilience of large French banks and
insurers to four climate scenarios including transition (banks and
insurers) and physical risks (insurers only).
Danmarks Nationalbank Link Financial system exposure
analysis (desk-based)
Analysis of how projected sea-level rise could aect nancial
institutions’ mortgage collateral in Denmark.
De Nederlandsche Bank Link Stress test (desk-based) Analysis of how the asset-side exposures of Dutch banks, insurers
and pension funds are aected in scenarios of a disruptive energy
transition.
ECB Link Financial system exposure
analysis (desk-based)
Analysis of large exposures of European banks, insurers, investment
funds and pension funds to climate-sensitive sectors.
ESRB/ATC-ECB/FSC Link Stress test (desk-based) A forthcoming example is one jointly conducted by the ECB,
the European Systemic Risk Board and the European System of
Central Banks: this stress-test is a ‘pilot’ exercise and ultimately aims
at identifying data gaps and methodological limitations for the
assessment of climate-related risks. It investigates the materiality
of transition risks for banks’ solvency and their lending capacity,
also looking at the implications for the overall economy using
a dynamic setting.
MAS Not published Stress test (rm-based) Selected general insurers were required to assess the impact on
their exposures through insured properties (by considering a list of
ood-prone areas in Singapore) as well as the possible implications
on their business lines under a climate variability scenario featuring
an extreme ooding event.
Norges Bank Link Identication of vulnerabilities
and triggers (desk-based)
Exploration of various scenarios for the oil industry, and of how
the Norwegian economy and nancial sector may be aected
in these scenarios.
NGFS
Secretariat